. 5
( 13)


exactly the same spatial area picked up 56 households that had also been in
the 1978 universe survey. In 2004, these same 56 households were tracked and
51 were re-interviewed (indicating a 9 per cent attrition rate).

5.4.2. Analysis
The Indio Guayas household asset index is based on the following sources:
those de¬ned in the literature, research based on local anthropological knowl-
edge of asset vulnerability in the community (Moser, 1996, 1997, 1998), and
the empirical data available from the panel data. The variables were adapted

Asset Accumulation in Ecuador

from the questionnaire data from the 1978“2004 panel data. Two types of
physical capital were identi¬ed: housing and consumer durables. Financial
capital was extended to incorporate productive capital, while human capital
was limited to education because of a lack of panel data on health. Finally,
social capital was disaggregated in terms of household and community social
capital. 6
Table 5.1 outlines each type of asset analysed, the category of capital that it
belongs to, and the speci¬c components that make up its index. The following
section describes in detail the construction of the index to measure each of
these capital types and the associated challenges. Polychoric PCA was used for
many but not all of the asset categories; as we elaborate in the subsequent
sections, its advantages and limitations become clearer when moving from
theory to practice. As with any statistical technique, the devil is in the details
and we lay out below exactly how and why we chose speci¬c variables out of
this detailed dataset for use with different techniques.

Physical capital is generally de¬ned as comprising the stock of plant equip-
ment, infrastructure, and other productive resources owned by individuals,
businesses, and the public sector (see World Bank, 2000). In this study, how-
ever, physical capital is more limited in scope. It is subdivided into two and
includes the range of consumer durables households acquire, as well as their
housing (identi¬ed as the land, and the physical structure that stands on it).
Housing is the more important component of physical capital. In Indio
Guayas households squatting in severe conditions on a mangrove swamp
rapidly constructed wooden stilts and a platform, and then incrementally
built very basic houses with bamboo walls, wood ¬‚oor, and corrugated iron
roofs. However, such houses were insecure”bamboo walls could easily be split
by knives, and the materials quickly deteriorated. Consequently, as soon as
resources were available, households upgraded their dwellings. This started
with in¬ll to provide land, followed by permanent housing materials such as
cement blocks and ¬‚oors. This very gradual incremental upgrading took place
over a number of years.
This process is re¬‚ected in the econometric ¬ndings on housing based on
the four indicators: type of toilet, light, ¬‚oor, and walls. These are ordered in
terms of increasing quality (for instance ˜incomplete™ walls are those in the
process of being upgraded from bamboo to either wood or brick/concrete),
A ¬fth type of capital, natural capital, is commonly used in the assets and livelihoods
literature. Natural capital includes the stocks of environmentally provided assets such as soil,
atmosphere, forests, water, and wetlands. This capital is more generally used in rural research.
In urban areas where land is linked to housing this is more frequently classi¬ed as productive
capital as is the case in this study. However, since all households lived on similar plots this
was not tracked in the dataset.

Caroline Moser and Andrew Felton

Table 5.1. Types of capital, asset categories, and components

Capital type Asset index categories Index components

Physical capital Housing Roof material
Walls material
Floor material
Lighting source
Toilet type
Consumer durables Television (none, b/w, colour, or both)
Washing machine
DVD player
Record player
Financial/productive capital Labour security Type of employment:
State employee
Private sector permanent worker
Contract/temporary worker
Productive durables Refrigerator
Sewing machine
Transfer/rental income Remittances
Rental income
Human capital Education Level of education:
Some primary school
Completed primary school
Secondary school or technical degree
Some tertiary education
Social capital Household Jointly headed household
Other households on solar
˜Hidden™ female-headed households
Community Whether someone on the solar:
attends church
plays in sports groups
participates in community groups

with the data showing a high degree of inter-household correlation. The
ordinal nature and positive correlation of the variables make this part of
the data highly suitable for analysis using the polychoric PCA technique (see
Table 5.2).
The estimated coef¬cients rise with the increasing quality of each asset,
and greater numbers (either positive or negative) mean that the variable
provides more ˜information™ on the household™s housing stock. For example,
the greatest negative coef¬cient is on having no electric lights. This means
that a household that lacks electric lighting is extremely likely to fall into the
lowest categories of the other types of assets: toilet, ¬‚oor, and walls. Similarly,
a household with a ¬‚ush toilet (the highest level within the toilet category) is

Asset Accumulation in Ecuador

Table 5.2. Housing stock polychoric PCA coef¬cients

Asset Coef¬cient

Toilet: hole
Toilet: latrine
Toilet: toilet 0.4541
Light: none
Light: illegally tapped electricity
Light: mains electricity 0.4063
Floor: earth/bamboo
Floor: wood
Floor: brick/concrete 0.3658
Walls: earth/bamboo
Walls: incomplete
Walls: wood
Walls: brick/concrete 0.3631

likely to have scored highly on the other items as well. This is because ¬‚ush
toilets were owned by the fewest people, and it was only in 2004 that almost
all households acquired one. In contrast, many people had connected to the
main electrical grid, and upgraded their ¬‚oors and walls to brick/concrete, by
The consumer durables variable illustrates a new type of dif¬culty with PCA.
Because the data cover multiple time periods, the ˜values™ of many of these
assets have changed between observations. For example, a black-and-white
television was relatively more valuable in 1978 than in 2004. In 1978, it was
a sign of wealth to own a black-and-white television, but in 2004, a sign of
poverty, as colour televisions had become available. By 2004, a number of
electronic items that had become available were simply not on the market
This issue can be addressed either by conducting a separate analysis for
each year, or by aggregating the data across time. The ¬rst three columns
of Table 5.3, which calculate values for each item in each year, illustrate
the changing values of many of the variables. In 1978, a black-and-white
television had a strongly positive coef¬cient, as it was a sign of wealth. Its
coef¬cient decreased during each time period as it became less indicative of
wealth. This demonstrates that in addition to its ability to create a single
variable, asset index construction is useful for tracking the relative value of
Aggregating the time periods proved to be the most ef¬cacious method of
combining the variables, as it allows relative comparisons across, as well as
within, time periods. Items that were once luxury items can receive a negative
score in later time periods, which means they are on average indicative of
poverty. However, because we estimate the value of not owning the asset
as well as the value of owning it, a household with a black-and-white TV

Caroline Moser and Andrew Felton

Table 5.3. Consumer durables polychoric PCA coef¬cients

Asset 1978 1992 2004 All years combined

’0.5358 ’0.4168 ’0.4687 ’0.4616
’0.0317 ’0.2939 ’0.0564
B/w TV 0.5797
Colour TV 0.2782 0.0194 0.3093
Both b/w and colour TV 0.5229 0.3778 0.7321
’0.8888 ’0.6856 ’0.2631 ’0.1069
No radio
Radio 0.1761 0.1358 0.0943 0.0277
’0.0402 ’0.0914 ’0.0492
No washing machine
Washing machine 1.4188 0.7685 0.7507
’0.1190 ’0.1802 ’0.1428
No bike
Bike 0.3009 0.1665 0.3973
’0.0949 ’0.0240 ’0.0253
No motorcycle
Motorcycle 0.7978 0.2020 0.3464
’0.0623 ’0.0258
VCR 0.6574 0.8706
’0.1477 ’0.0580
No DVD player
DVD player 0.6507 0.8844
’0.1738 ’0.1236 ’0.0639
No record player
Record player 0.4394 0.6239 0.3718
’0.1100 ’0.0519
No computer
Computer 0.4843 0.7910

in 1978, although receiving a ˜negative™ score in aggregate, still ranks much
higher in 1978 than a comparable household that does not own a TV at all. In
fact, the average household in 1978 had a negative score for their consumer
durables capital, but the ordinal rankings remain the same as the coef¬cients
were calculated separately for each year. Therefore, the rankings make sense
both within and across time periods.
This method produces a feasible and accessible continuous variable repre-
senting ownership of consumer durables. Figure 5.2 shows the kernel density
distributions for the consumer durables variable in each round. In 1978 and
1992, the variable is roughly normally distributed (when the households are
just beginning to diverge from their equal starting points), but by 2004 it
resembles the lognormal distribution commonly found in studies of income
distribution (which parallels the actual growth of income and asset inequality
in Guayaquil).

Human capital assets refer to individual investments in education, health,
and nutrition, which affect people™s ability to use their labour and change the
nature of their returns from their labour. Education is the only component in
this index and therefore provides only a partial picture of human capital. 7

The study contains detailed information on health status, particularly in terms of
shocks relating to serious illnesses or accidents, as well as the use and cost of health services.

Asset Accumulation in Ecuador


Kernel density



-1 0 1 2 3 4

1978 1992 2004

Figure 5.2. Consumer durables capital density estimates

Human capital presents a different challenge from previous categories
because it is usually measured at the individual, not household level. If we
want to measure human capital at the household level, we need to develop
a method of aggregation. Furthermore, we have only one key measure of
human capital at the individual level: years of education (or, alternatively,
level of completed education). Since there is only one variable, we cannot use
any of the varieties of PCA at the individual level because PCA measures the
correlation between two or more variables. We could assign an equal weight
to every year of education and add them up”but this brings us back to the
earlier methods described above, with the same attendant problems. Instead,
we make use of the fact that the survey contains the income earned by every
individual, so are able to estimate the monetary return to education. The
education variable was split into ¬ve levels: none, some primary, completed
primary, completed secondary, and some tertiary (see Table 5.4).
Income earned from wages is regressed on the level of education, age and
age squared to proxy for experience, and a gender dummy variable. The
regression is estimated separately for each year because the value of each type
of degree changes every year as the job market changes. Therefore, the value of
the education capital of a household can change even though the actual level

However, the lack of an adequate methodology to translate these into a health asset index
means that the information remains at the narrative level.

Caroline Moser and Andrew Felton

Table 5.4. Value of educational levels

Educational level 1978 1992 2004

Illiterate 3.52 2.15 3.18
Some primary 3.20 2.47 3.09
Completed primary 3.31 2.51 3.19
Completed high school or technical school 3.09 2.66 3.21
Tertiary education 3.98 3.12 3.37

Note: Coef¬cients for age, age squared, and gender not shown.

of education in the household did not. In 1978, there was very little difference
in terms of wages in the value of being illiterate, having some primary educa-
tion, or having completed primary school. These education levels accounted
for almost 90 per cent of the young settlers of Indio Guayas™s population at
the time. Those few that had higher education earned considerably more in
the labour market. Over time, however, being illiterate or without a primary
degree became more disadvantageous because less-educated people earned
lower wages. Meanwhile, the macroeconomic instability of 1992 decreased
wages for every educational group.
Human capital is usually valued for its use in the labour market, so it is
one type of capital that may be measured in monetary terms relatively easily
using techniques similar to those described above. Years of education and
salary are frequently available in surveys. On the other hand, endogeneity and
other issues are problematic in this methodology. For example, many people
with low education are not in the workforce”neither including them as zero
income nor not including them is wholly satisfactory. If low-educated people
are disproportionately absent from the formal economy, then the estimation
of returns from low levels of education might be biased up, because only
the most talented of the poorly educated have income. Table 5.4 shows that
illiterate people often earn more than those with more education, suggesting
that this problem may indeed exist in our data.
Furthermore, the use of other variables like age and gender, while impor-
tant, also leads to complications. Younger generations on average had more
educational opportunities, and the importance of education changes over the
years as the economy develops. By using income as the dependent variable,
we are measuring the market value of education, rather than some level of
inherent human capital speci¬c to the individual. Finally, we may disagree
with the values the labour market places on human capital. For example,
people with no education at all in 1978 earned more than any other group
except those with college educations (only one person). However, we want
to assign those people the lowest level of human capital. For these reasons,
estimating the level of human capital using other variables may produce worse
results than an arbitrary ranking.

Asset Accumulation in Ecuador

Ideally, PCA or a similar technique can be used, given a variety of data on
individuals assumed to be correlated with the unmeasurable ˜human capital™,
such as test scores, grades, education, etc. In fact, the literature on measuring
intelligence often uses PCA to collapse scores along a number of dimensions
into one variable (Jensen, 2002). However, data of this nature are not usually
available, especially in developing countries.

Financial/productive capital comprises the monetary resources available to
households. In developed countries, this usually translates into ¬nancial
assets such as bank holdings, stock and bond investments, house equity, etc.
that can be drawn on in case of need. However, few citizens of developing
countries have any of these. In this case, a monetary measure is actually less
useful than an asset index, because the assets are likely to be intangible and
not easily quanti¬ed in monetary terms.
The ¬nancial/productive capital asset index comprises three components:
labour security, which measures the extent to which an individual has security
in the use of their labour potential as an asset; transfer/rental income which is
non-earned monetary resources; and productive durables, which are durable
goods with an income-generating capability. Table 5.5 shows the results of the
polychoric analysis.
Labour security is undoubtedly the most challenging component in the
index. However, it represents an effort to include labour as an asset?omitted
so far in the work on asset indices”and to include employment vulnera-
bility as linked to stability of job status. The composite component derives
from combining two ILO work categories on employer type and work status,
and is ranked in terms of vulnerability in the Guayaquil context through
local anthropological knowledge: the most secure type of job is working for
the state, the second is as a ˜permanent worker™ (with a formal, stable job)

Table 5.5. Financial/productive capital polychoric PCA coef¬cients

Asset Coef¬cient (all years combined)

Sewing machine: no
Sewing machine: yes 0.0173
Refrigerator: no
Refrigerator: yes 0.3133
Car: no
Car: yes 0.8356
Home business income: no
Home business income: yes 0.4999
Rental income: no
Rental income: yes 0.9031
Remittances: no
Remittances: yes 0.4779

Caroline Moser and Andrew Felton

in the private sector, the third is self-employment, and the least secure is
contract/temporary work. The ordering of the top two job types should be
uncontroversial, but the latter two require some explanation. Entrepreneurs,
even on a small scale, build up business knowledge, contacts, and habits
that can help sustain them through a downturn. They can continue in their
business even during times of reduced demand (Moser, 1981). Temporary
workers, however, have less to fall back on when they are let go. Consequently
we make the judgement that the self-employed have more job security than
contract workers. Unfortunately, we must still arbitrarily assign weights to
each type of job: we give temporary work a four on the vulnerability scale
and move down to government work, which gets a weight of one. We then
aggregate up to the household level by computing the average vulnerability of
each household. Although this method retains some of the arbitrariness that
we have been trying to avoid, we at least manage to turn labour security into
an ordinal variable that can be used for polychoric PCA.
The main sources of unearned income are remittances, government trans-
fers, and rent. The ¬rst two are transfers of income within society and
the latter is a return on capital”similar to income from physical goods
as analysed above. Non-wage income has increasingly played an important
role in household income. Remittance income has risen most dramatically,
linked to the explosion of Ecuadorian migrants in the late 1990s following
dollarization and the banking crisis. The fact that this accounted for over
50 per cent of non-wage income in 2004 shows that having someone abroad
is a real household asset. Remittance income comprised more than half
total income for some households. Rental income is much smaller and more
recent as households have speci¬cally built on extra rooms to accommodate
renters either at the back of their plots, or in additional ¬‚oors to their
Finally, productive durable goods count as ¬nancial/productive capital
because they represent a current or potential income stream. In the context
of Guayaquil, sewing machines, refrigerators, and cars were popular examples
of this type of goods, with each predominating during different time periods.
Numerous families acquired sewing machines in the 1970s. Men primarily
used them in their work as tailors, either as self-employed or as subcontracting
outworkers. A lesser number of women had sewing machines for use both
within the family but also to generate income through work as dressmakers
(Moser, 1981). Refrigerators are generally used as the basis of a small enterprise
selling ice, frozen lollies, and cold drinks such as Coca Cola. Car ownership
is a more recent phenomenon and one that requires far more capital (usually
based on credit loans). Almost all local men who own cars use them as taxis
to generate an income. While in some cases these are full-time occupations,
in other cases they supplement other jobs particularly when there is high
demand”such as weekend nights.

Asset Accumulation in Ecuador

Table 5.6. Community social capital polychoric PCA coef¬cients

Asset Coef¬cient

Don™t attend church
Attend church 0.2744
Don™t participate in community activities
Participate in community activities 0.3650
Don™t participate in sports league
Participate in sports league 0.6050

Social capital, the most commonly cited intangible asset, 8 is generally de¬ned
as the rules, norms, obligations, reciprocity, and trust embedded in social rela-
tions, social structures, and societies™ institutional arrangements that enable
its members to achieve their individual and community objectives. Social cap-
ital is generated and provides bene¬ts through membership in social networks
or structures at different levels, ranging from the household to the market-
place and political system. The index differentiates between community-level
social capital and household social capital. The latter is based on detailed
panel data on changing intra-household structure and composition (see
Moser, 1997, 1998). Social capital is usually considered extremely dif¬cult for
social scientists to measure because the assets are non-physical and dif¬cult to
translate into monetary terms. In the asset index framework, however, they
are measured in terms of binary variables such as household participation in
various different activities and groups. 9
This dataset uses three variables to determine household social capital as
identi¬ed in Table 5.1. The results are shown in Table 5.6. The index was
constructed using polychoric PCA. The three variables are positively correlated
with each other and participation in a sports league was the best indicator of
social capital. Not attending church was the best indicator of a lack of social
capital, garnering a large negative coef¬cient. Of the twelve observations in
which a household had a member participating in a sports club, only one of
those did not also have someone who either went to church or participated
in community activities.
Household social capital as an asset is complex because it is both positive
and negative in terms of accumulation strategies. On one hand, households
act as important safety nets protecting members during times of vulnerability

Social capital is the most contested type of capital (Bebbington, 1999). The development
of the concept is based on the theoretical work of, for instance, Putnam (1993) and Portes
Again it important to note that the original study in 1978 was not designed to ˜measure™
social capital; consequently the groups identi¬ed do not represent the universe but are those
for which comparative data is available.

Caroline Moser and Andrew Felton

and can also create opportunities for greater income generation through
effective balancing of daily reproductive and productive tasks (see Moser,
1993). On the other hand, the wealth of a household may actually be
reduced by having to support less-productive members. Over time, house-
holds change in size and restructure their composition and headship in
order to reduce vulnerabilities relating both to lifecycle and wider external
Household social capital was de¬ned as the sum of three indicator variables.
The ¬rst component, jointly headed households, serves to indicate trust and
cohesion within the family between partners, and is applied to both nuclear
and couple-headed extended households. In 1978 when the community com-
prised young families nearly two-thirds were nuclear in structure. 10 By 1992
this had dropped to a third and in 2004 was only one in ten households.
In contrast, the reverse was true for couple-headed extended households,
growing from one-¬fth in 1978 to two-¬fths by 1992 and levelling off to
slightly more by 2004. Within many extended households there are also
˜hidden™ female heads of household: unmarried female relatives raising their
children within the household to share resources and responsibilities with
others. This has grown from less than one in ten in 1978 to more than one
in four in 2004. The third component is the presence of other family-related
households living on the solar”usually the households of sons or daughters.
Unfortunately, none of the varieties of PCA could be used here because
the variables are not all positively correlated, but we wanted to give them
all a positive value. PCA or a similar technique would have given at least one
of the coef¬cients a negative value. We therefore had to give them all equal
weight (or some other arbitrary weight). This is an area where more research
is needed.

5.4.3. Preliminary analysis
Longitudinal analysis of changing poverty levels based on income alone
provides one measure of well-being, and shows movement between poverty
levels. A more comprehensive understanding of household assets accumula-
tion complements income data in helping to identify why some households
are more mobile than others and how some households successfully pull
themselves out of poverty while others fail.
Most assets increased fairly steadily between 1978 and 2004 (Figure 5.3).
The greatest difference between households in the ¬rst and last periods was
A nuclear household comprises a couple living with their children; an extended house-
hold comprises a single adult or couple living with their own children and other related
adults or children; a female-headed household comprises a single-parent, nuclear, or extended
household headed by a woman; if married she identi¬es herself as the head usually because
her husband is not the main income earner; a woman is counted as a ˜hidden™ head of
household if she: (1) lives on the family plot; (2) is unmarried; and (3) has at least one child.

Asset Accumulation in Ecuador

Standard deviation
Human capital

0.5 0.5
social capital

0.0 0.0

social capital
’0.5 ’0.5

’1.0 ’1.0

1978 1992 2004

Figure 5.3. Household asset accumulation in Guayaquil, Ecuador, 1978“2004

in housing: the average household improved its housing stock by over two
standard deviations. Community capital actually decreased from 1992 to
2004. These quantitative observations are supported by the anthropological
Asset analysis can be particularly useful when used in conjunction with
income data. The following Figure 5.4 displays the level of housing and
consumer durables owned by income group during each time period and
demonstrate that households of all income levels have similar average lev-
els of housing, but very different levels of consumer durables (especially in
2004). This implies that poor households place a much greater emphasis on
accumulating housing than consumer durables.
These numbers are not adjusted for household size, although size is obvi-
ously signi¬cant. Poorer households tend to be larger households, with greater
needs for housing space and physical infrastructure. Also, larger households,
ceteris paribus, tend to have more people working and greater total income
than smaller households, although large households tend to have lower per
capita incomes than small households. This means that the larger household
may have an advantage in accumulating assets and therefore look wealthier,
but those assets have to be shared among a greater number of people. Some
assets can be shared without diminishing their utility for any one person. A
radio, for example, can be listened to by multiple people at once. Cars can be
shared to some extent, but they cannot be driven by more than one person at
the same time. Naturally, jobs and education are not shared. Finally, the index
is not adjusted for household size because PCA techniques used to calculate


Caroline Moser and Andrew Felton
Physical capital: consumer durables
Physical capital: housing

.5 1



’1.5 ’.5

1978 1992 2004 1978 1992 2004
round round
Very poor poor Not poor Very poor poor Not poor

Figure 5.4. Patterns of housing and consumer durables investment over time by income group
Asset Accumulation in Ecuador

Tradeoff between consumption in 1992 and kids™ education in 2004

2004 kids™ education





’1 0 1 2 3
1992 consumption

Residuals Fitted values
Both kids™ education and consumption adjusted for household income

Figure 5.5. Trade-off between consumption and kids™ education

the asset indices do not have units, and would therefore be unsuitable for
interpreting variables on a per capita basis.
We can also use asset indices to examine how individual households
make ˜portfolio™ choices between types and amounts of assets to accumulate.
Figure 5.5 illustrates a trade-off between the level of consumer durables in
a household in 1992 and the total amount of education that the house-
hold™s children receive in 2004. By 1992 the original generation of settlers
had reached middle age and most had school-age children. By 2004 many
of the second-generation children had ¬nished school and moved out on
their own. The data in Figure 5.5 are adjusted for household income because
wealthier households may be able to afford more consumer durables and
educate their kids better. The kids™ education is not per capita”it represents
the total investment that households have put into their kids™ human cap-
ital and is parallel to consumer durables ownership, which is also not per
capita. The ¬gure quanti¬es the stark choices that households”especially
those with large numbers of children”make between acquiring two types of
Figure 5.6 uses ˜star graphs™ to display the changing composition of house-
hold portfolios over time. Because this type of graph cannot display negative
numbers, the estimated levels of assets were scaled so that the minimum
score was 0 and average score was 1. The graphs display how the average
portfolios increased in size and changed in shape. For example, there is a clear
outward shift of ¬nancial capital and consumer durables by 2004, as well as a
noticeable increase in variation.


Caroline Moser and Andrew Felton
1978 1992 2004

Housing Housing
5 5
4 4
3 3
2 2 Household Consumer
Household Consumer Household Consumer
social capital durables
social capital social capital durables 1
1 1
0 0
-1 -1
-2 -2

Community Human
Community Human Community Human
social capital capital
social capital capital social capital capital

Financial Financial
capital capital

Figure 5.6. Star graphs of household asset portfolios
Asset Accumulation in Ecuador

These simple diagrams suggest several directions in which the work could
go. One would be the use of mathematical techniques to sort observations
into groups based on a number of variables, such as cluster analysis and fuzzy
set theory. By 2004, there is enough differentiation among the households
that they may have sorted themselves into identi¬able groups. One of the
advantages of the asset-based analysis presented above is that it enables
enough dimensional reduction to make other techniques more intuitive and
easier to apply, because they can incorporate fewer key variables.

5.5. Conclusion

This study of assets provides an overview of recent research on the use of asset
indices as well as illustrating a particular way of constructing an asset index.
Much progress has been made over the last ¬ve years, but a number of issues
remain. For example, principle components analysis, in all its variations, is
still dependent on the observed variables being positively correlated. Another
unresolved issue is how best to aggregate assets from the individual level
to the household level without involving arbitrary methods like summation
or averaging. Similarly, there is no clear way to adjust levels of assets for
household size. Finally, this is obviously not a methodology that can be
immediately applied to many datasets. It requires considerable knowledge of
the different variables in order to select and transform them into appropriate
subjects for polychoric PCA.
Nevertheless, existing techniques contribute to the accuracy and robustness
of asset accumulation analysis. This chapter has demonstrated how grouping
a large number of assets into a smaller number of dimensions facilitates an
intermediate level of analysis. By examining how households allocate their
resources using an asset index, the analysis of speci¬c poverty mechanics is
possible without examining an overwhelming quantity of individual vari-
ables. Asset indices are an important complement to pure income data because
they paint a clearer picture of the strategies households in various income
groups have employed to acquire different types of assets, and because they
provide clues to poverty alleviation. The next step is to use these indices to
understand poverty dynamics in greater detail.


Adato, M., Carter, M., and May, J. (2006), ˜Exploring Poverty Traps and Social Exclusion
in South Africa using Qualitative and Quantitative Data™, Journal of Development
Studies, 42(2): 226“47.
Asselin, L.-M. (2002), ˜Composite Indicator of Multidimensional Poverty™, unpublished
paper, Institut de Math©matique Gauss.

Caroline Moser and Andrew Felton

Barrientos, A. (2003), ˜Non-Contributory Pensions and Well-Being among Older People:
Evidence on Multidimensional Deprivation from Brazil and South Africa™, DEPP
Working Paper, University of Manchester, IDPM.
Bebbington, A. (1999), ˜Capitals and Capabilities: A Framework for Analysing Peasant
Viability, Rural Livelihoods and Poverty™, World Development, 27: 2021“44.
Booysen, F., et al. (2005), ˜Using an Asset Index to Assess Trends in Poverty in Seven
Sub-Saharan African Countries™, paper presented at conference on Multidimensional
Poverty hosted by the International Poverty Centre of the United Nations Develop-
ment Programme (UNDP), 29“31 August, Brasilia.
Carney, D. (ed.) (1998), Sustainable Rural Livelihoods: What Contribution Can We Make?,
London: Department for International Development (DFID).
Carter, M., and May, J. (2001), ˜One Kind of Freedom: Poverty Dynamics in Post-
Apartheid South Africa™, World Development, 29(12): 1987“2006.
Chen, S., and Ravallion, M. (2000), ˜How Did the World™s Poorest Fare in the 1990s?™,
Policy Research Working Paper 2409, Washington: World Bank.
Clark, D. A., and Qizilbash, M. (2002), ˜Core Poverty and Extreme Vulnerability in South
Africa™, Discussion Paper No. 2002“3, School of Economics, University of East Anglia.
(2005), ˜Core Poverty, Basic Capabilities and Vagueness: An Application to the
South African Context™, GPRG Working Paper No. 26, Universities of Manchester
and Oxford. Available online at <http://www.gprg.org/pubs/workingpapers/default.
Ellis, F. (2000), Rural Livelihoods and Diversity in Developing Countries, Oxford: Oxford
University Press.
Filmer, D., and Pritchett, L. (2001), ˜Estimating Wealth Effects without Expenditure
Data”or Tears: An Application to Educational Enrollments in States of India™, Demog-
raphy, 38(1): 115“32.
Green, M. (2006), ˜Representing Poverty and Attacking Representations: Perspectives on
Poverty from Social Anthropology™, Journal of Development Studies, 42(7): 1108“29.
and Hulme, D. (2005), ˜From Correlates and Characteristics to Causes: Think-
ing about Poverty from a Chronic Poverty Perspective™, World Development, 33(6):
Harriss, J. (2006), ˜Why Understanding Social Relations Matters More for Policy in
Chronic Poverty than Measurement™, paper presented at the Concepts and Methods
Workshop, University of Manchester, October.
Hulme, D., and McKay, A. (2005), ˜Identifying and Measuring Chronic Poverty: Beyond
Monetary Measures™, paper presented at conference on Multidimensional Poverty
hosted by the International Poverty Centre of the United Nations Development
Programme (UNDP), 29“31 August, Brasilia.
Jensen, A. (2002), ˜Psychometric g: De¬nition and Substantiation™, in R. Sternberg and
E. Grigorenko (eds.), The General Factor of Intelligence: How General Is It?, Mahwah, NJ:
LEA, Inc.
Jolliffe, I. T. (2002), Principal Component Analysis, New York: Springer-Verlag.
Kanbur, Ravi (ed.) (2002), Qual“Quant: Qualitative and Quantitative Methods of Poverty
Appraisal, Delhi: Permanent Black.
Klasen, S. (2000), ˜Measuring Poverty and Deprivation in South Africa™, Review of Income
and Wealth, 46: 33“58.

Asset Accumulation in Ecuador

Kolenikov, S., and Angeles, G. (2004), ˜The Use of Discrete Data in Principal Component
Analysis: Theory, Simulations, and Applications to Socioeconomic Indices™, Working
Paper of MEASURE/Evaluation project, No. WP-04“85, Carolina Population Center,
University of North Carolina.
Kullback, S., and Leibler, R. A. (1951), ˜On Information and Suf¬ciency™, Annals of
Mathematical Statistics, 22(1): 79“86.
Moser, C. (1981), ˜Surviving in the Surburbios™, Bulletin of the Institute of Development
Studies, 12(3): 19“25.
(1993), Gender Planning and Development: Theory, Practice and Training, New York:
(1996), ˜Confronting Crisis: A Comparative Study of Household Responses to
Poverty and Vulnerability in Four Poor Urban Communities™, Environmentally Sus-
tainable Development Studies and Monograph Series No. 8, Washington: World Bank.
(1997), ˜Household Responses to Poverty and Vulnerability, Volume 1: Confronting
Crisis in Cisne Dos, Guayaquil, Ecuador™, Urban Management Program Policy Paper
No. 21, Washington: World Bank.
(1998), ˜The Asset Vulnerability Framework: Reassessing Urban Poverty Reduction
Strategies™, World Development, 26(1): 1“19.
(2002), ˜ “Apt Illustration” or “Anecdotal Information”: Can Qualitative Data Be
Representative or Robust?™, in R. Kanbur (ed.), Qual“Quant: Qualitative and Quantitative
Methods of Poverty Appraisal, Delhi: Permanent Black, 79“89.
(2008), ˜Assets and Livelihoods: A Framework for Asset-Based Social Policy™, in
C. Moser and A. Dani (eds.), Assets, Livelihoods, and Social Policy, Washington: World
and Felton, Andrew (2007), ˜Intergenerational Asset Accumulation and Poverty
Reduction in Guayaquil, Ecuador 1978“2004™, in Caroline Moser (ed.), Reducing Global
Poverty: The Case for Asset Accumulation, Washington: Brookings Institutional Press.
Portes, A. (1998), ˜Social Capital: Its Origins and Applications in Modern Sociology™,
Annual Review of Sociology, 24: 1“24.
Putnam, R. (1993), Making Democracy Work: Civic Traditions in Modern Italy, Princeton:
Princeton University Press.
Ravallion, M. (1992), ˜Poverty Comparison: A Guide to Concepts and Methods™, Living
Standards Measurement Study Working Paper 88, Washington: World Bank.
Sahn, D., and Stifel, D. (2003), ˜Exploring Alternative Measures of Welfare in the
Absence of Expenditure Data™, Review of Income and Wealth, 49(4): 463“89.
World Bank (2000), World Development Report 2000/01: Attacking Poverty, Washington:
World Bank.
(2004), Ecuador Poverty Assessment, Washington: World Bank.

Looking Forward
Theory-Based Measures of Chronic Poverty
and Vulnerability

Michael R. Carter and Munenobu Ikegami

6.1. From Backward-Looking to Forward-Looking
Poverty Analysis

[The historian] becomes a crab. The historian looks backward; eventually
he also believes backward.
(Friedrich Nietzsche, Twilight of the Idols)

Conventional quantitative poverty analysis invariably looks backwards to
the most recent living standards survey to enumerate (the past) extent and
nature of poverty. Living standards surveys, with their 7-day, 30-day, and
12-month recall periods, look yet further backward. While there is no rea-
son to follow Nietzsche and assert that backward-looking poverty analysis
˜believes backwards™, there are clearly (forward-looking) questions that con-
ventional poverty analysis is ill equipped to answer. Perhaps the most impor-
tant of these questions concerns the future persistence of observed poverty
status: Are the observed poor chronically poor, or are they in a transitory
Others have struggled with this question. One approach (used by the
Chronic Poverty Research Centre, 2004) is empirical. With numerous repeated
observations of the same households, the chronically poor can be identi¬ed
as those who have been ˜frequently™ poor in the observed past. While this
approach has much to recommend it, it is expensive and has an ad hoc
element (how frequently must an individual be observed to be poor in order
to be classi¬ed as chronically poor?). More importantly, it is also backward

Looking Forward

The approach put forward in this chapter is rather different. Using guidance
from the microeconomic theory of poverty traps, this chapter uses the past to
identify structural patterns of change”asset dynamics”rather than past lev-
els of poverty. The statistical identi¬cation of these patterns then permits the
creation of forward-looking poverty measures that tell us where we expect the
poor to be in the future, not where they have been in the past. 1 While these
new measures do not eliminate the need for other approaches (indeed, when
combined with standard approaches they provide a more complete poverty
dialogistic for a particular economy), they do offer a promising approach for
the conceptualization and measurement of chronic poverty. They also carry
important policy implications.
Building on the work of Buera (2005) and Barrett, Carter, and Ikegami
(2007), section 6.2 of this chapter develops a theoretically grounded approach
to chronic poverty that emphasizes the role of individual heterogeneity and
clari¬es the role that vulnerability to economic shocks plays in producing
chronic poverty. The key theoretical construct that emerges from this analysis
is the Micawber Frontier, de¬ned as the level of assets below which an indi-
vidual of a particular skill level is unable to successfully accumulate assets and
move ahead economically over time.
Section 6.3 then shows how knowledge of the Micawber Frontier can be
used to generate two classes of chronic poverty measures. The ¬rst class
generalizes a suggestion put forward by Carter and Barrett (2006) and is based
on the individual™s distance from the Micawber Frontier. The second uses
information on the Micawber Frontier to simulate future asset (and income)
changes. When combined with the family of chronic poverty measures put
forward by Calvo and Dercon (this volume), these asset dynamics open yet
another window into chronic poverty that is forward looking and based
on a theoretically well-speci¬ed model. Numerical simulation of a stylized
100-household economy is used to illustrate both sets of measures. In addi-
tion, both sets of chronic poverty measures can be used to derive well-
structured measures of vulnerability, where vulnerability is understood as the
fraction of all chronic poverty that would be eliminated in a world without
economic shocks (or with perfectly well-positioned social safety nets).
Section 6.4 then takes some ¬rst steps toward implementing the ideas
put forward in this chapter, using estimated asset dynamics in South Africa
over the 1993 to 1998 period to calculate chronic poverty measures based
on distance from the asset threshold. While based on somewhat strin-
gent assumptions, these forward-looking estimated chronic poverty measures
provide a more in-depth look at the nature of poverty than do standard

This approach of course still relies on the past, but uses it to identify patterns of asset
dynamics. If those past patterns of change are not stable, providing a poor guide to future
patterns, then the approach put forward here also becomes backward looking.

Michael R. Carter and Munenobu Ikegami

FGT-based measures. At the same time, data from a later time period (2004)
illustrate weaknesses of the estimated chronic poverty measures and point
the way toward more reliable estimation needed to capture the full richness
of a theory-based approach to chronic poverty. The chapter closes with some
re¬‚ections on the implications of the analysis for the design of social protec-
tion policy which potentially will pay off with a double dividend of reduced
chronic poverty.

6.2. A Theory-Based Approach to Chronic Poverty

This section summarizes recent theoretical work by Barrett, Carter, and
Ikegami (2007) (hereafter cited as BCI) on the economics of poverty traps.
Building on the dynamic model of Buera (2005) that explicitly incorporates
the intrinsic capacity or ability differences of individuals, BCI show that there
are two types of chronic poverty:
r Intrinsic Chronic Poverty suffered by those of relatively low skill and pos-
sibilities who are inevitably trapped in a poor, low-level equilibrium trap
(given the structure of wages and opportunity in their economy); and
r Multiple Equilibrium Chronic Poverty suffered by a middle-ability group that
has the potential to be non-poor in their extant economy, but whose
histories have placed them below the minimum asset threshold needed
to initiate and sustain the accumulation needed to escape poverty.

In addition to these two groups, the BCI model identi¬es a third, high-ability
group that may be consumption poor for an extended period of time, but who
are expected to surmount a poor standard of living given a suf¬ciently long
period of time in which to accumulate assets. We refer to this third group as
the Intrinsically Upwardly Mobile.
This section proceeds in two steps. First, it considers the implications of the
BCI model in the absence of economic shocks. While unrealistic, this simpli-
¬cation underwrites basic insights into the economics of asset thresholds and
chronic poverty. In addition, when paired with the analysis of shocks and risk
pursued later in this section, the simpli¬ed model will suggest measures of
vulnerability and its effect on chronic poverty.

6.2.1. Heterogeneous ability and poverty traps
Building on the model of Buera (2005), BCI assume that each individual is
endowed with a level of innate ability (·) as well an initial level of capital
(k0 ). Every period t, the individual has the choice between two alternative

Looking Forward
Livelihood and income

fH (·i ,kit )

fL(·i ,kit)

k — (·)
k —(·) Capital, kit
k(·i) H

Figure 6.1. Assets and livelihood options

technologies for generating a livelihood, f :

fL (·, k) = ·k„ L under the low technology
f (·, k) =
f H (·, k) = ·k ’ E under the high technology

Both technologies are skill sensitive (for any given technology, more able
people can produce more than less able people). One technology (the ˜high™
technology) is subject to ¬xed costs, E , meaning that the technology is not
worth using with low amounts of capital. Figure 6.1 illustrates these technolo-
gies for an individual with a given skill level ·. As can be seen in the ¬gure,
the individual (interested in maximizing income or livelihood possibilities)
will optimally shift to employing the high technology only after k reaches the
critical level k(·) = {k| fL (·, k) = f H (·, k)}.
Using this basic set-up, BCI analyse when it is possible and desirable for
the individual to save and accumulate assets in order to surpass k(·), employ
the high technology, and reach a higher standard of living. As summarized
in Appendix 6.1, the BCI model assumes that individuals divide their total
income ( f ) between consumption (ct ) and investment (it ) in order to max-
imize their intertemporal stream of utility. Assets evolve according to the
following rule

kt+1 = it + (1 ’ °)kt . (6.1)

Michael R. Carter and Munenobu Ikegami

where it = f (·, kt ) ’ ct is investment and ° is the rate at which capital depreci-
ates. Critically, the model assumes that the individual cannot borrow against
future earnings to build up capital and can only pursue autarchic accumula-
tion strategies.
The solution to this intertemporal choice problem de¬nes an investment
rule, i — (kt |·). Using this rule, we can de¬ne expected capital stock of a house-
hold in year t + 1 as follows:

kt+1 (kt |·) = i — (kt |·) + (1 ’ °)kt

If individuals had access to only one technology, they would optimally accu-

mulate capital up to the steady state values shown in Figure 6.1, kL (·) for the
low technology and k— (·) for the high technology. 2 To make it easy to discuss

the model, we will assume that the poverty line just equals f (kL (·)) for a high-
skill individual. That is, individuals can only become non-poor if they adopt
the higher livelihood strategy. 3
The key question addressed by the BCI model is whether individuals whose
initial capital stock is below k(·) gravitate toward the high or the low technol-
ogy. Consider an individual who begins life with the asset position at the level
marked by the dot in Figure 6.1. Will this individual optimally move to the
right over time, accumulating assets and ending up at k— (·) and a non-poor
standard of living? Alternatively, will the individual de-accumulate, move to

the left, and settle into a poor standard of living with capital stock kL (·)? 4
More formally, is there an initial asset threshold, which we will denote k(·),
below which individuals stay at the low equilibrium (remaining chronically
poor), and above which they will move to the high equilibrium (eventually
becoming non-poor)?
As analysed by BCI, the answer to this question depends on the skill level
of the individual. In particular, there will be three classes of individuals
each exhibiting distinct dynamically optimal behaviour. Figure 6.2, which is
created through the numerical analysis of the BCI model, illustrates these
three classes (see Appendix 6.1 for the full speci¬cation of the dynamic
programming model). Along the horizontal axis are skill levels, ranging from
least to most able. The vertical axis measures the stock of productive assets.

Note that these steady state values are increasing in the level of skill, ·. The steady state
values are also in¬‚uenced by the individual™s discount rate, meaning his or her willingness to
sacri¬ce current consumption in order to save and gain higher future consumption.
If this assumption is not true, individuals may get trapped at the low equilibrium, but
they would not necessarily be poor.
Note that beyond the low-level equilibrium, k— (·), the immediate marginal returns to
additional capital are low and less than the value of the consumption that the individual
must sacri¬ce in order to accumulate additional capital. In the rationality of the model, an
individual will only make that sacri¬ce if future returns (when she ¬nally accumulates at least
k(·) <sp>) are large enough close enough (and suf¬ciently certain, when there is uncertainty
in the model).

Looking Forward

k(·), Micawber
Initial capital endowment, k0

Asset poverty line

Multiple Upwardly
equilibrium poor mobile

0.90 0.95 1.00 1.05 1.10 1.15 1.20
·L ·H Intrinsic ability, ·

Figure 6.2. The Micawber Frontier (non-stochastic case)

The dashed curve is the asset poverty line, de¬ned as the level of assets needed
to generate an expected standard of living equal to the poverty line. The solid
curve shows the asset level at which an individual is just indifferent between
staying with the low technology versus building up stocks of assets such that
a transition to the high technology eventually becomes feasible. Denote this
frontier as k(·). An individual with ability level · will attempt to accumulate
and move out of poverty if she enjoys a capital stock k0 > k(·). Otherwise,
she will only pursue the low technology, accumulating the modest levels of
capital that it requires. Following Carter and Barrett (2006), we label k(·) as
the Micawber Frontier as it divides those who have the wealth needed to
accumulate from those who do not. 5
As illustrated in Figure 6.2, the numerical analysis identi¬es three distinct
regions in the space of ability and initial asset holdings. High-skill individuals
are those with · > · H who will always move toward the high equilibrium
even if they ¬nd themselves with a zero stock of assets (as k(·) = 0 for these
individuals). When they reach k(·) they will optimally switch to the higher
technology. Irrespective of their starting position, these individuals steadily
converge to the steady state asset value for the high technology. These
As discussed by Carter and Barrett, the phrase Micawber Threshold was ¬rst used by
Michael Lipton, and was then subsequently adopted by Zimmerman and Carter (2003), who
give it a meaning similar to that used here.

Michael R. Carter and Munenobu Ikegami

individuals are the intrinsically upwardly mobile, perhaps consumption poor
over some extended period as they save and accumulate assets, but eventually
expected to become non-poor.
In contrast, those with an ability level below the critical level · < · L will
never move toward the high technology if they ¬nd themselves with any
¬nite stock of assets. These are intrinsically chronically poor individuals who
lack the ability or circumstance to achieve a non-poor standard of living in
their existing economic context (CPRC, 2004, gives examples of individuals
who suffer such fundamental disabilities). 6
Finally, and most interestingly, the intermediate skill group with · L < · <
· H have positive, but ¬nite, values k(·). If suf¬ciently well endowed with assets
(k0 > k(·)), these individuals”the multiple equilibrium poor”will accumulate
additional assets over time, adopt the high technology, and eventually reach
a non-poor standard of living. If they begin with assets below k(·), these indi-
viduals will no longer ¬nd the high equilibrium attainable and will settle into
a low standard of living. Like the intrinsically chronically poor, this subset
of the multiple equilibrium poor will be chronically poor. 7 The total number
of chronically poor in any society will thus depend on the distribution of
households across the ability“wealth space shown in Figure 6.2. The chronic
poverty measures developed below rely on this insight.

6.2.2. Shocks, risk, and poverty traps
While establishing the possibility of distinct types of poverty, the analysis in
the prior section has ignored the reality of the economic shocks that threaten
the well-being of less well-off people almost everywhere. In the presence of
asset thresholds and poverty traps, economic shocks take on particular signif-
icance as Carter, Little, Mogues, and Negatu (2007) explore in an empirical
analysis of Ethiopia and Honduras.
There are at least three types of shocks that could generate risks that
might have a major impact on the accumulation decisions of poor people.
The ¬rst type is income shocks where households receive more or less than the
expected amount of income from their assets at any point in time. Second, the
marginal utility of income is also subject to shocks. Households, for example,
may suffer a severe illness, creating new needs for cash that effectively drive
up the marginal utility of consumption. Third, assets themselves are subject
to shocks. Livestock may die, businesses may burn down, or productive equip-
ment may unexpectedly break or be stolen. All three types of shocks have the
Addressing the poverty of such individuals will require transfers and perhaps efforts like
the PROGRESA programme to assure that the next generation acquires adequate human
Unlike the disabled, this class can be helped to help themselves with safety nets and cargo

Looking Forward

capacity to derail households from the accumulation paths discussed in the
prior section. We will here focus only on asset shocks. 8
In the presence of asset shocks, next-period assets depend not only on
prior stocks plus investments, but also on realized shocks. To represent this
possibility, BCI rewrite the rule that determines the evolution of capital stock
(6.1) as follows:

kt+1 = Ët [it + (1 ’ °)kt ], (6.2)

where Ët is a random variable realized every period t. Note that if Ë = 1, there is
no shock, whereas Ë < 1 indicates a negative shock that destroys some fraction
of assets. While in principle shocks could be positive (Ë > 1), such events seem
unlikely and we will restrict the analysis here to the case where only negative
shocks are possible.
As summarized in Appendix 6.1, the individual intertemporal choice prob-
lem can be modi¬ed with (6.2) and the assumption that the individual knows
the distribution of Ë and chooses consumption and investment every period
in order to maximize the discounted stream of expected utility. Denote the

investment rule in the presence of asset shocks as is (kt |·, ), where rep-
resents the set of information on the probability distribution that generates
random shocks.
The impact of shocks on investment and the long-term evolution of poverty
can be broken down into two pieces, the ex post effect of realized shocks
and the ex ante effect of risk. The ex post effect of shocks comes about
simply because negative events may destroy assets, knocking people off their
expected path of accumulation. For intrinsically upwardly mobile individu-
als, such shocks may delay their arrival at the upper-level equilibrium, or
occasionally knock them down from it, necessitating a period of additional
savings and asset reaccumulation.
For multiple equilibrium households, the ex post consequences of shocks
can be rather more severe. Consider the case of a household that is initially
only slightly above the Micawber Frontier. A shock that knocks it below that
frontier will knock the household into the ranks of the chronically poor,
as the household will (optimally) alter its strategy and give up trying to
reach the high equilibrium. Figure 6.3 illustrates such a case derived from the
numerical simulation of the BCI model. The horizontal axis shows time, and
the vertical measures accumulated capital stock. The two illustrated time paths
show two different histories for a household that begins with initial assets
above the Micawber Frontier. Under the dashed line trajectory, the household
avoids severe shocks (at least early on) and manages a long-term escape from

The analysis of income and marginal utility shocks is more dif¬cult, raising interesting
issues of asset smoothing discussed theoretically by Zimmerman and Carter (2003) and
analysed empirically by Hoddinott (2006).

Michael R. Carter and Munenobu Ikegami


Capital stock,kjt

Micawber Threshold, k(·)



0 10 20 30 40 50 60
Time period

Figure 6.3. The irreversible consequences of shocks

poverty. The solid line trajectory shows that the household receives a more
severe shock in year 5 and falls below the Micawber Threshold. From that
point on, the household sinks into a long-term poverty trap. Under the more
fortunate history, the household recovers and continues to move toward the
high-equilibrium steady state.
While these ex post effects of shocks are important, the anticipation that
they might take place would be expected to generate ˜a sense of insecurity,
of potential harm people must feel wary of”something bad can happen and
“spell ruin” ™, as Calvo and Dercon (2005) put it. Analysis of the BCI model
shows that this sense of impending ruin will indeed discourage forward-
looking households from making the sacri¬ces necessary to reach the high
equilibrium. Numerical analysis of the model shows that the Micawber Fron-
tier shifts to the north-east once asset risk is introduced into the model. As
shown in Figure 6.4, the solid line is the Micawber Frontier in the absence of
risk (as in Figure 6.2 above), while the dashed curve is the Micawber Frontier
in the face of risk. The boundaries marking the critical skill levels at which
households move between the different accumulation regimes also shift right
(to · L and · H ), meaning fewer intrinsically upwardly mobile households and
more intrinsically chronically poor households.

Looking Forward

Micawber Frontier
Frontier (no risk)
Initial capital endowment, k0



0.90 0.95 1.00 1.05 1.10 1.15 1.20
· · · · Intrinsic ability, ·

Figure 6.4. Vulnerability shifts out the Micawber Frontier

The most dramatic effects of risk are seen by considering a household whose
skill and capital endowments place it between the two frontiers. Consider
a household whose skill and initial asset endowments place it at the solid
circle illustrated in Figure 6.4. Absent the risk of shocks, such a household
would strive for the upper equilibrium and eventually escape poverty. In the
presence of risk, such a household would abandon this accumulation strategy
as futile and settle into a low-level, chronically poor standard of living. In the
face of asset risk, the extraordinary sacri¬ce of consumption required to try to
reach the high equilibrium is no longer worth doing, and the household will
optimally pursue the low-level, poverty trap equilibrium. Again, shocks have
their largest effects on mid-skill households.
Simulation of the BCI model provides additional insight into the impact
of shocks and risk on dynamic behaviour and chronic poverty. Consider the
following three simulations:
r A Non-Stochastic Simulation in which repeated application of the accumu-
lation rule, i — (kt |·), can be used to de¬ne the sequence of optimal capital
stocks for any individual i with initial endowments ki0 and ·i . At each
point in time, the implied income and consumption of the individual
can also be calculated.
rA Risk without Shocks Simulation in which repeated application of the
risk-adjusted optimal accumulation rule, ir— (kt |·, ), is used to de¬ne a


. 5
( 13)