30183
I N s  I DE
_!  \         t -      ~~~~~~~~slumi(lltul C;11c('11 11(i
%i  r O\  tin  R\o l l11~ i oOIC
on 11  1\\ 11w  \\lIlc'Io s   PooiJS
E~~~~ WOL  BANK                            11\ I)i.l'Q('(i sil\ (SI( 111
D)ui\idI'. IOveOl.s
TNH E W OR L D  A N K                             Nvicl 0ll7('Shi2l
w   \ Cr x ]  n  b D<t j             1)~~~~~~onldt( 1: k.{IISOIn.
\ _ M1; | " kl~~~~~oc5k R.                        mid(tlSl.z1(
\  J O  i  J  l  J1.     1           t ~~~~~~Plzos \l"Igis
.   \            '        (>11 \II1lala7ion   Risk in
J  JU l L 'L  V  t   SL<PR           S\(l1\\l111mn 1. sildcHl(l
. , =, ' | \\~~~~~~~~~illid      " rstsl ) Alll
,~~~~~~~~( \m lI'llll (m, RI§i lsic
F(l((hic ()ll ill Dc(\clop)inZ-
1,1c()llOllli(s
--ms       PXToIH                               ISS\ 02,-)7)-,'(J):'2
/t'NIVERST 1\\\\ )(.),|j)l ]lS(§



*MI v: 
Research Observer
EDITOR
Shantayanan Devarajan, World Bank
COEDITORS
Gershon Feder, World Bank
Shahid Yusuf, World Bank
EDITORIAL BOARD
Paul Collier, World Bank
Susan Collins, Georgetown University
Angus Deaton, Princeton UJniversity
Barry Eichengreen, University of California-Berkeley
Emmanuel Jimenez, World Bank
Danny M. Leipziger, World Bank
Howard Pack, University of P'ennsylvania
Michael Walton. World Bank
Thie World Bank Researchi Observer is intended for anyone who has a professional interest in
development. Observer articles are written to be accessible to nonspecialist readers; con-
tributors examine key issues in development economics, survey the literature and the lat-
est World Bank research, and debate issues of development policy. Articles are reviewed by
an editorial board drawn from across the Bank and the international community of econo-
mists. Inconsistency with Bank policy is not grounds for rejection.
The journal welcomes editorial comments and responses, which will be considered for pub-
lication to the extent that space permits. On occasion the Observer considers unsolicited
contributions. Any reader interested in preparing such an article is invited to submit a
proposal of not more than two pages to the Editor. Please direct all editorial correspon-
dence to the Editor, Tlhe World Batik Research Observer. 1818 H Street, NW, Washington, DC
204 3 3, USA.
The views and interpretations expressed in this journal are those of the authors and do not
necessarily represent the views and policies of the World Bank or of its Executive Directors
or the countries they represent. The World Bank does not guarantee the accuracy of data
included in this publication and accepts no responsibility whatsoever for any consequences
of their use. When maps are used, the boundaries, denominations, and other information
do not imply on the part of the World Bank Group any judgment on the legal status of any
territory or the endorsement or acceptance of such boundaries.
For more information, please visit the Web sites of the Research Observer at
www.wbro.oupjournaLs.org, the World Bank at www.worldbank.org,
and Oxford University Press at www.oup.com/us.



Research Observer
Volume 19 . Number 2 . Fall 2004
How Have the WorLd's Poorest Fared since the Early 1980s?                 141
Shaohua Chen and Martin Ravallion
Much Ado about Nothing? Do Domestic Firms RealLy Benefit                  171
from Foreign Direct Investment?
Holger Gbrg and David Green away
Policies on Managing Risk in AgricuLturaL Markets                         199
Donald F. Larson. Jock R. Anderson, and Panos Varangis
Oil, Macroeconomics, and Forests: Assessing the Linkages                  231
Sven Wunder and William D. Sunderlin
Vouchers for Basic Education in DeveLoping Economies:                     259
An Accountability Perspective
Varun Gauri and Ayesha Vawda



Subscriptions
A subscription to The World Bank Research Observer (ISSN 02 57-3032) comprises 2 issues. Prices include postage: for
subscribers outside the Americas, issues are sent air freight.
Annual Subscription Rate (Volume 19, 2 issues, 2004)
Academic libraries
Print edition and site-wide online access: US$98/f67
Print edition only: US$9 3/t64
Site-wide online access only: US$88/t61
Corporate
Print edition and site-wide online access: US$121/t81
Print edition only: US$115 /£277
Site-wide online access only: US$10911f73
Personal
Print edition and individual online access: tJS$36/t28
Please note: £ Sterling rates apply in Europe, US$ elsewhere.
Readers with mailing addresses in non-OECD countries and in socialist economies in transition are eligible to receive complimen-
tary subscriptions on request by writing to the UK address below.
There may be other subscription rates available: for a complete listing, please visit www.wbro.oupjournals.org/
subscriptions.
Full prepayment in the correct currency is required for all orders. Orders are regarded as firm, and payments are not refund-
able. Subscriptions are accepted and entered on a complete volume basis. Claims cannot be considered more than four
months after publication or date of order, whichever is later. All subscriptions in Canada are subject to GST. Subscriptions in
the EU may be subject to European VAT. If registered, please supply details to avoid unnecessary charges. For subscriptions
that include online versions, a proportion of the subscription price may be subject to UK VAT. Personal rates are applicable
only when a subscription is for individual use and are not available if delivery is made to a corporate address.
Back issues: The current year and two previous years' issues are available from Oxford University Press. Previous
volumes can be obtained from the Periodicals Service Company,  1 Main Street. Germantown, NY 12526, USA. E-mail:
pscQwperiodicals.com. Tel: (51 8) 537-4700. Fax: ( 518) 537-5899.
Contact information: Journals Customer Service Department, Oxford tUniversity Press, Great Clarendon Street.
Oxford OX2 6DP, UK. E-mail: jnls.cust.serv(4ioupjournals.org. Tel: +44 (0)1865 353907. Fax: +44 (0)1865 353485.
In the Americas, please contact: Journals Customer Service Department, Oxford University Press, 2001 Evans Road,
Cary, NC 27513, USA. E-mail: jnlorders4oupjournals.org. Tel: (800) 852-7323 (toll-free in USA/Canada) or (919)
677-0977. Fax: (919) 677-1714. In Japan, please contact: Journals Customer Service Department, Oxford University
Press, 1-1-1 7-SF, Mukogaoka, Bunkyo-ku, Tokyo, 113-0023, Japan. E-mail: okudaoup@po.iijnet.or.jp. Tel: (0)3) 3813
1461. Fax: (03) 3818 1522.
Postal information: The World Batik Research Observer (ISSN 0257-3032) is published by Oxford University Press for
the International Bank for Reconstruction and Development/olE WORLD) BANK. Send address changes to The World Bank
Research Observer. Journals Customer Service Department, Oxford University Press, 2001 Evans Road, Cary, NC 27513-
20)09. Communications regarding original articles and editorial management should be addressed to The Editor, The
W4'orld Bank Research Observer. The World Bank, 3818 H Street. NW. Washington. D.C. 2)433, USA.
Permissions: For information on how to request permissions to reproduce articles or information from this journal.
please visit www.oupjournals.org/permissions.
Advertising: Inquiries about advertising should be sent to Helen Pearson, Oxford Journals Advertising, PO Box 347,
AbingdonOXl4 1GJ. UK. E-mail: helen@oxfordads.com. Tel/Fax: +44 0))1235 201904.
Disclaimer: Statements of fact and opinion in the articles in The World Bank Research Observer are those of the respective
authors and contributors and not of the International Bank for Reconstruction and Development/niv WORLD BANK or
Oxford University Press. Neither Oxford University Press nor the International Bank for Reconstruction and Develop-
ment/nwlfh WORDI) BANK make any representation, express or implied. in respect of the accuracy of the material in this jour-
nal and cannot accept any legal responsibility or liability for any errors or omissions that may be made. The reader
should make her or his own evaluation as to the appropriateness or otherwise of any experimental technique described.
Paper used: The World Batik Research Observer is printed on acid-free paper that meets the minimum requirements of
ANSI Standard Z 39.48-1984 (Permanence of Paper).
Indexing and abstracting: The World Bank Researchi Observer is indexed and/or abstracted by ABI/INFORI, CAB Abstracts,
Current Contents/Social and Behavioral Sciences, Journal of Economiic Literatiare/EconLit, PAIS International, RePEc (Research
in Economic Papers), Social Services Citation Index, and Wilson Business Abstracts.
Copyright ©D The International Bank for Reconstruction and Development/TnE WORLD BANK 2004
Al rights reserved: no part of this publication may be reproduced, stored in a retrieval system, or transmitted in any fonn or by
any means, electronic, mechanical, photocopying. recording, or otherwise without prior written permission of the publisher
or a license permitting restricted copying issued in the LJK by the Copyright Licensing Agency Ltd, 90 Tottenham Court Road,
London WIP 9HE, or in the USA by the Copyright Clearance Center,222 Rosewood Drive, Danvers, MA ()1923.



How Have the WorLd's Poorest Fared
since the EarLy 1980s?
Shaohua Chen * Martin Ravallion
A newv assessment is mnade of the developing world's progress against poverty. By thefriugal
SI a day standard there were 1.1 billion poor people in 200 1-almrrost 400 mnillion fewer
thaii 20 years earlier. During that period the numnber oJ'poor people declined by more than
400 millioni in China, though halJ the decline was in the early 1 980s and the nutmber out-
side Chinla rose slightly. At the samne timne the numrrlber of people in the world living on less
than $2 a day rose, so that there has been a mlarked bunchiing up of people livinig between
$1 andd$2 a day. Slub-Saharan Africa has becomne the region wvithi the highest incidence of
extremrle poverty and the greatest depth of poverty. If these trends continlue, the 1 990
aggregate $1 a day poverty rate will be halved by 2015, rneeting the Millenniumirl Develop-
nment Goal, though only East and South Asia will reach this goal.
A cloud of doubt hangs over our knowledge about the extent of the world's progress
against poverty. A widely cited World Bank (2002) estimate is that there were 200
million fewer poor people in the world in 1998 than in 1980. This figure has been
contested and for good reasons. Deaton (2002b) contrasts this seemingly optimistic
assessment with the estimate in World Developmient Report 2000/2001: Attacking
Poverty (World Bank 2000), which appeared to show little or no progress. Deaton
argues that the claim in World Bank (2002) was based on methodologically incon-
sistent estimates from two studies-Bourguignon and Morisson (2002) (up to 1992)
and Chen and Ravallion (2000) (beyond that).' With reference to the relevant chart
in World Bank (2002) Deaton writes:
The historical data in this chart were assembled by Francois Bourguignon
and Christian Morrisson.... They derive their estimates by applying (some-
times sketchy and outdated) distributional information to the consumption
figures from national accounts data, a technique that is almost certainly the
The World Bank Research Observer, vol. 19, no. 2,
Cc The International Bank for Reconstruction and Development / THE WORLD BANK 2004: all rights reserved.
doi:10.109 3/wbro/lkhO2()                                           19:141-169



only methodology that would allow the construction of data for a century
and a half.... After 1993, when the Bourguignon and Morrisson data end,
Globalization [World Bank 2002] uses the poverty estimates that were
assembled by Shaohua Chen and Martin Ravallion.... But Chen and Raval-
lion's data from 1987 to 1993, which is when poverty increased, are
dropped from the chart. In consequence. and without any new informa-
tion. we go from an assessment that the number of poor people in the world
was showing little or no decline from 1987 to 1998 in Attacking Poverty to
an assessment, in Globalizationz, of a continuous and accelerating decline
from 1980 to 1998. (see www.imf.org/external/pubs/ft/fandd/2002/06/
deaton.htm)
These concerns are too important to ignore. The splicing of these different data
sources is questionable. The only solution is to construct a new, internally consis-
tent series over the 1980s and 1990s.
This article offers a new assessment of progress in reducing poverty over 1981-
2001 using more consistent data and methods-closely following the methods
underlying the Attacking Poverty (World Bank 2000) numbers, which had been
based on Chen and Ravallion (2000). In common with our past estimates, we draw
on nationally representative surveys as much as possible. The article reviews our
methods of measuring poverty from those surveys and notes any changes from past
estimates, though we refer readers to other sources for further discussion of our
methods and alternatives. 2 The new estimates presented here supersede all our pre-
vious estimates in that we recalculate everything back in time on a consistent basis
incorporating the new data. 3
We summarize our results in a standard regional tabulation following previous
work. However, we have also created a Web-based interactive tool, PovcalNet, that
allows users to access the primary distributions and so estimate poverty measures
for alternative country groupings or for a selected set of individual countries (http://
iresearch.worldbank.org/povcalnet).
A notable feature of these new estimates is that they go back to the early 1980s,
allowing an assessment of the validity of the poverty reduction claim in World Bank
(2002). We have previously resisted going back this far because of concerns about
the coverage and quality of the survey data available for the early 1980Os. Our efforts
to expand coverage have helped allay our fears about reliability of the data from this
time period. However, it is clear that our estimates for the first year in our series,
1981, are not as reliable as those in the rest of the series.
The new estimates suggest that the World Bank (2002) figure of 200 million
fewer poor people is probably an underestimate. Indeed, our best estimates suggest
that the figure is almost twice that number. That is good news. However, closer
inspection of the data leaves little room for complacency about the world's progress
142                                The WVorld Bnink Research Observer, vol. 1 9. no. 2 (Fall 2004)



against poverty. Indeed, the picture that emerges is one of highly uneven progress,
with serious setbacks in some regions and time periods. The number of people living
on less than $2 a day has risen.
It should not be forgotten that there are limitations to our measures. There are
continuing concerns about aspects of the underlying data, including the purchasing
power parity (Ppp) exchange rates, the accuracy and comparability of the surveys
used, and intrinsic limitations of the welfare measures based on those surveys.
Potentially important examples of the limitations of the welfare measures is that our
definition of poverty does not directly reflect inequality within households or access
to public goods.
The article first describes the coverage of the survey data. It then discusses the
poverty line and exchange rates, followed by the measures of poverty, and then pre-
sents the main results.
Coverage of the Household Surveys
This is our first attempt to estimate global poverty measures back to the early 1980s.
Our previous estimates started in 1987 (Chen and Ravallion 2000, 2001). In retro-
spect, starting the series in 1987 was an unfortunate choice, because the late 1980s
and early 1990s were a difficult time for the world's poor, given sharply lower
growth in both China and India. Going back further in time should give a clearer
idea of the long-term trend.
We draw on 454 surveys covering 97 countries representing 93 percent of the
population of all low- and middle-income countries (Part 2 member countries of the
World Bank). Taking the most recent survey for each country, about 1.1 million
households were interviewed. The surveys were mostly done by government statis-
tics offices as part of their routine operations.
The poverty measures are estimated from the survey data. No secondary sources
are used for measuring poverty at each survey round (unlike all other compilations
of distributional data and global poverty measures that we know of), although other
data sources are used for interpolation, given that the surveys of different countries
do not coincide in time. Households are ranked by consumption or income per per-
son. The distributions are weighted by household size and sample expansion factors.
so that a given fractile (such as the poorest decile) has the same share of the country-
specific population across the sample. Thus the poverty counts give the number of
people living in households with per capita consumption or income below the pov-
erty line. The data come in various forms, ranging from micro data to specially
designed grouped tabulations from the raw data constructed following our guide-
lines. Datt and Ravallion (1992) and Chen and others (1994) describe our estima-
tion methods for grouped data.
Sliao/1ua Chten uid Mlartiii Rav'alliorn                             143



As in previous work we try to eliminate obvious comparability problems. either by
reestimating the consumption or income aggregates or by dropping a survey. How-
ever, there are problems that we cannot deal with. Differences in survey methods
(such as in questionnaire design) can create nonnegligible differences in the estimates
for consumption or income. For example, although one-week recall for food consump-
tion is common in surveys, some countries use a longer period, which is likely to give a
lower estimate of consumption and hence higher measured poverty. An unusual
case is China, in which households are surveyed frequently throughout the year,
allowing an estimate of annual income, whereas other countries typically use recall
data for a month or less, obtained from one or just a few interviews; there is evidence
that China's practice yields lower inequality measures (Gibson and others 2001).
A specific data problem that has received attention in the recent literature con-
cerns the 55th round of India's National Sample Survey (NSS) for 1999/2000, which
has created a potentially serious comparability problem with previousNss rounds
(see Datt and Ravallion 2002; Deaton 2002a, 2003). For greater comparability
with previous NSS rounds, we use Deaton's (2003) adjusted distributions. The official
distributions from the 55th round give a lower poverty rate in 1999/2000 (32.3
percent below $1 a day compared with 34.8 percent using Deaton's corrections).
(The distributions are, of course, the same in the previous large sample survey, for
which we obtain a $ 1 a day poverty rate of 41.9 percent.) However, Deaton's correc-
tion requires an unchanging probability of being poor conditional on consumption
of the goods that appear to have been unaffected by the change in survey design.
Changes in relative prices can cast doubt on this assumption (Datt and Ravallion
2002: Sen and Himanshu 2003).
Possibly, Deaton's method overestimates the decline in poverty in India between
the 55th and previous rounds. An alternative approach to comparing the surveys
for 1999/2000 with the previous large sample survey of 1993 has been proposed by
Sundaram and Tendulkar (2003). It entails comparing estimates over time based on
a mixed recall period instead of the uniform recall period used by the official data and
by Deaton (2003). The comparison is possible only between these two surveys. With
the Sundaram and Tendulkar distributions, the $1 a day poverty rate for India falls
from 38.7 percent in 1993 to 32.3 percent in 1999/2000-a 6.4-percentage-point
drop rather than our 7.1-percentage-point drop estimate using the Deaton-adjusted
distributions.
Appendix table A.1 lists the surveys used, their dates, and whether consumption
or income data are used. Population coverage varies greatly by region, ranging from
74 percent of the population of the Middle East and North Africa to 98 percent of the
population of South Asia. Not all available surveys are included. Surveys are
excluded if essential data are missing (such as for the Ppp exchange rates or con-
sumer price indices used to update poverty lines over time) or if there are serious
comparability problems with the rest of the data set.
144                               The Vlowld Bank Research Observer, vol. 19, 1o. 2 (Fall 2004)



Naturally, the further back one goes, the fewer the number of surveys. Coverage
deteriorates in the last year or two of the series, given the lags in survey process-
ing. A simple but useful guide to the reliability of our estimates is to count the
number of surveys by year and to compare the number with the three-year mov-
ing total centered on each year-given that having a survey last year or next year
can help greatly in estimating poverty this year (figure 1). By this measure, our
estimates are the most reliable for the mid- to late 1990s and the least reliable for
1981. We have only 15 surveys up to 1983, though the number rises sharply to
32 surveys for the period up to 1985. By contrast, we have 86 surveys during
1986-90.
Most regions are still quite well covered, from at least the second half of the
1980s (East and South Asia being well covered from 1981 onward). Two excep-
tions stand out. Unsurprisingly, country coverage in Eastern Europe and Central
Asia is weak for the 1980s, when most of these countries did not officially exist.
More worrying is the lack of coverage for Sub-Saharan Africa in the 1980s. The
estimates for the early 1980s rely heavily on projections based on distributions for
the late 1980s, as a list of the average survey year by region for each reference year
Figure 1. Number of National Household Surveys by Year, 1979-2002
Number of surveys
140(
120-
100-
80-
Three-year movinig total
60-
40-
20-
Number by year
1980      1985     1990      1995     2000
Source: National household surveys.
Shaohzia Clhen and Martin Ravallion                                   145



shows (table 1). Although survey coverage for Africa has improved considerably
over our previous estimates (Chen and Ravallion 2001), the weakness of the cover-
age of Africa should be kept in mind when interpreting our results.
Even in regions with seemingly good survey coverage in the 1980s, there are
questions about some of the data. The other side of the coin of improved house-
hold survey data for developing countries in the 1990s is the much weaker
survey data in the 1980s. Consider China, the country with the largest popu-
lation. The National Bureau of Statistics (NBS) has provided us with income distri-
butions for China from surveys done in 1980 (rural areas) and 1981 (urban).
However, the NBS was only beginning to resume doing national household sur-
veys at that time (it had ceased doing surveys during the Cultural Revolution,
1966-76) and it can be conjectured that these early efforts were not as good
(Chen and Ravallion 1996). The early NBS surveys (prior to 1985) did not include
30 percent of provinces. However, this does not appear to be a source of bias
because the provincial coverage was uncorrelated with poverty; we could con-
vincingly reject the null hypothesis that the first available estimates of all our
poverty measures were the same for provinces omitted as for those included
(Ravallion and Chen 2004). The sample sizes for these early surveys are also
smaller than the other NBS surveys for China that we use, though the sample sizes
are still adequate. For the 1980 survey in rural areas, 16,000 randomly sampled
households were interviewed, and for the 1981 urban sample, about 9,000. By
contrast, from 1985 onward the sample sizes were about 70,000 in rural areas
and 30,000 in urban areas.
Table 1. Average Date of the Surveys Used for Each Reference Year, 1981-2001
Region            1981    7984     1987    1990    1993    1996     1999    2001
EastAsia         1981.5   1984.6  1987.4  1990.4  1992.5  1996.1   1999.1  2000.4
Eastern Europe   1988.2   1988.1  1988.3  1989.6  1993.6  1995.7   1998.9  1999.7
and Central Asia
Latin America and  1983.2  1984.7  1986.6  1990.4  1992.5  1996.1  1998.1  2000.1
the Caribbean
Middle East      1987.1   1987.8  1988.1  1990.0  1993.0  1996.0   1998.0  1998.2
and North Africa
SouthAsia        1980.9   1983.9  1987.2  1989.7  1993.0  1995.9   1999.0  1999.4
Sub-SaharanAfrica  1988.6  1988.3  1989.6  1990.2  1993.0  1995.1  1996.6  1997.0
Total            1982.8  1985.3   1987.7  1990.2  1992.8  1995.9   1998.5  1999.4
Note: Population-weighted mean for all the surveys that were used to estimate the poverty measures for each
reference year.
Soi4rce: National household surveys.
146                                     The World Bank Research Observer, vol. 19. no. 2 (Fall 2004)



Exchange Rates and Poverty Lines
We use the same Ppp estimates for consumption as Chen and Ravallion (2001),
which were produced by the World Bank's Data Group (data sources and methods
are described in Ahmad 2003). For 69 of the sample 97 countries, the PPPs are based
on price and consumption basket data collected by the 1993 International Compari-
son Project (icP). For almost all countries that did not participate in the 1993 icP, the
bank's PPPs are based on interpolations from the cross-country regressions described
in Ahmad (2003). Two exceptions are China and India, for which the bank's P'pis are
based on other sources-India's on an update of the country's 1985 Ppp and China's
on a credible independent study of prices levels in 10 cities (Ahmad 2003). As in
Chen and Ravallion (2000), the Penn World Tables are used in preference to the
World Bank's PPPs for five countries (Ghana, Mauritania, Nicaragua, the Philippines.
and Uganda) for which the bank's PPPs give poverty rates that are implausibly low,
whereas the PPPs for 1993 from the Penn World Tables (version 5.6) give more
believable estimates.4
The international poverty line in our work prior to Chen and Ravallion (2001)
was set at $1 a day at 1985 Ppp (more precisely, at $31 a month or $1.02 a day: see
Ravallion and others 1991). The original $ 1 a day poverty line was chosen as being
representative of the poverty lines typical of low-income countries (Ravallion and
others 1991). The same principle was applied by Chen and Ravallion (2001) in
updating the poverty line using the new PPPs for 1993.
But the 1985 PPPs based on Penn World Tables are not comparable with the World
Bank's PPPs at base 1993, either in primary data or method used. So one cannot
update the poverty line simply by adjusting for inflation in the UJnited States between
1985 and 1993. Indeed, that gives a poverty line that is well above those typical of
low-income countries (Chen and Ravallion 2001).5 To be consistent with the original
aim of using a poverty line that is representative of the lines actually found in poor
countries, we recalculate the dollar value of the original set of poverty lines using the
new PPPs and compare this with mean consumption, also calculated by the new PPPs.
Following Chen and Ravallion (2001), the resulting poverty line is $1.08 a day
($32.74 per month) in 1993 Ppp prices. This is the median of the 10 lowest poverty
lines in the set of countries used by Ravallion and others (1991). This is the main pov-
erty line focused on here and is referred to as the "$1 a day" line or "extreme poverty."
However, the poverty rate on this basis must be deemed a conservative estimate,
because aggregate poverty in the developing world is defined by perceptions of pov-
erty rates in the poorest countries. (Not a new observation, this was argued explicitly
in World Bank 1990 and Ravallion and others 1991.) Thus two broader definitions
are also considered. In one the poor include all those who would be judged poor by
standards more typical of middle-income countries. For this purpose the poverty line
is set at twice the $1 a day line.
Slitolu Ci 'n azizd Alartiun Rav'allioiI                           147



The second definition allows for "relative poverty." Chen and Ravallion (2001)
proposed an operational approach for measuring relative poverty, building on
Atkinson and Bourguignon (1999). By this measure of relative poverty people are
deemed poor if they do not attain either the $1 a day consumption level (loosely
interpretable as "physical needs") or a given proportion of mean consumption ("social
needs"). The constant of proportionality was set at one-third, which gave the best fit
to the data on poverty lines for developed and developing economies used in setting
the $1.08 poverty line (Chen and Ravallion 2001). We fix the real value of the rela-
tive poverty line over time for each country. So these poverty lines are relative
between countries but absolute over time. (Chen and Ravallion 2001 discuss this
choice further.) Making the poverty lines relative over time would mean that for
countries with mean consumption of greater than $3.23 a day the poverty mea-
sures will be independent of the absolute levels of consumption (and depend solely
on the percentile of the population for which the Lorenz curve has a slope of 1/3).
Measuring Poverty from the Surveys
Three poverty measures are computed. The first measure is the headcount index,
given by the percentage of the population living in households with consumption or
income per person below the poverty line. This is the easiest measure to interpret,
but it has the well-known deficiency that it says nothing about differences in the
depth of poverty below the line. A second measure, the number of poor people, is
obtained by applying the estimated headcount index to the population of each
region (under the assumption that countries without surveys have the same head-
count index on average as those with surveys). A third measure, the poverty gap
index, gives mean distance below the poverty line as a proportion of the poverty line
(where the mean is taken over the whole population, counting the nonpoor as hav-
ing zero poverty gap). PovcalNet also gives estimates of the squared poverty gap, in
which individual poverty gaps are weighted by the gaps themselves, to reflect
inequality among the poor (Foster and others 1984).
In keeping with our previous work, we aim to measure poverty in terms of house-
hold consumption expenditure per capita. Of the 454 surveys used, 247 allow us to
estimate the distribution of consumption expenditures; this is true of all the surveys
used in the Middle East and North Africa, South Asia, and Sub-Saharan Africa.
Whenever there is a choice consumption data are preferred to income data, because
consumption is likely to be the better measure of current welfare. About one-quarter
of the cases without consumption distributions do have survey-based estimates of
mean consumption. For those cases the income mean is replaced by the consump-
tion mean, leaving the Lorenz curve the same (all incomes are scaled up by the ratio
of the consumption mean to the income mean). There is, however, no obvious basis
148                               The W7orld Banlk Research Observer, vol. 19, no. 2 (Fall 2004)



for adjusting the Lorenz curve; one expects higher inequality in an income distribu-
tion than in a consumption distribution for the same place and data.
Our previous estimates for China (Chen and Ravallion 2001) relied on income
Lorenz curves, but we used the survey means for household consumption expendi-
ture per capita supplied by the NBS. For this update we obtained complete consump-
tion distributions from the NBS back to 1990. To maintain consistency with our
methods for other countries, we switch to consumption (for both the distribution
and the mean) from 1990 onward, though we have no choice but to keep our old
method for the 1980s. This raises a concern about comparability between our esti-
mates for China for the 1990s and for the 1980s. To assess whether this is a prob-
lem, we also calculate our estimates for the 1990s using the old method. The two
sets of estimates for the 1990s match up quite closely (table 2), so the comparisons
over time do not appear to be of concern. Ravallion and Chen (2004) discuss this
and other issues concerning China's poverty and inequality data in greater detail.
One important difference with Chen and Ravallion (2001) is that when only an
income distribution is available, we do not follow our past practice of rescaling mean
income by one minus the national savings rate. This practice was questioned by
Deaton (2002a), and in Chen and Ravallion (2001) we noted the implications of
Table 2. Poverty Headcount Estimates for China by Old and New Methods, 1990-2001
Headcount index (%)
Old rrrethod: incomne distribution  New rnethod: consuniption distribution
Year  Poverty line ( I 993 PPPS) tvitlm adjustmzenit to the meanl (Y)  with consumption mean (')
1981         1.08                 63.76
2.15                 88.1 2
1984         1.08                 40.99
2.15                 78.49
1987         1.08                 28.45
2.15                 67.41
1990         1.08                 31.53                     33.01
2.15                 69.93                     72.64
1993         1.08                 29.46                     28.36
2.15                 64.59                     68.13
1996         1.08                 16.91                     17.38
2.15                 50.63                     53.36
1999         1.08                 16.42                     1/.X7
2.15                 47.19                     5(.05
2001         1.08                 16.51                     16.64
2.15                 44.45                     46.67
Source: Authors ealculations based on China's national household surveys.
Shaohuna Clze arnd Miartin Ravallion                                          149



dropping this rescaling. Since then we have assembled surveys for 27 countries for
which we have both consumption and income distributions, to test this assumption
in our past work by calculating the poverty measures using both consumption and
income for the same country. There is only a small and statistically insignificant dif-
ference between the two sets of estimates: Consumption has a lower mean but also
lower inequality, with the effect that poverty measures are quite close. For the $1 a
day line the mean headcount index is 17.8 percent for consumption and 21.2 per-
cent for income; the difference is not statistically significant (t = 0.7 3; n = 2 7). For the
$2 a day line the mean headcount index is slightly higher for consumption (48.2
percent) than for income (44.8 percent) but, again the difference is not statistically
significant (t=0.49). So we abandon our past practice of rescaling the mean
for income surveys. The main implication is that our poverty measures for Latin
America (where income surveys are more common than elsewhere) drop a few
percentage points.
Having converted the international poverty line at Ppp to local currency in 1993.
we convert it to the prices prevailing at each survey date using the country-specific
official consumer price index (CpI).6 The weights in this index may or may not accord
well with consumer budget shares at the poverty line. In periods of relative price
shifts, this will bias our comparisons of the incidence of poverty over time, depending
on the extent of utility-compensated substitution possibilities for people at the pov-
erty line.
To estimate regional poverty at a given reference year (say, 1998) we "line up"
the surveys in time using the same method as in our past work. We start the series in
1981 and make estimates at three yearly intervals, except for 2002, when there
were too few surveys, so we estimate for 2001 instead. We thus make estimates for
1981, 1984, 1987, 1990, 1993.1996,1999, and 2001. These are calledreference
years, as distinct from survey years, which are spread over the interval 1979-2002
(see figure 1).
For the 97 countries, 9 have only one survey, 19 have two surveys, and 69
have three or more surveys over the period. If there is only one survey for a coun-
try, measures for each reference year are estimated by applying the growth rate
in real private consumption per person from the national accounts to the survey
mean-assuming that the Lorenz curve for that country does not change.7 This
seems the best option for dealing with this problem, though there can be no guar-
antee that the Lorenz curve would not have shifted or that a survey-based meas-
ure of consumption would have grown at the same rate as private consumption
in the national accounts. For example, growth of private consumption in the
national accounts might reflect growth in spending by nonprofit organizations-
which are not separated from households in the national accounts for most
developing economies-rather than growth in household spending (Ravallion
2003b).
150                               Tlhe lVorld Boauk Research Observer. vol. 1 9, tio. 2  f(all 2004)



When the reference date (say, 1993) is between two surveys (say, 1989 and
1995), one option is simply to interpolate between the two surveys. However, this
could be problematic when there is a long time period between surveys, and it
ignores the extra information available from the national accounts data. To bring
the national accounts information into the picture, mean consumption is first esti-
mated at the reference year using the national accounts growth rate between the
survey year and the reference year. For the example here there are two means at the
reference year based on two surveys, M93(89) and M93(95), where M93(t) is the
estimated mean for 1993 using the survey for year t. lJsing the 1989 distribution
and M93(89) yields headcount index H93(89). Similarly, using the 1995 distribu-
tion and the 1993 mean yields H93(95). Then the poverty headcount for 1993 is
estimated as the weighted average of H9 3(89) and H9 3(95) .8
Results
Estimates of the headcount indices for $ 1.0(8 at 1993 Ppp for 1981-2001 show that
the percentage of the population of the developing world living below $1 a day was
almost halved over 1981-2001, falling from 40 percent to 21 percent (table 3).
Expressed as a proportion of world population, the decline is from 33 percent to 18
percent. This assumes that nobody in developed areas lives below $1 a day.) The
number of poor people fell by 390 million, from 1.5 billion in 1981 to 1.1 billion in
2001 (table 4).
There was clearly more progress in some periods than in others. As already noted,
the late 198()s and early 1990s were a difficult period for the world's poor, with low
growth in both China and India. Once growth was restored, the rate of poverty
reduction by the $ 1 a day standard in the 1990s had returned to its long-term trend.
The percentage below $1 a day fell from 28 percent to 21 percent over 199(0-200(1,
about the same trend decline (in percentage points per year) as for 1984-2001 as a
whole. The number of poor people fell by about 130 million in 1990s. The poverty
measures for $2 a day follow a broadly similar pattern, though with a less dramatic
decline in the early 1980s and even stronger signs of stagnation in the period
around 1990 (see tables 3 and 4).
The new estimates suggest less progress in getting over the $2 a day line. By this
higher standard the poverty rate has fallen from 67 percent in 1981 to 53 percent in
2001 (see table 3). This has not been sufficient to prevent a rise in the number of
people living below $2 a day, from 2.4 billion to 2.7 billion (see table 4). Thus the
number of people living between $1 and $2 has risen sharply over these two
decades, from about 1 billion to 1.6 billion. This marked bunching up of people just
above the $1 line suggests that a great many people in the world remain vulnerable
to aggregate economic slowdowns.
NSIholU a (Chen aidi  (lMatirn Ravallioni                          151



Table 3. Headcount Indices of Poverty by Region for Two International Poverty Lines,
1981-2001 ('%)
Poverty liie and regiorn    1981   1984   1987   7990   1993    1996   1999   2001
$1.08 aday (1993 pi'p)
EastAsia                    57.7   38.9   28.0   29.6   24.9    16.6   15.7   14.9
China                     63.8   41.0   28.5    33.0   28.4   17.4   17.8   16.6
EastAsiaexcludingChina    42.0   33.5   27.0   21.1    16.7   14.7   11.0   10.8
EasternEuropeandCentralAsia  0.7    0.5    0.4    0.5     3.7    4.3    6.3    3.6
Latin America and Caribbean  9.7   11.8   10.9   11.3    11.3   10.7   10.5    9.5
Middle East andNorthAfrica   5.1    3.8    3.2    2.3    1.6     2.0    2.6    2.4
SouthAsia                   51.5   46.8   45.0   41.3   4(0.1   36.6   32.2   31.3
India                     54.4   49.8   46.3   42.1    42.3   42.2   35.3   34.7
SouthAsiaexcludinglndia   42.2   37.0   41.0    38.7   33.1   19.7   22.9   21.0
Sub-Saharan Africa          41.6   46.3   46.8   44.6   44.1    45.6   45.7   46.4
Total                       4(0.4  32.8   28.4   27.9   26.3    22.8   21.8   21.1
Total excluding China       31.7   29.8   28.4   26.1   25.6    24.6   23.1   22.5
$2.15 aday(1993 ppp)
EastAsia                    84.8   76.6   67.7   69.9   64.8    53.3   50.3   47.4
China                       88.1   78.5   67.4   72.6   68.1    53.4   50.1   46.7
EastAsiaexcludingChina      76.2   72.0   68.4   63.2    56.7   53.2   50.8   49.2
Eastern Europe and CentralAsia  4.7  4.1   3.2    4.9    17.3   20.7   2 3.8  19.7
Latin America and Caribbean  26.9  30.4   27.8   28.4   29.5    24.1   25.1   24.5
MiddleEastandNorthAfrica    28.9   25.2   24.2   21.4   20.2    22.3   24.3   23.2
SouthAsia                   89.1   87.2   86.7   85.5   84.5    81.7   78.1   77.2
India                       89.6   88.2   87.3   86.1   85.7    85.2   80.6   79.9
South Asia excludinglndia   87. 3  84.0   85.0   8 3.5  81.0    71.3   70.5   69.0
Sub-SaharanAfrica           73.3   76.1   76.1   75.0   74.6    75.1   76.0   76.6
Total                       66.7   63.7   60.1   60.8   60.2    55.5   54.4   52.9
Total excluding China       58.8   58.4   57.5   56.6    57.4   56.3   55.8   54.9
Source: Authors' calculations based on national household survey data.
Regional Differences
Performance in poverty reduction has been far from uniform across regions. Indeed,
there have been notable changes in regional poverty rankings over this period. In
1981 East Asia had the highest incidence of extreme poverty in the world, with 58
percent of the population living below $1 a day. Next was South Asia, followed by
Sub-Saharan Africa, Latin America, Middle East and North Africa, and Eastern
Europe and Central Asia. Twenty years later, Sub-Saharan Africa had swapped
places with East Asia, where the headcount index had fallen to 1 5 percent. South
Asia remained in second place, but Eastern Europe and Central Asia had overtaken
the Middle East and North Africa. The ordering of regions is not, however, robust to
the choice of poverty line. At the $2 a day poverty line South Asia edges out East
152                                     Tlhel World Bank Researchl Observer. vol. 19. no. 2 (Fall 2004)



Table 4. Number of Poor People by Region for Two International Poverty Lines. 1981-2001
(millions)
Poverty line and region  1981  1984  1987   1990    1993     1996    1999     2001
$1.08 a day
(1993 ppp)
EastAsia           795.6   562.2    425.6   472.2    415.4   286.7    281.7   271.3
China              633.7   425.0    308.4   374.8    334.2   211.6    222.8   211.6
Eastern Europe       3.1     2.4      1.7     2.3     17.5    2(.1     30.1    17.0
and Central Asia
Latin America       35.6    46.0     45.1    49.3     52.0    52.2     53.6    49.8
and Caribbean
MiddleEast           9.1      7.6     6.9      5.5     4.0     5.5      7.7     7.1
and North Africa
SouthAsia          474.8   460.3    473.3   462.3    476.2   461.3   428.5    431.1
India              382.4   373.5    369.8   357.4    380.0   399.5    352.4   358.6
Sub-SaharanAfrica  163.6   198.3    218.6   226.8    242.3   271.4    294.3   312.7
Total            1,481.8  1.276.8  1,171.2  1,218.5  1,207.5  1,097.2  1,095.7  1,089.0
$2.15 aday
(1993 ppp)
EastAsia          1169.8  1108.6   1028.3  1116.3   1079.3   922.2    899.6   864.3
China              875.8   813.8    730.8   824.6    802.9   649.6    627.5   593.6
Eastern Europe      20.2    18.3     14.7    22.9     81.3    97.8    11 3.)   93.3
and Central Asia
Latin America       98.9   118.9    115.4   124.6    1 36.1  117.2    127.4   128.2
and Caribbean
MiddleEast          51.9    49.8     52.5    5(0.9    51.8    60.9     7(0.4   69.8
and North Africa
SouthAsia          821.1   858.6    911.4   957.5   1004.8  1029.1  1039.0   1063.7
India              63(0.0  661.4    697.1   731.4    769.5   805.7    804.4   826.0
Sub-SaharanAfrica  287.9   326.0    355.2   381.6    410.4   446.8    489.3   516.0
Total            2,450.0  2,480.1  2,477.5  2,653.8  2,763.6  2,674.1  2,738.8  2,735.4
Source: Authors' calculations based on national household survey data.
Asia for the highest headcount index in 1981, and it edges out Africa for the highest
headcount index in 2001.
The dramatic progress in East Asia has meant that by 2001 the region had
already reached the Millennium Development Goal of halving the 1990 $1 a day
poverty rate by 2015. China's progress against absolute poverty was a key factor
(given the country's population weight), though the rest of East Asia had the same
proportionate decline in poverty over 1981-2001 as did China (see table 3). In
1981 China's incidence of poverty at the $1 a day measure was roughly twice that
for the rest of the developing world; by the mid-1990s China's poverty rate had
Shaohua Clien and Martin Ravallion                                              153



fallen well below the average. There were 400 million fewer people living under $1 a
day in China in 2001 than 20 years earlier, though a staggering half of this decline
was in the period 1981-84 (see table 4). This was enormous progress for China's
(and the world's) poor people. The most plausible explanation would appear to be
China's reforms, starting in the late 19 70s, which decollectivized agriculture and, in
the "household responsibility system," gave farmers considerably greater control
over their land and output choices (Ravallion and Chen 2004 discuss this and other
explanations for China's success against absolute poverty). This was a one-off
reform, however, suggesting that the sharp drop in global poverty by the $1 a day
standard in the early 1980s was also unusual. China experienced a further drop in
the poverty count of 120 million between 1993 and 1996, which is generally att-
ributed to the substantial but short-lived increases in 1994 and 1995 in the pro-
curement prices for food grains paid by the government, which greatly reduced the
burden on farmers of this form of taxation (World Bank 1997; Ravallion and Chen
2004).
The long-run trend decline in the global $1 a day poverty rate over 1981-2001 is
0.86 percentage point a year (table 5). For 1984-2001, which is more indicative of
the overall trend given the unusual large decline in extreme poverty between 1981
and 1984 resulting from China's agrarian transition, the trend is 0.66 percentage
point a year.9 Focusing on the 1990s could also be deceptive, because the early
1990s had relatively high poverty as a result of the stalled growth in China and
India.
For the developing world outside China, the headcount index for the $1 a day
poverty line fell from 32 percent to 23 percent over 1981-2001. This was not suffi-
cient to prevent an increase in the total number of poor people, which rose from 850
million to 880 million. The decline in the headcount index over time in the develop-
ing world excluding China was close to linear (figure 2), with a trend decline of 0.42
percentage point a year (with a standard error of 0.029).
The number of poor people has also fallen in South Asia, from 475 million in
1981 to about 430 million in 2001 (figure 3), and the poverty rate fell from 52 per-
cent to 31 percent. The South Asia series suggests a remarkably robust trend rate of
decline in the $1 a day headcount index of 1 percentage point a year (see table 5).
(For South Asia a linear trend clearly fits better than an exponential trend.) If main-
tained, this will be sufficient to reach the Millennium Development Goal for poverty
reduction. The critical value needed to reach the goal is -0.83 percentage point a
year, which is outside the 95 percent confidence interval (-0.87, -1.09) for the esti-
mate of South Asia's trend rate of poverty reduction.
The extent of bunching up that has occurred between the $1 and $2 a day pov-
erty lines is particularly striking in East and South Asia, where the total number is
1.2 billion, roughly equally split between the two regions. Although this points
again to the vulnerability of the poor, it also suggests that substantial further
154                               Tl,e tWlorld Bii,ik Researhli Observer, vol. 19, no. 2 (Fall 2004)



Table 5. Trend Rates of Change by Region in the Headcount Index for the $1 a Day Poverty
Line, 1981-2001 and 1984-2001 (percentage points per year)
Critical ratefor ialvingq the 1 990
Regioni                        1981-2001       1 984-2001            headcont initldex
East Asia                       -1.8 7*          -1. 36*                 -0.59
(0.32)          ((.19)
China                           -1.99*           -1.37*                  -0.66
(0.40)          (0.26)
Latin America and Caribbean     -0.) 3           -().I(                  -0.2 3
(0.)4)          (0.0(3)
South Asia                      -0.98*           -0.9 5*                 -0.8 3
((.05)          10.05)
India                           -(.91*           -0.8 3*                 -0.84
0. 0 ()         (0.12)
Sub-SaharanAfrica                0.12             0.00                   -(0.89
(0.(9)          (0.08)
Total                           -0.86*           -0.66*                  -0.56
((0.12)         (0.06)
*SigniGicant at the 1 percent level.
Note: The numbers in parentheses are standard errors. Eastern Europe and Central Asia and the Middle East and
North Africa regions are dropped from this table because there were so few people living below the $ 1 a day poverty
line in 1990). Trends were estimated by linear regression on time. All regressions were tested for first-order serial
correlation in the errors using the Lagrange multiplier tests. The null hypothesis of serial independence could not be
rejected in any case.
Source: Authors calculations based on national household survey data.
impacts on poverty can be expected from economic growth, provided that it does not
come with higher inequality.
There is less sign of progress against poverty outside Asia. The number of poor
people increased in Latin America, where the poverty rate has been roughly con-
stant over time (10 percent for $1 a day and 2 5 percent for $2 a day, which is closer
to the national poverty lines in that region). The Middle East and North Africa region
experienced a marked downward trend in the poverty rate during the 1980s, but
the rate stabilized in the 1990s at around 2 percent for the $1 a day poverty line and
at a little more than 20 percent for the $2 a day line.
Both the incidence of poverty and the number of poor people rose in Eastern
Europe and Central Asia in the 1990s compared with the 1980s. Although very few
people live below $1 a day in this region, the poverty rate by the $2 a day standard
rose from almost 2 percent in 1981 to 20 percent in 2001. However, the paucity of
survey data for this region in the 1980s should not be forgotten. Thus the estimates
are based heavily on interpolations, which do not allow for any changes in dis-
tribution. Distribution was probably better from the point of view of the poor in the
SlIUao)is(a t'117 he  id' Altini Racallion                                           155



Figure 2. Headcount Indices of Poverty Incidence in the Developing World for the $1 a Day
International Poverty Line, with and without China, 1981-2001
Percent
50-
40- Developing world total
30 -     <~
20_   excluding China
10-
0-                       I          I 
1980       1985        1990       1995       2000
Note: The actual $1 a day poverty line is $1.08 (199 3 ppp).
Source: Authors' calculations based on national household survey data.
1980s, in which case poverty would have been even lower than estimated here.
There are also some signs of recent progress for the poorest in the region, though it is
too early to say if this represents a change in trend.
The incidence of poverty in Sub-Saharan Africa has fluctuated around a mean of
45 percent for the $ 1 a day line (75 percent for the $ 2 a day line), with no significant
trend in either direction (see table 5). The number of poor people almost doubled
over 1981-2001, from 164 million to 316 million. By the $1 a day measure the
share of the world's poor living in Africa has risen from 11 percent in 1981 to 29
percent in 2001.
The critical rates of decline in poverty needed to achieve the Millennium Deve-
lopment Goal by 2015 are given in table 5. The actual trend rates of decline in
the aggregate $1 a day poverty rate will be sufficient to achieve the Millennium
Development Goal if progress is maintained until 2015. However, the variations
found over time point to a need for caution. For the full time period studied. the
critical trend needed to reach the Millennium Development Goal is just outside the
95 percent confidence interval for the estimated trend, so it can be claimed with
156                                   The 'World Baok Research Ohserver, vol. 19, no. 2 (Fall 2004)



Figure 3. Number of Poor People by Region for the S I a Day International Poverty Line,
1981-2001
Millions
600 -
400 -
1200 -        3
600~~~~~~~~~~~~~~~~~~~~~~~~~ E                    f ut imal *orld 
100
.~ ~ ~~~~~~jtv _i  E.                               -A -
400._
200                  -                                        Souwn 4\,.a
1981      1984       1987      1990      1993       1996      1999      2001
Note: The actual $1 a day poverty line is $1.08 (1993 ewr).
Source: Authors' calculations based on national household survey data.
95 percent confidence that the trend over 1981-2001 exceeds that needed to
halve the 1990 headcount index for $ 1 a day. However, 198 1-84 was an unusual
subperiod, as we have noted. For the period 1984-2001 the critical trend for the
Millennium Development Goal is within the 95 percent confidence interval. For
the series starting in 1984, it can only be said with about 90 percent confidence
that the aggregate trend exceeds the critical value needed to halve the 1990
poverty rate.
Poverty Gap Indices
The poverty gap index, PG, is related to the headcount index, H, as PG= (1 - MP)H,
where MP is the ratio of the mean income of the poor to the poverty line. Regional
rankings for the poverty gap index (table 6) are the same as those for the headcount
index (see table 3), and they follow the same change in patterns over time.
Shaohua Chei arnd Martin Rtavallio                                           157



Table 6. Poverty Gap Indices by Region for Two International Poverty Lines. 1981-2001 (%)
Poverty line and region  1981  1984   1987     1990    1993    1996    1999    2001
$1.08 aday(1993 PPP)
EastAsia             20.58    11.11    7.69    7.65    6.13     3.52    3.57    3.35
China                23.41    11.82    8.17    8.87    7.34     3.82    4.18    3.94
Eastern Europe        0.18     0.14    0.11     ).11   0.84     1.25    1.86    0.75
and Central Asia
Latin America         2.75     3.45    3.36    3.57    3.52    2.36     4.03    3.36
and Caribbean
MiddleEast             1.0()   0.76    0.61    0.49    0.27    0.39     0.53    0.45
and North Africa
SouthAsia            16.06    13.86   12.35   11.00   10.21    8.97     6.63    6.37
India                17.27    14.99   12.68   11.09   10.86   10.55     7.22    7.08
Sub-SaharanAfrica    17.03    19.65   20.10   19.07   19.24   19.80    20.10   20.53
Total                13.92    10.20    8.64    8.23    7.62    6.44     6.20    5.99
$2.15 a day (1993 ppp)
EastAsia             47.20    36.45   29.36   30.55   27.22   19.95    18.94   17.78
China                50.82    37.92   29.67   32.94   29.85   20.33    19.79   18.44
Eastern Europe         1.43    1.16    0.87    1.35    5.49     7.06    8.25    5.87
and Central Asia
Latin America        10.66    12.44   11.48   11.81   12.04    9.25    10.97   10.20
and Caribbean
MiddleEast            8.81     7.36    6.80    5.69    5.05     5.74    6.54    6.14
and North Africa
SouthAsia            45.78    43.02   41.86   39.92   38.84    36.52   32.98   32.35
India                47.22   44.68    42.52   4(0.43  40.10   39.93    34.89   34.43
Sub-SaharanAfrica    38.54   41.77    42.14   40.77   40.53   41.24    41.79   41.42
Total                35.02    30.79   27.86   27.8(0  26.82   23.76    23.05   22.20
Source: Authors' calculations based on national household survey data.
The most striking finding for the regional poverty gap indices is the depth of
poverty in Africa, which has a $1 a day poverty gap index of 20 percent com-
pared with 6 percent for the developing world as a whole. Furthermore, the
mean income of Africa's poor has been falling over time, dropping from $0.64
per person per day in 1981 to $0.60 in 2001 for the $1 a day poverty line,
though remaining roughly constant for the $2 a day line (table 7). In the rest of
the world poverty became shallower. The mean income of the poor in the devel-
oping world as a whole rose for the $1 a day poverty line from about $0.70 in
1981 to $0.77 in 2001 and even more markedly for the $2 a day line, from
$1.02 to $1.25.
The fact that the mean income of the poor is lowest in Africa implies that unless
inequality falls sufficiently, it will take more growth in Africa than in other regions
158                                      The World Bank Research Observer, vol. 19, no. 2 (Fall 2004)



Table 7. Mean Daily Income of the Poor by Region for Two International Poverty Lines,
1981-2001 (1993 ppp$)
Poverty line and region     1981   1984   1987    1990   1993   1996   1999   2001
$1.08 a day (1993 ppp)
EastAsia                    0.69   0.77   0.78    0.80   0.81   0.85   0.8X3  0.83
China                       0.68   0.77   0.77   0.79    0.80   0.84   0.82   0.82
EasternEuropeandCentralAsia  0.81  0.80   0.76    0.85   0.83   0.76   0.76   0.91
LatinAmerica and Caribbean  0.77   0.76   0.74   0.73    0.74   0.84   0.66   0.70
MiddleEastandNorthAfrica    0.86   0.86   0.87    0.85   0.89   0.87   0.86   0.87
South Asia                  0.74   0.76   0.78    0.79   0.80   0.81   0.85   0.86
India                       0.73   0.75   0.78   0.79    (.80   0.81   0.86   0.86
Sub-Saharan Africa          (.64   0.62   0.61    0.62   0.61   0.61   0.60   0.60
Total                       0.70   0.74   0.75    0.76   0.76   0.77   0.77   0.77
$2.15 a day (1993 ppp)
EastAsia                    0.95   1.13    1.22   1.21   1.25   1.35   1.34   1.35
China                       0.91   1.11   1.21    1.18   1.21   1.33   1.30   1.30
EasternEuropeandCentralAsia  1.5()  1.55   1.57   1.56   1.47   1.42   1.41   1.51
Latin America and Caribbean  1.30)  1.27  1.26    1.26   1.27   1.33   1.21   1.26
Middle East and North Africa  1.50  1.52  1.55    1.58   1.61   1.60   1.57   1.58
SouthAsia                   1.(5   1.(9   1.11    1.15   1.16   1.19   1.24   1.25
India                       1.02   1.06   1.10    1.14   1.15   1.14   1.22   1.23
Sub-SaharanAfrica           1.02   0.97   0.96    0.98   0.98   0.97   0.97   0.99
Total                       1.0)2  1.11   1.15   1.17    1.19   1.23   1.24   1.25
Source: Authors' calculations based on national household survey data.
to have the same proportionate impact on the poverty gap. This is borne out by cal-
culations of the elasticities of the poverty gap index to growth in the mean, holding
inequality constant (so that all levels of income grow at the same rate). The higher
the mean income of the poor, the higher is the absolute elasticity of the poverty gap
index to the overall mean. (It is readily verified that when all levels of income grow
at the same rate, the elasticity of the poverty gap index to the overall mean is -MP/
[I - MP].) Thus although this elasticity for the $1 a day poverty line in 2001 is -
1.3 for Sub-Saharan Africa, it is -1.8 for Latin America and the Caribbean, -3.3 for
East Asia (-3.2 for China), -3.7 for Eastern Europe and Central Asia, -3.9 in South
Asia, and -4.3 for the Middle East and North Africa (though this is deceptive,
because proportionately fewer people live below the $1 a day line; elasticity falls to
-2.8 for the $2 a day line). The elasticity is -2.5 for the developing world overall.
The elasticity has fallen (in absolute value) over time in Africa, though only slightly
(from -1.4 in 1981), while rising in the developing world as a whole (from -1.9 in
1981).
Shaolhua Clieni and Nfartini Ravallion                                         159



Relative Poverty
The absolute poverty measures examined so far aim to treat the same consump-
tion level the same way no matter what country a person lives in. To see how the
results might be affected by making an allowance for relative deprivation, pov-
erty is estimated based on the relative poverty lines described above (table 8 and
figure 4).
As expected, the incidence of relative poverty is noticeably higher for Eastern
Europe and Central Asia, Latin America, and the Middle East and North Africa. By
this measure Latin America overtakes South Asia in the early 1990s, making it the
second poorest region. And Eastern Europe and Central Asia overtakes East Asia by
Table 8. Relative Poverty Measures by Region. 1981-2001
Region             1981    1984     1987    1990    1993    1996    1999    2001
Headcount index OX,)
EastAsia            63.15   44.45   33.92   35.31    30.17   21.48   20.86   19.69
China               63.76   41.01   28.45   33.01   28.36    17.38   17.77   16.64
Eastern Europe       8.11    7.53    6.41    7.77   22.65    23.17   27.17   21.49
and Central Asia
Latin America       40.55   45.37   42.34   43.28   44.97    39.39   38.98   39.77
and Caribbean
Middle East         37.36   33.40   21.80   19.29    17.58   17.16   18.26   16.91
and North Africa
SouthAsia           58.17   50).65  47.72   41.45   40.33    36.87   32.09   31.41
India               62.55   54.51)  49.43   42.07   42.31    42.25   35.33   34.71)
Sub-SaharanAfrica   45.93   5(0.48  51.27   47.61   47.56    48.71   49.66   50.18
Total               5().1   42.0)   36.6    35.3     34.9    30.6    29.8    28.8
Number of poor
(millions)
EastAsia           871.3   642.9   515.2   563.7    502.6   371.4   373.1   358.8
China              633.7   425.2   308.4   374.8   334.2    211.6   222.8   211.6
EasternEurope       34.9    33.3    29.2    36.2   1(06.8   109.6   128.9   102.0
and Central Asia
LatinAmerica       149.1   177.6   175.6   189.8   207.8    191.3   198.1   208.3
and Caribbean
MiddleEast          67.1    66.1    47.3    45.8    45.0     46.8    52.8    50.8
and North Africa
SouthAsia          536.2   498.6   501.4   464.5   479.4    464.1   426.9   432.8
India              439.6   408.6   394.8   357.4   380.1)   399.5   352.4   358.6
Sub-SaharanAfrica  180.5   216.4   239.3   242.2   261.6    290.0   319.5   338.2
Total             1839.2  1634.9  15()8.0  1542.1  1603.2  1473.2  1499.4  149(0.8
Note: The relative poverty line is $1.08 or one-third of mean consumption, whichever is larger.
Source: Authors' calculations based on national household survey data.
160                                     The World Bank Researclh Observer, vol. 1 9, no. 2 (Fall 2004 



Figure 4. Number of Poor People by Region for a Relative Poverty Line, 1981-2001
2000
1800
1600
0
° 1400
*c 1000
: ea 1000   | 15 | li | lil l | l | l |  _ _ |  .  S,,r,-Sar-aran Air,ca
-w
.2 E
0.
600
0
400       :
200                        Z   t
0
1981       1984      1987       1990      1993       1996       1999      2001
Note: Therelative poverty line is $1.08 a day or one-third of mean consumption, whicheveris targer.
Source: Authors' calculations based on national household survey data.
the mid-199Os. Although there are marked changes in regional rankings, the aggre-
gate trends over time are quite similar. (This is at least in part a consequence of the
fact that the relative poverty lines are absolute over time.) The incidence of relative
poverty in the developing world as a whole is 29 percent in 2001, down from 50
percent in 1981. By this measure the total number of poor people is 1.5 billion in
2001. The total number of poor people by the relative poverty measure has shown
no trend decline since the mid-1980s. Excluding China, the number of relatively
poor people has remained at around 1.2 billion, though showing a slight upward
trend in the 1990s.
Shaohua Chen and Martin Ravallioni                                                   161



Conclusions
In providing new estimates of the extent of poverty in the developing world over
1981-2001, we have followed previous practice in focusing primarily on an inter-
national poverty line that accords with poverty lines typical of the poorest countries.
We used a poverty line of about $1 a day, though we also considered a line set at
twice this value, as well as a relative poverty line that rises with average consump-
tion when it exceeds about $3 a day. The estimates drew on newly available house-
hold surveys, and all past estimates have been revised in the light of the new data.
Our estimates appear to be more internally consistent and comparable over time
than past estimates, including those of World Bank (2002), which argued that there
were 200 million fewer poor at the end of twentieth century than 20 years earlier.
We find that the 200 million figure is an underestimate and that the reduction in
the number of poor people in the world was almost twice that size, entailing a near
halving of the 1981 poverty rate of 40 percent by 2001.
The precise time period covered is crucial, however. Progress against extreme
poverty has been uneven over time. The most dramatic reduction in poverty was in
the early 1980s; about half of the 390 million drop in the $1 a day poverty count
between 1981 and 2001 occurred in the first three years of that period. This coin-
cided with the sharp drop in extreme poverty in China in the aftermath of the
reforms that abandoned the socialist mode of agricultural production in favor of
household-based farming. In contrast, during 1987-93 the number of people living
on less than $1 a day stayed roughly constant, at around 1.2 billion. There was
more progress in the 1990s, once growth had been restored in the most populous
countries, China and India. There were 100 million fewer poor people by the $1 a
day standard at the end of the 1990s than at the beginning. For assessing overall
trends, we argue that the focus should be on the period 1984-2001.
Although the overall picture is good news, it is no cause for complacency. The
390 million fewer poor people by the $1 a day poverty line over 1981-2001 are still
poor by the standards of middle-income developing economies and certainly very
poor by the standards of what poverty means in rich countries. Our estimates indi-
cate that the number of people living on less than $2 a day has risen. Clearly, a great
many people remain poor and vulnerable to aggregate economic slowdowns.
Nor has this aggregate progress for the poorest over the 1980s and 1990s been
shared by all regions. The dramatic progress against poverty in the early 1980s
owes much to China. If one focuses on the developing world outside China, the num-
ber of poor people by the $1 a day standard has changed little-indeed, it has risen
slightly.
The composition of world poverty has changed dramatically. The number of poor
people has fallen in Asia but risen elsewhere. The share of the world's poor people
living in Africa has risen appreciably. Not only has Africa emerged in the 1990s as
162                               The World Bank Researcih Observer, vol. 19, no. 2 (Fall 2004)



the region with the highest incidence of poverty, but the depth of poverty is also
markedly higher than in other regions-suggesting that future economic growth
will have a harder time reducing poverty in Africa than elsewhere unless inequality
falls.
If the trend rate of decline in the incidence of poverty by the $1 a day standard
over 1984-2001 is maintained over 2001-15, it will be sufficient to halve the 1990
aggregate headcount index by 2015, consistent with the Millennium Development
Goal. However, only one part of the world-East and South Asia-will have reached
the goal.
Appendix: Survey Data Sets by Country, Date,
and WeLfare Indicator
TableA.I. SurveyDataSet
Share of 2001
population
Region          represented (%) Country               Survey dates       it..   I,
EastAsia            96.11    Cambodia           1997                   Expenditure
China             1980,1981,1984,         Income
1985.1987,1990
1990,1992,1993,        Expenditure
1996,1999,2001
Indonesia         1981,1984,1987.1990,    Expenditure
1993,1996,1999,2002
Lao PDR           1992.1997               Expenditure
Malaysia          1984.1987,1989,1992,    Income
1995,1997
Mongolia          1995,1998               Expenditure
Philippines       1985,1988,1991,1994,    Expenditure
1997,.2000
Thailand          1981,1988               Income
1988,1992,1996,1998,   Expenditure
1999,2000,(2002
Vietnam           1992/93.1998            Expenditure
Eastern Europe      97.32    Albania            1997.2002              Expenditure
and Central Asia            Armenia           1988.1996              Income
1996,1998              Expenditure
Azerbaijan        1988                    Income
1995,2001              Expenditure
Belarus           1988,1993,1995,1998,    Income
1999
1996-2000              Expenditure
(Continued)
Shaolhua Clien and Martin Ravallion                                              163



Table A. 1. (Continued)
Share of 2001
population
Region           represented (%)  Countrij             Survey dates       I...  .... i,
Bulgaria           1989.1992,1994,1995.   Expenditure
1997.2001
1993,1996               Income
Croatia            1998,1999,2000,2001    Expenditure
1988,1998              Income
Czech Republic     1988,1993,1996         Income
Estonia            1988. 1993,1995. 1998  Income
Georgia            1989.1996,1997         Income
1996,1998-2001          Expenditure
Hungary            1989,1993-1996,1998    Income
Kazakhstan         1988.1993              Income
1993,1996,2000          Expenditure
Kyrgyz Republic    1988, 1993,1996,1998   Income
1993,1997-2001          Expenditure
Latvia             1988,1993,1995,1998    Income
Lithuania          1988,1993,1994,1996,   Income
2000
1996,1998,2000          Expenditure
Macedonia          1988                   Income
1998                    Expenditure
Moldova            1988,1992,1997         Income
1997-2001               Expenditure
Poland             1985, 1987, 1989,      Income
199 3.1998
1990,1992,1993-96       Expenditure
Romania            1989,1992,1994         Income
1998,2000               Expenditure
Russian Federation  1988,1993             Income
1993,1996,1998,2000     Expenditure
Slovak Republic    1988,1992,1996         Income
Slovenia           1987,1993,1996,1998    Income
Tajikistan         1998                   Expenditure
Turkey             1987,1994,2000         Expenditure
Turkmenistan       1988,1993,1998         Income
Ukraine            1988,1992,1997         Income
1995,1996.1999          Expenditure
Uzbekistan         1988,1993              Income
1998, 2000              Expenditure
Latin America and   95.31     Argentina         1980.1982,1989,1992,    Income
the Caribbean                                   1996.1998,200)1
Bolivia            1986,1990,1997,1999    Income
164                                      The World Bank Research Observer, vol. 19, no. 2 (Fall 2004)



Table A.1. (Continued)
Shlire of 2001
populatiorn
Region           represented (t) Coanctry              Szr-y dates      WVelfare indicator
Brazil             1981. 1984, 1985. 1987,  Income
1988-89,1991).1993,
1995-98.2()()1
Chile             1987,1990,1992,1994.    Income
1996,1998,2001
Colombia           1980,1988.1989.1991,   Income
1995-96,1998-1999
Costa Rica         1981,1986.1989.1990),  Income
1993.1996-1998.2001
Dominican Republic  1986, 1989. 1992,1996,  Income
1998
Ecuador            1988,1994-95,1998      Income
1994-95                 Expenditure
El Salvador        1989,1995-98,2000      Income
Guatemala          1986,1989,1998.2001)   Income
Guyana             1993,1998              Expenditure
Flonduras          1986,1989-90,1992,     Income
1994,1996-1999
Jamaica            1988-94,1996-200()     Expenditure
Mexico             1984,1992.21)00        Expenditure
1989,1995.1996,1998     Income
Nicaragua          1993,1998              Expenditure
Panama             1979,1989,1991,        Income
1995-98.2000
Paraguay           1990, 1995. 1997-1999  Income
Peru               1985.1994,1996         Expenditure
1994,1996,2001)        Income
St. Lucia          1995                   Income
Trinidad and Tobago  1988,1992            Income
IJruguay           1981,1989.1996-1998,   Income
2001)
Venezuela          1981,1987,1989. 1991,  Income
1993.1995-98
Middle East and     74.0)5    Algeria            1988,1995              Expenditure
North Africa                Egypt, Arab Rep.   1991,1995, 21)00       Expenditure
Iran               1986,199),1994, 1998  Expenditure
Jordan             1987. 1992,1997        Expenditure
Morocco            1985, 199), 1998/99    Expenditure
Tunisia            1985,199),1995,2000   Expenditure
Yemen              1992,1998              Expenditure
(Continued )
Sliaohua Chenz and Mlartin Raxcallion                                             165



Table A.1. (Continued)
Share oJ 2001
population
Region           represented (%) Counrtry              Survey dates      Welfare indicator
South Asia          98.09     Bangladesh         1983/84, 1984-85. 1988.  Expenditure
1992.1996.2000
India              19 77/78, 1983.         Expenditure
1986-91,1993/94,
1995/96. 1999/2000
Nepal              1985,1995               Expenditure
Pakistan           1986/87, 1990/91, 1992/  Expenditure
93, 1996/97. 1998
Sri Lanka          1980,1985,1990),1995    Expenditure
Sub-Saharan         77.86     Botswana           1985/86, 1993           Expenditure
Africa                      BurkinaFaso        1994,1998               Expenditure
Burundi            1992.1998               Expenditure
Cameroon           1996                    Expenditure
Central African    1993                    Expenditure
Republic
C6ted'Ivoire       1985-88,1993, 1995,     Expenditure
1998
Ethiopia           1981,1995,20)00         Expenditure
Gambia. The        1992,1998               Expenditure
Ghana              1987,1989,1998          Expenditure
Kenya              1992, 1994,1997         Expenditure
Lesotho            1986, 1993,1995         Expenditure
Madagascar         1980,1993,1999          Expenditure
Mali               1989,1994               Expenditure
Malawi             1997                    Expenditure
Mauritania         1988,1993,1995,2000     Expenditure
Mozambique         1996/97                 Expenditure
Namibia            1993                    Expenditure
Niger              1992,1995               Expenditure
Nigeria            1985,1992, 1997         Expenditure
Rwanda             198 3/85                Expenditure
Senegal            1991,1994               Expenditure
Sierra Leone       1989                    Expenditure
South Africa       1993,1995,2000          Expenditure
Swaziland          1994                    Expenditure
Tanzania           1991                    Expenditure
Uganda             1988.1992,1996,1999     Expenditure
Zambia             1991,1993,1996,1998     Expenditure
Zimbabwe           1990/91, 1995           Expenditure
Source: National household surveys.
166                                       The World Ban1k Research Observer. vol. 19, no. 2 (Fall 2004)



Notes
Shaohua Chen is senior information officer in the Development Research Group at the World Bank: her
e-mail address is schenaworldbank.org. Martin Ravallion is research manager in the Development
Research Group at the World Bank; his e-mail address is mravallion@worldbank.org. The authors are
grateful to numerous colleagues at the World Bank who helped them obtain the necessary data and
answered their many questions, and the staff of numerous governmental statistics offices who collected
the primary survey data. The able assistance of Prem Sangraula is gratefully acknowledged. Angus
Deaton, Johan Mistiaen, Dominique van de Walle, and the journal's editorial board made useful com-
ments on the first draft.
1. Wade (2004) also questions the 200 million figure. However, he misdiagnoses the problem by
confusing changes in the methods used to count the world's poor people with the methodological issues
related to the way World Bank (2002) used different data sources. In fact, the Chen and Ravallion esti-
mates used in World Bank (2002) would be judged internally consistent by Wade's criteria. The Deaton
critique is more persuasive because it is grounded on a well-researched understanding of the methods
involved.
2. For a critical overview of our estimation methods, see Deaton (2002a), which covers the main
issues raised in the literature. Ravallion (2002a) replies to Deaton's comments: also see Ravallion
(200 3a) for further discussion.
3. The latest individual country estimates can be found online at www.worldbank.org/research/
povmonitor. The latest year's estimates at the country level are also published in the World Bank's
WTorld Development Indicators (see, for example, World Bank 2004).
4. Because we are using the same Ppp rates as Chen and Ravallion (2001), we use Penn World
Tables 5.6, which was the latest available at that time. Version 6.1 has since become available.
5. Thus we do not accept the claims made by Reddy and Pogge (2002) and Wade (2004) that we
have lowered the real value of the poverty line. They ignore the fact that there has been (in effect) a Ppp
devaluation of poor countries relative to the tJnited States since the switch from the 1985- to 1993-
based Ppps, reflecting both the new icp price data and differences in methods of measuring the l'pP rate.
For further discussion of the Reddy and Pogge criticisms of our methods see Ravallion (2002b).
6. Note that the same poverty line is generally used for urban and rural areas. There are two excep-
tions. For China and India we estimate poverty measures separately for urban and rural areas and use
sector-specific cPis. For India we also use a corrected version of the rural cPI (the consumer price index
for agricultural laborers), as discussed in Datt and Ravallion (1998).
7. For Nigeria we used the co, per capita growth rate. Substantial changes in Nigeria's method of
calculating private consumption made it impossible to construct a consistent series for consumption.
8. ThusH93=[(1995- 1993)/(1995-1989)] H93(89)+I[(1993- 1989)/(1995-1989)]. H93(95).
In a small number of cases this method did not give sensible results in that either M93(89) or M93(95)
was outside the interval [M(8 9), M(9 5)] even though the national accounts growth rates were positive
for both 1989-93 and 1993-9 5. In these cases we ignored the national accounts data and fell back on
simply estimating M(93), using the growth rate in survey means between 1989 and 1995.
9. This assumes that the trend is linear rather than exponential (linear in logs). The exponential
trends are 2.9 percent a year using all eight years and 2.5 percent a year ignoring the first year.
References
Ahmad, Sultan. 2003. "Purchasing Power Parity for International Comparison of Poverty: Sources
and Methods." World Bank, Development Data Group, Washington, D.C. Available online at
www.worldbank.org/data/icp.
Shaolhua Cheri anid Nlartirn Raiallion                                              167



Atkinson, Anthony B., and Francois Bourguignon. 1999. "Poverty and Inclusion from a World Per-
spective." Paper presented for the Annual World Bank Conference on Development Economics,
European Conference, Paris.
Bourguignon, Francois. and Christian Morisson. 2002. "The Size Distribution of Income among World
Citizens, 1820-1990." Americanii Economiiic Review 92(4):727-44.
Chen. Shaohua, and Martin Ravallion. 1996. "Data in Transition: Assessing Rural Living Standards in
Southern China." Cihina Economic Review 7:2 3-5 6.
. 2000. "How Did the World's Poorest Fare in the 1990s?" Policy Research Working Paper
2409. World Bank, Washington, D.C. Available online at http://econ.worldbank.org/resource.
php?type= 5.
2001. "How Did the World's Poor Fare in the 1990s?" Review of Incorne and Wvealth 47(3):
28 3-300.
Chen, Shaohua, Gaurav Datt. and Martin Ravallion. 1994. "Is Poverty Increasing or Decreasing in the
Developing World?" Reviei of Inicomiie and Wealthi 40:359-76.
Datt. Gaurav. and Martin Ravallion. 1992. "Growth and Redistribution Components of Changes in
Poverty Measures: A Decomposition with Applications to Brazil and India in the 1980s." Journal of
Development Econiomrlics 38:275-95.
. 1998. "Farm Productivity and Rural Poverty in India." Journal oJ Developmlent Studies 34:
62-85.
.2002. "Has India's Post-Reform Economic Growth Left the Poor Behind?" Journal of Economic
Perspectives 16(31:89-1 08.
Deaton, Angus. 2002a. "Counting the World's Poor: Problems and Possible Solutions." World Bank
Research Observer 16(2):] 2 5-47.
2002b. "Is World Poverty Falling?" Finance andcDevelopment 39(2):4-7.
2003. "Adjusted Indian Poverty Estimates for 1999-(0." Economrlic and Political 11 , i li 38(4):
322-26.
Foster, James. J. Greer, and E. Thorbecke. 1984. "A Class of Decomposable Poverty Measures."
Econometrica 52:761-65.
Gibson, John, Jikun Huang, and Scott Rozelle. 2001. "Why Is Income Inequality So Low in China Com-
pared to Other Countries? The Effect of Household Survey Methods." Economnics Letters 71: 329-33.
Ravallion, Martin. 2002a. "Comment on 'Counting the World's Poor' by Angus Deaton." World Bank
Researchi Observer 16(2):149-56.
-(2002b. "How Not to Count the Poor: A Reply to Reddy and Pogge." World Bank, Development
Research Group, Washington, D.C.
- 2003a. "The Debate on Globalization, Poverty and Inequality: Why Measurement Matters."
Internzational Affairs 79(4):739-54.
. 2003b. "Measuring Aggregate Economic Welfare in Developing Countries: How Well Do
National Accounts and Surveys Agree?" Review of Econorrnics and Statistics 85:645-52.
Ravallion, Martin, and Shaohua Chen. 20)04. "China's (Uneven) Progress against Poverty." World
Bank, Development Research Group. Washington, D.C.
Ravallion. Martin, Gaurav Datt, and Dominique van de Walle. 1991. "Quantifying Absolute Poverty in
the Developing World." Review of Incomne and Wealth 37:345-61.
Reddy. Sanjay G., and Thomas W. Pogge. 2002. "How Not to Count the Poor." Working paper, Barnard
College, New York.
Sen, Abhijit, and Himanshu. 20)0)3. "Poverty and Inequality in India: Getting Closer to the Truth."
Working paper, Jawaharlal Nehru University, New Delhi.
168                                       The Wtorld Bank Researchl Observer, vol. 19, no. 2 (Fall 2004)



Sundaram, K.. and Suresh D. Tendulkar. 2003. "Poverty in India in the 1990s. Revised Results for All-
India and 15 Major States for 1993-94." Econornic and Political Weekly 38(46):4865-72.
Wade, Robert Hunter. 2004. "Is Globalization Reducing Poverty and Inequality?" World Development
32(4):567-89.
World Bank. 1990. WAlorld Developnlent Report: Poverty. New York: Oxford University Press.
. 1997. China 2020: Sharin.g Rising Income. Washington, D.C.
. 2000. World Developmenit Report 2000/2001: Attacking Poverty. New York: Oxford University
Press.
2002. Globalization, Growvth and Poverty. Washington, D.C.
. 2004. World Development lndicators. Washington, D.C.
S)iaoiiiUi C'heii aid MAartii R{avaflioru                                             169



I
I



Much Ado about Nothing? Do Domestic
Firms ReaLLy Benefit from Foreign
Direct Investment?
HoLger Gorg * David Greenaway
Governments the world over offer significant inducements to attract investment, motivated
by the expectation of spillover benefits to augment the primary benefits of a boost to
national income from newi, investment. There are several possible sources of induced spill-
overs from foreign direct investmrlent. This article evaluates the empirical evidence on pro-
ductivity, wage, and export spillovers in developing, developed, and transition economies.
Although theor.y can identify a range of possible spillover channels, robust empirical sup-
port for positive spillovers is at best ,mixed. The article explores the reasons and concludes
wvith a review of policy aspects.
Of all the drivers of globalization-arm's length trade, migration of workers, and
cross-border investment-the last is probably the most visible. This is probably why
public anxiety about globalization often manifests itself as hostility toward multina-
tional firms (see Deardorff 2003 for a recent appraisal of such anxieties). From an
economic standpoint. cross-border investment may also be at the margin the most
important manifestation of globalization. Annual flows of foreign direct investment
(FDI) now exceed $700 billion, and the total stock exceeds $6 billion. Over the past
decade FDI flows have grown at least twice as fast as trade.
As with arm's length trade, the FDT environment is policy distorted, but it is gradu-
ally becoming more liberalized. Of 145 regulatory changes made by 60 countries in
1998, 94 percent created more favorable conditions for FDI (IJN 1 999). In many cases
interventions have extended beyond creating a more liberal environment to provid-
ing substantial public subsidies. For example, Head (1998) reports that the govern-
ment of Alabama paid the equivalent of $150,000 per employee to Mercedes for
locating its new plant in the state in 1994. Across the Atlantic the UK government
The World Bank Research Observer. vol. 19, no. 2.
©) The International Bank for Reconstruction and Development / TH4E WORLD RANK 2004: all rights reserved.
doi:10).1093/wbro/lkhtl 9                                           19:1 71-197



provided an estimated $30,000 per employee to attract Samsung to the northeast of
England in the late 1990s and $50,000 per employee to attract Siemens (Girma and
others 2001). Some countries also provide tax incentives. For example, Ireland
offers a corporate tax rate of 12.5 percent to all manufacturing firms locating there.
There seems to be a widely held assumption that foreign firms more than pay
their way, bringing not only new investment that boosts national income but also
secondary spillovers, resulting in productivity growth or higher export growth.
Much econometric work has been done in this area, but the results on the impor-
tance of spillovers are mixed at best. There is some evidence from case studies of spill-
over benefits to domestic firms (see Moran 2001). but even at that level there is
disagreement in particular instances.'
The failure of econometric work to find unambiguously positive effects could be
due to a number of factors. Despite theoretical arguments supporting spillovers, they
may simply be unimportant in reality. Multinational corporations may be effective
at ensuring that firm-specific assets and advantages do not spill over. Another possi-
bility is that spillovers exist and make up some part of the residual that appears in all
growth equations, but current statistical methods and data sets are unable to iden-
tify them. Furthermore, there may be much heterogeneity in spillovers, and aggre-
gate studies may therefore fail to detect them. Moreover, the lack of good-quality,
comprehensive firm- and plant-level data sets is a serious impediment to research.
This article examines in detail the evidence for intraindustry productivity spill-
overs in theory and in econometric analyses, taking essentially a microeconomic
and microeconometric view.2 It updates earlier surveys, such as Blomstrbm and
Kokko (1998) and Lipsey (2002) and highlights methodological issues and the
scope for policymakers to enhance potential spillover effects. The review is more
focused on spillovers from FDI than are related studies by Keller (2001) or Saggi
(2002), who discuss the scope and evidence for international technology diffusion
more generally, without much detail on FDI.
This article looks first at what guidance theory can give, on two counts: the possi-
ble channels for transmission of spillover benefits and whether host country charac-
teristics are likely to make a difference in the extent or speed with which spillovers
occur. It examines the empirical evidence on spillovers in developed, developing.
and transition economies and then draws some implications for policy. Should gov-
ernments intervene, and if so, how? Does policy make any difference?
What Does Theory TeLl Us?
A well-developed and extensively surveyed (see Caves 1996: Markusen 199 5) body
of literature tries to explain why multinational corporations set up overseas rather
than export directly or license their product or technology. The most persuasive
172                                 T7he XVorld Baitik Research Obeserver, vol. 19. no. 2 (Fall 2004)



explanations emphasize the coexistence of proprietary knowledge and market failures
in protecting that knowledge. Thus the firm internalizes certain transactions to pro-
tect its brand, technology, and marketing advantages. These motives are taken as
given, in particular, the existence of some kind of firm-specific asset, usually some
kind of technological advantage, including innovative management and organiza-
tional processes as well as new production methods and technologies.
The first question, then, is having chosen a particular location, how might any
advantages spill over to the local economy through firms in the same industry?
Then, with potential transmission channels identified, are there particular host
economy characteristics that make benefits from spillovers more or less likely?
Spillover Channels
When a firm sets up a plant overseas or acquires a foreign plant, it does so in the
expectation of realizing a higher rate of return than a given home country firm with
an equivalent investment. The source of the higher return is the technological
advantage alluded to. Whatever its source, the only way domestic firms can gain
from external benefits is if some form of indirect technology transfer takes place-
multinational firms will not simply hand over the source of their advantage. The
theoretical literature identifies four channels through which spillovers might boost
productivity in the host country: imitation, skills acquisition, competition, and
exports (table 1).
Imitation. Imitation is the classic transmission mechanism for new products and
processes. One mechanism commonly alluded to in the theoretical literature on
technology transfer from developed to developing economies is reverse engineering
(Das 1987; Wang and Blomstrom 1992). Its scope depends on the complexity of
products and processes, with simple manufactures and processes easier to imitate
than more complex ones. The same principle applies to managerial and organiza-
tional innovations, although these are thought to be easier to imitate. Imitation is. of
course, not the same as replication, and it would be surprising if the rents accruing
Table 1. Potential Channels for Spillover from Foreign Direct Investment
Driver               Souirces oj productivityj qain
Imitation            Adoption of new production methods. Adoption of new management practices.
Skills acquisition   Increased productivity of complementary labor. Tacit knowledge.
Competition          Reduction in X-inefficiency. Faster adoption of new technology.
Exports              Scale economies. Exposure to technology frontier.
Source: Authors' compilation: see text for details.
HIliqer Gdrg cti(d David Greerzavail                                         173



to multinational firms were entirely dissipated by the process. However, any upgrading
to local technology deriving from imitation could result in a spillover, with conse-
quent benefits for the productivity of local firms.
Skills Acquisition. Adoption of new technology can also occur through the acquisi-
tion of human capital. Even when the locational pull for FDI is relatively low wages,
multinational firms tend to demand relatively skilled labor. Generally, they will
invest in training, and it is impossible to lock in such resources completely. (This
inability to fully protect investment in human capital has long been an argument for
infant industry protection as a response to potential first-mover disadvantages; see
Baldwin 1969.) The movement of labor from multinational firms to other existing
or new firms can generate productivity improvements through two mechanisms:
through a direct spillover to complementary workers and through knowledge carried
by workers who move to another firm. Haacker (1999) and Fosfuri and others
(2001) argue that the knowledge that workers bring with them is the most impor-
tant channel for spillovers, and some empirical work supports this (Djankov and
Hoekmann 1999; Gbrg and Strobl 2002c).
Competition. Many models emphasize the role of competition (Wang and Blomstrbm
1992; Glass and Saggi 2002). Unless an incoming firm is offered monopoly status, it
will produce in competition with indigenous firms. Even if indigenous firms are
unable to imitate the multinational's technology and production processes, entry of
the multinational firm puts pressure on them to use existing technology more effi-
ciently, yielding productivity gains. Greater competition leading to a reduction in
X-inefficiency is analogous to one of the standard gains from arm's length trade and
is frequently identified as one of the major sources of gain. In addition, competition
may increase the speed of adoption of new technology.
Exports. A further indirect source of productivity gain might be through exports.
Crudely, domestic firms can learn to export from multinationals (Aitken and others
1997; Barrios and others 2003; Greenaway and others forthcoming). Exporting
generally involves fixed costs to establish distribution networks, create transport
infrastructure, and learn about consumers' tastes, regulatory arrangements. and so
on in overseas markets. Multinational firms generally come already armed with
such information and exploit it to export from the new host country. Through col-
laboration, or more likely imitation, domestic firms can learn how to penetrate
export markets. There is a growing body of literature that links exporting and pro-
ductivity. Recent work on Germany, Mexico, Morocco, Spain, the United Kingdom,
the United States, and Venezuela suggests that productivity levels are higher in
exporting firms than in nonexporting firms (Clerides and others 1998; Bernard and
174                               The World Banik Researchi Observer, vol. 19. no. 2 (Fall 2004)



Jensen 1999; Bernard and Wagner 1997; Delgado and others 2003; and Girma and
others forthcoming). Central to this literature is whether firms self-select into export-
ing or increase their productivity after entering export markets.
Host Country Characteristics and Spillovers
The literature on the determinants of FI)I emphasizes locational characteristics as
important factors in multinationals' decisions on where to invest (Wheeler and
Mody 1992; Brainard 1997; Gorg 2002). The focus here, however, is on whether
there are locational characteristics that affect the speed of adoption of new technol-
ogy and the spillover of productivity gains.
In a pioneering contribution Findlay (1978) emphasizes the importance of rela-
tive backwardness and contagion. Findlay's model suggests that the greater is the
distance between two economies in terms of development, the greater the backlog of
available opportunities to exploit in the less advanced economy, and so the greater
the pressure for change and the more rapid the uptake of new technology. Speed of
adoption is also a function of contagion or the extent to which the activities of the
foreign firm pervade the local economy. Thus technology transfer will be more rapid
if the multinational firm quickly establishes upstream and downstream networks,
because domestic firms involved in supply and distribution chains gain exposure to
new technology, promoting its diffusion.
Glass and Saggi (1998) also see a role for technological distance between the host
and home country, but a different one from Findlay's. Any technology gap signals
something to the multinational firm about absorptive capacity. The bigger the gap,
the less likely the host is to have the human capital, physical infrastructure, and dis-
tribution networks to support inward investment. This influences not only the deci-
sion to invest but also what kind of technology to transfer. Specifically, the bigger
the gap, the lower the quality of technology transferred and the lower the potential
for spillovers. This seems more plausible than Findlay's notion of a lack of absorptive
capacity as the driver. Clearly, technological distance will be directly related to the
potential gains from spillovers, but it is also likely to be inversely related to the prob-
ability that domestic firms are actually able to access them.
Summary
Economic theory gives some guidance on what to expect from cross-border investment
and spillovers. In general, multinational firms have firm-specific advantages that
might be related to the production methods they use, the way they organize their
activities, the way they market their products and services, and so on. Once a multi-
national has set up a subsidiary, it may be unable to prevent some of the benefits of
these advantages from spilling over to indigenous firms through imitation, labor
lolger Gorg and David Greeniaway                                      1775



mobility, competition, or export. Such spillovers have the potential to raise produc-
tivity, and exploitation of these potential channels of spillovers might be related to the
structural characteristics of the host economy, in particular its absorptive capacity.
What Does the Evidence Tell Us?
Empirical studies on spillovers from FDI were pioneered by Caves for Australia
(1974), Globerman for Canada (1979), and Blomstrom for Mexico (1986). Since
then, their empirical models have been extended and refined, although the basic
approach remains. Most econometric analyses use a framework that regresses the
labor productivity or total factor productivity of domestic firms on a range of inde-
pendent variables. To measure productivity spillovers from multinational firms, a
variable is included that proxies the extent of foreign firms' penetration, usually cal-
culated as the multinational's share of total employment or sales in a given sector. 3
In other words, the regression allows for an effect of FDI on the productivity of domestic
firms in the same industry. If the regression analysis yields a positive and statistically
significant coefficient on the foreign presence variable, this is taken as evidence that
spillovers have occurred from multinational firms to domestic firms.4 Most studies
use either the contemporaneous level of foreign penetration or relatively short lags
(most often one year) as their explanatory variables. If anything, therefore, these
studies usually measure short-run effects of foreign presence on domestic productivity.
Table 2 sets out details of 40 studies of horizontal productivity spillovers in manu-
facturing industries in developing, developed, and transition economies. Of those,
22 report unambiguously positive and statistically significant horizontal spillover
effects (2 of which find positive and significant effects for only one of several countries
studied). All but eight of those reporting unambiguously positive spillovers use
cross-sectional data, which may lead to biased results.
Gorg and Strobl (2001) argue that panel data using firm-level data are the most
appropriate estimating framework for two reasons. First, they permit investigation
of the development of domestic firms' productivity over a longer time period, rather
than at one point in time. Second, they allow investigation of spillovers after control-
ling for other factors. Cross-section data, particularly if aggregated at the sectoral
level, fail to control for time-invariant differences in productivity across sectors that
might be correlated with foreign presence without being caused by it. Thus coeffi-
cients on cross-section estimates are likely to be biased. For example, if productivity
is higher in the electronics sector than in the food sector, multinationals may be
attracted to the electronics sector. Cross-sectional data would show a positive and
statistically significant relationship between the level of foreign investment and
productivity consistent with spillovers, even though foreign investment did not
cause high levels of productivity but rather was attracted by them.
176                                  Tlre WVorld Batik Research Observer, vol. 19, no. 2 (Fall 2004)



Table 2. Papers on Intraindustry Productivity Spillovers
Aggregation
Author                  Country            Period   Data          level'         Resultb
Developing econornies
1 Blomstrom and       Mexico              1970     Cross-sectional Industry       +
Persson ( 1983)
2 Blomstrom (1986)     Mexico           1970/1975  Cross-sectional Industry       +
3 Blomstrom and Wolff  Mexico          1970/1975   Cross-sectional Industry       +
(1994)
4 Kokko (1994)         Mexico             1970     Cross-sectional Industry       +
5 Kokko (1996)        Mexico              1970     Cross-sectional Industry       +
6 Haddad and Harrison  Morocco          1985-89    Panel         Micro and
(1993)                                                           industry
7 Kokko and others    Uruguay             1990     Cross-sectional Micro
(1996)
8 Blomstrom and        Indonesia          1991     Cross-sectional Micro          +
Sjoholm (19991
9 Sjbholm (1999a)      Indonesia         1980-91   Cross-sectional Micro          +
10 Sjoholm (1999bi      Indonesia         1980-91   Cross-sectional Micro          +
11 Chuang and Lin       Taiwan             1991     Cross-sectional Micro          +
(1999)
12 Aitken and Harrison  Venezuela         1976-89   Panel         Micro
(1999)
13 Kathuria (2000)     India              1976-89   Panel         Micro            ?
14 Kokko and others     Uruguay            1988     Cross-sectional Micro          ?
(2001)
15 Kugler (2001)        Colombia          1974-98   Panel         Industry
16 L6pez-C6rdova        Mexico            1993-99   Panel         Micro           -.?
(20(02)
17 Gorg and Strobl      Ghana             1991-97   Panel         Micro            +
(2002c)
Developed countries
18 Caves (1974)         Australia          1966     Cross-sectional Industry       +
19 Globerman (1979)     Canada             1972     Cross-sectional Industry       +
20 Liu and others       United Kingdom    1991-95   Panel         Industry         +
(2000)
21 Driffield (2001)     United Kingdom    1989-92   Cross-sectional Industry       +
22 Girma and others     United Kingdom    1991-96   Panel         Micro            ?
(2001)
23 Girma and Wakelin    United Kingdom    198(0-92  Panel         Micro
(20(1)
24 Harris and Robinson  United Kingdom    1974-95   Panel         Micro
(200)4)
25 Girma and Wakelin    United Kingdom    1988-96   Panel         Micro
(2002)
(Continued)
Holger Gorg ind David Greenawayj                                                      177



Table 2. (Continued)
Aggregationi
Authzor                  Country             Period   Datal         level'          Resultb
26 Haskel and others     United Kingdom   1973-92     Panel          Micro            +
(2002)
27 Girma (20(02)         United Kingdom   1989-99     Panel          Micro
28 Girma and Gdrg        United Kingdom   1980-92     Panel          Micro
(2002)
29 Ruane and Ugur        Ireland          1991-98     Panel          Micro            +
(2002)
28 Barrios and Strobl    Spain            1990-94     Panel          Micro             ?
(2002)
29 Dimelis and Louri     Greece              1997     Cross-sectional Micro           +
(2002)
30 Castellani and Zanfei  France. Italy.  1992-97     Panel          Micro        + for Italy;
(2002b)                Spain                                                   - for Spain;
P for France
31 Keller and Yeaple    United States     1987-96     Panel          Micro            +
(2003)
32 Gbrg and Strobl      Ireland           3 973-96    Panel          Micro            +
(2003)
Transitioni econorrmies
33 Djankov and          Czech Republic    199 3-96    Panel          Micro
Hoekman (2000)
34 Kinoshita (2001)      Czech Republic   199 5-98    Panel          Micro
35 Bosco(2001)           Hungary          1993-97     Panel          Micro
36 Konings (2001)       Bulgaria          199 3-97    Panel          Micro
Poland            1994-97
Romania           199 3-97
37 Damijan and others   Bulgaria.         1994-1998   Panel          Micro           ? or-
(20(01)              Czech Republic,                                           + only for
Estonia,                                                  Romania
Hungary, Poland.
Romania,
Slovakia,
Slovenia
38 Li and others (2001)  China               1995     Cross-sectional Industry        +
39 Smarzynska-Javorcik  Lithuania         1996-2000) Panel           Micro
(forthcoming)
40 Zukowska-Gagelmann    Poland           1993-97     Panel          Micro
(2000)
'Micro data are at the firm, plant. or establishment level.
bA + indicates positive and statistically significant, - indicates negative and statistically significant, and indicates
mixed or statistically insignificant results on the foreign presence variable for the aggregate sample.
178                                          The World Bank Researchi Observer, vol. 1 9, no. 2 (Fall 2004)



Taking this into consideration, the evidence on positive horizontal spillovers is
much weaker. There are only eight studies employing panel data that find unambig-
uously positive evidence in the aggregate, and almost all of these are for developed
economies: Liu and others (2000) and Haskel and others (2002) for the United
Kingdom, Castellani and Zanfei (2002b) for Italy, Keller and Yeaple (2003) for the
United States, Ruane and Ugur (2002) and Gorg and Strobl (2003) for Ireland,
Damijan and others (2001) for Romania, and Gbrg and Strobl (2002c) for Ghana.
Liu and others (2000), however, use industry-level data that aggregates over heter-
ogeneous firms, which may lead to biased results.5 This leaves only seven studies
using appropriate data and estimation techniques that report positive evidence for
aggregate spillovers.
Several studies using firm-level panel data find some evidence of negative effects
of the presence of multinationals on domestic firms in the aggregate. These include
Aitken and Harrison (1999) for Venezuela, L6pez-C6rdova (2002) for Mexico,
Castellani and Zanfei (2002b) for Spain, Djankov and Hoekman (2000) for the Czech
Republic, Konings (2001) for Bulgaria, Zukowska-Gagelmann (2000) for Poland,
and Damijan and others (2001) for seven countries in Central and Eastern Europe.
Many studies on transition economies find at least some evidence of negative results.
Nineteen of the studies find no statistically significant effects on average of multi-
nationals on domestic productivity.6
What Explains the Negative or Neutral Effects?
There have been several explanations for the negative results found by some studies.
The most plausible is that foreign firms reduce the productivity of domestic firms
through competition effects, as suggested by Aitken and Harrison (1999) and
Konings (2001). They argue that multinationals have lower marginal costs due to
some firm-specific advantage, which allows them to attract demand away from
domestic firms, thus forcing the domestic firms to reduce production and move up
their (given) average cost curve.
This argument is not necessarily inconsistent with the theoretical discussion in the
previous section, which presented competition as one of the channels through which
positive spillovers are transmitted. Some firms may experience negative competition
effects in the short run (moving up a given average cost curve), whereas other firms
may improve efficiency (shifting down their average cost curve) because of increased
competition in the short run as well as the long run. Evidence for positive effects of
competition are found by Kokko (1996) for Mexico and Driffield (2001) for the
United Kingdom.
There are also other explanations for a failure to find evidence of positive aggre-
gate spillovers in the short run. There may be lags in domestic firms' learning from
multinationals, which short-run analyses do not pick up. Multinational firms may
Holger G5rg atud David Greenawvai                                    179



be able to guard their firm-specific advantages closely, preventing leakages to
domestic firms and therefore spillovers as well. Positive spillovers may affect only a
subset of firms, so that aggregate studies underestimate the true significance of such
effects. Spillovers may occur not horizontally (intraindustry) but vertically through
relationships that are missed in conventional spillover studies.
The first two explanations are straightforward and plausible and require little
comment. More detailed discussion is warranted for the last two explanations.
Absorptive Capacity. As discussed, the theoretical literature suggests that not all
firms would be expected to benefit equally from knowledge spillovers from multi-
nationals. Whether a firm benefits depends on its relative backwardness and its
capacity for assimilating knowledge-its absorptive capacity. Some of the empirical
literature has also considered these issues.
Kokko (1994) advances the idea that spillovers depend on the complexity of the
technology transferred by multinationals and on the technology gap between
domestic firms and multinational firms. Using cross-section industry-level data for
Mexico, he finds no evidence for spillovers in industries in which multinationals use
highly complex technologies (as proxied by large payments on patents or high capi-
tal intensity). A large technology gap on its own does not appear to hinder technol-
ogy spillovers on average, although industries with large gaps and a high foreign
presence experience lower spillovers than others. Expanding on Kokko (1994),
Kokko and others (1996) hypothesize that domestic firms can benefit only if the
technology gap is not too wide so that domestic firms can absorb the knowledge
available from the multinational-an argument similar to that of Glass and Saggi
(1998). Thus domestic firms using very backward production technologies and low
skilled workers may be unable to learn from multinationals. Using a cross-section of
firm-level data for Uruguay, Kokko and others find evidence for productivity spill-
overs to domestic firms with moderate technology gaps (measured as the difference
between the domestic firm's labor productivity and the average labor productivity in
foreign firms) but not for firms that use considerably lower levels of technology. 7
Girma and others (2001), using firm-level panel data, find no evidence of produc-
tivity spillovers in UK manufacturing on average-under the assumption that spill-
overs are homogeneous across different types of domestic firms. There is evidence for
spillovers to firms with a small gap between their productivity level and the industry
frontier productivity level (called the technology gap). Productivity appears to
increase with increasing foreign presence for firms with a technology gap of 10 per-
cent or less, whereas it appears to diminish in firms with larger gaps. Girma (2002)
uses threshold regression techniques to quantify the significance of absorptive
capacity, and Girma and Gorg (2002) use conditional quantile regression tech-
niques to allow for different effects of FDI on establishments at different quantiles of
the productivity distribution. Both studies find support for the hypothesis that only
180                                 The World Bank Research Observer, vol. 19. no. 2 (Fall 2004)



firms with some minimum level of absorptive capacity benefit from productivity
spillovers.
In a similar vein, Barrios and Strobl (2002), in firm-level panel data for Spanish
manufacturing, find little evidence for any aggregate horizontal spillovers from mul-
tinational firms. There is evidence for positive spillovers from foreign presence to
domestic exporters but not to nonexporters, which they interpret as evidence that
absorptive capacity matters. They argue that exporting firms are more exposed to
international competition and therefore are more likely to use more advanced tech-
nologies and to benefit from positive spillovers than are nonexporters. Kinoshita
(2001), using firm-level panel data for the Czech Republic, also finds no evidence of
spillovers on average but finds positive spillovers for local firms that are research and
development (R&D) intensive. She interprets this as evidence that absorptive capacity
is important.
Damijan and others (2001) also define absorptive capacity in terms of local firms'
R&D activities. In their firm-level panel data for a number of Central and Eastern
European transition economies, they fail to detect evidence of productivity spillovers
affecting the average firm. Taking into account absorptive capacity, by interacting
the foreign presence variable with a firm's R&D expenditure. they find evidence of
negative spillovers for the Czech Republic and Poland but positive spillovers for
Romania, and no evidence for all other countries.
Regional Dimensions. Because human capital acquisition and imitation are considered
important channels for knowledge spillovers, domestic firms located near
multinationals may be more likely to benefit than other firms. For example, Audretsch
(1998:21) argues that geographic proximity is necessary to facilitate knowledge spill-
overs because "knowledge is vague, difficult to codify, and often only serendipitously
recognized." Therefore, transmission costs are assumed to increase with distance.
Several studies have investigated the geographic dimension of horizontal spill-
overs. Calculating proxies for foreign presence at the regional level and using cross-
sectional data for Indonesia, Sjoholm (1999a) fails to find evidence of a regional
component. Aitken and Harrison (1999), using firm-level panel data for Venezuela.
also fail to find positive spillovers from multinationals to domestic firms in the same
region, though they find negative spillovers from multinationals in the same sector
in any region in the country. From firm-level panel Girma and Wakelin (2002) find
evidence for positive spillovers from FlI in the same region and sector as domestic
firms in the United Kingdom, but the results are significant only for firms that have a
low technology gap vis-a-vis multinationals.
Importance of vertical linkages. If multinationals prevent the transfer of their firm-
specific knowledge to domestic competitors in the same industry, there is no scope
for intraindustry knowledge spillovers. It is possible, however, that multinational
Holger Gorq and Dav'id Greeiiawayj                                     181



firms voluntarily or involuntarily help increase the efficiency of domestic suppliers
or customers through vertical input-output linkages. Multinationals may provide
technical assistance to suppliers to help them raise the quality of the intermediate
products they produce (Moran 2001), or they may simply insist on high quality
standards for local inputs, providing incentives for local suppliers to upgrade their
technology. Multinationals may also provide active assistance or passive guidelines
to domestic customers on the most effective way to use the products the firms supply.
Several recent studies have empirically investigated vertical spillovers (table 3 ).8
Kugler (2001) worked with industry-level panel data for 10 Colombian manufactur-
ing industries during 1974-98, using an estimation framework that distinguishes
intraindustry and interindustry spillovers. He finds widespread evidence for positive
interindustry spillovers, but finds evidence for intraindustry spillovers only in one
sector (machinery equipment). However, his framework does not distinguish spill-
overs through backward or forward linkages. Smarzynska-Javorcik (forthcoming)
uses firm-level panel data for Lithuania for 1996-2000 to consider spillovers
through backward linkages. Although she found no evidence for aggregate horizon-
tal spillovers, she does find productivity spillovers through backward linkages.
Blalock and Gertler (2003) also find results suggesting positive productivity spill-
overs through backward linkages in their analysis of Indonesian plant-level panel
data. They do not find evidence for horizontal spillovers, however.
Driffield and others (2002) allow for spillovers through horizontal, backward,
and forward relationships. They examine the relative importance of each using
industry-level panel data for UK manufacturing during 1984-92. Their economet-
ric estimations show evidence for positive spillovers through forward linkages but
not of statistically significant spillovers through backward linkages. The results for
horizontal spillovers are inconclusive. In a further study for the United Kingdom,
Harris and Robinson (2004) use plant-level panel data to estimate productivity
equations for 20 manufacturing sectors separately. Like Kugler (2001), they distin-
guish only horizontal and vertical spillovers; they do not separate vertical spillovers
into backward or forward linkages. Their results suggest that interindustry spill-
overs are much more prevalent than intraindustry spillovers. None of the spillovers
are always positive, however, and there is evidence of negative spillovers in many
sectors. Girma and others (2003), using UK firm-level data, also find substantial
differences in whether domestic firms benefit from vertical linkages, depending on
their export activities.
Wage Spillovers
If there are positive productivity spillovers from multinational firms to domestic firms
and if some of these spillovers are due to increasing labor productivity, domestic
firms will pay higher wages in competitive labor markets. Another field of empirical
182                                  The World Bank Researchi Observer, vol. 19, no. 2 (Fall 2004)



Table 3. Studies on Vertical Spillovers
Resultb
Autlhor                                  Couuitryj               Period         Data       Aggreiartior leiePl  Horizontal     Backward      Forward
1 Kugler (2001)]                         Colombia             1974-98          Panel       Industry                 ?          Many +         NA
2 Smarzynska-Javorcik   rrl       I      Lithuania             1996-200()       Panel      Micro                    .             +            NA
3 Driffield and others (2002)            UJnited Kingdom      1984-92          Panel       Industry                .                           +
4 Harris and Robinson (2002)'            tlnitedKingdom        1974-95          Panel      Micro
5 Blalock and Gertler (2(0 3)            Indonesia            1988-96          Panel       Micro                   r              +           NA
'Micro data are at the firm, plant, or establishment level.
bA + indicates positive and statistically significant. - indicates negative and statistically significant. and ? indicates mixed or statistically insignificant results on the foreign
presence variable for the aggregate sample. NA = not applicable.
'Do not distinguish backward and forward spillovers.



Table 4. Studies on Wage Spillovers
Aggregation
Author                    Coulatqr         Period   Datn          level       Resultb
1 Aitken and others (1996)  Mexico        1984-90   Panel         Industry
Venezuela       19 77-89  Panel         Industry
United States    1987     Cross-sectional  Industry   +
2 LipseyandSjoholm(2001)  Indonesia        1996     Cross-sectional  Micro     +
3 Girma and others (2001)  UJnited Kingdom  1991-96  Panel        Micro
4 Driffield and Girma(2003)  United Kingdom  1980-92  Panel       Micro
aMicro data are at the firm, plant, or establishment level.
bA + indicates positive and statistically significant, - indicates negative and statistically significant, and ? indi-
cates mixed or statistically insignificant results on the foreign presence variable for the aggregate sample.
research focuses on this connection, emphasizing horizontal spillovers (table 4).9
Productivity spillovers are not the only channel for such wage spillovers, however.
Multinationals often pay higher wages, even after controlling for size and other firm
and sectoral characteristics (Girma and others 2001; Lipsey and Sjoholm 2001:
Gorg and others 2003). This is attributed to the multinational firms' ownership of
firm-specific assets, implying that they use higher levels of technology than domestic
firms. If multinationals and domestic firms compete in the same labor market, domes-
tic firms have to pay higher wages to attract workers. Wage spillovers can also be
negative, however, if there are negative productivity spillovers from multinationals.
As with productivity, identifying wage spillovers usually involves estimating the
determinants of wages in domestic firms and including a measure of foreign pres-
ence (multinationals' share of total employment) in the industry as a covariate.
Aitken and others (1996) use industry-level data for manufacturing industries for
Mexico (1984-90), Venezuela (1977-89), and the United States (1987). They find
positive effects in the United States, but negative effects in Mexico and Venezuela. As
with productivity spillovers, the result for the United States should be treated with
caution because the study uses cross-sectional data. Lipsey and Sjoholm (2001)
study the same effect for the Indonesian manufacturing sector using plant-level
cross-sectional data for 1996 and find that higher foreign presence in a sector leads
to higher wages in domestic firms in the same sector. Girma and others (2001),
using firm-level panel data for UK manufacturing for 1991-96, find no effect on
average of multinationals in a sector on the wage level in domestic firms but weak
evidence of a negative effect on wage growth.
Export Spillovers
A third strand in the literature focuses on whether multinationals spread their
knowledge of global markets to domestic firms, thus enabling them to become more
184                                      Thie World Banik Research Observer, vol. 19, no. 2 (Fall 2004)



successful exporters. Domestic firms can be affected through three primary chan-
nels. First, if multinationals have better access to information about foreign markets,
this can spill over through their export activities. Second, domestic firms can learn
the multinationals' superior production or management techniques through obser-
vation (demonstration effects), enabling the domestic firms to compete more success-
fully in export markets. Third, competition with multinational firms at home and in
foreign markets can induce domestic firms to improve their export performance.
Several studies have examined export spillovers (table 5). Aitken and others
(19 9 7) estimate a probit model using export activity by multinationals in the indus-
try and region as a proxy for export information externalities. Using plant-level
cross-section data for Mexican manufacturing industries for 1986 and 1989, they
find that export activities by multinational firms in a sector positively affect the prob-
ability of a firm in the same sector, foreign or domestic, being an exporter.
Using firm-level panel data for the United Kingdom for 1992-96. Greenaway
and others (forthcoming) also investigate whether spillovers affect a firm's proba-
bility of exporting but extend the analysis to examine what affects a firm's export
ratio. In a two-step Heckman selection model, they first estimate the probability of
exporting and then estimate the factors that affect a firm's export ratio. They
include three measures of multinational presence to capture the three spillover
channels.
Their results suggest that multinational firms' exports have a positive effect on
domestic firms' probability of exporting but do not affect their export ratio. They also
find that R&D spillovers from multinationals to domestic firms and the presence of
multinational firms in the sector positively affect the decision to export and the
export ratio. Thus, export information externalities appear to matter only for the
decision of whether to export. This is not surprising because these externalities can
be expected to aid domestic firms in overcoming the sunk costs of exporting, which
should affect their probability of exporting but not their export ratio.
Table 5. Papers on Export Spillovers
Auithlor(s)              Couintrj         Period   Data         zAggregationi level' Resoiltb
1 Aitken and others(1997)  Mexico       1986/1989  Cross-sectional Micro          +
2 Kokko and others (2001)  VTruguay        1998    Cross-sectional Micro          ?
3 Greenaway and others  United Kingdom  1992-96    Panel         Micro            +
(forthcoming)
4 Banga (200 3)          India          1994-2000  Panel         Micro/industry   +
5 Barrios and others (2(00)3) Spain     1990-98    Panel         Micro
"Micro data are at the lirm, plant, or establishment level.
+ indicates positive and statistically significant, - indicates negative and statistically significant. and ? indi-
cates mixed or statistically insignificant results on the foreign presence variable for the aggregate sample.
Ilaoger G(iVrg and David Greenaiva'y                                               185



Barrios and others (2003) also focus on export information externalities and on
demonstration effects through R&D spillovers. Using firm-level panel data for Spanish
manufacturing for 1990-98, they estimate a probit model to explain why firms
export and a tobit model to estimate what determines the firm's export ratio. They
find no evidence that either R&D activity or export activity by multinationals in a sec-
tor affects the probability that domestic firms export, although they find spillovers
from both types of activity on other foreign-owned firms. The tobit estimations, how-
ever, find evidence for positive effects of multinationals' R&D activity on domestic
firms' export ratios, but no spillovers from multinationals' export activities on
domestic firms. The export ratios of other foreign firms again benefit from both types
of spillovers. In an extension Barrios and colleagues (2003) discover that R&D spill-
overs increase domestic firms' exports only to other developed economies, which are
generally markets with a superior technological capability.
Kokko and others (2001) investigate the decision to export by domestic firms in
Uruguay using cross-sectional firm-level data for 1998. They include only a simple
measure of the presence of multinational firms-their output share in an industry,
not their export activity-and it is thus not clear which channel leads to spillovers.
However, they distinguish between the presence of multinational firms in import-
substituting and export-orientated industries and find evidence only for spillovers
from export-oriented multinationals. This suggests that the trade regime within
which multinationals operate may determine their potential for generating positive
export spillovers.
Summary
An extensive array of empirical studies have searched for productivity spillovers
from multinationals of various forms. Much of this work has relied on cross-section
methods. With the growing availability of longitudinal data at the plant and firm
level, however, more analysts are using panel techniques. This is a helpful develop-
ment for two reasons: first because the plant or firm is the most appropriate level of
scrutiny, and second because there are methodological shortcomings associated
with applying cross-section techniques.
Much of the work fails to find positive horizontal spillovers on aggregate,
with some studies reporting negative effects of multinational presence on
domestic productivity. Evidence on horizontal effects on wages and export spill-
overs is also mixed. However, studies that further disaggregate data into more
homogeneous groups of firms or plants find more encouraging results. In par-
ticular, there is evidence that the absorptive capacity of domestic firms and
their geographic proximity to multinationals are important determinants of
whether domestic firms benefit from FDI in the same sector. This suggests that
186                                 Thze World Bank Research Observer, vol. 1 9, no. 2 (Ftall 2004)



spillovers may not affect all firms equally but may benefit only firms with high
levels of absorptive capacity or close proximity to multinationals. Furthermore,
the few studies that have looked at the potential for vertical (interindustry)
spillovers find evidence suggesting that vertical spillovers may be a more impor-
tant channel for knowledge externalities than horizontal spillovers (see also
Moran 2001).
Is There a Role for Policy?
In general, most governments see I'o as having greater potential to improve total
factor productivity than an equivalent amount of domestic investment. This
would be taken as axiomatic in developing and transition economies and,
depending on the origin of the multinational firm, in at least some developed
economies. Add to this the potential spillovers from multinational firms to domes-
tic firms that are believed to raise their productivity, yielding a second growth
bonus, and it becomes clear why attracting inward investment figures promi-
nently in the policy priorities of so many governments. This leads naturally to
three questions: Can active policy intervention influence the level and composi-
tion of inward investment? Can particular policies maximize the potential for
spillovers, both by encouraging multirlationals to transfer technologies and by
improving the absorptive capacity of domestic firms? Do targeted policies yield net
benefits?
Policy and the Level and Composition of FDI
The role of policy in influencing the level and composition of FDI has been reviewed
extensively (see, for example, Balasubramanyam and Salisu 2001: Pain 2000;
Hanson 2001). Most work relates to developing economies because policy interven-
tions there have in general been more active, though a growing volume of research
relates to industrial countries, where most F1IX originates.
Several key points emerge from this work.
. Trade policy is relevant. In general, economies with more open trade regimes
have done better at attracting FDi and benefiting from it than countries with
inward-oriented regimes (Balasubramanyam and others 1996).
. Although there is some evidence that investment incentives can affect the loca-
tion choice of multinationals, the effect appears to be small (Coughlin and others
1991; Head and others 2000). Head and colleagues (2000) even argue that
competition between potential host governments may render incentives ineffec-
tive as they offset each other. Also, this form of competition for FDI may have
Holger G&ry ,w,l David Greeniawi'ay                                187



affected the distribution of incentives and is likely to have redistributed income
from host countries to multinational firms (Haaland and Wooton 1999).
. Trade-related investment measures (TRIMS), such as local content requirements
and minimum export requirements, are often introduced to recapture some of
the rents that accrue to multinational firms. Although these measures can have
positive welfare effects on the host country, the evidence does not point to major
effects on levels of inward investment in developing economies (Greenaway
1992).
. The quality of local infrastructure is vitally important, in particular communi-
cation and transportation facilities, both in attracting initial investments and in
sustaining clusters (Coughlin and others 1991; Coughlin and Segev 2000).
. The availability of relatively skilled labor is an important magnet (Coughlin and
Segev 2000) as well as a key driver of agglomeration (Ottaviano and Puga 1998).
It has also been argued that host countries are more likely to benefit from spill-
overs if they have a large supply of skilled labor (Keller 1996) and if domestic
firms have a high level of technological capacity (Glass and Saggi 1998).
Overall the evidence seems to suggest that interventions should strive largely to
provide a supportive economic environment. More specifically, this flags a role for
education and training policies aimed at upgrading general skills, technology poli-
cies aimed at developing clusters, and public investment policies aimed at develop-
ing efficient and reliable transportation and communication networks.
Policy and Spillovers
The evidence on spillovers reported here is mixed at best. There are no clear results
that domestic firms always and unambiguously gain from the presence of multi-
national firms. Several factors could be at play. Under the optimistic view that spill-
overs occur but measurement instruments are not fine enough to identify them, the
question is whether governments can implement policies to maximize the prospects
for extracting benefits from multinational firms. General policies-designed to
change the environment within which multinationals operate-include industrial
policy, infrastructure development, trade policy, exchange rate policy, and so on.
There is evidence to suggest that such policies are related to the overall level of
inward investment into an economy over a period of time. General policies may turn
out to be the most effective means of boosting the probability of positive spillovers. If,
for example, absorptive capacity is the critical driver, education and training policy
is likely to be key to facilitating spillovers.
As for specific policies, many TRIMS are targeted at encouraging spillovers (table 6).
Local content requirements, which have been widely used, are intended to raise the
share of local value added in subsidiary production and in the process encourage
188                                 Thie World Banik Research Observer. vol. 19. nio. 2 (Fall 2004)



Table 6. TrRimS Targeted at Spillovers
fo2Strlo17erit                 .ll .  i.  I,
Input TRIMs
Local content requirements     Specify that some proportion of value added or intermediate inputs is
locally sourced.
Local equity participation     Specifies that some proportion of the equity must be held locally.
Local hiring targets           Ensure specified employment targets are hit.
Expatriate quotas              Specify a maximum number of expatriate staff.
National participation         Specifies that certain staff must be nationals or sets a schedule for the
in management                  indigenisation' of the management.
R&D requirements               Commit multinationals to investment in research and development.
Technology transfer            Commits multinationals to local use of specified foreign technology.
Output TRIMS
Export controls                Specify that certain products may not be exported.
Licensing requirements         Oblige the investor to license production of output in the host country.
Technology transfer            Commits multinationals to a specified embodied technology.
Soiirce: Derived from Greenaway 1992 .
upstream development, with the intention of stimulating interindustry spillovers.
Because one can argue that spillovers are more likely if there is some local owner-
ship, local equity requirements are geared to that end. Local hiring targets and expa-
triate quotas are intended to raise the share of employment accounted for locally,
with a view to encouraging spillovers through the transfer of human capital. R&D
and technology transfer requirements are intended to make multinational firms
commit to some minimum level of R&D expenditures or transfer technology to local
firms. 1()
The economics of TRIMS is not straightforward. In general they are second-best
measures. For example, analytically a local content requirement is equivalent to an
input tariff, though less efficient. What little work has so far been completed on TRIMS
has failed to establish a direct link between them and the transfer of useful technolo-
gies (Blomstrom and others 1994: Greenaway 1992). This appears to be because
many of the measures are difficult to specify precisely and to monitor. But it is also
because the more general policies referred to are in practice rather more important.
ConcLusions
FDI is a key driver of economic growth and development. Most governments consider
attracting FDI a priority, particularly in developing and transition economies. It is
given this emphasis not just because it boosts capital formation but also because it
can enhance the quality of the capital stock. The reason is that multinationals are
Holger G6rg and Daivid Green?awayJ                                                   189



assumed to bring with them best practice or, as a minimum, better practice tech-
nology and management. Moreover, it is possible (even probable) that a given multi-
national firm will not be able to protect its superior technology or management fully
to prevent some elements from being absorbed by indigenous firms. If spillovers
occur, they provide an external benefit from FDI, one that governments are hoping to
secure when they offer inducements.
Theory points to reasons why spillovers might arise, but finding robust empirical
evidence to support their existence is more difficult. This could indicate that the bene-
fits are in fact illusory, in that multinational firms are effective in protecting their
assets. But it could as well be that researchers are looking in the wrong place and
with the wrong lens. Many studies focus on the industry rather than the firm or
plant. The growing availability of survey data at the firm and plant level makes such
study increasingly feasible. Most studies use cross-section data when panel data are
required for proper analysis.
Because research on disaggregated data with both cross-sectional and longitudi-
nal variation is still limited, the message is clear: More systematic research is needed.
More discriminating work is also required, analysis that probes what really mat-
ters-form of entry (greenfield or acquisition), ownership characteristics, corporate
governance, absorptive capacity of domestic firms, and so on.
The consensus in the policy literature is also clear: General policies aimed at alter-
ing the fundamentals are more important than specific policies aimed at attracting
particular investments. Such specific policies seem to affect primarily the distri-
bution of rents. Governments compete in offering investment incentives and in the
process create rents for multinational firms. They then use (at least some) TRIMS to try
to reclaim some of those rents.
Both econometric evidence and survey and case study work suggest that the
characteristics of the economic environment are generally much more important:
infrastructure, local labor market conditions, reliability of communications systems,
and so on, as well as the overall macroeconomic and trade policy climate. That, of
course, does not mean that selective interventions will cease to be extensively
deployed. Governments will no doubt continue to see opportunities for targeted
measures, and multinational firms will stand ready to accept them. This is another
area for future work. Very little is known about the comparative impact of different
instruments.
Notes
Holger Gorg is lecturer in the School of Economics and a research fellow at the Leverhulme Centre for
Research on Globalisation and Economic Policy, University of Nottingham. He is also affiliated with the
German Institute of Economic Research, Berlin: his e-mail address is holger.gorg@nottingham.ac.uk.
190                                   The Wlorld Bank Research Observer. vol. 19, no. 2 (Fall 2004)



David Greenway is professor of economics and director of the Leverhulme Centre for Research on Glo-
balisation and Economic Policy, School of Economics. University of Nottingham; his e-mail address is
david.greenawayanottingham.ac.uk. Earlier versions of the article were presented at a [TN Economic
Commission for Europe/European Bank for Reconstruction and Development meeting in Geneva, a [TN
Economic Commission for Latin America and the Caribbean conference in Mexico City, an Inter-
American Development Bank/World Bank conference in Washington, D.C., and a Centre for Economic
Policy Research meeting in Turin. The authors wish to thank in particular Wolfgang Keller, Bob
Lipsey, Mauricio Mesquita Moreira, Andres Rodriguez-Clare. and Ian Wooton for helpful comments. In
addition, reports from four anonymous referees and the managing editors of the journal were
extremely helpful. Financial support from the Leverhulme Trust under Programme Grant Fl 14/BF and
the European Commission under grant no. SERD-2002-00077 is gratefully acknowledged.
1. For example, Larrain and others (2000) conclude that the location of Intel in Costa Rica has had
positive effects on the local economy. Hanson (2t)01) argues that there is little evidence for spillovers
from Intel on domestic firms. He also argues that Ford and General Motors plants in Brazil have failed to
show the expected spillover benefits.
2. A related body of literature examines the macro effect of inward Fl) on growth in the framework of'
cross-country growth regressions. See, for example, Balasubramanyam and others (1996), Borensztein
and others (1998), and Alfaro and others (forthcoming) for recent evidence. DeMello (1997) provides a
review of that literature.
3. Castellani and Zanfei (2002a) argue for the use of the absolute level of foreign activity in the sec-
tor rather than foreign activity as a proportion of total activity. because changes of the same magnitude
in foreign and aggregate activities within a sector would have no effect on the dependent variable.
Although this is an interesting econometric argument, it is not clear what the economic rationale for
using absolute rather than relative FrI penetration would be.
4. Interpretation of this coefficient of course hinges on the assumption that the FDI variable does not
merely pick up the effect of other correlated factors on productivity-it assumes that a full vector of
productivity-augmenting activities is included in the empirical model. Although this may be problem-
atic in some of the studies reviewed herein, it is beyond the scope of this article to discuss this in detail.
The article therefore assumes that the estimated FDI coefficient adequately reflects spillovers.
5. Dunne and others (1989) note this for the case of measuring the growth performance of manu-
facturing plants in the United States.
6. The magnitude of the coefficients, which indicates the strengths of the spillovers, also differs
across studies. Gorg and Strobl (2001) attempt to explain the differences in magnitude in a meta-
regression analysis. using characteristics of the studies (such as data, variables used, countries covered)
as explanatory variables.
7. By contrast. Sjoholm (1999b), using cross-sectional data for Indonesian manufacturing firms,
finds that productivity spillovers from foreign to domestic firms are larger the larger the technology gap
(also defined in terms of differences in labor productivity) and the higher the degree of competition in
the industry.
8. Related theoretical models by Rodriguez-Clare (1996) and Markusen and Venables (1999) show
that multinationals can have positive effects on the development of domestic firms through vertical
input-output linkages. Gorg and Strobl (2002a, b) show empirical evidence that the presence of multi-
national firms has fostered the entry and development of domestic firms in the Republic of Ireland.
Alfaro and Rodriguez-Clare (forthcoming) point out that the Rodriguez-Clare (1996) model makes a
case for expecting horizontal (intraindustry) rather than vertical spillovers.
9. A related issue is whether FDI contributes to the shift in labor demand toward skilled labor in the
host country. See, for example, Feenstra and Hanson (1997) for empirical analysis for Mexico, Figini
and Gorg (1999) for Ireland, Blonigen and Slaughter (2001) for the United States, and Taylor and
Driffield (forthcoming) for the tJnited Kingdom.
10. The Uruguay Round obligated countries to phase out certain TRIMS (those that violate Articles III
and XI of the General Agreement on Trade and Tarrifs), with local content requirements being the most
prominent. The key issue here, however, is whether they work.
Holger G(,rg anid David Greenawazj                                                   191



References
Aitken, Brian J.. and Ann E. Harrison. 1999. "Do Domestic Firms Benefit from Direct Foreign Invest-
ment? Evidence from Venezuela." Armerican Economic Review 89(3):605-18.
Aitken, Brian, Gordon H. Hanson, and Ann E. Harrison. 1997. "Spillovers, Foreign Investment, and
Export Behavior." Journal of International Economics 43(1-2):103-32.
Aitken, Brian, Ann Harrison, and Robert E. Lipsey. 1996. "Wages and Foreign Ownership: A Com-
parative Study of Mexico, Venezuela, and the IJnited States." Journal of Internzatiornal Economnics
40(3-4):345-71.
Alfaro, Laura, and Andres Rodriguez-Clare. Forthcoming. "Multinationals and Linkages: An Empirical
Investigation." Economrlia.
Alfaro, Laura, Areendam Chanda, Sebnem Kalemli-Ozcan, and Selin Sayek. Forthcoming. "FDI and
Economic Growth: The Role of Local Financial Markets." Journal of International Economnics.
Audretsch, David B. 1998. "Agglomeration and the Location of Innovative Activity." Oxford Review of
Economzic Policy 14(2):18-29.
Balasubramanyam, V. N., and Mohammed Salisu. 2001. "Foreign Direct Investment and Globalisation."
In S. Lahiri, ed., Regionalism and Globalisation. London: Routledge.
Balasubramanyam, V. N., Mohammed Salisu, and David Sapsford. 1996. "Foreign Direct Investment
and Growth in EP and is Countries." EconomicJournial 106(434):92-105.
Baldwin, Robert. 1969 "The Case against Infant Industry Protection." Journal of Political Econorny
77:295-305.
Banga, Rashmi. 2003. "The Differential Impact of Japanese and U.S. Foreign Direct Investments on
Exports of Indian Manufacturing." Working paper, Delhi School of Economics, India.
Barrios, Salvador, and Eric Strobl. 2002. "Foreign Direct Investment and Productivity Spillovers:
Evidence from the Spanish Experience." Weltiwirtschaftliches Archiv 138:459-81.
Barrios, Salvador, Holger Gorg, and Eric Strobl. 2003. "Explaining Firms' Export Behaviour: R&D,
Spillovers and the Destination Market." Oxford Bulletini of Economnics and Statistics 65(4):475-96.
Bernard, Andrew. and J. Bradford Jensen. 1999. "Exceptional Exporter Performance: Cause, Effect or
Both?" Journal of Interniational Economics 47(1): 1-26.
Bernard, Andrew, and Joachim Wagner. 1997. "Exports and Success in German Manufacturing."
W1, lis-i.,[ kilt,,ie b. Arclhiv 133:1 34-57.
Blalock, Garrick, and Paul Gertler. 200)3. "Technology from Foreign Direct Investment and Welfare
Gains through the Supply Chain." Working paper, Cornell University, Ithaca, N.Y.
Blomstrom, Magnus. 1986. "Foreign Investment and Productive Efficiency: The Case of Mexico."
Journial of Indu(strial Economics 355(1):9 7-112.
Blomstrom, Magnus, and Ari Kokko. 1998. "Multinational Corporations and Spillovers." Journal of
Economic Surveys 12(3):247-77.
Blomstr6m, Magnus, and Hakan Persson. 1983. "Foreign Investment and Spillover Efficiency in an
Underdeveloped Economy: Evidence from the Mexican Manufacturing Industry." World Development
11(6):493-50)1.
Blomstrbm, Magnus, and Fredrik Sjoholm. 1999. "Technology Transfer and Spillovers: Does Local
ParticipationwithMultinationals Matter?" Europeani Econzomizic Review43(4-6):915-23.
Blomstrbm, Magnus, and Edward N. Wolff. 1994. "Multinational Corporations and Productive
Convergence in Mexico." In William J. Baumol, Richard R. Nelson, and Edward N. Wolff, eds., Con-
vergence of Productivity: Cross National Studies and Historical Evidence. Oxford: Oxford University
Press.
192                                        The World Bank Research Observer, vol. 19. no. 2 (Fall 2004)



Blomstr6m, Magnus. Ari Kokko. and Mario Zejan. 1994. "Host Country Competition, Labor Skills, and
Technology Transfer by Multinationals." Weltwirtsclhaftliches Archiv 130:521-33.
Blonigen, Bruce A., and Matthew J. Slaughter. 2001. "Foreign-Affiliate Activity and U.S. Skill Upgrading."
Review of Econornics anid Statistics 83(2):362-76.
Borensztein. E., J. De Gregorio. and J.-W. Lee. 1998. "How Does Foreign Direct Investment Affect
Economic Growth?" Journal of International Economnics 45(1): 115-35.
Bosco, Maria Giovanna. 2001. "Does FDi Contribute to Technological Spillovers and Growth? A Panel
Data Analysis of Hungarian Firms" Transnational Corporations 10(l) :4 3-68.
Brainard, S. Lael. 1997. "An Empirical Assessment of the Proximity-Concentration Trade-Off between
Multinational Sales and Trade." American Economic Review 87(4):520-44.
Castellani, Davide. and Antonello Zanfei. 2002a. "Multinational Companies and Productivity Spillovers:
Is there a Specification Error? " Working paper, University of Urbino, Italy.
- 2002b. "Technology Gaps, Absorptive Capacity and the Impact of Inward Investments on
Productivity of European Firms." Working paper, University of Urbino, Italy.
Caves, Richard E. 1974. "Multinational Firms, Competition, and Productivity in Host-Country Markets."
Econiomrlica 41(162):176-93.
-   1996. Multiniational Enterprise and Economic Analysis, 2nd ed. Cambridge: Cambridge Univer-
sity Press.
Chuang, Yih-Chyi, and Chi-Mei Lin. 1999. "Foreign Direct Investment. R&D and Spillover Efficiency:
Evidence from Taiwan's Manufacturing Firms." Journal of Development Studies 35(4): 117-37.
Clerides, S. K., S. Lach. and J. R. Tybout. 1998. "Is Learning by Exporting Important? Micro-Dynamic
Evidence from Colombia, Mexico. and Morocco." Quarterly Journzal of Econornics 113(3):903-48.
Coughlin, Cletus C.. and Eran Segev. 2000. "Location Determinants of New Foreign-Owned Manufac-
turing Plants." Journal of Regional Science 40(2): 323-51.
Coughlin, Cletus C., Joseph V. Terza, and Vachira Arromdee. 1991. "State Characteristics and the
Location of Foreign Direct Investment within the United States." Review of Economics and Statistics
73(4):675-8 3.
Damijan, Joze P., Boris Majcen, Mark Knell, and Matija Rojec. 2001. "The Role of FDI, Absorptive
Capacity and Trade in Transferring Technology to Transition Countries: Evidence from Firm Panel
Data for Eight Transition Countries." UN Economic Commission for Europe, Geneva.
Das, Sanghamitra. 1987. "Externalitites, and Technology Transfer through Multinational Corpor-
ations: A Theoretical Analysis." Journal of International Economics 22(1-2):171-82.
Deardorff, Alan. 2003. "What Might Globalisation's Critics Believe?" World Economy 26(5):639-58.
Delgado. M., J. Farinias. and S. Ruano. 2002. "Firm Productivity and Export Markets: A Non-Parametric
Approach." Journal of International Econonmics 57(2): 397-422.
DeMello, Luiz R. 1997. "Foreign Direct Investment in Developing Countries and Growth: A Selective
Survey." Journal of Development Studies 34(1): 1-34.
Dimelis, Sophia, and Helen Louri. 2002. "Foreign Direct Investment and Efficiency Benefits: A Condi-
tional Quantile Analysis." Oxford Econiomnic Papers 54:449-69.
Djankov, Simeon. and Bernard Hoekman. 2000. "Foreign Investment and Productivity Growth in
Czech Enterprises." World Bank Econornic Review 14(l):49-64.
Driffield. Nigel. 2001. "The Impact on Domestic Productivity of Inward Investment in the UK."
Manchester School 69(1):1 t) 3-19.
Driffield, Nigel, and Sourafel Girma. 2003. "Regional Foreign Direct Investment and Wage Spillovers:
Plant Level Evidence from the Electronics Industry." Oxford Bulletin of Economics and Statistics:
65(4):453-74.
Holger- Gorg and Dav'id Greena way                                                     193



Driffield, Nigel, Max Munday, and Annette Roberts. 2002. "Foreign Direct Investment, Transactions
Linkages, and the Performance of the Domestic Sector." International Journal of tlhe Economics of
Business 9(3):335-51.
Dunne, Timothy, Mark J. Roberts, and Larry Samuelson. 1989. "The Growth and Failure of U.S.
Manufacturing Plants." Quarterly Journial of Economnics 104(4):671-98.
Feenstra, Robert C.. and Gordon H. Hanson. 1997. "Foreign Direct Investment and Relative Wages:
Evidence from Mexico's Maquiladoras." Journal of Internzationial Econornics 42(3-4): 371-93.
Figini. Paolo. and Holger Gbrg. 1999. "Multinational Companies and Wage Inequality in the Host
Country: The Case of Ireland." Weltwirtschaftliclhes Arclyiv 135:594-612.
Findlay, Ronald. 1978. "Relative Backwardness, Direct Foreign Investment, and the Transfer of
Technology: A Simple Dynamic Model." Ouarterly Journial of Econzomllics 92(1):1-16.
Fosfuri. Andrea, Massimo Motta, and Thomas Ronde. 2001. "Foreign Direct Invesment and Spillovers
through Workers' Mobility." Journzal of Interniationzal Economlics 5 3(1):205-22.
Girma, Sourafel. 2002. "Geographic Proximity. Absorptive Capacity and Productivity Spillovers from
1'DI: A Threshold Regression Analysis." GiP Research Paper 02//06. University of Nottingham, UK.
Girma, Sourafel, and Holger Gbrg. 2002. "Foreign Direct Investment, Spillovers and Absorptive
Capacity: Evidence from Quantile Regressions." Geh Research Paper 02/14. University of
Nottingham, UK.
Girma, Sourafel, and Katharine Wakelin. 2002. "Are There Regional Spillovers from IDi in the UK?" In
David Greenaway. Richard Upward, and Katharine Wakelin, eds., Trade, lnvestment, Migration and
Labour Markets. Basingstoke, UK: Macmillan.
Girma, Sourafel, David Greenaway, and Katharine Wakelin. 2001. "Who Benefits from Foreign Direct
Investment in the UK?" Scottish Journal of Political Economrly 48(2):119-33.
Girma. Sourafel. Holger Gorg, and Mauro Pisu. 2003. "The Role of Exports and Foreign Linkages for FDI
Productivity Spillovers." Working paper, University of Nottingham, United Kingdom.
Girma, Sourafel, David Greenaway, and Richard Kneller. Forthcoming. "Entry to Export Markets
and Productivity: A Microeconometric Analysis of Matched Firms." Review of Internzational
Economics.
Glass, Amy. and Kamal Saggi. 1998. "International Technology Transfer and the Technology Gap."
Journal of Developmnent Economnics 55(2): 369-98.
. 2002. "Multinational Firms and Technology Transfer." Scandinavian Journal of Econorics
104(4):495-514.
Globerman, Steven. 1979. "Foreign Direct Investment and 'Spillover' Efficiency Benefits in Canadian
Manufacturing Industries." Canadian Journal of Economtiics 12(4):42-56.
Gorg, Holger. 2002. "Fancy a Stay at the 'Hotel California'. Foreign Direct Investment, Taxation and
Exit Costs." GEI' Research Paper 02/28. University of Nottingham. tJK.
Gorg, Holger, and Eric Strobl. 2001. "Multinational Companies and Productivity Spillovers: A Meta-
Analysis." Economnic Journial 111(473):F72 3-39.
.2002a. "Multinational Companies and Entrant Start-up Size: Evidence from Quantile Regres-
sions." Review of Industrial Organizationi 20)(1): 15-31.
20)02b. "Multinational Companies and Indigenous Development: An Empirical Analysis."
European Economic Review 46(7):1305-22.
.20)0)2c. "Spillovers from Foreign Firms through Worker Mobility: An Empirical Investigation."
GEP Research Paper 02/13. University of Nottingham, UK.
-   2003. "Multinational Companies, Technology Spillovers and Plant Survival." Scandinavian
Journal ofEconomnics 105(4):581-95.
194                                        The Vl'orld Banlk Resear-ch Observer, vol. 19, rio. 2 (Fall 2004)



Gorg, Holger, Eric Strobl, and Frank Walsh. 2003. "Why Do Foreign Firms Pay More? The Role of
On-the-job Training." GEP Research Paper 02/15. University of Nottingham, UK.
Greenaway, David. 1992. "Trade Related Investment Measures and Development Strategy." Kylklos
45(2):1 39-60.
Greenaway, David. Nuno Sousa, and Katharine Wakelin. Forthcoming. "Do Domestic Firms Learn to
Export from Multinationals?" European Journal of Political Economy.
Haacker. Marcus. 1999. "Spillovers from Foreign Direct Investment through Labour Turnover: The
Supply of Management Skills," cEP Discussion Paper. London School of Economics.
Haaland, Jan I.. and Ian Wooton. 1999. "International Competition for Multinational Investment."
Scandinviaan Journal of Economrlics 101(4):631-49.
Haddad, Mona. and Ann Harrison. 1993. "Are There Positive Spillovers from Direct Foreign Investment?
Evidence from Panel Data for Morocco." Journal oJ Developmernt Economnics 42(1):51-74.
Hanson, Gordon H. 2001. "Should Countries Promote Foreign Direct Investment?" G-24 Discussion
Paper 9. llNCTAD), Geneva.
Harris, Richard, and Catherine Robinson. 2004. "Productivity Impacts and Spillovers from Foreign
Ownership in the United Kingdom. " National Institute Econornic Review 18 7:58-75.
Haskel, Jonathan E., Sonia C. Pereira, and Matthew J. Slaughter. 2002. "Does Inward Foreign Direct
Investment boost the Productivity of Domestic Firms?" NBER Working Paper 8724. Cambridge.
Mass.
Head, Keith. 1998. "Comment on 'Comparing Wages, Skills, and Productivity between Domestically
and Foreign-Owned Manufacturing Establishments in the tJnited States.'" In R. Baldwin, R. Lipsey.
and J. D. Richardson. eds., Geograplhy and Ownership tis Bases for Economic Accounting. Chicago:
Chicago University Press.
Head, Keith C.. John C. Ries, and Deborah L. Swenson. 2000. "Attracting Foreign Manufacturing:
Investment Promotion and Agglomeration." Regional Scienceaand Urban Economics 29(2):197-218.
Kathuria, Vinish. 2000. "Productivity Spillovers from Technology Transfer to Indian Manufacturing
Firms." Journal of InlternationzalDevelopmrlenit 12(2):343-69.
Keller, Wolfgang. 1996 "Absorptive Capacity: tJnderstanding the Creation and Acquisition of Tech-
nology in Development." Journalof Developrnerlt Economnics49:199-227.
.2001. "International Technology Diffusion." NBER Working Paper 8573. Cambridge, Mass.
Keller, Wolfgang, and Stephen Yeaple. 2003. "Multinational Enterprises, International Trade, and
Productivity Growth: Firm-Level Evidence from the United States." GEP Research Paper 03/0)3.
University of Nottingham. IJK.
Kinoshita, Yuko. 2001. "R&D and Technology Spillovers through 'DI: Innovation and Absorptive Capacity.
CEPR Discussion Paper DP2775. Centre for Economic Policy Research, London.
Kokko. Ari. 1994. "Technology. Market Characteristics, and Spillovers." Journal of Developmnent Eco-
niomtics 43(2):279-93.
.1996. "Productivity Spillovers from Competition between Local Firms and Foreign Affiliates."
Journal of Internzationial Developmnent 8(4):517-30.
Kokko, Ari, Ruben Tansini, and Mario C. Zejan. 1996. "Local Technological Capability and Productiv-
ity Spillovers from FDI in the Uruguayan Manufacturing Sector." Journal of Development Studies
32 :602-11.
Kokko, Ari, Mario Zejan, and Ruben Tansini. 2001. "Trade Regimes and Spillover Effects of FDI:
Evidence from Uruguay." Weltwirtscluaftlic hes Arcliiv 137:124-49.
Konings. Jozef. 2001. "The Effects of Foreign Direct Investment on Domestic Firms: Evidence from Firm
Level Panel Data in Emerging Economies." Economrlics of Transition 9(31:619-33.
Hol,ger G6ra arnd David Greenavway                                                  195



Kugler, Maurice. 2001. "The Diffusion of Externalities from Foreign Direct Investment: The Sectoral
Pattern of Technological Spillovers." Working paper, University of Southampton, UKm.
Larrain, Felipe B., Luis F. Lopez-Calva, and Andres Rodriguez-Clare. 2000. "Intel: A Case Study of
Foreign Direct Investment in Central America." Center for International Development Working
Paper 58. Harvard University, Cambridge, Mass.
Li. Xiaoying, Xiaming Liu, and David Parker. 2001. "Foreign Direct Investment and Productivity
Spillovers in the Chinese Manufacturing Sector." Economic Systems 25(4): 305-21.
Lipsey, Robert E. 2002. "Home and Host Country Effects of FDI." NBER Working Paper 9293. Cambridge,
Mass.
Lipsey, Robert E., and Fredrik Sjoholm. 2001. "Foreign Direct Investment and Wages in Indonesian
Manufacturing." NBER Working Paper 8299. Cambridge, Mass.
Liu, Xiaming, Pamela Siler, Chengqi Wang, and Yingqi Wei. 2000. "Productivity Spillovers from
Foreign Direct Investment: Evidence from UK Industry Level Panel Data." Journal of International
Business Studies 31(3):407-25.
L6pez-C6rdova, J. Ernesto. 2002. "NAFTA and Mexico's Manufacturing Productivity: An Empirical
Investigation Using Micro-Level Data." Working paper, Inter-American Development Bank,
Washington, D.C.
Markusen, James R. 1995. "The Boundaries of Multinational Enterprises and the Theory of Inter-
national Trade." Journal of Economic Perspectives 9(2): 169-89.
Markusen, James R., and Anthony 1. Venables. 1999. "Foreign Direct Investment as a Catalyst for
Industrial Development." European Economnic Review 43(2):335-56.
McCullogh, Neil, Alan Witners, and Xavier Cirera. 2001. "Trade Liberalisation and Poverty: A Hand-
book." Working paper, Centre for Economic Policy Research. London.
Moran, Theodore. 200)1. Parental Supervision: The New Paradigm for Foreign Direct Investment and
Development. Washington, D.C.: Institute for International Economics.
Ottaviano, Gianmarco I. P., and Diego Puga. 1999. "Agglomeration in the Global Economy: A Survey
of the New Economic Geography." World Economy 21(6): 707-31.
Pain, Nigel, ed. 2000. Inward Investment Technological Chanige and Growth, London: Palgrave.
Rodriguez-Clare, Andres. 1996. "Multinationals, Linkages, and Economic Development." American
Economic Review 86(4):852-73.
Ruane, Frances, and Ali Ugur. 2002. "Foreign Direct Investment and Productivity Spillovers in Irish
Manufacturing Industry: Evidence from Firm Level Panel Data." Trinity Economic Papers 02/06.
Trinity College, Dublin.
Saggi, Kamal. 2002. "Trade, Foreign Direct Investment, and International Technology Transfer:
A Survey." World Bank Research Observer 17(2): 191-235.
Sj6holm, Fredrik. 1999a. "Technology Gap, Competition and Spillovers from Direct Foreign Invest-
ment: Evidence from Establishment Data." Journal of Development Studies 36(1): 53-73.
-    . 1999b. "Productivity Growth in Indonesia: The Role of Regional Characteristics and Direct
Foreign Investment." Economic Development and Cultural Change 47(3): 559-84.
Smarzynska-Javorcik, Beata K. Forthcoming. "Does Foreign Direct Investment Increase the Productiv-
ity of Domestic Firms? In Search of Spillovers through Backward Linkages." American Economnic
Review.
Taylor, Karl, and Nigel Driffield. Forthcoming. "Wage Inequality and the Role of Multinationals:
Evidence from UK Panel Data." Labour Econornics.
UN (United Nations). 1999. World Investment Report. Geneva: United Nations.
196                                       The World Bank Research Observer, vol. 19, no. 2 (Fall 2004)



Wang. Jian-Ye. and Magnus Blomstrom. 1992. "Foreign Investment and Technology Transfer: A Simple
Model." Europeato Econornic Review 36(1): 13 7-5 5.
Wheeler, David, and Ashoka Mody. 1992. "International Investment Location Decisions: The Case of
U.S. Firms." Jourral of Irnternational Economics 3 3(1-2): 5 7-76.
Zukowska-Gagelmann, Katarzyna. 2000. "Productivity Spillovers from Foreign Direct Investment in
Poland." Economic Systemns 24(3 ):22 3-56.
Holger Grqg and David Greenaw'auy                                                    197






Policies on Managing Risk
in AgricuLturaL Markets
Donald F. Larson * Jock R. Anderson * Panos Varangis
Over the past dozen years, policyrniakers have largely abandoned long-standing popular
approaches for addressing risk in agriculture without fully resolving the question of IIow
best to manage the negative consequences of volatile agricultural markets. The article
reviews the transition from past policies and describes current approaches that distingutish
between the trade-relatedfiscal consequences of commodity market volatility and the con-
sequences of price and production risks for vulnerable rural households and commu nities.
Currenit policies rely mnore heavily on markets, even tlhouglh marketsfor risk are incomplete
in numerous ways. The benefits and limitations of market-based instruments are examined
in tlhe context of risk managemenit strategies, and innovative approaclhes to extend the reach
of risk markets are discussed.
For decades governments have intervened to reduce risk in markets for internation-
ally traded commodities. Over the past dozen years, however, many of the policies
and supporting institutions used for that purpose were reevaluated and found to be
not only ineffective and unsustainable but impediments to growth. Most of these
approaches have now been abandoned, but governments and policymakers still seek
to understand how best to manage the negative consequences of volatile commodity
markets because traded commodities remain an important source of export earnings
for many developing economies and an important component of income and con-
sumption for the poor.
This article reviews the historical relationship between the work of applied
economists and policymakers and the commodity and risk market instruments
introduced in the 1980s. It explores how a growing body of analytical work con-
tributed to a change in thinking that moved the emphasis of policy from stabiliza-
tion to management of risks, and it surveys current thinking on managing rural
risks.
The World Bank Research Observer. vol. 19, no. 2.
© The International Bank for Reconstruction and Development / THE WORLD BANK 2004: all rights reserved.
doi:10.109 3/wbro/lkhO22                                            19:199-230



Efforts Aimed at Stabilization
By the 1980s unilateral and multilateral interventions in commodity markets to
achieve macroeconomic and commodity market stabilization were common. Inter-
national commodity organizations employed buffer stocks or managed trade, trying
to bring order to unstable commodity markets. The United States used support prices
and inventories to manage domestic prices. The European Union had a similar
scheme and also operated a set of commodity-specific exchange rates-"green
rates"-for trade among EU members. Marketing boards and stabilization stocks
and funds were common in both developed and developing economies. For develop-
ing economies, lines of credit were available at the International Monetary Fund
(IMF) and elsewhere to support these operations.
Both historical and practical considerations led to such interventions (Larson and
others 1998; Akiyama and others 2001). but a great many of the interventions-
especially the multilateral interventions-were also motivated by prevailing economic
thought.
At the close of World War II, policymakers focused on ways to prevent short-term
trade shocks from turning into the widespread economic decline that characterized
the Great Depression years. In an age of gold-standard fixed currencies, central
banks were often challenged to manage shocks in export earnings. At Bretton
Woods, Keynes (1943) proposed a world currency based on a price index of the 30
most traded commodities. With currencies linked to the index, commodity prices
and price-related swings in trade earnings would be stabilized largely automatically.
Although Keynes's ideas were not incorporated into the charters of the Bretton
Woods institutions, an alternative approach of linking lending to commodity volatil-
ity emerged. A succession of internationally backed compensatory financing
schemes were proposed in the years following the Bretton Woods conference,
including the 1953 Olano Proposal for a Mutual Insurance Scheme, the 1961
Development Insurance Fund, the 1962 Organization of American States Proposal,
and the 1963 Swedish and Brazilian proposals at the Committee for International
Commodity Trade meetings and the French Proposal for Market Organizations.
Also in 1963 the iMF began to offer compensatory financing to countries experienc-
ing unexpected temporary declines in export earnings. The program is based on net
export earnings. rather than a single set of commodities, thus taking advantage of any
natural portfolio effect that might arise from diversified exports and imports. With mod-
ifications, this program remains the primary instrument among the Bretton Woods
institutions for handling the effects of volatile commodity trade. Later, as part of the first
Lome Agreement in 1975, the European Union offered its own compensatory financing
schemes, STABEX and SYSMIN, to the African, Caribbean, and Pacific group of states.
In 1950 Prebisch (1950) and Singer (1950) independently offered the hypo-
thesis that the prices of primary commodity exports would fall relative to those of
200                                The World Bank Research Observer, vol. 19. no. 2 (Fall 2004)



manufactured imports because of differing elasticities of income and demand. Conse-
quently, the net barter terms of trade for commodity-producing developing econo-
mies would decline. This contradicted the long-standing notion of increasing
scarcity put forward by Malthus, Ricardo, and Hotelling. Soon thereafter, the two
ideas became entwined so that the "commodity dependency" problem was charac-
terized by declining terms of trade and volatile export earnings.' Generally, econo-
mists argued that instability of export earnings limited development through
adverse effects on income, inflation, savings, and investment. As a consequence,
international efforts were extended to key commodity markets as well. Later, Massell
(1969, 11970) integrated these ideas in a model for estimating the benefits of stabiliz-
ing commodity markets. More than a decade after that, Newbery and Stiglitz (1981,
p. 18) commented:
The predictions of the (Massell 1969) model are readily derived ... and are
quite precise: (i) producers gain and consumers lose from price stabilization
if the source of instability lies on the supply side: (ii) consumers gain and
producers lose from price stabilization if the source of instability lies on the
demand side; (iii) in both cases, gainers could afford to over-compensate the
losers, so there are net benefits from price stabilization.
Policies of the era were motivated by macroeconomic considerations, and analy-
sis was largely sectoral. Price levels, price stability, and spatial price differences were
the principal policy targets. The macroeconomic focus derived in part from the belief
that the domestic income effects or trade effects associated with volatile agricultural
and other commodity markets were an impediment to economic growth. The focus
was conditioned as well by the dominant view that agriculture's role in development
was as a supplier of "surplus" labor (Lewis 1955). The emphasis on price emerged in
part from the empirical observation that commodity prices were the primary source
of commodity income volatility (Johnson 1947; Massell 1969). There were practical
considerations as well, especially for developing economies, where few alternative
instruments are available for redirecting resources (Timmer 1989). For traded com-
modities governments could frequently gain some measure of control by imposing
quotas or tariffs while allowing domestic markets to function with fewer interven-
tions. Moreover, information gathering is simpler for prices than for supply, allow-
ing government managers to react more quickly.
A series of international interventions were structured for dealing with price vola-
tility, a concern for both developed and developing economies. Under UN auspices,
five international commodity agreements were signed by producing and consuming
countries: the International Sugar Agreement (1954), the Tin Agreement (1954),
the Coffee Agreement (1962). the Cocoa Agreement (1972), and the Natural Rub-
ber Agreement (1980). A 1968 UN Conference on Trade and Development proposal
resulted in a 1975 resolution calling for an Integrated Program for Commodities
Donaild F. Larsoni. Jock R. Aitiersoi. ind Panos taranqgis           201



covering 10 core commodities. The Common Fund for Commodities was established
to provide liquidity for the integrated program. Moving more rapidly, the IMF estab-
lished its Buffer Stock Financing Facility on June 25, 1969.
Governments began to run unilateral domestic stabilization programs as well.
The European Union, Japan, and the United States all operated systems that com-
bined minimum price guarantees with government-controlled inventories. Develop-
ing economies took up a variety of stabilization efforts as well (Knudsen and Nash
1990). These included buffer stock schemes (Bangladesh, India, Indonesia, Republic
of Korea, Mexico, the Philippines), buffer funds (C6te d'Ivoire, Republic of Korea,
Papua New Guinea), marketing boards with monopolies on trade (Ecuador, India,
Malaysia, most of Africa), and variable tariff schemes (Chile, Malaysia, Venezuela).
Although price stabilization was a common objective, multiple, often competing objec-
tives emerged as well, complicating administration of the programs and speeding
their demise.
In Asia and Africa policymakers faced additional challenges. Following a series
of devastating famines between 1967 and 1975 in Biafra (Nigeria), Bangladesh,
Ethiopia, and throughout the Sahel, self-sufficiency and food security became driv-
ing concerns for many developing economies. Controls on trade or domestic supply
were modified to meet multiple objectives. Large state enterprises, such as Conasupo
in Mexico. the Bulog in Indonesia, and grain boards throughout Africa, were
charged with managing domestic inventories of food.
Challenges to Stabilization Approaches
Although the argument that volatility reduced investment and subsequently led to
lower rates of growth in commodity-dependent countries was appealing, economists
found only weak empirical evidence of a direct link. UJsing data on Malaysian rubber
plantations, Caine (19 54) challenged the negative link between revenue instability
and investment. Later, MacBean (1966) also challenged the findings using cross-
country data, as did Knudsen and Parnes (1975). Deaton (1992) found for Africa
overall that the expansions in investments and income during periods of increasing
export prices were greater than the contractions during periods of falling prices.
Dawe (1996) calculated instability indices for a cross-section of countries based on
shares of exports and found that export instability was negatively associated with
growth and investment.
Increasingly, economists began to wonder whether the link between low growth
and volatile commodity markets had more to do with government mismanagement
than with private investment. Bourguignon and Gelb (1988) found evidence for
this view for Venezuela, as did Bevan and others (1990) for Nigeria and Indonesia.
Hausmann and Gavin (1996) argued that uncertainty over commodity-dependent
202                               Tlze lVorld Bank Research Observer, vol. 19. nio. 2 (Fall 2004)



government revenues (and fiscal management) had a cascading negative effect on
the economies of Latin America. Rodrik (1998) argued that the link between short-
term economic shocks and growth was determined by the ability of domestic institu-
tions to manage the political strife that such shocks initiate.
Some economists also argued that the benefits of stabilization were overstated.
Friedman (1954) stressed the importance of private savings rather than public stabi-
lization schemes in solving the "producer income problem." Newbery and Stiglitz
(1981) argued that the Massell model was flawed in ways that exaggerated the ben-
efits of stabilization. An important and frequent finding is that the welfare gains that
are possible from price stabilization are small (Anderson and others 1981; Myers
and Oehmke 1988: Wright 1988; Kannapiran 2000). Moreover, the practical
implementation of stabilization schemes raises many thorny problems for program
administrators (Anderson and others 1977b).
Early on, Bauer and Parish (1952) noted that the stabilization objectives of most
marketing boards were ill defined and potentially a guise for taxation. Quiggin and
Anderson (1979, 1981) discussed the limits of price bands and buffer funds. Wright
and Williams (1990) noted the widespread failure of domestic stabilization schemes
of all sorts and linked the failure to the nature of commodity prices and underlying
models of storage. Examining the time-series properties of commodity prices, Deaton
and Laroque (1992) argued that the series tended to be mean-reverting-a condi-
tion for a successful stabilization fund-but that the reversion took place over years;
consequently, successful stabilization funds needed impractically large lines of
credit. Townsend (1977) and Larson and Coleman (1993) discussed the likelihood
of bankruptcy for stabilization schemes, even when hedging markets are used.
As the poor performance of stabilization schemes became more evident, analysts
began to emphasize the distinction between policies to change the distribution of
prices internationally or domestically and policies to manage uncertainty using
markets for price risk. McKinnon (1967) explored futures markets as an alternative
to buffer stocks. Later, Gilbert (1985) demonstrated that hedging on forward mar-
kets could achieve some of the welfare gains normally associated with buffer stocks.
Gemmill (1985) argued that futures markets for cocoa, coffee, and sugar were an
attractive mechanism for hedging export earnings risks and that forward contracts
could be substantially cheaper than buffer stock operations. O'Hara (1984) looked
at the use of commodity bonds to stabilize consumption. Rolfo (1980) investigated
the use of futures for cocoa producer prices and calculated the optimal hedge ratio
when there was both production (output) and price volatility. Overdahl (1987) demon-
strated the benefits of oil futures markets for oil-producing states.
The volume edited by Priovolos and Duncan (1991) brought together much of
the new thinking on the use of market instruments to manage government debt.
Myers and Thompson (1991) presented a model of external debt management that
included commodity-linked bonds. Claessens (1991) pointed out that commodity
Doaltd F. larson, lock R. Anderson, and Panos Vtara(igis             203



bonds can be used to hedge debt management problems associated with volatile
export earnings. Wright and Newbery (1991) proposed commodity-linked financial
instruments to smooth commodity export revenue, and Anderson and others ( 19 91)
looked at the role of partial guarantees and commodity contingency. A related vol-
ume, edited by Claessens and Duncan (1993), presented case studies showing how
markets could be used to achieve many of the sectoral stabilization objectives of
existing programs in a sustainable way.
About the same time, evidence was mounting that many of the interventions
intended to facilitate growth had instead become impediments. Many stabilization poli-
cies limited competition and misdirected resources. But more important, many policies
had inconsistent objectives. The same governments that sought to protect producers
from the negative consequences of volatile commodity markets also taxed their produc-
ers directly or indirectly to spur industrialization and favor urban interests. Influential
studies by Krueger and others (1991) and Mundlak and others (199 3) documented the
negative consequences of these policies and recommended sweeping reforms.
Most telling was the growing evidence that the interventions failed to have their
intended effect, most often because they could not be sustained. Growing expendi-
tures on stabilization and support programs led to a lessening of support in the
European Union, Japan, and the United States, whereas untargeted efforts to subsi-
dize consumers or producers proved expensive for developing economies (Akiyama
and others 2001). The international commodity agreements were unable to adapt
to changes in the market, and by 1996 their economic clauses had all lapsed or
failed (Gilbert 198 7, 1996), victims of politics and economics (table 1). Funds dedi-
cated to buffer stock management at the Common Fund have never been used for
that purpose. The IMF eliminated its Buffer Stock Financing Facility in February
2000. Signaling an end to the era of international commodity market interventions,
sTrAMx and SYSMIN were abolished with the Cotonou Agreement in June 2000.
By the 1990s domestic regimes were being reformed as well in response to the
changing view of government's role in commodity markets; a series of political
changes, market events, and crises; and the urging of multilateral lenders such as
Table 1. Lifc Cycles of Recent International Commodity Agreements
Itl'             Siugair    C( . ,:.        Cocoai         RU1bber
Initialagreementdate 1954    1962           1972           1980
Status of economic  Lapsed in 1963 Suspended in 1989 Suspended in 1988 Suspended in 1996;
clauses          and 198 3                                 revived in 1 99 7;
suspended in 1 999
Number of treaties  4        4              4              4
Souirce: Gilbert (1995) and Varangis and Larson (1996), updated by the authors.
204                                  The ladorld Banik Researcih Obserer. cal. 1 9. no. 2 (Fall 2004)



the World Bank (Meerman 1997). How quickly reforms were undertaken and how
well they succeeded depended greatly on the timing and initial conditions. Some
reforms came as a result of fiscal crises. spurred in part by events in international
markets, and were implemented with little planning. Some reforms yielded quick
results, with smallholders' incomes rising rapidly as interventions came to an end.
The extent of reforms differed greatly by commodity as well, with reforms common
in coffee markets and rare in sugar markets (Akiyama and others 2003). But as
domestic and international institutions charged with stabilizing commodity prices
were systematically dismantled, policymakers became concerned that reforms had
exacerbated volatility problems.2 New approaches were sought.
Thus from the time of the Bretton Woods Conference through the structural
reform period of the 1990s, policymakers and development economists seeking to
manage the risks associated with commodity markets focused first on macroeco-
nomic issues of stability and growth and later on the efficacy and sustainability of
stabilization policies. There was frequently a tendency to blur the distinctions
between two sets of problems related to commodity risks: the inability of some gov-
ernments to prudently manage volatile revenues and expenditures and the limited
capacity of vulnerable rural households to diminish the consequences of risks and
the high costs of trying to do so.
Though earlier approaches relied on sectoral instruments (such as buffer stocks or
trade interventions) for both problems, current approaches advocate separate treat-
ment. This is not to dismiss the link between macroeconomic stabilization policies
and rural incomes. Macroeconomic policy associated with stabilization still signifi-
cantly affects rural incomes through rural-urban terms of trade, remittances from
urban and overseas workers with links to rural communities, and government
financing of rural development projects and safety nets. But agricultural commodities
today are less frequently a source of macroeconomic instability than they were two
decades ago because of greater diversification and declining international prices.
Dependence on agricultural commodities is more common for households. communi-
ties, and regions than for national economies. Exceptions, where export earnings for
agricultural commodities remain an important component of national income, are
concentrated among the poorest countries, where aid flows and debt muddle the links
between commodity stabilization and macroeconomic stabilization (Collier 2002).
Toward RuraL Risk Management Strategies
Helping make the conceptual distinction between the problems of macroeconomic
stabilization and rural risk management was ongoing work on the role of risk in
farming households and communities. Early examples include Schultz (1945),
Johnson (1950), and Heady (1952) on the United States. Risk also began to receive
Donald F. Larp-son., Jock R. Ainderson. anid Panos Varanzgis         205



explicit treatment in agricultural modeling exercises (Freund 1956) and early theo-
retical models of preference (Pratt 1964; Arrow 1965). Dillon (1971), Anderson
and others (19 77a), and Musser and Patrick (2002) survey the literature. Similar
themes were addressed, first by anthropologists and later by economists, in the con-
text of how traditional peasant societies managed risk (Schultz 1964: Stern 1989).
Related work examined sharecropping arrangements (Cheung 1969).
Concurrently, household-focused studies provided additional empirical infor-
mation about the relationship between risk and behavior. This research was made
practical by newly available village and household surveys, especially a series
undertaken by the International Crops Research Institute for the Semi-Arid Tropics
(JcCRISAT). From that research came strong evidence that farmers in poor rural com-
munities are risk-averse and take reasonable actions that result in lower but more
stable incomes. Researchers began to distinguish between planning strategies and
coping strategies. They also distinguished among price, production, and income
risks; between idiosyncratic and systemic risk: and between rare natural disaster
risk and more frequent risks. Researchers also explored the capacity of households
to self-insure and both formal and informal arrangements for sharing risks within
communities. Early examples are Moscardi and de Janvry (1977), Dillon and
Scandizzo (1978), Binswanger (1980), Hazell (1982), andAntle (1989). Fafchamps
(1999) and Dercon (2002) review informal mechanisms, and Moschini and
Hennessy (2001) review agricultural risk studies in the context of formal and
incomplete markets.
At the same time, development institutions and policymakers began to search for
ways to measure the microeconomic effects of macroeconomic and sectoral policies
meant to spur economic growth. This required finding better ways to measure pov-
erty and analyze its causes. This new approach led to several institutional changes
by development organizations. Examples from the World Bank include the treat-
ment of uncertainty in project appraisal, World Developient Report 1990: Poverty
and subsequent related research, the rise in lending for targeted safety nets, and the
sponsorship of household surveys and related analytical tools. 3
From these developments emerged a more systemic view of risk and risk manage-
ment that involves multiple strategies (prevention, mitigation, coping, management)
and arrangements (informal, market-based, public) for dealing with risk and instru-
ments that take account of the sources and characteristics of rural risk (Robison and
Barry 1987; Anderson and Dillon 1992; McConnell and Dillon 1997; Hardaker and
others 2004). The theme was featured in the 2001 updating of the World Bank
rural development strategy (Anderson 2001) and social protection strategy (World
Bank 2001).
Because the type and severity of the risks confronting farmers and others vary
greatly with farming systems, climate, infrastructure, policy, and institutional settings,
there is a certain boundlessness to the issues and few general prescriptions about
206                                T7hle Wl'orl(d Bank Research Observer. vol. 1 9, Io. 2 I FaIl 2004)



strategies to manage risk are possible. Rather, the sections that follow discuss some
of the formal and informal mechanisms for managing rural risk.
Although the relevance of these mechanisms depends on local circumstances,
there are some typical characteristics of development that affect the availability of
these instruments and the demand for them. Keeping key differences among coun-
tries in mind is especially important when drawing generalizations from research
based on specific household surveys, particularly those in India, Ethiopia. and
Thailand.
Generally, as countries develop, rural communities become more diversified in
their sources of income, the share of the population engaged in agriculture shrinks
(Lanjouw and Lanjouw 1995: Larson and Mundlak 1997), and farmers become
more commercialized and specialized. Formal markets of all sorts-for inputs, credit,
insurance, information, and outputs-are more likely to arise and mature as coun-
tries develop. In part all this comes about because governments are better able to
safeguard the security of persons and property and to enforce contracts (North
1994). Finally, the most obvious differences are in accumulations-of wealth, infra-
structure, human capital-that affect the vulnerability of rural households and
communities and the role of policy in managing agricultural risks. This means that
the hurdles to accessing risk markets are higher in countries and communities
where there is less capacity to self-insure.
Vulnerability, Ex Ante Choices, and Ex Post Adjustments
The consequences of risk for rural households are related both to the types of risks
associated with specific income-earning activities and to the capacity of households
to absorb losses. Because all economic activity entails some measure of risk, the effi-
cacy of risk management efforts by the household can be judged in the aggregate by
examining how well households are able to maintain consistent levels of consump-
tion over time.4 Households can achieve this in several ways. They can prepare for
times of need by saving, holding inventories, and accumulating productive assets
that can be sold in times of need. Because risks are likely to affect different activities
in different ways, diversification of activities-especially the diversified use of land
and labor-can lessen the consequence of any particular risk. Thus for farming
activities, for example, choices about production technologies can affect the conse-
quences of weather-related risks. Because there are separate risks associated with
markets, decisions concerning self-sufficiency and market participation affect income
risks as well.
A growing applied literature on these topics has frequently found shocks related
to crop and livestock yields, prices, health, and labor markets. Fafchamps (1999)
and Dercon (2002) survey studies characterizing the sources of rural risks in poor
Doriald F. Larson, Jock R. Anderson, and Panos Varangis               207



countries. Wolpin (1982), Paxson (1993), and Rosenzweig (2001) examine consump-
tion smoothing in low-income countries. Dercon (1996), Larson and Plessmann
(2002), and Morduch (2002) find evidence that farmers choose to diversify into less
profitable crops or choose less productive technology.
These strategies and their outcomes act in reinforcing ways. Greater wealth
allows households to diversify their income sources by starting nonagricultural
businesses or entering into more profitable agricultural activities (Reardon and
others 1988; Dercon 1998). Greater wealth can also allow farmers to make invest-
ments that can limit the consequences of risks (Rosenzweig and Binswanger 1993).
The consequences of rural risk can also reverse the accumulations of human and
physical capital that contribute to income and wealth. Households may attempt to
supply more labor (Kochar 1995: Rose 2001), and parents may choose to have their
children work rather than attend school (Jacoby and Skoufias 1997). In difficult
times households may liquidate productive assets (Binswanger and McIntire 1987).
Not all risk management strategies rely exclusively on household resources.
Arrangements for risk sharing and mutual insurance are a common feature of rural
communities in low-income countries. Most arrangements are informal-they
rarely involve written contracts and are not enforced by governing institutions.
Alderman and Paxson (1992) and Morduch (1998) provide useful surveys of infor-
mal insurance mechanisms. The ubiquity of risk-sharing arrangements among the
rural poor in low-income countries is consistent with studies that find rural risks in
low-income countries to be largely idiosyncratic (Udry 1991: Townsend 1995; Jalan
and Ravallion 1999). The nature of risk may be partly an outcome of diversification
strategies. Systemic risk is more significant in developed economies' agriculture,
where greater specialization is evident (Miranda and Glauber 1997).
When risks are idiosyncratic, there is scope for mutual insurance arrangements.
In turn, the practical distinction between what constitutes systemic or idiosyncratic
risks depends on how risks are pooled. Risks that affect all household members may
not affect all households in a community. Risks that affect communities may not
affect the entire region, state, or country. For agriculture, underlying risks associ-
ated with production stem from the interplay between climate-related events and
natural endowments. Consequently, geographically diverse pools of agricultural
risk exhibit less systemic risk. At the same time, performance risks related to con-
tract enforcement in low-income countries limit the scope for mutual insurance
arrangements that extend beyond families, tribes, or villages.
The personal relationships that characterize mutual insurance arrangements are
also a dominant feature of other forms of exchange in poor rural areas (Fafchamps
and Minten 2001). Where low-cost enforcement mechanisms are absent, perfor-
mance risks associated with all forms of contracts can be high.5 However, the scope
for pooling and sharing risks of all kinds is dramatically reduced for forms of
exchange and insurance based on personal relationships. Storage markets, liquid
208                                The lW'orld Bank Research Observer, vol. 19, no. 2 (Fall 20041



markets for output and for assets, financial markets for savings and credit, and for-
mal insurance markets all depend on impersonal trade accounts to varying degrees.
It is not surprising then that constrained household and community insurance
mechanisms are only partly effective (Dercon 2002). Moreover, the poorest appear
to benefit the least from informal insurance (Reardon and others 1988; Jalan and
Ravallion 1999: Goldstein and others 2002). For all participants, informal insur-
ance schemes in rural areas work least well when they are needed most for several
reasons. First, the most severe agricultural shocks tend to be systemic, so that all
farmers in the community are less able to afford transfers. This is exacerbated when,
as is frequently the case, nonfarm income can be covariate with systemic shocks to
agriculture (Czukas and others 1998). Because of illiquid markets, shortfalls in food
supply can cause food prices to rise rapidly while prices for livestock or other produc-
tive assets fall (Rhamato 1991). In situations where household and community risk
strategies fail completely, households lack the ability to command food supplies, and
famines result (Sen 1981; Ravallion 199 7).
Extending Formal Markets for Risk
Though varied, comprehensive in types of risks addressed, and able to solve informa-
tion problems related to transaction costs and moral hazard, informal mechanisms
for managing rural risks also reduce the scope for pooling risks outside of communi-
ties.6 They are unable to take advantage of the potentially beneficial portfolio effects
of including individuals who face uncorrelated or negatively correlated natural risks
or who have different capacities or willingness to assume risk.
There have been many recent efforts to extend the reach of rural risk markets in
developing economies to supplement informal systems. The focus here is on instru-
ments that address agricultural production and price risks, rather than on credit and
income risks, for which there is also an extensive literature (for surveys, see Bell
1988 and Besley 1994: for policy issues see Yaron and others 1997; and for lessons
on informal systems, see Udry 1995 and Eboh 2000). The primary markets that
uniquely address agricultural risks are futures and options markets, which can be
used to hedge against price movements, and crop insurance markets, which focus
on yield shortfalls.
Risk Markets, Households, and Small Businesses
Small-scale farmers and other small businesses in developing economies rarely use
market-based risk instruments. For one thing, the very poor lack assets insurable by
crop insurance, price insurance, or similar products. For another, most agricultural
risk markets are incomplete in many aspects-time, space, and type of commodity.
Donald F. Lairson. jock R. Anderson. and Panos Vartngis               209



Nonetheless, there is some scope for making use of such markets in poor rural areas
(Rolfo 1980; Myers 1988: Lapan and Moschini 1996).
There are at least three types of potential benefits of using risk markets. One
relates to the indirect effects of insuring individuals. Poverty can be transient and
often market related (Jalan and Ravallion 1998). As a result, market-related price
and output insurance, delivered directly or indirectly to community households and
businesses, may have direct income effects for the insured and economic multiplier
effects in the community. In addition, both effects may help fund informal risk sys-
tems. There is some related evidence from food aid programs, discussed later.
A second type of benefit relates to funding formal public safety nets and services.
During periods of systemic risk, when informal systems are most likely to fail, gov-
ernments are often called on to provide relief and to step up related services. In some
instances governments are unable to fully afford to do so, particularly provincial or
local governments that may be obliged to share in the cost of relief. Thus risk mar-
kets have a potential role when governments do not have the resources to self-
insure.
A third potential benefit is that extending the reach of formal risk markets can
help extend the temporal and spatial aspects of related markets for output, storage,
inputs, and credit. This is important because it is the emergence from self-sufficiency
and the opportunity to participate in markets that marks the earliest stages of devel-
opment in rural areas. '
Price Risk
The most transparent markets for price risk are futures and options contracts traded
on organized exchanges. Contracts for coffee, cocoa, maize, soybeans and soybean
products, sugar, wheat, and some livestock are traded on an international basis.
Most volume is traded in Japan, the United States, and the United Kingdom, but
there are long-standing exchanges in Argentina, Brazil, and India, as well as emerg-
ing commodity exchanges in China, Hungary, Poland, and Russia.8 There are also
active over-the-counter markets for options on traded agricultural products, includ-
ing rice and palm oil. Significant barriers stand between these risk markets and the
risks faced by farmers, however, including issues of scale, basis risk, information,
and enforcement mechanisms (Frechette 2000).9 Moreover, options are short-term
instruments, rarely exceeding nine months, and they are inappropriate for hedging
annual income fluctuations (Gardner 1989). However, some mechanisms have
emerged to extend price risk markets.
Using inventories. In some countries, warehouses and grain elevators are delivery
points for price risk instruments. In the simplest form grain elevators post schedules
of forward purchase prices. Elevator companies then hedge their net exposure on
210                                 The llWorld Bank Research Observer. vol. 19, no. 2 (FIall 2004)



commodity exchanges or related price risk markets. Harwood and others (1999,
p. 7 5) provide an example from the tJnited States.
Contractual arrangements can also be used to link price hedging, inventories, and
credit. For rural communities the chief benefit is lowering transaction costs. In ware-
house receipt or inventory financing systems, inventory owners place their crops in
certified warehouses and enter into a repurchase agreement with the warehouse
company, which then offers a loan based on the value of their inventories. When
price futures, options, or related markets are available, warehouse companies can
use the markets to hedge the value of the inventory collateral.
Such arrangements are frequently accorded special treatment in the law, allow-
ing warehouse managers to quickly liquidate inventories when loan payments are
due. In the United States, for instance, certified warehouses can issues transferable
receipts (Budd 1991). In Latin America two documents perform the same economic
function: the certificatdo de deposito (certificate of deposit), a claim stating the quantity
and quality of goods stored, and the bono de prenzda (product bond), used to assign
ownership. Although this institutional arrangement has a long history in Latin
America, efforts to establish warehouse receipts are more recent in countries where
inventories were long managed by government agencies or state enterprises.
Coulter and Onumah (2002) provide examples from Africa; Martin and Bryde
(I1999) discuss warehouse receipt systems in transition economies.
Aggregating demand. Several characteristics of price risk markets create obstacles to
access by small-scale farmers. The contracts themselves are associated with large
underlying volumes, and futures and options trading involves large fixed costs. Sev-
eral efforts have been made to provide access to price risk markets by aggregating
local demand. Varangis and Larson (1996) report on Mexico's Agricultural Products
Options Program (Programa de Covertura de Precios de Productos Agricolas), which
offers subsidized dollar-denominated futures options to maize, cotton, sorghum, and
soybean farmers. Guatemala's national coffee association, Asociacion Nacional de
Cafe. facilitates price-hedged credit for smallholder producers.
Though the mechanisms of such programs are understood, there has been little
analysis of their effects on household decisions and outcomes. To further investiga-
tion of this type of risk management program, the World Bank, the European Union,
the Netherlands. and Switzerland convened an International Task Force on Com-
modity Risk Management in Developing Countries (1999) to study the benefits of
facilitating access by farmers and small businesses in developing economies to price
risk-management instruments. The program focuses on finding local intermediaries,
mostly producer cooperatives, to aggregate demand. pay option premia, and dissem-
inate payouts. Ten case studies are available on the program's Web site (www.itf-
commrisk.org/itf.asp?page= 104).
Do,niald F. Ltar-so. ock R. A.i R derson, anid Panos Varl-anis       211



Yield Risks
In addition to price risks, farmers face yield risks, due to adverse weather, pests, and
diseases. Roumasset and others (1989) review studies on the sources of yield risk for
grains.
Crop insurance. Formal agricultural insurance institutions are better suited to
address yield risks than are informal mutual insurance mechanisms, which have
limited reach. Ray (1967) has an early discussion. However, moral hazard and
adverse selection problems plague traditional crop insurance schemes. Moral hazard
arises when farmers are able to take actions that affect insurance payouts. Inspec-
tion and monitoring can limit the problem, but they boost costs-especially in the
context of smallholder production in developing economies. Adverse selection
results when those purchasing insurance have better information about risks than
the insurance provider. As a consequence, unless insurance is compulsory, the self-
selected pool of insurance participants will have above-average risks. The adverse
selection problem has both spatial and temporal aspects. For example, farmers may
choose to insure only crops grown on inferior land (spatial) or participate only when
preseason rainfall is low (temporal). In practice, because moral hazard and adverse
selection problems both stem from asymmetric information, both types of problems
will emerge (Quiggin and others 1993).
Information problems related to crop insurance are well studied in part because
efforts to provide comprehensive crop insurance on a commercial basis have so con-
sistently failed (Valgren 1922; Krammer 1983).'() Thus crop insurance relies
heavily on government subsidies. Pomareda (1986), Hazell (1992), and Wright and
Hewitt (1994) compare crop insurance in developed and developing economies.
Detailed histories of crop insurance and related policies are provided by Lopes and
Dias (1986) for Brazil, Yamauchi (1986) for Japan, and Glauber and Collins (2002)
for the United States.
The combination of moral hazard, adverse selection, and public subsidy has wel-
fare consequences related to how farmers allocate resources, how natural resources
are used, and how public support is allocated. There is an extensive theoretical and
applied literature, much of it related to North American experiences. Knight and
Coble (1997) and Moschini and Hennessy (2001) survey the literature, and Skees
(2000) provides lessons based on U.S. experiences.
Index insurance. Because of the high costs associated with administering traditional
crop insurance, interest has grown in alternative instruments for insuring a portion
of farm yield risks. These products look to contain administrative costs by eliminat-
ing moral hazard and reducing adverse selection and thereby the associated moni-
toring costs (Skees and others 1999). They do so by using an insurance trigger
212                                 The World Bnnk Researcih Observer. vol. 19, no. 2 (Fall 2004)



linked to farm losses that cannot be manipulated by the insured. Contracts and
indemnity payments are the same for all buyers per unit of insurance, so the prob-
lems of moral hazard and adverse selection associated with traditional crop insur-
ance are lessened. These features can significantly reduce administrative costs
because contracts can be standardized and on-site inspections and damage assess-
ments are avoided.
Conceptually, these products are similar to exchange-linked price insurance for
price takers. These products address risks that are systemic to the insured group but
leave idiosyncratic risks uninsured, much as basis risks remain with price insurance.
In addressing systemic risk only, these products are less appealing to farmers but
more attractive to a broad range of people and institutions whose incomes are also
affected by the insured events, including banks, agribusinesses, rural financial insti-
tutions, agricultural traders, and transport industries. Banks and rural finance insti-
tutions could purchase such insurance to protect their portfolios against defaults
caused by severe weather events, thereby addressing an important constraint on
rural credit.
Area-yield and weather-based insurance are two of the better known examples of
index insurance, although other products have also been studied. Skees and Enkh-
Amgalan (2002) examine a proposed index-based insurance for livestock in Mongolia.
Area-yield crop insurance is a contingent contract that pays out when average yield
for a group, usually defined geographically, falls below a specified trigger. Payouts
are based on probabilities of the event occurring, which are usually based on histor-
ical data. If the pool that defines the trigger is sufficiently large, decisions to enroll or
actions on the farm will not affect outcomes, resolving moral hazard and most types
of adverse selection problems.
Halcrow (1949), Industries Assistance Commission (1978), and more recently
Miranda (1991) discuss proposals for area-yield insurance. Much of the literature is
devoted to optimal contract design, including discussions by Miranda (1991), Smith
and Baquet (1996), Skees and others (1997), Mahul (1999), and Vercammen (2000).
Chambers and Quiggin (2002) consider the influence of area-yield insurance on
producer decisions. Miranda (1991) and Skees and others (1997) describe area-
yield programs in Canada, Sweden, and the United States. Mishra (1996) describes
area-yield insurance in India.
Experience with area-yield products has been mixed for several reasons. Product
coverage, based on county or other administrative boundaries, may not correspond
to the spatial outcomes of climatological risk (Wang 2000). There is a natural ten-
sion in the optimal size of the pool of farmers included in the trigger. Large pools are
needed to preclude moral hazard and help offset the fixed costs of offering an insur-
ance product, but large pools may increase the basis risk associated with the insur-
ance trigger. Detailed data are needed to construct the underlying probabilities, and
such data may not be available in many developing economies. Experience with
Donald F. Larsoni, lock R. Anderson, and Patios Varanigis             213



area-yield programs comes primarily from developed economies, especially the United
States, where government policies may crowd out market-based products and intro-
duce policy-related risks that market-based products cannot address (Gardner
2002).
A relatively new insurance instrument is weather-based index insurance, with
payouts based on the occurrence of a weather event rather than on actual crop
losses (Lee 1953: Bardsley and others 1984; Quiggin 1986; Skees 1999; Turvey
2001). Thus insurance contracts could be written against severe rainfall shortfalls
(say, 30 percent or more below a defined norm) measured at agreed regional
weather stations. The insurance would be sold in standard units, and all buyers
would pay the same premium and receive the same indemnity payment per unit of
insurance if a shortfall occurs.
Although the theoretical aspects of contract design have been explored, and cli-
mate-related risks are well understood in developing economies, less is known about
whether farmers are willing or able to pay for such insurance. What little evidence is
available suggests that a residual unmet demand for insurance against weather
risks remains despite the myriad risk management mechanisms employed by house-
holds (Gautam and others 1994; Sakurai and Reardon 1997). Additionally, survey
evidence suggests that farmers are willing to pay for certain contract designs based
on reasonable premia assumptions (McCarthy 2003).
Still, there are certain challenges in creating weather insurance instruments, par-
ticularly in developing economies. First, reliable historic data are needed to allow
accurate pricing of the insurance. Though such data are generally available in most
developing economies, coverage can be limited. Second, secure weather stations are
needed to increase the confidence of those providing the insurance. Automated
weather stations and increasingly remote sensing could reduce the risk of tampering
with weather observations at local weather stations. Third, farmers may face basis
risk if rainfall at the weather station is not highly correlated with rainfall at the
individual farm. This is more of a problem for farmers in regions with diverse micro-
climates than for banks purchasing insurance to protect their regional portfolio or
for agribusinesses that have weather exposure over a wider area. Skees and others
(2002) discuss examples from Mexico.
There have been few agricultural applications of weather-based index insurance
in the world. Weather indices are used for agricultural insurance in Ontario and
Alberta, Canada. In Mexico AGROASEMEX, the agricultural reinsurance company,
recently used weather derivatives to protect part of its crop reinsurance portfolio
exposed to weather risks. In Argentina a rainfall insurance contract is used by a
milk-producing cooperative (there is positive correlation between rainfall and milk
yields). In 2001 the International Finance Corporation, the private-sector lending
arm of the World Bank Group, joined with private investors to create an $80 million
facility to reinsure weather risks in developing areas. Agriculture, however, is only
214                                 The World Bank Research Observer, vol. 19, 1o. 2 (Fall 2004)



one of several industries affected by weather risks. Markets in North American and
Europe have developed primarily to hedge energy-related risks (Muller and Grandi
2000). Consequently. weather markets may prove more liquid than markets for
area-yield insurance.1I
Revenue insurance. There has been considerable discussion about how incomplete
markets limit the ability of farmers to hedge income risks (Rolfo 1 980; Anderson and
Danthine 1983: Lapan and Moschini 1994). Several studies have assessed the
potential for revenue insurance, primarily in North America (Miranda and Glauber
1991: Turvey 1992; Babcock and Hennessy 1996: Hart and others 2001).
Conceptually, revenue insurance is a combination of price and traditional crop
insurance-although area-yield variations have been considered. A U.S. pilot pro-
gram uses the average February price of a December settlement corn futures and
historical yields to price an insurance product. Pavouts are based on a combination
of assessed yields and the average price of the December settlement in November.
Because of this construction, high prices can compensate for low yields, lessening
the contract payout. Application to developing economies is limited because the
insurance requires both price and yield risk markets.
Contractual Arrangements and Market Instruments
Pre-export financing. Where domestic credit systems are expensive or function
poorly, exporters frequently enter into pre-export financing arrangements to share
performance, price, and exchange risks. These arrangements, which combine
aspects of formal and informal markets, typically include a local domestic agent and
an international trading company.
The two parties enter into a contract for a specific purchase-so many tons of
coffee, for example, for a fixed fee. The cost of credit is implicit in the fee. The inter-
national company deposits dollar-denominated funds in a domestic bank, and the
domestic agent purchases domestic crops and prepares them for export. The inter-
national company approves the release of funds to the domestic agent based on certi-
fication by a local warehouse or inspection company. Although the contracts usually
cover very short periods, they are often repeated when performance has been good.
The parties rely on economic incentives (off-shore credit is typically less expen-
sive) and the prospects of repeated future business, a feature also associated with
informal personal transactions. Price risks are shared: the agent takes on price and
exchange risks during the purchasing period (because local purchases will be in
local currencies), whereas the international company takes on the price risks from
the time it acquires supplies in country until the goods reach their eventual market.
Doniald F. Lairsoii, Jock R. A-n derson. aizd Ptinos l 'ti-ar gis    215



The international company may enter into additional risk management arrange-
ments by hedging the good or entering into forward contracts.
Producer clubs. Even though contract enforcement is often difficult in developing
economies, sharecropping and labor-sharing arrangements are commonplace in
many rural communities. Robertson (1987) provides many examples from Africa.
Where production takes place near a processing center, contracts are sometimes
structured so that processors provide smallholders inputs on credit in exchange for
future product deliveries. Such contracts are common in the sugar and palm oil
industries, where crop characteristics create a binding relationship between pro-
ducer and processor.
More sophisticated arrangements can also be found for specialized products that
include types of risk management. For example, in Mexico starch companies will
enter into arrangements with producer groups for high-starch maize. Group mem-
bers are jointly obligated to meet contract requirements. In return, the starch com-
pany offers a forward purchase agreement for the maize at a fixed price and partly
prefinances inputs and technical advice. The company also partially prefinances
crop insurance, so that farmers are not penalized when crops fail. By addressing
both price and yield risk, the arrangement allows farmers to adopt newer techno-
logies more readily.
Governments, Insurance, and Rare Events
Some events can overwhelm the actions taken by individuals, families, and commu-
nities to manage risks. When the scale of loss of life and property is large, govern-
ments step in to limit further loss and help compensate victims. One area where such
intervention is important is in providing assistance to prepare for and recover from
natural or human-caused disasters, when human and economic losses can be large.
Developing economies suffer the greatest costs from disasters: More than 9 5 percent
of all deaths caused by disasters occur in developing economies, and losses due to
natural disasters are 20 times greater (as a proportion of gross domestic product) in
developing economies than in industrial countries. Poorly planned development can
turn a recurring natural phenomenon into human and economic devastation with
irreversible effects.12
But along with such cases there has been a tendency for public outcry to drive
governments to intervene in ways that are demonstrably ineffective and that distort
individual incentives to plan more carefully for occasional inevitable bad outcomes,
including through selective purchase of insurance contracts. If governments rush to
bail people out of the effects of insurable natural disaster risks whenever there is
political clamor to do so, commercial insurance markets will be stunted. Such "free"
216                                Tlre World Banlk Researchl Observer, vol. 1 9, no. 2 (Fall 2004)



but unreliable public insurance entices some to take up risky activities with incom-
plete safeguards and without full consideration of the public and private costs of
doing so (for an example related to drought management in Australia, see Drought
Policy Review Task Force 1990). Often, when political factors (including the avail-
ability of donor assistance) trigger disaster-relief programs, the timing and level of
relief is uncertain (Shaw 2002) and relief may be poorly targeted.
Public policy research has long considered questions about the role of govern-
ments in sharing responsibilities for insurance and insurance markets (for example,
Kunreuther 1978). A potentially good policy would swing into action as needed.
without requiring (or allowing) political largesse or creating disincentives for pro-
ducers to plan for and manage their own natural disaster risks (Anderson and Dillon
1988).
Food Programs and Informal Insurance
Food programs are a special type of public insurance. In wealthy countries, food
assistance is part of a general safety net, provided continuously and not only as part
of natural disaster relief efforts. Not so in poor countries, where food assistance gen-
erally is a response to acute shortages. Natural disasters are often a triggering event,
but their consequences are heavily influenced by previous private and public invest-
ments and policy. When food shortages occur in their most acute form, as famine,
they are almost always associated with a combination of natural disasters and bad
policy. Ravallion (199 7) reviews public action failures related to famine, from inade-
quate government responses to food shortages-sometimes deliberately so-to mar-
ket interventions that have made households more vulnerable to famine.
There is an extensive body of literature on programs that provide food in response
to disaster-related shortages, with surveys by Maxwell and Singer (1979), Clay
(1986), and Barrett (2001). In addition to the nutritional impact of such programs.
researchers have addressed problems of moral hazard and adverse selection, includ-
ing disincentives for households to produce food and to work. Food deliveries can
also displace food and storage markets (Schultz 1960; Sen 1960: Fisher 1963:
Barrett 2001). The disincentive effects are intensified by the tendency for some
program food supplies to go to unintended recipients, which also reduces the effi-
ciency of the programs. Considerable effort has gone into finding appropriate meth-
ods of targeting food assistance to avoid these problems (Besley and Kanbur 1990:
Ravallion 1991; von Braun 1995: Alderman and Lindert 1998).
More recently. researchers have considered the relationship between informal
insurance systems and food programs. Attanasio and Rios-Rull (2 000) provide a numer-
ical analysis drawing on data related to Mexico's PROGRESA program, and Dercon and
Krishnan (2003) offer econometric evidence based on surveys from Ethiopia. Both
Donal,ld 1F. Lairson, Jo(k R .Anderson, aid Panos V aranfqis         217



studies suggest that formal food assistance can crowd out informal insurance,
because food assistance changes the consequences of not participating in mutual
insurance arrangements. This tension between informal insurance incentives and
food assistance prompted Dercon and Krishnan to suggest that food assistance pro-
grams use village mutual insurance systems to target assistance.
Nevertheless, informal systems fail when risks are systemic-for example, when
extensive or sequential multiple crop failures occur-and formal aid may have lim-
ited effects on decisions to participate in informal systems that address idiosyncratic
risks. Even when systemic natural events are the cause of food shortages, other types
of moral hazard problems related to the frequency of the events remain. When short-
ages and food aid occur frequently, households' choices may be affected in ways that
create a continuing dependency on aid-for example, households may continue to
work marginal lands. Alternatively, when shortages are infrequent, aid interven-
tions can provide immeasurable benefits without affecting long-term choices,
because households will view the possible risk and the subsequent food aid as rare.
From this Skees and others (2002) argue that food aid policies should be designed
around the frequency and systemic nature of disaster events. They suggest establish-
ing objective measures for triggering the prompt release of food aid in response to
relatively rare and systemic events, such as droughts or earthquakes, to minimize
moral hazard. The use of objective measures also opens up the possibility of finding
insurance-based methods to fund such relief. Analogous relationships exists
between disaster relief and crop insurance, discussed next.
Disaster Relief and Private Insurance
Natural disasters can have devastating effects for farmers and spawn cascading
effects throughout the local economy as agricultural workers and those employed in
support services lose jobs and income. The farm income losses associated with the
failed crops can sometimes be privately insured. When parametric insurance is con-
sidered, such as temperature or area-yield policies, the distinction between disaster
relief and insurance as compensating mechanisms diminishes. IHowever, publicly
financed disaster relief creates incentives different from those of privately priced
insurance and can induce risk-taking and other problems related to moral hazard.
Mexico's Fund for Natural Disasters (FONDEN) illustrates some of these points. The
government established FONDEN in 1996 to finance postdisaster reconstruction of
public infrastructure and compensate low-income producers for crop and livestock
losses arising from natural disasters. FONDEN targets beneficiaries and limits the
amounts it disburses per beneficiary. FONDEN was never intended to compete with
private insurance, but analysis of historic reinsurance payouts reveals that weather
events similar to those covered under FONDEN are the primary source of systemic pay-
outs by farmer mutual insurance funds. The funds were formed to provide mutual
218                                 The WVorld Bank Researcih Observer, voi. 19, no. 2 (Fall 2004)



crop insurance to their members. Reserves for paying indemnities and covering
operational costs come from member premiums and government contributions (equal
to one-third of member premiums). The funds also purchase traditional reinsurance
for times when claims deplete reserves. Because privately insured farmers are not eli-
gible to receive payments from FONDEN, the result is a choice between public and pri-
vate insurance, a situation fraught with moral hazard. Recently, to make disaster
relief more transparent and predictable, FONDEN has started to adopt objective rules
for declaring catastrophic events, using triggers analogous to those used for index
insurance. For example, technical guidelines are given for drought- or frost-related
natural disasters.
Parametrization of public disaster programs has several advantages. First, it
removes an ad hoc dimension from the declaration of catastrophes, which helps pro-
grams provide appropriate levels of relief in a timely and consistent manner. Second,
the information required to parametrize relief programs may help governments
design better policies. Third, parameterization enables governments to assess the
potential program costs. If governments, especially provincial or local governments,
find that under some conditions they would be unable to fund their share of the relief
program, they may decide to curb benefits, seek more federal support, or purchase
reinsurance from private markets.
Government insurance of risks that are sufficiently rare will not induce moral
hazard, even when the consequences of the event are considerable. Nevertheless,
when rare events are associated with extensive damage, the cost of private insur-
ance may be high. This suggests an appropriate way of sharing risks on the basis of
event likelihood, with governments insuring rare but catastrophic risks and private
markets insuring more frequently occurring risks.
Conclusion
During much of the past 50 years, agricultural risk management policies advo-
cated by governments and multinational organizations focused on stabilizing
national and sectoral incomes, primarily through interventions meant to stabilize
the prices of key food or export commodities, such as rice, maize, coffee, and sugar.
Frequently, these policies also set out to change relative prices. Successful efforts
to manage these markets were rare and usually short-lived. Most countries came
to view these interventionist policies as an impediment to growth and subse-
quently moved to sharply reduce or eliminate domestic market interventions.
often in times of fiscal crisis. Contemporaneously, researchers and policymakers
gained a better understanding of the role of risk and the negative consequences of
volatility for growth and development, in large part through new household-
based surveys.
Dotiald 1. larson. lock R. Anderson. aindti Panios  artinagis         219



Currently, strategies for managing the negative consequences of volatile rural
incomes are emerging that rely more heavily on markets to supplement the many
ways governments and households manage risks. These strategies face demanding
challenges because the costs of risk avoidance to households and businesses can be
high: the capacity for self-insurance by households, groups. and even governments
can be limited; and markets that support risk sharing are incomplete. Finding ways
that extend the reach of risk markets is therefore important. These strategies must
also address risks that markets cannot address, and key questions remain concern-
ing the roles for formal, informal, private, and public insurance markets and the
links between them. Issues include the extent to which informal risk-sharing sys-
tems benefit from formal insurance markets, how climate risk information can be
used to understand the implicit value of publicly provided insurance and the capa-
city of governments to meet those obligations under all circumstances, and what
components of price and yield risk should fall to individuals, governments, and
multinational donors.
Notes
Donald F. Larson is senior economist in the Development Research Group at the World Bank: his e-mail
address is dlarsonCworldbank.org. Jock R. Anderson was an adviser and is presently a consultant in
the Agriculture and Rural Development Department at the World Bank: his e-mail address is janderson
gworldbank.org. Panos Varangis is lead economist in the Agriculture and Rural Development Depart-
ment at the World Bank: his e-mail address is pvarangisCa).worldbank.org. Seniority of authorship is not
assigned. but a novel ordering of authors is used. An earlier version of this article was prepared for a
seminar held in honor of Ronald C. Duncan (former Chief of the Commodities Division, World Bank) on
the occasion of his retirement as executive director of the National Centre for Development Studies and
Director, Asia Pacific School of Economics and Management. Australian National University, Canberra,
in July 2001 (Anderson. Larson, and Varangis 2002).
1. See modern treatments by Sapsford and Balasubramanyam (1999) and Lutz (1 999).
2. As it turns out, evidence of increased commodity price volatility is hard to find (Dehn 20(00:
Akiyama and others 2(003.)
3. See Reutlinger (1970) and Anderson (1989) for discussion concerning uncertainty and public
project appraisal. See Grosh and Glewwe (1 995), Deaton (1997). and Grosh (1998) for discussions of
household surveys, the World Bank and poverty analysis.
4. This test is based on the notion that households that can freely save and borrow will ignore tran-
sitory changes in income and base consumption on the permanent component of income. Friedman's
(1954) response to Caine (1954) contains early elements of the permanent income hypothesis.
5. Kranton (1996) discusses this in the context of personal exchange generally: Morgai and others
(2002) discuss transaction and enforcement costs specifically in the context of mutual insurance
schemes. See also related studies by Coate and Ravallion (1993), Platteau (1997), and Ligon and others
(2002)
6. Studies of pastoralists show the limits of diversifying climatic risks. See for example, Little and others
(2001), McPeak and others (2001), and Smith and others (2001) and references therein.
7. Reinikka and Collier (2001) document how increased market participation by households char-
acterized Uganda's recovery in the 1990s.
220                                      Thle tt'orld Batik Researcih Obserter, v,ol. 19. nio. 2 (FaDl 2004)



8. Williams (2001) provides a survey on the organization of commodity futures market. See Peck
(2001) for a discussion of emerging exchanges in the former Soviet IJnion. Eastern Europe, and China.
9. Basis risk refers to a situation where local prices fail to move exchange prices. This topic is dis-
cussed in the context of West African cotton by Satyanarayan and others (1993): coffee in Costa Rica
by Claessens and Varangis (1993): and wheat in Pakistan by Faruqee and Coleman (1996).
1(0. Insurance against specific risks-especially hail and some multiperil livestock insurance pro-
grams-have proven successful (Harwood and others 2(000).
1 1. Lack of liquidity, partly a result of crowding-out by other tT.S. programs, was a primary reason
for a failed attempt by the Chicago Board of Trade to market area-yield options.
12. Christiansen and others (2003) show sobering results that quantify the stunting effects of
drought on children in Ethiopia.
References
Akiyama. T.. J. Baffes, D. F. Larson. and P. Varangis. 2001. Comtnriodityj Market Reformls: Lessons oJ Twvo
Decades. Washington, D.C.: World Bank.
2()03. "Commodity Market Reform in Africa: Some Recent Experience." Econonmic Sy/stemr?s
27)1):83-1 15.
Alderman, H.. and K. Lindert. 1998. "The Potential and Limitations of Self-Targeted Food Subsidies."
W1orld Baook Researclh Observer 13(2):21 3-29.
Alderman. H., and C. Paxson. 1992. "Do the Poor Insure? A Synthesis of the Literature on Risk
and Consumption in Developing Countries." Policy Research Working Paper 10()8. World Bank.
Washington, D.C.
Anderson. J. R. 1989. "Reconsiderations on Risk Deductions in Public l'roject Appraisal." Austroliao
Journal of Agricultiural Economzics 33(2): 136-4(1.
20)0)1. Risk la qagemenit in Rural Developmnent: A Reviewv. Washington, D.C.: World Bank, Rural
Development Department.
Anderson. J. R., and J. L. Dillon. 1988. "Socioeconomic Impacts of Climatic Variability: Implications for
Policymakers and Planners." In The Imipacts of Clirrmatic Voriotions on Agriculture, vol. 2: Scemi-Arid
Regions. ed. M. Parry, T. Carter. and N. Konijn. Dordrecht, Netherlands: Kluwer.
1992. "Risk Analysis in Dryland Farming Systems." Farm Systems Management Series 2. Food
and Agriculture Organization of the United Nations, Rome.
Anderson, J. R., J. L. Dillon, and J. B. Hardaker. 1977a. Agricultural Decision Anal,rsis. Ames: Iowa State
tJniversity Press.
Anderson. J. R.. P. Hazell, and P. Scandizzo. 197 7b. "Considerations in Designing Commodity Stabiliza-
tion Schemes." Amlen-ican Joiirnal of Agricultural Economics 59(5):908-1 1.
Anderson, J. R., C. Blitzer. E. Cauchois, and E. Grilli. 1981. "A Dynamic Simulation Model of the World
Jute Economy." Europeanr Econiomic Reviewv 16(2/3):303-31.
Anderson. J. R., D. F. Larson, and P. Varangis. 2002. "Managing Risks Rather than Markets: An
Institutional View." In Resource AMlanagemtienit in Asia Pacific Developing Counltries, ed. R. Garnaut.
Canberra: Asia Pacific Press.
Anderson, R., and J. Danthine. 1983. "The Time Pattern of Hedging and the Volatility of Futures
Prices." Revienw of Economric Studies 50)2) :249-66.
Anderson, R., C. Gilbert. and A. Powell. 1991. "Securitizing Development Finance: The Role of Par-
tial Guarantees and Commodity Contingency." In Commnodity Risk MWvanagemtienit anld Finance., ed.
T. Priovolos and R. C. Duncan. Oxford: Oxford Utniversity Press.
Douald F. Larson. Joc k R. Anderson., tid l'a,,os l'araommqis                          221



Antle, J. M. 1989. "Nonstructural Risk Attitude Estimate." Journial of Agricultural Economics 71(3):
774-84.
Arrow, K.1965. Aspects of the Theory of Risk-Bearing. Helsinki: YrioJahnsson Foundation.
Attanasio, O., and J. Rios-Rull. 2000. "Consumption Smoothing in Island Economies: Can Public
Insurance Reduce Welfare?" European Economic Review 44(7): 1225-58.
Babcock B., and D. Hennessy. 1996. "Input Demand under Yield and Revenue Insurance." American
Journal of Agricultural Economics 78(2):416-27.
Bardsley, P., A. Abbey, and S. Davenport. 1984. "The Economics of Insuring Crops Against Drought."
Australian Journal of Agricultural Economics 28(1):1-14.
Barrett, C. 2001. "Food Security and Food Assistance Programs." In Handbook of Agricultural Econom-
ics, vol. 2, ed. B. Gardner and G. Rausser. Amsterdam: Elsevier Science.
Bauer, P., and F. Parish. 1952. "The Reduction of Fluctuations in the Incomes of Primary Producers."
Economic Journal 62(248):750-80.
Bell. C. 1988. "Credit Markets, Contracts and Interlinked Transactions." In Handbook of Development
Economics. vol. 1. ed. H. Chenery and T. N. Srinivasan. Amsterdam: Elsevier.
Besley, T. 1994. "Savings, Credit and Insurance." In Handbook of Development Economics, vol. 3. ed.
A. J. Behrman and T. N. Srinivasan. Amsterdam: Elsevier.
Besley, T., and R. Kanbur. 1990. "The Principles of Targeting." Policy Research and External Affairs
Working Paper 385. World Bank, Washington, D.C.
Bevan, D., P. Collier, and J. Gunning. 1990. Controlled Open Economies: A Neo-Classical Approach to
Structuralism. Oxford: Clarendon.
Binswanger, H. 1980. "Attitudes toward Risk: Experimental Measurement in Rural India." American
Journal of Agricultural Economics 62(3):39 5-407.
Binswanger, H., and J. Mclntire. 1987. "Behavioral and Material Determinants of Production Rela-
tions in Land-Abundant Tropical Agriculture." Economic Development and Cultural Change 36(1):
73-97.
Bourguignon, F., and A. Gelb. 1988. "Venezuela: Absorption without Growth." In Oil Windfalls: Bless-
ing or Curse?, ed. Alan Gelb and Associates. New York: Oxford University Press.
Budd, N. 1991. "Financing Operation for Commodity Producers: A Guide to Pre-Export Financing of
Physical Commodities in Origin Countries." International Financial Law Review 10(1):24-8.
Caine, S. 1954. "Instability of Primary Product Prices: A Protest and a Proposal." Economic Journal
64(2 5 5):610-14.
Chambers, R., and J. Quiggin. 2002. "Optimal Producer Behavior in the Presence of Area-Yield Crop
Insurance." American Journal of Agricultural Economics 84(2): 320)-34.
Cheung, S. 1969. The Theory of Share Tenancy with Special Application to Asian Agriculture and the First
Phase of Taiwan Land Reform. Chicago: University of Chicago Press.
Christiansen, L., H. Alderman, and T. Yamono. 2003. "Child Growth, Shocks and Food Aid in Rural
Ethiopia. " Policy Research Working Paper 3128. World Bank, Washington, D.C.
Claessens, S. 1991. "Integrating Commodity and Exchange Rate Risk Management: Implications for
External Debt Management." In Commodity Risk Management and Finance, ed. T. Priovolos and
R. Duncan. Oxford: Oxford University Press.
Claessens, S.. and R. C. Duncan, eds. 1993. Managing Commodity Price Risk in Developing Countries.
Baltimore, Md.: Johns Hopkins University Press.
Claessens, S.. and P. Varangis. 1993. "Implementing Risk Management Strategies in Costa Rica's
Coffee Sector." In Managing Commodity Price Risk in Developing Countries, ed. S. Claessens and
R. Duncan. Baltimore, Md.: Johns Hopkins University Press.
222                                        The World Bank Research Observer, vol. 19, no. 2 (Fall 2004)



Clay, E. 1986. "Rural Public Works and Food-for-Work: A Survey." World Developmnent 14(1()-11):
1237-52.
Coate, S., and M. Ravallion. 1993. "Reciprocity without Commitment: Characterization and Perfor-
mance of Informal Insurance Arrangements." Jouirnzal of Developmernt Econtomics 40)(1): 1-24.
Collier, P. 2002. "The Macroeconomic Repercussions of Agricultural Shocks and Their Implications for
Insurance." WIDER Working Discussion Paper 202/46. World Institute for Development Economics
Research, Helsinki.
Coulter, J., and G. Onumah. 2002. "The Role of Warehouse Receipt Systems in Enhanced Commodity
Marketing and Rural Livelihoods in Africa." Food Policy 27(4): 319-37.
Czukas, K., M. Fafchamps, and C. Udry. 1998. "Drought and Savings in West Africa: Are Livestock a
Buffer Stock?" Jouirnal of Developmenit Economics 55(2 J:2 73-305.
Dawe, D. 1996. "A New Look at the Effects of Export Instability on Investment and Growth." iWorld
Developmnent 24(12):1905-14.
Deaton, A. 1992. "Commodity Prices, Stabilization, and Growth in Africa." Center of International
Studies Discussion Paper 166. Princeton University, Princeton, N.J.
-   1997. The Analysis of Househlold Surveys: A Microeconomtietric Approach to Development Policy.
Baltimore. Md.: Johns Hopkins University Press.
Deaton, A., and G. Laroque. 1992. "On the Behaviour of Commodity Prices." Review of Economnic Studies
59(1):1-23.
Dehn. J. 2000. "Commodity Price Uncertainty in Developing Countries." Policy Research Working
Paper 2426. World Bank, Washington. D.C.
Dercon, S. 1996. "Risk, Crop Choice and Savings: Evidence from Tanzania." Economic Development and
Cultural C blt.w's 44(3):385-514.
.1998. "Wealth. Risk and Activity Choice: Cattle in Western Tanzania." Journal of Development
Economics 55(1(:1-42.
.2002. "Income Risk. Coping Strategies. and Safety Nets." World Bank Research Observer 17(2):
141-66.
Dercon, S., and P. Krishnan. 200 3. "Food Aid and Informal Insurance." Centre for the Study of African
Economies Working Paper 200-01. Oxford University.
Dillon. J. L. 1971. "An Expository Review of Bernoullian Decision Theory in Agriculture: Is Utility
Futility?" Review of'Marketin.q and Agricultural Economics 39(1): 3-80.
Dillon, J. L., and P. L. Scandizzo. 1978. "Risk Attitudes of Subsistence Farmers in Northeast Brazil:
A Sampling Approach." American Journal of Agricultural Econornics 60(3):425-35.
Drought Policy Review Task Force. 1990. National Droughlt Policy. Final Report, vol. 1. Canberra,
Australia: Australian Government Publishing Service.
Eboh, E. 2000. "Rural Informal Savings and Credit Associations as Risk Managers and the Lessons for
the Design and Execution of Rural Credit Schemes in Nigeria." Africani Developmenit Review 12(2):
23 3-62.
Fafchamps, M. 1999. Rural Poverty, Risk and Development. FAo Economic and Social Development Paper
144. Food and Agriculture Organization, Rome.
Fafchamps, M., and B. Minten. 2001. "Property Rights in a Flea Market Economy." Econornic Develop-
mernt and Cultural Clhange 49(2):229-68.
Faruqee, R., and J. Coleman. 1996. "Managing Price Risk in the Pakistan Wheat Market." World Bank
Discussion Paper 334. Washington, D.C.
Fisher F. 1963. "Theoretical Analysis of the Impact of Food Surplus Disposal on Agricultural Produc-
tion in Underdeveloped Countries." Journial of Farm Economics 45(3):863-65.
Donald F. Larson, lock R. AnIderson, anld Panos lVarargis                             223



Frechette, D. 2000. "The Demand for Hedging and the Value of Hedging Opportunities." American Jour-
ial of Agricultuiral Econooiics 82(4): 897-907.
Freund. R. 1956. "The Introduction of Risk into a Programming Model." Econometrica 24(3):
25 3-6 3.
Friedman, M. 1954. "The Reduction of Fluctuations in the Incomes of Primary Producers: A Critical
Comment." Econornic Journal 64(256):698-703.
Gardner. B. 1989. "Rollover Hedging and Missing Long-term Futures Markets." Americanl Journal of
Agricultlural Economics 71(3):311-18.
.2002. "Risk Created by Policy in Agriculture." In A Comnprehensive Assessment of the Role of Risk
in U.S. Agricolture. ed. R. E. Just and R. D. Pope. Boston: Kluwer.
Gautam, M.. P. Hazel]. and H. Alderman. 1994. "Rural Demand for Drought Insurance." Policy
Research Working Paper 1383. World Bank, Washington. D.C.
Gemmill, G. 1985. "Forward Contracts or International Buffer Stocks? A Study of Their Relative
Efficiencies in Stabilizing Commodity Export Earnings." EconomicJourlal 95(378):400-417.
Gilbert, C. 1985. "Futures Trading and the Welfare Evaluation of Commodity Price Stabilisation."
Economic Journal 95(379):637-61.
. 198 7. "International Commodity Agreements: Design and Performance." Wlorld Developmenit
15(5):591-616.
.1995. "International Commodity Control: Retrospect and Prospect." Policy Research Working
Paper 1545. World Bank, Washington. D.C.
- 1996. "International Commodity Agreements: An Obituary Notice." World Developmenlt
24(1):1-19.
Glauber. J., and K. Collins. 2002. "Risk Management and the Role of the Federal Government." In
A Comprehensive Assessmenit of the Role of Risk inz U.S. Agricuilture. ed. R. E. Just and R. D. Pope.
Boston: Kluwer.
Goldstein, M.. A. de Janvry. and E. Sadoulet. 2002. "Is a Friend in Need a Friend Indeed?: Inclusion and
Exclusion in Mutual Insurance Networks in Southern Ghana." WIDER Discussion Paper 20)02/25.
World Institute for Development Economics Research, Helsinki.
Grosh, M. 1998. "The World Bank's Living Standards Measurement Study Household Surveys." Jour-
nal of Economic Perspectives 12(1): 187-96.
Grosh. M.. and P. Glewwe. 1995. "A Guide to Living Standards Surveys and Their Data Sets." Living
Standards Measurement Study Working Paper 120. World Bank, Washington, D.C.
Halcrow. H. 1 949. "Actuarial Structures for Crop Insurance." Journal of Farm Economnics 21(2):418-43.
Hardaker. J.B., R.B.M. Huirne, J.R. Anderson, and G. Lien. 200)4. C. l-j.i ,iWi1 Risk inz Agriculture. 2nd ed.
Wallingford, UK: CABI Publishing.
Hart, C., B. Babcock, and D. Hayes. 2001. "Livestock Revenue Insurance." Journal of F'utuires Markets
21(6):5 5 3-80.
Harwood, J., R. Dismukes. M. Vandeverr. and R. Heifner. 2000. "From Risk-Pooling to Safety Nets:
Crop and Revenue Insurance in the United States." In Incomwe Risk Manaclgemtient in Agricultuire. Paris:
Organisation for Economic Co-operation and Development.
Harwood, J.. R. Heifner, K. Coble, J. Perry. and A. Somwaru. 1999. "Managing Risk in Farming: Con-
cepts, Research. and Analysis." Economic Research Service Agricultural Economic Report 774. U.S.
Department of Agriculture, Washington, D.C.
Hausmann, R., and M. Gavin. 1996. "Securing Stability and Growth in a Shock-Prone Region:
The Policy Challenge in Latin America." In Security Stability andtl Growth in Latin Anme'ica, ed.
R. Hausmann and H. Reisen. Paris: Organisation for Economic Co-operation and Development.
224                                       The Wlorld Bank Research Observer, vol. 19, nmo. 2 (Fall 2004)



Hazell, P. B. R. 1982. "Application of Risk Preference Estimates in Firm-Household and Agricultural
Sector Models." American Journal of Agricultural Economlics 64(2):384-90.
- 1992. "The Appropriate Role of Agricultural Insurance in Developing Countries." Jourlnal of
International Development 4(4): 567-81.
Heady, E. 0. 1952. EconomizicsojfAgriculturalProductionandResource Use. EnglewoodCliffs, N.J.: Prentice-Hall.
Industries Assistance Commission. 1978. Rural Incomeus Fluctuations. Report 161. Canberra, Australia:
Australian Government Publishing Service.
International Task Force on Commodity Risk Management in Developing Countries. 1999. Dealing
llith Commodity Price Volatility in Developing Countries: A Proposal Jor a Market-Based .4pproach.
World Bank, Washington, D.C.
Jacoby, H., and E. Skoufias. 1997. "Risk, Financial Markets, and Human Capital in a Developing Coun-
try." Review of Economic Studies 64(3): 311-35.
Jalan, J., and M. Ravallion. 1998. "Transient Poverty in Post-Reform Rural China." Journal of Compara-
tive Economics 26(2):338-57.
. 1999. "Are the Poor Less Well Insured? Evidence on Vulnerability to Income Risk in Rural
China." Journal of Development Economics 58(1):61-8 1.
Johnson, D. G. 1947. Forward Pricesfor Agriculture. Chicago: University of Chicago Press.
.1950. "Resource Allocation under Share Contracts." Journal of Political Economy 58(2):111-23.
Kannapiran, C. 2000. "Commodity Price Stabilization: Macroeconomic Impacts and Policy Options."
Agricultural Economizics 2 3(1): 17-30.
Keynes, J. 1943. "War Cabinet Committee on Reconstruction Problems: The International Reg-
ulation of Primary Products." Reprinted in Collected Wllriting of John Al\aljnard Kezynes, vol. 27, ed.
D. Moggridge. London: Macmillan and Cambridge UJniversity Press. 1980.
Knight, T., and K. Coble. 1997. "A Survey of Literature on U.S. Multiple Peril Crop Insurance Since
1980." Revieiw of Agricultural Econornics 19(1): 128-56.
Knudsen, O., and J. Nash. 1990. "Domestic Price Stabilization Schemes in Developing Countries."
Economic Developrnent and Cultural Chanige 3 8(3): 539-58.
Knudsen, O., and A. Parnes. 1975. Trade Instability and Economiiic Development. Lexington, Mass.:
Lexington Books.
Kochar. A. 1995. "Explaining Household Vulnerability to Idiosyncratic Income Shocks. " American Eco-
nomizic Review 85(20:159-63.
Krammer, R. 1983. "Federal Crop Insurance." Agricultural History 97:1 81-200.
Kranton, R. 1996. "Reciprocal Exchange: A Self-Sustaining System." Americani Economic Review, 86(4):
830-51.
Krueger. A., M. Schiff, and A. Valdes. 1991. The Political Ecotnonnjy oJ Agricultural Pricing Policy. vols.
1-3. Baltimore, Md.: Johns Hopkins UJniversity Press.
Kunreuther, H. 1978. Disaster Inisuranice Protection: Public Policy Lessons. New York: Wiley.
Lanjouw J., and P. Lanjouw 1995. "Rural Nonfarm Employment: A Survey." World Bank Working
Paper 1463. Washington, D.C.
Lapan, H., and G. Moschini. 1994. "Futures Hedging under Price. Basis and Production Risk." American
Journal of Agricultural Economics 76(3):465-77.
-         1996. "Optimal Price Policies and the Futures Markets." Economic Letters 5 3(1):175-82.
Larson, D., and J. Coleman. 1993. "The Effects of Option Hedging on the Costs of Domestic Price Stabil-
ization Schemes." InCommnoditye,\ I '"i',l,. m1. ,am and Finanice, vol.2, ed. S. Claessens andR. Duncan.
Oxford: Oxford University Press.
Donald F. Larson., lock R. Anderso,,. anid Patios Varangis                             225



Larson, D., and Y. Mundlak. 1997. "The Inter-Sectoral Migration of Agricultural Labor." Econornic
DevelopmnentantdCulturalClhange45(2):295-319.
Larson, D., and F. Plessmann. 2002. "Do Farmers Choose to be Inefficient? Evidence From Bicol,
Philippines." Policy Research Working Paper 2787. World Bank. Washington, D.C.
Larson, D., P. Varangis. and N. Yabuki. 1998. "Commodity Risk Management and Development." Pol-
icy Research Working Paper 1963. World Bank. Washington, D.C.
Lee, 1. 1953. "Temperature Insurance-An Alternative to Frost Insurance in Citrus." Journal of Farm
Econornics 35(1):15-28.
Lewis, S. 1955. The Theory of Economnic Growthl. London: Allen and Unwin.
Ligon, E., J. Thomas, and T. Worrall. 2002. "Informal Insurance with Limited Commitment: Theory
and Evidence from Village Economies." Review of Economic Studies 69(l):20)9-44.
Little P.. K. Cellarius, D. Coppock, and C. Barrett. 2001. "Avoiding Disaster: Diversification and Risk
Management among East African Herders." Developmnent and( I'. ',.' 321(3):401-33.
Lopes, M., and G. Dias 1986. "The Brazilian Experience with Crop Insurance Programs." In Crop Inisuir-
anice Jbr Agricultural Development: Issues and Experience. ed. P. Hazell. C. Pomareda, and A. Valdes.
Baltimore, Md.: Johns Hopkins tJniversity Press.
Lutz, M. 1999. "Commodity Terms of Trade and Individual Countries' Net Barter Terms of Trade: Is
There an Empirical Relationship?" Journial of Interniational Developmrzenit l 1(6):8559-70.
MacBean, A. 1966. Export Instability and Economic Developnient. Cambridge, Mass.: Harvard University
Press.
Mahul, 0. 1999. "The Design of an Optimal Area Yield Crop Insurance Contract." Geneva Papers on Risk
and Insuranice Theory 24(2):1 59-71.
Martin. E., and P. Bryde. 1999. "Grain Receipts in Economies in Transition: An Introduction to Fin-
ancing of Warehouse Receipts." In Agricultural Finance and Credit Infrastructure in Transitioni Econo-
mies: Proceedings of OECD Expert Meeting. Paris: Organisation for Economic Co-operation and
Development.
Massell, B. 1969. "Price Stabilization and Welfare." Quarterly Journall of Economrlics 83(2):234-89.
19770. "Some Welfare Implications of International Price Stabilization." Jou1rnll of Political
Economny 78(2):404-17.
Maxwell, S., and H. Singer. 1979. "Food Aid to Developing Countries: A Survey." World Developuient
7(2):225-46.
McCarthy, N. 2)03. "Demand for Rainfall-Index Based Insurance: A Case Study from Morocco.' Envi-
ronment and Production Technology Division Discussion Paper 1()6. International Food Policy
Research Institute, Washington. D.C.
McConnell, D.. and J. L. Dillon. 1997. Farm Managemiientfor Asia: A Systems Approach. FAo Farm Systems
Management Series 1 3. Rome: Food and Agriculture Organization.
McKinnon. R. 1967. "Futures Markets. Buffer Stocks, and Income Stability for Primary Producers."
Journial of Political Econiomy 75(6):844-61.
McPeak, J.. C. Barrett, and B. Christopher, 20)01. "Differential Risk Exposure and Stochastic Poverty
Traps among EastAfrican Pastoralists." AmerincanJournal oJAgricultlural Economrics 83(3):674-79.
Meerman, J. 1997. Reforminig Agricultlure: The World Banl;k Goes to Market. OE) Evaluation Study.
Washington. D.C.: World Bank, Operations Evaluation Department.
Miranda. M. 1991. "Area-Yield Crop Insurance Reconsidered." Amlericanrl Joura-ltl oJ Agricultuiral Eco-
nomnics 73(3):233-42.
Miranda, M.. and J. Glauber. 1991. "Providing Crop Disaster Assistance through a Modified Deficiency
Payment Program." Amnerican Jou-rnal ojfAgricuiltuir(al Economnics 7 3(4):12 3 3-43.
226                                         The lWorld Ban7k Resetrchli Observer. vol. 1 9. no. 2 (Fall 2004)



-   1997. "Systemic Risk, Reinsurance, and the Failure of Crop Insurance Markets." American
Journial of Agricultur/al Econiomitics 79(1):20)6-15.
Mishra, P. 1996. Agricultural Risk. Insuranzce antd Incomiie: A Study of the Impact and Design of India's
Comprehlensive Crop Insuiraniice Schzemiie. Brookfield, Vt.: Avebury.
Morduch, J. 1998. "Between the Market and State: Can Informal Insurance Patch the Safety Net?" Devel-
opment Discussion Paper 621. Harvard Institute for International Development, Cambridge, Mass.
.20))2. "Consumption Smoothing Across Space: Testing Theories of Risk-Sharing in the ICRISAT
Study Region of South India." WIl)ER Discussion Paper 2)(02/5 5. United Nations University. World
Institute for Development Economics Research, Helsinki.
Morgai. R., P. Winters. E. Sadoulet. and A. de Janvry. 2002. "Localized and Incomplete Mutual Insur-
ance." Journll1 oj Developinemit Economnics 67(2):24 5-74.
Moscardi, E., and A. de Janvry. 1977. "Attitudes Toward Risk among Peasants: An Econometric
Approach." Americanii Jourmadil of Agricultural Economiiics 59(4):710-16.
Moschini. G.. and D. Hennessy. 200)1. "Uncertainty, Risk Aversion and Risk Management for Agricul-
tural Producers" In Handbook of Agricultural Economnics. vol. 1A. Agricultural Production. ed. B. Gardner
and G. Rausser. Amsterdam: Elsevier Science.
Muller. A.. and M. Grandi. 2000. "Weather Derivatives: A Risk Management Tool for Weather-
Sensitive Industries. " Geneva Papers on Risk antd Insurance: lssues and Practice 25(2):273-87.
Mundlak, Y.. D. Cavallo. and R. Domenech. 1993. Agriculture and Economic Growvthl in Argentina,
1 91 3-84. Research Report 76. Washington. D.C.: International Food Policy Research Institute.
Musser, W., and G. Patrick. 2002. "How Much Does Risk Really Matter to Farmers?" In A Comprehensive
Assessmnent of thie Role of Risk in U.S. Agricultlure. ed. R. E. Just and R. D. Pope. Norwell. Mass.: Kluwer.
Myers. R. 1988. "The Value of Ideal Contingency Markets in Agriculture." Aniericani Journial of Agricul-
tu'al Economiiics 70(2):255-67.
Myers. R., and J. Oehmke. 198 8. "Instability and Risk as Rationales for Farm Programs." In Agricultural
Stability amid Farimi Programis: Concepts, Evidence, and Imuplicatiomis, ed. D. Sumner. Boulder, Colo.:
Westview.
Myers. R., and S. Thompson. 1991. "Optimal External Debt Management with Commodity-linked
Bonds." In Commrlodity Risk A\anzagemtienit amid Finanice. ed. T. Priovolos and R. Duncan. Oxford: Oxford
UJniversity Press.
Newbery, D.. and J. Stiglitz. 1981. The Theory oJ Commiodity Price Stabilizatiomi. Oxford: Clarendon Press.
North, D. 1994. "The Evolution of Efficient Markets" In History in Capitalismn in Context: Essays on
Economiic Devce1opmnent anld Cultuiral ( l.-lj' iln Honor of R. M Hartwell. ed. J. James and M. Thomas.
Chicago: lJniversity of Chicago Press.
O'Hara. M. 1984. "Commodity Bonds and Consumption Risks." Journal of Finance 39(3):193-20)6.
Overdahl. J. 1987. "The t]se of Crude Oil Futures by the Governments of Oil-Producing States." Journal
of Futures Markets 7(6):60)3-18.
Paxson, C. 199 3. "Consumption and Income Seasonality in Thailand." Journal of Political Econonmy
10)(1):39-75.
Peck, A. 21)()1. "The Development of Commodity Exchanges in the Former Soviet Union, Eastern
Europe. and China." Australiant Economnic Papers 40)(4):437-60.
Platteau, 1. 1997. "Mutual Insurance as an Elusive Concept in Traditional Rural Communities.." Jourmal
of Developmnent Studies 3 3(6):764-96.
Pomareda. C. 1986. "The Financial Viability of Agricultural Insurance." In Cr-op Insuranicejor Agricul-
tural Developmnent: Issutes and Experience, ed. P. Hazell, C. Pomareda, and A. Vald6s. Baltimore,
Md.: Johns Hopkins University Press.
Donald F. tarsoni. Jock R. Ataiderson. tit)t Poios Varariiis                              227



Pratt, J. 1964. "Risk Aversion in the Small and in the Large." Econometrica 32(1/2): 122-36.
Prebisch, R.1950. The Econonmic Development of Latin America and Its Principal Problems. New York: UN
Economic Commission for Latin America.
Priovolos, T., and R. Duncan, eds. 1991. Commodity Risk Management and Finance. Oxford: Oxford
University Press.
Quiggin, J. 1986. "A Note on the Viability of Rainfall Insurance." Australiani Journal of Agricultural
Economics 30(l():63-9.
Quiggin, J. C.. and J. R. Anderson. 1979. "Stabilisation and Risk Reduction." Australian Journal of
AgriculturalEconiomizics 23(3):191-206.
.1981. "Price Bands and BufferFunds." Economk'icRecord 57(196):67-73.
Quiggin, J., G. Karagiannis. and J. Stanton. 1993. "Crop Insurance and Crop Production: an Empirical Study
of Moral Hazard and Adverse Selection." AustralianjJournalofAgricultuiralEconomics 37(2): 95-11 3.
Ravallion, M. 1991. "Reaching the Rural Poor through Public Employment: Arguments, Evidence,
and Lessons from South Asia." Worid Bank Research Observer 6(2): 153-75.
.1997. "Famines and Economics." Journral of Economic Literature 35(3): 1205-42.
Ray. P. 1967. Agricultural Insuirance: Theory and Practice and Application to Developing Countries. Oxford:
Pergamon.
Reardon, T.. C. Delgado, and P. Matlon. 1988. "Determinants and Effects of Income Diversification
Amongst Farm Households in Burkina Faso." Journal of Development Studies 28)(1):264-96.
Reinikka, R., and P. Collier, eds. 20)01. Uganda's Recovery : The Role of Farms, Firms and Government.
Washington, D.C.: World Bank.
Reutlinger, S. 1970. "Techniques for Project Appraisal under Uncertainty." World Bank Staff Occa-
sional Papers 10. Washington. D.C.
Rhamato, D. 1991. Famine and Survival Strategies: A Case Study fromn Northeast Ethiopia. Uppsala.
Sweden: Nordiska Afrikainstitutet.
Robertson. A. 198 7. Thze Dyjnamiiics of Productive Relationshlips: African Share Contracts in Comparative Per-
spective. New York and Melbourne: Cambridge tJniversity Press.
Robison, L., and P. Barry. 1987. The Competitive Firmn's Response to Risk. New York: Macmillan.
Rodrik, D. 1998. "Where Did All the Growth Go? External Shocks, Social Conflict, and Growth Col-
lapses." NBER Working Paper 6350. National Bureau of Economic Research, Cambridge, Mass.
Rolfo. J. 1980. "Optimal Hedging under Price and Quantity Uncertainty: The Case of a Cocoa Pro-
ducer." Journal of Political Economni 88(11): 100-116.
Rose, E. 2)01. "Ex Ante and Ex Post L.abor Supply Response to Risk in a Low-Income Area." Journal of
Developmyient Economics 64(2): 371-88.
Rosenzweig, M.2001. "Savings Behaviour in Low-Income Countries." Oxford Review of Economic Policy
17(1 ):40-54.
Rosenzweig, M., and 11. Binswanger. 1993. "Wealth, Weather Risk and the Composition and Profit-
ability of Agricultural Investments." Econiomic Journal 103(1): 56-78.
Roumasset, J. A., M. Rosegrant, U. Chakravorty. and J. R. Anderson. 1989. "Fertilizer and Crop Yield
Variability: A Review." In V7ariability in Grain Yields, ed. J. R. Anderson and P. B. R. Hazell. Baltimore.
Md.: Johns Hopkins University Press.
Sakurai. T., and T. Reardon. 199 7. "Potential Demand for Drought Insurance in Burkina Faso and Its
Determinants." American Journal ojfAgricultural Economnics 79(4):1193-207.
Sapsford. D., and V. N. Balasubramanyam. 1999. "Trend and Volatility in the Net Barter Terms of
Trade, 190(0-92: New Results from the Application of a (Not So) New Method." Journal of Inter-
nationial Development 11(6):851-57.
228                                         The World Bank Research Observer, vol. 19. no. 2 (Fall 2004)



Satyanarayan, S., E. Thigpen, and P. Varangis. 1993. "Hedging Cotton Price Risk in Francophone African
Countries." Policy Research Working Paper 1233. World Bank. Washington. D.C.
Schultz. T. W. 1945. Agricnltnire in an Unstable Econiomiy. New York: McGraw-Hill.
- 1960. "Impact and Implication of Foreign Surplus Disposal on Underdeveloped Countries:
Value of [l.S. Farm Surpluses to Underdeveloped Countries." Journzal of Farmn Econiomnics 42(4):
1019-30.
.1964. Transforminig Traditional Agriculture. New Haven, Conn.: Yale University Press.
Sen, A. 1960. "Impact and Implication of Foreign Surplus Disposal on Underdeveloped Countries-The
Indian Perspective." Jourl nal of Farmt Economics 4214): 10 31-42.
.1981. Poverty and Faminies: An Essay on Entitlement antd Deprivation. Oxford: Oxford University
Press.
Shaw. J. 2002. "Food Aid in Sub-Saharan Africa: Policy Lessons for the Future." Calladianl Jourlnal of
Developmnent Studies 2 3(3):571-99.
Singer. H. 1950. "The Distribution of Gains between Investing and Borrowing Countries." Anmerican
Economiic Review 4()(2):473-85.
Skees. J. 1999. "Opportunities for Improved Efficiency in Risk Sharing tJsing Capital Markets."
Americani Journal of Agricultural Economiiics 81(5):1228-33.
- )2000. "Agricultural Insurance Programmes: Challenges and Lessons Learned." In Incomne Risk
Manzageinent in Agriculture. Paris: Organisation for Economic Co-operation and Development.
Skees, J.. and A. Enkh-Amgalan. 2002. "Examining the Feasibility of Livestock Insurance in Mongolia."
World Bank Working Paper 2886. Washington, D.C.
Skees. J., J. Black, and B. Barnett. 1997. "Designing and Rating an Area Yield Crop Insurance Con-
tract." Amnericani Journzal oJ Agricultzural Economics 79(2):430-38.
Skees. J., P. Hazell. and M. Miranda. 1999. "New Approaches to Crop Yield Insurance in Developing
Countries." Environment and Production Technology Division Discussion Paper 55. International
Food Policy Research Institute. Washington, D.C.
Skees, J., P. Varangis, D. F. Larson. and P. Siegel. 2002. "Can Financial Markets Be Tapped to Help Poor Peo-
ple Cope with Weather Risks? " Policy Research Working Paper 2812. World Bank, Washington. D.C.
Smith. K., C. Barrett. and P. Box. 200 1. "Not Necessarily in the Same Boat: Heterogeneous Risk Assess-
ment among East African Pastoralists." Jourlal of De'elopmtienit Studies 3 7(5):1-30.
Smith. V.. and A. Baquet. 1996. "Almost Ideal Area Yield Crop Insurance Contracts." Agricnltnral and
Resource Econiomttics Review 23(1 (:75-83.
Stern. N. 1989. "Peasant Economy." In The Ne"' Palgrave: Economniic Dcve0lopmcnmt, ed. J. Eatwell,
M. Milgate. and P. Newman. New York: Norton.
Timmer, C. 1989. "Food Price Policy: The Rationale for Government Intervention." FoodPolicy 14(1):1 7-27.
Townsend, R. 1977. "The Eventual Failure of Price Fixing Schemes." Journial of Political Economyiy
91(2):190-99.
.199 5. "Consumption Insurance: An Evaluation of Risk-Bearing Systems in Low-Income Econ-
omies." Journatl of Econonmiic Perspectives 9(3):83-10)2.
Turvey. C. 1992 "An Economic Analysis of Alternative Farm Revenue Insurance Policies." Canadian
Jlournzal ojfAgricuiltniral Econtomics 40(3):40)3-26.
-(2001. "Weather Derivatives for Specific Event Risks in Agriculture." Revietw oJ Agricultural Eco-
nzomilics 23(2):33 3-51.
tJdry, C. 1991. "Risk, Insurance and Default in a Rural Credit Market: An Empirical Investigation
in Northern Nigeria." Yale Economic Growth Center Discussion Paper 636. Yale University,
New Haven, Conn.
Donialtd F. Larsoni, btck R. Anidcrson, amid 1'aiios V arangis                         229



.1995. "Risk and Saving in Northern Nigeria." Americani Economic Review 85(5):1287-300.
Valgren, V. N. 1922. "Crop Insurance: Risks, Losses and Principles of Protection." USDA Bulletin 1043.
U.S. Department of Agriculture, Washington, D.C.
Varangis, P., and D. F. Larson. 1996. "Dealing with Commodity Price Uncertainty." Policy Research
Working Paper 1667. World Bank. Washington. D.C.
Vercammen, J. 2000. "Constrained Efficient Contracts for Area Yield Crop Insurance." American Jour-
nal of Agricultural Economics 82(4):856-64.
von Braun, J., ed. 1995. Employment for Poverty Reductioni and Food Security. Washington, D.C.: Inter-
national Food Policy Research Institute.
Wang, H. 2000. "Zone-Based Group Risk Insurance." Journral of Agricultural and Resource Economics
25(2):411-31.
Williams, J. 2001. "Commodity Futures and Options." In Handbook of Agricultural Economics, vol. 1B,
Marketing, Distribution and Consumllers. ed. B. Gardner and G. Rausser. Amsterdam: Elsevier Science.
Wolpin, K. 1982. "A New Test of the Permanent Income Hypothesis: The Impact of Weather on the
Income and Consumption of Farm Households." Internzational Economic Review 23(3):583-94.
World Bank. 1990. World Development Report 1990: Poverty. New York: Oxford IJniversity Press.
.2001. Social Protection Sector Strategy: From Safety Net to Springboard. Washington, D.C.
Wright, B. 1988. "Storage, Stability, and Farm Programs." In Agricultural Stability and Farm Programs:
Concepts. Evidence, and Implications. ed. D. Sumner. Boulder. Colo.: Westview.
Wright, B., and J. Hewitt. 1994. "All Risk Crop Insurance: Lessons from Theory and Experience."
In Economics of Agricultural Insurance: Theory and Evidence. ed. D. Hueth and W. Furtan. Boston:
Kluwer.
Wright, B., and D. Newbery. 1991. "Financial Instruments for Consumption Smoothing by Commodity-
Dependent Exporters." In Coimnodity Risk Management and Finanice, ed. T. Priovolos and R. C. Duncan.
Oxford: Oxford University Press.
Wright, B., and J. Williams. 1990. "The Behavior of Markets for Storable Commodities." Paper
presented at the Annual Conference of the Australian Agricultural Economics Society. Brisbane,
Australia.
Yamauchi, T. 1986. "Evolution of the Crop Insurance Program in Japan." In Crop Insurancefor Agricuil-
tural Development: Issues and Experience, ed. P. Hazell, C. Pomareda, and A. Valdes. Baltimore, Md.:
Johns Hopkins University Press.
Yaron. J., M. P. Benjamin, and G. L. Piprek. 1997. "Rural Finance: Issues, Design. and Best Practice."
Environmentally and Socially Sustainable Development Studies and Monographs 14. World Bank,
Washington. D.C.
230                                        The World Bank Research Observer. vol. 19. no. 2 (Fall 2004)



Oil, Macroeconomics, and Forests:
Assessing the Linkages
Sven Wunder . William D. Sunderlin
How does an oil boom affect theforest cover of tropical oil-exporting countries? Wlhat ,mac-
roeconomnic linkages and policies are decisive? A comparison of research findings on long-
run land-use changes in eight tropical developing economies reveals that the direct plhysical
impacts of the oil industry on forests are unquestion7ably less than its derived macroeco-
nomic impact. In most cases oil wealth indirectly but significantly protects tropicalforests.
The core mechanism is that oil rents cause macroecono,mic "Dutch disease" decreasing the
price competitiveness of agriculture and logging, strongly diminishing pressures for forest
degradation and deforestation. But domestic policy responses to oil wealth are also vital
determinanzats of the forest outcome. When governments use oil wealth for urban spending
sprees, this reinforces tlhe core effect by pulling more labor out of land-using and forest-
degrading activities. When oil revenuesfinanice road construction or frontier colonization,
however, the core forest-protective effect can be reversed. Repeated currency devaluation
and import protection of land-using domestic sectors also increase pressures on forests.
Other international capital transfers, like bilateral credits, aid, or debt relief, can have
impacts si,milar to those of oil wealtlh, either alleviating pressures on forests or aggravating
specific forest-detri,mental policies. These insiglhts point to forest-friendly safeguards that
can realistically be mnade in thze design of structural adjustment programs, considering the
important tradeoffs between development and conservation objectives.
Over the past decade there has been growing interest in the relationship between
macroeconomics and the environment, including the links to deforestation and for-
est degradation. This is due to a growing recognition that the fate of tropical forests
is determined by factors outside forests and the forestry sector, factors which have
often been called the underlying causes of deforestation (Contreras-Hermosilla
2000). A wealth of theoretical economic models has been developed, with multi-
country regression models as the predominant category (Angelsen and Kaimowitz
The World Bank Research Observer, vol. 19, no. 2,
Cc The International Bank for Reconstruction and Development / THE WORLD BANK 2004: all rights reserved.
doi: 10.1(093/wbro/lkhOI8                                             19:2 31-2 57



1999). However, the United Nations Food and Agriculture Organization (FAO) def-
orestation data normally used in these exercises are highly problematic (see later
discussion). Thus, it is dubious to what extent the observed cross-country patterns
can be given a time-series interpretation (Kaimowitz and Angelsen 1998). Cross-
country analysis should thus be complemented by country-comparative case studies
(Kaimowitz and others 1998; Reed 1992, 1996: Wood and others 2000). These
studies, however, have analyzed mostly short-run effects of macroeconomic change,
notably the environmental effects of structural adjustment.
The comparative research described here was conducted by the Center for Inter-
national Forestry Research (CIFOR) and is documented more extensively elsewhere
(Wunder 2003b). It looks at the macroeconomic determinants of changes in land
use and forest cover since the 19 70s. Eight developing economies with tropical for-
ests were analyzed, five with detailed case studies (Cameroon, Ecuador, Gabon,
Papua New Guinea, and Venezuela) and referred to as primary cases and three
(Indonesia, Mexico, and Nigeria) with more summary-type or secondary case stud-
ies. The countries were selected to cover a variety of conditions. including geo-
graphic location, country size, income per capita, and proportion of forest coverage.
A subsidiary criterion was the existence of previous CIFOR research on land-use
change (Cameroon, Indonesia, and Ecuador).
This article focuses mainly on the five primary case study countries. For forest
impacts, the concentration is on forest conversion to other land uses and deforesta-
tion, defined as a (temporary or permanent) removal of trees to less than 10 percent
crown cover, which is similar to the FA)'S definition (FAO 2000). Selective logging is
thus not deforestation but may degrade forests and enable conversion. The link from
oil wealth and macropolicies to logging and forest degradation is analyzed elsewhere
(Wunder 2003a).
All of the case study countries are tropical countries that export oil, a choice made
for two reasons. First, the macroeconomic "laboratory" of oil exporters offers a good
opportunity to study links between external economic changes and forests. Oil econ-
omies often fluctuate dramatically due to heavy reliance on a single export commod-
ity with unstable world market prices. The policy responses to frequent boom and
bust cycles also differ enormously, both over time and across countries (Bevan and
others 1999a; Collier, Gunning, and Associates 1999; Little and others 1993: Neary
and van Wijnbergen 1986). This is because oil wealth buys governments higher
degrees of freedom in defining individual policy responses (Gelb and Associates
1988: Karl 1997). This large variation makes it easier to track the eventual effects of
different policies on forests.
Second, earlier studies provide support for the hypotheses that on average oil- and
mineral-exporting tropical countries have more forests left and lose them at a slower
rate than nonmineral-exporting countries. This is true not only for a bivariate group
comparison but also for multivariate cross-country deforestation models controlling
232                                hlle lVorld Baunk Research Observer, vol. 19. nio. 2 (Fall 2004)



for third causal factors (Mainardi 1998; Sunderlin and Wunder 2000). The FAO'S
Forest Resource Assessmlenit 2000 data indicate that specialized mineral exporters,
which derive more than 50 percent of their export revenues from oil and minerals,
have no less than 47.7 percent of the area of remaining tropical forests (Wunder
2003b:chap. 2). This estimate is fairly conservative.' In other words, the cross-
country data suggest that there are some characteristics of oil and mineral econo-
mies that in most (but not all) cases inadvertently protect forests. This article seeks to
explain this forest-protecting mechanism and understand how it is affected by
domestic policy responses. The core hypothesis is that in the oil-exporting countries,
deforestation pressures are lower when oil revenues are booming than during oil
bust periods. This shifts the reasoning from a static cross-country framework to an
intertemporal, comparative one.
The article briefly reviews hypotheses and methodologies and outlines the defor-
estation data problems and how they were dealt with. It presents empirical results
for the five primary case study countries and takes a closer look at three of the key
causal linkages in the model. It then examines the role of different policy instruments
and finally discusses policy insights that extend beyond the oil-exporting countries.
Theories and Methods
Why should oil wealth reduce pressures on forests? Sizable foreign exchange earn-
ings from oil, sometimes supplemented by foreign borrowing, raise government
spending, increasing aggregate demand. This causes the real exchange rate to appreci-
ate through growing inflation and nominal currency revaluation. Nontraded and
quasi-nontraded (import-protected) sectors expand, while nonoil traded goods lose
competitiveness and decline. This is the classical Dutch-disease pattern (Corden
1984), named after the Netherlands' experience with gas revenues in the 1960s and
1970s (Tle Econzomist 1977).
In developing economies, primary sectors have been the main victims of Dutch
disease, especially agriculture and the "purely traded" export sectors (Collier, Gunning,
and Associates 1999; Neary and van Wijnbergen 1986; Roemer 1984). To the extent
that large parts of the agricultural and timber sectors are exposed to foreign competi-
tion, an oil boom generally discourages the expansion of agriculture and logging.
That also tends to reduce area expansion in these sectors and hence curbs deforest-
ation for agricultural conversion and for a number of other primary activities that
potentially claim forested lands (shrimp farming, small-scale mining).
What impact has there been on forests from the oil-led expansion of the nontrad-
able sectors? Most nontradables (public and private services, construction, and so
on) and trade-protected quasi-nontradables (manufacturing) are urban, so most oil
booms are accompanied by accelerated urbanization. Rural labor is drawn to the
Svoi R'un uder ti,uc Williami D. Siuliclrliii                         233



cities, abandoning natural-resource based livelihoods for urban activities that tend
to have more benign forest impacts. Conversely, in times of urban crisis some of
these people return to their villages, reexpanding food-crop production into forests
as a default rural safety net. Growing cities also leave "ecological footprints" on for-
ests, through demand for timber for construction, fuel wood for cooking, or protein-
rich foods (meat, dairies) that can lead to forest conversion. Often peri-urban forests
are most affected by these pressures. Still, urban sprawl in itself was a negligible
source of deforestation in the case study countries.2 In most cases, agricultural pro-
duction for urban markets also became more sedentary and land-intensive (fewer
land inputs per unit of output), and part of urban demand was also satisfied through
imports.3 Rapid urbanization is therefore good news for the conservation of a coun-
try's forest cover in most cases.4
As this rudimentary overview shows, it is important to take into account a variety
of linkages between macroeconomic factors and forests. In a recent analogous study
on the root causes of biodiversity loss, all macro-level changes were assumed to jeo-
pardize biodiversity (Wood and others 2000). But it is more realistic to assume that
some underlying factors increase environmental pressure whereas others reduce it,
the net effect depending on the balance between them. One way of analyzing multi-
level linkages with opposing forest cover effects is to use a computable general equi-
librium (CGE) model (see Kaimowitz and Angelsen 1998:61-68, for an overview).
However, such models have difficulty dealing with the spatial specificity of deforesta-
tion. They also generally require a wealth of information. The data gaps for the case
study countries proved to be too large to use a CcE model without first bridging data
gaps by making "creative" CGE parameter assumptions. For the sake of greater trans-
parency, a partial recursive approach was used instead (figure 1).
The analysis runs from the external level (1) of foreign-exchange inflows and the
macroeconomic level (2) down to sectoral production (3), land use (4), and the forest
levels (5 ). At the external level when oil wealth, or foreign borrowing against future
oil revenues, finances large increments in road budgets or resettlement projects or sub-
sidies for transport or agriculture, forest loss may increase (marked by a plus sign in
figure 1). The opposite could be the case if budget allocations for conservation and
forestry regulations are strengthened (marked by a minus sign).
Structural changes outside the policy sphere also affect the forest outcome in an
ambiguous way (left side of figure 1). For instance, if higher incomes induce con-
sumers to buy more meat and dairy products, that tends to increase deforestation for
cattle raising, as in Latin America. But if higher income leads to a shift from tubers
produced on swidden land to more land-intensive rice production, that can reduce
forest loss, as in Central Africa. Reduced poverty in a booming economy goes hand in
hand with higher labor costs (reducing deforestation), but there will also be more
funds available for investment, which can promote more forest clearing in capital-
scarce frontiers (right side). The relative strength of these counterhypothetical,
234                                 The lWorld Batik Research Observer, vol. 19. no. 2 (Fall 2004)



Figure 1. Linking Resource Booms to Forest Cover Outcomes
Level of                                    Causal mechanism
analvsis
(I) External        Oil btom                             Etra
(price or quantitv)         -         orwn
(2) Macreconomi  Higher national income
(transitory or permanent
demand    /--)                                                 Agiutrlbdgets ()                   caost
Higher domestic spending               ~                udgets (4              Sainsts    
(conisumptioni, investment)   ~               rad protectio buge'-Svig
Ni,~~~~~~~~~~~Frsr u
Real currency appreciation?
Relative price of nontraded goods rises
Nontraded            Quasi-nontraded      Semitraded sector    Traded production
production rises*    production rises,    -ambiguous*          declines*
(3) SectoralI
Accelerated        Construction      I       Agricultural      TimberOtelad
urbanization*      booni*            L       production        production        using etr
I       I                  ~~~~~~~~~~~declines*  delines*       declie
IL  -----           -  -   -----
Expansion of      Foet area        clearin land
(4) Land use                                              ae   sr                e*etato
DefoetainZoest degradation is
(5) Forest                                                is redud*reduced*
*Relative to preexisting trends (growth and structural change).
>   Core-lItypothesis causality: oil wealth protects forests (trend indicated in target box).
-     Counter-hypothetical causality: oil wealth eliminates forests (couinteracting trend indicated in target
box).
Expected to accelerate forest loss anid degradation.
Expected to decelerate forest loss and degr-adation.
*     Ambiguous in effect.
Source: Authors' construction.
Sveen Wuntder anid Williauot D. Sninderlin                                                                       235



forest-eliminating partial effects differs from case to case (dotted lines in figure 1). In
addition, there is the macroeconomic Dutch disease effect described already (center-
left), causing real currency appreciation and changes in relative prices.
At the sectoral level (3) Dutch disease causes traded sectors to decline while non-
traded (services, construction) and quasi-nontraded sectors (industry) rise. If timber
production and agriculture are fully traded sectors with perfect foreign substitutes,
Dutch disease at the land-use level (4) will unambiguously alleviate pressures to log
new forest areas and convert forests to agriculture and other traded-sector land
uses (such as shrimp farming impacting mangroves)-at least compared with pre-
existing sectoral trends. If these extrasectoral effects are dominant in their impact
on forests (5), deforestation and forest degradation would also be reversed or at
least reduced. Whether the forest-protection effect is proportional to the land-use
effect depends on specific parameters, such as the number of species extracted per
hectare or the extent to which abandoned agricultural areas convert back into forests.
Contrary effects may occur to the extent that logging and agriculture are semi-
traded or quasi-nontraded, for example, through import protection of part of the sec-
tor. In that case greater urbanization may cause a construction boom with higher
domestic timber demand or a beef-and-dairy boom accelerating deforestation for
domestic ranching. The partial approach does not produce a bottom-line calculus of
whether these contrary effects in net terms reverse the forest-protecting effect. Yet it
is a flexible framework for pulling together the patchwork of land-use factors, in
spite of incomplete data coverage and variable spatial scenarios.
Deforestation and Land-Use Data
Although hundreds of deforestation models and analyses have been developed over
the past decade, country-level statistics on forests and deforestation tend to be lack-
ing or unreliable. The adequacy of FAO forest statistics, the most frequently used
source, has been discussed critically elsewhere (Grainger 1996; Matthews 2001;
Rudel and Roper 1997). Here, observations concern only the five case study coun-
tries. A comparison of forest stock estimates fromn the European Union's T'REES project
for the early 199()s with those from FAO (FAO 199 3, 2001) and the World Conserva-
tion Union (IUCN) reveals typical stock variations in the 5-1 5 percent range, but the
gap can be 30-50 percent for countries with large forest-savannah transition zones,
such as Cameroon, or forest fragments in deforested landscape, such as Ecuador
(figure 2 and table 1) s Accounting for much of these large disparities are differences
in canopy cover criteria, spatial scales of resolution, time scale, and sample coverage
(Wunder 2003b:chap. 3).
There are no anniual deforestation estimates for any of the case study countries,
and it is thus not possible to do a proper testing of time-series relationships. The F,AO
236                                 The World Bank Research Observer, vol. 19, no. 2 (Fall 2Q04)



Figure 2. Differences in r,\o and ItICN Tropical Forest Cover Estimates from TREES Estimates
[ IUCN (most estimates from 1980s)         * FAO Forest Resources Assessment 1990 (1990 estimate)
C FAO Forest Resources Assessment 2000 (1990 baseline)
Venezuela
Papua New Gulnea
Ecuador
l        l                      ~~~~~~~~~ ~  ~~~Cameroon|ll     l
-35      -25       -15        -5         5        15        25        35        45
Difference from TREES estimates 1991-95 (percent)
Note: 5R0s: estimates are from maps between 1991 and 1 995.
Source: rloREES Mavauxand others 1998); It CN Conseration Atlas figures cited in Mayaux and others 1 1998): rAO
1990) estimates Ieo 1993:table 1J: l'Ao 1990) baseline Ipso 20)01).
produces average deforestation estimates for every full decade, but even these tend
to be highly problematic for the case study countries. The deforestation figures from
Forest Resouirces Assessmletit 2000 (F,\o 2001) for the 1990s are strikingly different
from those for the 198()s in Forest Resources Assessmtient 1 990 (FAO 199 3). The differ-
ence is explained more by changing methods and assumptions than by measured
differences. According to these data, forest cover loss appears to have doubled in the
1990s in Cameroon and decreased by two-thirds in Venezuela and by more than 90
percent in Gabon. Papua New Guinea's annual forest loss remains at 113,000 ha,
but Forest Resolurces Assessmlent 2000 retrospectively set deforestation in the previ-
ous period (1975-85) to zero, so that the flow estimate would still be compatible
with the new forest stock figures. For Venezuela the large reduction in estimated
deforestation reflects the changes in methods from a population-based model prediction
to a national expert consensus around a much more conservative figure. As explained
in the notes to table 1, Forest Resouirces Assessmnmt 2000 deforestation figures are based
Sven WAndrl(ler aniid 1W7i7i,ttlli  D. Sziniderlini                                   237



Table 1. Comparing Forest Cover Estimates: FAO'S Forest Resources Assessment (FRA) 2000 and 1990 and Authors' Estimates (thousands of ha
unless otherwise noted)
I,lol--,. ,,,1990
jorest area betieen USA
FRA 2000)                              I'R 199(0             200() ind FRA 1 99()      Auithors best guess
Annuaitil                                   Aninuaitil         Nuimiiber of
Total     Annuial    deforestation           Total    Annulal    deforestation       deforestation
forests  deJorestation    rate      Btaseline  forests  deJorestation  rate             years           Forest c     Deforestation
Counitrn        2000'   1990-2000   1990-2000(%)    1990'    1 99(1   1980 9(1   1980- 90 (%f,)  %    (I R 200()    over _ I 99q( b    1990s
Cameroon       23,858      222'          (.9       26.078   20.372      122          0.6       28.0      26      17.000-24,0)00       15()-200
Ecuador        10,(557     137d          1.2        11,927  12,026      238          1.8      -0.9       -1      15,500                >180
(11.500-1 7,000)
Gabon          21.862       10e          0.()      21,962   18,265      116          0.6      20.2      370)     20,500-23,000           0
Papua          30,601      1131          0.4        31,731  36,043      113          0.3     -12.0      -38      32,000-37,000         50-70
New Guinea
Venezuela      49,506      218g          0.4        51,686  46,052      599          1.2       12.2      26      49,000               250-400
(44,000-49,000)
Sou1rce: FAO (1993, 2001): online data at www.fao.org/forestry/fo/country/index.jsp (accessed 31 July 2001); and authors' estimates.
aIncluding plantations.
bPreferred point estimate or likely range under variable assumptions and definitions.
cCameroon, 197 5-99 interpolation and extrapolation.
dEcuador, 198 5-92 interpolation and extrapolation.
'Gabon, 1970-99 interpolation and extrapolation.
fPapua New Guinea, 19 75-2000 interpolation.
gVenezuela, 1985-95 interpolation and extrapolation.



on long-term interpolation and extrapolations from other decades, so at best they say
something about forest loss in the recent past rather than specifically in the 1990s.
If the amount of forest lost in the study countries is unknown, along with how the
pace of loss has changed in subperiods, how can the causes of forest loss be estimated?
The analysis here employs what FAO calls the "convergence of evidence" method. This
involves gathering all available data and using basic math and common sense to
deduce what land-use changes are most likely to have occurred. All national esti-
mates, subnational trends specified in two global deforestation study databases (Geist
and Lambin 2001; unpublished collection of Thomas Rudel) and country sources,
and trends in predesignated deforestation "hot spots" were critically examined. For
Cameroon, primary data gathered for a CIFOR project were used, providing three- to
four-period estimates. Following Houghton and others (1991) and Barbier (2001),
cultivated area expansion was also examined as an alternative deforestation indica-
tor. Except for the savannah regions of northern Cameroon and the Llanos region in
Venezuela, most unoccupied land in the study countries is forested, and by far most
forest loss can be attributed to the expansion of agriculture (including livestock).
Estimates of deforestation in the 1990s for the five primary case study countries
based on this screening of a large number of land-use sources are reported in the last
column of table 1. In some cases, estimated ranges were preferred to point estimates.
For Cameroon. Ecuador, and Venezuela the estimates fall between those of Forest
Resources Assessment 1990 and Forest Resources Assessment 2000; for Gabon and
Papua New Guinea the estimates are significantly lower than both FAO figures.
Despite the absence of hard data (except for Cameroon), this yielded a rough idea of
historical forest trends. This article is limited to simply identifying periods of "high"
or "low" forest loss and employing them in the analysis below. (Wunder
200 3b:chaps. 4-9, provides a detailed quantitative assessment of the intricate pic-
ture of land-use changes and deforestation in each of the study countries.)
The country-specific land-use screening exercises also gave a clear indication of
the uses to which converted forestlands were being put. Although agriculture was
always the principal alternative land use, the types of land-hungry subsectors dif-
fered greatly across the cases. In Ecuador and Venezuela, pastures for cattle ranch-
ing were overwhelmingly dominant. In Cameroon, Gabon, and Papua New Guinea,
extensive swidden food crop systems were the primary land use. The multitude of
export crops grown in Cameroon accounted for only a minor share of deforested
area, because in Cameroon export crops are cultivated in a more land-intensive
manner than food crops.
The screening also gave a more realistic view of which sectors were definitely not
to blame for the bulk of deforestation. The direct land-conversion effects from oil and
mining, judged to be a significant cause in some past campaigns to save the rainforest
(RAN and Project Underground 1998; WWF 2002), were found to be negligible sources
of forest loss. In Cameroon oil was produced off-shore. In Venezuela oil is mainly from
Sven liiWinder and Williamn D. Sunderlin                              239



nonforested savannahs. In Gabon most oil sources are in forests, but the industry
occupies less than 10,000 ha (0.05 percent of forest area). In Papua New Guinea the
industry occupies just 1,200 ha. In Ecuador 99 percent of oil production comes
fromthe Amazon forest, but accumulated direct conversion has been only about
3,000-6,500 ha (0.04-0.09 percent of Ecuador's Amazon forest). Indirect access-
provision and pollution effects have in some cases been much larger than the direct
effects. But these effects were much reduced when the industries used best practice.
Country Stories
This section examines the five primary case study countries for evidence of the
impact of oil wealth on forests.6
Gabon
Gabon provides a textbook case of confirmation of the core hypothesis. Its transfor-
mation to an oil exporter in the early 1970s gave this sparsely populated country
the highest per capita oil rent in the case study countries. Despite a mini-bust in
1986-89 and highly fluctuating oil revenues in the 1990s, the entire post-1973
period was an era of high oil wealth.
Oil revenues, which were unequally distributed, caused massive, economywide
structural changes. An appreciating real exchange rate eliminated smallholder cash
crops like coffee and cocoa. Nontraded sectors thrived, in particular oil-financed
public employment and urban construction. With the total neglect of rural areas
and road building, and with substantial urban rent-seeking, there was a rural exo-
dus of people in the most productive age groups. The agricultural lands they aban-
doned grew back into forest. A village chief in northeast Gabon describes a situation
in which reduced cropping and lower human presence go hand in hand with forest
rehabilitation: "Nobody lives here any more.... The young are leaving, and the ele-
phants and gorillas run freely through our gardens, destroying what little we grow
to eat."' A resettlement program that brought people out of the forest into concen-
trated roadside settlements (rather than moving them into the "empty" forestland)
reinforced that trend.
Two national forest inventories indicate that high oil wealth likely led to a mar-
ginal absolute expansion of national forest area between 1970 (20,567,338 ha) and
1990 (20,677,000 ha; Wunder 2003b:86-90). For the 1990s FAO's Forest Resources
Assessmlenit 2000 revised annual deforestation estimates downward to only 10,000
ha (0.05 percent). Within this overall picture, there were local deforestation pro-
cesses in peri-urban areas, especially during recent mini-crises, when some people
lost their urban jobs and increasingly turned to peri-urban "weekend farming" to
240                                 The W'orld Bink Research Observer, vol. 19. no. 2 (Fall 2004)



improve their livelihoods. On the other hand, forests away from urban areas
returned to abandoned agricultural lands, as indicated by a number of village case
studies using remote sensing data (Wunder 2(00 3b).
Should oil revenues begin to decline rapidly. reducing the real value of Gabon's
currency, wages, and nontraded prices, there would be an increase in the domestic
production of cash and food crops, which are now mostly imported from neighbor-
ing countries. Because most of Gabon is covered by forests, this higher land demand
would be a significant stimulus to deforestation.
Venezuela
Venezuela's transformation from a specialized agricultural exporter in the 1920s to
a mono-exporter of oil from the 1930s onward has many similarities to Gabon's
story. It is another case that confirms the core hypothesis in absolute terms.
National forest maps show marked net forest regrowth in abandoned agricultural
areas from 1920-50, when people rushed into the cities to take advantage of eco-
nomic opportunities arising from oil wealth. But in the 19 50s Venezuela invested
heavily in roads. which together with fuel subsidies greatly reduced the cost of trans-
porting agricultural products. (Even in the early 1990s a liter of gasoline was
cheaper than a liter of bottled water.) This promoted the expansion of the cattle sec-
tor, which has accounted for almost all of the country's expansion of cultivated area
since World War II. Cultivated area expanded from 2.4 million ha in 1945 to 19.5
million ha in 1978. But 16.4 million ha (95.6 percent) of the 17.1 million ha expan-
sion were for cultivated pasture and only 0.76 million ha (4.6 percent) for cropland.
No pasture data were collected after 1978. but data on the number of cattle indicate
that the strong expansion continued. By 1996 pastures accounted for 95 percent
of cultivated-area expansion since 1945 (Wunder 2003b:144-47: data from the
Ministry of Agriculture and Breeding).
Cattle ranching was a semi-traded sector, with import protection during some
periods and for some products. It expanded procyclically with urban incomes due to
higher demand for protein-rich food. Meanwhile, cropped area stagnated. despite
large population growth. Unlike pasture, growth in cropped area was negatively
correlated with both rising urban incomes and declining price competitiveness. In
this sense, crop cultivation was the true victim of Dutch disease.
Although Venezuela still had low forest loss rates over 1950-80, both deforesta-
tion assessments and agricultural area data indicate that the loss has accelerated in
the past two decades. With the deep economic and political crises of the 1980s and
199()s, the speed of forest extraction and clearing increased, even in the areas south
of the Orinoco River that had previously been remote from the bustling urban cen-
ters. This spatial expansion was closely related to a depreciating real exchange rate,
which stimulated logging, artisan mining, and even food crop cultivation. Since
Sveun 4'uinder and lVilliarn D. Su,derlirz                         241



World War II Venezuela has thus witnessed increasing forest loss, though still at a
slower pace than its neighbors, which have stronger agricultural sectors.
Cameroon
Cameroon had the most pronounced economic cycles in the sample countries. The
prosperity of 1978-85. with a convergence of high international prices for its main
exportables (oil, coffee, and cocoa) and access to foreign capital, gave way to misery
in 1986 when the external environment suddenly reversed sharply. with a collapse
in world market prices for oil and crops and soaring real interest rates. During
1986-94 Cameroon could not unilaterally devalue the CFA franc, whose value was
set at the regional level. This clearly intensified the economic crisis. Cameroon has
been recovering slowly since the 50 percent CFA devaluation in 1994.
Specific analysis, tailored to compare the effects of boom and bust and with good
coverage of the entire humid forest zone, allowed a more exhaustive evaluation of
the core hypothesis. The boom (1979-85), though less pronounced than that in
neighboring Gabon, had a clear urban bias. Public sector employment rose by two-
thirds in just six years, and urban construction also increased. This accelerated
rural-urban migration, which led to a slowdown in deforestation. In the crisis
period, when the exchange rate was overvalued (1986-94), cash crops in the
humid forest zone (mainly cocoa) were stagnant. But an upsurge in food-crop pro-
duction caused a strong increase in deforestation in the forest zone.
The assessment of the shift in deforestation trends is highly reliable for Cameroon,
because satellite imagery and agricultural survey data were collected for periods
directly coinciding with the macroeconomic cycles. Research by CIFOR and its part-
ners combined time-series satellite imagery with socioeconomic survey data on 5,000
households in 125 villages, generating primary data on both land use and rural house-
hold economies. In a block in NMlOOl, East Province, annual deforestation quadru-
pled between 1973-86 and 1986-96. In the Bertoua area in East Province, forest
loss almost tripled, from 0.3 percent in 1973-86 to 0.8 percent in 1986-91. In the
peri-urban area of Yaoundeforest, clearing doubledbetween 1973-88 and 1987-95.
Household survey data from villages in all three provinces of the humid forest
zone confirm that forest clearing increased markedly after the onset of the crisis,
as farmers increased their cultivation of food crops (Mertens and Lambin 2000;
Sunderlin and others 2000). Reduced urban labor absorption and net return
migration to rural areas were found to be key factors in the dramatic resurgence of
deforestation.8 Since the devaluation. cash crops have also experienced a slow revival,
and deforestation continues to be high. Unlike in Gabon, oil wealth was not large
enough to reverse forest loss and cause forest area to expand. The economic rents
were not large enough to fully compensate for other factors, notably high rural pop-
ulation growth. But they caused an abrupt reduction in the pace of deforestation.
242                                Tlhe  World Baik Research Observer, vol. 1 9. no. 2 (Fall 2004)



Cameroon thus provides a strong relative confirmation of the core hypothesis:
During the oil-wealth period, the rate of forest loss dropped markedly.
Ecuador
The per capita size of the boom was larger in Ecuador than in Cameroon but much
smaller than in Gabon and Venezuela. Due to rising production, oil revenues
remained high until the mid-1980s. A bust followed, accompanied by a political cri-
sis in the 1]990s. Although there have been mini-oil bonanzas, the entire post-1985
period is best characterized as an economic downturn.
Nevertheless, deforestation actually accelerated during the oil boom, stayed high
for a decade before likely decelerating. The government spent a large share of oil rev-
enues in ways that promoted more extensive land uses. It built new and better roads
connecting the highlands with the lowlands.9 Fuel subsidies fostered extensification
by making transport cheaper, though compared with roads, the effect of subsidies is
nonspatial and reversible.?t) The government also used some oil motey to support
both spontaneous and planned frontier colonization that caused forest loss. " As in
Venezuela, cattle were also a crucial factor, reinforcing the spatial extensification
impacts of government policies. A growing urban middle class used part of its new
purchasing power to buy cattle-derived proteins, thus creating higher demand match-
ing the supply-side factors that enabled a large-scale expansion of pastures. Government
trade policy partially protected meat and dairy products from import competition,
allowing domestic producers to expand into marginal forestlands with meager returns.
In Ecuador these counterhypothetical factors were stronger than the core effect.
and policy was decisive in creating that outcome. Dutch disease effects did reduce
cropland expansion in the 1970s, and after the devaluations of the mid-] 9880s these
reductions were reversed. But on the whole the forest eliminating pasture policy
nexus was stronger.
Hence. the Ecuador country case study was contrary to the hypothesis in both
absolute and relative terms. Forest cover decreased in absolute terms during the oil
boom, and the pace of deforestation also picked up over the preboom period. This is
indicated by both national forest cover data (with some comparability problems) and
by agricultural area statistics, which are more reliable. The slowdown in deforesta-
tion in the 1990s, indicated by various regional forest data on hot spots of deforesta-
tion, may eventually be enhanced by the recent dollarization, which is likely to
cause loss of competitiveness that will lharm land-using sectors.
Papua New Guinea
Papua New Guinea became a specialized mineral exporter after 1972. Copper and
gold predominated until 1992, when significant petroleum production began. The
Svcn Wunder tndiii WilliamlJ D. Sunderlin                            243



size of these mineral rents relative to the size of the economy was somewhat greater
than in Ecuador. Firm macroeconomic management succeeded in controlling the
impact of rents until the early 1990s, but since then fiscal control has been increas-
ingly lost. The fixed kina exchange rate introduced in 1975 greatly hampered non-
mineral exports until 1994, when a tloating kina coupled with economic
mismanagement led to a large real devaluation. In other words, there were fairly
high mineral revenues throughout the whole period, but with different economic
management cycles.
The most reliable land-use data, from the Papua New Guinea Resource Informa-
tion System (PNGRIS) and covering the 1975-96 period, show that cash crops never
became a significant cause of deforestation. Even after the kina devaluation, rural
violence, inflexible land tenure arrangements, and infrastructure problems were
large impediments. Thus in Papua New Guinea improved price competitiveness
was not enough to induce expansion in the cash crop sector. An exception is the oil-
palm sector, which caused some deforestation. As in Gabon, government policies
tended to ignore rural infrastructure, which greatly reduced forest loss. A large
share of mineral revenues went into consumption. Some mineral revenue was wasted,
some was turned into capital flight, and little was invested to enhance production.
At the national level forest loss was clearly driven by swidden cultivation of
food crops, the demand for which grew in tandem with the population. But the
country's rigid land tenure structure, which strongly favored land intensification
over extensification, was discouraging expansion into forestlands. Hence the
increase in demand for land was much less than the increase in food crop produc-
tion. Between 1975 and 1996 intensive land use expanded at 0.7 percent a year,
but total land use increased by only 0.2 percent, far below rural population growth
rates of around 2 percent (Allen and others 2001). Intensification is more labor-
demanding than swidden cultivation in the forest fringes, but new forest clearing for
swidden cultivation meets severe sociocultural constraints. Intensification is achieved
by technically improving cultivation (terracing, improving drainage, composting,
rotating crops, growing nitrogen-fixing fallow species), but also by introducing new
crops or varieties (such as potatoes, sweet potatoes, and African yams; Bourke 2001).
Annual forest loss rates were 50,000-70,000 ha (about 0.2 percent). This is a
low rate and was probably stable over time, though with respect to the small size of
the population and the economy it was significant.12 Few value-added and service
sectors have developed that might have drawn people out of food crop agriculture
and into urban activities to uncouple deforestation from population growth.
On the whole, mineral wealth in Papua New Guinea did not stop deforestation.
It did slow forest loss, in particular by curtailing the production of cash crops.
None of the other cases was as complex as that of Papua New Guinea. Factors such
as land tenure, crime, and political instability were equally important in shaping
land use.
244                               Tlhe World Bank Researcih Observer. vol. 19. no. 2 (Fall 2004)



Summarizing the Country Evidence
Developing economies are subject to long-run processes of structural change, such
as demographic and economic growth, urbanization, shifts of resources from primary
to secondary and tertiary sectors, changing consumption patterns, and expanding
land use. The core hypothesis of the forest-protecting impact of mineral rents has to
be evaluated against these normal development trends. Table 2 summarizes the
Table 2. Comparing Macroeconomic and Forest Cover Trends
lVainr sources use(d to
Forest cov,er          deterrinie Jorest cover Evialuatioii of tie
(ounitrnj   Macroecononiic ctcles  chaige cycles        ctllil7ge         core Ilt pOtli(sis
Gabon       196(0-73: Preboom:   1 970-1990s: Net forest  Remote sensing   Absolute
1974-85: boom:       regrowth: recent: Some  (two national    confirmation
1986-89: mini-bust:  periurban clearing,    maps,* peri-urban  (short and
1990s: fluctuations  probably little net change and village studies) long run)
Venezuela   ] 92201/3(s: rise of  192(0-5(): forest     National forest    Absolute
petro-economy:       regrowth; 1950-1980s:  maps 1920.1950, confirmation,
1956-58: mini-boom: slow loss: 1980s-1 990s:  1991]:* regional  before World
1974-83: boom:       more rapid loss        forest maps: agri-  War II: relative
1984-: crisis and                           cultural area data  confirmation.
mini-booms                                  and estimations*  after World War II
Cameroon    1960-78: preboom:    1973-85: slow loss:    Remote sensing     Relative
1979-8 5: boom:      1986-94: rapid loss:   images (national.*  confirmation
1986-94: bust. fixed  after 1994: probably  regional) and
exchange rate: after  high loss             primaryagriculture
1995: devaluation,                          survey data from
recovery                                    the humid forest
zone
Ecuador     1960-73: preboom:    Before 1975:           National forest    Rejection, absolute
1974-8]: boom        moderate loss:         maps (1 962. 1987 . and relative
(rising): 1982-85:   1975-90):              20)0))*): regional
boom (declining):    rapid loss:            forest maps: agri-
1986-95: bust: after  1990s: probably       cultural area data.
1996: mini-booms     slower loss
Papua       1972-94: mineral     Whole period:           National maps     Relative confirmation
New Guinea boom (rising), fixed  Probably stable,       from aerial photos
overvalued kina:     slow loss, linked to   (1975) and satellite
after 1995: oil boom.  food-crop expansion:  images (1 992-93)*:
financial capital    alter 1994: Perhaps loss  M'CRIS land-use
outflows, devalued kina acceleration        change database
Souirce: Wunder (200(3b:chap. 1)(.
*Limitations in data comparability over time.
Sveni WA'uoder anld 1V illaim D. Suinderlini                                              245



evidence for the five primary country case studies by juxtaposing their macroeco-
nomic cycles and their land-use cycles.
Gabon and pre-World War II Venezuela are hypothesis-confirming cases in abso-
lute terms. Large oil rents basically wiped out agriculture, and many abandoned
areas grew back into forests. Forest cover increased in absolute terms. In Cameroon
the oil effects were not strong enough to produce such an outcome. The boom was
srnaller than those in Venezuela and Gabon and thus could not nullify the forest-
cover impact of rural population growth. But there was a clear relative reduction.
The pace of deforestation decreased dramatically in the oil boom period and then
picked up again during the crisis.
Papua New Guinea was also a case of relative confirmation, though a more
uncertain one. Real exchange rate effects were mediated by shifts in economic pol-
icy, and several nonmineral factors had a crucial influence. Still, mineral wealth is
one of the explanations for low deforestation rates. Finally, Ecuador was the only
case that rejects the core hypothesis, mainly due to the policy of extensive road
building and a nexus of factors that linked this expansion and land-extensive cattle
ranching (including dairy farming) to oil wealth.
Tracing Linkages
The juxtaposition of periods in table 2 exhibits some (but not all) of the patterns one
would expect for the core hypothesis. This is hardly surprising, given the long causal
chain described in figure 1, the extended period of study, and the presence of coun-
try-specific peculiarities. The main transmission stages are from oil wealth to the
real exchange rate and relative prices, from prices to sectoral production (agricul-
ture and timber), and from sectoral production to forests. Detailed analysis of all
these linkages for the study countries, including how to deal with various data prob-
lems related to deforestation and to macroeconomic variables, is available in
Wunder (2003b). Some general comments on the strength of the linkages are pre-
sented here.
In most countries oil wealth produced the expected appreciation of the real exchange
rate, with a shift in relative prices in favor of nontradables, though sometimes with a
time lag (Wunder 200 3b:chap. 10). For several countries, foreign borrowing was an
integral feature of oil wealth. They often borrowed against their new oil wealth
either to increase already high foreign exchange inflows or to prolong a boom period
and so delay the downward adjustment in government expenditure linked to oil busts.
Higher capital inflows from borrowing caused more real currency appreciation.
Management of nominal exchange rates also proved important. In the simple
Dutch disease model currency devaluation has no impact on relative prices or com-
petitiveness, because domestic prices adjust instantaneously so that the real
246                               The World Bank Research Observer. vol. 19, no. 2 (Fall 2004)



exchange rate remains unchanged. In the reality of the case study countries, domes-
tic prices were sticky. and overvalued exchange rates could persist for years. Hence,
repeated devaluations (as in Ecuador and Indonesia in the mid-19 8Os) were vital in
reviving traded sectors, whereas fixed exchange rate regimes exposed them to
reduced competitiveness in other countries, notably in Cameroon and Papua New
Guinea before 1994.
Agriculture was a prime victim of Dutch disease, but often it was the purely
traded export sectors that were hardest hit by declining competitiveness. In Gabon
and Venezuela appreciation of the exchange rate practically liquidated the cash crop
export sectors. In Cameroon and Papua New Guinea it was probably the most
important among several factors leading to economic crisis. The food crop sector
was not always unambiguously hurt by appreciation of the real exchange rate. At
the macro level food crops were semi-traded because of imperfect substitutes and
partial trade protection over time, but also because of "natural" protection from
trade provided by the high costs of transport to remote or inaccessible regions. To the
extent that semi-traded goods were strongly linked to higher income (meat, dairy
products, luxury staples), domestic suppliers profited from higher demand, as was
the case with producers of nontraded goods. To the extent that semi-traded goods
were close substitutes for importables and had low income elasticities (such as tubers
and plantains), producers lost market shares, similar to those in traded sectors.
Did effects on agricultural production translate directly into forest impacts? As
explained, forests were the default vegetation cover in most regions, so higher demand
for land would in most cases cause corresponding loss of forests. However, the linkage
of higher agricultural production with higher demand for land could be somewhat
variable. In Papua New Guinea lack of physical access and legal rights to new land
promoted intensification on already occupied soils, so that land demand increased less
than production. The opposite was the case for cattle ranching. In both Ecuador and
Venezuela the aggregate cattle carrying capacity declined over time, reflecting the
spread of ranching to still more marginal lands with lower per hectare returns.
Still, in all the case study countries agriculture was the big land-use competitor
with forests. Gains for agriculture meant losses for forests. The somewhat uneasy
implication is that most of what is good news for agriculture tends to be bad news for
forests, and vice versa. As the next section shows, this also means that policies that
are biased against agriculture may, unintentionally, be forest conserving-and
more effectively so than policies aimed directly at forest conservation.
Decisive Policy Responses
The country analysis shows that the net impact of oil wealth on forests is often to
enhance forest protection, as predicted by the core hypothesis. Yet specific case study
Svenz Vvlunder cnid Ililltt/ia  D. Sunderlini                         247



outcomes ranged from absolute confirmation to relative confirmation and, in one
case, to rejection.'3 Country-specific preconditions were in part responsible for the
variability of outcomes. For instance, large oil booms are conducive to forest protec-
tion, but in Latin America a meat and cattle tradition in the region exerted signifi-
cant pressures on forests when incomes rose. In Central Africa and Papua New
Guinea deforestation is mainly for land-extensive swidden cultivation, though in
Papua New Guinea there was a strong intensification response.
Yet beyond these preconditions. distinct country-specific policy responses strongly
shaped the variation in outcomes. Governments were not passively reacting to the
changing fortunes of the world market or, as some claim, simply adopting adjust-
ment programs designed by the Bretton Woods institutions. On the contrary, the
rent-rich oil states developed an ability to steer clear of external pressures and influ-
ences, delaying adjustment and resisting cutbacks and reforms. Internally, they
were also able to buy off powerful stakeholder groups, though this came at the cost
of increased rent-seeking and corruption. In some cases oil wealth in itself was
conducive to armed conflict with separatist movements trying to appropriate the
rich natural resources (Collier 2000; Ross 2001: Wunder 2003b:chap. 10). In the
aggregate it is fair to say that the oil-rich countries had greater room for political
maneuver than most developing economies and that they used this political space in
ways that had important consequences for forests.
Which policies, then, proved most effective in protecting forests, regardless of
whether there was any intent to protect them? Experience in Gabon, where long-run
land-use determinants have favored conservation, can be combined with policy
ingredients from other study countries to create a list of factors that have tended to
result in forest conservation (table 3). This is not a list of policy recommendations-
few of these policies are propoor development policies. It is simply a list of policies
that de facto have protected forests in the case study countries.
Conclusion and Perspectives
This study supports the core hypothesis that oil and other mineral wealth tends to
protect forests (though not in all cases) through indirect, economywide effects that
discourage land- or forest-based commodity production and reduce their competitive-
ness. This macroeconomic impact of mineral rents on forests is bound to be more signif-
icant than the direct, physical effects of mineral production in forested areas. Most of the
countries studied conformed to the core hypothesis in relative terms (oil wealth redLuces
pressures that degrade and convert forests), some even in absolute terms (oil wealth
reverses pressures that degrade and convert forests), and only one (Ecuador) of the pri-
mary case study countries was contrary to the hypothesis. The absolute conformers
(Gabon, pre-World War II Venezuela) had high oil rents per capita. The only negative
248                                 Thie Wlorld Bi3tik Research Observer. vol. 1 9 no. 2 (FIall 2004)



Table 3. Policies in Case Study Countries That Inadvertently Resulted in Forest Conservation
Polioz                                    Cotlniooeots
Neglecting demands for new road           Countries with large road-building programs had high
building and road maintenance             deforestation rates (Ecuador, fndonesia) and those
in rural areas.                           neglecting rural roads had low deforestation rates
(Gabon, Papua New Guinea). This macro-level
observation confirmed micro results strongly
tying roads to forest loss.
Selling fuel domestically at its "normal."  This policy is forest-conserving. Conversely, selling cheap,
nonsubsidized price.                      subsidized fuel reduces transport costs for goods (crops.
livestock, wood products) from remote areas and thus
promotes land extensification. This had an impact
especially in countries where the fuel subsidies reached
extreme levels. as in V7enezuela.
Spending large amounts of funds on cities.  A strong urban development bias, such as that seen in
Gabon. protected forests, by attracting labor migration
out of rural areas. Conversely, spending on social
infrastructure in forested frontiers eventually helps
attract migrants, sustain settlement, and promote
forest conversion to alternative uses.
Keeping overvalued exchange rates.        Most nontraded sectors are urban, so overvalued exchange
rates helped sustain urban biases and jeopardized cash
crop production. On the other hand. sharp, repeated
currency devaluations were a powerful tool to make
agriculture and timber extraction more competitive.
Overtaxing export agriculture by confiscatory  C. vll,-l. l F,o-h,i that de facto transferred cash crop
price stabilization schemes.              earnings from farmers to the government or to an
administrative bureaucracy impeded a larger scale
expansion of cash crops. especially in the African case
study countries, which in most cases would have
caused additional deforestation.
Heavily taxing logging companies to capture  In Papua New Guinea in the I 990s. a sharp rise
stumpage values.                          in taxes reduced the speed of harvesting. However.
if logging concessions are overly generous (in the
sense of low taxes), as occurred for certain periods in
Cameroon, tindonesia, and Venezuela, companies
rush to cash in high timber rents, and this tends
to accelerate forest conversion.
Intensifying food production and liberalizing  The net impact of general trade liberalization on forests
food imports.                             is ansbiguous, but imnport protection of cattle and certain
land-extensive (typically, swidden) food-crop sectors
helped these sectors expand production into marginal
forestlands with very low per hectare returns.
(Continued)
Sreei tVli,niter no,.d lIllilon D. Srinderliu                                             249



Table 3. (Continued)
Policy                                 Comments
Forcing rural people to settle in concentrated  Many colonial and postcolonial settlement programs,
roadside agglomerations.               especially in Central Africa. concentrated swidden
cultivators in sedentary roadside settlements with
more land-intensive methods, thus reducing forests
clearing. Conversely, an Indonesian-style trausmtiigrasi
resettlement of people without land to a (forest)
land without (a lot of) people causes deforestation.
Wasting agricultural budgets on        Oil-financed modernization strategies to promote
agro-industrial "white elephants"      sedentary, capital-intensive cropping tended to require
and ignoring smallholders.             less land per unit of value added than traditional
farming. Yet much funding for agro-industry was
diverted or wasted, so lack of agricultural development
meant avoided forest loss.
Nourishing a rent-seeking environment  In high-rent oil countries personal income is increasingly
in which few people find it worthwhile to  divorced from production and determined instead by the
produce.                               success of rent-seeking efforts. This has reinforced urban
biases and has provided disincentives for commodity
production-further relieving pressures on forests.
Source: Author's compilation.
secondary case was Mexico, and as in Ecuador (and recently in Venezuela), a Latin
American roads-colonization-cattle nexus drove the counterhypothetical outcome.
Deforestation data for all study countries are poor and involve much guesswork,
but agricultural census and land-use survey data can sometimes compensate for this
lack of reliable information. This is because in all study countries conversion to agri-
cultural land (including pastures, fallows, and degraded croplands) was the domi-
nant ultimate use of cleared forestland. Forests receded basically because people
decided to cultivate tree-covered lands. Within agriculture, production and land
used for "purely traded" export crops was clearly held back by oil-led Dutch disease
effects on competitiveness. For semi-traded food crop sectors, trends were ambigu-
ous, but in most cases land expansion was also constrained by rising wages and
urban migration, which were especially important when these crops were produced
in swidden, land-extensive systems. Cattle ranching was often linked procyclically
to oil wealth through import protection and a high income elasticity of consump-
tion for cattle-derived, protein-rich foods. In regions where this cattle effect was
dominant (Latin America), the core hypothesis was less likely to hold than in others
(Central Africa, Papua New Guinea) where the reduced expansion of land-extensive
crops was the dominant force. This underscores regional peculiarities in the relation
between higher national income and forest loss.
250                                       T1ue Worldi Batik Researcih Observer, vol. 19. no. 2 (Fall 2004)



Did governments have an important say in how oil wealth affected forestsP They
certainly did, even though these influences were unintentional. Governments nor-
mally are the sole custodians of a country's mineral riches.'4 An urban bias in the
spending of oil revenues (increased public employment and urban investments) would
generally protect forests. whereas spending on new and better rural roads, infrastruc-
ture in frontier areas, and (especially) agricultural and fuel subsidies counteracted the
core forest-protecting impact. Notably, expansions or cuts in the budgets of gov-
ernment forestry and conservation agencies-sometimes observed to be a key envir-
onmental factor in crisis-ridden countries (Reed 1996; Sunderlin and Rodriguez
1996)-had little effect. In most case study countries the implementation capacity of
these agencies was low, irrespective of the level of resources available (see also
Kaimowitz and others 1998). In addition to spending decisions on oil wealth, several
policies significantly influenced the use of land and forests, notably exchange rate
policy, import protection of selected sectors, and forestry concession policies.
The list of 10 forest-protecting policies in table 3 invites two general observations.
First, only one of the factors-taxing logging concessions-affects forests directly.
All the other measures are outside the sector, side effects of policies not targeted at
forests-what could be called "blind conservation strategies." This underscores the
crucial importance of understanding the underlying causes of deforestation.
Second, the list is somewhat pessimistic in that most of these inadvertent conser-
vation policies are not conducive to economic growth or poverty alleviation. Many
are directly harmful. Compared with a standard of best practice development poli-
cies, arguably only 3 of the 10 policies-elimination of fuel subsidies, high taxes on
logging stumpage value, and intensification and import liberalization of resource-
wasteful sectors-would receive a positive evaluation. This indicates that the hard
tradeoffs between tropical forest conservation and economic development occur not
only at the micro level but also at the macro policy level. On the other hand, the list
identifies three tangible areas in which positive synergies may be present.
Implications beyond Oil
The policy results reported in the last section may be provocative and raise legit-
imate questions about whether, first, the chosen countries are representative of oil
countries and, second, to what extent results from oil-producing countries in tropi-
cal areas are applicable to developing economies that are not endowed with oil. On
the first question, the five primary case study countries represent a fairly broad mix-
ture of developing countries in terms of geography, gross domestic product level,
agricultural traditions, and ecological conditions. Perhaps they inadequately repre-
sent dry forest countries, where firewood overharvesting and overgrazing can be
causes of deforestation. They are also all small to medium-size economies, which is
SvetI Wunder niruld IVtilliai I D. Suniderlin                         251



why the analysis was eventually supplemented with three big-country cases from
each southern continent: Mexico, Nigeria. and Indonesia. Nigeria and Indonesia were
cases of relative confirmation of the core hypothesis (deforestation rates dropped dur-
ing the oil boom but did not turn negative), whereas Mexico was a case of rejection.
The second question is harder to answer. Developing economies specializing in oil
production are on average richer, more urbanized, more high-cost, and less agricul-
tural than other developing economies (Sunderlin and Wunder 2000). In the polit-
ical economy sphere, many oil countries suffer from "resource curse" problems of
underinvestment in human capital, ill governance, rent-seeking, institutional decline,
and economic stagnation. These features definitely make them a special sample. For
instance, rent-seeking and corruption related to illegal logging are known to cause
severe forest degradation. However, in our group of countries, the fact that rent-
seeking was happening in the urban sectors and that corruption was connected with
agro-industrial programs meant that corruption came to have a forest-protecting
effect. something that is obviously not fully generalizable to other countries.
Still, many of the macroeconomic phenomena and policies analyzed here are
equally common in nonoil-producing countries. Perhaps the main differences in
terms of policy responses is that in nonoil-producing countries a wider range of pri-
vate sector activities would be expected (with differential impacts on forests), in par-
ticular a higher diversity of agricultural scenarios with different land-use intensities,
as well as stronger production responses. However, there is no reason to believe a pri-
ori that nonoil-producing countries would systematically face either weaker or stron-
ger forest protection effects from a substantial foreign exchange transfer from abroad.
Do these findings have implications for countries that are not blessed with min-
eral wealth; As suggested by the wide applicability of the Dutch disease model, the
impact of a mineral boom's spending effect is similar to that of other international
rent transfers, such as remittances. foreign aid, and debt relief. All of these foreign
exchange inflows tend to increase aggregate demand, cause real currency apprecia-
tion, and speed urbanization. A large-scale global reduction or write off of foreign
debt under the Heavily Indebted Poor Countries Debt Initiative, for example, would,
on the whole, also lead to less tropical deforestation. However, the deviating case
studies show that this effect would not be guaranteed in every country. Hypothet-
ically, the forest protecting effect of debt reduction would be expected to be stronger in
a poor African country dominated by swidden cultivation than in a middle-income Latin
American country where most forest clearing is for pasture. In the Latin America
case rising incomes for an urban middle class with a large appetite for meat can
increase forest loss.
A second field of application concerns the claim that structural adjustment pro-
grams have strong negative effects on forests and the environment and thus need to
be environmentally screened and redesigned (Reed 1992, 1996). Counterargu-
ments have been that many environmental effects are unavoidable or too complex
252                                'flite h1,orld Baitik ResearLch Obscrer, n1l. 7 9, nio. 2 (FaIl 2004)



to analyze or that there are internal contradictions among different environmental
objectives. This study confirms that many actions to improve foreign exchange
inflows by revitalizing traded sectors will almost inevitably have negative impacts on
forests. In particular, it confirms the powerful role of currency devaluation, a key
policy tool in most structural adjustment programs in promoting land- and forest-
using sectors. Our study gives less support to the assertion that specific fiscal cut-
backs in forest regulating agencies have detrimental effects (see previous discus-
sion). Instead, it points to the danger that sharp general declines in public spending
can partially "reruralize" the economy at the expense of forests.
Finally, trade liberalization and export incentives have been singled out as highly
ecologically damaging components of structural adjustment programs. The results
of this study suggest that the argument is often valid but that import liberalization
can also restrict highly wasteful and forest-damaging domestic sectors, thus increas-
ing the natural resource efficiency of land uses to the benefit of forests. For instance,
it may well be preferable to fill the barbecue of an Ecuadorian family with meat from
the Argentine pampas. rather than from deforested erosion-prone and nutrient-poor
slopes in the Ecuadorian Andes. Likewise. it might well be better for global forest
conservation if a Central African urban family builds its staple diet on imported Thai
high-yield variety rice rather than on local plantains. whose production requires
reiterative burning of extensive peri-urban forest plots. The net balance of trade
impacts on forests will differ across countries, and the impact on forests will differ
from that on other environmental assets. This underscores that asset- and country-
specific assessments of structural adjustment effects are necessary, reducing the
scope for generalization. However, this study also shows that, despite the complexi-
ties, it is possible to analyze these effects. This should also serve as an encourage-
ment to those interested in modifying macroeconomic tools for the purpose of
improving environmental outcomes.
Notes
Sven Wunder and William D. Sunderlin are Senior Scientists at the Center for International Forestry
Research (CIFoR) in Bogor, Indonesia. Their e-mail addresses are s.wunderCacgiar.org and w.sunderlin@
cgiar.org, respectively. They thank Arild Angelsen, Ken Chomitz, David Kaimowitz, Tom Rudel, Stuart
White. and three anonymous reviewers for comments on earlier drafts of this article (and correspond-
ing book chapters). They are also grateful to Ahmad Dermawan and Ambar Liano for research assis-
tance and secretarial help.
1. If Brazil, now a low mineral exporter but endowed at the state level with important mineral
exports, is excluded from the denominator, that share rises to 72.1 percent (Wunder 200 3b:chap. 2).
2. Caracas is said to grow at the rate of about 1 ha a day, yet the additional land demand of 3 6 5 ha a
year corresponds to only 0.09-0.15 percent of Venezuela's forest loss. In Greater Jakarta. one of the
world's fastest-growing metropolitan areas, forests receded from 7 3,266 ha in 1969 to 43,691 ha in
Sven VtIuider antid Wl'illihutii D. Switiderlini                                253



1989, yet the annual clearing of about 2,000 ha contributed only 0.1-0.3 percent to Indonesia's defor-
estation (Prasetyo 1992).
3. The highest level of analysis in this study is the nation-state, so this article does not look at the
global consequences of oil wealth and of changed international trade patterns.
4. Sunderlin and Wunder (2000:318-20) demonstrate, using comparative data for 66 countries,
that high urbanization in a subset of mineral-rich countries is one of the structural characteristics
related to a higher share of forest cover and a lower rate of forest loss.
5. Most sources are from the early 1990s. rREEs data from 1991-95 are used as a benchmark. IUCN
data are from the latest available nationwide map (various years), whereas the Forest Resources Assess-
inent 2000's 1990 estimate (FAO 2001) is calculated backward to its 1990 base.
6. This section summarizes the main lessons from chapters 4-8 in Wunder (2003b).
7. Chief Mboula Thaopile, Nioungou village, cited in Adams and McShane (1996:207).
8. Demographic surveys showed that average annual population growth exploded from 0.75 per-
cent in 1976-86 to 4.6 percent in 1987-97. During the boom period the urban economy had
absorbed most of the rural population surplus. In 1986-92 rural to urban migration stagnated. and as
the crisis continued (after 1992), there was net urban-rural return migration (Sunderlin and Pokam
2002).
9. In 1974, the first full oil boom year, Ecuador spent 48.4 percent of public investments on roads
(Gelb and Marshall-Silva 1988:184). Many roads were built through forests, causing immediate defor-
estation as well as continuing deforestation for about a decade following completion of the roads
(Pich6n 1997; Rudel and Horowitz 1993: Wunder 2000).
10. As in Venezuela, massive fuel subsidies in Ecuador were politically difficult to reverse after the
boom, and in 1986 their cost reached an astonishing 9 percent of gross domestic product (Mosley
1991:414). Together with the new roads. subsidies drastically increased the mobility of goods and pro-
duction factors in the rural economy. Although only about 20 percent of peasant production was mar-
keted in the 1960s, 60 percent was in the 1970s (Larrea 1992:3 50-51).
11. Development banks specialized in cattle, the most extensive form of land conversion, and
expanded their credit volumes thanks to earmarked oil revenues. Several regional and national land
colonization institutes received oil money that had been earmarked by the government for agricultural
development (Rudel and Horowitz 1993, pp. 56-7: Wunder 2003b, pp. 237-8).
12. Data from PNGRIS need to be translated first to obtain deforestation information defined accord-
ing to FAO standards. There have probably been few fluctuations over time. For example, the increase
in logging triggered comparatively little additional forest clearing. But with Papua New Guinea's
small population of 4.7 million (1999), a point estimate of 60,000 ha is larger on aper capita defores-
tation basis, at 12.8 ha/1,000 people. than Ecuador's (12.1) and Cameroon's (11.9) (Wunder
2003b:chaps. 8, 10).
13. Among the three secondary cases-Indonesia, Nigeria, and Mexico-the first two were prob-
ably cases of relative confirmation. Forest loss was not reversed, but was weakened during oil boom
periods. As in other cases, cash crop production declined precipitously, while food crops were semi-
traded with an intermediate production performance. The same happened in Mexico, but as in Ecuador
and Venezuela, the Latin American nexus of road building-colonization programs-cattle meant that
Mexico contradicted the core hypothesis. There was procyclical deforestation for pasture creation,
although evidence was also found at the subnational level (Chiapas state) that supported the hypothesis.
14. A partial exception here was Papua New Guinea, where both regional governments and local
landowners directly received a nontrivial share of oil and mining revenues.
References
Adams, J. S., and T. 0. McShane. 1996. The Mytih of Africa: Conservation withioutt Illusion. Berkeley, Calif.:
University of California Press.
254                                       The World Bank Researcih Observer, vol. 19, no. 2 (Fall 2004)



Allen, B. J.. R. M. Bourke. and L. Hanson. 2001. "Dimensions of Papua New Guinea Village Agricul-
ture. " Paper presented at Papua New Guinea Food and Nutrition 2000 Conference, PNG tUniversity
of Technology, Lae, Papua New Guinea.26-30 June 2)000.
Angelsen, A.. and D. Kaimowitz. 1999. "Rethinking the Causes of Deforestation: Lessons from Eco-
nomic Models." World BankResearclh Observer 14(1):77-98.
Barbier, E. B. 20)01. "The Economics of Tropical Deforestation and Land tTse." Land Economlics
77(2):155-72.
Bevan, D.. P. Collier, and J. W. Gunning. 1999. Nigeria antd Indoniesia. Oxford: World Bank.
Bourke, M. 2001. "Intensification of Agricultural Systems in Papua New Guinea." Asia Pacific View-
point 42(2/3):219-35.
Collier, P. 2000. "Economic Causes of Civil Conflict and Their Implications for Policy." Global
Policy Forum. Online document available at www.globalpolicy.org/security/issues/diamond/
wb.htm.
Collier, P.. W. Gunning, and Associates. 1999. Trade Shiocks in Developinig Countries, vols. 1 and 2. New
York: Oxford University Press.
Contreras-Hermosilla, A. 2000. The Underlying Causes of Forest Decline. Bogor, Indonesia: Center for
International Forestry Research.
Corden, W. M. 1984. "Booming Sector and Dutch Disease Economics: Survey and Consolidation."
Oxford Economic Papers 36(3): 3 59-80.
Thze Economist. 1977. "The Dutch Disease." 26 November.
FAO (Food and Agriculture Organization of the United Nations). 1993. Forest Resources Assessmernt
1 990: Tropical Counltries. Rome.
20)00. Forest Resources Assessment 2000: OCn L- (i'. i,- ., oJ Forest anld Forest Clharnge. Rome.
2001. "Results of the Global Forest Resources Assessment 2000." Secretariat note. Item 8(b) of
the Provisional Agenda. Fifteenth Session. Committee on Forestry. Rome.
Geist. H. J., and E. F. Lambin. 200 1. .Il4hat Drives Tropical Fr.  i'. i.,. ,, A Meta-Analysis of Proximate
and Underlyinzg Causes of .-1. .- .iot l.- i Based oni Subizatiornail Case Study E vidence. Louvain-la-Neuve:
Ciaco.
Gelb. A.. and Associates. 1988. Oil I1' ii,lkulI'i Blessings or Cuirse? New York: Oxford tTniversity Press.
Gelb. A., and J. Marshall-Silva. 1988. "Ecuador: Windfalls of a New Exporter." In A, Gelb and Associ-
ates, eds.. Oil li ,,,dt,  Blessing or Curse? New York: Oxford [Jniversity Press.
Grainger. A.1996. "An Evaluation oftheFAo Tropical Forest ResourceAssessment 1990)." Geographi(alI
Journal 162(1l):7 3-9.
Houghton. R. A., D. S. Lefkowitz. and D. L. Skole. 1991. "Changes in the Landscape of Latin America
between 1850and 1985." ForestEcology andMManlagemernt 38:143-72.
Kaimowitz. D.. and A. Angelsen. 1998. Economiiic Models oJ Tropical Dejorestationi: A Review. Bogor.
Indonesia: Center for International Forestry Research.
Kaimowitz. D., Erwidodo, 0. Ndoye, P. Pacheco, and W. D. Sunderlin. 1998. "Considering the Impact
of Structural Adjustment Policies on Forests in Bolivia, Cameroon, and Indonesia." Unasylva
49) 149): 57-64.
Karl. T. L. 1997. The Paradox ofPlecnnj: Oil Booms aid Petro-States. Berkeley: University ofCalifornia Press.
Larrea. C. 1992. "The Mirage of Development: Oil, Employment, and Poverty in Ecuador." PhD diss..
Graduate Programme in Social and Political Thought. Y'ork University, Ontario.
Little. I. M. D., R. N. Cooper, W. M. Corden, and S. Rajapatirana. 199 3. Boomn. Crisis anld Adjust-
ment: The Macroeconomic Experience oJ' Developin,g Counltries. New York: Oxford IUniversity
Press.
Svem,11 Wander id Wlilliat D. Sunizderli,i                                                255



Mainardi, S. 1998. "An Econometric Analysis of Factors Affecting Tropical and Subtropical Deforesta-
tion. " Agrekon 3 7(1):23-45.
Matthews, E. 2001. Understanidinig the Forest Resources Assessrnent 2000. Washington D.C.: World
Resources Institute.
Mayaux, P.. F. Achard, and J.-P. Malingreau. 1998. "Global Tropical Forest Area Measurements
Derived from Coarse Resolution Satellite Imagery: A Comparison with Other Approaches." Environ-
mnental Conservation 25(1):37-52.
Mertens. B., and E. F. Lambin. 2000). "Land-Cover-Change Trajectories in Southern Cameroon." Annals
of the Association of Arnerican Geographers 90 (3):467-94.
Mosley, P. 1991. "Ecuador." In P. Mosely, J. Harrigan, and J. Toye, eds., Aid and Power: The World Bank
and Policy-Based Lendinig. London: Routledge.
Neary, P., and S. vail Wijnbergen, eds. 1986. Natural Resources and the Macroeconomy: A Theoretical
Framneu7ork. Oxford: Basil Blackwell.
Pich6n, F. J. 1997. "Settler Households and Land-Use Patterns in the Amazon Frontier: Farm-Level Evi-
dence from Ecuador." World Development 255(1):67-91.
Prasetyo. L. B. 1992. "Application of a Geographical Information System (GIS) for Land-Use Changes
Study towards Better Understanding of Deforestation Processes: Case Studies in Jabotabek and
South Sumatra, Indonesia." PhD diss., University of Tsukuba, Japan.
RAN (Rainforest Action Network), and Project Underground. 1998. Drilling to the Ends of the Earth: The
Ecologicail. Social and Climate Imperative for Ending Petroleumrl Exploration. San Francisco: RAN and
Project Underground.
Reed, D., ed. 1992. Structural Adjustmiient and the Environmiienit. Boulder, Colo.: Westview Press.
. 199 6. Structural Adjustmrlent, the Enviromnent, and Sustainiable Development. London: Earthscan.
Roemer, M. 1984. "Dutch Disease in Developing Countries: Swallowing Bitter Medicine." Working
paper, Harvard Institute for International Development, Cambridge, Mass.
Ross, M. L. 2001. Timber Boonis and Institutiornal Breakdown in Southeast Asia. Cambridge: Cambridge
University Press.
Rudel, T., and B. Horowitz. 1993. Troi, ., i V) ,', . ali . , Small Farmners and Land Clearing in the Ecuadorian
Amazonz. New York: Columbia University Press.
Rudel, T., and J. Roper. 1997. "The Paths to Rain Forest Destruction: Cross-National Patterns of Tropi-
cal Deforestation. 1975-1990." World Development 25(1): 53-65.
Sunderlin, W. D., and J. Pokam. 2002. "Economic Crisis and Forest Cover Change in Cameroon: The
Roles of Migration. Crop Diversification, and Gender Division of Labor." Economic Development and
Cultural e It-,-., 50( 3):581-606.
Sunderlin, W. D., and J. A. Rodriguez. 1996. Cattle. Broadleaf Forests and the Agricultuiral Modlerniza-
tion Law of Hondurais: The Case of Olancho. Bogor, Indonesia: Center for International Forestry
Research.
Sunderlin, W. D., and S. Wunder. 2000. "The Influence of Mineral Exports on the Variability of Tropi-
cal Deforestation." Environmenital amid Development Economics 5:309-32.
Sunderlin, W. D., 0. Ndoye. H. Bikie. N. Laporte. B. Mertens, and J. Pokam. 2000. "Economic Crisis,
Small-Scale Agriculture, and Forest Cover Change in Southern Cameroon." Environtmental Conserva-
tion) 27(3):284-90).
Wood, A., P. Stedman-Edwards, and J. Mang, eds. 2000. The Root Causes of Biodiversity Loss. London:
Earthscan.
Wunder, S. 2000). The Econommmics of F,. ,.  iii."  The Examnple of Ecuador. New York: Houndmills,
Macmillan, and St. Martin's.
256                                          The lA'orld Bank Researclh Observer, vol. 19. no. 2 (Fall 2004)



. 2003a. "Foreign Exchange Inflows, Competitiveness and the Expansion of Logging in Five
Tropical Countries." Paper presented at the IUFRO workshop Forest Science/Policy Interface in
Europe, Africa and the Middle East, June, Copenhagen.
-   200(3b. Oil Wealth and the Fate of the Forest: A Comiparativ,e Study oJ Eighit Tropical Developing
Countries. London: Routledge.
wwr (World Wide Fund for Nature). 200)2. To Dig or Not to Dig. Criteria.lor DeteriningiilirJ the Suitalbility or
Acceptability of Mineral Exploration, Extraction and Tran1spor-t froom Ecological and Social Perspectilves.
Gland, Switzerland.
Sven Wiunder and WVillianm n. Swiderliti                                                    257



I
i
I



Vouchers for Basic Education
in Developing Economies: An
Accountability Perspective
Varun Gauri * Ayesha Vawda
Advocates argue that voucher programs can correct the incentive problemns of education
systems in developing economies. An accountability perspective, based on a principal-agent
framework, was developed to clarify the argumentsfor and against education vouchers. An
assessmen7t offindings on vouchier programs in industrial countries and a review of voucher
or1 quasi-voucher experiences in Bangladeslh, Chile, Colombia, C6te d'voire, and the Czech
Republic support the usefulness of the analytic framework. The assessment concludes that
the policy relevance of vouclher programs for developing economies remains uncertain.
Major voucher initiatives have been attempted only in countries with a well-developed
instituitional infrastructlure. Some studies findfavorable benefits for at least some popula-
tion groups, but others find limited effects and evidence of increasing social stratification in
sclhools. Whetlher vouchers lead to better outcomes or greater stratification appears related
to specific contexts, institutional variables, and program designs.
Education systems in developing economies face enormous problems. In many of the
poorest countries, there are large gaps in affordable access and staggering differ-
ences in attainment between children from poorer and richer households. Even in
many middle-income countries that have achieved nearly universal coverage in pri-
mary and secondary education, the quality of instruction and learning is low, parti-
cularly for poor people. Government spending is often inadequate and inefficiently
and inequitably allocated across education inputs and levels of education. Even
where education systems are adequately funded, they can be indifferent to families'
concerns because they are controlled by inattentive officials or by particular social
classes and interest groups. In these settings, high rates of teacher absenteeism and
The World Bank Research Observer. vol. 19, no. 2,
© The International Bank for Reconstruction and Development / TlHE WORLD BANK 2004: all rights reserved.
doi:10.10)9 3/wbro/lkhOl 7                                          19:2 59-28(0



low enrollment rates for girls are symptomatic of systems with inadequately aligned
incentives.
What kind of an education system is best able to address these problems? In the
broadest terms, there are two archetypes for education systems. In the public arche-
type, the government finances education and manages all aspects of schooling,
including hiring teachers and constructing schools. The public archetype is useful
for building a system from the ground up and for ensuring uniformity among
schools, but it is typically less effective at motivating teachers and schools to respond
to families' concerns. In the private archetype, schools are privately owned, and
families pay the entire cost of tuition. The private archetype gives parents choices
among schools, and schools have a strong financial motivation to keep their clients
happy. But families unable to pay for a private school lose out. Using the vocabulary
of the World Bank's (2003) World Developmtent Report 2004: Making Services Work
for Poor People, the public archetype entails the long route of accountability in which
students' interests are secured when citizens' voice influences the compacts that pol-
icymakers establish with providers, and the private archetype entails the short route
of accountability in which the client power of family preferences influences provider
organizations and frontline professionals.
Analysts and reformers have argued that a combination of the two archetypes-
public financing with private provision-is the best way to combine universal access
with systemic responsiveness. Parents receive vouchers that can be redeemed at
qualifying private schools, or private schools are reimbursed by the state based on
their enrollment rates. Thus parents choose the schools their children attend, and
schools are financed based on the number of students they attract and are allowed
substantial management autonomy to make themselves more attractive. Critics
worry that a decline in financing for public schools will eventually destroy the sys-
tem of public education, resulting in difficult-to-measure but real consequences for
social cohesion, and that voucher programs will exacerbate rather than lessen what
are already significant inequalities in access to good education.
Despite the popularity (and notoriety) of voucher proposals, there is little empiri-
cal evidence on their effects, especially in developing economies. Programs that sub-
sidize education are not uncommon in developing areas, particularly those that aim
to allay the direct and indirect costs of schooling for girls, low-income groups, and
ethnic minorities. 1 Generally speaking, the objective is to increase the likelihood that
children in particular groups enroll in and complete school. Voucher programs, on
the other hand, generally aim to increase the quality of education among students
who are already enrolled by expanding the range of choices.
Using a principal-agent framework, this article examines the available studies on
the impact of voucher programs and draws inferences about what might be expected
from voucher programs in developing economies, what kinds of voucher programs
might be most useful, and what elements of the institutional infrastructure will be
260                                The World Batk Research Observer, vol. 19. no. 2 (Fall 2004)



important for their implementation. The analysis compares the theoretical predic-
tions to the evidence on the operation and impact of voucher plans in industrialized
countries and then reviews five voucher and quasi-voucher experiments in develop-
ing and transition economies. It draws inferences about the ways in which educa-
tion vouchers might be useful in low- and middle-income countries.
Vouchers and the Principal-Agent Problem
In the public archetype, a centralized education department designs a national cur-
riculum, finances education out of general revenues, and pays capital and operating
costs directly. It also makes all managerial and staffing decisions; negotiates teacher
salaries directly with national unions; remunerates and promotes teachers on the
basis of negotiated, experience-related criteria; employs evaluations as indicators of
need and not performance; and assigns students to schools. Critics contend that the
system results in overstaffing, too many unmotivated teachers, unnecessarily high
taxes, school administrators indifferent to families, and passive parents with no
option but to enroll their children in schools with which they are dissatisfied.
These critics advocate voucher systems that depart from the public archetype in
three key aspects. First, students are not assigned to a public school based on resi-
dence but depending on the details of the program's design, have choices among a
variety of other public or private schools. Although elements of choice and competi-
tion also exist in public systems-parents choose a school by choosing a residence,
and private school options usually exist-vouchers significantly augment them.
Second, schools have strong incentives to expand student enrollments because a
significant part of school financing is proportional to the number of students
enrolled in a given school. The idea is to shift schools' primary accountability for per-
formance from the education department, which is politically compromised in its
efforts to enforce quality standards, to parents and students, who are the best judges
of the education they want for their children. In theory, these incentives will lead the
better schools to take steps to improve quality, enhance efficiency, and develop inno-
vative approaches to learning. Poorly performing schools will not attract students
and will either close or be forced to find a market niche. Some schools might then
satisfy demands that the bureaucratically constrained archetypal system cannot
meet, such as specializing in computers, the arts, specific languages, certain ethnic
traditions or religions, or learning methods for socially disadvantaged students.
This aspect of voucher systems focuses not on the simple existence of subsidies to
private schools but on the degree to which payments vary with enrollment. In sev-
eral European countries, for instance, political resolutions of historical religious
struggles entail notions of equal treatment of different denominations, and the state
reimburses sectarian schools for some portion of their costs. Most of these are not
IVaru  Gauri a(nd AYIesha Va1I'da                                    261



really voucher systems because the reimbursements are not proportional to enroll-
ment, are proportional to enrollment but relatively small in magnitude, or are adjusted
irregularly, with the result that schools do not have strong incentives to expand
enrollments.2
Third, voucher systems require schools to have enough managerial control to
respond to parental demand. To be responsive to demand, schools need substantial
control over staffing and personnel decisions, budget, and instructional methods. At
least limited control over capital expansion, calendar, and curriculum also provide
schools with needed flexibility. Voucher programs and proposals differ significantly
in the extent of control granted to local schools, leading to large variations in
expected outcomes. How much leeway schools have in selecting students or charg-
ing additional fees on top of the per student allocation. for example, is likely to result
in substantial differences in how much socioeconomic sorting a voucher system will
exhibit.
Such sorting is one of the principal concerns of voucher opponents. They argue
that if vouchers are available to all students, children from advantaged backgrounds
(greater income, education, and access to information) would have a significant
edge in enrolling in better-performing schools. As more advantaged students choose
these schools, performance would continue to improve, leaving students from disad-
vantaged backgrounds behind in increasingly less attractive education settings.
These dynamics would accelerate not only if students chose schools but if schools
were allowed to choose students, either openly or indirectly. New schools would be
reluctant to serve disadvantaged students because the costs of educating them
would be relatively high. Opponents of vouchers also argue that for a variety of
social and political reasons, voucher programs might not function as supporters
imagine: Politicians might be reluctant to let public schools close, parents might not
value or discern differences in educational quality, and professional norms might
impede competitive practices among schools.
How can one assess the relevance of the voucher debate for developing econo-
mies, where educational institutions are more varied and resource constraints more
severe than in industrial countries? Setting aside, at least provisionally, the assump-
tions that are difficult to prove or disprove generally-parents know their children's
best interests, parents are in a better position to observe school effort than adminis-
trative officials, and schools respond to financial incentives just as most other organ-
izations do-how well voucher programs can be expected to function depends on
key parameters of the institutions and resources in a given developing area.
To identify these parameters. it is useful to cast the voucher reform as a payment
mechanism in a principal-agent relationship: The state is a principal that pays
schools, its agents, for providing educational services that further its objectives. Con-
sistent with the voucher idea, agents are assumed to have sufficient management
autonomy to deliver the objectives. Clients are free to choose from among the
262                                 The WA'orld Banik Research Observer. vol. 19, 1o. 2 (Fall 2004)



Table 1. The Expected Effect of Key Social and Institutional Parameters on Voucher Programs
Social or institutiorial cloaracteristic           i - v., .1. ii. .  , vou cher programs
Parents use student body composition as a proxy for school  Increases socioeconomic or racial sorting
quality
Enrollment is weakly correlated with school effort  Weakens productivity effect
Teachers and school owners are poorer and more risk averse  Reduces welfare gains
Determinants of school quality or enrollment are exogenous  Weakens productivity effect
to the school
Professional norms are weak                        Discourages curricular innovation
Monitoring of reported attendance is ineffective   Weakens productivity effect
.9ource: Authors' compilation.
agents. The state rewards schools in a linear compensation contract: Payments are
proportional to the number of students they enroll. As a result, the compensation
scheme employs intense incentives.3 The state's education objectives are, stated
loosely, the same for most governments in the contemporary world: to impart liter-
acy, numeracy, a stock of cultural and scientific knowledge, and cognitive skills; to
transmit social norms and world views; and to provide educational opportunities
that structure life chances based on local interpretations of freedom, equality, and
nondiscrimination. In a voucher system. the state usually continues to rely on a
national curriculum, the broader institutional landscape, and civil rights laws to
guide schools toward those objectives, but it uses student enrollment as the exclu-
sive indicator of the effort a given school has exerted to achieve them.4
The principal-agent literature is now fairly advanced, and its findings can be
applied directly to the architecture of voucher programs in education (Milgrom and
Roberts 1992). The findings can be used to assess key institutional parameters that
are likely to affect voucher systems, particularly as they might function in develop-
ing economies (table 1).
Bias and Noise in the Estimate of Effort
Theory indicates that compensation with intense incentives should be based only on
performance measures that are strongly correlated with the agent's effort. 5 It is rea-
sonable to ask, then, how well enrollment estimates a school's effort at achieving the
state's objectives. Parents probably observe aspects of school effort that ministry offi-
cials might overlook, such as the attentiveness of staff, the dedication of teachers,
and help in securing placement in higher level schools. But parents' assessments
might also have systematic biases, such as the tendency to conflate the composition
of a school's student body, particularly along socioeconomic and racial dimensions,
I'ariui Gauri and Ayjesla VTlavda                                               263



with the quality of the education offered, and to conflate the quality of education
offered with the effort a school exerts.6
Because family background, not school performance, usually explains most of the
variance in student performance, such a confusion would bias the state's indicator
of school effort. That bias would also create a powerful incentive for schools to
attract a student body that parents would find desirable. The preconditions for sig-
nificant socioeconomic or racial segregation would be in place. If the potential for
such bias is substantial, it will make sense for the compensation formula to include
additional variables that correct for the effect of student body composition on demand
for a school and on the learning outcomes for which the school is responsible.
Moreover, enrollments can also be a noisy estimate of effort. Remoteness or trans-
portation costs, changes in the birth rate, migration, returns to education in the econ-
omy, and (arguably) cultural expectations about who should be in school are all
exogenous determinants of enrollment that are uncorrelated with school effort. In
developing economies these exogenous determinants of enrollments could weaken the
desired effect of voucher programs on school productivity. Weights for these factors
would not be hard to devise, and they would not require frequent adjustment because
these exogenous factors change slowly, and changes are generally predictable.
Risk-Averse Agents
The less comfortable an agent is with risk, the higher the welfare cost the agent
bears from intense incentives.' In voucher programs, schools that suffer enrollment
declines receive lower revenues, with commensurate impact on the salaries and pro-
fessional reputations of staff. The poorer the country, the more likely it is that teach-
ers are poor and risk-averse, and the higher the welfare cost of voucher programs to
them and their families. A related problem is the transaction costs of transferring
ownership of the facility, or the opportunity cost of letting it sit idle, which carries
higher welfare costs in countries where capital is in short supply.
The Productivity of More Effort
Intense incentives are worthwhile only if increments of additional effort produce bet-
ter performance. Intense incentives for schools to increase enrollment, although
they might elicit greater effort at first, might not improve academic outcomes if, for
example, the quality of teacher training is inadequate nationwide, textbooks are
routinely stolen, it is difficult to recruit teachers to certain areas, teacher time is
siphoned off for political purposes, population density is too low, cultural norms limit
girls' attendance, or students are malnourished.
There might also be circumstances in which the marginal effect of school effort is
powerful, at least over a certain range of effort, as when modest numbers of public
264                                  Thle 1l7orld( Bank Research Observer, vol. 19, no. 2 (Fall 2004)



school students are given vouchers to attend private schools, which often have the
capacity to accommodate small numbers of new students with just a little more
effort. In that program design (vouchers are awarded to some but not all students),
an issue emerges about whether the voucher scheme includes intense incentives for
the public schools the students have left and what effects incremental effort on the
part of these schools might have on school quality and on the willingness of the stu-
dents to stay. In other words, it is important to distinguish a partial equilibrium
effect (the incremental gain in learning for students who use vouchers) from the
general equilibrium effect (which sums the effect for students who use vouchers and
the effect on students left behind in public schools).
Relative Compensation among Activities
If a compensation scheme with intense incentives rewards certain activities at lower
rates, the agent will not perform them. In voucher programs schools are not
rewarded for stressing norms or principles of fairness with which many parents
might disagree. So if the rules governing, say. equal treatment or ethnic pluralism
are not well enforced, the payment scheme will not give schools a reason to pursue
those objectives. Of course, the same can be true of compensation schemes without
intense incentives, such as those based on subjective evaluation. It can be argued
that because processes are difficult to observe, they should not be included in the
compensation scheme.9
A related phenomenon is that voucher programs might not stimulate innovation in
instruction and curricula, as proponents hope, and might even dampen it, because
parents in many contexts are conservative about teaching methods, the selection
criteria of higher-level schools and universities remain traditional, and the formula
for compensation does not pay schools for innovation or professional development.
The Value of Monitoring
With intense incentives it pays to monitor the agent's performance carefully (Milgrom
and Roberts 1992). In a voucher program, a school will be rewarded if it can exag-
gerate enrollment rates and get away with it. It will also be rewarded if it successfully
lowers standards for expelling students or if it raises the average grades students can
expect to receive, enticing more students to choose the school. Without reliable
inspectors and a strong current of professionalism among educators, education
departments in developing economies will be hard pressed to monitor how schools
behave in response to intense incentives, which in turn will lower the usefulness of
the incentives.
1Val-71 G1ui a11  d AnAesht? 1Vwll'da l                              265



Voucher Programs in Industrialized Countries
Although the framework described here comes directly from theory, empirical work
on voucher programs in industrial countries supports its usefulness. In applying the
framework, it should be emphasized again that voucher programs take a variety of
forms, depending on such things as rules on student eligibility, student admissions.
information dissemination, fiscal and pedagogical accountability, the mean and the
distribution of the voucher value, and school-level management autonomy. All
these influence the expected impact of a given program on academic achievement,
equity, innovation, and the state's other educational objectives. Impact evaluations
of voucher programs remain sparse. Even so, experiences can be analyzed to deter-
mine how key institutional and social variables influence program effects.
United States
If enrollments are a biased or noisy indicator of school effort, the expected impact of
voucher programs on school effort is likely to be weak, with little final impact on out-
comes. In the United States, studies of the academic effect of voucher programs in
small, privately funded efforts in New York City; Washington, D.C.; Dayton, Ohio;
and Charlotte, North Carolina, and targeted, publicly financed programs in Milwaukee,
Wisconsin; Cleveland, Ohio; and Florida have found limited or mixed results, consis-
tent with that expectation. 1t
In New York a private foundation offered scholarships of up to $1,400 to elemen-
tary school students enrolled in low-quality public schools. Some 20,000 students
applied for the scholarships, and 1,300 were selected through a lottery. Using the
lottery as an instrumental variable for private school attendance, Myers and others
(2000) compared average reading and math test scores of students who enrolled in
private schools with a comparable group of students who did not, and found no sta-
tistically significant difference after the second year of the program. Mayer and
others (2002) also found no statistically significant difference overall after the third
year. But both studies found significant and positive differences for African Americans
enrolled in the scholarship program, though not for other racial and ethnic groups.
Working with the same data, Krueger and Zhu (2002) argue that the positive effect
for African Americans disappears if student with missing baseline data are included
and that the effect is sensitive to the definition of race that is used.
Studies of the experiments in Dayton and Washington, D.C., also found no differ-
ences in achievement between voucher and control group students overall but some
possible benefit for African American students (Howell and others 2000; Howell and
Peterson 2002). Evaluation of the publicly financed voucher program in Milwaukee
has generated considerable controversy, even though it involved only 341 students in
7 nonsectarian schools. Witte (2000) found no consistent difference in achievement
266                                Tlhe World Batik Reseoarchi Observer. vol. 1 9. no. 2 (Fall 2004)



scores among voucher program participants and public school students, but Greene
and others (1997, 1998) reanalyzed the data, comparing results with those for a
group of voucher applicants who could not find space in a participating school. They
found that participating students did better after four years of enrollment. Rouse
(1998) used the same data and found that voucher students did better in math but
not in reading and that the math advantage increased over time.
Studies of the effects of these voucher programs on public schools and on the stu-
dents who remain in the public system have been hard to conduct because the num-
bers of students involved have been so small. In Florida's voucher program students
in schools deemed to have "failed" two years in a row are given the option of using
vouchers in private schools. One study of the program found that all 78 failing
schools improved to avoid the voucher threat and argues that this is evidence that
voucher programs have positive effects for participants and nonparticipants alike
(Greene 200(1). But it is unclear whether the improvements in those schools can be
attributed to vouchers per se or to the social and political pressure following a failing
grade.
New Zealand
More direct evidence on bias in the estimate of school effort comes from New Zealand.
In the late 1980s and early 1990s successive liberal and conservative governments
established a "quasi-voucher" system in basic education (Wylie 1998). Control over
teacher hiring, operating budgets, selection of academic missions, student fees, and
local fundraising was transferred to self-governing school boards composed of the
principal, one teacher, and elected parents. Residency rules were loosened so that
parents could choose any school for their children. The system did not establish
intense incentives based on enrollment, but schools started to compete for students
anyway because they received more operating funds as enrollment increased, salary
scales and prestige for principals were linked to school size, and schools gaining stu-
dents were widely perceived to be successful, which motivated school boards to
expand enrollments.
Studies of parental choice in New Zealand (controlling for census-based predicted
enrollments and nonmandatory certificate exam scores in secondary schools) found
that schools serving largely disadvantaged groups experienced declining enrollment
and rising shares of minority students (Fiske and Ladd 2000, 2003). Schools serving
advantaged populations were more likely to experience rising enrollment and steady
or falling shares of minority students. It appeared that white students were either
fleeing schools with large minority populations or pursuing schools with advan-
taged students. Additional research suggested that minority students were less likely
to consider high-status schools as an option and less likely to attend a nonlocal
school (Smithfield Project 199 5).
t'aruin Gauri atid .A1esha Vaxwda                                    267



Although parents' decisions to flee schools with high minority populations might
be individually rational, such actions are also consistent with the notion that in
some contexts enrollment rates are biased estimates of school quality. There is also
evidence of noise in the correlation between enrollment and quality: when one sec-
ondary school suffered a number of suicides, enrollment declined dramatically even
though it was far from clear that the school atmosphere was causally associated
with the suicides (Fiske and Ladd 2000). Schools also used a variety of clever but
legal techniques to recruit more advantaged populations. In 1997 these enrollment
schemes were used by more than 50 percent of schools in Auckland and
Christchurch and 24 percent in Wellington (Fiske and Ladd 2000). The apparent
increase in socioeconomic polarization in New Zealand was probably also related to
the fact that schools were able to set the level of student fees, as well as the fact that
less advantaged parents had a harder time accessing information about the better
schools and traveling to them.
Sweden
Though systematic studies are not available, similar accounts have surfaced about
the voucher program in Sweden. From the early 1990s onward, Sweden transferred
control and management of schools to municipalities, which began to finance edu-
cation based on their own revenues and block transfers from the central govern-
ment. Municipalities now grant privately managed "independent" schools, some of
which had been receiving state assistance, per pupil subsidies equal to the per pupil
financing granted to public schools.
The enrollment share of independent schools increased from 1 percent in 1991 to
4 percent in 2002. Parents with children in independent schools appear to be better
educated than parents of children in public schools, and they seem to be making
choices that avoid schools with large shares of non-Nordic immigrants (Daun
2()0 3). In one large municipality, some schools with large losses in enrollment were
unable to attract better teachers, but others were able to establish new market
profiles (Daun 2003).
United Kingdom
Evidence on systematic bias in the estimate of effort is mixed in the United Kingdom,
and evidence on the outcomes of the voucher program is unclear. A series of laws
beginning in 198(0 created a quasi-market in government-financed basic education
(LeGrand 1991: Walford 1997, 2003). The 1988 Education Reform Act transferred
control of budgets and personnel hiring to state-maintained schools, linked school
funding to age-weighted enrollment figures, allowed parents to send children to any
school with available space, and required schools to follow the national curriculum.
268                                The lorld Bank Research Observer. vol. 1 9, nio. 2 (Vall 2004)



In 1993 schools were allowed to select up to 10 percent of their students on the basis
of specific abilities, and existing private schools or independent sponsors were per-
mitted to establish grant-maintained schools. Seven new schools were formed from
1993 to 1997, and seven more have been established under the Labour govern-
ment, which has retained the voucherlike system.
A series of case studies has argued that the choice system has increased socioeco-
nomic segregation among schools, as students sorted based partly on the socioeco-
nomic mix in schools and as schools started to select students (Ball and Vincent
1998; Gerwitz and others 1995). But statistical analyses of segregation by class
have been inconclusive (Gibson and Asthana 2000; Gorard and Fitz 2000; Levacic
and Woods 1999), partly because in the previous system, which assigned students
to schools based on residence, "selection by mortgage" had already created substan-
tial sorting.
Achievement scores of students in the system increased during the 1990s, and
some studies argue that competition among schools, as measured by private sector
enrollment share, number of schools, or a Herfindahl index, is associated with
higher test scores (Bradley and Taylor 2002; Millington and Bradley 1998). but
other studies find mixed results. One study used interviews with headmasters to try
to establish a link between structural competition in a market, measured by the
number of firms (or, in this case, schools) or a Herfindahl index, and competitive
behavior and found that the relationship is looser than believed because of such
attenuating factors as product differentiation (type of education offered) and amount
of space available (Levacic 2001). It also found that the impact of competition on
performance is sensitive to the measure of competition used. In the early 1990s, the
government introduced nationwide testing and required publication of test results
and school inspections-a system of "naming and shaming." It is possible that the
accountability system, rather than competition for students, motivated school effort.
Incentive Problems
Some European countries with voucher systems, recognizing the incentive that flat
per pupil payments create for schools to select relatively advantaged students and for
parents to choose those "successful" schools, transfer additional resources to schools
based on the composition of the student body. Holland pays 1.9 times the standard
voucher value for each minority student and 1.25 times the value for an economically
disadvantaged student (Patrinos 2002a). Sweden also transfers additional resources
based on numbers of minority students and students with learning disabilities.
In almost all countries, constellations of teachers, local politicians, and voters
have successfully resisted efforts to close schools, even schools with low enrollments
and declining quality. That is suggestive of the risk aversion (and the political
power) of the actors in the education sector, including teachers. Experiences in
Varuni Gauri and Ai esha Vaxvla                                       269



industrial countries also suggest that voucher programs can result in less attention
to noncompensated school activities, such as curricular innovation.
Consistent with the predictions, case studies have found that pedagogy in
England and Wales has become more uniform as schools have come to be ranked on
the same hierarchical academic scale. Schools have seemingly expended less effort
on pedagogical innovation because they were not rewarded for it and may even
have been punished for it because parents tend to be pedagogically conservative
(Walford 2003). Fiske and Ladd (2000) also report that some teachers felt that colle-
giality had declined in New Zealand schools and that parents, regardless of socioeco-
nomic status, rated the quality of schools similarly, resulting in a consistent focus on
traditional academics in most schools and little curricular innovation.
Both findings are consistent with the notion that when certain activities-in this
case professional relations among teachers and pedagogical innovation-are
rewarded at lower rates, they will receive less attention. In the United States voucher
schools were less likely to have students with disabilities than the public school popu-
lation (Myers and others 2000), probably because voucher schools were not com-
pensated for taking on students with disablities.
Voucher Programs in Developing and Transition Countries
Although there have been experiments with demand-side financing in a large number
of developing and transition countries (Patrinos 2002b), true voucher programs are
rarer. The reforms in Chile and Colombia are probably the only genuine examples.
Bangladesh, Cote d'Ivoire, and the Czech Republic have had some quasi-voucher ini-
tiatives, which help illustrate the variety of contexts in which voucher programs
might be implemented and the results they might achieve.
Chile
Starting in 1980, Chile transferred management of public primary and secondary
schools to municipalities, abolished pay scales and civil servant protections for
teachers, started to finance municipal and nonfee-charging private schools at equal
rates tied strictly to attendance, and encouraged parents to choose schools based on
performance. The democratic government that assumed power in 1989 reestablished
salary scales and employment protection for teachers, but otherwise maintained
Chile's voucher system. It is probably the closest national-level approximation of a
voucher system among developing economies (Gauri 1998).
The Chilean voucher system led to a dramatic increase in students enrolled in pri-
vate voucher schools, from 15 percent of total enrollment at the beginning of the
1980s to 33 percent in 1996, almost entirely at the expense of municipal schools.
270                               The WVorld Bantk Research Obserrer, vol. 19, no. 2 (Fall 2004)



Although many teachers shifted to the private subsidized sector, almost no muni-
cipal schools closed. There is evidence that the private subsidized schools perform
better, based on test scores, than municipal schools (Aedo and Larrafiaga 1994;
Rodriguez 1988), but this private advantage disappears or turns negative when stu-
dent-level socioeconomic data are included as controls (McEwan 2001; Mizala and
Romaguera 1998). McEwan and Carnoy (2000) find that Catholic schools perform
better than municipal and nonreligious private voucher schools, but they spend
more per student than the other groups of schools.
A study that attempted to measure the effects of private school competition on
municipal schools found that communities with higher concentrations of private
school enrollment had modestly higher positive test scores in the national capital (a
net 0.2 standard deviation gain over 15 years) but small negative effects in the rest
of the country (McEwan and Carnoy 1999). Another study found that communities
with higher private enrollment shares saw public school performance fall from 1982
to 1988 (though the point estimate was significant only at the 10 percent level;
Hsieh and Urquiola 2002). The study concludes that even if a voucher system cre-
ates incentive effects that lead to improvements in municipal schools, the fact that
the better students migrate from municipal to private schools could lead to an over-
all decline in municipal school performance.
That study also argues that although small numbers of public students might
benefit from vouchers to attend private schools, when a voucher system induces
large number of students to move to private schools the overall impact can be
assessed only by taking the sum of the (assumedly positive) effects for the migrating
students, the (assumedly positive) effect of competition on public schools, and the
(assumedly negative) effect of the loss to public school students of their most talented
peers. It finds no change in national aggregate indicators of achievement, including
Chile's ranking in international test scores and the achievement gap between subsi-
dized and nonsubsidized elite schools. There is also evidence that both municipal and
private schools in Chile took active measures to select more advantaged students
during this period (Gauri 1998; Espinola 1995).
Chile's experience with vouchers, consistent with the predictions mentioned pre-
viously, suggests that when enrollments are the indicator for school effort, schools
take steps to make themselves more attractive to parents by enrolling a more advan-
taged clientele. New Zealand's experience shows that such sorting effects can be
problematic when the supply of schools is fixed. Chile's experience illustrates that
sorting can also occur when there is a strong supply response. In Chile the effects of
intense incentives based on enrollment were modest in the national capital and zero
or negative elsewhere. That is also consistent with the expectation that when school
effort is not directly related to productivity, intense incentives do not lead to better
performance. That no municipal schools were closed is consistent with the fact that
intense incentives can carry costs for risk-averse agents, who will fight efforts to
17arun Gnari and Ayeslia V7awdaz                                     271



relocate or close schools. Finally, Chile's voucher formula included adjustments for
rural schools, in recognition of the need to include covariates of enrollment in
intense incentives.
Colombia
Starting in 1992, Colombia offered vouchers to entering sixth-grade students resid-
ing in low-income neighborhoods who had previously attended a public school. The
vouchers, renewable through the end of secondary school, were cofinanced by the
central government and participating municipalities. Where local demand exceeded
the municipal allotment, vouchers were assigned by lottery. The program was
designed to help poor students make the transition to secondary school in areas
where public schools were filled to capacity. Financing to public schools was not
reduced when they lost students to the program.
The value of the voucher was initially high enough to pay for tuition at low-cost
private schools, but it was not indexed to inflation. By 1998 participating students
were typically making out-of-pocket payments equal to the voucher value to cover
tuition costs. Most elite private schools in Colombia elected not to accept the vouchers.
The program provided vouchers to 125,000 students, or about 1 percent of national
secondary enrollment, before it was discontinued in 1997 (Angrist and others
2001; Calder6n 1996: King and others 1997,1998; Montenegro 1995).
A quasi-experimental study compared educational and other outcomes among
samples of students who won and lost the voucher lottery (Angrist and others
2001). Lottery winners had lower grade repetition rates than losers but were not
more likely to be enrolled in school. A sample of lottery winners also scored 0.2 stan-
dard deviations higher on standardized tests, controlling for age, gender, parents'
schooling, and residence, and almost 0.3 standard deviations higher in a two-stage
least squares regression that took into account private school scholarships obtained
by lottery losers and scholarships not used by lottery winners. The effects were
slightly larger and estimated more precisely for girls than for boys. Lottery winners
spent US$52 more on school fees than losers, gave up $41 in reduced earnings
because they remained in school longer, and received scholarships that were $74
higher, for a net household contribution of $19. The public spent an extra $24 for
each voucher student, net of reduced public school costs.
A complete analysis of the net impact of the program in Colombia would estimate
not only the direct gains for students who used vouchers but also the program's
effect on public school students who lost classmates to the voucher program and the
effect on the new classmates of lottery winners. Short of research of that scope, how-
ever, Angrist and others (2(001) at least make plausible the argument that in certain
circumstances the marginal impact of incremental private school effort can be sub-
stantial. That is consistent with the predictions mentioned already, though it should
272                                The World Bank Research Observer, v ol. I 9, no. 2 (Fall 2004)



be noted that programs like Colombia's do not create intense incentives for schools
but instead exploit incentives in existing private schools. The findings from the
Colombia program are also consistent with the predictions that the marginal impact
of school effort will be lower in rural settings (most participating schools were in the
large urban areas) and that schools behave as if families use the socioeconomic com-
position of their student body as an indicator of quality (most elite schools in Colombia
refused to accept the vouchers).
Bangladesh
Community-managed, not-for-profit schools enrolled 80-95 percent of secondary
school students in Bangladesh in 2001. depending on how religious schools
(mnadrassas) are classified. The government subsidized 80 percent of the base teacher
salaries at these nongovernment schools, accounting for 79 percent of total govern-
ment expenditures at the secondary level in 1997. Because the schools were sup-
posed to follow state criteria on number and recruitment of students, the subsidies
functioned as a sort of voucher: When schools attracted enough students to warrant
hiring another teacher, the government subsidized that increase in enrollments
by paying for most of the additional teacher's salary. In private secondary schools
student fees for admission, tuition, books, uniforms, stationery, sports, after-hours
tutoring, and other activities constituted an average of half to two-thirds of private
school revenues. Demand for secondary education was strong enough. and entry
requirements in the sector liberal enough, that overall enrollments doubled from
1990 to 199 7, when they reached 7 million students.
But the norms for the transfer of the subsidies were weakly enforced. One survey
found that in 1998 only 29 percent of principals reported a visit by an inspector in
the recent past (World Bank 2000). As a result, schools applied for and received sub-
sidies without complying with enrollment and other norms, and the quasi-voucher
system did not necessarily create intense incentives to expand school enrollments.
The private fees, however, did serve that function.
By all accounts, the quality of secondary education in Bangladesh was inadequate
despite high per pupil expenditures by international standards. The private schools
did not train their teachers or develop new teaching methods, instead relying on the
traditional pedagogy of memorization. Textbooks were often unavailable, and charges
of cheating, by both students and schools, were not uncommon. The principal pur-
pose of secondary education was to screen and winnow university applicants on the
basis of the all-important examinations. Promotion rates were as low as 7 5 percent
between grades, 1 5 percent of students who enrolled in the first year of secondary
school completed a secondary degree, and 6 percent went on to a university.
The Bangladeshi experience with a quasi-voucher system is consistent with sev-
eral theoretical predictions. Intense incentives can fail to achieve objectives without
V"aruno GaiCuri and, Alesha Vawda                                    273



a functioning monitoring system. School effort will not be related to academic out-
comes if exogenous factors. such as textbook shortages and examination pressures,
limit returns to effort. Risk-averse actors will lobby to reduce the risk to which
voucher programs expose schools. External pressures, such as university exams, will
limit curricular innovation and teacher training (World Bank 2000).
Cote d'Ivoire
Cote d'Ivoire subsidizes private primary schools at rates that are negotiated with
umbrella groups for religious (Catholic and Protestant) and secular schools, that
vary with school location and tuition charges, and that are loosely tied to the num-
ber of students enrolled. These subsidies ranged from $40 to $66 per pupil in 1999.
At the secondary school level reimbursements to private schools even more closely
resembled a voucher system. In 1999 the government reimbursed private second-
ary schools $200-$2 3 3 per year for each "state-sponsored" student. These voucher
values were above market tuition rates at some private schools, but about one-tenth
the fees at the top ones.
Qualifying for sponsorship was not transparent. but it was tied to test scores,
which had been the subject of corruption controversies in previous years. Generally,
private schools had small permanent teaching staffs and hired public sector teachers
on a contractual basis or used "rental" teachers from the government. Private
enrollments were 1 3 percent of the national total at the primary level and 3 6 percent
at the general secondary level. Some 42 percent of private students at the primary
and secondary levels received either direct or indirect state support. The per capita
private education subsidy was significantly higher for families in the top income
quintile than for families in the lower quintiles. but the same was true for public
education, which favored urban areas.
Outcomes studies of education subsidies in Cote d'Ivoire are not available. One
analysis compared the results of a standardized test administered to five Francophone
Sub-Saharan countries (Michaelowa 2001). It noted that although primary edu-
cation expenditure as a share of gross domestic product (cil)0 was 50 percent
more in Cote d'Ivoire than in Cameroon and three times more than in Madagascar,
and its per capita (;t)P was higher than that of both countries, test scores among
fifth graders in Cote d'Ivoire were lower. Enrollment rates in C(te d'Ivoire were
54 percent, lower than in Cameroon but higher than in Madagascar. In some
regions of Cote d'Ivoire first-grade enrollment rates for girls were as low as 12
percent.
Explanations for the inefficiency in expenditures included the short academic year
(at 820 hours, one of the shortest in the world); double-shift classes for some stu-
dents. which further reduced the hours of schooling: fees and other obstacles to text-
book access that resulted in pupil:textbook ratios of 5:1 in urban areas and 10:1 in
274                                 The l, -l ( 13dllaiik Research Observer. vol. 1 9, 11o. 2 (FaIll 2(04)



rural areas: high teacher salaries (an average of nine times gross national product
per capita); and small class sizes. which were the lowest among the five comparator
nations (LaRocque 1999; World Bank 1998).
Although detailed analyses of the voucher program in C6te d'lvoire are not avail-
able. the case illustrates that even if schools are given intense incentives, external
constraints on performance, such as the labor market for teachers. national rules
regarding the school year, and expectations regarding girls' education, can diminish
the returns to school effort. In the circumstances, compenisation based on enroll-
ment will be weakly productive in raising achievement.
Czech Republic
Private schools were legalized in the Czech Republic in 1990. At the same time,
state-run schools were given authority over enrollment and curricula. Although pri-
vate schools at first received 'equal treatment" with respect to public financing, in
1995 their public funding was lowered to 60-9() percent of the funding for state
schools. Following complaints about the arbitrariness of the state's funding deci-
sions, a new formula in 1999 directed that private schools should receive funding
based on a set of defined quality measures.
In that scheme state-run schools received separate recurrent budget allocations,
determined on a per student, voucherlike basis. The recurrent component had two
parts: a base support level, which varied by type of school (on average 50 percent of
the per student payment to state schools), and a quality-based payment, with a max-
imum total set at 90 percent of the state sector per student subsidy. The average pri-
vate school received just under 80 percent of the per student allocation that public
schools received. Private secondary schools could charge tuition, which in 1 998
averaged 55-65 percent of the per student allocation going to public secondary
schools. Although the enrollment share remained tiny for private primary schools,
the secondary school share went from 0.1 percent in 1990 to 9.8 percenit in 1999
(Filer and Munich 200 3: World Bank 2001).
Assessments of the voucherlike system of education financing in the Czech
Republic remain provisional. One study found that parents witli higher earnings
and more education were more likely to choose a private school for their children,
but many private schools found their market niche by enrolling students who
did not gain admission to the most desirable, oversubscribed public secondary
schools. There is little published data on the criteria private schools use to select
students or on whether enrollments were functioning as a useful estimate of
school effort. Provisional evidence suggested that areas with relatively weak pub-
lic schools were more likely to have private schools established (Filer and Munich
200 3).
I'uIt ilii Gaurli (((it AqYesha Van'd                                275



Conclusion
Voucher programs are a potential solution to the problem that governments face
in motivating schools, as their agents, to put forth effort. Making school payments
proportional to enrollment creates intense incentives to satisfy parents. Enroll-
ment is a useful measure of school effort. But because parents frequently conflate
student body composition with learning opportunities, voucher programs can
create a strong incentive for schools to select advantaged students. Findings in
Chile and New Zealand are consistent with this concern. Where administratively
feasible, it makes sense to set the voucher value for disadvantaged students at
higher levels. as Holland does. Where that is not feasible, which is likely in most
developing economies, eligibility for voucher programs might be restricted to the
poor.
Intense incentives are worthwhile only if more school effort produces better
results. For small voucher programs that send modest numbers of students to urban
private schools with surplus capacity, as in Colombia's experiment, that might well
be true, at least for participating students. (Evidence on aggregate results is still
unavailable.) In programs with wide eligibility, the ability of schools to improve
learning among large numbers of students is likely to be limited in many developing
areas by factors beyond the schools' control, such as the number of qualified teach-
ers, the availability of textbooks, and pedagogical traditions.
Voucher programs that do not reward innovation may reduce diversity and inno-
vation in teaching if parents are pedagogically conservative. Voucher programs in
Bangladesh, Chile, Cote d'lvoire, New Zealand, and the United Kingdom did not pro-
mote pedagogical innovation. Case studies suggest that pedagogy might even have
become more uniform in those countries than before the programs. If implementing
voucher programs, governments will need to continue to finance and support
teacher training and professional development in private schools.
Teachers, school owners, and parents in developing economies are often risk-
averse. As a result, strict voucher programs in which poorly performing schools are
closed and teachers are laid off carry higher welfare costs and might encounter
strong political resistance.
Finally, voucher programs reward schools that can get away with exaggerating
enrollment rates or loosening education standards. So the effect of intense incentives
to expand enrollments will be distorted unless the government establishes a func-
tioning monitoring system, which can be expensive.
The usefulness of voucher programs for developing economies remains uncertain.
Major voucher initiatives have only been attempted in two countries with relatively
well-developed institutional infrastructures, Chile and Colombia. Some studies have
found favorable benefits for at least some population groups, but others have found
limited effects and evidence of increasing social stratification in schools. Whether
276                                The Wt'orld Bank Reserech Observer. vol. 1 9, no. 2 (Fall 2004)



vouchers lead to better outcomes or greater stratification appears related to specific
contexts, institutional variables, and program design.
Notes
Varun Gauri is Economist in the Development Research Group at the World Bank; his e-mail address is
vgauriCaworldbank.org. Ayesha Vawda is Education Specialist at the World Bank: her e-mail address is
avawda@.worldbank.org. The authors thank Shantayanan Devarajan. Harry Patrinos, David Plank,
Miguel Urquiola. and three anonymous reviewers for useful comments on earlier drafts and Imran
Hafiz and Hedy Sladovich for help in preparing the manuscript.
1. Patrinos (2002b) reviews the evidence on subsidies to increase education demand in developing
economies, including. for example. Bolsa Escola in Brazil, Progressa in Mexico, Food for Education in
Bangladesh, and a community grants program in Pakistan.
2. It is possible that schools will exhibit competitive behavior even in the absence of explicit, intense
incentives. In some cases, rivalry for professional status can lead to a "culture ol competition. " as in the
New Zealand system.
3. Formally, incentives in a voucher system are modeled as r = ac + j3(e + x + myy. where r represents
total payments to the school, a is a base payment for capital costs or in-kind inputs. e is school effort as
measured by the number students enrolled, .r is noise in the measurement of enrollments, y is a set of
factors that varies with enrollment, andy is a parameter that varies from 0 to 1. The higher the voucher
value, 3, the more "intense" the incentives the school faces (Milgrom and Roberts 1992).
4. Another approach would model parents as principals, either alongside or instead of the state
as a principal. with the school remaining the agent. But that approach, although useful as an exer-
cise, would have limited value for policy. The analytic virtue of the principal-agent framework is
that it identifies important contractual parameters of the voucher scheme in the hope that they
might be modified to improve social outcomes. If parents are not concerned with broad social out-
comes, or if they cannot shape the parameters of contracts with schools--there are reasons to
believe that both are true-an approach that makes them principals has limited value for policy-
making. The state might design a voucher program in which schools are paid not only for each stu-
dent enrolled but an additional amount for students from minorities or from disadvantaged
backgrounds, or less if an inflow of immigrants raises average enrollment rates and lowers marginal
costs for all schools in a neighborhood. or it may link payments to a weighted average of enroll-
ments and test scores. But parents are concerned primarily with the education of their own children
and are not likely to pay less if the school fails to fulfill a social objective. Moreover, because school-
ing, for economies of scale and social and historical reasons, is provided collectively, parents would
have to negotiate the terms of the contract with schools among themselves first, resulting in a diffi-
cult collective action problem. As a result, they would be hard pressed to design and modify key
parameters of contracts with schools. The framework adopted here focuses on, to the use the lan-
guage of World Bank (2003), the policymaker-provider link, and it assumes that the policymakers
follow through on their commitment to reward schools on the basis of parental choices. In other
words, it does not address issues related to interest group influence, which of course do not disap-
pear in a voucher system.
5. The use of a poor measure of effort and the exclusion of observable, exogenous correlates from the
contract both raise the expected variance with which effort will be measured. The higher the variance,
the higher the implicit costs of contracting (Holmstrom 19 79).
6. This bias on the part of parents is rational if enrolling a child in a school with advantaged peers
helps the child learn-if there are peer effects-or if enrolling in such a school is construed as a sign of
high achievement by employers and higher level schools-if there are signaling effects.
V'arsnti Gaziiri rid Ayeslma Vaw'da                                                 277



7. With a risk-neutral principal and a risk-averse agent, a fee schedule in which the agent bears all
the risk is never pareto optimal (Shavell 1979), but a small degree of risk aversion in a linear compen-
sation contract will permit a solution close to the first best (Holmstrom and Milgrom 1987).
8. For a formal presentation of this point, see Hsieh and Urquiola (2002).
9. Holmstrom and Milgrom (1987) argue that if two of an agent's activities can be separately
observed, the weight to attach to each in the optimal compensation scheme depends not only on the
costs and benefits of the activities but on their variances. The higher the variance, the lower is the
weight attached to the activity.
10. For a summary of research findings on vouchers in the United States. as well as on charter
schools, see Gill and others (2001). That volume argues that differences in academic achievement
between conventional public and charter schools in Arizona, Michigan, and Texas are relatively small
and mixed in direction, though that might be a consequence of the relative youth of charter schools.
Hoxby (2001) argues that areas with greater competition from charter schools exhibit higher achieve-
ment. Other reviews include Ladd (2002) and Neal (2002).
References
Aedo, C., and 0. Larranaga. 1994. "Sistemas de Entrega de los Servicios Sociales: La Experiencia
Chilena." In C. Aedo and 0. Larrafiaga, eds., Sistenia de Entrega de los Servicios Sociales: Una Agenda
Para la Reforma. Washington, D.C.: Banco Interamericano de Desarrollo.
Angrist, J. D., E. Bettinger, E. Bloom, E. King. and M. Kremer. 2001. "Vouchers for Private Schooling in
Colombia: Evidence from a Randomized Natural Experiment." NBER Working Paper 8343. National
Bureau for Economic Research, Cambridge, Mass.
Ball. S.. and C. Vincent. 1998. "'I Heard it on the Grapevine': 'Hot' Knowledge and School Choice."
British Journal of Sociology of Education 19(3): 377-400.
Bradley, S., and J. Taylor. 2002. "The Effect of the Quasi-Market on the Efficiency-Equity Trade-off in
the Secondary School Sector." Bulletin of Economic Research 54(3):295-314.
Calder6n, A. 1996. "Voucher Programs for Secondary Schools: The Colombian Experience." Human
Capital Development and Operations Policy Working Paper. World Bank, Washington, D.C.
Daun, H. 2003. "Markets Forces and Decentralization in Sweden: Impetus for School Development or
Threat to Comprehensiveness and Equity?" In D. Plank and G. Sykes. eds., Choosing Chioice: School
Choice Policies in Internzationial Perspective. New York: Teachers College Press.
Espinola, V. 1995. "The Decentralization of the Educational System in Chile." cIDLWorking Paper. Cen-
ter for Research and Teaching Economics, Mexico City.
Filer. R., and D. Munich. 2003. "Public Support for Private Schools in Post-Communist Central Europe:
Czech and Hungarian Experiences." In D. Plank, and G. Sykes, eds., Choosing Choice: School Choice
Policies in International Perspective. New York: Teachers College Press.
Fiske, E., and H. Ladd. 2000. WVhen Schools Compete: A C..Li' i.Zj,i Tale. Washington, D.C.: Brookings
Institution Press.
- 2003. "School Choice in New Zealand: A Cautionary Tale." In D. Plank and G. Sykes, eds.,
Chzoosing Choice: School Choice Policies in Iuter,iationzal Perspective. New York: Teachers College Press.
Gauri, V. 1998. School Choice in Chile: Two Decades of Educational Reform. Pittsburgh. Pa.: University of
Pittsburgh Press.
Gewirtz, S.. S. J. Ball. and R. Bowe. 1995. Markets, Choice and Equity in Education. Buckingham, UK:
Open University Press.
Gibson. A.. and S. Asthana. 2)00. "What's in a Number? Commentary on Gorard and Fitz's 'Investigat-
ing the Determinants of Segregation between Schools.'" Research Papers in Education 15(2): 133-53.
278                                        TI'he WVorld Batik Researchl Observer. vol. 19, nio. 2 (Fall 2004)



Gill, B.. P. Timpane, K. Ross, and D. Brewer. 2001. Rhetoric versus Reality: Wlhat Wtre Knlotw and lWmat lte
Need to Know abouit Vouchiers and Clharter Sclhools. Santa Monica. Calif.: Rand Education.
Gorard, S.. and J. Fitz. 2000. "Investigating the Determinants of Segregation between Schools."
Research Papers in Educationi 15(2): 11 5-32.
Greene, J. 2001. "An Evaluation of the Florida A-Plus Accountability and School Choice Program."
Center for Civic Innovation at the Manhattan Institute, New York.
Greene, I., P. Peterson. and l. Du. 1997. "Effectiveness of School Choice: The Milwaukee Experiment."
Harvard University Program in Education Policy and Governance. Cambridge, Mass.
- 1998. "School Choice in Milwaukee: A Randomized Experiment." In Paul E. Peterson and
Bryan C. Hassel, eds., Learnin2gfromri Sclhool Choice. Washington, D.C.: Brookings Institution Press.
Holmstroni, B. 19 79. "Moral Hazard and Observability." Bell Journal of Economtiics 1o: 74-91.
Holmstrom, B., and P. Milgrom. 1987 . "Aggregation and Linearity in Provision of Intertemporal Incen-
tives." Econoinetrica 55(2):303-28.
Howell, W., and P. Peterson. 2002. Thze Education Gap: Vouchers anrd Urban Schools. Washington, D.C.:
Brookings Institution Press.
Howell, W.. P. Wolf, P. Peterson, and D. Campbell. 2000. "Test-Score Effects of School Vouchers
in Dayton, Ohio. New York City, and Washington, D.C.: Evidence from Randomized Field Trials."
Paper presented at the American Political Science Association, September. Washington. D.C.
Hoxby. C. M. 2001. "How School Affects the Achievement of Public School Students." Hoover Institu-
tion, Stanford, Calif.
Hsieh, C.-T., and M. Urquiola. 2002. "When Schools Compete, How Do They Compete? An Assessment
of Chile's Nationwide School Voucher Program." Occasional Paper 43. National Center for the
Study of Privatization in Education. Teachers College, Columbia University, New York.
King. E.. P. Orazem, and D. Wolgemuth. 1998. "Central Mandates and Local Incentives: The Colombia
Education Voucher Program." Series on Impact Evaluation of Education Reforms Working Paper 6.
World Bank, Development Economics Research Group, Washington. D.C.
King. E., L. Rawlings, M. Gutierrez, C. Pardo, and C. Torres. 1997. "Colombia's Targeted Education
Voucher Program: Features, Coverage and Participation." Series on Impact Evaluation of Education
Reforms Working Paper 3. World Bank, Development Economics Research Group, Washington, D.C.
Krueger, A., and P. Zhu. 2002. "Another Look at the New York City School Voucher Experiment." NBER
Working Paper 9418. National Bureau for Economic Research, Cambridge. Mass.
Ladd. H. 2002. "School Vouchers: A Critical View." Journal of EconomicPerspectives 16(4):3-24.
LaRocque. N. 1999. "An Assessment of the Private Education Sector in Cote d'Tvoire." World Bank,
Human Development Network, Education Group. Washington. D.C.
Le Grand. J. 1991. "Quasi-Markets and Social Policy. " Econornic Journal 101: 12 56-67.
Levacic, R. 200]. "An Analysis of Competition and its Impact on Secondary School Examination
Performance in England." National Center for the Study of Privatization in Education. Teachers
College. Columbia University, New York.
Levacic. R., and P. A. Woods. 1999. "Polarisation and Inequality between Secondary Schools in
England: Effects on School Practice and Performance." Paper presented at the British Educational
Research Association Annual Conference, University of Sussex.
Mayer, D., P. Peterson, D. Myers, C. Tuttle, and W. Howell. 2002. "School Choice in New York City
after Three Years: An Evaluation of the School Choice Scholarship Program." Mathematica Policy
Research, Washington, D.C.
McEwan, P. J. 2001. "The Effectiveness of Public, Catholic, and Non-Religious Private Schooling in
Chile's Voucher System." Education Economiics 9(2):] () -28.
t'a1ruiw GarN-i and Ayesha VaIvda                                                   279



McEwan, P. J., and M. Carnoy. 1999. "The Impact of Competition on Public School Quality: Longitudi-
nal Evidence from Chile's V7oucher System." Stanford University, Stanford. Calif.
.2000. "The Effectiveness and Efficiency of Private Schools in Chile's Voucher System." Educa-
tional Evaluation and Policlj Analyqsis 22(3):213-39.
Michaelowa, K. 2001. "Primary Education Quality in Francophone Sub-Saharan Africa: Determinants
of Learning Achievement and Efficiency Considerations." World Development 29(10): 1699-716.
Milgrom. P., and J. Roberts. 1992. Economics, Organization, and Managemnent. Englewood Cliffs. N.J.:
Prentice Hall.
Millington, I., and S. Bradley. 1998. "The Effect of Spatial Competition on Secondary School Exam Per-
formance." CREED Discussion Paper 4/98. Collaborative Research in Economics and Development.
Lancaster University, Department of Economics, Lancaster, UK.
Mizala, A., and P. Romaguera. 1998. "Desempeno Escolar y la Eleccion de Colegios: La Experiencia
Chilena." University of Chile. Santiago.
Montenegro. A. 1995. "An Incomplete Educational Reform: The Case of Colombia." Human Capital
Development and Operations Policy, Working Paper 60. World Bank, Washington. D.C.
Myers, D., P. Peterson, D. Mayer. J. Chou. and W. Howell. 2000. "School Choice in New York City after
Two Years: An Evaluation of the School Choice Scholarship Program." Mathematica Policy
Research, Washington, D.C.
Neal, D. 20))2. "How Vouchers Could Change the Market for Education." Journal of Econornic Perspec-
tives ] 6(4):25-44.
Patrinos. H. 2002a. "Private Education and Public Finance: The Netherlands as a Possible NModel."
Occasional Paper 59. National Center for the Study of Privatization in Education, Teachers College,
Columbia UJniversity. New York.
- 2002b. "A Review of Demand-Side Financing Initiatives in Education." World Bank, Washington,
D.C.
Rodriguez, 1. 1988. "School Achievement and Decentralization Policy: The Chilean Case." Revista de
Analisis Econioinico 3(1):75-88.
Rouse. C. 1998. "Private School Vouchers and Student Achievement: An Evaluation of the Milwaukee
Parental Choice Program." Quarterly Journal of Economic 113(2):553-602.
Shavell, S. 1979. "Risk Sharing and Incentives in the Principal and Agent Relationship." Bell Journal of
Economnics 10(1):55-73.
Smithfield Project. 1995. Trading in Futures: The Nature of Chzoice in Educational Markets in New Zealand.
Wellington. New Zealand: Ministry of Education.
Walford. G. 1997. "Diversity, Choice and Selection in England and Wales." Educationial Administration
Quarterly 33(2):31-44.
.200(3. "School Choice and Educational Change in England and Wales." In D. Plank and G. Sykes.
eds., Choosinig Choice: School Choice Policies in Interniational Perspective. New York: Teachers College Press.
Witte, J. 20)0)0. The Market Approach to Education. Princeton. N.J.: Princeton University Press.
World Bank. 1998. Cote d'Ivoire: Educatiorn and Training Support Project. Washington, D.C.
2000. Bangladesh: Education Sector Review. Washington, D.C.
200(1. Czech Republic: Enh1anicinig the Prospects for Growth with Fiscal Stability. Country Study.
Washington, D.C.
- (2003. World Developmenit Report 2004: Making Services Workfor Poor People. Washington, D.C.
Wylie, C. 1998. "Can Vouchers Deliver Better Education, A Review of the Literature, with Special
Reference to New Zealand." New Zealand Council for Educational Research, Wellington.
280                                        The WVorld Batik Researcli Observer. vol. 19, no. 2 ( Fall 2004)



Oxford Journals Online
uuiiectuun                                                    *"             7I77'I _71-
- --   -                    ;~    T~ i    '¢5;^n             j   \   rw           - -.
Are you 1ookioi6fnd1&                     f
subject area qutcldy and estj fOm a w                   ma   -
af different sources?
liFp.pl,'r.,tso c.e, I'D ,neIrro3tnally accIaimed I0utrnM illies onirre n
IhV 3 arr- .:.1
, HUMANI, If
. CiiiAL ':cliIK:ES
1- A.WJ
*C (pMI,i C  Eu LI  Ai J  ON F IrJa CPE
LS IVL  :'', 1':  1   '  1:  f I: 1-:.
* MEDICINE AND HEALTH SCIENCES
. PUBLIC HEAUH AND EPIDEMIDLOGY
MATHEMATICS, STATISTICS, COMPUTING, AND PHYSICS
Your library can subscribe to the whole collection, a particular subject
area, or a tailor-made package and enable you to receive full access to
journal articles (full text HTML or POF), all with advanced online
functionality.
Your librarian can control your online subscriptions quickly and easily and
can review your institution's usage statistics, 24 hours a day.
Pass this information to your librarian and encourage them to visit
www.oupjournals.org/librarians
F  s  o                       .         - i. ni,i                        _ 
w ww. o u pj o u r n a                              s .o rg
OXFORD
UNIVERSITY PRESS



I *_   = ;wz&w
~CD + --i Q>U> 0 E  ~5.Om   >C.
aD)      CCD °X- o   X   1CD  CD  (  =  °
<     Q  - M      < 4 -  ) zr a     0.
(D        -    Co  CD  CD  c D ,  D cg
3  CD  X N)             3
o~~~~~~~~~~~~~~~~~~~~~ Co 00
CD
C)  C:
0                    Co~~~~~~~~~~~.
3



HUMAN DEVELOPMENT REPORT 2004
rrs,-ar            (.ltilm I eh'ilmenlb ( Imat/c,s (,ttih 1) ndsv   PerS/
@  i         ~~~United Nations Development Prograomme
.i  /fill rl/S   O)wi/iipm-ltt   /f?ip ol 201,1i (i-bunilts  p zildis  iiiitt Iq il.ti t   listItit ,a
d1wilsts, ni   ti tti t  itl i btit, tIsitiltiw 1 s'  spi n o itisit, it a ti ittIliteaiige ta ohAuli
o deia insoiltl siigttwiis  wi;tk.ilhlni lig  lifi at itlirritig lil ti( li ss  bIt tiltlloatit
-\a1filtiv, i'sil", that 0w   (IN 1'.17111c..   of globi i/aTill  hine l ought to i; l 1lwl r
i7             t)l-:l\\-~fielf"11 t   mi(N.lt: gvf ida, lIIi,l.l II)'fid  pool   cooXllu itl>irv El]l1l  Inchuh. i
rw~~~~~/?, ps9ttlopo- >l)szNidltlwe-bawd coistwiutiona:l. juilidi(;sJ an  ocliovi<(-onolinlic
si |-jt I)e~~~~~~plic% optlifon to plolnXlo ctilui(llr.l]ibvfn tad :111111 11iill-lifa:l dvi llo( lal(N inl tti
EE       -            u<>~~~~~~~~crlt'X   o  Y t *, ili slilNc   1 s% a,  )litd   hilt,ililli I ighl
WORLD DEVELOPMENT REPOSRT 200
wotledrlain>n$orr i   7  Et halfstwent Cnf Wtiate (srow/h, nnit/ /n7overiv
WorW Baank
c18olloi-/ /llil, ldviwr,,l-lld acadc, 2ui5 ,I , w it  al.  I ,i&-t   u ,,,k,d e l'tning ss (Is l<le 
To o ilrde  o ilrl iot lp gmor nllirclmia tolit,it c  ll  1 -0 45  -7556
imell^ lill lit "11lell(,b 11 e\i;1miliclv   whatR:   gowl llillenlt s .1im   Ill) to  ilil)(l ow   Ill,
illv,i  cm   hilat.  of dili   omit i-,,purgl,il Ih,aill((I r t,e1v* t  TlW
(iN WR'ItY PRESS " In Canada.  I1 -8038o- 1cp0 llc   d1111g0ilct'd % (.lwlopillg
.itittit Eztt~srttt~ ett J I liVt nedsile t  oggsel prlcasl wilot-pgiom /ura(cchrati/g s/hgi9hs. eI builds off
,3           ~~~~ilifoslilial:olsl fslom paste W'lrld lh(-kclopmuntl2 ReprXtl-.mid dim on-ls(l nc%,N
z          lnI    o   lur-, ew   d a.  t  a ,'1dc   it   e   e iih
# jji 5 sE 5; j ~~~~IN DEFENSE OF                           CEOUIUAliATION
il 4         t     --   ^    ;    1      WHO~~~~~'S AFRAIDi OF THE WTO?
' _s;E- -- i il~~~~~~~~~~~ent Jones
= = s *|EEX-it ~~~THE NEW                            RUTNLESS ECONOMY
_ * * ~~~~~~~~~~~~Simon Head
2uti 140pp p<ff
OXFOnr                To rOrder or for more information, call l1-800-451-7556.
IVNIAEtSlT'Y PESS     In Cn"ado, call 1-800-387-8020.
Ifirimit Erfu(Ario)\S GROUvP  Visit cilt- website otl wwwoup corn/us/higherecf



T H E W   R L D  3 K             - A
~~,~  =A --4
V    -~~~~~~~.
BN 019 8530668
078019"83387"2