Background Paper

What are the Sources of Risk and How do people cope? Insights from household surveys in 16 countries Rasmus Heltberg, Ana María Oviedo & Faiyaz Talukdar The World Bank

What are the sources of risk and how do people cope? Insights

from household surveys in 16 countries

Rasmus Heltberg, Ana María Oviedo, and Faiyaz Talukdar

November 24, 2013

Abstract: We report on a major multi-country comparison of household surveys on shocks and coping.

Natural disasters, health shocks, economic shocks, and asset loss are the most commonly reported types

of shocks. People often cope using costly responses that increase their vulnerability to future shocks. We

conclude that household survey modules on shocks and coping largely fulfill their objective of providing

information on risk exposure yet do little to inform policy beyond providing broad diagnostics.

1. Risks are important, and can be researched and documented in many ways

People in developing countries are surrounded by risks of many kinds and large numbers of people

remain vulnerable amidst rapid growth and unprecedented reduction in extreme poverty. The literature

on risk and vulnerability has established that shocks from many sources strike frequently and hit hard,

causing loss of life, assets, and livelihoods. The literature has also established that the cost of risk

exceeds the impact of shocks, including also ex-ante adjustments people make in the knowledge that

there is risk (Morduch 1995, Kochar 1995, Ligon and Schechter 2003, Christiaensen & Subbarao 2005).

Risk management tools such as microfinance, social protection, and preventative health can both

mitigate poverty and serve as a springboard to enable pursuit of productive opportunities. However, the

best design of risk management tools and the best balance between different tools and policies has been

subject to debate and research for years (Ashraf, Karlan, and Yin 2006, Duflo, Dupas, Kremer and Sinei

2006, Banerjee, Cole, Duflo, and Linden 2007).

Hopes are high that understanding the more frequent and costly sources of risk, and documenting the

detrimental coping people are often forced to take, can underline appropriate policy responses. After all,

better data have often led to advances in the understanding of the most useful and desirable risk

management tools and policies. This hope has guided much research and data collection effort. National

statistical agencies, often supported by the World Bank and other agencies, have included modules on

shocks and coping in household surveys in a large number of countries. These surveys ask respondents

about types of shocks experienced by their household within a set reporting period (usually either one or

five years), how they responded to those shocks (consumed less, worked more, borrowed, migrated,

sought assistance, and so on), and sometimes what impacts resulted (loss of income or productive assets,

for example). Surveys have been collected in normal times and after major disasters (for example Carter

and others, 2007 after Hurricane Mitch in Honduras). Such surveys of self-reported shocks are fairly

simple and cost-effective to collect and particularly favored in social protection where they have been

used to inform several reports and strategies (the World Bank’s Africa Social Protection Strategy 2012-

2022 on Managing Risk and Promoting Growth is a recent example).

Self-reported shock data is not the only source of data on risks faced by people. The academic literature

on risk and vulnerability has often relied on “natural experiments” to identify the short and long-term

impacts of shocks, for example identifying life-long health and income consequences for children born

during severe droughts and other major systemic shocks (Hoddinott and Kinsey 2001). Another strand

of the literature, starting with Townsend (1994 and 1995), has focused on informal insurance

mechanisms used by communities to smooth consumption, often relying on panel data. This literature

sheds light on the extent to which idiosyncratic risks are shared within communities, finding that

households are partially able to reduce income variability, either through informal credit from other

members of the community (Kochar 1995), or by establishing risk-sharing networks with other

community members who are not faced with the same set of shocks (Ligon 2002). Determining causal

chains and patterns of behavior are key strengths of the academic literature. Since much of this research

draws on data from rural areas, and sometimes focused on specific events such as war, epidemic, or

drought, it often does not permit the broad comparative diagnostic of risk from all sources and covering

both rural and urban areas that ideally would inform policy.

In this context, the appeal of self-reported shock data is that it allows for an empirical comparison of the

frequency with which various types of shocks occur, thereby infusing a sense of the priority of various

risks into policy discussions. There are many types of risk: idiosyncratic shocks such as health shocks,

family breakup, and some loss of employment affect households in isolated incidents and are not

simultaneously experienced by other members of the community. Systemic (or covariate) shocks such as

drought, flood, price shocks, and recession simultaneously affect all or many households in the

community. Shocks and coping modules can document the perceived occurrence of all shock types and

the many and varied ways in which households respond to these shocks.

The literature on risk and vulnerability has debated the relative importance of different sources of risk in

the lives of poor and vulnerable people in developing countries. Some authors argue that idiosyncratic

shocks are by far the most frequent and costly to people (Deaton 1997, Gertler and Gruber 2002, Udry

and Kazianga 2006). Other authors are somewhat dismissive of idiosyncratic risk, arguing that although

such shocks are fairly common, informal insurance mechanisms often allow people to manage them

relatively well, unlike systemic (or covariate) risk, for which informal mechanisms are often inadequate

(Kochar 1995, Dercon 2002, Gunther and Harttgen 2009, Hoogeveen, van der Klaauw and van Lomwel

2011). This debate has important implications for social protection and other policies aiming to assist

households avoid costly coping that damage their long term livelihoods and human capital. In the first

view, policy should focus on addressing idiosyncratic risk, for example via health insurance and long-

term social protection for the chronically poor. In the second view, there is a more acute need for policy

response to systemic risk, for example via rainfall insurance and social protection that is more

responsive to crises and natural disasters.

We report in this paper on an analysis of surveys from 16 countries in all developing regions. The

analysis was undertaken as part of preparing the World Development Report 2014 on Risk and

Opportunity: Managing Risk for Development. Although we cover 16 surveys, only 15 of them have a

shock module and only 15 of them contain questions on coping. We report major cross-country findings

emerging from this analysis on the extent of household exposure to different types of idiosyncratic and

systemic shocks, and the coping responses used by households.

We conclude that both idiosyncratic and systemic sources of risk are frequent and potentially

impoverishing. Health shocks are a universal problem, while systemic shocks—drought and other

disaster events in particular—tend to be more pronounced in rural than in urban areas. People affected

by shocks commonly respond with a mix of consuming fewer food and nonfood items, working more,

seeking credit and assistance from formal and informal sources, and relying on savings and sales of

assets.

Reflecting on the analysis, we also conclude that these survey modules largely fulfill their objective of

providing information on shocks and coping responses, although there is room to improve and

standardize survey instruments. Yet beyond providing broad diagnostics, the information culled from

these surveys largely disappoint the aspiration to inform policy. This is mostly because these surveys

shed so little light on the obstacles to risk management, that is, the reasons that people and societies

often fail to take commonsense precautions in the face of known threats. Without major innovation in

this area, shock and coping surveys will likely remain more useful for broad diagnostics than for specific

policy recommendations.

2. Data

We review and analyze household survey data from Afghanistan, Bangladesh, China, Iraq, Laos,

Malawi, Maldives, Mexico, Nigeria, Peru, Sudan, Tajikistan, Tanzania, Uganda, Uzbekistan and

Vietnam. Surveys comprise a mix of Living Standard Measurement Surveys (LSMS), budget surveys,

and special-purpose surveys designed for social protection analysis. The main selection criterion was the

availability of a module asking respondents to list shocks they have experienced.

To facilitate comparison across surveys, we constructed aggregate categories of shock types (health,

employment, price shocks and so on) and of coping responses (work more, consume less, and so on).1

The surveys were:

Afghanistan National Risk and Vulnerability Survey (2005): The sample size is 30,822

households, taken from the 2003 FAO Livestock Census using a systematic sampling with a

random start. The sample is nationally representative and the module on shocks and coping has a

recall period of 1 year.

Bangladesh Welfare Monitoring Survey (2009): The sample size is 14,000 households, drawn

from the 2001 Population Census using a two-stage cluster design. The sample is nationally

representative and the module on shocks and coping has a recall period of 1 year.

China Rural Social Protection Survey (2005): A sample of 6,165 households from the Fujian,

Gansu, Guangxi and Zheijiang provinces were surveyed using a multi-stage sampling design.

The sample is only representative of the 4 provinces it was drawn from, and the module on

shocks and coping has a recall period of 1 year.

Iraq Household Socio-Economic Survey (2006/07): A sample of 18,144 households was drawn

from the 1987 Iraq Census using a two-stage sampling design. The sample is nationally

representative and the module on shocks and coping has a recall period of 1 year.

Lao PDR Vulnerability and Shocks Survey (2008): A sample of 600 households from the

Attapeu, Phongsaly and Viantiane provinces were surveyed using a multi-stage sampling design.

The sample is only representative of the 3 provinces it was drawn from, and the module on

shocks and coping has a recall period of 1 year.

1 Annex A has a detailed description of how the categories were constructed and of survey design and questionnaire quality.

Malawi Third Integrated Household Survey (2010/11): A sample of 12,271 households was

drawn from the 2008 Malawi Population and Housing Census using a random systematic

stratified two-stage sample design. The sample is nationally representative, and the module on

shocks and coping has a recall period of 1 year.

Maldives Vulnerability and Poverty Survey (2004): A total of 2,840 households from 200

inhabited islands were surveyed using a systematic sampling with a random start. The sample is

nationally representative, and the module on shocks and coping has a recall period of 5 years.

Mexico Family Life Survey (2002): A sample of 8,440 households was surveyed using the

sampling framework from the 2002 Mexican National Employment survey. The sample is

nationally representative and the module on shocks and coping has a recall period of 5 years.

Nigeria General Household Survey 2010/11: A sample of 4,986 households from the 2006

Housing and Population Census was surveyed using a two-stage probabilistic sampling design.

The sample is nationally representative and the module on shocks and coping has a recall period

of 5 years.

Peru Encuesta Nacional de Hogares Sobre Condiciones de Vida y Pobreza (2011): A sample of

26,456 households was selected from the 2007 Population and Housing Census using a stratified

three-stage sampling design. The sample is nationally representative, and the module on shocks

and coping has a recall period of 1 year.

Sudan National Baseline Household Survey (2009): The sample size is 7,920 households, who

were selected from the 2008 Population and Housing Census using a two-staged stratified

sampling design. The sample is nationally representative and the module on shocks and coping

has a recall period of 5 years.

Tajikistan Living Standards Measurement Survey (2009): A sample of 1,500 households was

surveyed using the sampling framework from the 2005 Tajikistan Multiple Indicator Cluster

Survey. The sample is nationally representative. Contrary to the other surveys, it does not

contain a shock module. It does, however, contain a module on coping (with unspecified shocks),

with a recall period of 1 year.

Tanzania National Panel Survey (2010/11): Some 3,924 households from the 2002 Population

Census were surveyed using a multi-stage clustered sample design. The sample is nationally

representative. It does not contain a coping module; its (quite informative) module on shocks has

a recall period of 5 years.

Uganda National Household Survey (2009/10): A total of 6,800 households from the 2002

Population and Housing Census were surveyed using a two-stage stratified sampling design. The

sample is nationally representative, and the module on shocks and coping has a recall period of 1

year.

Uzbekistan Regional Panel Survey (2005): A sample of 2,948 households was drawn from a

countrywide population review conducted in 2002 using a three-stage stratified sampling design.

The sample is nationally representative, and the module on shocks and coping has a recall period

of 1 year.

Vietnam Household Living Standard Survey (2008): Drawn from the 1999 Population and

Housing Census, 45,945 households were surveyed using a multi-staged sampling design. The

sample is nationally representative, and the module on shocks and coping has a recall period of 1

year.

Table 1 presents a summary of the types of information on shocks and coping collected by each survey.

The design of the shock module varies significantly across countries.

Table 1: Typology of questions asked in shock and coping modules

Experienced

shock

Shock

timing

Multiple

shocks

Costs

(type of loss)

Costs

(currency)

Did others

experience

it

Severity Individual

who was

affected

Has the

household

recovered

Coping

type

Coping

ranking

Afghanistan x

x

x x

Bangladesh x

x

China x

x x

x x

Iraq x

x

Lao PDR x x

x x

x x x

Malawi x

x x

x

x x

Maldives x x x x x

x

Mexico x x x

x x

Nigeria x x x

x x x

Peru x

x x

x x

Sudan x

x

x

x

Tajikistan

x

Tanzania x x

x x x x

Uganda x x

x x

x

Uzbekistan x

x

x x x

Vietnam x x x x x x

7

3. Frequency and magnitude of shocks

The share of households reporting any shock varies significantly across countries; in some cases defying

explanation. Table 2 reports the percentage of households that reported experiencing a shock in any of

the eight broad categories created for comparative purposes. Disasters, asset losses, and health shocks

appear to be the most commonly reported shocks, which is consistent with most of the literature on risk.

However, the percentage of households reporting shocks seems surprisingly low in fragile states like

Afghanistan and Iraq (particularly in the category of crime and safety). In as Bangladesh, the

questionnaire design omitted disasters. A similar observation can be made from looking at Table 3,

which groups all answers about different shocks into indicators of whether the household experienced a

single shock or multiple shocks over the recall period. There is large variation in the percentage of

households reporting any shock across countries. And although surveys with 5 year recall tend to show a

higher incidence of shocks than surveys with one year recall, that difference is lower than expected. It

seems clear that not only objective variation in risk levels, but also survey design, survey

implementation, and respondents’ subjective interpretation of shocks affect the observed cross-country

patterns in Table 2 and 3.

Table 2: Percentage of all households reporting the following shocks

COUNTRY Recall

Period

Prices

(inputs,

outputs,

food)

Disasters

(natural)

Employment

(jobs, wages)

Health

(death,

illness)

Asset and

crop loss

(house,

land,

livestock)

Household

breakup

Crime &

safety Other

Afghanistan 1 year 2.2 34.2 4.4 11.5 15.8 - 4.9 -

Bangladesh 1 year - - 2.8 4.9 4.1 2.5 0.7 4.3

China 1 year - - 0.6 11.1 25.2 2.6 1.9 3.1

Iraq 1 year - - 10.1 2.0 - - 8.1 2.7

Lao PDR 1 year 5.3 18.5 5.0 25.2 23.5 0.6 3.5 -

Malawi 1 year 32.8 38.8 3.1 13.9 2.3 7.2 8.5 3.1

Peru 1 year - 6.7 4.2 8.8 - 0.8 3.3 1.7

Uganda 1 year 1.5 32.4 1.4 11.4 0.6 - 13.9 2.6

Uzbekistan 1 year - - 38.6 20.1 3.5 9.8 4.6 7.8

Maldives 5 years - - 0.9 14.5 1.6 0.8 - -

Mexico 5 years - 1.4 8.0 13.4 4.7 - 1.4 -

Nigeria 5 years 6.5 5.5 3.9 13.7 4.1 0.7 2.8 1.0

Sudan 5 years - 33.7 - 25.7 32.7 - 5.6 2.4

Tanzania 5 years 43.5 34.2 4.0 46.4 7.7 18.4 15.9 3.4

Vietnam 5 years 39.5 29.3 2.2 18.3 9.2 1.7 2.5 2.4

Note: Tajikistan (2009) does not ask questions on shocks.

Note: Countries are grouped by the survey recall period, where the first group has a recall period of 1 year while the latter group has a

recall period of 5 years.

However, looking within countries, rural households report more shocks than do urban ones: they are

more likely to report having experienced at least one shock than urban households, especially in the

lowest quintile (true in 9 countries out of 15 total for which we have shock data). Rural households also

report a higher number of shocks on average (with the exception of Maldives and Uzbekistan). The

8

urban-rural difference is particularly strong in Afghanistan, Lao PDR and Malawi. It is safe to conclude

that rural areas and rural livelihoods tend to be more risky than urban ones.

Table 3: Percentage of surveyed households reporting the incidence of single/multiple distinct shocks over the

specified recall period

COUNTRY Recall

Period

Experienced a

single shock

Experienced

multiple shocks

Mean number of

shocks reported

during recall period

T-statistic of the

difference in

means across

region

Urban Rural Urban Rural Urban Rural

Afghanistan 1 year 8.3 9.8 8.1 39.2 0.3 1.4 (-)52.6*

Bangladesh 1 year 11.1 11.5 2.9 4.4 0.2 0.2 (-)3.4*

China 1 year - 25.9 - 11.7 - 0.5 -

Iraq 1 year 9.0 8.2 8.0 6.8 0.3 0.3 3.7*

Lao PDR 1 year 22.5 36.0 11.9 36.1 0.5 1.2 (-)6.6*

Malawi 1 year 27.3 26.4 12.7 40.4 0.6 1.5 (-)37.5*

Peru 1 year 19.3 32.6 1.4 1.9 0.2 0.4 (-)14.6*

Uganda 1 year 24.1 40.6 5.6 15.6 0.4 0.8 (-)13.0*

Uzbekistan 1 year 29.8 27.7 20.9 17.7 0.9 0.4 9.1*

Maldives 5 years 19.0 16.9 4.6 1.1 0.3 0.2 4.1*

Mexico 5 years 22.4 22.2 6.7 9.4 0.4 0.5 (-)4.2*

Nigeria 5 years 17.9 18.3 5.8 12.0 0.3 0.5 (-)6.1*

Sudan 5 years 32.2 29.7 16.5 40.3 0.8 1.4 (-)19.56*

Tanzania 5 years 31.1 27.8 52.3 54.5 2.1 2.2 (-)1.02

Vietnam 5 years 34.9 27.2 24.8 39.3 1.0 1.5 (-)13.7*

Note: Tajikistan (2009) does not ask questions about shocks.

Note: The t-statistic represents the difference in mean number of shocks experienced by urban/rural regions. * denotes significance at the

1% level.

The Tanzania survey asked people whether a given shock was also experienced by other members of the

community. This permits an assessment of how covariate different shock types are (Figure 1). The

results confirm our priors: food price hikes, water scarcity, droughts, floods, and crop disease are mostly

covariate; death, illness, crime, household breakup, and business failures are mostly idiosyncratic.

However, the results also indicate that, in practice, the distinction between covariate and idiosyncratic

risk is rather graduated: most types of agricultural risk, for example, tend to affect the entire village, but

sometimes they affect many farmers, and occasionally just a single farmer (Christiaensen and Sarris

2007). Overall, six of the seven most commonly reported shocks are mostly covariate, reflecting the

agricultural nature of life in Tanzania.

9

Figure 1: Sources of shocks in Tanzania

Source: Authors based on data from the Tanzania National Panel Survey 2010/2011.

We group survey responses into broad categories that permit us to compare across countries. Price

shocks comprise input, output, and food price shocks. Disasters include drought, water scarcity for

various reasons, flood, crop disease, storms, and more. Employment shocks comprise reduced earnings

and wages and loss of job. Asset shocks denote loss of land, house, livestock, and machinery for various

reasons (of which livestock disease is one of the most common). Health shocks comprise death, illness,

accidents, and disability, while crime and safety comprise common theft and violence of all kinds.

Household breakup includes separations but also incidents involving the police and other authorities.

Table 4 shows how the incidence of these major shock types vary across countries and between rural

and urban areas.

Natural disasters, health shocks, economic shocks, and asset loss are the most commonly reported types

of shocks across countries, and can often result in the loss of life, health, property and livelihoods.

Natural disasters such as drought, water scarcity, and flooding are among the most frequently reported

type of shock in all countries for which we have data. These shocks are often the single most common

risk in rural areas. In Lao PDR, Malawi and Vietnam, almost half of the poorest rural population suffers

from disasters. Drought is the most common type of natural disaster, with flooding and crop disease also

important. Although more prevalent in rural areas, they remain high risks in urban areas as well. Where

surveys (Malawi, Nigeria and Tanzania) ask respondents to rank shocks by severity, rural households in

0 10 20 30 40 50 60

Loss of salaried employment

Loss of own land

Household business failure

Household breakup

Illness/accident of a working member

Crime

Death of a household member

Livestock death or theft

Large rise in agricultural input prices

Large fall in crop sale prices

Crop disease or pests

Drought or floods

Insufficient water

Death of an extended family member

Large rise in food prices

% of households that have experienced each type of shock in the past 5 years

Shock was mostly systemic (affected all/most other households in the community) Shock was idiosyncratic in most cases

39

12

11

27

24

30

27

34

22

8

5

3

8

5

58

10

particular tend to rank disasters highly. In rural Malawi, for example, nearly three-fifths of all

households rank disasters as the most severe shock experienced.

Death, illness, and accidents are another major risk category that ranks high in all countries with shock

data. Health shocks is the most commonly reported shock type in Maldives, Mexico, and Nigeria;

second only to natural disasters in rural India and in Peru and Uganda; and second to asset loss in rural

China. Health shocks are ranked as severe in Malawi and Tanzania, where they are the most severe

shock for more than half of the households who report them.

Price shocks are also very common: they are the most commonly reported type of shock in Tanzania and

Vietnam, and the second-most common in Malawi. This is hardly surprising since several of these

surveys were conducted during the height of the food price crises (Vietnam during the 2008 crisis and

Malawi, Peru, Nigeria and Tanzania during the 2nd

crisis in 2010/11); these numbers confirm other

studies finding widespread impacts of the food price crisis on both rural and urban households,

particularly in Africa (Heltberg, Hossain and Reva 2012; World Bank 2011).

Compared to price shocks, far fewer households report loss of employment as a major shock, perhaps

because there is so little regular employment to begin with. Employment shocks are more commonly

reported in urban areas, probably because that is where paid jobs are to be found. Loss of assets is the

fourth major type of shock looking across the countries for which we have data. A wide variety of risks

can threaten a household’s livelihood through the loss of productive assets (farm equipment, livestock or

stored harvest) and of durable goods. Asset loss is almost always more common in rural than in urban

areas, and it is the most common type of shock in rural China, Lao PDR, Mexico, and Sudan.

Other types of shocks are important only in specific countries. Crime and violence, including common

theft, are fairly common in some places such as Iraq, rural Afghanistan, and the African countries, and

relatively minor in most other places. Of course, this frequency data do not show the impact of shocks,

and there is reason to believe that crime and, in particular, collective forms of violence may have deeper

repercussions on trust, social cohesion, and business climate than many other types of risk (Petesch

2013). Household issues is a residual category that includes divorce or separation, dowry and marriage

payments, legal issues, getting arrested, and having trouble with the police. In a few places such as

Tanzania and Uzbekistan, these numbers are relatively high.

Looking across our sample, we conclude that disasters, health, and price shocks and asset losses clearly

stand out as the major types of shocks. However, in order to understand the nature of exposure that

affects the household’s reporting rate, we turn to regression analysis. We use ordinary least squares

regressions to link shock reporting rates to key household characteristics, namely: the size of the

household, the gender, occupation and education level of the household head, share of working

household members, region (rural or urban), and economic conditions measured by quintiles of per

capita consumption. We are interested in how household’s internal conditions are related to the

household’s likelihood to report shocks. We run OLS instead of a binary outcome model (such as probit)

or a latent variable model (such as multinomial probit) in order to minimize the number of imposed

assumptions, particularly on the distribution of the error term. By merely reporting correlates, OLS

provides coefficients that are more robust to specification errors.

11

We use the following model:

where ,our dependent variable, is a dummy that captures the type of shock experienced by the

household, and depends on a set of household and household head characteristics , and district-level

fixed effects . Unobservables are grouped into the error term . We run the regressions on the entire

sample of survey respondents in each country.

Table 4: Most common shocks for rural and urban households

Type of household most likely to report experiencing shock

Country Price shocks Disasters Employment

shocks

Asset/crop

losses Illness/death Safety

Household

breakup

Afghanistan Rural Rural

Rural Rural Rural

Bangladesh

Rural

Urban

Iraq

Urban

Rural

Lao PDR

Rural Urban

Malawi Rural Rural Rural Rural Rural Urban Rural

Maldives

Urban Urban

Urban

Mexico

Rural Urban Rural Urban Rural

Nigeria

Rural

Rural

Peru

Rural Urban

Urban

Sudan

Rural

Rural Rural Urban

Tanzania Rural Rural Urban Rural Urban Urban Urban

Uganda

Rural

Uzbekistan

Rural

Urban Rural

Vietnam Rural Urban Rural

Regressing a host of covariates pertaining to household (and individual level) characteristics on shock

incidence yields some consistent results across countries; the main (significant) results are listed below

and the full set of results are attached in annex B.

Household size is positively correlated with shock reporting rates across the board, as larger

households are exposed to more shocks from multiple dimensions. This is largely the case across

the typology of shocks considered with the exception of crime and safety shocks (which were

negatively or not correlated with household size for most countries).

Female headed households are more likely to report a shock, particularly in the case of health

shocks and natural disasters (Mexico is an exception).

Households whose heads are employed in agriculture report more shocks on average as agrarian

households are often exposed to or affected by a larger set of shocks than their urban

counterparts, particularly when it comes to price of inputs, outputs or staple food items, and

natural hazards. Employment shocks, on the other hand, are more frequently reported by

12

households whose heads are employed in non-agricultural sectors in some survey countries such

as Afghanistan and Uzbekistan.

The education level of the household head is negatively correlated with shock reporting rates,

particularly when dealing with asset or crop loss, in some countries.

Rural households report more systemic shocks. (there is no clear pattern for idiosyncratic

shocks).

Dwelling characteristics, particularly relating to the quality of the house, are often associated

with idiosyncratic shocks.

4. Households use a wide variety of coping responses, often not very appealing ones

Households cope with shocks in many ways, some more effective than others. Our study of coping

responses is made complicated by the fact that response categories were not uniform. We use two

different analytical strategies to achieve a degree of comparability across countries. First, we group

coping responses into comparable functional categories based on what households did, shown in Figure

2 for the countries with the most directly comparable data. The figure makes it clear both that there is

huge variation in surveys’ response categories, and that people affected by shocks commonly respond

with a mix of consuming fewer food and nonfood items, working more, seeking credit and assistance

from both formal and informal sources, and relying on savings and sales of assets.

Figure 2 Responses to shocks

Source: Authors based on data from household surveys, various years 2004–11.

0% 20% 40% 60% 80% 100%

Nigeria

Sudan

Maldives

Iraq

Afghanistan

Uzbekistan

Tajikistan

Uganda

Malawi

% of all coping responses when faced with a shock

Informal credit and assistance Formal credit and assistance Consumption reduction

Savings and sale of assets Employment or migration

13

Second, we distinguish ‘good’ and ‘bad’ forms of coping. ‘Good’ coping comprises use of savings,

credit, asset sales, additional employment and migration, and assistance (for example, from friends,

family, community members, and social safety nets); it often requires a degree of ex-ante preparation.

‘Bad’ coping comprise responses that may increase vulnerability to future shocks and include

compromising health and education expenses, productive asset sales, and consumption reductions; as

already noted, the rationale for anti-vulnerability policies such as social protection, disaster risk

reduction, and micro-insurance often center on avoiding ‘bad’ coping and the associated perpetuation of

poverty. This classification is shown in Table 5.

Using savings is a common coping strategy. Savings, as argued in the 2014 World Development Report,

is a key component of risk preparation, acting as an instrument for absorbing some of the losses

associated with hard-hitting negative shocks and reducing a household’s reliance on costly coping. Rural

households generally report a higher reliance on sales of non-productive assets such as furniture, basic

appliances and durable items than urban households do; this is most noticeable in the case of Bangladesh

and Uganda where the rural reporting rate is four times as high as with urban households. Migration is

also a common form of coping in some countries, particularly amongst rural households; this is

consistent with migratory patterns reported in most of the migration literature (Fields 1975, Stark and

Bloom 1985, Lucas 1997. In rural China and Tajikistan, every other household reports having a family

member who has had to take on additional employment after a shock, primarily through migration. The

emergence of micro-finance has paved the way for greater access to both formal and informal credit,

particularly in parts of Asia and Africa; in Bangladesh, 3 out of every 4 households hit by a shock used

credit to cope; in rural Iraq, half did so. Informal assistance, both monetary and non-monetary, from

friends, relatives, and neighbors is also common, especially in some of the African and East European

countries.

Turning to ‘bad’ (or costly) coping, reductions in food and non-food consumption are common. Some

households also rely on sales of productive assets, which often reduce future income earnings for

agricultural households; we see this particularly in the case of rural Afghanistan, Bangladesh, Nigeria,

Sudan and Tajikistan. In contrast, reductions in household expenditure on health and education are

relatively uncommon in most countries, with the exception of Tajikistan.

14

Table 5: Coping responses

Percentage of households reporting responses conditional on having experienced a shock

COUNTRY Savings

Sale of

non-

productive

goods

Employment

or migration

related

Credit Assistance

Reduction in

health and

education

spending

Reduction in

food

consumption

Reduction in

non-food

consumption

Sale of

productive

assets

Other

Afghanistan 20.9 2.4 10.1 26.1 9.6 2.8 35.3 37.9 14.9 4.3

Bangladesh - 4.0 - 77.3 - - - - 13.0 2.4

China 20.4 3.0 51.0 - - 5.5 56.3 - 10.2 -

Iraq 48.4 13.9 3.1 42.3 20.6 3.0 60.2 74.5 2.9 2.6

Lao PDR 18.4 0.7 - 2.2 4.4 - 43.8 0.6 29.8

Malawi 24.6 1.6 4.8 1.8 25.4 - 4.4 0.5 1.2 8.9

Maldives 31.1 0.7 23.0 25.2 9.7 - - - 0.9 6.6

Mexico 38.7 7.0 16.8 25.9 2.6 - 3.3 - 1.2

Nigeria 3.9 8.8 16.1 36.5 33.3 2.9 21.3 12.9 19.9 8.6

Peru 17.2 3.7 9.3 16.4 0.9 - 11.9 - 10.6

Sudan 14.3 10.4 20.4 15.2 23.6 0.6 4.2 2.8 13.7 -

Tajikistan 9.2 3.8 47.9 25.5 34.1 16.9 28.9 59.7 23.6 23.3

Uganda 33.2 3.0 23.5 12.0 21.7 - 38.7 0.5 4.7 34.4

Uzbekistan 24.8 - 14.1 26.1 15.5 0.5 0.7 0.7 13.7 52.5

Vietnam 38.3 0.6 - 12.1 20.3 - 35.2 2.7 -

Note: Tanzania (2011) has no information on coping mechanisms.

Note: We cannot distinguish between reductions in food and nonfood consumption for Lao PDR, Mexico, Peru and Vietnam.

15

We use ordinary least squares regressions to link good and bad coping to the type of shock reported and

key household characteristics, namely: the size of the household, the gender, occupation and education

level of the household head, share of working household members, region (rural or urban), and

consumption quintiles. We are interested in how households’ internal conditions affect their ability to

cope with shocks. We run OLS instead of a binary outcome model (such as probit) or a latent variable

model (such as multinomial probit) in order to minimize the number of imposed assumptions,

particularly on the distribution of the error term. Again, by merely reporting correlates, OLS provides

coefficients that are more robust to specification errors.

We use the following model:

where ,our dependent variable, is a dummy that captures the type of coping strategy used by the

household, and depends on the occurrence of any shock , a set of household and household head

characteristics , and district-level fixed effects . Unobservables are grouped into the error term . We

run the regressions on the set of households who have experienced at least one shock.

Type of household most likely to report coping strategy

Country

Use

savings/credit/

assets

Work

more/

migrate

Assistance

(government/family/community/NGOs)

Sell

productive

assets

Reduce

consumption

quantity/quality

Afghanistan Richer Richer Richer Poorer Poorer

China

Richer

Richer Richer

Iraq Richer Richer

Poorer

Malawi Richer

Mexico

Poorer

(credit/asset

sales) Poorer

Nigeria Richer

Peru Richer

Sudan Richer Poorer Poorer Poorer Poorer

Tajikistan

Richer Poorer

Uganda

Richer

(savings/sell

assets), poorer

(credit)

Richer

Poorer

Uzbekistan Poorer (credit) Poorer

Poorer

Vietnam

Richer

(savings),

poorer (credit)

Poorer Richer

16

The regression results for each survey country are attached in annex B, while the main results of note

(addressing, where relevant, the relationship between the utilized form of coping and each covariate) are

discussed in the following.

Gender of the household head: There is no strong cross-country pattern of female-headed

households using more costly coping strategies. In Afghanistan, for example, female-headed

households are more likely to use savings and sell durable goods in response to shocks; whereas

in Malawi they are more likely to increase their labor supply, sell productive assets or reduce

food consumption. In Uganda, they are more likely to resort to credit; while in Uzbekistan they

are less likely to do so (they also report using their savings and labor supply relatively more).

Occupation of the household head: The sale of productive assets is reported as a significant

coping strategy for farming households in 5 out of 12 countries. In Sudan, Uganda and Vietnam,

these households also report using their savings; while in Uganda, Uzbekistan and Vietnam they

also report having access to assistance (they are less likely to report using assistance in

Afghanistan and Nigeria). Despite the strong correlation between the location of the household

(urban or rural) and its economic activity (farming vs. other), there are some differences in the

reported coping strategies in a few countries. For instance, increasing labor supply is a common

coping strategy for rural households (typically through migration to urban areas) in 5 countries,

but in Afghanistan, Peru, and Sudan farming households are less likely to increase their labor

supply.

Household size: Larger households are more likely to rely on ‘good’ forms of coping (mostly

using savings or selling assets, and in some cases obtaining credit). In Afghanistan, Iraq and

Vietnam, however, these households are more likely to sell productive assets; while in Mexico

and Uzbekistan they are less likely to receive assistance.

Wealth: Richer households generally are more likely to use ‘good’ coping practices, although

contrasting results are observed in the case of credit (Mexico, Uzbekistan and Vietnam). Richer

households in Afghanistan, China and Iraq; and poorer households in Sudan and Uzbekistan are

more likely to increase their labor supply either by migrating or by increasing the hours worked.

Poorer households (except in China and Vietnam) are more likely to reduce the quantity and

quality of consumption. In Afghanistan and Uganda richer households are more likely to rely on

assistance from friends, family, the community, and NGOs, but this is true for poorer households

in Sudan and Vietnam. The sale of productive assets is more likely for richer households in

China and Tajikistan, and poorer households Sudan and Uzbekistan.

We conduct additional analysis for a subset of survey countries (Malawi, Nigeria and Uganda), where

we measure the conditional correlates between shocks, coping responses, and access to safety nets,

preventive measures, and financial services. These include receipt of conditional cash or food transfers,

income generating schemes and public loan schemes, access to bednets and preventive healthcare

services, and access to informal and formal credit and savings arrangements. We believe that these

indicators are suitable proxies for an enabling environment to conduct efficient risk management.

However, an important caveat is that correlates between these indicators and either shocks or coping

mechanisms cannot reveal whether such an environment causes risk management to improve or not (and

with no counterfactual such assessment is not possible). Rather, these estimates should be interpreted as

purely descriptive. The results are discussed below:

Access to safety nets: In Malawi and in Nigeria households that benefit from access to safety

nets are less likely to report increasing their labor supply. In addition, in Nigeria these

17

households are less likely to report using their savings, and reducing consumption and human

capital investment.

Access to preventive measures (access to bednets and preventive healthcare): In Malawi,

households that report using bednets also report fewer health shocks, whereas the opposite is

true in Nigeria and Uganda.

Access to financial services: In Uganda, households with access to financial services are more

likely to report greater use of credit and assistance.

5. Conclusions and recommendations for improved survey instruments

The analysis above has elucidated the types of shocks that people experience and how they respond. We

found that natural disasters, health shocks, economic shocks, and asset loss are the most commonly

reported types of shocks and often result in ‘bad’ coping responses that perpetuate vulnerability. On the

whole, we conclude that the self-reported survey modules on risk fulfill their purpose of providing

relevant information on shocks and coping. Yet we also have to conclude that the surveys leave room for

improvement and are somewhat disappointing from a policy perspective: little if any detailed insight on

how to conduct anti-vulnerability policy can be derived from these results.

An obvious room for improvement would be to set up a broadly harmonized format for shock and

coping modules to use across countries. This would help generate more complete and more comparable

data. In countries lacking panel data, suitably improved shock modules can be a strong second best for

poverty and vulnerability diagnostics. A good practice would be to ask which household members were

subject to each shock. The Vietnam and Lao PDR surveys did this by explicitly asking about shock

consequence for individual household members. Another small improvement would be to better match

the survey recall periods across modules so that, for example, information on health care and social

protection can be linked to shocks and coping. Further, more detail on post-shock credit and assistance

would be helpful. Only a handful of surveys clarify whether the source of credit used to cope was formal

or informal, a crucial distinction from a policy perspective. Sources of credit and assistance used to cope

should be included in future surveys. Likewise, access to and timing and usefulness of government and

NGO support post-shock would be useful information. Although safety nets are captured in most

household surveys, greater emphasis could be placed on how they help in coping with shocks.

However, the main problem comes from what was not measured in the surveys, and sometimes may not

even be measurable. Key details not contained in the surveys include shock frequency, damage caused,

and coping costs incurred. This information is measurable in principle, but may be hard to assess given

weak precision in respondents’ recall. Further, the surveys contain nothing on preventive measures and

risk preparation: what families did to avoid risk, such as engaging in low risk-low return livelihood

strategies. The full cost of risk is the sum of the cost of risk avoidance and the cost of coping with

shocks. We did not see a survey that attempted to assess all of these costs, nor do we know how to do so.

Actions of risk avoidance are hard to ask about in these types of surveys, and may not always be

measurable.

To address risk, preparation has to improve (World Bank 2013). Policies to reduce vulnerability need to

be rooted in an understanding not only of key risks facing the poor and near-poor, but also of the

constraints and obstacles to better risk management. These constraints operate at the level of individuals,

households, communities, enterprises, and government (World Bank 2013). The surveys offer no real

18

insights on this—and in some cases could not. For example, a stable well-paid job is ultimately the best

source of financial protection, yet household surveys are neither designed, nor able to, shed light on the

factors impeding job creation. We would argue that information on shocks and coping have served at

best to provide broad policy recommendations, for example that health shocks and natural disasters are

impoverishing and need to be addressed. They offer little on the specifics of how countries might

achieve this.

Reflecting on the apparent paradox that reported data on shocks and coping largely fulfill the intended

objective of providing a broad information base on risks and its costs but largely disappoint the

aspiration to inform policy beyond broad generalities leads us to discuss what additional pieces of

information might help to move policy forward. We contend that the nature of the public response

depends not only on the type of risk, but also on how well individuals, families, and communities

manage risks on their own and the reasons they sometimes fail to manage them. As argued by the WDR

2014, it is frequently the case that risks are fairly well-understood and that simple cost-effective steps to

address them are available and known. Yet people, families, communities, and societies often fail to

enact risk management. Seen this way, the key bottleneck for better policy design may be less about risk

information than with understanding the constraints and obstacles to better risk management, that is, the

behavioral, cognitive, social, and political reasons for apathy in the face of risk.

Therefore, we recommend greater attention to assessing people’s knowledge about risk and risk

preparation, and obstacles to risk management. Such research may well go beyond routine household

surveys and require specialized instruments and a combination of quantitative and qualitative methods.

19

References

Ashraf, Nava, Dean Karlan, and Wesley Yin. 2006. “Tying Odysseus to the Mast: Evidence from a

Commitment Savings Product in the Philippines.” Quarterly Journal of Economics 121 (2): 635-

72.

Banerjee, Abhijit V., Shawn Cole, Esther Duflo, and Leigh Linden. 2007. “Remedying Education:

Evidence from Two Randomized Experiments in India.” Quarterly Journal of Economics 122

(3): 1235-64.

Bertrand, Marianne, and Sendhil Mullainathan. 2001. “Do People Mean What They Say? Implications

for Subjective Survey Data.” American Economic Review 91 (2): 67-72.

Carter, Michael R., Peter D. Little, Tewodaj Mogues, and Workneh Negatu. 2007. “Poverty Traps and

Natural Disasters in Ethiopia and Honduras.” World Development 35 (5): 835-56.

Christiaensen, Luc J., and Alexander Sarris. 2007. “Rural Household Vulnerability and Insurance

against Commodity Risks : Evidence from the United Republic of Tanzania.” FAO commodities

and trade technical paper 10, Food and Agriculture Organization of the United Nations, Trade

and Markets Division, Rome.

Christiaensen, Luc J., and Kalinidhi Subbarao. 2005. “Towards an Understanding of Household

Vulnerability in Rural Kenya.” Journal of African Economies 14 (4): 520-58.

Deaton, Angus. 1997. The Analysis of Household Surveys : A Microeconometric Approach to

Development Policy. Baltimore, MD: Johns Hopkins University Press.

Dercon, Stephan. 2002. “Income Risk, Coping Strategies, and Safety Nets.” World Bank Research

Observer 17 (2): 141-66.

Duflo, Esther, Pascaline Dupas, Michael Kremer, and Samuel Sinei. 2006. “Education and Hiv/Aids

Prevention : Evidence from a Randomized Evaluation in Western Kenya.” World Bank Policy

Research Working Paper 4024, Washington, DC.

Fields, Gary. 1975. “Rural-Urban Migration, Urban Unemployment and Underemployment, and Job-

Search Activity in Ldcs.” Journal of Development Economics 2 (2): 165-87.

Gertler, Paul, and Jonathan Gruber. 2002. “Insuring Consumption against Illness.” American Economic

Review 92 (1): 51-70.

Gunther, Isabel, and Kenneth Harttgen. 2009. “Estimating Households Vulnerability to Idiosyncratic

and Covariate Shocks: A Novel Method Applied in Madagascar.” World Development 37 (7):

1222-34.

Heltberg, Rasmus, Naomi Hossain, and Anna Reva. 2012. Living through Crises : How the Food, Fuel,

and Financial Shocks Affect the Poor. New Frontiers of Social Policy. Washington, DC: World

Bank.

Hoddinott, John, and Bill Kinsey. 2001. “Child Growth in the Time of Drought.” Oxford Bulletin of

Economics and Statistics 63 (4): 409-36.

20

Hoogeveen, Johannes , Bas Van der Klaauw, and Gijsbert Van Lomwel. 2011. “On the Timing of

Marriage, Cattle, and Shocks.” Economic Development and Cultural Change 60 (1): 121-54.

Kazianga, Harounan, and Christopher Udry. 2006. “Consumption Smoothing? Livestock, Insurance and

Drought in Rural Burkina Faso.” Journal of Development Economics 79 (2): 413-46.

Kochar, Anjini. 1995. “Explaining Household Vulnerability to Idiosyncratic Income Shocks.” American

Economic Review 85 (2): 159-64.

Ligon, Ethan. 2002. “Targeting and Informal Insurance.” United Nations University Wider Discussion

Paper DP2002/08, World Institute for Development Economics Research.

Ligon, Ethan, and Laura Schechter. 2003. “Measuring Vulnerability.” Economic Journal 113 (486):

C95-C102.

Lucas, Robert E.B. 1997. “Internal Migration in Developing Countries.” In Handbook of Population and

Family Economics, edited by Mark Richard Rosenzweig and Oded Stark. Amsterdam ; New

York: Elsevier.

Morduch, Jonathan. 1995. “Income Smoothing and Consumption Smoothing.” Journal of Economic

Perspectives 9 (3): 103-14.

Petesch, Patti. 2013. “How Communities Manage Risks of Crime and Violence.” Background paper for

the World Development Report 2014.

Stark, Oded, and David E. Bloom. 1985. “The New Economics of Labor Migration.” American

Economic Review 75 (2): 173-78.

Townsend, Robert M. 1994. “Risk and Insurance in Village India.” Econometrica 62 (3): 539-91.

———. 1995. “Consumption Insurance - an Evaluation of Risk-Bearing Systems in Low-Income

Economies.” Journal of Economic Perspectives 9 (3): 83-102.

World Bank and International Monetary Fund (2011). Improving the Odds of Achieving the MDGs.

Global Monitoring Report. Washington, DC, World Bank; IMF.

World Bank 2013. World Development Report 2014 Risk and Opportunity: Managing Risk for

Development.