_)__ __\A 1G66
POLICY RESEARCH WORKING PAPER 1666
Is Growth in Bangladesh's The current level of per capita
production of rice in
Rice Production Sustainable? Bangladesh can be sustained
only through increased yields
of modern rice varieties.
John Baffes
Madbur Gautam
The World Bank
International Economics Department
Commodity Policy and Analysis Unit
October 1996
POLICY RESEARCH WORKING PAPER 1666
Summary findings
The recent growth of foodgrain (primarily rice) level of per capita production can be sustained only
production in Bangladesh has outpaced population through increased yields from modern rice varieties.
growth largely because of the spread of green revolution Other factors that could affect growth in per capita
technology. The transition from being labeled a "basket rice production are population control (which is found
case" in rhe early 1970s ro the virtual elimination of rice to have significant long-term benefits) and faster
imports in the early 1990s is particularly remarkable conversion of remaining areas to modern varieties (which
considering the severe land constraint in Bangladesh. is found to have important short-run payoffs).
Two decades of concerted government efforts to achieve But population control and faster conversion to
rice self-sufficiency have created both an atmosphere of modern varieties are only complements for the most
optimism and concerns about whether rice self- important factor: efforts to increase the yields from
sufficiency is sustainable. modern rice varieties. If policies designed to raise the
Baffes and Gautam find that rice production grew in overall rate of economic growth and reduce poverty
Bangladesh between 1973 and 1994 mainly because of succeed, it will be even more critical to focus on
the conversion of rice-growing areas from local to increasing productivity.
modern varieties. Simulations suggest that the current
This paper - a product of the Commodity Policy and Analysis Unit, International Economics Department - is part of a
larger effort in the department to analyze global food security issues. Copies of the paper are available free from the World
Bank, 1818 H Street NW, Washington, DC 20433. Please contact Pauline Kokila, room NS-030, telephone 202-473-3716,
fax 202-522-3564, Internet address pkokila@worldbank.org. October 1996. (29 pages)
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about
development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The
papers carry the names of the authors and should be used and cited accordingly. The findings, interpretations, and conclusions are the
authors' own and should not be attributed to the World Bank, its Executive Board of Directors, or any of its member countries.
1Produced by the Policy Research Dissemination Center
Is Growth in Bangladesh's Rice Production
Sustainable?
John Baffes
Madhur Gautam
The authors would like to thank Jock Anderson, John McIntire, and Don Mitchell for helpful
comments and suggestions on earlier drafts. The paper has greatly benefited from comments by
seminar participants in Dhaka, Bangladesh (April 1995) and the World Bank (May 1996).
TABLE OF CONTENTS
1. Introduction .............................. 1
2. Identifying the Sources of Growth ............. .. 2
2a. Growth Decomposition ............................... 3
2b. Estimating the Rate of Growth ................................: 4
3. Results ..............................5
4. Simulations ............................... 8
4a. Projecting Per Capita Rice Production ............................... 8
4b. Projecting Per Capita Rice Demand ............................... 10
4c. Simulation Results ............................... 10
5. Conclusions .............................. 14
Endnotes ..15
Tables .......... 16
Figures.. 20
References ..23
Appendix A: Kinked Growth Rate Regression .26
Appendix B: Growth Rate Estimates .27
I
1. Introduction
The growth of foodgrain production in Bangladesh - primarily rice - has outpaced population
growth largely because of the spread of green revolution technology. The transition from being
labeled a "basket case" in early the 1970s to the virtual elimination of rice imports in the early
1990s is particularly remarkable considering the severe land constraint in Bangladesh. A
culmination of two decades of concerted efforts by the Government towards rice self-sufficiency,
this feat has simultaneously created an atmosphere of optimism and raised concerns about its
sustainability.
The discussion on the sustainability of output growth in Bangladesh is topical in light of
the current debate on the Earth's carrying capacity. Projections of future global food demand and
production levels, although based on the same historical data, differ depending on the
assumptions maintained (McCalla). For example, Mitchell and Ingco argued that if past trends in
crop yields continue and population growth rates decline as projected, then improvements in the
world food situation as seen in the past 30 years should continue. Brown and Kane, on the other
hand, argue that several constraints are emerging to prevent past trends from continuing.
Additional research also suggests that the current, relatively comfortable food situation
may not last for long and that complacency at this time is misplaced and premature (Pinstrup-
Anderson and Pandya-Lorch). With limits on cultivable area, rapid population growth,
stagnating yields, and potential resource degradation, the future does not appear to be promising.
It is generally argued that raising yields are the key to promoting and sustaining growth in food
output. In the long run, this would entail increased research to develop superior grain varieties;
in the short run, it requires dissemination of accumulated knowledge, delivery of appropriate
technology, and input intensification to close existing yield gaps (Anderson et al.).
The objective of this paper is to evaluate the future prospects of rice for Bangladesh under
alternative strategic policy directions. The analysis serves two purposes. One is to assess the
future growth prospects for Bangladesh's agriculture. The analysis is focused on rice because it
is the most important crop, contributing 70% of crop GDP, and over 50% of the agricultural GDP
which in turn accounts for a third of national GDP. More importantly, past growth in gross crop
revenue has been due almost entirely to the growth in rice production, raising concerns whether
past performance can be maintained, and if so, for how long.
l
The second purpose of the analysis is to use Bangladesh as a case study to contribute to
the global debate on the sustainability of food output. With abundant water resources and low
rice yields, Bangladesh has a relatively greater potential for expanding food production by
maintaining the momentum of the green revolution than many other developing countries (Brown
and Kane). Thus, with an a priori optimistic outlook, the results would be indicative of the
prospects for other developing countries.
To identify the potential for further growth, the first part of the paper decomposes past
growth in rice production to identify its major sources. The analysis is restricted to the growth in
area and yields of rice varieties to study the impact of physical constraints on the spread of the
current technology, and hence on the growth process. This framework allows identification of
variables for broad policy direction. It is also complementary to micro-economic analyses; e.g.,
the total factor productivity approach which accounts for the contribution of inputs to output
growth and-duality applications that study the structural properties of production technology.
The second part of the paper is concerned with the conditions under which past growth
can be sustained in the future. These conditions embody options affecting population growth,
rate of conversion from traditional to modern rice varieties, rice yields, and income growth. The
alternatives give profiles of per capita rice production and consumption over the next 20 years
and are used to identify areas on which policy must focus to sustain growth in output. It must be
emphasized that the purpose of these simulations is to qualitatively assess the effects of alternative
policy thrusts on long-run outcomes. The conditions underlying each simulation are intended to
reflect the necessary actions that might be needed to achieve the intended outcomes, without
making any judgment on the desirability of specific policies.
2. Identifying the Sources of Growth
There are three rice cropping seasons in Bangladesh. Aus is a short-duration crop which is
directly seeded in March-April and harvested in July-August, utilizing the pre-monsoon rain
water. Aman, from June-August to November-January, is the monsoon crop. It grows with the
flood waters and is harvested after the floods recede. Shorter duration modern Aman varieties
have made room for short-duration non-rice crops (oilseeds and pulses) before the high-yielding
dry season crop is transplanted. Boro, from November-January to April-June, is the dry season
crop. With the development of controlled (groundwater) irrigation, Boro modern variety rice has
expanded rapidly at the expense of Aus.
2
Both local (Lv) and modern (Mv) varieties can be grown in all three seasons.2 However,
because of overlapping production cycles, area under the more profitable Boro expands at the
expense of Aus and broadcast Aman. Since Boro rice is almost entirely irrigated, area allocated
to it can also be viewed as irrigated area. Table 1 gives the 1973-75 and 1992-94 average rice
production, area, and yield for the three seasons and two rice varieties.3
2a. Growthl Decomposition
4
This section decomposes past growth in rice production. Production in year t, Q, is the sum of
the output of the two varieties in the three seasons:
23J
i=I j=I
where i = MV, LV and j = Aus, Aman, and Boro; thus Q,Ij denotes production of rice variety i in
seasonj. Expressing Q,j as the product of area, A,' and yield, Y,IJ, (1) can be written as:
2 3
Q, = Y AY1'j. (2)
i=1 j=I
The growth rate of total rice production, p, for a particular time interval (t = m, ...,
can be expressed as the weighted sum of growth rates of each component:
2 3
p = Y, E U ±p (pA'J + p!/), (3)
1=1 j=I
where PAU and p,'J denote growth rates of area and yield of variety i in seasonj and wv'j is the
corresponding weight, with all weights summing to one. The weights are defined as:
1 " 0 (4)
n - m +1,,
Depending on the specific objective of the decomposition, n and m can take any value within the
sample range.5
Inserting (4) into (3) and setting n = T and m = 1 gives the final formn of decomposition:
P EtT Q 0(P P.) (S)
3
Growth in total rice production in (5) is thus decomposed into 12 parts defined by the two
varieties (Mv and LV), the three seasons (Aus, Aman, and Boro), and the two components (area
and yield). Finally, dividing both sides of (5) by p and multiplying them by 100 gives the
decomposition in percentage terms.
2b. Estimating the Rate of Growth
Suppose that Q,, which grows at rate p, is subject to stochastic shocks, u,. That is,
Q, = (J+p)Q,.u,. (6)
It is assumed that a multiplicative error, representing proportional shocks, is a more appropriate
representation than an additive error, representing similar magnitudes of shock at all levels of
output (Banerjee, et al.). From (6) it follows that Q,-1 = (1 +P)Q,U2U.1. Inserting Q, 1 into (6)
gives Q, = (1 +p)2 Q,.2u1u,, while recursive substitution yields:
Q, = ( + p) Qou,u,, ... uo (7)
Taking logarithms in (7) and setting u = ln(Qo), ,l? = ln(1 +p), and s, = X,ln(u,) gives,
ln(Q,) = A+ -t + s, (8)
,B can be estimated with OLS and the growth rate can be calculated as p = e6 - 1, where e is the
base of the natural logarithm. To what extent 18 is a reliable estimate, and hence reflective of the
true growth rate, depends on the properties of F,. If s, is a stationary process, then p reflects the
true average growth rate over the sample period. If, on the other hand, £, is nonstationary then
the average growth rate as hypothesized in (6) may not exist (in the sense that the stochastic
component drives Q, in (6) rather than the deterministic one).
In the case where s, is I(1), an alternative way to estimate the growth rate would be to take
logarithms in (6), rearrange terms and substitute as before:
ln(Q, / Q,,) = , + v, (9)
where v, = ln(u,). The growth rate as given in (9) is simply the arithmetic mean of logarithmic
changes of Q,. Thus, for an I(1) process, only the information contained in the first and last
observations of the sample is useful in estimating the rate of growth.6
To determine if the estimated coefficients represent firm estimates of the underlying
growth rates, the performance of (8) is assessed by both conventional and stationarity statistics.7
4
Further, although this single equation framework implicitly imposes the weights (4) used in the
decomposition (5), the validity of this restriction can be formally tested using the nonlinear cross-
equation restriction that the weighted contribution of each growth component adds up to the
growth rate of total output.
3. Results
Bangladesh Bureau of Statistics (BBS) data on rice area, yield, and production for the 1973-94
period are used for the analysis.8 The full set of regression results are reported in appendix B
(table B.1). The test for the restriction on weights (4) yielded a x2 statistic of 0.32, far lower
than its 5% critical value counterparts. Hence, the restriction that the weighted contribution of
12 components add up to the estimated total growth rate could not be rejected. The estimated
growth rates, based on unrestricted OLS estimates, are reported in table 2.
Total rice production grew at an annual rate of 2.65%. Rosegrant and Pingali report that
the rice output growth rate for Asia increased from 2.60% per annum during the pre-green
revolution period (1958 to 1966) to 3.30% during the post-green revolution period (1966 to
1982). Thus, the 2.65% growth rate places Bangladesh well below the Asian average for the
post-green revolution period. This is also consistent with evidence of low technical efficiency in
the production of MV rice in Bangladesh (Sharif and Dar). Seasonal estimates show that Boro
rice production has grown substantially (7.07%) while Aman production has grown at a much
lower rate (1.94%) and Aus production has declined (-1.36%).
The growth in gross area allocated to rice production has been negligible (0.20%). That
is, almost all growth in rice production has been due to increased average yields. At the seasonal
level, area allocated to Aman rice has remained unchanged while area allocated to Boro rice has
increased at the expense of Aus (at rates of 5.55% and -2.66%, respectively).
In all three seasons average yields experienced steady growth: 1.33%, 1.98%, and 1.44%
for the Aus, A man, and Boro season, respectively. The growth rates of all seasonal yields are
below the overall growth rate of 2.44%, because a significant part of the overall yield growth has
been due to the increase in Boro area (from 1,065 to 2,605 thousand hectares) largely at the
expense of Aus (from 3,072 to 1,767 thousand hectares).
Distinguishing local and modern variety yields highlights some important features.
Average Mv yields (figure 1) remained virtually unchanged while LV yields (figure 2) grew at an
average of 1.29% for the entire 1973-94 period. Disaggregating at the seasonal level, Aus MV
5
yields declined at the rate of 1.92% while they remained almost unchanged for Boro and Aman
(0.03 % and 0.10%, respectively). Growth in LV yields for the Aus and Aman seasons was higher
(1.25% and 1.29%, respectively) relative to the Boro LV yield growth rate (0.42%).
In the mid 1980s, the Government of Bangladesh liberalized agricultural input markets
and deregulated the import of minor irrigation equipment (Hossain). This led to a rapid
expansion of irrigated acreage (especially Boro) in the late 1980s. To allow for a structural break
due to the policy reforms, growth rates were re-estimated for the two subperiods 1973-87 and
1987-94 using a kinked growth regression (see appendix A for a description of the model). As in
the constant growth rate model, the hypothesis that the weighted contribution of the 12
components add up to the estimated total growth rate in each subperiod could not be rejected with
a i statistic of 0.33. The growth rates of the two subperiods are reported in the middle and
lower panels of table 2.
For the pre- and post-87 periods, annual growth in total rice production was estimated at
2.60% and 2.84%, respectively, while the growth in rice area was estimated at 0.47% and -
0.81%, respectively. Annual MV yields in the pre-87 period experienced a decline of 0.41%,
while they increased by 1.42% in the post-87 period. Overall, the results confirm a structural
change in MV yields but not in LV yields. The -0.41 % growth rate of MV yields in the earlier
period is consistent with Rosegrant and Evenson's finding that there was little total factor
productivity growth in Bangladesh between 1965 and 1985.
The negative growth rate of MV yields in the pre-1987 period raises the important issue of
declining yields due to resource degradation. For example, Pingali, Moya and Velasco present
evidence from the Philippines and other countries indicating stagnating and declining yields in
intensively cultivated irrigated rice even in experiment station fields. Pagiola presents similar
evidence of declining yields using data from Bangladeshi experiment station fields.
The upper panel of table 3 reports the weighted contribution of area and yield to overall
growth by seasons and varieties for the 1973-94 period. Boro contributed 68% to output growth;
Aman contributed 40% while Aus contributed a negative 10% to overall growth. Further
decomposition reveals that: MV Aman and MV Boro yield contribution was negligible to overall
growth (1 % and 0%, respectively) while MV Aus yields reduced it by 4%. On the other hand,
both LV Aus and MV Aman yields made a noticeable contribution (7% and 19%). MV and LV
yields together contributed about one fourth to overall growth. The major portion of growth
6
(three fourths) was the result of conversion from LV to MV within the Aman season as well as
from LV in the Aus season to MV in the Boro season.
The contributions by season and variety for the two subperiods present a clearer picture of
the movement from LV to MV and from Auis to Boro production. The difference is the
significantly greater contribution of Boro and Aman MV area and yields to overall growth in the
post-87 period, while MV Aman yields declined in the pre-87 period. It is evident from the lower
panel of table 3 that the increase in post-87 MV output was at the expense of LV output in both
Aus and Aman, primarily due to the decline in the area devoted to these crops.
To summarize: conversion from LV Aus and LV Aman to MV Boro contributed the most
to the growth of rice production followed by conversion from LV to MV during the Aman season
and by LV yield growth in the Aman season. Thus, the success of the green revolution in
Bangladesh should be attributed mostly to the conversion from local to modern rice varieties and
not to increase in modern variety yields. Moreover, the fact that in the last 10 years real crop
GDP grew at 1.87% annually and that rice contributes about 70% to crop GDP suggests that
almost the entire growth in crop GDP has been due to the conversion from local to modern rice
varieties.
Wheat production in Bangladesh has followed a similar pattern. In 1973 it covered
119,000 hectares with 82% under local varieties. By 1987 the entire wheat area (540,000
hectares) was under modern varieties. Production before 1987 grew at an annual rate of 25.52%.
Of this growth, area expansion accounted for 16.59% and the conversion from LV to MV and
yield increases accounted for the rest. Between 1987 (when the completion of the conversion to
modern varieties was completed) and 1994 wheat production grew at only 1.28% annually, with
three-fourths of this growth due to area expansion and one-fourth to yield increases.
Similar findings are reported in a recent study by Byerlee and Siddiq regarding wheat
production in Pakistan. They write that "The most striking trend is the slow rate of progress in
yields in areas already sown to Modern Varieties ..." (p. 1347; emphasis in original). In
particular, they found no trend over the 1967-90 period (roughly our sample period) in MV
yields. The main growth in wheat production came from either conversion of rainfed to irrigated
land or the switch to modern varieties. Furthermore, their evidence "... raises serious concerns
about Pakistan's ability to sustain even present levels of food self-sufficiency to 2000 and
beyond". (p. 1358). Faruqee also reports that MV rice yields in Pakistan have stagnated during
the 1980s while they grew at only 0.20% annually during the 1970s.
7
In a study of rice production in the Philippines, Bouis reports that increases in proportion
of area planted to MV rice was one of the major sources of growth (in addition to shifts in the
fertilizer response function). Furthermore, MV rice in the Philippines has approached 90% of all
rice area and can no longer be counted on as a source of growth in the future.
The decomposition findings have important policy implications for the simple reason that
each of the growth components has a limited potential for expansion. For example, the land
potential has already been exhausted. Thus, conversion from local to modern varieties is limited
by the total area allocated to rice. When this potential is exhausted, assuming that current yield
trends continue, the growth in rice production will decline significantly. Coupled with a growing
population, this will imply a decline in per capita production.
4. Simulations
Based on the findings of the previous section, this section examines whether and under what
conditions growth in rice production can be sustained. This is done by simulating the likely
impact on per capita rice production of alternative assumptions about MV yield changes, the rate
of and limits to adoption of Mv technology, and population growth. Finally, to evaluate future
excess supply/demand of rice, we also simulate changes in aggregate per capita demand for rice
using the historical GDP growth and average income elasticity estimates.
4a. Projecting Per Capita Rice Production
Projections of per capita rice production (q,) are based on the following relationship:
q= A, y',(+ p' - pPP) (10)
i-l
where ,JP0 denotes the population growth rate while y, is per capita yield. Ideally, one would use
the growth rates and rates of conversion from LV to MV for the three seasons. However, that
would complicate matters. To keep the analysis focused on main issues, simulations use
aggregate levels of the key determinants of q, in (10): MV yield growth, MV adoption, MV upper
bound, and population growth. A brief discussion of each follows.
MV YIELDS: The rate of growth for the entire period in MV yields was found to be -0.03%
(which for practical purposes is viewed as zero). Accounting for the structural change, the
growth of MV rice yields was -0.41% in the pre-87 period and 1.42% in the post-1987 period.
These provide 3 potential MV yield growth paths for the future. To these is added a relatively
8
more optimistic scenario whereby Bangladesh is expected to increase its MV yields from the
current level of 2.41 metric tons per hectare (MTH) to 3.50 MTH by the end of the 20-year period
under consideration, matching the current performance of other Asian countries. This would
entail an MV growth rate of 2.00%. Coincidentally, the projected 2.00% growth in MV yields is
also the actual growth rate between 1987-91, reflecting the fact that although this rate might
appear to be optimistic, it is not an unrealistic objective. Thus, the four assumptions regarding
potential MV yield growth rates are: (i) -0.41%, (ii) 0.00%, (iii) 1.42%, and (iv) 2.00%.
MV ADOPTION: Between 1973 and 1994, the proportion of rice area allocated to MV grew from
11 to 50% reflecting a rate of growth of 7.61 % (figure 3). Under this historical rate of
conversion, in less than 8 years 90% of rice area will be under modern varieties. However, as
the conversion is closely associated with irrigation expansion, which is becoming progressively
more costly, it is unlikely that the historical rate of conversion can be maintained. Accordingly,
the two rates of conversion from LV to MV considered are 3 % and 5 %. These may be viewed as
(relatively) pessimistic and optimistic forecasts, although both are less than the historical trend.
MV UPPER BOUND & TOTAL RICE AREA: A third important consideration is the upper bound to
the spread of MV technology. This is determined by the total area allocated to rice and limits on
the maximum attainable MV intensity. The gross area allocated to rice has remained virtually
constant over the past two decades. The agro-ecological circumstances of Bangladesh are a major
constraint to further increase the rice cropping intensity, which at 182% is already among the
highest in the world. Short duration varieties are already being used in the mid to higher
elevations. In the lower elevations, over a third of the area is prone to annual flooding, limiting
the potential for increasing the cropping intensity. For simulation purposes it is assumed that
gross area allocated to rice in the future will remain constant at the 1992-94 average of 10,135
thousand hectares. Given this, two limits to the expansion of MV intensity are considered: 65%
and 80% of rice area. To a limited extent, the 65/80% assumptions implicitly incorporates minor
increases in gross rice area.
POPULATION GROWTH: The last parameter considered is population growth (an important policy
variable). Two population growth rates are assumed: (i) a fixed rate of 2.05%, which was the
1980-90 inter-census growth rate, and (ii) a declining schedule which assumes an aggressive but
attainable population policy aiming at 1.79% growth from 1996 to 2000, 1.59% from 2011 to
2005, 1.36% from 2006 to 2010, and 1.16% from 2011 to 2014 (Bos et al.). This schedule is
9
consistent with an average annual growth of 1.49%. The first assumption represents the more
pessimistic outlook (reflecting recent trends), while the latter is more optimistic.
The four MV yield assumptions, along with the assumptions corresponding to 3/5% MV
adoption rate, 65/80% MV ceiling, and 2.05/1.49% population growth, give a total of 32 per
capita rice production profiles, as depicted in figures 5 through 12. To put these graphs in
perspective, figure 4 presents the 1973-94 trend in per capita rice production.
4b. Projecting Per Capita Rice Demand
To assess the rice imbalance associated with each production scenario, per capita rice demand is
also projected. Using a semi-log specification, per capita demand at period t (denoted d,) is
projected as:
,= d,-, [1 + (PG -P _ pop) 7] (11)
where pGDP is the GDP growth rate and q is the income elasticity of the demand for rice.
The GDP growth rate over the last two decades has been 4.34%.9 The income elasticity of
rice is set to 0.30. With a zero rice imbalance in the base period (consistent with token net
exports in 1994), per capita demand is set at 158 kg (the 1992-94 average per capita rice
production). With these parameters, two demand profiles corresponding to fixed and declining
population growth are obtained. They are also depicted in figures 5 through 12.10
It is clear from (11) that besides population growth, GDP growth and the income elasticity
of rice will affect demand projections. We assess the sensitivity of the results to a change in GDP
growth rate (by raising it to 5.00%) and in the rice income elasticity of demand (by raising it to
0.40). The various production and demand scenarios outlined above allow us to evaluate the rice
imbalance that is likely to emerge in the medium and long terms.
4c. Simulation Results
The simulation results in figures 5 through 12 are summarized in table 4, which gives the
projected rice surplus/deficit as a percentage of per capita demand. In particular, the top two
panels present surplus/deficit per capita under the assumption that demand per capita remains
constant at the average 1992-94 level. In other words, it gives the change in production per
capita (PPc) over the simulation period." The lower two panels in table 4 present surplus/deficit
per capita (sPc) under the assumption that per capita demand changes according to (11). The
results presented are for years 2004 and 2014, reflecting medium and long run outcomes.12
10
The first general result is that of population control. In the medium run (i.e., 10 years),
an aggressive population policy has a marginal beneficial impact on both PPC (ranging from 2 to
3%) and SPC (from 1 to 2 %). The long-run impact is significantly greater, ranging from 7 to 14
for PPC and from 4 to 7% for SPC, depending on the various yield and MV intensity scenarios.
Thus, from a policy perspective, population control has substantial long run pay-offs.
The second noticeable result is the impact of the speed at which MV intensity proceeds. In
the long run, both paths result in the same levels of PPC/SPC for corresponding yield and
population growth scenarios - this is expected since the maximum attainable intensities of 65 %
and 80% are reached well within the 20-year time horizon considered here.13 Even in the
medium run, identical results are obtained for the 65% Mv ceiling case, which is reached in less
than 10 years under both rates of adoption. The benefits of a faster rate of adoption for this case
are limited to the short run (i.e., over the next 5 years or so). Raising the MV ceiling 80%,
produces a significant difference in the profile of the two paths to intensity over the medium run.
The accelerated path creates surplus for both demand scenarios that are 6 to 9% higher than the
slower path, depending on the assumptions regarding other parameters. These short to medium
run gains offer an opportunity for initiating a structural change in agriculture. The surplus under
the more optimistic scenarios could be mobilized either for growth in the nonagricultural sector
or to set the stage for diversification into nonrice crops. In short, increasing MV production
could open a valuable window of opportunity for agricultural development.
The impact of changes in MV yields is highly significant in both the medium and long
runs. As expected, assumption (i) on yield changes, which reflects resource degradation, has a
large negative impact on both PPC and SPC, irrespective of the assumptions regarding other
parameters. The magnitudes, even in the medium run, highlight the importance of addressing
emerging natural resource management issues promptly or even preemptively.
Assumption (ii), which reflects the zero MV yield growth, also leads to substantial
declines in Ppc and SPC in the long run. Even for the medium run, PPC is marginally positive
only in the most optimistic case with 80% MV ceiling, accelerated MV adoption, and aggressive
population control; in all other cases, it is nonpositive. Rice surplus is consistently negative, and
quite substantially so, in all cases in the medium run. These results point to the high cost of
maintaining status quo with respect to yield changes, irrespective of the value of other
parameters. They indicate that other policies are unlikely to compensate for stagnating yields.
11
The MV yield growth assumption (iii), which reflects the post-liberalization performance,
does better in terms of production per capita, but does not offer much promise in terms of
generating a surplus. In the long run, PPC is higher in all cases except that of lower MV ceiling
and higher population growth. In the most optimistic case (i.e., higher ceiling and lower
population growth) it is significantly higher. The corresponding performance of SPC is not as
good; it is marginally lower than the current level even in the most optimistic case and gets worse
with increasing pessimism about other parameters. In the medium run, PPC is higher in all cases,
but SPC is positive only with higher ceiling and adoption rate scenarios.
The MV yield growth assumption (iv) is the most optimistic. However, as noted earlier, it
is not overly ambitious since the MV yield growth rate of 2% was achieved between 1987 and
1991, the years immediately following the deregulation of agricultural input markets in
Bangladesh. Not surprisingly, this assumption provides the best results under both demand
profiles. PPC is positive under all parametric assumptions, with the magnitude increasing
considerably for the more optimistic cases. SPC is also positive for all cases in the medium run,
although only marginally with the slower adoption rate and lower adoption ceiling. In the long
run, SPC is positive only for the higher MV adoption ceiling, with population control also making
a significant contribution.
From a policy perspective, it is important to note the relative impacts of different
parameters. To keep the discussion focused, consider the sPC associated with yield growth
assumption (iv) for the 80% ceiling case. An aggressive population control policy increases SPC
by 2 % in the medium run and by 7 % in the long run; increasing the rate of MV adoption raises it
by 9% in the medium run, although it does not matter in the long run. The effect of increasing
Mv yield growth rate from 1.42 to 2.00% raises SPC by 5% in the medium run, but raises it by
11 % in the long run. Similarly, the increase in sPC relative to the outcome corresponding to
yield growth assumption (ii) is 18% in the medium run and 33% in the long run. 4
These results highlight that it is imperative to focus on yields if PPC and SPC are to be
sustained in the long run. In addition, population control has long run benefits. In the short run,
in addition to increasing yields, accelerating MV adoption will be the key to further growth.
The discussion has so far concentrated on the assumptions affecting production growth.
On the demand side, the projections were made under two assumptions: (a) with constant per
capita demand, and (b) with per capita demand changing in response to GDP growth at the
historical rate of 4.34% and an aggregate income elasticity for rice of 0.30. However, if
12
Bangladesh is successful in meeting the main objectives of its development policy, i.e.,
accelerating growth and poverty alleviation, these assumptions may not remain valid. Therefore,
we simulated the likely impacts of the success of these policies on the SPC.
Raising GDP growth rate from 4.34 to 5.00% raises demand per capita and reduces the
surplus. Thus, in the long run, sPc is positive only for the most optimistic scenario of MV yield
growth (assumption (iv)), accelerated MV adoption, aggressive population control and 80% MV
adoption ceiling; the magnitude of change, however, is reduced by half. In the medium run, the
SPC is reduced by 2% in all cases, with the rankings of different cases remaining as in table 4.
The impact of structural poverty reduction policies (e.g., growth with redistribution or
other pro-poor growth interventions) is summarily captured as an increase in the average
aggregate income elasticity from 0.30 to 0.40 since lower income groups have a higher income
elasticity of demand. The impact is to reduce spc, with magnitudes of change being
coincidentally similar to the case of increasing GDP growth from 4.34% to 5.00%.
To maintain or increase output, both these simulations on the demand side underscore the
earlier conclusion that attention has to focus on raising yields. They also have important
implications for food self-sufficiency as an indicator of economic performance. The results
suggest that achieving self-sufficiency can be very misleading when considered in isolation from
other indicators such as production growth. This is because a slow-down in the rate of growth of
GDP, or a more unequal distribution of income, will tend to produce a higher surplus (or lower
deficit). Thus, while the economic performance may be deteriorating and/or poverty increasing
(as indicated by a worsening income distribution), it might appear as if the country were doing
well as measured in terms of achieving self-sufficiency, a cherished goal in a number of
developing countries.'5
Finally, one methodological point related to the global food security debate is in order.
As mentioned in the introduction, studies have reached substantially different conclusions despite
using the same historical evidence. The results obtained here show that holding all other
parameters constant, varying the period over which the growth rate of modern variety yields is
estimated gives dramatically different outcomes. For example, for the population growth of
1.49% and MV ceiling of 80% in the changing demand case, MV yield growth based on the 1973-
94 or 1987-94 periods (0.00 and 1.42%) give a deficit of 32% or a surplus of 10%, respectively,
20 years hence. Or, relative to the 1992-94 average production of 10.9 million metric tons
(MMT), Bangladesh could face a deficit of 3.5 MMT or a surplus of 1.1 MMT, depending on which
13
estimate is used. Clearly, the former would qualify as a Malthusian outcome while the latter
would be a no problem outcome.
5. Conclusions
With the area allocated to rice remaining virtually unchanged during the 20-year period
examined, almost the entire growth in rice production in Bangladesh can be attributed to rising
average yields. Further decomposition, however, revealed that almost two thirds of the growth
was attributable to the conversion from local to modern varieties rather than to increases in
varietal yields. Before the 1987 deregulation of the minor irrigation equipment sector, modern
variety yields experienced an annual decline of 0.47%, raising concerns of resource degradation.
Simulation results reveal significant insights on the prospects for further growth in rice
output in Bangladesh. Population control has long-run benefits, and speeding up the conversion
from LV to MV could yield valuable surplus in the short to medium run. However, in the long
run, these measures by themselves are unlikely to be able to compensate for a failure to raise
yields. Further, if policies designed to raise the overall rate of economic growth and reduce
poverty succeed, it will be even more critical to focus on yields. The results also suggest that
focusing on self-sufficiency as a primary policy objective, in isolation from other indicators such
as economic growth and income distribution, can be very misleading.
These findings have important implications for the global debate on the future food
supply-demand imbalance. As an optimistic case for further gains from the current green
revolution technology, the outlook for Bangladesh is quite revealing. There is clearly no
alternative to increasing yields. This would entail a fuller exploitation of available technology by
closing the existing yield gaps, increased research efforts to develop varieties with higher yield
potentials, and additional research on the likely impact of current practices to pre-empt declining
yields due to resource degradation.
14
Endnotes
Farmers have little choice for diverting Aman land to other crops because of high soil moisture and poor drainage.
2 As Byerlee recently noted, in many developing countries there has been a continuous release of MVs since the 1980s.
Between 1967 and 1994 a total of 36 modern varieties have been released in Bangladesh - two by the International Rice
Research Institute (1967 and 1969) and 34 by the Bangladesh Rice Research Institute (1970 through 1994). At least 18
releases of the semi-dwarf and short-duration varieties have been unpopular because of lower yield potentials as well as
agro-ecological constraints (Hossain). The most popular variety, BR-I I (introduced in 1980), currently covers over 73 %
of total rice area. Almost all the remaining popular varieties have yields similar to BR- 1l. Therefore, to avoid
unnecessary complexity, a homogenous modern variety (MV) is assumed.
Years refer to crop years; for example, 1973 refers to the crop year 1972-73.
Alauddin and Tisdell (1986) review the growth decomposition literature and evaluate alternative methods.
For example, for a 2-period change, the weights are calculated as ii = (1/2)[QY'j/Q, + Q, ,jl/Q1,.
To see this, consider a 3-period case where growth rates between periods I and 2 and between 2 and 3 are given by
ln(Q2/Q,) and ln(Q3/Q2), respectively. The average growth rate is (1/2)[ln(Q3/Q2)+1n(Q2/Qd) or (l/2)1n(Q3/Q2), hence
not using information on Q2.
The augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) stationarity statistics were calculated for the
disturbance term of each trend regression. They are reported in appendix B.
Pray in examining the quality of the Bangladeshi agricultural statistics, argued that the statistics - especially in the
pre-independence period and a few years after independence - are not accurate. Given the desire to include as much
information as possible in the post-green revolution period but at the same time retain an acceptable quality of data, the
analysis starts in 1973, although complete rice data are available since 1969. In addition to the quality problems, there is
some evidence that official rice statistics may contain artificial noise, a result of attempts by the Ministry of Food to
attract foreign aid. Boyce (1987) and Alauddin and Tisdell (1987) discuss the data quality issue.
The GDP trend regression coefficient was 0.0425 (r-value = 32.35, R2 = 0.98, ADF =-3.99, PP = -4.01). The
income elasticity of rice using the 1988 Household Expenditure Survey (BBS, 1991) was estimated to be 0.40. Ahmed
and Goletti have used elasticities equal to 0.40 and 0.35. However, local researchers recommend that a lower figure
would be a more realistic and updated estimate.
10 It should be noted that prices are assumed to remain constant at the current level. With the current domestic price of
rice between import and export parity levels, projecting domestic price changes would, in effect, refect assumptions on
world price movements which, although important scenarios, are beyond the scope of the present study.
The assumption of constant per capita demand is equivalent to assuming that GDP and population growth rates are
identical, income elasticity of demand is zero, or a combination of the two factors.
12 Rice imbalance is calculated as q, - do for constant and q, - d, for variable demand. The percentage changes are
calculated as: PPC, = 100*(q, - do)ldo and SPC, = 100*(q, - d,)1d1, which are the figures reported in table 4.
For the slow (3%) and accelerated (5%) rate of MV adoption, the 65% ceiling is reached in 9 and 6 years, while the
80% ceiling is reached in 16 and 10 years.
14 At the suggestion of one reviewer, we considered the possibility of increasing cropping intensity, effectively
increasing total area. Simulation results indicate that holding MV yields at the current levels (scenario ii) and population
following the low growth path, the cropping intensity would have to increase to 2 to maintain a zero rice imbalance in the
year 2014 in the changing demand case (compared to the projected deficit of 23% if cropping intensity remained at the
current level of 1.82). However, as noted earlier, the potential for inceasing cropping intensity significantly beyond the
current levels is limited.
15 Ahmed presents a similar argument by attributing the recent state of rice self-sufficiency in Bangladesh in part to a
worsening income distribution (in addition to urbanization and cross-price effects).
15
Table 1. Rice Surmary Statistics: Production, Area, and Yields
Total Rice Modern Varieties Local Varieties
Output Area Yield Output Area Yield Output Area Yield
1973-75 Average
Annual 10,920 9,766 1.12 3,277 1,351 2.41 7,683 8,414 0.91
Aus 2,644 3,072 0.86 414 160 2.61 2,230 2,912 0.77
Aman 6,095 5,628 1.08 1,336 629 2.09 4,759 5,000 0.95
Boro 2,180 1,065 2.05 1,527 562 2.75 654 503 1.30
1992-94 Average
Annual 18,213 10,135 1.80 11,973 4,969 2.41 6,239 5,166 1.21
Aus 2,035 1,767 1.15 733 396 1.85 1,301 1,371 0.95
Aman 9,456 5,763 1.64 4,898 2,233 2.19 4,558 3,530 1.29
Boro 6,722 2,605 2.58 6,342 2,340 2.71 380 265 1.43
NOTES: Area and output are in thousands of hectares and thousands of metric tons; yields are in MTH. Averages may
not add up exactly because of rounding. Reported yields are period averages of production divided by area. SOURCE:
Calculated by the authors. Original data from Bangladesh Bureau of Statistics.
16
Table 2. Growth Rates
Total Rice Modern Varieties Local Varieties
Output Area Yield Output Area Yield Output Area Yield
1973 to 1994
Annual 2.65 0.20 2.44 7.80 7.83 -0.03 -1.28 -2.54 1.28
Aus -1.36 -2.66 1.33 1.97 3.97 -1.92 -2.28 -3.49 1.25
Aman 1.94 -0.04 1.98 8.58 8.55 0.03 -0.74 -2.00 1.29
Boro 7.07 5.55 1.43 9.06 8.95 0.10 -3.03 -3.43 0.42
1973 to 1987
Annual 2.60 0.47* 2.12 7.42 7.86 -0.41* -0.13* -1.41* 1.30
Aus 0.59* -0.72' 1.32 5.17* 8.06* -2.67* -0.42* -1.57' 1.17
Aman - 1.73 0.04 1.69 7.15 7.68 -0.49 0.28* -1.13* 1.43
Boro 6.57 4.86 1.63 8.86 8.86 -0.00 -2.75 -3.49 0.76
1987 to 1994
Annual 2.84 -0.81 3.68* 9.24 7.71 1.42* -5.50* -6.66* 1.24
Aus -8.40' -9.63* 1.36 -9.26* -10.13* 0.97* -9.00* -10.38* 1.54
Aman 2.72 -0.35 3.08 14.11 11.86 2.01 -4.51 -5.23* 0.75
Boro 8.98 8.22 0.71 9.82 9.29 0.48 4.05 -3.23 -0.86
NOTES: Growth rates have been estimated using regression (8) - tables B 1 and B2 (appendix B) report the full set of
results. Asterisks in the lower two panels denote rejection of the hypothesis of equality of the pre- and post-1987 period
growth rates (5% level). SOURCE: Calculated by the authors. Original data from Bangladesh Bureau of Statistics.
17
Table 3. Growth Decomposition
Total Rice Modern Varieties Local Varieties
Output Area Yield Output Area Yield Output Area Yield
1973 to 1994
Annual 100 8 92 126 126 0 -28 -56 28
Aus -10 -20 10 4 8 -4 -13 -19 7
Aman 40 -1 41 50 50 0 -11 -29 19
Boro 68 53 14 74 75 1 -4 -5 1
1973 to 1987
Annual 100 18 82 99 105 -5 -3 -35 33
Aus 5 -6 12 12 19 -6 -3 -10 8
Aman - 37 1 36 33 36 -2 5 -19 24
Boro 53 39 13 55 55 0 -5 -6 1
1987 to 1994
Annual 100 -29 130 180 151 28 -86 -105 19
Aus -46 -53 7 -16 -17 2 -34 40 6
Aman 50 -6 56 103 86 15 -49 -57 8
Boro 103 94 8 104 98 5 -4 -3 -1
NOTES: Figures represent the weighted contributions to overall growth, expressed as percentages. Bold-face figures
represent the 12 components of the overall growth rate - text equation (5). SOURCE: Calculated by the authors.
Original data from Bangladesh Bureau of Statistics.
18
Table 4. Projected Rice Surplus(+)/Deficit(-) as a Percentage of Demand
Percentage change in 2004 Percentage change in 2014
Population Population Population Population
Growth: 2.05% Growth: 1.49% Growth: 2.05% Growth: 1. 49%
65% 80% 65% 80% 65% 80% 65% 80%
Constant Demand (PPC) -Low MV Adoption Rate (3%)
i (-0.41%) -10 -9 -8 -7 -27 -23 -19 -15
ii (0.00%) -7 -6 -5 -4 -22 -17 -14 -9
iii (1.42%) 3 5 6 8 -3 6 7 17
iv (2.00%) 8 10 11 13 6 18 17 29
Constant Demand (PPC) -Accelerated MV Adoption Rate (5%)
i (-0.41%) -10 -4 -8 -1 -27 -23 -19 -15
ii (0.00%) -7 0 -5 3 -22 -17 -14 -9
iii (1.42%) 3 13 6 16 -3 6 7 17
iv (2.00%) 8 19 11 22 6 18 17 29
Changing Demand (SPC) -Low MV Adoption Rate (3%)
i (-0.41%) -16 -15 -15 -14 -36 -33 -32 -28
ii (0.00%) -14 -13 -12 -11 -32 -28 -27 -23
iii (1.42%) -3 -2 -2 0 -15 -7 -10 -1
iv (2.00%) 1 3 3 4 -7 3 -1 10
Changing Demand (SPC) - Accelerated MV Adoption Rate (5%)
i (-0.41 %) -16 -10 -15 -8 -36 -33 -32 -28
ii (0.00%) -14 -7 -12 -5 -32 -28 -27 -23
iii (1.42%) -3 6 -2 8 -15 -7 -10 -1
iv (2.00%) 1 11 3 13 -7 3 -1 10
NOTES: 65% and 80% refer to the MV ceiling. The figures in parentheses of the first column represent the four scenarios
for MV yields. SOURCE: Calculated by the authors. Original data from Bangladesh Bureau of Statistics.
19
FIGURE 1: Modern Variety Rice Yield FIGURE 2: Local Variety Rice Yield
0 0 0
2.40 0 1.30
2.20 0 ~~~~~ ,~~ A 0A 0 .00.0 -
< 2 .20 ___° _° 0 g-o- -
0 MV Rticc Yield
0 - LV Rice Yield
2.00 - - 0.70 _-_ _ _ __- __ __
1.80 i I I I - 1 I I I I 0.50 … -- - .i- -- i i i . - i - i i f . i -
1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993
FIGURE 3: Gross Area Allocated to Rice FIGURE 4: Per Capita Rice Production
12,000 .-__ ___ ___ ____ _.___ 200 ---- - - --- - ------ - --- ---
0 0 0° 0 0 0 0 0 0 0-0- 0 0O0 0 0 0 0 o
9,000 -- 180
O-- Per Capita Rice Production
6,000 . --- _-___ , - o- O
3 000 ___ _ __ _~ __ ._ __160 t-- ---
1973 1975 1977 1979 1981 1983 /985 1987 1989 1991 1993 1973 1975 1977 1979 198! 1983 1985 1987 1989 1991 1993
20
FIGURE 5: 2.05% Population Growth, FIGURE 6: 2.05% Population Growth,
3% MV Adoption, 65% MV Cciling 5% MV Adoption, 65% MV Ceiling
220 . 220 7
200 200
180 -i180
_ I --~~~~~~~~~~T T--- - -
t 160 FT+ :+ + +++ + + + + +- + + +-
/60~~~~~~~
140 OC6--~~~~~~~-.. -~~ - L0-
0 00 6 o00
'-~ 120 00 -6
1oo I t-- I I--t - tI I +t-+- ---tI ----_|100 -- t- I--- -+t- -F I ---+ - I -+_ - - -f-- , I _
1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 1994 1996 /998 2000 2002 2004 2006 2008 2010 2012 2014
FIGURE 7: 1.49% Population Growth, FIGURIE 8: 1.49% Population Growth,
3% MV Adoption, 65% MV Ceiling 5% MV Adoption, 65% MV Ceiling
220 - _.220 -. .
200 200
180 - - - -F - - - --+d 180 , -- - -c - C
160 - O -I-..O -1 - ---- - I -60 F -+ - t I-- F -+ - + + - + +
140- F- - + -- - -40 20 00 -0 L 2 20/
~.120
120
9199 16 1998 2000 2002 2004 2006 2008 2010 2012 2014
1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 19 96 19 00 20 04 20 08 21 02 21
-0- (i): MVyield growth = -0.41% -- (ii): MVyield growth = 0.00%
--+-- (iii): MVyield growth = 1.42% _- (iv): MVyield growvth 2.00%
Per capita rice de,,,and
21
FIGURE 9: 2.05% Population Growth, FIGURE 10: 2.05% Population Growth,
3% MV Adoption, 80% MV Ceiling 5% MV Adoption, 80% MV Ceiling
220 _ ___-___
220 ___ _ -- --
19 200 /2 20
3180 MV AM
~z) 60 r-0++~1
~140 14
2240 -00 _0 0 _- - E40 j _ _ 0 _ _
-- 1996199820002002 2004~~~~~~0 2
100 --e-- -----------1d ---f-t--f------- - --- 100 - dn- -- i- --d--F-I- f -- -f - f --I - f -
1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014
FIGURE 11: 1.49% Population Growth, FIGURE 12: 1.49% Population Growth,
3% MV Adoption, 80% MV Ceiling 5% MV Adoption, 80% MV Ceiling
220 ---_______________ ______________ 220 -- ------ --- - -_-- 2
200 ~200 ---
.~180 4 + 180 -
160 - 160
0-0-
140 '~~~~~~~~000--00 0- -*0-- 0 400 00
14 0- 0. 14
120 120
100 100~)~~ -+--ti f----- i-----I--- ----F *1~
1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014
-0- ('i): MVyield growvth = -0.41% ---- (iQ): MVyield growvth =0. 00%
+ (ii): AMVylicdgrowth =1.42%- (iv): A4Vyelddgrowtih = 2.00%
Per capita rice demand
22
References
Ahmed, R. "A Sluggish Demand could Be as Potent as Technological Progress in Creating Surplus in
Staple Production: The Case of Bangladesh." Mimeo, International Food Policy Research
Institute, 1994.
Alauddin, M. and C. Tisdell. "Trends and Projections for Bangladeshi Food Production: An
Alternative Viewpoint." Food Policy, 12(1987):318-33 1.
Alauddin, M. and C. Tisdell. "Decomposition Methods, Agricultural Productivity Growth and
Technological Change: A Critique Supported by Bangladeshi Data." Oxford Bulletin of
Economics and Statistics, 48(1986):353-372.
Anderson, J. R., P. G. Pardey, and J. Roseboom. "Sustaining Growth in Agriculture: A
Quantitative Review of Agricultural Research Investments." Agricultural Economics,
10(1994): 107-123.
Banerjee, A., J. Dolado, J. W. Galbraith, and D. F. Hendry. Co-Integration, Error-Correction,
and the Econometric Analysis of Non-Stationary Data. Great Britain: Oxford University
Press, 1994.
Bangladesh Bureau of Statistics. Report on the Household Expenditure Survey, 1988-89. Dhaka,
Bangladesh: Ministry of Planning, 1991.
Bangladesh Bureau of Statistics. Yearbook of Agricultural Statistics. Dhaka, Bangladesh: Ministry
of Planning, various issues.
Bos, E., M. T. Vu, E. Massiah, and R. A. Bulatao. World Population Projections: 1984/85
Edition, Johns Hopkins University Press, Baltimore, 1994.
Bouis, H. E. "Measuring the Sources of Growth in Rice Yields: Are Growth Rates Declining in
Asia?" Food Research Institute Studies, 22(1993):305-330.
Boyce, J. K. "Trends and Projections for Bangladeshi Food Production: Rejoinder to M. Alauddin
and C. Tisdell." Food Policy, 12(1987):332-336.
Boyce, J. K. "Kinked Exponential Models for Growth Rate Estimation." Oxford Bulletin of
Economics and Statistics, 48(1986):385-391.
Brown, L. R. and H. Kane. Full House: Reassessing the Earth's Population Carrying Capacity.
W.W. Norton and Company: New York, 1994.
Byerlee, D. "Modern Varieties, Productivity, and Sustainability: Recent Experience and Emerging
Challenges." World Development, 24(1996):697-718.
23
Byerlee, D. and A. Siddiq. "Has the Green Revolution Been Sustained? The Quantitative Impact of
Seed-Fertilizer Revolution in Pakistan Revisited." World Development, 22(1994):1345-1361.
Dickey, D. and W. A. Fuller. "Distribution of the Estimators for Time Series Regressions with Unit
Roots. " Journal of the American Statistical Association, 74(1979):427-43 1.
Faruqee, R. "Pakistan's Agriculture Sector: Is 3 to 4 Percent Annual Growth Sustainable?" Policy
Research Working Paper, No. 1407, The World Bank, 1995.
Gallant A. R. and D. W. Jorgenson. "Statistical Inference for a System of Simultaneous, Nonlinear,
Implicit Equations in the Context of Instrumental Variables Estimation." Journal of
Econometrics, 11(1979):275-302.
Goletti, F. The Changing Public Role in a Rice Economy Approaching Self-Sufficiency. The Case of
Bangladesh. International Food Policy Research Institute, Research Report No. 98, 1994.
Hossain, M. "Rice Research, Technoligical progress and Impact on Rural Economy: The Bangladesh
Case." Paper presented at the International Conference on the Impact of Rice Research,
Bangkok, Thailand, June 3-5, 1996.
McCalIa, A. "Agriculture and Food Needs to 2025: Why We Should Be Concerned." Sir John
Crawford Memorial Lecture, Consultive Group for International Agricultural Research.
Washington, D.C., October 27, 1994.
Mahmud, W., S. H. Rahman, and S. Zohir. Agricultural Growvth through Crop Diversification in
Bangladesh. International Food Policy Research Institute, Working Paper No. 7, 1994.
Mitchell, D. 0. and M. D. Ingco. The World Food Outlook. International Economics Department,
The World Bank, November 1993.
Pagiola, S. "Environmental and Natural Resource Degradation in Intensive Agriculture." Mimeo,
SAlAN, The World Bank, 1995.
Phillips, P. C. B. and P. Perron. "Testing for a Unit Root in Tirne Series Regression." Biometrica,
65(1988):335-346.
Pingali, P. L., P. F. Moya, and L. E. Velasco. "The Post-Green Revolution Blues in Asian Rice
Production: The Diminished Gap between Experiment Station and Farner Yields."
International Rice Research Institute, Social Science Division Papers, No. 90-01, Manila,
Philippines, 1990.
Pinstrup-Anderson P. and R. Pandya-Lorch. "Enough Food for Future Generations?" Choices,
(1994): 13-16.
Pray, C. E. "An Assessment of the Accuracy of the Official Agricultural Statistics of Bangladesh."
Bangladesh Development Studies, 7(1980): 1-38.
24
Rosegrant, M. W. and R. E. Evenson. "Agricultural Productivity and Sources of Growth in South
Asia. " American Journal of Agricultural Economics, 74(1992):757-761.
Rosegrant, M. W. and P. L. Pingali. "Policy and Technology for Rice Productivity Growth in
Asia." Journal of International Development, 6(1994):665-688.
Sharif, N. R. and A. A. Dar. "An Empirical Study of the Patterns and Sources of Technical
Inefficiency in Traditional and HYV Rice Cultivation in Bangladesh". Journal of
Development Studies, 32(1996):612-629.
25
APPENDIX A: KINKED GROWTH RATE REGRESSION
Consider regression (5),
ln(Q,) = + fit + s- (Al)
and let k be the year in which a structural change has taken place. Typically, to account for such
change (Al) is transformned as:
ln(Q,) = p,uD + P20 -D) + /3,Dt +,62(I-D)t + (A 2)
where D is a dummy variable taking the value of unity up to year k and zero elsewhere.
Since (A2) is equivalent to running two separate regressions, the trend lines may not
necessarily intersect at the break point k. To eliminate this discontinuity we follow Boyce (1986) by
imposing the following linear restriction:
i + 3, k = P2 + B2 k. (A3)
Restriction (A3) ensures that the trend lines intersect at k. Solving (A3) for p2, substituting the
resulting expression in (A2), and rearranging termns results in:
ln(Q,) = pi + /1,(Dt+(I-D)k) + 82(I-D)(t-k) + (A-4)
The hypothesis that /j =/,2 is then tested; rejection would indicate that a structural break did occur in
year k. As Boyce has argued, in the absence of special circumstances (A4) is preferable to (A2).
Further, (A4) has the advantage of ruling out the possibility that the growth rate derived from (Al)
falls outside the interval (fi, 8,2) as derived from (A2).
26
APPENDIX B: GROWTH RATE ESTIMATES
Table B1. Constant Growth Rate Model
Total Rice Modern Varieties Local Varieties
Output Area Yield Output Area Yield OutpUt Area Yield
Annual
IU 9.26 9.21 0.05 7.79 6.95 0.85 9.12 9.20 -0.08
(526.1) (905.9) (3.27) (132.1) (126.5) (54.08) (205.9) (266.0) (-3.83)
a6 0.0262 0.0020 0.0241 0.0751 0.0754 -0.0003 -0.0129 -0.0257 0.0128
(19.52) (2.63) (19.62) (16.71) (18.04) (-0.24) (-3.83) (-9.76) (8.50)
R2 0.95 0.26 0.95 0.93 0.94 0.01 0.42 0.83 0.78
ADF -7.25 -2.54 -4.25 -5.86 -4.35 -3.41 -2.34 -1.60 -4.09
PP -7.75 -2.74 -4.59 -2.85 -2.57 -3.80 -2.47 -1.55 -4.34
Aus
IU 8.09 8.23 -0.14 6.40 5.45 0.95 7.86 8.17 -0.31
(128.5) (148.1) (-6.44) (36.86) (29.70) (31.47) (136.9) (150.5) (-14.17)
a3 -0.0137 -0.0269 0.0132 0.0195 0.0389 -0.0194 -0.0231 -0.0355 0.0123
(-2.86) (-6.37) (7.76) (1.48) (2.78) (-8.49) (-5.29) (-8.59) (7.39)
R2 0.29 0.67 0.75 0.10 0.28 0.78 0.58 0.79 0.73
ADF -2.44 -1.05 -5.24 -5.16 -4.53 -3.01 -1.42 -1.32 -4.05
pP -2.72 -1.68 -5.44 -4.28 -3.83 -3.23 -1.78 -1.33 -4.27
Aman
8.73 8.67 0.06 6.66 5.92 0.73 8.67 8.67 0.01
(267.9) (489.6) (2.51) (49.80) (41.19) (18.64) (159.8) (252.8) (0.15)
/3 0.0192 -0.0004 0.0196 0.0823 0.0820 0.0003 -0.0075 -0.0202 0.0127
(7.75) (-0.29) (10.71) (8.09) (7.49) (0.10) (-1.81) (-7.75) (6.03)
R2 0.75 0.01 0.85 0.77 0.74 0.01 0.14 0.75 0.65
ADF -4.60 -2.98 -4.65 -3.99 -3.94 -4.22 -2.74 -2.33 -3.51
PP -4.92 -3.08 -4.84 -2.71 -2.52 -4.48 -2.70 -1.89 -3.72
Boro
7.37 6.71 0.66 6.95 5.99 0.96 6.58 6.30 0.28
(119.2) (124.9) (25.01) (100.3) (99.35) (37.35) (123.1) (178.1) (7.20)
a3 0.0683 0.0540 0.0142 0.0867 0.0858 0.0010 -0.0307 -0.0349 0.0042
(14.52) (13.22) (7.09) (16.43) (18.67) (0.50) (-7.55) (-12.97) (1.42)
R2 0.91 0.90 0.72 0.93 0.95 0.01 0.74 0.89 0.09
ADF -3.02 -2.54 -2.74 -3.76 -3.47 -3.91 -5.37 -5.53 -3.31
PP -2.96 -2.51 -2.94 -2.40 -2.23 -4.04 -5.54 -5.92 -3.54
NOTES: The numbers in parentheses are t-values. ADF and PP refer to the augmented Dickey-Fuller and Phillips-Perron
statistics. SOURCE: Authors' estimates. Original data from Bangladesh Bureau of Statistics.
27
Table B2. Kinked Growth Rate Model
Total Rice Modern Varieties Local Varieties
Output Area Yield Output Area Yield OutpUt Area Yield
Annual
A 9.26 9.19 0.08 7.82 6.94 0.88 9.04 9.12 -0.08
(444.6) (1079) (4.59) (113.0) (106.6) (61.48) (239.8) (475.6) (3.26)
fil 0.0257 0.0047 0.0209 0.0716 0.0757 -0.0041 -0.0013 -0.0142 0.0128
(12.59) (5.68) (13.13) (10.58) (11.87) (-2.96) (-0.36) (-7.59) (5.63)
132 0.0280 -0.0082 0.0361 0.0884 0.0743 0.0141 -0.0560 -0.0689 0.0123
(4.78) (-3.41) (7.89) (4.55) (4.06) (3.55) (-5.34) (-12.29) (1.87)
R2 0.95 0.63 0.96 0.94 0.94 0.42 0.70 0.96 0.78
Fl, 19 0.10 19.19 7.28 0.50 0.01 13.85 17.96 68.05 0.01
ADF -7.60 -4.28 -7.55 -5.49 -4.38 -6.47 -3.28 -2.71 -4.08
PP -8.71 -4.53 -7.87 -2.92 -2.57 -6.49 -3.42 -2.32 4.33
Aus
7.89 8.09 -0.14 6.18 5.19 1.00 7.72 8.03 -0.31
(192.1) (376.9) (-5.41) (34.11) (28.74) (38.39) (237.0) (478.2) (-11.79)
fli 0.0059 -0.0072 0.0132 0.0504 0.0775 -0.0271 -0.0042 -0.0159 0.0116
(1.46) (-3.45) (5.06) (2.84) (4.39) (-10.65) (-1.33) (-9.65) (4.75)
/62 -0.0878 -0.1013 0.0135 -0.0972 -0.1068 0.0096 -0.0943 -0.1096 0.0153
(-7.55) (-16.78) (1.81) (-1.91) (-2.11) (1.32) (-10.30) (-23.21) (2.09)
R2- 0.78 0.97 0.75 0.30 0.51 0.88 0.91 0.99 0.73
F-,J19 42.87 160.45 0.01 5.55 8.73 16.68 63.92 260.65 0.17
ADF -5.05 -3.92 -5.26 -4.78 -4.37 -5.97 -4.92 -4.68 -4.12
PP -5.13 -4.12 -5.46 -4.18 -3.89 -7.68 -5.24 -5.56 -4.33
Aman
p 8.74 8.66 0.08 6.75 5.98 0.77 8.60 8.61 -0.01
(229.2) (415.4) (2.96) (44.09) (35.42) (17.62) (154.6) (291.9) (-0.17)
fli 0.0172 0.0004 0.0168 0.0691 0.0740 -0.0049 0.0028 -0.0114 0.0142
(4.61) (0.21) (6.32) (4.61) (4.48) (-1.15) (0.52) (-3.95) (4.43)
/62 0.0268 -0.0035 0.0303 0.1320 0.1121 0.0199 -0.0462 -0.0537 0.0075
(2.50) (-0.59) (3.98) (3.07) (2.36) (1.63) (-2.95) (-6.48) (0.81)
R2 0.76 0.02 0.87 0.78 0.74 0.13 0.36 0.87 0.65
Fl 19 0.54 0.30 2.10 1.41 0.43 2.71 6.49 17.21 0.35
ADF -4.88 -3.01 -5.75 -3.74 -3.75 -5.42 -2.97 -2.90 -3.46
PP -5.34 -3.10 -6.34 -2.80 -2.49 -5.51 -2.91 -2.23 -3.64
28
Boro
7.40 6.75 0,65 6.96 6.00 0.97 6.56 6.31 0.26
(103.1) (112.1) (21.04) (84.86) (83.84) (31.88) (104.4) (150.3) (5.71)
a,1 0.0636 0.0474 0.0162 0.0849 0.0849 -0.0001 -0.0279 -0.0355 0.0076
(9.05) (8.05) (5.37) (10.57) (12.13) (-0.01) (-4.54) (-8.65) (1.73)
/32 0.8600 0.0790 0.0071 0.0936 0.0889 0.0048 -0.0414 -0.0328 -0.0086
(4.26) (4.67) (0.82) (4.06) (4.42) (0.56) (-2.34) (-2.78) (-0.68)
R2 0.92 0.91 0.73 0.93 0.95 0.02 0.75 0.89 0.14
Fl, 19 0.82 2.29 0.73 0.10 0.03 0.21 0.38 0.04 1.10
ADF -2.88 -2.48 -2.92 -3.66 -3.43 -3.80 -5.55 -5.53 -3.48
PP -2.89 -2.48 -3.08 -2.37 -2.22 -3.95 -5.73 -5.95 -3.67
NOTES: Fj, 19 is the F-statistic ofthe null hypothesis that the pre- and post- 1986/87 gro-th rates are equal. Values
higher than 4.34 and 8.18 imply rejection of the equality hypothesis at the 5 and 1% level of significance, respectively. For
other notes see Table B I. SOURCE: Authors' estimates. Original data from Bangladesh Bureau of Statistics.
29
Policy Research Working Paper Series
Contact
Title Author Date for paper
WPS1651 Bank-Led Restructuring in Poland: Cheryl W. Gray September 1996 B. Moore
Bankruptcy and Its Alternatives Arnold Holle 38526
WPS1652 Intra-Industry Trade, Foreign Direct Bernard Hoekman September 1996 F. Hatab
Investment, and the Reorientation Simeon Djankov 35853
of Eastern European Exports
WPS1653 Grants and Debt Forgiveness in Leonardo Hernandez September 1996 R. Vo
Africa: A Descriptive Analysis 31047
WPS1654 Indonesia's Palm Oil Subsector Donald F. Larson September 1996 P. Kokila
33716
WPS1655 Uncertainty and the Price for Crude Timothy J. Considine September 1996 P. Kokila
Oil Reserves Donald F. Larson 33716
WPS1656 The Investment Decision: A Cherian Samuel September 1996 C. Samuel
Re-Examination of Competing Theories 30802
Using Panel Data
WPS1657 Is There an Optimal Structure for Frannie Humplick September 1996 J. Williams
Decentralized Provision of Roads? Azadeh Moini-Araghi 82557
WPS1658 Decentralizing Structures for Frannie Humplick September 1996 J. WVilliams
Providing Roads: A Cross-Country Azadeh Moini-Araghi 82557
Comparison
WPS1659 Unemployment Insurance in Algeria: Elizabeth Ruppert September 1996 H. Osselyn
Implications for a Labor Market in 36039
Transition
WPS1660 Mind Your P's and Q's: The Cost of Lant Pritchett October 1996 S Fallon
Public Investment is Not the 38009
Value of Public Capital
WPS1661 Determinants of Public Expenditure Susan Randolph October 1996 Z Bogetic
on Infrastructure: Transportation and Zeljko Bogetic 623-7292
Communication Dennis Heffley
WPS1662 From Learning to Partnership: Giorgio Barba Navaretti October 1996 M. Patena
Multinational Research and Carlo Carraro 39515
Development Cooperation in
Developing Countries
WPS1663 Internal Finance and Investment: Cherian Samuel October 1996 C. Samuel
Another Look 30802
WPS1664 Pensions in Germany Monika Queisser October 1996 H. Arbi
34663
Policy Research Working Paper Series
Contact
Title Author Date for paper
WPS1665 How Important Are Labor Markets Andrew D. Mason October 1996 D. Ballantyne
to the Welfare of Indonesia's Poor? Jacqueline Baptist 87198
WPS1666 Is Growth in Bangladesh's Rice John Baffes October 1996 P. Kokila
Production Sustainable? Madhur Gautam 33716