WPS7337


Policy Research Working Paper                            7337




     The Consumption, Income, and Wealth
               of the Poorest
Cross-Sectional Facts of Rural and Urban Sub-Saharan Africa
                    for Macroeconomists

                               Leandro De Magalhaes
                              Raul Santaeulalia-Llopis




Development Research Group
Poverty and Inequality Team
June 2015
Policy Research Working Paper 7337


  Abstract
 This paper provides new empirical insights on the joint                            to wealth, which suggests that SSA households face a larger
 distribution of consumption, income, and wealth in three                           inability to save and accumulate wealth compared with US
 of the poorest countries in the world — Malawi, Tanzania,                          households; and (ii) a lower transmission from income to
 and Uganda — all located in Sub-Saharan Africa (SSA).                              consumption inequality, which suggests the presence of
 The first finding is that while income inequality is similar                       powerful institutions that favor consumption insurance to
 to that of the United States (US), wealth inequality is barely                     the detriment of saving. These features are more relevant
 one-third that of the US. Similarly, while the top of the                          for rural areas, which represent roughly four-fifths of the
 income distribution (1 and 10 percent) earns a similar share                       total population. The paper identifies the few successful
 of total income in SSA as in the US, the share of total wealth                     pockets of the SSA population that are able to accumulate
 accumulated by the income-rich in SSA is one-fifth of its                          wealth by exploring sources of inequality such as age, edu-
 US counterpart. The main contributions of the paper are                            cation, migration, borrowing ability, and societal systems.
 to document: (i) this dwarfed transmission from income




  This paper is a product of the Poverty and Inequality Team, Development Research Group. It is part of a larger effort by
  the World Bank to provide open access to its research and make a contribution to development policy discussions around
  the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be
  contacted at rauls@wustl.edu and Leandro.DeMagalhaes@bristol.ac.uk.




         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 cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
         of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
         its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


                                                       Produced by the Research Support Team
  The Consumption, Income, and Wealth of the Poorest:
  Cross-Sectional Facts of Rural and Urban Sub-Saharan
               Africa for Macroeconomists∗

            Leandro De Magalh˜  aes                    Ra¨ul Santaeul`alia-Llopis
             University of Bristol                 Washington University in St. Louis
                                                        Universitat de Val`encia




       JEL: E20; E21; O10; O55; I32

       Keywords: Inequality, Consumption, Income, and Wealth, Sub-Saharan Africa




   ∗
     We thank Kathleen Beegle, Taryn Dinkelman, Pascaline Dupas, Francisco Ferreira, John Kaboski, Talip Kilic,
and John Temple for comments and suggestions at different stages of this project. We also thank the seminar
participants at the World Bank LSMS group and the SED in Seoul 2013.
1       Introduction

Cross-sectional facts on the distribution of consumption, income, and wealth (CIW) are readily
available for a large set of modern industrialized economies, see the special issue of the Review of
Economic Dynamics on “Cross Sectional Facts for Macroeconomists” (Krueger et al. (2010)) and
the more recent studies of D´  ıaz-Gim´enez et al. (2011) and Piketty (2014). Such distributional
facts have been extensively used to build and test macroeconomic theories that incorporate
heterogeneous household saving behavior for the study of, almost invariably, rich economies
(Heathcote et al. (2009)). For these macroeconomic frameworks to be useful in poor countries
they need to be fully contextualized and informed by the behavior of households in these countries.
Hence, a good understanding of household-level CIW inequality for poor countries is required and
to date has been missing. The main contribution of this paper is to help close this gap by providing
a careful and comprehensive dissection of CIW behavior in three of the poorest countries in the
world in 2010 — Malawi, Tanzania and Uganda — using new and unique nationally representative
data. To construct the whole distribution of CIW a series of technical issues had to be addressed:
converting consumption from own production into kilograms, deseasonalizing consumption, and
assigning a monetary value to the agricultural production that is not sold on the market.

     To gain a perspective on the relative poverty of the countries investigated note that in 2010,
income per capita was US$359 in Malawi, US$524 in Tanzania, and US$471 in Uganda. That
is, our households live with country averages below (or close to) US$1 per capita per day, and
less so in rural areas, where the overwhelming majority of the population lives: 84% in Malawi,
71% in Tanzania, and 85% in Uganda.1 For comparison purposes, in Mexico, the poorest country
studied in Krueger et al. (2010), the income per capita in 2010 was US$8,920 and the population
living in rural areas was 22%. In Thailand, a country extensively investigated in the development
literature, these figures were US$4,802 and 56% respectively.

   Our main finding is a large and widespread inability to accumulate wealth in both rural and,
to a lesser extent, urban areas of Sub-Saharan Africa (SSA). First, while income dispersion in
SSA is of similar magnitude to that of the US and Europe, wealth inequality in SSA is close to
one-third that of the US.2 Further, while the transmission from income to wealth inequality is
not apparent from the separate inspection of their respective distributions, our study of the joint
behavior of income and wealth suggests a substantially lower transmission in SSA than in the
    1
     Indeed, Malawi is currently the poorest country in the world in terms of income per capita according to the
2014 World Development Indicators (World Bank). Malawi has been disputing this unfortunate position over the
past 10 years with the Democratic Republic of Congo and Niger, also sub-Saharan African countries.
   2
     For these cross-sectional comparisons between SSA and the US we largely draw from the US statistics reported
in D´ıaz-Gim´enez et al. (2011) and Piketty (2014).



                                                       2
US. This phenomenon is clearly starker in the rural areas than in the urban areas of SSA: we
find that the top 1% income-richest households in rural and urban SSA, respectively, hold 4% of
total wealth in rural areas and 11% in urban areas, while this figure is 26% in the United States.3
That is, the income-richest households in rural SSA accumulate only as much as 15% of the
share of total wealth that the income-richest US households are able to accumulate in terms of
US wealth. The equivalent figure for urban areas in SSA is 46%, which also suggests a reduced
ability to accumulate wealth in SSA than in the US. The behavior of the joint density of income
and wealth strengthens this message as we find significant positive correlations between income
and wealth only for the top 20% of the income distribution. The correlation is stronger in urban
areas, 0.30, than in rural areas, 0.12, while this figure is much larger, 0.57, for the entire US
economy.

    The direct exploration of household saving provides further insights. We find that only house-
holds in the top 1% and 5% of the income distribution in rural and urban areas are able to save,
and they do so with high saving rates similar to those of the top income earners in the US. One
way to relate this finding (i.e., saving rates similar to those in the US for the top income earners
in SSA) with the result discussed in the previous paragraph (i.e., lower share of total wealth
held by the top income earners in SSA than in the US) is through the lesser persistence of high
income in SSA than in the US. Indeed, we find that within a span of four years, 51% of rural
households at the top quintile of the income distribution fall to a lower quintile, as do 46% of
urban households, while this figure is only 23% in the US.4 That is, households in the top quintile
of the income distribution in SSA leave that quintile twice as fast as their counterparts in the
US, and this downward mobility occurs somewhat faster in rural than urban areas. This lesser
persistence of high income — and hence, of saving — helps to explain the low transmission from
income to wealth and the low wealth accumulation in SSA compared with that in the US, even
for the top income earners who actually save.

    Interestingly, despite the generalized inability to save that we find in SSA, we find a concurrent
lower transmission from income to consumption inequality in SSA than in the United States.
Precisely, economy-wide, the inequality of consumption is 47% that of income (in terms of
the variance in logs). The figure for the United States is larger, roughly 55% (Blundell et al.
(2008)). Furthermore, the joint density of consumption and income implies a correlation between
consumption and income of 0.30 in rural areas and 0.53 in urban areas, suggesting that this
insurability of consumption is larger in rural areas than in urban areas. Combining our two
findings — lower transmission from income inequality to both wealth inequality and consumption
  3
      For brevity, the precise numbers reported in the introduction refer to Malawi.
  4
      We use data from Uganda for these mobility statistics, as it has a panel component not available for Malawi.



                                                         3
inequality in SSA than in the US — suggests that the greater ability to contain the dispersion
in consumption in SSA compared with the US must occur through insurance mechanisms other
than self-insurance (i.e., own savings).

    The higher ability to insure in rural areas is further revealed from direct evidence on self-
reported information about shocks, coping strategies, and ability to borrow. Among rural house-
holds 71% report being hit by a shock in the past 12 months (most commonly weather-related
shocks), and among urban households 39% report a shock (most commonly high food prices).
Despite the higher occurrence of shocks in rural areas, of the rural households that report a
shock, 51% have some form of insurance to cope with the shock, while this figure is smaller in
urban areas, 40%. The ability to insure consumption can also depend on the ability to borrow.
In this case, conditional on needing a loan, urban households show higher application rates, 40%,
than rural households, 27%.5 While this urban-rural differential in the ability to borrow suggests
urban households might be better able to self-insure, we note that in urban areas borrowing is
used 3.6 times more often for start-up capital than for consumption, while this ratio is 1.6 for
rural areas. This suggests that such borrowing is used mostly for production and not necessarily
for consumption insurance.

    In decomposing CIW we find unambiguous signs of a subsistence economy where food con-
sumption requirements can severely limit the ability to save. Specifically, the share of food in
total consumption is above 50% throughout the income distribution, except for the top 1% in
rural areas and the top 5% in urban areas. Coincidentally, these two groups are the only ones
with positive saving rates across the income distribution. The composition of income (mostly
agricultural) in rural areas also points towards a subsistence economy. Only in the top 1% of
the income distribution in rural areas does the share of agricultural income fall below 50%. The
similarity in the wealth portfolio across income groups (except for the top 1%) further reflects an
inability to accumulate wealth. The only components of wealth that show signs of accumulation
are livestock and non-housing durables.

    While the overall power to accumulate wealth is dismal in the SSA countries we consider,
the examination of cross-sectional sources of inequality allows us to identify pockets of wealth
accumulation. First, over the life cycle, wealth in the rural areas of SSA (i.e., mostly land) follows
a hump pattern that increases, in Malawi, by a factor of 2.65, while this increase is by a factor
of 7.13 in urban areas, further suggesting a greater ability to accumulate wealth in urban areas
than in rural areas. However, again, even the urban increase is small compared to the life cycle
behavior of wealth in the United States that increases by a factor of roughly 20. Second, we find
   5
    As we discuss below, these figures deal directly with self-selection as households are asked whether they
needed a loan and whether they applied or not, independently of the need.


                                                     4
that while education tends to allow for more income and consumption, it does not necessarily
imply higher wealth accumulation, particularly in rural areas. In contrast, in urban areas, the
more-educated households (i.e., “secondary education or more”) are overrepresented in the top
1% of the wealth distribution. Indeed, only the more-educated urban households have positive
savings on average. Third, CIW averages in rural areas and urban areas are very similar across
countries. The large discrepancy is between CIW averages between rural and urban areas. Indeed,
the difference in average CIW between rural and urban areas is an order of magnitude larger than
the difference in average CIW across countries. Such large differences should trigger important
rural-to-urban migration flows, which we do observe as 60% of current urban household heads
report having migrated from rural areas. We find that while rural-to-urban migration improves
their consumption and income, we also find that rural-to-urban migrants are not better able to
accumulate wealth than the local households in the hosting urban region. The only migration
group with a positive saving rate is the one formed by urban-to-urban migrants, who represent
18% of the urban population. Finally, we find that the ability to accumulate wealth can also be
related to gender and to the particular societal systems a household belongs to (i.e. matrilineal
or patrilineal). In particular, for Malawi, the only group with average positive savings is the the
one formed by matrilineal households in urban areas in the center of the country.

    Our paper relates to a vast literature in development economics. First, our results on the
inability to save and accumulate wealth in Malawi, Tanzania, and Uganda are directly related
to the experimental results in Dupas and Robinson (2013a,b) for Kenya on savings constraints.6
Similar to Rosenzweig and Wolpin (1993) findings for India, we find that the households in rural
areas of East Africa that save do so in livestock and other nonhousing durables. This may be due
to a lack of alternative saving options. Second, in terms of consumption insurance, Townsend
(1994) and Udry (1994) suggest the presence, respectively, of informal arrangements in villages
in India and northern Nigeria. These results confirm our findings of a weaker transmission from
consumption to income inequality in SSA than in the United States, and in rural rather than
urban areas. The phenomenon that the ability to insure consumption is stronger in rural areas
than in urban areas has been previously explored in Morten (2013) and Munshi and Rosenzweig
(2014) for India, and in Santaeul` alia-Llopis and Zheng (2015) for China. Two aspects set our
work apart from previous studies. The first is the use of nationally representative data, which is
a natural approach from our macroeconomic perspective. The second, and most important, is
that we can establish the joint behavior of the triplet of CIW and document — as far as we know
for the first time — the joint phenomenon of a low ability to accumulate wealth (i.e., generate
growth in the economy) and a low transmission from income to consumption (i.e., a high ability
  6
      See also the recent review in Karlan et al. (2014).



                                                            5
to insure consumption).

   The rest of the paper is sectioned as follows. In Section 2, we provide a full account of
the data construction. In Section 3, we discuss in detail the distributional facts of CIW for
Malawi, Tanzania, and Uganda, with more emphasis on Malawi. The emphasis on Malawi is
due to the sample size being four times larger and the data more detailed than for the other
two countries. The corresponding tables for Tanzania and Uganda are available in the online
appendix. In Section 4, we decompose the consumption basket, the income sources, and the
wealth portfolio in rural and urban areas. In Section 5, we investigate the sources of inequality
and identify pockets of the population with positive savings. In Section 6, we explore economic
mobility using panel information for Uganda in 2005 and 2009. We conclude in Section 7.


2       ISA Data and Measurement Issues

The Integrated Surveys on Agriculture (ISA) are conducted under the umbrella of the Living
Standards Measurement Study (LSMS). ISAs greatly improve previous LSMS household surveys
in use for decades. In particular, ISAs are unique in the level of detail in CIW for households that
make their living through agriculture. The ISAs allow us to recover — for any practical purpose
—the entire deseasonalized household-level budget constraint in a set of the poorest countries
of SSA.7 However, in addition to the ISA improvements in data collection and availability, a
correct measurement of CIW requires further adjustments. In particular, due to the importance
of agricultural production, estimating the monetary value of income and consumption requires us
to convert physical units in which production and consumption are reported into a single unit,
assign prices to items that are produced by the household for own consumption, and deseasonalize
consumption. As we show, the choice of which price to use to value non-sold production — either
the price at the gate (i.e., production price) or consumption price — can have large impacts on
the estimated monetary values of agricultural production.

2.1     Units Conversion of In-Kind Items: From Pails to Kilograms

In household surveys from poor countries, it is standard to report amounts of consumption (in-
cluding from own production), production (mainly agricultural) and additional sources of income
(e.g., gifts and transfers) in units that are not necessarily harmonized across time or space. For
example, in the Malawi ISA, households are asked to report the amount they produce of a given
item in any unit they wish, and this varies from bags, dishes, bunches, and pails, to kilograms
    7
    To the best of our knowledge, the availability of the CIW triplet is rather unique as these three items are not
supplied at the same time by any other single dataset, not even in developed countries. See also the discussion
in Heathcote et al. (2009).


                                                        6
(kg). It is then necessary to deal with the measurement issue of converting all these reported
units into a single unit, say kilograms.8

    Here, we use a simple price-unit conversion method separately for production and consump-
tion. In terms of production, this method uses the information on prices from households that
produce a given item and at the same time sell the same item in different units.9 We use the
median unit price for a given sold item in a given region, residential area, and season to gener-
ate household-specific conversion rates. We merge all household-specific conversion rates from
all regions, residential areas, and seasons for 2004-05 and 2010-11. Then, we pick the median
conversion rate (if there are at least 7 household specific conversion rates) for each item-unit
pair (i.e., conversion rates are item-specific). With the resulting conversion rates, items are first
converted into the modal unit, and then into kilograms. In terms of consumption, we proceed
similarly, using the information on prices from households that consume a given item and also
report bought quantities of the same item in different units. Finally, we need to attribute prices
to these quantities. To measure both household production and consumption (that includes own
produced and gifts that are collected in in-kind units) we use the median consumption prices in a
given season-region separately for rural and urban areas, a matter that is discussed in more detail
in the next section.

    While agricultural production is reported in annual terms through the recollection of the
harvest output from the last dry and rainy seasons separately,10 food consumption is reported
with a weekly recall (i.e., seven days before the interview). This introduces an artificial seasonal
component to the measurement of consumption that needs to be netted out. Indeed, Malawi,
Tanzania, and Uganda have clearly demarcated lean and plenty seasons that largely determine
food consumption; that is, seasonality is crucial to understanding consumption behavior. One
of the improvements of the ISAs is that data collection is rolled out throughout the entire year.
This means that we can remove month effects with two year-month observations. Here, we use
   8
      For Tanzania and Uganda unit-conversion rates are not necessary as in Tanzania all quantities are reported
in kilograms and in Uganda households provide unit-conversion rates (we use the median for a given region and
season). This does not mean that households report in kilograms, but that under several procedures, interviewers
transform all units into kilograms.
    9
      For more details, see De Magalh˜ aes et al. (2015). There we show that this price-unit conversion method is
able to recover more item-unit pairs and has similar precision as the standard market surveys conducted on the
field that weighs produce in local markets to obtain physical-unit conversion. This is important because market
surveys are costly in monetary terms and are usually not rolled out throughout the year (e.g., the market survey
for Malawi in 2014). However, the conversion units might differ across months during the year, as the same crop
may have different kilogram weights in two different harvests. All prices are deflated using consumer price indexes
(CPI) for food and non food items separately for urban and rural areas — CPI available at the National Statistics
Office of Malawi.
   10
      This information is stored in separate submodules of the agriculture module. A submodule for a permanent
crop season is also provided.



                                                       7
the deseasonalized consumption measures computed in such manner from De Magalh˜  aes et al.
(2015), who find that the amount of consumption in the pre-harvest month of March is 0.4 log
points below the annual average in Malawi.11

2.2     Unsold Agricultural Production and Extended Household Income

The extraordinarily detailed information on agricultural production is one of the most novel as-
pects of the ISAs. The new agricultural modules substantially extend previous LSMS surveys.
The correct account of agricultural production (and income) has proven difficult in the past, as in
such settings most rural households represent small-scale farms that sell only a small fraction of
what they produce in the open market. ISAs provide detailed collection of agricultural production
(separately for each main harvest season, including permanent crops) and their costs (including
rentals of capital equipment and structures), fishery production and their costs, as well as business
income, wages, and informal sources of revenue (e.g., “ganyu” in Malawi). Agricultural informa-
tion is available mostly at the plot level (household may farm non-contiguous plots), and fishery
and business income are available at the household level. Information on wages for each type of
formal and informal activity are available for each and all members in the household (not only
husbands and wives). This set of information allows us to construct precise extended-household
income measures.

2.2.1    Unsold Agricultural Production

In very poor agricultural economies such as the ones we study, assigning a monetary value to
unsold agricultural production is essential to the measurement of household income. The reason
is that unsold production represents the majority of total household production. We use maize
in Malawi to illustrate this issue. First, we convert maize production into the same units to find
that maize represents 69% of the total agricultural production in kilograms. Most households
produce maize as their main source of food and calorie intake,12 but few sell it. Of the 9,280
households in the Malawi survey who report producing maize, only 1,618 (17%) report selling
any maize. Among the top 20% of the income distribution in rural areas, 30% of households
report selling maize; among the bottom 20%, only 6% sell maize. It is noteworthy that even the
rich keep their own production for consumption. Moreover, sales among the poor may indicate
  11
     There, we further show that consumption measures that are not deseasonalized tend to underestimate in-
equality. We use the Integrated Household Survey of 2004-05 (also rolled over the entire year) as the second
year-month observation to deseasonalize consumption for Malawi. For Tanzania, we use the ISA 2008 and for
Uganda, the ISA 2005.
  12
     See the Malawi Bioenergy and Food Security country brief by the United Nations Food and Agriculture
Organization.



                                                     8
desperation rather than a good business strategy.13

    Second, we need to assign prices to unsold production. The price at the gate is normally
used for this purpose (Deaton and Zaidi (2002)). For valuing maize, this implies using the price
of shelled maize reported by households mostly in the immediate post-harvest season. There are
two reasons this price underestimates the value of unsold production. The shelled maize is a
different good than what farmers produce, as farmers use the cobs, husks, and stems for other
intermediate purposes such as fuel and animal feed, and such uses have an important monetary
value in subsistence households. Further, maize prices are lowest immediately after harvest (see
Kaminski et al. (2014)), whereas consumption takes place throughout the year. While the storage
possibilities are limited,14 almost all households report storing some maize for their own use later.
That is, maize has an additional value when consumed when the consumption prices are high,
and this option value is not captured by the post-harvest price at the gate.

     The underestimation of the value of production by the use of the price at the gate can be
illustrated by focusing on rural households that sell neither maize nor tobacco, tobacco being the
main cash crop.15 These are the households closest to an autarkic model, in which the maize
they consume is what they have produced. In this sample, the mean estimated quantity of maize
produced is 124 kg and the mean estimated quantity of maize consumed is 130 kg.16 The small
6-kg difference between production and consumption represents their total purchases (or gifts).
This comparison is just an illustration as the maize produced is reported in shelled maize and the
maize consumed is mostly reported as flour or green maize (i.e., maize on the cob). Nevertheless,
since the quantities of production and consumption are similar, it becomes clear that assigning
prices at the gate to production and consumption prices to consumption can create an artificial
wedge between their monetary values if these two prices differ. Indeed, the price of shelled maize
at the gate for these households is approximately US$0.16 per kg, and the consumption price of
green maize on the cob is approximately US$0.36 per kg.17 If we use the price at the gate, the
average estimated production value is US$22, which is less than half of the average consumption
value, US$47. That is, the price at the gate underestimates the value of unsold production.

       In Figure 1 we compare the distribution of the monetary value in dollars of household total
  13
     See Manda (2010).
  14
     Only 9% of households have a dedicated storage structure; most crops are stored in the house or in open
drums or sacks.
  15
     The sample includes 4,384 households that sell neither maize nor tobacco and report producing a positive
value of maize and consuming a positive value of home-produced maize. These households represent 36% of the
entire sample and 45% of all rural households.
  16
     For the entire sample of rural households, these numbers are, respectively, 129 kg and 137 kg.
  17
     We use the following exchange rates for 1US$ in March 2010: Malawi, 152; Tanzania, 1,350; and Uganda,
2,110.


                                                     9
food consumption in logs (dotted line) with the value of household agricultural production valued
with consumption prices (dashed line) and valued with the price at the gate (solid line). Since we
are focusing on households that sell neither tobacco nor maize and these are rural households close
to subsistence, one would expect the distribution of agricultural production and food to overlap.
This is actually the case in raw quantities and, hence, also the case if we value home production
with consumption prices. However, if we value production with prices at the gate, the total value
of food production is estimated to be considerably lower than food consumption.18 Even if we
value production with consumption prices, the distribution of the value of food production is
slightly shifted left in relation to consumption. This is to be expected as these households may
have other sources of income, albeit small, such as informal labor and business. In light of our
results, it is our view that the shadow price of unsold agricultural production is best captured by
consumption prices.

    Finally, the remaining issue is how to value the part of agricultural production that is actually
sold on the open market. We have chosen to use the price at the gate to value sold production
as sold items lose their storage value for the producer household, but this is of relatively small
consequence given the low share of sold production in the sample. The estimated average per
capita value of maize production for the sample of all rural households is US$101 under our
preferred measure — that is, if we use consumption prices for the unsold production and prices
at the gate for the sold production. This figure is US$109 if we use the consumption price for all
production, sold and unsold.

2.2.2    Extended Household Income

The main resource in rural areas is agricultural production. Production is reported by crops per
plot and by season (rainy season, dry season, and permanent crop). After converting unsold
production into the same physical units and assigning prices to it as described in the previous
section, we can measure the entire agricultural production, sold and unsold, in monetary terms,
which is household agricultural income. In our computations, we use net measures of income and
subtract the full set of production costs from intermediate inputs (seeds and fertilizers), rental cost
of plots, rental costs of capital equipment and structures, hired labor, and transportation costs
  18
    That is, the consumption price is higher than the price at the gate, and it should be so, as in addition to the
value of the other parts of the plant that are not eaten, it includes the storage value for future consumption, or
barter. This is also reflected in the change of consumption prices across seasons. Even market prices for shelled
maize (sold in bulk) are approximately 40% higher in the lean pre-harvest season compared with the immediate
post-harvest prices; see Kaminski et al. (2014). It is our view that the shadow price of unsold maize is best
captured by the consumption price of green maize (maize on the cob), as this is the closest to the produce that is
actually consumed by the household. For other food items that do not store well or that do not produce valuable
by-products, this difference between the consumption price and the price at the gate should be less than for maize.



                                                        10
associated with inputs purchases and production sales.19 Note that in net agricultural income
we include the contribution of household labor to agricultural production.20 Livestock sales and
animal produce are also reported for the past 12 months, and we include this in agricultural income
after netting out their associated costs (e.g., animal feed, vaccinations, veterinary services, and
hired labor). As in the previous section, we include in agricultural income the value of livestock
produce that is not sold.

    Labor income is reported by occupation (main, secondary, and informal). For each activity
there is information on the average hours worked per day, average days worked per week, and
number of weeks worked per year. This allows for an estimate of yearly labor supply.21 Wages
are reported by activity but potentially in different units of time, mostly on a monthly basis for
those with steady labor income (specially in urban areas), and on a weekly or daily basis for those
working on ganyu or informal activities (e.g., landowners’ seasonal labor supply outside their own
farm in the lean season). Wage payments include salaries plus additional allowances. These
allowances could be in kind (mostly in maize) but are reported in monetary value. By combining
the wages and the labor supply in a consistent unit of time we build an estimate for annual labor
income for all individuals in the household. By summing individual labor income of all members
within households we construct a measure of annual household labor income.

    We define annual household business net income using information from all enterprises owned
by the household.22 For each enterprise we compute net income as total annual sales minus
costs. In the Malawi ISA, households report the average net income for a bad, standard, and
good month. Households are then asked how many of each type of month occurred in the past
year. In Uganda, households report both gross income and costs. In Tanzania households report
net income directly.23
  19
     In Malawi, the majority of households receive seed and fertilizer subsidies by the Farm Inputs Subsidy Program.
We use the subsidized (coupon) prices reported by each household.
  20
     This is innocuous for our purposes of measuring household income because the labor income generated by
household members in agricultural production is also part of household income. Under some assumptions, this
labor income can be separated from net profits, but this is beyond the scope of our paper.
  21
     Previous LSMS datasets provide information on labor supply with the reference period of “the past 7 days”.
ISAs complement that information with recalled hours worked per day, week, and month over the past 12 months,
which greatly facilitates the determination of the annualized labor supply. In particular, this avoids potential
measurement error from labor supply seasonality. See Rosenzweig and Udry (2014) for a discussion on how wages
are affected by seasonal weather patterns.
  22
     Privately held businesses per household (potentially more than one) include owned nonagricultural businesses
that process/sell agricultural byproducts (e.g. flour, juice, beer, jam, oil, seed, and livestock by-products),
sales of forest-based products, street or market trading businesses, taxi or pickup truck drivers, bar/restaurants,
professional services (e.g., doctor, accountant, lawyer, and midwife) etc..
  23
     As was the case for agricultural net income, business net income includes the contribution of household labor
to household businesses.




                                                        11
    Fishing net income (by fish species) is also collected. These are provided separately for each
of the two landing seasons in a year, high and low. We transform the total quantity per species
in kilograms depending on the units reported and the form of packaging, which we use to value
sold and unsold production. We net fishing income from intermediate input costs such as rented
gears (e.g., mosquito nets, beach seine, long/hand line, gillnet, fish traps), rented boats/engines,
fuel, oil, and maintenance, hired labor salaries and other in-kind payments.

    Household annual capital income includes net interest income, pension income, rental income
from nonagricultural land rental; apartment and/or house rental; shop and/or store rental; car
and/or truck, other vehicle rental; capital gains (including sales) from real estate; nonagricultural
asset sales; agricultural/fishing asset sales; and other income from inheritance, lottery, or gambling
winnings. This information is available for the past 12 months. We include in capital income
agricultural land rentals (per season) and income from renting fishing equipment (gears). As we
describe in the next section, despite the level of detail, household capital income is negligible
compared with other sources of income.

    Finally, household annual net transfers are defined as income transfers/gifts received from
rural areas/urban areas/other countries minus income transfers/gifts given in the past 12 months.
Further, the value of received aid (e.g., free maize, other free food, food/cash-for-work programs
such as Malawi Social Action Fund or Public Works Program) provided by social safety nets is
added to transfers received. The survey also records remittances in cash received from children 15
years of age or older who no longer live in the household. Neither the Tanzania nor Uganda ISAs
have a specific question on in-kind food transfers received by the household. The Malawi survey
has such a question, but less than 20% of households report receiving in-kind food transfers in
the past 12 months. Nevertheless, in the consumption questionnaire a much higher proportion of
households report eating food gifts in the past 7 days. In this 7-day recall data, 62%, 41%, and
25% of households, respectively, in Malawi, Tanzania, and Uganda report consuming food gifts.
We therefore include food gifts (deseasonalized and annualized) in our definition of disposable
income. In Malawi, we can compare the contribution of this source of income with other reported
transfers. Food gifts represent approximately 6% of total disposable income and dwarfs the 1%
contribution of net transfers in Malawi.24

2.3      Wealth and Its Portfolio

Our definition of wealth (i.e., net worth) includes land, housing, livestock, agricultural equipment
and structures (e.g., tools and barns), fishing equipment, other durables (e.g, cars, furniture, and
  24
       See further details on the computation of household income in Appendix Section A.3.



                                                       12
household electrical appliances), minus debt.25 The main difference between ISAs and previous
LSMS surveys is that in ISAs the quality of capital and its depreciation is measured in a straight-
forward way by asking: “How much would you get for this piece of equipment if you sold it now?”
This selling price accounts for capital quality and depreciation and avoids the alternative use of
the age of the assets (with potential recall error) plus assumptions on the depreciation factor to
impute the current value of capital (see Deaton (1997)). While we prefer the use of the selling
price, this is not without potential measurement error, particularly for assets such as land, for
which the market is largely underdeveloped (an issue that we discuss next).26

    The most important components of wealth are land and house values, which are self-reported.
This poses a potential measurement problem, particularly for land, which is only partially marke-
tized in rural areas. For example, in the 2010 survey in Malawi households are asked to provide
an estimate of the value of their land and all households do so, but more than four-fifths of
households live in areas where no market for land operates (see Restuccia and Santaeul`  alia-Llopis
(2015)). Most land is either granted by a village chief, inherited, or obtained as bride price.27

    The valuation of land is available for all three countries, Malawi, Tanzania, and Uganda. We
find the price of land varies considerably across countries. In Malawi, the median and mean prices
of an acre are, respectively , US$214 and 473. In Tanzania, the price distribution is much wider,
the median and the mean are, respectively, US$136 and 1,762. In Uganda, prices are consistently
higher; the median and mean are, respectively, US$ 582 and 1,811. Some of these differences
in prices may be driven by differences in land quality. In Uganda and in the north of Tanzania,
there are two rain-seasons, the territory is hillier, and there is access to Lake Victoria; the staple
crops are also more varied. The average price of land in northern Tanzania, US$2,463 per acre, is
similar to Uganda’s. In south-west Tanzania the land and climate are similar to those in Malawi
(e.g., there is one rain-season and the main crop is maize). The average price of land in southern
Tanzania, US$ 255 per acre, is closer to Malawi’s. It is reassuring that land quality and prices
are correlated as this suggests that price differences tend to capture genuine variation.

                                         alia-Llopis (2015) show that the correlation between land
       For Malawi, Restuccia and Santaeul`
  25
      A detailed construction of our wealth measures is available in Appendix Section A.4.
  26
      An alternative to measure wealth is the construction of an asset index based on ”yes/no” answers to questions
about the ownership of each individual asset. However, note that an index cannot be measured in the same
units as consumption and income, which limits the comparison between wealth and the remaining economic
variables discussed here. Further, an asset index misses some informative aspects of quintile analysis such as the
measurement of how much wealth is held by any given proportion of the population (e.g., the top 1% wealth-rich),
which we pursue in detail.
   27
      In the 2004 survey in Malawi, households are first asked whether there is a market for land in their area; only
if the answer is “yes” are they then asked to provide an estimate for land value. We find that 83.0% of households
report there is no market for land.



                                                        13
quality (at the plot level) and its price is positive and increasing with the amount of marketed
land. For instance, for rural households that do not operate any plot obtained through the
market, the effects of land quality on land price is a significant .116, while for rural households
that operated purchased land (with title), this figure increases to .503. Analogously, the difference
in the reported price of land across countries might also be due, in part, to differences in market
development. In Uganda, on the other hand, in 1998 the government enacted the Land Act with
the explicit aim of turning dwellers on land held under customary tenure into freeholders. Even if
the policy was not fully successful (see McAuslan (2003)), the development of a land market in
Uganda may help explain the disparity in valuations compared with Malawi. This issue deserves
further attention beyond the scope of this paper. Nevertheless, we should keep the issue of land
markets (or lack thereof) in mind when comparing the reported monetary value of land, hence,
wealth — across poor countries.

    Finally, we also note that the survey for Uganda does not report the value of outstanding
debt, whereas the surveys for Malawi and Tanzania do so. While debt is a minor component
of net worth in Malawi and Tanzania, Uganda’s wealth measure is likely to be overestimated
for this reason. Also, the survey for Tanzania does not report the value of housing and other
durables; thus, the monetary value of wealth reported for Tanzania is likely to be underestimated,
particularly for urban areas.

2.4   Further Measurement Issues

An important feature of the ISAs is that they are not top coded and there are very few missing
observations. Our understanding from discussions with the World Bank survey managers is that
the surveys were conducted in conjunction with national statistics offices. Ground staff were
provided with official government identification badges and respondents understood the survey to
be official government business. While this largely resolves survey response biases, the quality of
the data merits further scrutiny. Previously, we discussed two potential sources of measurement
error of annual quantities and our efforts to mitigate their effects: (i) deseasonalizing consumption
(see Section 2.1) and (ii) comparing our monetary measures of wealth with physical measures
(i.e., acres for land) to overcome the presence of underdeveloped markets for assets, particularly
for land (see Section 2.3).

    Food consumption is the lion’s share of household consumption in our settings and hence,
this measurement is perhaps the most important aspect of consumption. The ISA collection of
food consumption is based on a 7-day recall questionnaire. These short-recall periods tend to




                                                14
yield better consumption measures (Beegle et al. (2012)),28 but at a cost. Given that ISAs are
spread over 12 months, the surveys will do a good job in recovering average food consumption
in the population, but they will potentially do a poor job in measuring annual dispersion as part
of this dispersion will be artificially due to seasonal variation. This shortcoming can easily be
dealt with using standard deseasonalization techniques (see Section 2.1). Measurement error for
other types of consumption, particularly information on durables collected with 12 month recall,
is potentially still present, although we note that durable consumption represents a minor share
of total consumption (4% in rural areas and 6% in urban areas; see below).

    Income is based on recall of the entire production per crop and plot for the past two harvests.
The harvest referred to in the questionnaire may have taken place months earlier. We conduct
robustness testing for potential recollection bias for production using measures of household
income only for those households interviewed within 3 months after the rainy season harvest
has been completed (May, June, July); the mean and median of total agricultural production in
kilograms is virtually identical for these three months and for the yearly values. The rainy season
represents 93% of total agricultural production. In addition to potential recall bias in agricultural
production, the underreporting of income is a recurrent issue in household surveys in both rich
and poor countries (Deaton (1997)). This is potentially more stringent for the self-employed and,
hence, of particular relevance for our rural settings, where most households operate as farms. In
this context, it is reassuring that when we focus on rural households that we can categorize
as autarkic (i.e, with no production sales) the reported agricultural production (with potentially
months of recall bias) and the reported annualized consumption imply very similar quantities
(see Section 2.2.1). That is, if we use the measure of food consumption from own production
as external validation for our measure of agricultural production, we find only a small scope for
measurement error in the ISA’s agricultural production (from recall or elsewhere). Further, in
many instances surveys provide checks for internal consistency. For example, interviewees are
asked about both total sales, and also sales by crop. The interviewer then must check in situ that
the sum of the crops coincides with the total or otherwise reinterview. This design is not applied
to nonfarm business income. Hence, the problem of underestimating income potentially persists,
in particular, for those households with a large proportion of business income (i.e., urban areas).

    Finally, our trimming strategy consists of two steps that mitigate the presence of outliers. As
a first step, we exclude households with zero calorie consumption or with an intake per person
above the maximum daily of 10,000 Kcal.29 As a second step, we trim clear outliers after a visual
inspection by subitems and then by aggregated measures. We finalize with an implied trimming of
  28
       See also Gibson et al. (2014).
  29
                                                                         aes et al. (2015).
       We describe in detail how we construct caloric intake in De Magalh˜



                                                         15
2% of households for Malawi and Tanzania and 4% for Uganda.30 The final samples for Malawi,
Tanzania, and Uganda include, respectively, 12,015, 3,012, and 2,337 households.

2.5    Household Survey Data vs. National Accounts

Table 1 compares our household survey data from ISAs with the national accounts data for 2010
from the World Development Indicators at the World Bank database. The national accounts
figures are reported in panel (a), and the mean income and consumption per capita computed
with the ISA household survey data adjusted as described in this Section are reported in panel
(b). Focusing on Malawi, we find that the mean income per capita from our household survey
data is US$343, which is very close to the national accounts number of US$359. However, note
that the composition of income as estimated by the household survey implies that agricultural
ouptut represents 43% of total income, while this figure is solely 29% in the national accounts.
Further, also note that in Section 2.2 we showed that ISAs do an excellent job of measuring
unsold agricultural production. Combining these two findings suggests that national accounts are
underestimating agricultural income by 41% in Malawi.31 After such an adjustment, the national
accounts estimate would result in a higher income than the household survey estimate. This
then also implies that, if the national accounts are doing a good job in measuring nonagricultural
income, the household survey is potentially underestimating (or not observing) some component
of nonagricultural income. For example, illegal income from the diversion of international aid (e.g.,
in the form of bribes) could potentially account for some of this discrepancy.32 Finally, while in
Tanzania the household survey estimates of mean income are lower than those from the national
accounts, in Uganda the household survey estimates and national accounts are not significantly
different. Further, in Tanzania and Uganda the household survey measures of agricultural income
are similar to the national accounts counterparts.33

   The main disparity between the household survey and the national accounts estimates is
found in consumption. The ISA estimates of consumption per capita are higher than the national
accounts estimates for all three countries. But, in fact, the high quality of consumption data
  30
      The 2010 surveys for Tanzania and Uganda are the second waves of panel surveys. Splitter households were
dropped to compute the tables in this paper and are not accounted for in the percentage of trimming reported.
   31
      That is, the ratio between the agricultural income from the household surey and the national accounts is,
US$343×0.43/US$359×0.29=1.41.
   32
      The reason is that international aid shows up in national accounts but not necessarily in household sur-
vey data. See, for example, the cashgate scandal covered by The Guardian, http://www.theguardian.
com/global-development/2014/nov/11/malawi-official-jailed-cashgate-scandal-aid. This way, it
is likely that the household survey will capture some, but not all, of the international aid, which according to WDI
represents 26% of the total income in Malawi in 2010.
   33
      The ratio between the agricultural income from the household survey and the national accounts is,
US$378×0.34/US$524×0.28=0.87 for Tanzania and US$509×0.28/US$471×0.25=0.99 for Uganda.



                                                        16
in the household survey is consistent with the value of consumption per capita being higher
in household survey data than in the national accounts.34 For example, the national accounts
are likely to underestimate the value of consumption from own agricultural production, which we
estimate to be 20% of total consumption from the household survey data in Malawi. This suggests
that national accounts might not only be underestimating consumption growth, as suggested in
Young (2012), but also the consumption level.35 The ranking of consumption across countries
is maintained in the macro and the micro data: Malawi is the poorest, following Tanzania, and
Uganda.


3        Cross-Sectional Inequality in Sub-Saharan Africa

Here we investigate the distributions of CIW separately for the rural and urban areas of Malawi,
Tanzania, and Uganda. First, we investigate the level and distributional behavior of CIW in
Section 3.1. Second, we explore the behavior of saving rates and CIW conditional on income in
Section 3.2. Third, we study the top and bottom of the distributions and their respective welfare
in Section 3.4.

3.1      Rural and Urban Differences in Levels and Inequality

We present the mean for CIW split between rural and urban areas in Table 2. We find that
differences in the levels of income and consumption between rural and urban areas are an order
of magnitude larger than the differences across these countries. Across rural areas, the largest
difference in mean per capita consumption is between Tanzania (US$280) and Uganda (US$321)
— a 15% difference36 — while the mean per capita consumption in Malawi is US$297. The largest
difference in mean income per capita in rural areas is between Tanzania (US$221) and Malawi
(US$246) — an 11% difference.37 In contrast, in urban areas, mean per capita consumption
and income in urban Malawi is, respectively, US$648 and US$618, which implies a difference of
118% and 151%, respectively, compared with its rural Malawi counterparts. The magnitude of
the difference between rural and urban areas is even higher for Tanzania and Uganda. In all, the
rural-urban gap in consumption and income within countries is at least 10 times larger than the
    34
     ISA provides close to gold standard measures are based in short-recall periods (Deaton (1997) and Gibson
et al. (2014)), see our discussion Section 2.
  35
     However, note that our measure of consumption is entirely different from that in Young (2012), who uses
the ownership durables (and other education and health measures) to proxy for real consumption. We use actual
consumption data.
  36
     These two estimates are statistically different. The confidence intervals are, respectively, [US$271; US$290]
and [US$299;US$343].
  37
     The respective confidence intervals are [US$203; US$239] and [US$236; US$256].




                                                      17
country differences across Malawi, Tanzania, and Uganda.38

   While the differences in the level of CIW between rural and urban areas are perhaps well
understood, whether distributional differences exist across rural and areas is a lesser-understood
phenomenon. We find that inequality is higher in urban areas than in rural areas (see Table 3).
This is true for the triplet of CIW in all three countries.

     Focusing on consumption, we find that the log variance in rural areas is 0.49 in both Malawi
and Tanzania, which is lower than their respective variances 0.59 and 0.60 in urban areas. That
is, consumption is more unequal in urban areas than in rural areas, by roughly a factor of 1.20 for
Malawi and Tanzania. The similarity between the shape of the entire distribution of consumption
in Malawi and Tanzania can also be seen in Figure 2. In Uganda, consumption inequality is
roughly 70% larger than in Malawi and Tanzania in both rural and urban areas.39 Hence, as in
Malawi and Tanzania, urban consumption inequality in Uganda is roughly 1.26 times larger than
in its rural areas.

    The variance of logged income varies across all three countries; Malawi has the lowest in-
equality (rural: 0.99 and urban: 1.60), following by Tanzania (1.44 and 1.90) and Uganda (1.67
and 2.01). Income is also more unequal in urban areas than in rural areas, by a factor of 1.6 for
Malawi (i.e., a larger factorial rural-urban difference in log variances than that of consumption
inequality, 1.2). Further, the inequality of income is larger than that of consumption in both rural
and urban areas, indicating a substantial degree of partial insurance. Precisely, the log variance
of consumption is between 32% and 50% of the log variance of income across countries and
residency areas.

    Finally, wealth inequality is larger than income and, hence, consumption inequality. This can
also be observed in Figure 2; the densities for wealth are flatter than for consumption and income.
  38
      In more detail, urban Malawi and Tanzania have virtually identical per capita consumption — respectively,
US$648 and US$641 — and very similar income — respectively, US$618 and US$578. Neither comparison is
statistically different. Moreover, the composition of income and consumption is also very similar (see Table 2). In
rural areas, agricultural income represents more than 50% of total income, followed by labor and then business.
In urban areas, labor represents more than 50% of total income followed by business and agriculture. In contrast,
Uganda is relatively different from Malawi and Tanzania. Even though the estimated income per capita in
rural areas is very similar across all three countries, the main source of income in rural Uganda is business, not
agriculture. Business, instead of labor, is also the dominant source of income in urban areas. The difference in
urban consumption is even more striking: Uganda’s urban per capita consumption is around two-thirds higher
than Malawi’s and Tanzania’s. Moreover, consumption of durables in Uganda is relatively more important than in
the other two countries. According to the household survey data, Uganda as a whole is clearly richer on average
than the other two countries, and this difference is related to relative higher business income and more durable
consumption in urban Uganda.
   39
      The higher consumption inequality in Uganda than in Malawi and Tanzania is largely driven by the share of
durable consumption, see Table 2. See also Table B-5 in the online appendix.



                                                       18
For example, focusing on Malawi,40 wealth inequality is roughly 1.5 larger than income inequality
(in terms of variance of the logs) in rural areas and roughly 3 times larger in urban areas. Further,
there is certainly much more wealth inequality in urban than in rural areas, by a factor of 3 (i.e.,
a larger factorial rural-urban difference than that of income inequality, 1.6), suggesting a stronger
ability to accumulate wealth, and hence, to create dispersion in urban areas than in rural areas.41

    Comparing our results with US inequality, we find that income dispersion does not transmit
as much into wealth dispersion in SSA as it does in the United States. Precisely, in terms of
household income, the United States has a log variance of 0.99 for the year 2010, as reported
in D´ıaz-Gim´enez et al. (2011) using the Federal Reserve’s Survey of Consumer Finances, while
the variance of logs for urban income for Malawi, Tanzania, and Uganda is, respectively, 1.60,
1.90, and 2.01. That is, urban income inequality is larger in SSA than in the United States. In
contrast, in terms of wealth the United States has a log variance of 4.53, which is surprisingly
similar to the inequality measures of urban Malawi, Tanzania, and Uganda, respectively, 4.52,
4.44, and 4.67. That is, using urban Malawi as a benchmark, the Unites States is able to generate
the same amount of wealth dispersion as urban Malawi with about 60% of its income dispersion.
This lower transmission from income to wealth inequality in urban SSA compared with the United
States is also present in rural SSA. For example, rural Malawi has income inequality similar to
that in the United States, with a variance of logs of 0.99, while wealth inequality in rural Malawi
is 1.49: one-third of the US wealth inequality. That is, with the same income dispersion as rural
Malawi, the United States is able to generate three times the wealth dispersion of rural Malawi.
Similar relations are attained for Tanzania and Uganda.42

    To summarize, the inability of income dispersion in SSA, which is at least as large as that
of the United States, to translate into wealth dispersion as large as that of the United States,
points to an inherent inability to save and accumulate wealth in both urban and, more so,
rural SSA. Moreover, despite this evidence on a lesser ability to save, we find that consumption
inequality is 49% and 37% of income inequality (in terms of the variance in logs) in rural and
urban Malawi, while this figure is larger and close to 55% in the United States (Blundell et al.
(2008)).43 Combining these two dimensions suggests that the greater ability to contain the
   40
      In terms of wealth (i.e., net worth), we focus on Malawi because Tanzania lacks housing and other durables
data and Uganda debt data.
   41
      Since most of the variation in wealth in rural areas comes from land (roughly 80%), this suggests that either
land cannot be accumulated (or create wealth dispersion) in rural areas or rural households do not have access
to the accumulation of other types of assets.
   42
      Similar insights are obtained from alternative measures of inequality such as the Gini index, mean to median
ratios, and 90th/10th percentile ratios, also reported in Table 3.
   43
      These figures are 34% and 33%, respectively, for rural and urban Tanzania and 50% and 52%, respectively,
for rural and urban Uganda. For the United States, we use the reported average from the mid-1980s to the early
1990s in Blundell et al. (2008).


                                                        19
dispersion in consumption in SSA compared with the United States must involve consumption
insurance mechanisms other than self-insurance (i.e., own savings), a matter to which we return
in Section 5.4.

3.2    Rural-Urban Differences As Income Increases: Conditional and Joint Distributions

Here, we rank households by their income and carefully study the behavior of their CIW. This
analysis is conducted for rural and urban areas, respectively, in panels (A) and (B) in Table 4.44
Note that while we focus on Malawi in this Section, the analogous partitions for Tanzania and
Uganda can be found in the online appendix.

    The average difference in the level of consumption between rural and urban areas translates
into the entire distribution of income. That is, on average consumption is 2.20 times larger in
urban than in rural areas, and this rural-urban consumption gap is relatively stable across all
income quintiles with roughly a factor of 2, except for the top income quintile for which urban
consumption is larger by a factor of 2.64 (and 3.53 for the top 1%). If we look at consumption
per capita, urban consumption remains two times higher than rural consumption, which shows
that the difference in levels is not driven by the larger size of rural households. Note also that
the US$1 dollar per capita a day threshold is only achieved by the fourth quintile of the income
distribution in rural areas, whereas in urban areas, all quintiles have per capita consumption above
the US$1.

    In contrast to consumption, while the rural-urban income gap is on average similar to that
of consumption (i.e., 2.23), this gap increases with income. Starting at the bottom 20% of the
income distribution we find similar income per capita levels between rural and urban areas (a
higher level in urban areas by a factor of 1.09), but this gap increases monotonically to 2.82 for
the top 20% and to 3.83 for the top 1%. Finally, as in income, the rural-urban gap in wealth also
increases with income, starting with urban wealth that is 1.41 times larger than rural wealth for
the bottom 20% of the income distribution and increasing to a factor of 3.62 for the top 20% of
income earners and to 6.46 for the top 1%. Thus, the growth in the rural-urban gap as income
increases is virtually identical for income and for wealth except for the top 1% and top 5% of the
income distribution.45
  44
      Hereafter, we focus on Malawi because its sample size is large enough to maintain accuracy in our estimates
when we partition the distribution of CIW. The sample size for Malawi is 12,015 households, while in Tanzania it
is 3,012 and in Uganda 2,337. Note that our sample size is roughly 2.5 times larger than the Survey of Consumer
Finances used to conduct a similar partition for the United States in D´     ıaz-Gim´enez et al. (2011). Analogous
distributions of CIW conditional on wealth, land, and consumption can be found in the online appendix.
   45
      Precisely, the rural-urban gap from the respective bottom 20% to the top 1% (and top 1% to 5%) of income
earners grows by 251% (186%) in income and by a larger 355% (237%) in wealth. Instead, this increase for the
second, third, and fourth quintiles of the income distribution is on average lower for wealth (21%) than for income


                                                        20
    Similar findings are extracted from the joint densities of CIW for rural and urban Malawi
plotted in Figure 3. The correlation between income and wealth is higher in urban areas, 0.36,
than in rural areas, 0.17.46 In rural areas income and wealth have a correlation of 0.10 and 0.12,
respectively, for the bottom 80% and top 20% of the income distribution. In urban areas, we find
a slightly opened L-shaped joint density, which implies that for the bottom 80% of the income
distribution, the correlation between income and wealth is 0.06, while this correlation is 0.30 for
the top 20% of income earners. Thus, only the income-rich households in urban areas are able
to accumulate wealth. Overall, there is a weaker link between income and wealth in SSA than
in the United States, where the correlation of these two variables is 0.57 (D´  ıaz-Gim´enez et al.
(2011)). Finally, the correlation of consumption and income is also higher in urban areas, 0.53,
than in rural areas, 0.30.

    To summarize, only the households at the top of the income distribution in urban areas are
able to transform their rural-urban income differential gain (i.e., their higher income compared
with the top income counterparts in rural areas) into a somewhat sizable rural-urban differential in
wealth. The joint densities of income and wealth reinforce this result, implying that the income-
rich households in urban areas are able to accumulate more wealth than their rural counterparts.
This suggests that the high saving rates of the top-income earners in rural areas might be less
persistent than those of the top income earners in urban areas. Finally, we find less growth for
the rural-urban consumption gap than for income, which suggests an ability to partially insure.
The joint density of consumption and income suggests that this insurability of consumption is
larger in rural areas than in urban areas, an argument further developed in De Magalh˜   aes et al.
(2015).

3.3    Saving Rates

New insights emerge by looking directly at the saving rates — that is, one minus the consumption-
to-income ratio. As is the case in the United States, a large fraction of the population has a
negative saving rate in Malawi. Using the Consumer Expenditures Survey (CEX), Sabelhaus and
Groen (2000) find that approximately half of the households in the United States have a negative
saving rate. In rural Malawi, we find that the saving rates are, on average, positive for only
the top 10% of the income distribution, while in urban Malawi the saving rates are, on average,
positive only for the top 5% of the income distribution (see Table 5). This implies that while in
the United States the average saving rate is 0.12 for the 1992 CEX, the saving rates in rural and
urban Malawi are, on average, negative and, respectively, -0.23 and -0.15.
(32%).
  46
     In Tanzania, this correlation is 0.45 for urban households and 0.40 for rural households. For Uganda, the
numbers are 0.42 and 0.32.

                                                     21
    Interestingly, for the top income earners in Malawi we find saving rates that are very similar
to (or slightly above) the saving rates of the top income earners in the United States. The saving
rates for the top 10% of the income distribution are 42% for rural Malawi, 46% for urban Malawi,
and 36% for the United States.47 In rural and urban areas, the top 1% and top 5% of the income
distribution share a very similar saving rate. For the top 1%, the saving rate is 0.72 in rural areas
and 0.74 in urban areas; for the top 5%, the saving rate is 0.37 in rural areas and 0.36 in urban
areas. Across the income-poor households, we find a larger negative saving rate in Malawi than
in the United States. The saving rates for the bottom 10% of the income distribution are, on
average, -5.0 in rural Malawi and -12.0 in urban Malawi, and this figure is -1.30 for the United
States. That is, unlike at the top of the income distribution, the rural-urban differences in saving
rates are large at the bottom of the income distribution.

    In contrast, when we rank households by wealth, we find larger discrepancies in the saving
rates between rural and urban areas across the top wealth groups. The top 1% (and 5%) of
the wealth distribution saving rate is 0.08 (and 0.09) in rural areas and 0.42 (and 0.08) in urban
areas. This difference reinforces the idea of a larger persistence of wealth in urban areas compared
with rural areas. The bottom of the wealth distribution in rural and urban areas behave similarly
with a negative saving rate of -0.60 and -0.87, respectively. In terms of land, only the top 1%
land-richest households show positive saving rates in rural areas.

    To summarize, only the top income earners in Malawi are able to save, and they do so with
saving rates that are high and similar to those of the top income earners in the United States.
Now, with such similar saving rates, can the top income earners in Malawi accumulate as much
wealth as the top income earners in the United States? This is the question we turn to next.
  47
     Here, we compare in a bit more of detail the saving rates by income decile between the United States from
the 1992 CEX as documented in Sabelhaus and Groen (2000) and the saving rates that we find for Malawi (ISA
2010):

                                                    Saving Rates by   Income   Deciles
                              1       2       3        4       5          6        7     8      9      10
        U.S. CEX 1992        -1.3   -0.63   -0.66    -0.12 0.00         0.05     0.10  0.19    0.26   0.36
        Malawi — Rural       -5.0   -2.48   -1.57    -1.27 -0.97       -0.76    -0.46 -0.30   -0.06   0.44
        Malawi — Urban      -12.0   -3.81   -2.47    -1.65 -1.41       -1.12    -0.90 -0.32   -0.22   0.47

The income decile 1 represents the bottom 10% of the income distribution and the income decile 10 represents
the top 10% of the income distribution. This table excludes 73 observations with either zero or negative income.




                                                        22
3.4    The Top and Bottom of the Income and Wealth Distributions and Their Respec-
       tive Welfare

Here, we closely explore the top and bottom of the income and wealth distributions. In Table 6
we use the entire country data, combining rural and urban areas, to compare the concentration
of income and wealth in SSA with rich and emerging economies as reported by Piketty (2014).
In both the United States and in Malawi, the top 1% of the income distribution earns 20% of
the total income. In Uganda, the concentration is even higher at 31% of total income, by far the
highest concentration of income in all countries reported. In Tanzania, the top 1% earns 15%
of total income, the same as in Britain, and similar to the average of emerging countries. If we
consider the top 10%, it becomes clear that income is at least as concentrated in SSA as in the
US; the top 10% in the United States earn 48% of total income. In Malawi, the top 10% earn
50% of total income, in Tanzania 51%, and in Uganda 65%. It is not the case, however, that
higher income inequality translates to higher wealth inequality than in the United States. The top
10% of the wealthiest households in Tanzania and Uganda hold 72% and 69%, respectively, of
the total wealth, a similar number to the United States, 71%. In Malawi, wealth concentration is
lower; the top 10% holds no more than 57% of total wealth, a concentration lower than Sweden’s.

    We now take a closer look at the entire distribution of income separately for rural and urban
Malawi in Table 7, panels A1 and B1. Those panels show the share of total CIW ranked by
household income. In rural Malawi, the top 10% of the income distribution earns 43% of total
rural income. In urban areas, the top 10% earns 62% of total urban income. The top 1% of the
income distribution earns 14% and 25% of total income, respectively, in rural and urban areas.
On the other end of the distribution, the bottom 20% in rural areas earns 3% of total rural
income and the bottom 20% in urban areas earns 2% of total urban income. In urban Malawi,
the income gap between the rich and poor is very similar to that of the United States in 2010,
where the top 1% of the income distribution earns 21% of total income and the bottom 20%
earns 3% of total income (D´        enez et al. (2011)).
                             ıaz-Gim´

    In terms of wealth, we find that the top 10% of the wealth distribution holds 49% of total
wealth in rural Malawi and 73% in urban areas (see, respectively, panels A2 and B2 in Table 7).
Further, the top 1% of the wealth distribution holds 17% and 32% of total wealth in rural and
urban areas, respectively. Strikingly similar to urban Malawi, the top 10% and top 1% of the
                                                                                     ıaz-Gim´
wealth distribution in the US holds, respectively, 71% and 34% of total wealth (see D´      enez
               48
et al. (2011)). However, the current concentration of wealth at the top of the distribution in
  48
   In Tanzania, in terms of income, the top 10% and top 1% rich in rural (urban) areas earn, respectively, 50%
(41%) and 14% (18%), and 62% (66%) and 27% (26%) in Uganda. In terms of wealth, the top 10% and top
1% rich in rural (urban) areas hold, respectively, 67% (90%) and 20% (44%) in Tanzania, and 67% (70%) and


                                                     23
Malawi is much lower than that attained by developed countries when these economies were
experiencing growth takeoffs and industrialization. For instance, in Britain the top 10% and top
1% of the wealth distribution held 82% and 53%, respectively, of the total wealth in 1810; in
France these figures were 80% and 46%, and in Sweden 83% and 57% (Piketty (2014)). If these
high levels of inequality are an integral part of the process of growth takeoff, then it seems that
none of the three SSA countries studied herein is experiencing such a takeoff.

    Further, exploiting the joint information about each household’s income and wealth, we find
that income inequality transmits to wealth inequality, but much less so than in the United States.
In Malawi, the top 10% income-rich households hold 27% of total wealth in rural areas and 49% in
urban areas. In contrast, in the United States the top 10% of the income distribution holds 60%
of total wealth — a substantial difference between the United States and Malawi, particularly
rural Malawi. This differential is largest for the top 1% income-rich households, who hold 26%
of total wealth in the United States, but merely 4% of total wealth in rural Malawi and 11%
in urban Malawi. This implies a nationwide average of 5%, that is, the top income-richest 1%
households in SSA hold a share of total wealth that is one-fifth of its US counterpart. This is
further direct empirical evidence on the weaker connection between income and wealth in SSA.
This happens despite the similar saving rates between Malawi and the US for the top income
earners as reported in Section 3.3. These two factors together suggest that high saving rates
do not last long enough for households to accumulate wealth — for example due to a lower
persistence of high incomes in SSA than in the US, a phenomenon discussed in Section 6.

                                              ıaz-Gim´
    Finally, unlike previous studies such as D´      enez et al. (2011) and Piketty (2014), our data
allow us to link income and wealth to a more direct measure of welfare: consumption. We find
that in Malawi the top 10% of the income distribution accounts for 21% of total consumption in
rural areas and 30% in urban areas, while the bottom 10% of the income distribution accounts for
6% of total consumption in both rural and urban areas. Therefore, in terms of consumption, the
difference between the top 10% and the bottom 10% of the income distribution drops substantially
from a 90th -10th percentile ratio of 43 in income to 4 in consumption in rural areas and from a
ratio of 63 in income to 5 in consumption in urban areas.

     To summarize, while the disparities between the top and bottom of the income distributions
are large in SSA and indeed are similar to those previously documented for the United States and
Europe, we find that these income disparities do not translate into wealth or consumption inequal-
ity in SSA as they do in the United States and Europe. The inability of high-income households to
either accumulate more wealth or consume more suggests powerful redistributive arrangements,
28% (23%) in Uganda. See the online appendix.



                                                24
particularly in rural areas, that mitigate the high income inequality and serve as consumption
insurance mechanisms. For example, the somewhat uniform distribution of consumption across
land quintiles (see Table 7 Panel A3) suggests that the nonmarketed land distribution is likely to
be used to mitigate inequality of consumption.


4         Deconstructing Consumption, Income, and Wealth

In this section, we investigate the composition of consumption by item, the sources of income,
and the wealth portfolio (Table 8) for Malawi.49

    The main component of consumption is food. In rural areas, the food share remains above
60% throughout the entire income distribution except for the top 1%. In urban areas, the food
share varies between approximately 50% and 60% across the income distribution except for the
top 20% and above (e.g., food consumption represents 21% of total consumption for the top 1%
income earners). In rural areas, food consumption from own production and purchases account
for similar shares of total consumption, and consumption from own production stays above 25%
of total consumption across all income quintiles. In urban areas, food consumption from own
production is no higher than 5% of total consumption throughout the quintiles and purchased
food accounts for 46% of total urban consumption.

    Most other components of consumption (i.e., clothing, utilities, and health) are relatively
stable across the income distribution in both rural and urban areas except for durables (other
than housing), which remains very low across the distribution except for the top 20% and, in
particular, the top 1% of income earners. Precisely, the average share of durables consumption
in total consumption is 2% in rural areas and 4% in urban areas, while the top 1% of income
earners consume a share of 13% and 20% of durables, respectively, in rural and urban areas.
A similar phenomenon occurs for other nondurable consumption (candles, matches, cigarettes,
public transportation, soap, paraffin lamps, mats, carpets, bricks, cement). We also find a
moderate increase in schooling expenditures as income increases, particularly in urban areas:
from an average share of 3% of total consumption to a share of 10% for the top 1% of income
earners.

    In terms of sources of income, we find that the share of agriculture and labor income in rural
areas is similar across the entire income distribution. On average, agricultural income represents
60% of all income and labor represents 19%. In rural areas, the difference across the income
distribution is that the income-rich households have a high share of business income (the top
20% of households have a business share of 14% and the top 1% of 29%, while the bottom 20%
    49
         The analogous analysis for Tanzania and Uganda can be found in the online appendix.


                                                         25
have a business share of 3%) and the poor have a high share of food gifts (the bottom 20% of
households have a food gift share of 17%, while for the top 20% the food gift share is 3%).50
In urban areas, the main components of income are labor, varying from 50% to 60% across the
quintiles, and business income with a share that is (roughly) 20% across income quintiles except
for the top 20% of households where business income represents 40% of total income. For the
top 1% of income-rich households, the business share is highest (i.e., 57%) and represents a
higher share than labor (i.e., 38%). There is still a sizable share of agricultural income in urban
households, between 14% and 19% for the bottom 80% of the income distribution, that declines
to 5% for the top 1% of the income distribution. The share of food gifts as a source of urban
income is 11% among the poor and decreases to 1% among the rich.

    The wealth portfolio in rural areas has three major components: land (44%), housing (30%),
and livestock (11%). The share of land wealth varies between 46% and 52% in the bottom
80% of the income distribution and decreases to 36% for the top 20% of the distribution and
to 21% for the top 1%. Note that the bottom 1% of the income distribution in rural areas
has a low share of its wealth derived from land, 30%. This is not surprising as precisely 49%
among the bottom 1% owns no land. The share of housing wealth tends to decline with income
in rural areas, from a share of 50% for the bottom 1% of the income distribution, to 38% for
the bottom 20%, and to 24% for the top 1%. Instead, the share of livestock (a mechanism
to accumulate wealth explored in Rosenzweig and Wolpin (1993)) and other durables seems to
correlate positively with the income in rural areas. Other durables and livestock each represent
5% of the wealth portfolio for the bottom 20% in the wealth distribution, whereas for the top
20% they are, respectively, 15% and 20%. In urban areas, the major components of wealth are
housing (59%), other durables (27%), and land (14%). To put these figures into context, note
                                                               ıaz-Gim´
that the share of housing wealth is 24% in the United States (D´       enez et al. (2011)); thus,
the share of housing in Malawi is about 2.5 larger than in the United States. This suggests
that saving constraints in SSA might also take the form of limitations in the available savings
portfolio.51 Further, the share of housing wealth is roughly invariant to income, ranging from
45% to 61% across income quintiles in urban areas. The share of other durables is substantially
larger for the top 20% of the income distribution (33%) and top 1% (37%) than for the rest
of the income distribution (approximately 20%). The share of land among urban households is
largest for the middle quintile (37%) and lowest for the bottom and top quintiles (respectively,
  50
     Note that our measure of food gifts as a source of income is compiled from the consumption questionnaire.
The results in Table 8 suggest that the questions on annual transfers in the income questionnaire are not able to
capture the most relevant transfers received by the poorest 20%: food gifts.
  51
     Santaeul`alia-Llopis and Zheng (2015) provide a similar argument for China, where growth over the past 20
years is associated with a portfolio largely dominated by lifetime savings (i.e., housing wealth and investment in
children) that deter consumption insurance.


                                                       26
8% and 5%). The highest reported debt is -3% of total net worth for the top 1% of the income
distribution.

    To summarize, the composition of consumption (largely food) reflects that these are largely
subsistence economies. Not surprisingly, the ability to save and accumulate wealth can be limited
by the need for food consumption. This constraint is relaxed only for the top 1% of the income
distribution in rural areas and the top 5% in urban areas, for whom the food shares falls below
50%. The composition of income (mostly agricultural) in rural areas reinforces the idea of a
subsistence economy. In rural areas, only for the top 1% of the income distribution is the share
of agricultural income less than 50%. Finally, the similarity in the wealth portfolio across income
groups (except for the top 1%) further reflects an inability to accumulate wealth. Only livestock
and durables show signs of accumulation.


5        Sources of Cross-Sectional Inequality

We study several sources of economic inequality and their implication for wealth accumulation.
These sources include the usual suspects such as age, household composition, and education.
Further, we provide a cross-sectional assessment of the relationship between self-reported types
of risk, insurance mechanisms, and inequality. We pay particular attention to the ability to borrow,
controlling for self-reported selection in loan applications, and to the implications of the degree
of capital market incompleteness for inequality. Last, we find substantial variation in societal
systems (i.e., matrilineal or patrilineal) and study their gender-specific implications.52

5.1      Demographics

Household composition turns out to be crucial to understanding the life cycle behavior of con-
sumption and income, particularly in rural areas.

5.1.1      Age

Table 9 shows the age distribution of household heads in rural and urban areas. The average age
of a household head in rural areas is 43, while it is 39 in urban areas.

    The average CIW (in current USD) by each age group is documented in Table 9. We find a
clear hump over the life cycle for CIW, particularly in urban areas. In urban areas, panel B2 shows
that consumption increases by a factor of US$3,684/US$1,854 = 1.98 from the youngest age
    52
     In this section, we focus on rural and urban Malawi. The analogous tables for Tanzania and Uganda are
available in the online appendix.



                                                   27
group (15-24 years) to the age group (45-54 years) associated with highest consumption. Income
follows a similar pattern, which in this case implies an increase by a factor of US$3,623/US$939
= 3.85. This hump behavior over the life cycle has also been documented for industrialized
economies and for some middle-income countries, see Deaton and Paxson (1994). Average
positive saving rates (close to zero) are observed only among prime age households (35-44 years),
who represent 25% of urban households. In urban areas, the hump in wealth is even steeper
than that of consumption and income and peaks at a later age group (55-64 years) at a level
US$6,661/US$933 = 7.13 times larger than the wealth of the youngest age group. While this
ability of urban households to accumulate wealth over the life cycle is non-negligible, it is dwarfed
by the ability of US households to accumulate wealth. In the United States, the level of wealth
at ages 55 to 64 is 20 times higher than the wealth of the youngest group (D´    ıaz-Gim´ enez et al.
(2011)).

    In rural areas, there is a much smaller hump in all three categories of CIW (see panel A2 in
Table 9). First, consumption increases by roughly a factor of US$1,559/US$1,137 = 1.37 from
the youngest age group (15-24 years) to the age group associated with highest consumption that
now peaks at relatively younger ages (35-44 years) than in urban areas (45-54 years). Second,
income increases from US$868 at ages 15 to 24 to US$1,327 at ages 45 to 54 — a factor of
1.52. None of the age groups in rural areas has a positive average saving rate. Wealth, driven
mostly by land, increases by a factor of US$1,699/US$639 = 2.65 between the ages of 15 to 24
and 55 to 64. Note that land holdings peak in the 45 to 54 age group in rural areas — slightly
younger than the peak in wealth — suggesting a wealth portfolio that changes over the life cycle
moving away from land.

    To summarize, the main difference between rural and urban areas over the life cycle is the
ability to generate income and accumulate wealth. While income at the ages of 15 to 24 is only
1.08 times larger in urban areas than in rural areas, at its peak — ages 45 to 54 — income is 3.73
times larger in urban areas. This is also the case for wealth accumulation. While at young ages
(15 to 24 years) wealth is only 1.46 times larger in urban areas than in rural areas, at its peak —
ages 55 to 64 — wealth is 3.93 times larger in urban areas. Thus, the ability to generate income
and accumulate wealth over the life cycle is almost four times larger in urban areas than in rural
areas. This large discrepancy between rural and urban areas, however, is much less pronounced
for consumption, for which we find that urban areas enjoy 2.36 times more consumption than
rural areas — less than half the rural-urban income differential — at their respective life cycle
consumption peaks.




                                                 28
5.1.2    Household Structure

Across Malawi, the average household size is 4.58 and the dependency ratio, which is largely
driven by the presence of children, is 1.16. This ratio is higher in rural areas than urban areas:
1.22 versus 0.88.53 The larger share of children in rural households is also present at the tails
of the distribution. While the population share of childless households is 18% in rural areas,
this share is higher in urban areas, 24% (see panels A1 and B1 in Table 10). Further, while the
population of households with at least four children is 22% in the rural areas, this share is 15%
in urban areas. Indeed, the proportion of households with a dependency ratio higher than 1 is
55% in rural areas, while this figure is 44% in urban areas.

    We investigate the CIW conditioning on the number of children living in the household. Panel
B2 of Table 10 shows that in urban areas, households with a dependency ratio lower than 1
have clearly higher average consumption and income, respectively, by a factor of 1.38 and 1.89,
compared with households with a dependency ratio above 1. This and the lack of a positive
relationship between consumption/income and the number of children in urban areas suggests
that children represent, on average, a substantial consumption/income cost to urban households.

    In contrast, in rural areas, where the vast majority of the population lives, we find that house-
hold size is positively related to income (see panel A2 of Table 10). In the same direction, mean
consumption and income are similar between households with dependency ratios above and below
1 in rural areas. Our facts suggest that children, taken at face value, are not necessarily a bur-
den on rural household consumption/income. Indeed, children might be important in generating
income, helping with farm activities from early ages, and also serving as an insurance mechanism
for the short run through labor-supply adjustments (see Beegle et al. (2006) and Jacoby and Sk-
oufias (1997)). Finally, in the environments that we study where wealth accumulation is difficult,
children can be viewed as a viable form of asset that substitutes for social security as proposed
in Boldrin and Jones (2002) and more recently pursued in Banerjee et al. (2014).54
  53
     ISA household rosters provide individual demographic information about all members of the household.
In particular, the relationship between each member of the household and the household head is identified.
Relatives who are members of the household include children (i.e., son/daughter-in-law, niece/nephew, grand-
children), wife/husband, father/mother, father/mother-in-law, brother/sister, brother/sister-in-law, and grandfa-
ther/grandmother. Non-relatives who are members of the household include servants and lodgers living in the
household. We are interested in household size (the sum of all members), as well as the household structure
by age and sex groups. We define the household-level dependency ratio as the number of dependants, that is,
                                                                    own−in          −in
children younger than 15 years of age in the household, nc = nc            + nother
                                                                               c        , and adults older than 52
years of age (average life expectancy), over the number of working-age adults between 15 and 52 years of age.
The household size that we obtain for the 2010 Malawi ISA is similar to the one obtained from the nationally
representative 2010 Demographic and Health Survey for Malawi, 4.72. If we include all individuals available on
the household roster who do not necessarily belong to the household, we obtain a household size of 5.64.
  54
     An alternative explanation that merits further investigation is whether larger households are allocated more


                                                       29
5.1.3    Consumption and Income in Adult-Equivalent Scales

An important aspect of household structure is its potential role in explaining the hump in house-
hold consumption, in particular, nondurable consumption (see Deaton and Paxson (1994) and
Attanasio et al. (1999)). For completeness, here we reproduce the results in De Magalh˜  aes et al.
(2015) that find the life cycle behavior of consumption in adult-equivalent scales shows practically
no hump in SSA. That is, the evolution of the life cycle household structure (see the panels in
Figure 4) is the main determinant of the observed hump in the age profile of household con-
sumption (see the right panes in Figure 4). This is particularly the case for rural areas where,
while household consumption shows an increase from ages 25 to 45 of a factor of roughly 1.25,
adult-equivalent consumption does not grow over the life cycle and, if at all, slightly declines
to almost 0.90 at age 50 (see the center panels in Figure 4). In urban areas, while household
consumption shows an increase from ages 25 to 45 of a factor of 1.70, this factor drops to 1.15
in terms of adult-equivalent consumption (see bottom panels in Figure 4). Combining rural and
urban areas (i.e., nationwide) adult-equivalent consumption shows absolutely no trend over the
life cycle (see top panels in Figure 4).

    Finally, the behavior of income per adult over the life cycle displays a clear hump that increases
from the age of 20 to a peak at 35 years of age by a factor of 1.20/0.77=1.55 in rural areas,
while in urban areas the hump in income per capita is larger and increases from the age of 20
to a peak at 45 years of age by a factor of 2.25/0.80=2.81 (see the right panel in Figure 4).
Our results echo those in Bils and Klenow (2000) and Lagakos et al. (2012) that focus on the
behavior of wages and document flatter age profiles in poor countries. Here we instead focus on
household total income to consider the fact that in the poor countries that we investigate, there
is a large rural population (approximately 80%), for whom agricultural production is the main
source of income. Our findings suggest that the flatter country-wide life cycle profiles of income
in developing countries may be the result of a composition effect driven by a large share of rural
households, which indeed have flat age profiles, and balance the steeper age profiles of urban
households.

5.2     Migration

The process of economic growth and structural transformation (Gollin et al., 2002, 2004) is
inevitably related to migration. We find clear evidence of this structural shift as 60% of the
current urban residents consist of migrants from rural areas (see panel B1 in Table 11), where
rural-to-urban migrants are defined as those who resided in rural areas prior to their current
resources (i.e., land) by the chief/community.


                                                 30
urban residency since childhood and independently of the year of migration. Further, there is
also urban-to-urban migration that accounts for another 18% of the current urban residents,
which implies that nonmigrants represent only 22% of the current urban population. In contrast,
the nonmigrants account for the largest share of the current rural population, 65% (see panel
A1 in Table 11). Finally, rural-to-rural migration represents 31% of households and urban-to-
rural migration is minor and represents merely 4%. This implies that the absolute rural-to-urban
                      ×18.3%
migration flow is 60%4%×81.7%
                              = 3.36 times larger than the opposite urban-to-rural flow, which
generates a structural shift that decreases rural population from a counterfactual of 89.4% (i.e.,
a scenario under which all migrants in the data return to their original residency) to the current
81.7% in 2010/11.55 Interestingly, our migration patterns differ substantially from the averages
recently documented in Young (2013) for 39 countries for which the rural-to-urban migration flows
                                                                                       .9%×41.5%
are much more similar to their urban-to-rural counterpart, precisely, by a factor of 23
                                                                                     13.2%×58.5%
                                                                                                 =
     56
1.28.

    To explore migration as a potential source of inequality, we study whether migrants fare better
in terms of average CIW than nonmigrants. First, note that urban residency is associated with
better CIW than rural residency even within the nonmigrant population (see the first columns of
panel A2 and B2 in Table 11). Urban nonmigrants enjoy consumption that is higher than the
consumption of their rural nonmigrant counterparts by a factor of 1.70, higher income by a factor
of 1.59, and higher wealth by a factor of 2.09. Second, exploring the incentives of rural-to-urban
migration — that is, comparing the rural-to-urban migrants with the nonmigrants in rural areas
— we find that this type of migration boosts consumption by a factor of 2.17, income by a
factor of 2.10, and wealth by a factor of 1.89. Further, we find that the rural-to-urban migrants
tend to do slightly better than the urban-born who do not migrate but worse than urban-born
that migrate within urban areas. Indeed, the urban-to-urban migrants are the group that does
  55
      Indeed, using IHS2 data for Malawi we find that in 2004-05 the rural population accounted for 86.8% of the
total population, suggesting that a proportion of our migration patterns occurred in a period that spans more
than 5 years. Indeed, we find that 50.70% of households have migrated in the past 10 years, 32.1% in the past 5
years, 15.5% in the past 2 years, and 8.86% within the past year. If we focus on the urban residents and define
as migrants as only those who migrate in the past five years, rural-to-urban migrants account for 18% of total
current urban residents, 8% are urban-to-urban migrants, and 74% are nonmigrants. If we focus on the rural
residents and define as migrants as only those who migrate in the past five years, rural-to-rural migrants account
for 8% of total current rural residents, 1% are urban-to-rural migrants, and 91% are nonmigrants. Last, migration
is a phenomenon of the young, with a median age at migration of 25, and more educated, with 1.54 years of
schooling more than the nonmigrants of their cohort; see also De Magalh˜    aes et al. (2015).
   56
      This last ratio for Young (2013) is computed using statistics reported in his Table II. We find reasonable
larger rural-to-urban migration flows for the economies that we study, which are currently in the earliest stages of
economic development and still hold a large fraction of their populations in rural areas, roughly 80%. Instead, the
average figures in Young (2013) incorporate middle-income countries with substantial completion of the rural-to-
urban transition. In those middle income countries the self-selection mechanisms studied in Lagakos and Waugh
(2013) are likely to be operative with less frictions and costs to migration than those potentially present in the
economies that we investigate, a feature that we acknowledge deserves further exploration.


                                                        31
best with a consumption that is 1.80 times larger than that of urban nonmigrants, an income
that is 2.62 times larger, and wealth that is 2.05 times larger. Third, in rural areas, within
rural migration also helps improve CIW, although these gains are less substantial than within
urban areas. Specifically, rural-to-rural migrants enjoy higher consumption and income than their
rural nonmigrant counterparts by respective factors of 1.17 and 1.15. There are, however, no
significant differences in terms of wealth or land. Finally, the few urban-to-rural migrants tend
to do, on average, as well as the rural-to-rural migrants but worse than their urban counterparts.
That is, moving from the city to the countryside is, on average, not a good deal.57

    It is obvious that migrants might not necessarily attain the average CIW profiles of the hosting
locations. First, we focus on the urban areas as hosting locations. Rural-to-urban migrants are
underrepresented among the poorest and richest in both consumption and income (compared
with their average population, 60%). Thus, rural-to-urban migrants are concentrated in the
mid-consumption and mid-income quintiles, and are more underrepresented in the top income
quintiles, although they still represent a sizable 49% and 39% of the population in the top
1% of the respective consumption and income distribution. In terms of wealth, the rural-to-
urban migrants are slightly overrepresented in the bottom 20% of the wealth distribution and
underrepresented at the top 20% of the wealth distribution.58 The overrepresented group among
the consumption and income poorest are the nonmigrants (e.g., 35% of the bottom 1% in
consumption are nonmigrants) compared with a 22% share of nonmigrants in the total urban
areas. Thus, in terms of consumption and income it seems that rural-to-urban migrants do
better than urban nonmigrants, although not in terms of wealth. A potential explanation is that
part of the migrant population is temporary and does not stay in urban areas long enough to
save and accumulate wealth. Finally, it is clear that urban-to-urban migrants do better as they
are heavily overrepresented among the richest in all CIW dimension — precisely, 50% of the top
1% of the consumption distribution, 56% of the top 1% of the income distribution, and 48%
of the top 1% of the wealth distribution — while they simply represent 18% of the total urban
population. Interestingly, urban-to-urban migrants are also overrepresented at the bottom of the
wealth distribution.
  57
     The reasons for migration into or within urban areas are almost exclusively work related, either to start a new
job or business (36%), to look for a job (21%), or a job transfer (13%). The reasons to migrate within rural areas
are mostly “marriage” (34% of those migrating), “parents move” (21%), “looking for land to farm” (11%), and
work-related reasons (15%). For those who moved from urban areas, marriage plays a much smaller role (14%).
A substantial part of this movement is due to marriage customs of moving to the spouse’s village. In Section 5.5,
we discuss how Malawi is divided between a societal system where the wife joins the husband in his village and
another where the husband joins the wife in her village.
  58
     In more detail, 28% of the population in the top 20% of the wealth-richest are nonmigrants, 46% in the
top 10% to 5%, and 24% in the top 5% to 1%, although only 3% of the top 1% of the wealth distribution are
nonmigrants.



                                                        32
    Second, focusing on rural areas as hosting locations, we find that rural nonmigrants, similar
to their urban nonmigrant counterparts, populate the bottom of the consumption and income
distributions and also the top of the wealth distribution (see panel A3 in Table 11). That is, the
nonmigrants are relatively more consumption and income poor, but also relatively more wealth
rich. Similar to the urban-to-urban migrants, the rural-to-rural migrants are overrepresented in the
top of the consumption and income distributions but, unlike urban-to-urban migrants, the rural-
to-rural migrants are not overrepresented in the top of the wealth distribution. That is, the rural-
to-rural migration does not help accumulate wealth as much as the urban-to-urban migration.
Finally, the urban-to-rural migrants, like the rural-to-rural migrants, are overrepresented in the
top of the consumption and income distributions and underrepresented at the top of the wealth
distribution.

    To summarize, we find migration is associated with an improvement of average CIW. In
particular, rural-to-urban migration and urban-to-urban migration boosts CIW by a factor of
roughly two compared with their reference scenarios. However, migration does not always help
wealth accumulation. Rural-to-urban migrants still have a difficult time accumulating wealth and
are underrepresented at the top of the wealth distribution, while urban-to-urban migrants and,
to a lesser extent, urban nonmigrants are somewhat more able to accumulate wealth.59 Similar
patterns arise within rural populations, the top 1% is largely overrepresented by rural nonmigrants
instead of migrants, suggesting a lesser scope for wealth accumulation for those who migrate to
rural areas (from either rural or urban areas).

5.3    Education

In 2010, the average number of schooling years among household heads was 5.5, 4.8 in rural
areas, and 9 in urban areas.60 This rural-urban differential in education is also attained in terms
of education composition: 71% of rural household heads have incomplete primary schooling at
  59
      As a cautionary note, a more complete picture of the welfare gains of migration would incorporate the possi-
bility of higher income risks and potentially less consumption insurance associated with rural-to-urban migration
(e.g., Morten (2013); Munshi and Rosenzweig (2014); and Santaeul`     alia-Llopis and Zheng (2015)).
   60
      Malawi has an 8-4 educational system (IPUMS international). This implies 8 years in primary school and
4 in secondary school. Individuals who ever attended school record the highest class level attended and the
highest educational qualification acquired. For individuals with completed education , we use this information
to group individuals into six levels of completed education: no education, primary dropouts, primary, secondary,
college, and training college. If the maximum class level attended is nursery/preschool, we consider these cases
as never attended school. Primary school dropouts include individuals who attended school for at least one year
and do not hold a Primary School Leaving Certificate (PLSC); primary school achievers include individuals have
attended school for at least 8 years and hold a PLSC; secondary school achievers are individuals who hold a Junior
Certificate (JC) (ages 13-15) or a Malawi School Certificate of Education (MSCE) (ages 16-17), university college
individuals have attended at least one year of university and hold a university diploma or a post-graduate diploma;
and training college includes individuals with at least one year of training college and who hold a non-university
diploma.


                                                        33
most and only 8% have secondary education or more, while these figures are, respectively, 31%
and 34% for urban household heads (see panels A1 and B1 in Table 12). This is consistent with
smaller returns to schooling in rural areas than in urban areas (see panels A2 and B2 of Table 12).
The increase in income is twofold from “primary dropout” to “secondary or more” in rural areas,
whereas in urban areas the increase is more than fourfold.

    Regarding the ability to save, we note that in rural areas not even the more educated have
positive saving rates. This suggests that human capital accumulation might not necessarily
be linked to wealth accumulation in the rural settings that we study. In this same direction,
we find that in rural areas the more educated (ie., those with secondary education or more),
form the group that is most overrepresented (with respect to their proportion of the total rural
population: 8%) in the top 20% and top 1% of the income distribution, by a factor of 18/8=2.25
and 30/8=3.75, respectively, and similar figures are attained in the consumption distribution.
However, the more educated are only slightly overrepresented in the top 20% and 1% of the
wealth distribution, by a factor of 10/8=1.25 for both the top 20% and 1%. That is, overall,
education seems to be associated with more income and consumption, but not necessarily wealth
in rural areas. However, in urban areas, we note that households with “secondary or more”
(34% of the urban population) have a nonnegative saving rate of 3.1%. In this case, households
with secondary or more education are overrepresented (with respect to their proportion of total
urban population: 34%) in the top 1% of the CIW distributions (by factors of 82/34=2.41,
75/34=2.20, and 80/34=2.35, respectively). Hence, in contrast to rural areas, education seems
to be positively related with wealth accumulation in urban areas.

    Finally, we note that education can have farther-reaching implications for inequality through its
interaction with health as more educated individuals in developing countries tend to have access to
more treatment in case of illnesess/injuries and are more likely to take preventive action to reduce
the probability of illness, see Dupas (2011). This extends to the HIV epidemic that ravages African
economies (Young, 2005, 2007; Santaeul`    alia-Llopis, 2008). For example, Iorio and Santaeul` alia-
Llopis (2015) explicitly study the relationship between human capital and HIV and find that while
the probability of being HIV-positive increases with human capital at the early stages of the
HIV epidemic, this probability decreases as the more educated switch to safer sex behaviors by
decreasing the number of sexual partners,61 and finally, the relationship between human capital
and HIV status rises again at subsequent stages of the epidemic as the more educated gain
more access to antiretroviral (ARV) drugs than the less educated. This phenomenon potentially
increases income inequality further given the effects of HIV on productivity (Levinsohn et al.
  61
    Related to this sexual margin, Greenwood et al. (2013) recently consider short- versus long-term relationships
to explicitly model the HIV epidemic when the information about the HIV status of potential partners is not
perfect.

                                                       34
(2011)) by making the more educated with relatively more access to ARVs even more productive
than the less educated.62 Overall, these patterns between human capital and HIV suggest that a
universal ARV policy in countries like Malawi — where about 10% of the population is infected
with HIV — could help mitigate income inequality but we acknowledge this deserves further
exploration. Note that ISA does not test the sampled households for HIV.63

5.4     Risk and Insurance

Here we exploit the large household sample in the Malawi ISA 2010 to study the cross-sectional
relationship of risk, insurance, and inequality, relying on both the risks and consumption insurance
mechanisms self-reported by households. We pay particular attention to the self-reported ability
to borrow.

5.4.1    Self-Reported Risks and Insurance Mechanisms

In the past 12 months, in rural areas, 43% of households have suffered an aggregate shock, 6%
an idiosyncratic shock, and 22% both types of shocks in the past 12 months, leaving 29% of
the population without shocks. In urban areas, 15% of households have suffered an aggregate
shock, 13% an idiosyncratic shock, and 11% both types of shocks in the past 12 months, leaving
61% of the population without shocks. The most common aggregate shock in rural areas is rain,
43% of households report a rain shock (too little or too much), followed by agricultural costs
(33%) and food prices (26%) (see panel A1 in Table 13).64 In urban areas, a smaller number
— 49% of households — report some shock. The most common aggregate shocks in urban
areas are unexpected high food prices, reported by 18% of households, which highlights the tight
link between rural and urban areas. In both rural and urban areas, health shocks are the most
important idiosyncratic shocks, followed by death and theft.65 In general, households that report
suffering one of these main shocks (in particular, rain in rural areas) are poorer in consumption
   62
      This might not be straightforward as education subsidies that increase schooling do not reduce sexually
transmitted infection by themselves and require a combination with a governmental HIV curriculum to do so (see
Duflo et al. (2015)). Further, by decreasing life expectancy and the returns to schooling, it is natural to think
that HIV potentially reduces schooling and, hence, human capital accumulation (see Fortson (2011)). Indeed,
the increasing availability of ARVs tends to increase physical and human capital investments through a lower
perception of mortality risk (Baranov and Kohler (2015)).
   63
      See also Durevall and Lindskog (2012) for further discussion on the effects of HIV on inequality.
   64
      Here we define rain as a type of aggregate shock, but we acknowledge there is rainfall dispersion across
households. If we redefine rain as an idiosyncratic shock, then we find that in rural areas 14% of households have
suffered only an aggregate shock, 25% only an idiosyncratic shock, and 31% both types of shocks in the past 12
months. The planting of maize needs to be timely. Rains that come too early or too late are as problematic as
droughts and floods. Our measure of shock includes all these possibilities.
   65
      In Table B-17 in the online appendix we examine the compositions of each quantile according to health
indicators.


                                                      35
and income compared with those that report no shock (see panels A2 and B2 in Table 13).
However, households have the same average wealth and land holdings independently of having
self-reported a shock or no shock. Further visual inspection suggests that these self-reported
shocks are not correlated with wealth. This apparent lack of correlation between shocks seems
to hold for both rural and urban households (see panels A3 and B3 in Table 13).

     Table 14 shows the self-reported insurance mechanisms used to cope with shocks. We group
insurance mechanisms as specific strategies to cope against these shocks into “no insurance”,
“self-insurance”, and “mutual insurance”. We focus on what households declared as their main
coping strategy. The items “no Insurance”, “mutual Insurance”, and “self-Insurance” do not sum
to 100%; instead, they sum up to the percentage of households that reported having a shock
(see Table 13). Precisely, in rural areas, 71% of households reported suffering a shock: 35% of
total rural households reported using no insurance to deal with that shock, 28% reported some
form of self-insurance, and 7% reported some form of mutual insurance. In urban areas, 39% of
households reported suffering a shock: 22% of total urban households report using no insurance,
15% reported some form of self-insurance, and 2% reported some form of mutual insurance. That
is, in rural areas, 35/71=49% of households report not using an insurance mechanism conditional
on having reported a shock, while this figure is slightly higher (22/39=56%) in urban areas.

   The results in previous sections imply that only those at the top of the income distribution
have some ability to accumulate wealth. This suggests that only those households will be able to
use their savings or accumulated wealth to smooth shocks as a form of self-insurance. Indeed,
we find that households that report report self-insurance are better off in CIW than those that
report no insurance or some form of mutual insurance (see panel A2 in Table 14). In particular,
note from panel B2 Table 14 that urban households that report using “savings” as their main
coping strategy are the only ones with a positive saving rate of 0.12.66

    We note that the households that resort to mutual insurance in the event of a shock are those
that are the worst off in terms of CIW. This seems to suggest that those that report not having
used insurance mechanisms are self-selected; that is, the households that report no insurance
are possibly not in bad enough shape to resort to mutual insurance. Indeed, among the bottom
of the CIW distributions, self-insurance (mainly “savings”) is less common as a coping strategy.
Among the bottom 20% of the CIW distributions in rural areas, “mutual insurance” is cited by
10% of the households as the most important coping strategy (see panel A3 in Table 14).67 This
  66
     Udry (1995) provides evidence of the use of savings as an insurance mechanism to smooth consumption in
northern Nigeria.
  67
     Note that mutual insurance is exclusively family help; the percentage of households that report receiving
mutual insurance from government and nongovernmental organizations is negligible in both rural and urban
areas.


                                                     36
is similar to the number of households that report using “savings” (12-15%) as their main coping
strategy. This is further evidence that using accumulated wealth to deal with shocks is more
problematic in rural areas compared with urban areas. In urban areas, even among the bottom
20% in CIW, “savings” is more than twice as likely to be cited as the main coping strategy than
“mutual insurance” (see panel B3 in Table 14). Nevertheless, “savings” as a coping strategy is
also problematic for the bottom 20% of the CIW distribution in urban areas. The main coping
strategy for the bottom 20% of the urban distribution of CIW is “diet” restrictions.

5.4.2    Credit and Ability to Borrow

Here, we explore the ability to borrow from self-reported needs, actions, and outcomes of loan
applications. This provides direct evidence of the degree of capital market incompleteness at the
household level.68 First, households are asked whether they have applied or not applied for a
loan in the past year. Second, households are asked whether or not a loan was needed. The
main reasons for not applying for a loan when needing one (i.e., direct evidence of capital market
incompleteness) are “not knowing any possible lender” and “having no collateral”. Note that
combining “need” and whether or not they applied, we directly address the issue of self-selection.
Third, households are also asked whether they succeeded in receiving the loan and the amount
borrowed if they were successful.

    We investigate the differentials between rural and urban areas in the ability to borrow. We
find that a slightly higher percentage of households report needing a loan in rural areas, 75%,
than in urban areas, 67% (see panels A1 and B1 of Table 15). However, we can see that in both
rural and urban areas only a very small group of households actually receive a loan: 13% of rural
and 20% of urban households. This suggests that loans are not easily available in either urban
or rural areas. The large majority of households do not even try to obtain a loan: 56% in rural
areas and 40% in urban areas. As expected, rural households are less likely to meet the necessary
requirements to apply for a loan. This can be seen directly from the data as the application rate
conditional on needing a loan is 27% in rural areas and 40% in urban areas. However, once we
condition on the households that do meet the requirements to formally apply for a loan, the data
show that the success rate is nearly identical in rural areas, 48%, and urban areas, 50%. This
suggests that if we deal with the selection of who applies for a loan, being in an urban or rural
area makes no difference on the odds of obtaining a loan.

    In rural areas, those households that obtain a loan are richer on average in all dimensions.
In particular, those that obtain a loan are richer than those that claim not to need a loan (see
  68
    See Buera et al. (2011) for an study of the implications of credit constraints for occupational choice and
growth.


                                                     37
panel A2 of Table 15). In urban areas those that claim not to need a loan are the ones who
are richer in all three dimensions (see panel B2 of Table 15). Across the CIW distribution, the
median interest rate charged is considerably higher for the poor than for the rich in both rural
and urban areas (see panels A3 and B3 of Table 15). For example among the consumption poor
in rural areas the interest rate is 10%, while among the rich it is 5%.

    Finally, it must be noted that the fact that urban households have higher loan application rates
and an overall larger ability to borrow does not necessarily imply larger consumption insurance
in urban areas than in rural areas. To link credit to consumption insurance we must dig deeper
into the motives to borrow, which are available in the survey. We note that there are 45/18=3.6
times more loans acquired for start-up capital than for consumption in urban areas, while this
ratio is much less, 1.6, in rural areas. This suggests that in urban areas, relative to rural areas,
borrowing is used to target production activities rather than for consumption insurance purposes.

5.5    Societal Systems

Malawi is a particularly interesting country to explore the role of gender in macroeconomics69
as it has two clearly distinct societal systems: one matrilineal and matrilocal, and the other
patrilineal and patrilocal. In patrilineal societies the wives join their husbands in the husband’s
village. In matrilineal societies, the husbands join their wives in the wife’s village. We find that
a large proportion of households in Malawi are headed by women, and quite a few by divorced
women. Among the rural households in our sample, 26% are headed by women: 38% of these
women are divorced and 46% are widows. In urban areas, the percentage of households headed
by women is lower, 18%, but the proportion of divorced and widowed women is similar. On
the other hand, there are very few households headed by divorced or widowed men: 3% of all
households, and a negligible proportion is headed by men or women who never married. The
majority of households are monogamous households headed by men, 65%, and there is also a
sizable minority of polygamous households, 6%.70 As we discuss next, the phenomenon of women
as household heads is more prominent in the matrilineal south of Malawi.

   In Malawi, different regions are associated with different societal systems. The north is
patrilineal, the south is matrilineal, and the center has a mix of societal systems.71 The clear
  69
      See a general discussion in Tertilt (2012).
  70
      In the online appendix we explore the composition of each quantile by its gender and marital status for Malawi
in Table B-20, Tanzania in Table B-21, and Uganda in Table B-22
   71
      In the IHS3 survey the information on societal systems is not provided by household but in the community
questionnaire, in which a local knowledgeable informant reports the percentage of households in each enumeration
area that are matrilineal and patrilineal. We have defined an enumeration area as patrilineal if the informant
reported that at least 80% of households in that enumeration area were patrilineal; we have then coded these
households as patrilineal. We have defined matrilineal households analogously. Households have been declared


                                                        38
regional demarcation can be seen in the panels A1 and B1 of Table 16. The north is exclusively
patrilineal in both rural and urban areas (i.e., no household in our sample lives in an enumeration
area where less than 80% of households are patrilineal). The south is overwhelmingly matrilineal
but there is a nonnegligible minority of patrilineal households. The center has a similar number
of households that live in patrilineal and in mixed areas, and a minority of households that live in
matrilineal areas.

    Separation arrangements also differ substantially across societal systems, and separations have
important implications for wealth accumulation. In matrilineal societies, in case of divorce, the
man leaves behind the land to be reassigned within the woman’s village, the dwelling goes to the
woman, and the children stay with the woman in her village. In patrilineal societies the women
leave without the children, leaving behind the land and housing (a bride price may be paid back).72
These differences imply a higher bargaining power for women in matrilineal societies, evidence
of which can be seen in much higher divorce and remarriage rates in matrilineal areas of Malawi
(Reniers (2003)). Specifically, we find that households headed by divorcees (80% of which are
women) represent 7% in north Malawi and twice as much, 13%, in south Malawi. We also find
that that 38% of all households are headed by women in south Malawi, compared to 26% for the
entire country.73

    In rural areas, in terms of CIW, we find that the rural south is the poorest region and the
rural center is the richest region (see panel A2 in Table 16). In the rural center households in
all three societal systems have similar consumption, but the patrilineal areas are slightly richer
in income, wealth, and land. The only societal group (with non-negligible population) that has
a clear positive saving rate are the matrilineal households in the urban center (see panel B2 in
Table 16).74 These households represent 12% of the urban population, are the richest group in
all CIW dimensions, and have a saving rate of 33%, which is similar to the saving rate of the top
income decile in the United States. In fact, this societal group is overrepresented in the top 1%
in all dimensions: 50% of the consumption top 1%, 42% of the income top 1%, and a staggering
75% of the wealth top 1% (see panel B3 in Table 16). All this suggests that the restrictions
mixed if located in districts where neither systems had more than 80%.
   72
      Different societal systems may also have implications regarding the relative population growth of lineages
with each group. As noted by Miers and Kopytoff (1979), the growth rate of a particular matrilineage is fixed by
the number of sisters, whereas in a patrilineal group the men can marry more wives and therefore increase the
lineage’s growth rate. This way, the institution of slavery allowed matrilineal societies to relax this constraint, as
slave wives were accepted as “acquired” sisters, and could therefore carry on the lineage. This has implications
on the role of slavery in growth models such as Lagerl¨ of (2009).
   73
      In this direction, Gneezy et al. (2009) find experimental evidence that women take more risk in matrilineal
societies — as much as men in patrilineal societies.
   74
      In the north, mixed and matrilineal households also show positive saving rates, but they represent a minority
of households as almost all households living in the north are patrilineal.



                                                         39
to wealth accumulation are also likely to be related to societal system affiliation and the norms
within that group.


6        Income and Wealth Mobility

Why are the top income earners in SSA not able to accumulate wealth despite high saving rates?
Indeed, top income earners in SSA save at rates as high as the top income earners in the United
States (see Section 3.3). However, they are not able to accumulate as much wealth as their US
counterparts. Specifically, the top 1% income earners hold 4% of total wealth in rural Malawi and
11% in urban Malawi, while this figure is 26% in the US (see Section 3.4). In this context, note
that wealth accumulation is the result of past saving behavior and, hence, at saving rates similar
to those of the US, the top income earners in SSA will accumulate sizeable wealth only if they
remain at the top for a period of time comparable with their US counterparts. Therefore, a simple
way to reconcile these two facts — high saving rates but relatively low wealth accumulation for
the top income earners in SSA — is through the study of income (and wealth) mobility.

    Table 17 shows the economic mobility patterns we find. To discuss mobility we focus on the
fraction of households that leave a given income or wealth quantile. We focus on two cases, the
United States from 2001-2007 PSID data reported in D´       ıaz-Gim´enez et al. (2011) and Uganda,
for which we have available panel ISA data for two waves: 2005-06 and 2009-10.75 First, we find
much less persistence in the bottom and the top of the income distribution in Uganda than in the
United States. Precisely, in rural Uganda, 51% of households in the top quintile leave it within
four years, whereas this figure is 46% in urban Uganda; that is, there is more income persistence
in urban areas than in rural areas. Instead, in the United States only 23% of households at the
top quintile leave it within four years; that is, the income earners at the top quintile in both rural
and urban Uganda are likely to leave that quintile at least twice as fast as their US counterparts.
This implies a larger downward mobility in SSA than in the United States. Second, focusing on
the bottom of the distribution, we also find larger upward mobility in the SSA than in the United
States. Specifically, around 70% of households in the bottom 20% of the income distribution
leave that quintile in both rural and urban Uganda, whereas only 23% of households at the
bottom of the income distribution move upwards from that quintile in the United States. Third,
wealth mobility is large in absolute terms, suggesting a substantial amount of wealth risk. At
the same time, wealth is more persistent than income. For instance, while 71% of households
move upwards from the bottom income quintile in rural Uganda within a span of four years, this
    75
                                                                          ıaz-Gim´
    We transform the mobility statistics for the 6-year span reported in D´       enez et al. (2011) for the PSID
data into 4-year span statistics using simple averages to make the mobility statistics between the United States
and Uganda comparable.



                                                       40
figure is 58% for wealth. Similarly, while 51% of households would move downward from the top
income quintile in rural Uganda, this figure is 45% for wealth. The scenario for urban Uganda is
very similar to the rural scenario but with higher persistence in both income and wealth.

    To summarize, the fact that the rich remain rich and the poor remain poor is more a feature
of the United States than of SSA; there is larger economic mobility in SSA for both income
and wealth. These results suggest that the fact that top income earners in SSA are not able to
transform their saving into wealth at the rate that US households do can be explained, at least
partially, by a relatively lower persistence at the top of the income distribution.


7   Conclusion

We provide a comprehensive study of the cross-sectional joint distributions of consumption, in-
come, and wealth using new and unique nationally representative ISA data for three of the poorest
countries in the world: Malawi, Tanzania, and Uganda. First, we find that the transmission from
income to wealth is low, which we relate to a large inability to save and accumulate wealth in
SSA compared with other world regions. Second, we provide evidence of a low transmission
from income to consumption, which suggests that despite being unable to persistently save —
and hence, self-insure — SSA households are able to insure their consumption relatively well, a
phenomenon that requires the presence of powerful informal insurance arrangements. Combining
these two findings suggests a trade-off between growth and insurance in SSA. This raises the
question of whether the process of accumulation (i.e., saving) and growth potentially requires the
disruption of consumption insurance for SSA, a phenomenon already experienced by China (see
Santaeul`alia-Llopis and Zheng (2015)).

    Looking ahead, our analysis opens the road to potentially more empirical and theoretical
assessments of how the joint distributions of consumption, income, and wealth are determined
across aggregate stages of economic development. This implies expanding our analysis not only
to other SSA countries for which ISA data are becoming available (Burkina Faso, Ethiopia, Mali,
Niger, and Nigeria) but also to other developing countries worldwide. We are currently exploring
this type of analysis using LSMS cross-sections beyond those under the ISA umbrella.



References
Attanasio, O., Banks, J., Meghir, C., and Weber, G. (1999). Humps and Bumps in Lifetime
  Consumption. Journal of Business and Economic Statistics, 17:22–35.

Banerjee, A., Meng, X., Porzio, T., and Qian, N. (2014). Aggregate Fertility and Household

                                               41
  Savings: A General Equilibrium Analysis using Micro Data. Technical report, National Bureau
  of Economic Research.

Baranov, V. and Kohler, H.-P. (2015). The Impact of AIDS Treatment on Savings and Human
  Capital Investment in Malawi. Working Paper, University of Melbourne and University of
  Pennsylvania.

Beegle, K., De Weerdt, J., Friedman, J., and Gibson, J. (2012). Methods of Household Con-
  sumption Measurement Through Surveys: Experimental results from Tanzania. Journal of
  Development Economics, 98(1):3–18.

Beegle, K., Dehejia, R. H., and Gatti, R. (2006). Child Labor and Agricultural Shocks. Journal
  of Development Economics, 81(1):80–96.

Bils, M. and Klenow, P. J. (2000). Does Schooling Cause Growth? American Economic Review,
  90(5):1160–1183.

Blundell, R., Pistaferri, L., and Preston, I. (2008). Consumption Inequality and Partial Insurance.
  American Economic Review, 98(5):1887–1927.

Boldrin, M. and Jones, L. E. (2002). Mortality, Fertility, and Saving in a Malthusian Economy.
  Review of Economic Dynamics, 5(4):775–814.

Buera, F. J., Kaboski, J. P., and Shin, Y. (2011). Finance and Development: A Tale of Two
  Sectors. American Economic Review, 101(5):1964–2002.

De Magalh˜                              alia-Llopis, R. (2015). Consumption and Expenditures in
          aes, L., Koh, D., and Santaeul`
  Poor Countries: A Life With No Humps and Rural-Urban Insurance Traps to Growth. Working
  Paper, University of Bristol and Washington University in St. Louis.

Deaton, A. (1997). The Analysis of Household Surveys. A Microeconometric Approach to De-
  velopment Policy. Johns Hopkins University Press.

Deaton, A. and Paxson, C. (1994). Intertemporal Choice and Inequality. Journal of Political
  Economy, 102(3):437–467.

Deaton, A. and Zaidi, S. (2002). Guidelines for Constructing Consumption Aggregates for Welfare
  Analysis. Living Standards Measurement Study, Working Paper No. 135, Washington, D.C.:
  World Bank.

 ıaz-Gim´
D´                                 ıos-Rull, J. V. (2011). Facts on the Distributions of Earnings,
        enez, J., Glover, A., and R´
  Income, and Wealth in the United States: 2007 Update. Federal Reserve Bank of Minnepolis
  Quarterly Review, 34(1):2–31.

Duflo, E., Dupas, P., and Kremer, M. (2015). Education, HIV, and Early Fertility: Experimental
  Evidence from Kenya. CEPR Discussion Papers - 10338.

Dupas, P. (2011). Health Behavior in Developing Countries. Annual Review of Economics, 3.



                                                42
Dupas, P. and Robinson, J. (2013a). Saving Constraints and Microenterprise Development:
  Evidence from a Field Experiment in Kenya. American Economic Journal: Aplied Economics,
  5(1):163–192.

Dupas, P. and Robinson, J. (2013b). Why Don’t the Poor Save More? Evidence from Health
  Savings Experiments. American Economic Review, 103(4):1138–71.

Durevall, D. and Lindskog, A. (2012). Economic inequality and HIV in Malawi. World Develop-
  ment, 40(7):1435–1451.

Fortson, J. G. (2011). Mortality Risk and Human Capital Investment: The Impact of HIV/AIDS
  in Sub-Saharan Africa. The Review of Economics and Statistics, 93(1):1–15.

Gibson, J., Beegle, K., De Weerdt, J., and Friedman, J. (2014). What Does Variation in Survey
  Design Reveal About the Nature of Measurement Errors in Household Consumption? Oxford
  Bulletin of Economics and Statistics.

Gneezy, U., Leonard, K. L., and List, J. A. (2009). Gender Difference in Competitions: Evidence
  from a Matrilineal and a Patriarchal Society. Econometrica, 77(5):1637–1664.

Gollin, D., Parente, S. L., and Rogerson, R. (2002). The Role of Agriculture in Development.
  American Economic Review, 92(2):160–164.

Gollin, D., Parente, S. L., and Rogerson, R. (2004). Farm Work, Home Work, and International
  Productivity Differences. Review of Economic Dynamics, 7(4):827–850.

Greenwood, J., Kircher, P., Santos, C., and Tertilt, M. (2013). An Equilibrium Model of the
  African HIV/AIDS Epidemic. NBER Working Paper 18953.

Heathcote, J., Storesletten, K., and Violante, G. L. (2009). Quantitative Macroeconomics with
  Heterogeneous Households. NBER Working Paper No. 14678.

                      alia-Llopis, R. (2015). Education, HIV Status, and Risky Sexual Behavior:
Iorio, D. and Santaeul`
   How Much Does the Stage of the HIV Epidemic Matter? Working Paper, University of Bologna
   and Washington University in St. Louis.

Jacoby, H. G. and Skoufias, E. (1997). Risk, Financial Markets, and Human Capital in a Devel-
  oping Country. Review of Economic Studies, 64(3):311–335.

Kaminski, J., Christiaensen, L., and Gilbert, C. L. (2014). The End of Seasonality? New Insights
  from Sub-Saharan Africa. Policy Research Working Paper, World Bank.

Karlan, D., Ratan, A. L., and Zinman, J. (2014). Savings by and for the Poor: A Research
  Review and Agenda. Review of Income and Wealth, 60(1):36–78.

Krueger, D. and Perri, F. (2006). Does Income Inequality Lead to Consumption Inequality?
  Evidence and Theory. Review of Economic Studies, 73:163–193.

Krueger, D., Perri, F., Pistaferri, L., and Violante, G. L. (2010). Cross Sectional Facts for
  Macroeconomists. Review of Economic Dynamics, 13(1). Special Issue, January 2010.

                                              43
Lagakos, D., Moll, B., Porzio, T., Qian, N., and Schoellman, T. (2012). Experience Matters:
  Human Capital and Development Accounting. NBER Working Papers 18602.

Lagakos, D. and Waugh, M. E. (2013). Selection, Agriculture, and Cross-Country Productivity
  Differences. American Economic Review, 103(2):948–80.

      of, N.-P. (2009). Slavery and Other Property Rights. The Review of Economic Studies,
Lagerl¨
  76(1):319–342.

Levinsohn, J. A., McLaren, Z., Shisana, O., and Zuma, K. (2011). HIV Status and Labor Market
  Participation in South Africa. Working Paper 16901, National Bureau of Economic Research.

Manda, E. L. (2010). Price Instability in the Maize Market in Malawi. PhD. Thesis, School of
 International Development Studies - University of East Anglia.

McAuslan, P. (2003). A Narrative on Land Law Reform in Uganda. Lincoln Institute of Land
 Policy - Conference Paper.

Miers, S. and Kopytoff, I. (1979). Slavery in Africa: Historical and Anthropological Perspectives.
  Univ of Wisconsin Press.

Morten, M. (2013). Temporary Migration and Endogenous Risk Sharing in Village India. Working
 Paper, Stanford.

Munshi, K. and Rosenzweig, M. (2014). Networks and Misallocation: Insurance, Migration, and
 the Rural-Urban Wage Gap. Working Paper, Cambridge and Yale University.

Piketty, T. (2014). Capital in the Twenty-First Century. Belknap Press of Harvard University
  Press, Cambridge, MA.

Reniers, G. (2003). Divorce and Remarriage in Rural Malawi. Demographic Research, 1(6):175–
  206.

                          alia-Llopis, R. (2015). Land Misallocation and Productivity. Working
Restuccia, D. and Santaeul`
  Paper, University of Toronto and Washington University in St. Louis.

Rosenzweig, M. and Wolpin, K. (1993). Credit Market Constraints, Consumption Smoothing,
  and the Accumulation of Durable Production Assets in Low-Income Countries: Investment in
  Bullocks in India. Journal of Political Economy, 101(2):223–244.

Rosenzweig, M. R. and Udry, C. (2014). Rainfall Forecasts, Weather and Wages over the Agri-
  cultural Production Cycle. NBER Working paper 19808.

Sabelhaus, J. and Groen, J. A. (2000). Can Permanent-Income Theory Explain Cross-Sectional
  Consumption Patterns? Review of Economics and Statistics, 82(3):431–438.

        alia-Llopis, R. (2008). Aggregate Effects of AIDS on Development. Working Paper,
Santaeul`
  Washington University in St. Louis.



                                               44
        alia-Llopis, R. and Zheng, Y. (2015). The Price of Growth: Consumption Insurance in
Santaeul`
  China 1989-2009. Working Paper, Washington University in St. Louis and City University of
  Hong Kong.

                                             ele Tertilt on Gender Economics. Economic
Tertilt, M. (2012). The Research Agenda: Mich´
  Dynamics Newsletter - November -2012.

Townsend, R. M. (1994). Risk and Insurance in Village India. Econometrica, 62(3):539–91.

Udry, C. (1994). Risk and Insurance in a Rural Credit Market: An Empirical Investigation in
  Nothern Nigeria. Review of Economic Studies, 61(3):495–526.

Udry, C. (1995). Risk and Saving in Northern Nigeria. American Economic Review, 85(5):1287–
  1300.

Young, A. (2005). The Gift of the Dying: The Tragedy of AIDS and the Welfare of Future
  African Generations. Quarterly Journal of Economics, (120):423–466.

Young, A. (2007). In Sorrow to Bring Forth Children: Fertility amidst the Plague of HIV. Journal
  of Economic Growth, (12):283–327.

Young, A. (2012). The African Growth Miracle. Journal of Political Economy, 6(120):696–739.

Young, A. (2013). Inequality, the Urban-Rural Gap and Migration. The Quarterly Journal of
  Economics.




                                              45
         Figure 1: Agricultural Production: Consumption Prices vs. Prices at the Gate




Note: Sample of 4,385 households in rural Malawi that sell neither maize nor tobacco and report consumption
from own production.




                                                    46
Figure 2: Density of Consumption, Income, and Wealth in Malawi, Tanzania, and Uganda (ISA
2010)




Notes : The construction of household consumption, income, and wealth is described in Section 2. All variables
have been logged.




                                                     47
       Figure 3: Joint Densities of Consumption, Income, and Wealth (Malawi ISA-2010)




               (a) Income and Wealth, Rural                                (c) Income and Wealth, Urban




            (b) Income and Consumption, Rural                           (d) Income and Consumption, Urban


Notes : The construction of household consumption, income and wealth is described in Section 2. All variables
have been logged.




                                                     48
                                          Figure 4: Life Cycle Profiles

                  Household Structure                             Adult-Equivalent Consumption and Income




                      (a) Nationwide                                                (b) Nationwide




                    (c) Rural Residency                                          (d) Rural Residency




                    (e) Urban Residency                                          (f) Urban Residency

Notes : The life cycle profiles refer to adult-equivalent consumption and directly extracted from De Magalh˜       aes
et al. (2015). They are normalized to 1 at age 25. In the left columns, KP refers to the equivalence scales used in
Krueger and Perri (2006), children are defined as household members younger than age 15, adults above the age
of 15, and DR refers to the weak dependency ratio defined as the ratio of adults to children. In the right columns,
C refers to household consumption, C/KP refers to adult-equivalent consumption that uses the KP scales, I refers
to household income, and I/A refers to income per adult. The definition of household consumption and income are
in Section 2. These life cycle profiles control for time effects using a second cross-sectional dataset for Malawi, the
Integrated Household Survey in 2004-05 (IHS2) of similar quality to our current survey (i.e, IHS3). We apply to
IHS2 data exactly the same methodology discussed in section 2 to construct household consumption and income.
          Table 1: World Development Indicators and LSMS-ISA (Current USD, 2010)



                        (a) Macro Data: World Development Indicators, 2010

                                        Malawi      Tanzania        Uganda     Thailand      Mexico             US

  Income per capita                         359               524      471           4,802    8,920 48,377
  Agricultural share (% Income)              29                28       25              12        3      1
  Consumption per capita                    257               328      376           2,577    6,023 32,783
  Rural population (%)                       84                71       85              56       22     19
  Life expectancy                            53                59       57              73       77     79



                                   (b) Micro Data: LSMS-ISA, 2010

                                               Malawi                  Tanzania                 Uganda

  Income per household                          1,384                     1,625                   1,623
                                             [1,314; 1,453]            [1,465; 1,786]          [1,358; 1,888]
  Income per capita                               343                        378                   509
                                               [328; 358]               [342; 414]              [335; 682]
  Agricultural share (% Income)                   43                         34                     23

  Consumption per household                     1,601                     1,931                   2,366
                                             [1,569; 1623]             [1,869; 1,994]          [2,240; 2,491]
  Consumption per capita                          416                        456                   589
                                               [407; 425]               [436; 475]              [545; 633]
  Rural households (%)                            82                         69                     77

  Sample size                                  12,015                     3,012                   2,337


Notes : Statistics in panel (a) are provided by the World Development Indicators at the World Bank and are based
on national accounts data. Statistics in panel (b) are produced from the ISA household surveys data provided by
the World Bank and adjusted as described in Section 2. Brackets denote 95% confidence intervals.
             Table 2: Rural and Urban Levels: Cross-Country Comparison (ISA 2010)



                                              Malawi           Tanzania            Uganda

                                          Rural    Urban    Rural    Urban     Rural    Urban

                   Consumption            1,366    2,912    1,545     2,891    1,809     4,910
                    ” (p.c.)                297      648      280       641      321     1,043
                      Nondurables         1,272    2,561    1,466     2,615    1,254     3,205
                      Durables               40      173        4        77      328     1,060

                   Income                 1,131    2,781    1,225     2,610    1,268     3,217
                     ” (p.c.)               246      618      221       578      225       676
                      Agriculture           665      247      690       204      426       132
                      Labor                 212    1,630      248     1,390      183       846
                      Business              128    1,052      178       800      534     1,843

                   Wealth                 1,309    3,976    3,361     1,760    6,148    10,256
                     Assets
                        Land                575      401    2,341     1,588    4,421     4,774
                        House               404    2,690      n.a.      n.a.   1,190     4,336
                        Land (acres)        2.3      0.4         6       2.5      4.7       1.5
                     Debt                     5       37       11         22     n.a.      n.a.

                   Sample size            9,820    2,195    2,067       945    1,809        528



Notes: All variables except land acres are in current USD. The construction of the measures of household
consumption, income, and wealth is discussed in Section 2, with further details in the online appendix. Per capita
variables are computed dividing by total household size; alternative adult-equivalent measures are discussed in
Section 5.1.
                                Table 3: Rural and Urban Inequality: Cross-country Comparison (ISA 2010)



                                                   Malawi                           Tanzania                            Uganda

                                                         mean     90th                      mean     90th                      mean     90th
             (A) Rural Residency:       Var.     Gini    med.     10th
                                                                           Var.     Gini    med.     10th
                                                                                                              Var.     Gini    med.     10th

              Consumption              0.49     0.39        1.3     6      0.49     0.37      1.3      6      0.83     0.48      1.5      10
              Income                   0.99     0.54        1.7    11      1.44     0.61      2.0     21      1.67     0.74      2.6      24
              Wealth                   1.49     0.60        2.0    16      2.95     0.76      4.3    110      2.81     0.75      3.7      52
                 Land                  1.15     0.64        2.1     -      2.25     0.79      4.5      -      2.35     0.81      4.7       -

                                                         mean     90th                      mean     90th                      mean     90th
             (B) Urban Residency:       Var.     Gini    med.     10th
                                                                           Var.     Gini    med.     10th
                                                                                                              Var.     Gini    med.     10th

              Consumption              0.59     0.44        1.5     6      0.60     0.40      1.3      7      1.05     0.48      1.4     14
              Income                   1.60     0.71        2.9    22      1.90     0.66      2.3     33      2.01     0.76      2.8     98
              Wealth                   4.52     0.84        6.8   282      4.44     0.94        -      -      4.67     0.80      6.5    369



Notes: All variables are in current USD. The measures of inequality that we study are the variance of logged variables (i.e., Var.), the Gini index, the
                                                                                                                 90th
mean-to-median ratio (i.e., mean
                            med ), and the ratio between the top 10% and bottom 10% of the distribution (i.e., 10th ). The construction of the measures of
household consumption, income, and wealth is discussed in Section 2, with further details in the online appendix.
       Table 4: Consumption, Income and Wealth by Income Partition (Malawi ISA 2010)




                        Bottom (%)                     Quintiles                     Top (%)               All
                      0-1    1-5 5-10       1st    2nd    3rd    4th     5th    10-5    5-1       1     0-100


                                                   (A) Rural Residency

    Consumption      911     738    796     832   1002   1269 1498 2218        2140     2596 4501        1364
      ” (p.c.)       297     236    258     267    312    336 367 466           473      510   905        350
    Income           -40      90    165     183    416    679 1091 3182        2397     4092 15983       1110
    Wealth           419     632    592     642    854    971 1396 2521        2331     3544 5243        1277
       Land          125     241    291     320    396    475 723 919           955     1069 1118         567


                                                  (B) Urban Residency

    Consumption 1990 1903 1708            1761 1771 2182 2677 5877             5309     8300 15902       2851
      ” (p.c.)   623 550 453               524 581 662 720 1381                1066     2028 2939         773
    Income      -121   62 169              201 577 979 1674 8979               5220    12881 61655       2478
    Wealth      1567 800 747               910 1930 1527 1888 9134             6154    16954 33881       3074
       Land       17   67   66              77 445 564 520 490                  415      521 2287         419


Notes: All variables are averages in current USD. The construction of the measures of household consumption,
income, and wealth is discussed in Section 2, with further details in the online appendix. Per capita variables are
computed dividing by total household size; alternative adult-equivalent measures are discussed in Section 5.1.
        Table 5: Saving Rate by Income, Wealth, and Land Partition (Malawi ISA 2010)




                             Bottom (%)                        Quintiles                        Top (%)             All
                          0-1     1-5 5-10       1st     2nd      3rd    4th    5th     10-5       5-1     1     0-100


                                                       (A) Rural Residency

 Saving rate :
   By income partition    n.a.   -7.20 -3.82   -3.55   -1.41    -0.87   -0.37    0.30    0.11 0.37 .0.72         -0.23
   By wealth partition   -0.79   -0.66 -0.50   -0.60   -0.25    -0.26   -0.28   -0.06   -0.11 0.09 .0.08         -0.23
   By land partition     -0.22    n.a.  n.a.   -0.26   -0.33    -0.32   -0.25   -0.09    0.00 -0.10 0.13         -0.23


                                                       (B) Urban Residency

 Saving rate:
   By income partition    n.a. -29.69 -9.11    -7.76   -2.07 -1.23 -0.60        0.35    -0.02     0.36    0.74   -0.15
   By wealth partition   -0.77 -0.56 -0.54     -0.87   -1.03 -0.38 -0.33        0.18     0.18     0.34    0.42   -0.15
   By land partition     -0.21   n.a.  n.a.     n.a.    n.a. -0.21 -0.31        0.06     0.24     0.08    0.15   -0.15


Notes: Household saving rates are defined as 1 minus the household consumption to income ratio.
   Table 6: Cross-Country Comparison: Top of the Income and Wealth Distributions (2010)


                                                 Micro Data                      Macro Data
                                       Income         Wealth                    Income (p.c.)
                 Countries        Top 1% Top 10% Top 1% Top 10%                     WDI

                 SSA:
                  Malawi             20%         50%        24%         57%               359
                  Tanzania           15%         51%        24%         72%               524
                  Uganda             31%         65%        28%         69%               471

                 Rich:
                  US                 20%         48%        34%         71%           48,377
                  Britain            15%         42%        28%         70%           38,363
                  France              9%         33%        24%         62%           40,706
                  Sweden              7%         28%        20%         59%           52,076

                 Emerging:
                  India              12%          n.a.       n.a.        n.a.          1,417
                  Indonesia          13%          n.a.       n.a.        n.a.          2,946
                  China              11%          n.a.       n.a.        n.a.          4,433
                  South Africa       17%          n.a.       n.a.        n.a.          7,175
                  Argentina          17%          n.a.       n.a.        n.a.         11,460
                  Colombia           20%          n.a.       n.a.        n.a.          6,179

Notes: The figures for rich and emerging countries are retrieved from Piketty (2014). All numbers refer to 2010,
except for Argentina, which refers to 2005. The construction of the measures of household income and wealth
for SSA countries is discussed in Section 2, with further details in the online appendix.
            Table 7: Shares of Total Consumption, Income, and Wealth by Rural and Urban Residency, Malawi (ISA 2010)



                                    (A) Rural Residency                                                    (B) Urban Residency


                    Bottom(%)               Quintiles                  Top(%)        All     Bottom(%)                 Quintiles                  Top(%)          All
                   0-1 1-5 5-10      1st 2nd 3rd 4th 5th            10-5 5-1 1    0-100     0-1 1-5 5-10      1st   2nd 3rd 4th         5th    10-5 5-1 1      0-100


   (A1) Income Partition:                                                                  (B1) Income Partition:
                                       Shares of   Total (%)                                                    Shares of   Total (%)
   Consumption       1      2   3     12   15      19 22       33     8   10 3      100       1    2     3 12       13      15 19         41     9   15 6        100
   Income            0      0   1      3    7      12 20       57    11   18 14     100       0    0     0     2     5       8 14         72    11   26 25       100
   Wealth            0      2   2     10   13      15 22       39     9   14 4      100       1    1     1     5    13      10 12         59    10   28 11       100
     Land            0      2   3     11   14      17 26       32     8    9 2      100       0    1     1     4    21      27 25         23     5    6 5        100


   (A2) Wealth Partition:                                                                  (B2) Wealth Partition:
                                       Shares of   Total (%)                                                    Shares of   Total (%)
   Consumption       1      3   4     15   16      18 21       31     7   10 2      100       2   1      3 11       14      17 19         39     9   14    4     100
   Income            1      2   3     11   15      17 20       36     8   13 3      100       1   0      2     7     8      14 16         54    12   24    9     100
   Wealth           -0      0   0      2    6      10 18       63    11   21 17     100      -0   0      0     0     1       4 10         84    12   29   32     100
     Land            0      0   0      2    7      12 21       59    11   19 13     100       0   0      0     0     1       5 20         74    24   20    8     100


   (A3) Land Partition:                                                                    (B3) Land Partition:
                 0-10                    Shares of Total (%)                               0-75                   Shares of Total   (%)
   Consumption      16      -   -     20    16 19 20           25     6    7 2      100      75    -      -     -      - 75         5     20     4    6 3        100
   Income           16      -   -     20    14 17 20           28     7    8 2      100      72    -      -     -      - 72         4     24     7    7 4        100
   Wealth            6      -   -      8     8 14 19           51    10   16 13     100      64    -      -     -      - 64         3     33     7   13 7        100
     Land            0      -   -      1     5 10 19           65    12   22 18     100       0    -      -     -      -    0       3     97    15   46 26       100


Notes: All variables are averages in current USD. The construction of the measures of household consumption, income, and wealth is discussed in Section 2,
with further details in the online appendix.
                   Table 8: Deconstructing Consumption, Income, and Wealth by Income Partition, Malawi (ISA 2010)
                                        (A) Rural Residency                                                       (B) Urban Residency

                          Bottom(%)             Quintiles                  Top(%)         All     Bottom(%)                  Quintiles                Top(%)           All
                        0-1 1-5 5-10     1st 2nd 3rd 4th         5th   10-5 5-1 1      0-100    0-1 1-5 5-10       1st    2nd 3rd 4th       5th   10-5 5-1 1        0-100

                                          Consumption Type (%)                                                       Consumption Type (%)
    Food                 67 62     65    65   67 68 68 64               68    61 47       65     51    48    53     57   62 60 59 43               51    36    22      51
     (a) Purchased       46 35     37    36   33 32 31 33               31    36 33       32     47    46    50     53   55 53 51 40               47    34    21      46
     (b) Own prod.       21 24     25    25   27 30 30 26               31    21 12       27      4     1     2      2    4    5     4    2         3     1     1       3
     (c) Received         0   2     3     4    6    7     6    5         6     4   2       6      0     1     1      1    3    3     3    1         1     1     0       2
    Clothing              2   2     2     2    2    3     3    3         3     4   3       3      2     2     2      3    3    3     3    4         4     4     3       3
    Utilities            19 23     21    20   18 16 14 12               12    12 10       15     18    24    15     16   15 14 15 15               13    17    11      14
    Other Nondurables     9 10      9     9    9 10 10 13               11    14 24       11     20    14    21     17   16 17 17 23               18    26    34      19
    School                1   1     1     1    1    1     1    2         1     2   3       1      2     1     5      2    1    2     2    6         6     7    10       3
    Health                1   1     1     1    1    1     1    1         1     1   1       1      4     8     1      2    1    1     1    2         2     2     0       1
     (a) Prevention       0   0     0     0    0    0     0    0         0     0   0       0      0     0     0      0    0    0     0    0         0     0     0       0
     (b) Treatment        1   1     1     1    1    1     1    1         1     1   1       1      4     8     0      2    1    1     1    1         2     2     0       1
    Durables              1   1     1     1    2    2     2    5         4     5 13        3      3     2     2      3    3    3     3    8         5     9    20       5
     (a) Housing own      2   2     2     2    2    2     2    2         2     2   2       2      2     2     1      1    2    2     3    2         2     2     3       2
     (b) Housing rent     0   0     0     0    0    0     0    0         0     0   0       0      0     1     1      1    1    1     1    1         1     1     1       1
     (c) Other            1   1     1     1    2    2     2    5         4     5 13        3      3     2     2      3    3    3     3    8         5     9    20       5
                        100 100   100   100 100 100 100 100            100   100 100     100    100   100   100    100 100 100 100 100            100   100   100     100

                                              Income   Sources (%)                                                       Income   Sources (%)
    Labor               -22 22     17    20      19     17 17 19        17    26 13       19    -10    41    47     53      56     59 60 50        66    52    38      53
    Agriculture          24 54     61    57      60     63 66 57        66    57 41       60    -28     2    13     16      14     19 16      8    12     7     5      10
    Fishing             114  -0     0    -1       -0     0     0   5     1     2 16        3     72    -1     0     -3       -0     0     0   0     1     0     0       0
    Business             -2   4     3     3        4     5     6 14      8    11 29       10      6    53    21     21      20     12 19 40        19    40    57      34
    Capital              -7   1     1     1        1     1     2   2     2     3   0       2     -1     2     1      1        1     0     0   0     1     0     0       0
    Transfers             1   2     3     3        2     1     1   0     0    -0   0       1     64   -12    10      2        2     3    -0   1    -0     2     0       1
    Food Gifts           -9 18     16    17      14     12     9   3     5     2   0       7     -4    15     9     11        8     6     5   1     1     0     0       2
                        100 100   100   100     100    100 100 100     100   100 100     100    100   100   100    100     100    100 100 100     100   100   100     100

                                           Wealth Portfolio (%)                                                       Wealth Portfolio (%)
    Housing              50 50     39    38   42 32 27 25               25    27 24       30     73    64    61     61   63 45 51 61               62    62 56         59
    Other Durables       10   5     4     5    5     6      7 15        11    16 39        9     17    26    26     22   13 16 19 33               31    36 37         27
    Land                 30 38     49    50   46 49 52 36               41    30 21       44      1     8     9      8   23 37 28           5       7     3   7        14
    Agric. structures     0   0     0     0    0     1      1    1       1     1   1       1      0     1     0      0    0     0      0    0       0     0   0         0
    Agric. equipment      2   1     1     2    1     2      2    3       3     3   4       2      0     0     0      4    0     0      0    0       0     0   1         1
    Fishing equipment     1   0     0     0    0     0      0    1       2     0   1       0      3     0     0      0    0     0      0    0       0     0   0         0
    Livestock             8   6     6     5    5 10 12 20               18    23 11       13      6     0     4      4    1     1      2    1       1     1   2         2
    Debt                 -0  -0    -0    -0   -0    -0     -0   -1      -1    -1  -0      -0     -0    -0    -1     -0   -0    -0     -1   -2      -1    -1  -3        -1
                        100 100   100   100 100 100 100 100            100   100 100     100    100   100   100    100 100 100 100 100            100   100 100       100


Notes: All variables are averages in current USD. The construction of the measures of household consumption, income, and wealth is discussed in Section 2,
with further details in the online appendix.
               Table 9: Age of Household Head and Inequality, Malawi (ISA 2010)

                                               (A) Rural Residency

                        (A1) Population Shares (%) by Age Groups
                                         15-24 25-34 35-44 45-54         55-64   65+
                        Population 2010    9      29      22     15       11      13

                        (A2) Average US$ by Age Groups
                                        15-24 25-34      35-44   45-54   55-64   65+
                        Consumption      1137 1379       1559    1526    1347    1008
                        Income           868    1073     1294    1327    1106    823
                        Wealth           639     965     1399    1604    1699    1484
                          Land           336     471      674     711     660    522


                                               (B) Urban Residency

                        (B1) Population Shares (%) by Age Groups
                                         15-24 25-34 35-44 45-54         55-64   65+
                        Population 2010    8      40      25     13        7      6

                        (B2) Average US$ by Age Groups
                                        15-24 25-34      35-44   45-54   55-64 65+
                        Consumption      1854 2390       3385    3684    3226 2728
                        Income           939    1858     3401    3623    2023 2850
                        Wealth           933    1488     3413    6247    6661 3409
                          Land           311     389      458     233     928  371


Notes: All variables are averages in current USD. The construction of the measures of household consumption,
income, and wealth is discussed in Section 2, with further details in the online appendix.
          Table 10: Household Structure, Children, and Inequality, Malawi (ISA 2010)

                                         (A) Rural Residency

                   (A1) Population Shares (%) by Number of Children and Dependency Ratios
                                           Number of Children         Dependency Ratio
                                      0      1     2      3     ≥4 DR<1        DR≥1
                   Population 2010 18       19     22    19     22      45      55

                   (A2) Average US$ by Number of Children and   Dependency Ratios
                                         Number of Children           Dependency Ratio
                                     0     1      2      3       ≥4 DR<1        DR≥1
                   Consumption     1058 1242 1409 1417          1631 1313        1406
                   Income           905 975 1073 1212           1342 1114        1107
                   Wealth          1043 1237 1303 1274          1482 1397        1178
                     Land           444 539 592 622              620   589        548


                                         (B) Urban Residency

                   (B1) Population Shares (%) by Number of Children and Dependency Ratios
                                           Number of Children         Dependency Ratio
                                      0      1     2      3     ≥4 DR<1        DR≥1
                   Population 2010 24       21     21    20     15      56      44

                   (B2) Average US$ by Number of Children and   Dependency Ratios
                                         Number of Children           Dependency Ratio
                                     0     1      2      3       ≥4 DR<1        DR≥1
                   Consumption     2631 3045 3109 2699          2761 3244        2349
                   Income          2041 2799 3306 2214          1880 3125        1651
                   Wealth          2573 3681 4089 2552          2256 4023        1863
                     Land           377 367 379 460              561   452        377


Notes: All variables are averages in current USD. The construction of the measures of household consumption,
income, and wealth is discussed in Section 2, with further details in the online appendix.
                                                    Table 11: Migration and Inequality, Malawi (ISA 2010)

                            (A) Rural Residency                                                                  (B) Urban Residency

 (A1) Population Shares (%) by Migration History                                             (B1) Population Shares (%) by Migration History:
                  Not Migrant Migrant:         R-to-R   U-to-R     R-to-U      U-to-U        Not Migrant Migrant:       R-to-R      U-to-R      R-to-U      U-to-U
 Population 2010       65           35           31       4           -           -              22            78          -           -          60          18

 (A2) Averages US$ by Migration History:                                                     (B2) Averages US$ by Migration History
                  Not Migrant    Migrant      R-to-R    U-to-R     R-to-U      U-to-U        Not Migrant    Migrant    R-to-R       U-to-R      R-to-U      U-to-U
 Consumption         1278          1522        1505      1645         -           -             2173         3047          -           -         2782        3929
 Income              1048          1224        1209      1337         -           -             1670         2711          -           -         2206        4389
 Wealth              1296          1241        1245      1208         -           -             2711         3180          -           -         2457        5583
   Land               588           526         530       501         -           -              849         294           -           -         292          302

 (A3) Population Shares (%) by Consumption, Income, and Wealth Partitions:                   (B3) Population Shares (%) by Consumption, Income, and Wealth Partitions:
                      Bottom (%)                Quintiles               Top (%)        All      Bottom (%)                 Quintiles             Top (%)           All
                     1     1-5 5-10 1st 2nd 3rd 4th 5th 10-5 5-1 1                  0-100      1     1-5 5-10 1st 2nd 3rd 4th 5th 10-5 5-1 1                    0-100
                                       Consumption Quintiles                                                       Consumption Quintiles
 Not migrant       77       76    70   72     66 67 61 59            58 55 41          65     35      40     32    29    27 27 17 13           15 13       1       22
    R-to-R         18       22    27   25     30 29 34 36            36 39 49          31      -       -      -     -     -     -      -  -           -    -         -
    U-to-R           4       2     3    3      4    4      4   6       5     6  9       4      -       -      -     -     -     -      -  -           -    -         -
    R-to-U           -       -     -     -     -     -      -  -       -     -  -       -     51      49     56    54    60 62 63 59           62 49 49            60
    U-to-U           -       -     -     -     -     -      -  -       -     -  -       -     14      11     12    18    13 11 20 28           22 39 50            18
                  100      100 100 100 100 100 100 100 100 100 100                    100    100     100 100 100 100 100 100 100 100 100 100                      100
                                           Income Quintiles                                                           Income Quintiles
 Not migrant       70       75    69   71     67 65 63 59            57 55 54          65     31      23     24    25    30 19 22 16           17 15       6       22
    R-to-R         28       23    28   26     29 31 33 35            37 39 36          31      -       -      -     -     -     -      -  -      -    -    -         -
    U-to-R           2       2     4    3      4    4      4   6       6     6 10       4      -       -      -     -     -     -      -  -      -    -    -         -
    R-to-U           -       -     -     -     -     -      -  -       -     -  -       -     52      65     65    61    56 65 60 57           54 55 39            60
    U-to-U           -       -     -     -     -     -      -  -       -     -  -       -     18      12     12    14    14 16 18 27           29 29 56            18
                  100      100 100 100 100 100 100 100 100 100 100                    100    100     100 100 100 100 100 100 100 100 100 100                      100
                                           Wealth Quintiles                                                           Wealth Quintiles
 Not migrant       38       43    60   59     68 67 66 65            65 64 72          65     10      13     17    14    16 24 29 28           46 24       3       22
    R-to-R         55       51    34   36     29 30 29 31            31 32 24          31      -       -      -     -     -     -      -  -      -    -    -         -
    U-to-R           6       6     6    5      3    3      5   4       4     4  5       4      -       -      -     -     -     -      -  -      -    -    -         -
    R-to-U           -       -     -     -     -     -      -  -       -     -  -       -     59      53     71    67    67 59 58 48           36 44 49            60
    U-to-U           -       -     -     -     -     -      -  -       -     -  -       -     31      34     12    19    17 17 13 24           18 33 48            18
                  100      100 100 100 100 100 100 100 100 100 100                    100    100     100 100 100 100 100 100 100 100 100 100                      100


Notes: Conditional on migration history, R-to-R denotes rural to rural migration, R-to-U denotes rural to urban migration, U-to-R denotes urban to
rural migration, and U-to-U denotes urban to urban migration. All variables are averages in current USD. The construction of the measures of household
consumption, income, and wealth is discussed in Section 2, with further details in the online appendix.
                                     Table 12: Adult Educational Attainment and Inequality, Malawi (ISA 2010)

                            (A) Rural Residency                                                                    (B) Urban Residency

 (A1) Population Shares (%) by Educational Attainment                                           (B1) Population Shares (%) by Educational Attainment
                        No Education    Primary Dropout     Primary       Secondary and More     No Education     Primary Dropout       Primary      Secondary and More
 Population 2010             26                 45             21                 8                    8                 23                35                34

 (A2) Averages US$ by Educational Attainment                                                    (B2) Averages US$ by Educational Attainment
                       No Education     Primary Dropout     Primary       Secondary and More     No Education    Primary Dropout        Primary       Secondary and More
 Consumption               981               1307            1601                2360                1579             1851               2393                4329
 Income                    743               1035            1316                2255                1073             1105               1805                4469
 Wealth                    872               1309            1618                1535                1417             1627               2254                5332
   Land                    434                621             628                 532                 213              533                542                 261

 (A3) Population Shares (%) by Consumption, Income, and Wealth Partitions                       (B3) Population Shares (%) by Consumption, Income, and Wealth Partitions
                         Bottom (%)               Quintiles               Top (%)         All     Bottom (%)                Quintiles             Top (%)              All
                         1 1-5 5-10 1st 2nd          3rd 4th 5th 10-5 5-1 1            0-100      1 1-5 5-10 1st 2nd           3rd 4th 5th 10-5 5-1 1              0-100
                                          Consumption Quintiles                                                      Consumption Quintiles
 Av. Schooling Years     2    2     3    3      4      5     5   7       7     7 10        5      3    3      6     6     8      6 10 12        13 13 16                9
 No education           55 54      48   45     31     24 20 12           9 11     6       26     66 19       17    21     7      4     5    2    1     2     0          8
 Primary dropouts       34 38      39   41     47     47 46 41          41 40 25          45     11 36       38    33    28     36 12       9    5     5     0         23
 Primary                10    7    12   12     18     22 25 30          34 26 30          21     23 34       32    34    45     35 38 22        19 16 18               35
 Secondary or more       1    1     1    2      4      6     9 17       16 23 39           8      0 11       13    12    19     24 45 67        76 77 82               34
                      100 100 100 100 100           100 100 100 100 100 100              100    100 100 100 100 100           100 100 100 100 100 100                 100
                                              Income Quintiles                                                          Income Quintiles
 Av. Schooling Years     4    4     3    4      4      5     5   7       7     8  8        5      6    6      7     7     8      9     9 12     13 13 15                9
 No education           42 38      40   38     34     27 20 14          15     7 13       26      9 13        5    13    10      8     8    2    1     5     0          8
 Primary dropouts       27 42      42   43     45     46 48 41          40 39 27          45     49 37       49    38    32     21 17       9    7 11        0         23
 Primary                21 16      15   16     17     23 25 27          28 27 29          21     24 38       28    38    38     39 37 22        24 12 25               35
 Secondary or more      10    4     3    4      4      5     7 18       17 27 30           8     18 12       18    11    20     33 38 67        68 72 75               34
                      100 100 100 100 100           100 100 100 100 100 100              100    100 100 100 100 100           100 100 100 100 100 100                 100
                                              Wealth Quintiles                                                          Wealth Quintiles
 Av. Schooling Years     6    5     5    4      4      5     5   6       6     6  6        5      7    5      7     7     9      9     9 11      9 13 15                9
 No education           15 27      30   32     33     27 25 15          15 13     9       26      8 26        9    11    10      7     6    6    6     4     0          8
 Primary dropouts       43 41      41   42     43     47 45 46          47 44 47          45     38 25       31    30    20     23 28 16        27 12        0         23
 Primary                25 20      19   18     19     20 22 28          26 30 35          21     38 44       44    39    36     34 35 30        30 19 20               35
 Secondary or more      16 12       9    8      5      6     8 10       12 13 10           8     16    5     15    20    33     36 31 49        37 66 80               34
                      100 100 100 100 100           100 100 100 100 100 100              100    100 100 100 100 100           100 100 100 100 100 100                 100


Notes: All variables are averages in current USD. The construction of the measures of household consumption, income, and wealth is discussed in Section 2,
with further details in the online appendix.
                                                            Table 13: Risk and Inequality, Malawi (ISA 2010)

                                     (A) Rural Residency                                                                                    (B) Urban Residency

 (A1) Population Shares (%) by Risk:                                                                            (B1) Population Shares (%) by Risk
                                   No. of Shocks            Idiosyncratic Shocks         Aggregate Shocks               No. Shocks                 Idiosyncratic shocks       Aggregate Shocks
                                                                    Type                       Type                                                        Type                     Type
                          No shock    Both   Ind.   Agg.    Sick     Theft Death       Rain     AgriC FoodP     No shock Both Ind. Agg.            Sick     Theft Death      Rain    AgriC FoodP
 Population 2010             29        22      6     43      12         6      5        43       33       26       61       11     13    15         6         6       3       9        9       18

 (A2) Averages US$ by Risk:                                                                                     (B2) Averages US$ by Risk
                                  No. of Shocks              Idiosyncratic shocks    Aggregate Shocks (Agg.)             No. Shocks                Idiosyncratic shocks        Aggregate Shocks
                                                                     Type                     Type                                                         Type                      Type
                          No shock    Both Ind. Agg.        Sick      Theft Death     Rain     AgriC FoofP      No shock Both Ind.       Agg.     Sick      Theft Death      Rain     AgriC FoodP
 Consumption               1369       1346 1773 1315        1468      1751 1234       1193      1451    1229     3169    2017 2933       2122     2323      3132 2469        1999     2164    1977
 Income                    1213       1013 1549 1032        1082      1426     811     928      1157    1006     2763    1846 2525       1751     1618      2218 1303        1619     1995    1653
 Wealth                    1166       1259 1639 1315        1362      1965 1073       1192      1384    1033     3743    1959 2947       1296     2059      3678 3061        1923     2243    1410
   Land                     506        622  575 578          715       810     482     541       638    469       364    629  637         285      844       663     475     493       791     378

 (A3) Population Shares (%) by Consumption, Income and Wealth Partitions:                                       (B3) Population Shares (%) by Consumption, Income, and Wealth Partitions:
                                Bottom (%)                    Quintiles                  Top (%)          All        Bottom (%)                     Quintiles                 Top (%)             All
                                  1   1-5 5-10      1st 2nd 3rd         4th   5th    10-5 5-1    1     0-100           1     1-5 5-10     1st 2nd 3rd         4th  5th   10-5 5-1       1      0-100
                                                    Consumption Quintiles                                                                  Consumption Quintiles
 No shocks                       26    30     28    28     31 30         29     29    29   26     36      29          75      66    36     48    59 62         66    70    68 84          78      61
 Both shocks                     42    23     26    25     20 22         21     23    23   23     17      22           7      11    19     17    14 14          8      5    4    4         6      12
 Idiosyncratic shocks only        6     4      3      4     3    6        6      8     8   12     14       6           5       9    20     12    13    9       14    17    23    7        11      13
        Illness                  18    10     11    11     11 13         12     15    16   18     11      12           0      11     5      7     5    9        4      5    4    4         0       6
        Theft                     7     2      3      4     4    6        6      9     7   12     11       6           0       4     4      5     4    3        8      7   16    3         6       6
        Death                    13     7      5      6     4    3        5      4     5    3      1       4           2       4     8      4     4    2        2      4    6    1         0       3
 Aggregate shocks only           27    43     44    43     46 42         44     40    39   39     33      43          13      13    25     23    15 14         12      9    5    5         5      15
        Rain/Drought             55    54     52    51     48 44         41     32    33   30     14      43           1      21    16     15    11 10          7      4    2    3         6       9
        High Ag. Costs           29    28     32    31     32 31         34     39    39   43     39      33           7       6    21     14     6 13          7      6    6    4         0       9
        High Food Price          51    34     32    31     27 26         23     22    23   21     14      26          19      14    26     29    19 19         13      9    4    5         0      18
                                                        Income Quintiles                                                                      Income Quintiles
 No shocks                       41    27     30    30     26 29         30     33    32   34     36      29          91      69    59     67    60 57         55    65    60 71          84      61
 Both shocks                     22    21     20    22     23 23         23     19    23   17     12      22           1      10    11     11    13 10         16      9    7    7         6      12
 Idiosyncratic shocks only        4     6      3      4     5    5        6      8     7   11     12       6           0      10    16     10    12 13         15    14    16 14          10      13
        Illness                  17    11     11    12     12 13         13     13    15   13      6      12           1       5     4      4     7    5        9      6    5    3         0       6
        Theft                     4     3      3      4     5    6        7      8     9    6      7       6           0       9     3      6     4    7        6      4    6    3         6       6
        Death                     4     7      4      5     5    4        5      3     2    3      0       4           0       0    12      5     4    2        2      3    3    1         0       3
 Aggregate shocks only           33    46     47    44     46 44         41     39    38   38     39      43           9      10    14     12    15 20         14    12    16    9         0      15
        Rain/Drought             46    54     52    53     50 43         38     31    31   26     34      43           5       7     9     10     8 12          9      8   10    4         0       9
        High Ag. Costs           15    18     25    24     33 36         39     36    37   35     31      33           1       4     3      4     5 12         16    10     9    6         0       9
        High Food Price          25    29     26    27     28 27         25     21    22   16     23      26           5      18    17     17    21 16         23    12    13    6         6      18
                                                        Wealth Quintiles                                                                      Wealth Quintiles
 No shocks                       39    43     33    33     29 29         27     30    28   30     22      29          61      51    73     64    57 59         63    61    59 71          94      61
 Both shocks                     31    22     29    25     23 21         22     21    21   20     28      22          15      18     7     13    11 13          9    12    12 10           0      11
 Idiosyncratic shocks only        3     7      4      6     5    5        6      7     6    6      9       6          11       3     4      8    15 12         13    17    24 15           6      13
        Illness                  11    11     12    12     13 12         13     13    13   13     15      12          14       7     4      5     6    7        5      6   11    5         0       6
        Theft                    11     6      4      5     4    5        6      8    10    8     13       6           0       1     2      4     5    7        3      8    4    8         6       6
        Death                     1     6      5      5     5    4        4      3     3    3      6       4           6       1     1      2     4    2        4      4    7    7         0       3
 Aggregate shocks only           27    28     35    36     43 46         46     43    45   44     41      43          13      28    16     14    17 16         16    10     5    4         0      14
        Rain/Drought             27    24     42    38     48 48         45     38    39   38     34      43           4       7     3      5    11 13         10      9    5 10           0       9
        High Ag. Costs           10    19     33    31     33 32         35     37    40   36     39      33           2       2     3      2     6 13         14    10    13    6         0       9
        High Food Price          38    30     36    33     28 23         24     20     2   17     20      26          23      44    19     23    22 18         13    14    12    9         0      18


Notes: “AgriC” stands for costs of agricultural inputs and “FoodP” stands for the price of food. All variables are averages in current USD. The construction
of the measures of household consumption, income, and wealth is discussed in Section 2, with further details in the online appendix. The construction of
the indicators for shocks reported by the households is discussed in Section 5.4.
                                            Table 14: Insurance Mechanisms and Inequality, Malawi (ISA 2010)
                                  (A) Rural Residency                                                                                  (B) Urban Residency
 (A1) Population Shares (%) by Insurance Mechanisms                                                          (B1) Population Shares (%) by Insurance Mechanism
                                                   Self-Insurance                     Mutual Insurance                                   Self-Insurance                       Mutual Insurance
                         No Insurance Save Diet Ext. Int. Credit        Sell   Fam.    Gov NGO Child.        No Insurance Save Diet Ext. Int. Credit Sell              Fam.    Gov NGO Child.
 Population 2010              35        19     2     2       2    1      2      7        0      0      0          22         9      3      0       1     1     1        2        0      0      0

 (A2) Averages US$ by Insurance Mechanism                                                                    (B2) Averages US$ by Insurance Mechanism
                                                  Self-Insurance                    Mutual Insurance                                    Self-Insurance                        Mutual Insurance
                       No Insurance   Save Diet    Ext. Int. Credit Sell       Fam. Gov NGO Child.           No Insurance Save Diet Ext. Int. Credit            Sell   Fam.    Gov NGO Child.
 Consumption               1318       1583 1136    1135 992 1388 1558          1160 1529 941 1135                2362      2751 1717 2063 3027 1787            2454    1911    1032 1253 1620
 Income                    983        1351 902     719 972 1088 1365            770 1752 1198 1005               1982      3124 772 1862 1722 1291             1529    1268     289    911     586
 Wealth                    1343       1512 1089     711 862 1228 1461          1022 1687 535 2189                2096      2443 969 3264 2278 1876             1330    1706     599    469     132
   Land                     609       631 588      358 428       638 630        509   454    248 1079             669       346 111 305 522            566      344     393      0      42      55

 (A3) Population Shares (%) by Consumption, Income, and Wealth Partitions                                    (B3) Population Shares (%) by Consumption, Income, and Wealth Partitions
                            Bottom (%)                    Quintiles                 Top (%)            All      Bottom (%)                     Quintiles                  Top (%)               All
                          1      1-5 5-10      1st 2nd 3rd          4th 5th    10-5    5-1    1     0-100     1      1-5 5-10       1st 2nd      3rd      4th 5th    10-5    5-1      1      0-100
                                                   Consumption Quintiles                                                                Consumption Quintiles
 No Insurance            36       35    38      37     37 31         36   33     34      34    22      35    22       13      45     29     25    23       19   18     21      9       11       23
 Self-Insurance          30       25    24      25     24 30         28   32     29      35    39      28     1       15      15     19     14    14       14   11     11      7       11       15
       Own Savings (%) 15         16    13      15     15 19         20   25     22      29    29      19     1       11       4      7      7    10        8    8     11      6        6        8
       Diet (%)           0        2     3       3      2     3       2    1      1       1     2       2     0         0     10      9      3     0        3    1      0      0        0        3
       Labor Ext. (%)    11        3     4       3      2     3       2    2      2       0     1       2     0         2      1      1      0     0        0    0      0      0        0        0
       Labor Int. (%)     4        1     2       2      2     1       1    1      2       0     1       2     0         1      1      1      0     1        1    0      0      0        5        1
       Credit (%)         0        1     1       1      1     1       1    1      1       1     0       1     0         2      0      1      3     2        1    0      0      0        0        1
       Sell Assets (%)    0        3     1       2      2     3       2    2      2       3     7       2     0         0      0      1      2     1        2    0      1      0        0        1
 Mutual Insurance         9       10    11      10      8     8       6    6      7       5     3       8     2         5      4      5      4     3        2    1      0      0        0        3
       Family Help (%)    8        8    11       9      7     8       6    5      6       4     2       7     2         5      3      3      3     3        1    1      0      0        0        2
       Gov. Help (%)      1        1     0       0      0     0       0    1      1       1     1       0     0         0      1      1      0     0        0    0      0      0        0        0
       NGO Help (%)       0        0     0       0      0     0       0    0      0       0     0       0     -         -      -      0      0     0        0    0      -      -        -        0
       Send Child. (%)    0        0     0       0      0     0       0    0      0       0     0       0     -         -      -      1      1     0        1    0      -      -        -        1
                                                     Income Quintiles                                                                      Income Quintiles
 No Insurance            30       44    42      40     38 37         31   29     28      24    34      35     5       10      21     17     24    26       26   20     24     18        5       23
 Self-Insurance          19       23    19      21     26 27         32   34     35      39    27      28     5       11      13     13     15    14       17   14     14     12       12       15
       Own Savings (%) 14         11    11      12     16 17         23   26     27      32    23      19     4         8      3      6      6     8       11   10     10     11       12        8
       Diet (%)           2        2     3       2      3     3       1    1      1       1     1       2     0         3      5      5      5     4        2    0      1      0        0        3
       Labor Ext. (%)     1        5     2       3      3     3       2    1      1       0     1       2     0         0      0      0      1     0        1    0      0      0        0        0
       Labor Int. (%)     0        1     0       1      3     1       1    1      0       1     1       2     1         0      2      1      0     1        1    1      2      0        0        1
       Credit (%)         0        2     1       1      1     1       1    1      1       1     0       1     0         0      2      1      3     1        0    2      0      0        0        1
       Sell Assets (%)    3        2     1       1      2     2       2    3      6       3     1       2     0         0      0      1      1     1        2    1      1      0        0        1
 Mutual Insurance         9        6     9      10      9     8       7    5      5       3     3       8     0         0      8      4      3     4        2    2      1      0        0        3
       Family Help (%)    8        6     8       9      8     7       7    4      5       2     2       7     0         0      2      2      2     3        2    2      0      0        0        2
       Gov. Help (%)      0        0     1       1      0     1       1    0      0       0     1       0     0         0      2      1      0     0        0    0      0      0        0        0
       NGO Help (%)       1        0     0       0      0     0       0    0      0       1     0       0     -         -      -      0      0     0        0    0      -      -        -        0
       Send Child. (%)    0        0     0       0      0     0       0    0      0       0     0       0     0         0      3      1      0     1        0    0      0      0        0        1
                                                     Wealth Quintiles                                                                      Wealth Quintiles
 No Insurance            27       29    31      33     36 35         36   36     34      33    41      35    24       29      15     19     22    24       26   21     32     14        6       23
 Self-Insurance          27       20    25      25     26 29         30   30     32      32    26      28    13       17      11     15     17    13       13   16      8     12        0       15
       Own Savings (%) 17         12    13      13     17 19         22   22     26      23    22      19    10         1      5      6      8     8        8   11      6      9        0        8
       Diet (%)           6        2     4       3      2     2       2    1      1       2     3       2     2       14       4      6      4     2        1    2      0      0        0        3
       Labor Ext. (%)     1        3     2       3      3     3       2    1      0       1     0       2     0         0      0      0      1     1        0    0      0      0        0        0
       Labor Int. (%)     1        1     3       2      1     2       1    1      0       1     1       2     1         0      0      0      2     0        1    1      0      1        0        1
       Credit (%)         1        2     1       1      1     1       1    1      1       2     0       1     0         0      0      0      2     1        1    2      1      0        0        1
       Sell Assets (%)    0        0     2       1      2     2       2    3      4       3     1       2     0         2      2      2      1     0        2    0      0      1        0        1
 Mutual Insurance         9        9    11      10      9     8       7    5      5       5    11       7     5         2      3      3      4     5        1    2      0      3        0        3
       Family Help (%)    9        8    10       9      8     7       6    4      4       4    10       7     5         2      1      2      2     4        1    2      0      3        0        2
       Gov. Help (%)      0        0     0       0      1     1       0    0      1       0     1       0     -         -      -      0      0     0        0    0      -      -        -        0
       NGO Help (%)       0        0     1       1      0     0       0    0      0       0     0       0     -         -      -      0      0     0        0    0      -      -        -        0
       Send Child. (%)    0        0     0       0      0     0       0    0      0       1     0       0     0         0      2      2      1     0        0    0      0      0        0        1

Notes: All variables are averages in current USD. The construction of the indicators for insurance mechanisms is discussed in Section 5.4.
                                     Table 15: Capital Market Incompleteness and Inequality, Malawi (ISA 2010)

                                 (A) Rural Residency                                                                         (B) Urban Residency

 (A1) Population Shares (%) by Ability to Borrow                                                       (B1) Population Shares (%) by Ability to Borrow
                                      Applied                    Not Applied                                  Applied                         Not Applied
                             Got Loan Loan Denied         Needed      Not needed                       Got Loan Loan Denied         Needed         Not Needed
 Population 2010                13             7            56             25                             20            7             40               33

 (A2) Averages US$ by Ability to Borrow                                                                (B2) Averages US$ by Ability to Borrow
                                      Applied                    Not Applied                                  Applied                        Not Applied
                              Got Loan Loan Denied        Needed      Not needed                       Got Loan Loan Denied          Needed       Not needed
 Consumption                    1635         1391          1237           1481                           2884         2179            2145           3781
 Income                         1390         1085          912            1358                           2933         1299            1359           3682
 Wealth                         1566         1200          1089           1573                           3333         1888            1509           5117
   Land                          834          558           516            550                            420         583              310            507

 (A3) Population Shares (%) by Consumption, Income, and Wealth Partitions                              (B3) Population Shares (%) by Consumption, Income, and Wealth Partitions
                               Bottom (%)               Quintiles                Top (%)         All     Bottom (%)                Quintiles                 Top (%)            All
                               1 1-5 5-10        1st 2nd 3rd 4th 5th          10-5 5-1 1      0-100      1 1-5 5-10         1st 2nd 3rd 4th 5th        10-5     5-1   1      0-100
                                                  Consumption Quintiles                                                        Consumption Quintiles
 App. Accepted (%)             8    7     7        8   10 12 15 18             19    17 27       13     14 15       22      24    22 20 12 23             22     19    32       20
     Median i > 0             14 11      10      10     7     8     7     5     5     7   3       7      -    7     25        5    5     5    3      2     2      2     5        4
     Length loan m. i > 0      7    9     4        6    6     4     5     6     6     6 18        5      - 12        2      10     5     6    6      8    11     14     4        8
     Length loan m. i ≤ 0      4    6     2        2    4     4     2     3     3     3   6       3      -    5      1        4    4     5 10        6    23     23     -        5
     Borrowing US$ i > 0      34    7    22      28    31 36 27 96             71    96 432      41      - 66       33        7 134 68 263 333          302 1345 672           124
     Borrowing US$ i ≤ 0       7 15      10        8   14 31 17 21             21    26 185      15      - 14        3      13    35 27 202 201         269 2158        -       42
 App. Denied (%)              11 10       5        7    5     6     7     7     6    10   3       7      3    5      4        8    7     9    8      3     4      1     1        7
 Not App.: Needed (%)         58 60      61      60    60 59 56 46             48    40 24       56     76 65       59      49    46 45 36 23             23     13     7       40
 Not App.: No Need (%)        23 24      26      25    25 23 23 29             36    46 28       25      7 15       15      19    25 26 44 52             51     67    60       33
                             100 100 100        100 100 100 100 100           100   100 100     100    100 100 100         100 100 100 100 100          100 100 100            100
                                                     Income Quintiles                                                            Income Quintiles
 App. Accepted (%)             6    7     9        9    9 14 14 17             15 19 16          13     21 20       26      21    22 20 17 21             22     27    35       20
     Median i > 0               -   6    10      13    10     7     6     4     4   4   3         7      2    6      8        4    5     5    3      2     2      2     2        4
     Length loan m. i > 0       -   6     5        4    8     4     5     6     7   6   6         5     15    4      6        6   10     4    7      9     9      8    16        8
     Length loan m. i ≤ 0      4    1     3        3    3     4     2     4     3   4   -         3      -    4      4        4    5     5    5 16         1     23     -        5
     Borrowing US$ i > 0        - 22     15      15    21 36 74 74             71 100 133        41     68 135      34      34    21 134 34 329         269 484 2698           124
     Borrowing US$ i ≤ 0       7    5     7        7    8 22 15 28             26 28    -        15      - 14       13      14    35 30 132 483           17 2158       -       42
 App. Denied (%)               5    7     8        7    6     6     7     7     7   6   4         6      2    7      9      10     7     7    9      2     2      2     0        7
 Not App.: Needed (%)         54 68      62      63    63 59 55 41             44 33 28          56     40 37       46      47    49 42 39 23             26     14     7       40
 Not App.: Not Needed (%) 35 19          21      22    22 21 25 35             35 42 52          25     37 36       18      23    22 35 53 56             51     57    58       33
                             100 100 100        100 100 100 100 100           100 100 100       100    100 100 100         100 100 100 100 100          100 100 100            100
                                                     Wealth Quintiles                                                            Wealth Quintiles
 App. Accepted (%)            31 17      10      12    11 10 13 16             19    20 15       13     39 12       14      22    19 21 17 23             26     28    17       20
     Median i > 0              2    7     6        6   10 10        8     4     5     4 12        7      4     -     7        7    5     3    3      2     2      2     5        4
     Length loan m. i > 0     12    5     7        7    4     5     4     6     8     6   4       5      9     -     3        5   10     9    5      9    16     12     4        8
     Length loan m. i ≤ 0      2    4     3        3    3     5     4     3     3     3   4       3      5     -     5        4    6     1 10        7     6     23     -        5
     Borrowing US$ i > 0     295 31 106          43    36 37 28 71             62    50 133      41    202     - 197       129    47 35 34 329          672 672 1008           124
     Borrowing US$ i ≤ 0       8 32      14      14    14 22 14 25             37    26 41       15     30     -    52      30    68 17 67 201          201 2158        -       42
 App. Denied (%)               6    6     8        6    6     7     7     7     7     7   4       6      5 16       10        9    6     6    7      6    10      3     0        7
 Not App.: Needed (%)         29 55      58      58    60 57 57 48             47    41 39       56     37 67       43      45    43 43 44 24             17     15    16       40
 Not App.: Not Needed (%) 33 22          25      24    23 25 23 29             26    32 41       25     19    5     34      24    32 29 32 48             47     54    67       33
                             100 100 100        100 100 100 100 100           100   100 100     100    100     - 100       100 100 100 100 100          100 100 100            100


Notes: All variables are averages in current USD. The construction of the variables on the credit market is discussed in Section 5.4.
                                                 Table 16: Societal Systems and Inequality, Malawi (ISA 2010)

                                  (A) Rural Residency                                                                          (B) Urban Residency

 (A1) Population Shares (%) by Societal System                                                       (B1) Population Shares (%) by Societal Systems
                               North                       Center                   South                         North                       Center                       South
                         Patri Mixed Matri         Patri   Mixed Matri      Patri   Mixed   Matri         Patri Mixed Matri            Patri Mixed Matri           Patri   Mixed    Matri
 Population 2010          12     0        0         18      17    6          5        8      33            11       0       0           10      19   12             1       25       21

 (A2) Averages US$ by Societal   System                                                              (B2) Averages US$ by Societal System
 Consumption           1403       1718 1278        1603    1517   1568      799     1204    1234          2316 5675 4489               2857   2156   3876          1443    3296     2619
 Income                1351       1688  863        1414    1322   1294      774      760     854          1987 5726 7310               2309   1739   5796           943    2495     1452
 Wealth                1445       1949  133        1583    1372   1331      690     1132    1119          1823 3735 4129               2396   2049   8336           508    2500     2666
   Land                 524        838   0          830     605   627       261      447     488           291    197     62           692    1077    392           105    101       186

 (A3) Population Shares (%) by Consumption, Income, and Wealth Partitions                            (B3) Population Shares (%) by Consumption, Income, and Wealth Partitions
                      Bottom (%)                   Quintiles                  Top (%)          All       Bottom(%)                    Quintiles                    Top(%)              All
                   0-1    1-5 5-10       1st 2nd     3rd     4th 5th 90-95 95-99 99-100     0-100    0-1    1-5 5-10        1st 2nd     3rd     4th 5th 90-95 95-99 99-100          0-100
                                              Consumption Quintiles                                                              Consumption Quintiles
 North: Patri        5      9     11     10    11     12      15      14   15    11     7      12      0     14      10      14   11     11      10      8     6        8       0      11
        Mixed        0      0      0       0    0      0       0       0    1      0    0       0      0      0       0       0    0      0       1      1     1        2       0       0
        Matri        0      0      0       0    0      0       0       0    0      0    0       0      0      0       0       0    0      0       1      1     1        2       0       0
 Center: Patri      11     12     11     13    16     17      19      24   23    29    31      18     28     17       3       8    8     10      11     11    17        7       0      10
        Mixed       16     16     16     17    15     15      16      21   23    21    34      17      5     21      41      27   21     21      16      9     3        4      14      19
        Matri        0      2      2       4    5      6       7       7    9    10     5       6      5      4      13       9   13     10      11     19    15      19       50      12
 South: Patri       36     19     10     12     6      4       3       2    2      1    0       5      0     10       1       3    1      1       1      0     1        0       0       1
        Mixed       12     10     11     10    11      9       7       6    5      7    8       8     19     15      12      16   20     26      30     33    35      36       26      25
        Matri       20     32     39     35    36     36      33      25   23    21    16      33     43     20      21      23   25     18      20     19    21      22       10      21
                  100    100     100    100 100      100     100    100   100   100   100     100    100    100     100    100 100      100     100    100   100     100      100     100
                                                 Income Quintiles                                                                   Income Quintiles
 North: Patri        5      4      5       6    9     14      16      17   20    17     5      12      9     12       7      10   10     10      13     10     8        4       7      11
        Mixed        0      0      0       0    0      0       0       0    0      1    0       0      0      1       0       0    0      0       0      1     1        1       3       0
        Matri        0      0      0       0    0      0       0       0    0      0    0       0      1      0       0       0    0      0       0      1     0        2       2       0
 Center: Patri      15      7     11     10    14     16      22      28   26    34    24      18      4      2       7       6    9     11      11     12    11      12        5      10
        Mixed       31     15     14     14    13     17      19      21   20    19    26      17      9      3      20      12   19     22      31     12    10      16        8      19
        Matri        1      2      2       3    5      6       8       7   10      6    9       6     11     12       5       7   10     11       9     24    25      27       42      12
 South: Patri       19     14      9     11     7      3       3       3    4      1    4       5      0      1       0       2    2      2       1      0     1        0       0       1
        Mixed       10     17     14     14    11      8       5       4    3      4    9       8     30     26      30      32   25     22      22     24    28      23       34      25
        Matri       20     40     44     42    41     36      27      19   16    18    22      33     37     43      30      31   24     21      14     17    16      14        0      21
                  100    100     100    100 100      100     100    100   100   100   100     100    100    100     100    100 100      100     100    100   100     100      100     100
                                                 Wealth Quintiles                                                                   Wealth Quintiles
 North: Patri       10     11      9     10    11     11      13      16   19    19     9      12     15      8      12      10   10     13      11      9     7      10        2      11
        Mixed        1      0      0       0    0      0       0       0    0      0    1       0      0      0       0       0    0      1       1      1     1        1       0       0
        Matri        1      1      0       0    0      0       0       0    0      0    0       0      0      0       1       0    0      0       0      1     1        1       0       0
 Center: Patri      18     17     14     15    15     15      20      24   23    25    27      18     15      8       8      10    8     11       9     11     6      13        0      10
        Mixed       13     22     23     19    16     17      16      17   16    16    27      17     16      8      21      17   16     17      25     20    30      11        0      19
        Matri        1      4      6       5    7      6       5       6    6      6    9       6     16     11       4      10   12     14      10     16    12      20       75      12
 South: Patri        9      4      8       7    9      6       3       2    2      2    1       5      0      2       1       1    3      2       0      0     0        0       0       1
        Mixed        9      9      8     10     9      9       9       7    6      4    3       8     15     21      39      26   30     23      23     23    27      27        7      25
        Matri       39     33     32     33    33     37      33      29   29    27    23      33     23     43      15      27   20     19      22     18    16      17       16      21
                  100    100     100    100 100      100     100    100   100   100   100     100    100    100     100    100 100      100     100    100   100     100      100     100


Notes: All variables are averages in current USD. The construction of the measures of household consumption, income, and wealth is discussed in Section 2,
with further details in the online appendix. The construction of the indicators for societal systems is discussed in Section 5.5.
                 Table 17: Income and Wealth Mobility, Uganda (ISAs 2005-2010)


                                                     Fraction (%) of Households
                                                        that Left the Quintile

                                                     1st 2nd     3rd   4th   5th
                           Income

                              US                      23    37    40    39     23

                              Uganda        Rural     71    71    74    72     51
                                            Urban     70    69    72    71     46

                           Wealth

                              US                      25    39    41    37     20

                              Uganda        Rural     58    69    71    66     45
                                            Urban     53    66    69    64     39


Notes : The figures of economic mobility use the 2005-6 — 2009-10 Uganda ISA panel. The definition of household
consumption, income, and wealth is discussed in Section 2. The data are divided by quintiles with 1st denoting
the poorest quintile and 5th the richest.