Precautionary Wealth and Financial Access: Evidence from Afghanistan 1

Transcription

1 Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized Precautionary Wealth and Financial Access: Evidence from Afghanistan 1 Leila Aghabarari, Ahmed Rostom and Rishabh Sinha 2 Households accumulate wealth as a reserve against unexpected contingencies. We employ a detailed household survey in Afghanistan to study how non-housing wealth accumulation of households varies with labor income uncertainty. We find that households facing higher income uncertainty accumulate significantly larger quantities of non-housing wealth. Exploiting variation in availability of banks and micro-finance institutions across provinces, we find lower wealth accumulation in provinces having better access to financial institutions. Keywords: Precautionary Wealth, Access to Finance, Afghanistan, Livestock, Jewelry JEL: O16, D14, D31 Public Disclosure Authorized 1 The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. This paper is part of a larger research effort under the Saving and Investment under Uncertainty (P159317) ESW that is delivered under the AFG: Navigating Risk and Uncertainty (P157288) PA. 2 Leila Aghabarari is a Consultant South Asia Region, Ahmed Rostom is a Senior Financial Sector Specialist South Asia Region (all Finance and Markets Global Practice) and Rishabh Sinha is an Economist in the Development Economics Research Group (Macroeconomics & Growth). Authors would like to thank Norman Loayza, Claudia Nassif, Aminata Ndiaye, Subika Farazi and Christina Wieser for useful comments and suggests. Authors also benefited from useful comments of participants in the World Bank Seminar Series: Navigating Risk and Uncertainty in Afghanistan: Promoting Savings and Investment under Uncertainty, that was held in Kabul. Wed, Jun 1, Corresponding author: Ahmed Rostom (arostom@worldbank.org).

2 1 Introduction The precautionary savings motive is argued to be an important driver of household savings that act as a reserve against unexpected contingencies. 2 While a vast literature has analyzed the implications of precautionary motive in different settings, with some studies indeed focusing on developing economies, there is a gap when it comes to the coverage of economies trapped in the environment of fragility, conflict, and violence (FCV). 3 This is especially unfortunate as precautionary wealth can potentially serve as an effective instrument of self-insurance in such economies, arguably more so when compared to other developing countries not being impacted by FCV. 4 Self-insurance helps households cope with frequent idiosyncratic shocks of smaller magnitude. In absence of a self-insurance mechanism, many households fare much worse by resorting to lower food intake, migration and depending on child labor. In this paper, we take a step towards bridging this gap by directing our attention to Afghanistan where a range of factors packed under the broad umbrella of FCV creates an environment of utmost uncertainty. The existence of precautionary motive has important implications for welfare. Welfare costs associated with precautionary savings are especially much higher for households that are liquidity constrained (Deaton (1991)). 5 Households in FCV economies like Afghanistan are much likely to suffer from liquidity constraints as they have limited sources from where they can borrow. Financial access remains depressed in many FCV economies. The situation is particularly stark for Afghanistan where number of bank branches per 100,000 people stands at a paltry 2.2 compared to 3.6 for the median country (Figure 1). Not surprisingly, financial inclusion remains dismal in these economies with around 10 percent of Afghans having an account at a bank of a financial institution (Demirgüç-Kunt et. al. (2014)). 6 Even this low level of inclusion is concentrated among the wealthy who are better positioned to self-insure against 2 See Marshall (1920) and Keynes (1936) for early foundations. 3 In a large cross-country study consisting of both industrial and developing countries, Loayza, Schmidt-Hebbel and Serven (2000) find evidence of precautionary saving motive. 4 We use the concept of savings and wealth interchangeably throughout the paper. Savings and wealth at any point of time refer to the stock accumulated up till then. In contrast, saving (or period saving) corresponds to flows into the accumulated stock over some time duration. In our context, period saving will pertain to annual flows. 5 Though still better than no insurance, self-insurance can generate significant private and social losses under certain circumstances. Private losses are bound to large when the idiosyncratic negative shock is of larger magnitude (for example, death of a family member). On the other hand, social losses are large when nonproductive assets are used as instrument of self-insurance. For a detailed discussion on related issues, see World Development Report (2014). 6 Including mobile accounts. 2

3 shocks. As such, instruments that can effectively help households to insure against shocks can deliver large welfare gains in FCV economies. Figure 1 7 In contrast to the welfare costs of self-insurance, it is tempting to see a precautionary motive to have a positive impact on economic growth through investment. Yet, there is limited evidence of saving leading to growth in the macroeconomic context (Deaton (1989)). Deaton suggests that the missing link between saving and growth is because not all saving gets translated into productive investment. Looking at saving practices in Afghanistan, with jewelry and livestock being assets of choice, Deaton's claim seems reasonable. Even in the event that saving leaves household via informal channels, it is hard to argue that saving gets transformed into productive investment. Within this context, the central objective of our inquiry is to ascertain whether the precautionary savings motive is a significant driver of wealth accumulation in the country or not. To answer this, we identify the nature of the relationship between wealth and labor income uncertainty of households (Guiso, Jappelli, and Terlizzese (1992), Carroll and Samwick (1997, 1998) etc.). We use the data from the Afghanistan Living Conditions Survey (ALCS) (Central Statistics Organization (2016)) which is large sample survey covering more than 20,000 households and includes information about their income and expenditure. While the ALCS has been conducted sporadically for more than a decade now, the cross-section is of particular importance with respect to the question at hand. The cross-section was 7 Source: World Development Indicators. Figures refers to Sample contains 24 countries with available data that feature in the World Bank's List of Fragile Situations FY17 (World Bank (2017)). 3

4 designed to capture household wealth holistically and contains information on ownership of gold and silver, together with livestock owned by households. Including such forms of assets is important as it has been well-documented that precious metals in the form of jewelry together with livestock form a vital share of household wealth in poor economies (Rosenzweig and Wolpin (1993), Townsend (1995), Alderman (1996), Fafchamps, Udry, and Czukas (1998), Park (2006)). More significantly, these assets might bear singular importance in protecting against future shocks as they are a better store of value than cash which can lose value quite quickly due to high inflation, or in an extreme scenario due to the complete breakdown of the banking system. Additionally, jewelry and livestock are more liquid compared to other real assets like a house or land and can readily serve as collateral if a loan needs to be taken out. Our preferred measure of wealth reflects the value of jewelry and livestock held by the households, net of any debt. The measure also accounts for any net saving, calculated as excess income over expenditure reported over the previous year, to account for any lag in converting saving to wealth given the volatile conditions in the country. 8 Next, we measure labor income uncertainty by measuring the cross-sectional variation in household income after classifying households into different groups depending on the sources of household income. The variation in labor income uncertainty is derived from the degree of income variability associated with different occupations (Skinner (1998), Dardanoni (1991)). We note that our measure of income uncertainty is limited by the cross-sectional nature of the ALCS data and hence accounts for only a share of the total uncertainty associated with labor income. We find a strong positive relationship between our preferred measure of household wealth and labor income uncertainty. To see if this result is driven by our definition of wealth, we also perform the analysis for many different wealth measures. The relationship between wealth and uncertainty, though noisy for these other measures, is still positive. There exists a challenge when constructing any measure of wealth that some forms of wealth which should have been included in the measure go unaccounted for due to lack of information or due to the definition of wealth imposed. To address this, we also look at the relationship between annual household expenditure and labor income uncertainty. We find further strong support of the existence of precautionary saving motive using this specification. Households with higher income uncertainty indeed report making a lower annual expenditure. 8 Although our measure of wealth does not account for housing wealth explicitly, we control for various forms of house ownerships in our analysis. 4

5 We next turn our focus on the state of financial development in Afghanistan and ask whether the presence of formal financial institutions bears association with the precautionary savings motive. The exercise attempts to shed light on whether financial access can provide effective insurance against certain income shocks weakening the precautionary motive. 9 Unfortunately, the ALCS does not contain information on financial participation. To overcome this challenge, we exploit the variation in availability of financial institutions (bank branches and microfinance institutions (MFIs)) across the Afghan provinces. We have two main findings here. First, we find that households in provinces with a relatively lower number of bank branches per person accumulate up to 2 to 4 times more wealth, depending on the measure of wealth used. Nonetheless, a challenge with interpreting lower wealth accumulation in provinces with better financial access is the fact that our measure of wealth does not include financial wealth. Hence, it is possible that households in provinces with better financial access simply substitute their wealth in the form of jewelry and livestock with financial wealth with no change per se happening in the aggregate wealth levels. Nonetheless, we find evidence of lower annual expenditure by households in provinces with lower financial access indicating higher periodsaving. This observation casts doubt on the substitution argument. Specifically, households with lower financial access spend around 60 percent less than their counterparts in provinces with higher branches per person. Second, though financial access is associated with lower wealth accumulation, we find that wealth continues to be positively correlated with income uncertainty. In other words, the inclusion of financial access variable is not enough to break the strong relationship between wealth and uncertainty. In regards to the effect of financial access on precautionary wealth, we find that counterfactually changing the financial access/development status of households residing in provinces with worse outcomes decreases the holding of non-housing wealth by 34 percent for the aggregate economy in our most conservative estimates. Finally, we also show that the reduction in wealth accumulation in provinces with better financial access operates via the uncertainty channel further providing evidence in favor of precautionary motive. To verify the importance of uncertainty mechanism, we interact income uncertainty with the financial access variable and find that the coefficient on lower financial access is larger with the difference between the two coefficients being significant. We conclude our analysis by performing three robustness tests on our findings. A potential concern with including livestock in the analysis is that it may contribute to household 9 Beck and Demirgüç-Kunt (2008) suggest that financial inclusion and outreach at the household level can have a meaningful impact on development goals (for example, poverty reduction). 5

6 utility via the consumption channel. Though our analysis of period saving rules out the possibility that the evidence of precautionary motive is driven by inclusion of livestock in our wealth measure, we perform another set of regressions in which we only consider the value of gold and silver held by household as wealth. The trade-off from using this restricted concept of wealth is that we end up dropping many households as ownership of jewelry is highly concentrated. Nonetheless, we find that the main findings of the paper are robust to using a narrower definition of wealth. Second, we check the sensitivity of our results to the exclusion of Kabul households given that financial access in the province is significantly higher than other provinces. Second, we test for the sensitivity of our results to alternative definition of financial access. We also admit the possibility of endogeneity in our regression specifications. It is likely that at least some share of wealth accumulated by households are used productively to derive income (or consumption utility) and hence not only affect the level of income but the its variation as well. Yet, we find the evidence of precautionary motive to be a useful starting point and hope to estimate causal estimates in our future endeavors. The rest of the paper is organized as follows. We begin by giving a brief description of the data and the construction of the main variables. Following, we outline our econometric specifications, report our findings and check for the robustness of our findings. We conclude the paper summarizing the main findings and the policy relevance of the study. 2 Data We use data from many sources to conduct our analysis. Essential to study the question at hand, we use household level data from the ALCS to construct measures of labor income uncertainty and household wealth. The ALCS is a large-sample household survey conducted by the Central Statistics Organization of Islamic Republic of Afghanistan (CSO) which is representative at the province level. The detailed nature of the ALCS data also allows us to control for many household level characteristics that are considered important for saving decisions. The ALCS data unfortunately does not contain information on financial participation of households. To measure financial access of households, we use two additional data sources. We use data from Da Afghanistan Bank (DAB) and Microfinance Investment Facility Support for Afghanistan (MISFA) to aggregate number of bank branches and Micro Finance Institutes (MFI) at the province level. Next, we use population data from the Central Statistics Organization (CSO) to normalize financial access by population. Finally, we use data from the (2016) to control for terrorism related events at provincial level. 6

7 2.1 Wealth The ALCS has been conducted sporadically for more than a decade now. Yet, the cross-section is of particular importance as this round was designed to capture household wealth holistically. This round contains information on ownership of gold and silver, together with livestock owned by households. As mentioned before including such forms of assets is important given the importance of such forms in aggregate household wealth in poor economies. The round of the ALSC covers more than 20,000 households. To construct our measure of wealth, we aggregate the value of gold, silver, and livestock owned by households less any household debt. We further add any excess saving of previous year which is estimated as income in excess of expenditure. Our motivation for including saving of previous year is to account for any lag in converting such savings to wealth. Nonetheless, we build two additional measures for the wealth excluding saving from previous year. The first alternative to our preferred measure of wealth excludes the saving from previous year and considers only the contribution of gold, silver and livestock less any household debt (GSL Debt). The second alternative wealth measure GSL further excludes the reported debt of households. Around two-thirds of the households in our sample report any ownership of livestock. Livestock includes ownership of a comprehensive list of animals including cattle, oxen, yaks, horses, camels, goats, sheep, chickens. However, chicken and sheep are the most common type of livestock that households hold. The ALCS provides us with exact number of each type of livestock owned by households. We obtain an average price for each livestock in 2014 and calculate the total value of livestock using a common average price for all households. 10 There is a significant heterogeneity in the livestock ownership across provinces. Ownership of livestock ranges from about 90 percent in Paktika and Khost to 39 percent in Herat and 25 percent in the more urban Kabul province. In contrast to livestock ownership, only 14 percent households report owning either gold or silver. The ownership of gold or silver is relatively higher in Kunarha and Khost provinces where around 40 percent of households report having some amount of gold and/or silver. Interestingly, livestock ownership is also relatively higher in these provinces. Yet, in many provinces there is almost no ownership of precious metals. Ownership is less than a percent in Baghlan, Ghor and Zabul with no household in the sample from Nuristan and Daykundi reporting any ownership of gold or silver. 11 Figure 3 shows the average value of jewelry and 10 See appendix Table C for more information on price data. 11 See appendix Table B for the variation in livestock and gold/silver ownership across provinces. 7

8 Paktika Khost Ghazni Herat Wardak Parwan Kunarha Nangarhar Kabul Laghman Balkh Nimroz Jawzjan Faryab Paktya Kunduz Kapisa Panjsher Badakhshan Takhar Helmand Farah Sar-e-Pul Logar Bamyan Kandahar Samangan Badghis Urozgan Baghlan Zabul Daykundi Ghor Nooristan livestock owned across provinces. Though livestock is more widely held, we find that it accounts for only a small share of the combined wealth. Figure Average Value of Jewelry and Livestock (Provencial Level) Gold and Silver (in AFN) Livestock (in AFN) Around half of the households in the sample report having any debt and in line with assets ownership, there is a huge variation in debt accrual across the provinces. Almost every household in Ghor in our sample reported having a debt obligation with debt ownership being high in provinces of Kunarha and Urozgan as well. In sharp contrast, households in Paktika and Fryab had a minimal debt obligation. Urozgan and Nangarhar are on top of the list in terms of actual value of debt owed. Finally, in figure 4 we report the average annual income and expenditure which we use to estimate annual saving across the provinces. Average annual income ranges from more than AFN 200 thousand in Kabul and Khost to just around AFN 50 thousand in Ghor and Daykundi. In line with income, Kabul also has highest level of annual expenditure. Average expenditure is particularly high in Helmand and lies close to the average annual income resulting in a low saving rate. Daykundi and Ghor have the lowest level of average expenditure in line with lower average income observed in the provinces. 8

9 Khost Helma Ghazni Faryab Herat Kanda Logar Badak Nimroz Paktya Kunduz Nooris Balkh Panjsher Nanga Jawzjan Parwan Bamyan Kunarha Laghm Paktika Sar-e- Kapisa Takhar Urozgan Badghis Saman Zabul Baghlan Farah Wardak Ghor Dayku Figure 4 Average Annual Income and Expenditure (Provincial Level) Average Annual Income (in AFN) Average Annual Expenditure (in AFN) Though our measure of wealth does not include housing wealth, in part due to lack of information needed to infer housing values, we do control for house ownership in our analysis. Data from the ALCS not only allows us to identify if a particular household owns a house or not, but also the nature of house owned. Houses in Afghanistan range from temporary shacks and tents to the traditional durable forms. We list the distribution of house ownership in table D of appendix. 2.2 Labor Income Uncertainty Next, we discuss the construction of the measure of income uncertainty using crosssectional variation in household income. We classify households into various groups according to their source of incomes. However, in most cases the household income is derived from multiple sources. This is not only due to more than one member contributing to the household income but also because of a single household member being deriving income from multiple sources. In the ALCS survey households report their three main sources of income. To derive our measure of labor income uncertainty, we use a more attentive approach by including all three sources of income instead of relying on the main source contributing to household income. In regards to the sample selection, we exclude the households that did not report their income and those with their heads below 18 or above 64 years old. Also, we eliminate those households in which the household head is retired or unemployed. We further exclude those households for which either of the three main sources is derived from non-labor sources. These sources of income relate to pension earnings, rental income, Zakat, and borrowing. Table A in the appendix lists the 28 different sources of household incomes which are essentially the occupations from which this income is derived. Production and sale of field-corps (non-opium) is the most common source of income and is the main source of income of 21 percent of the households in the sample. On the other 9

10 extreme, road and building construction has limited representation and is reported as the primary income source by 0.01 percent of the households. The average annual income in our sample is AFN 124K while the median household income earned stands at AFN 93K. Across provinces, households in Kabul and Khost have the highest average level of income (AFN 235K and AFN 211K respectively) while households in Daykundi and Ghor have the lowest average income of around AFN 50K. The next step is to classify households into groups based on their three main sources of income. We group households that report the same first, second and third source into a group. Note that to be in the same group, it is not only required that two households have the same three occupations as the three main sources of income but also that the relative contribution of each occupation relative to others is the same for the two households as well. We follow a more restrictive classification to separate households that may have same sources but may differ in how important the source with higher variability source is to the household income. To reduce noise in income variance associated with a group, we drop groups containing less than 40 observations. We end up with 13,157 households classified into 84 different occupational groups for the final analysis. Table B reports the distribution of our selected sample of households across the Afghan provinces. 2.3 Financial Access A second point of inquiry of the paper is to find out whether financial access has some association with lower precautionary wealth. The household survey data from ALCS covers households residing across all 34 provinces in Afghanistan, which is the finest level of geographical location of households that we can obtain. Unfortunately, the household dataset does not contain information about financial access of the families. To overcome this issue, we use another dataset containing information on the number of bank branches in regions provided by Da Afghanistan Bank (DAB) to exploit the variation in bank branches across provinces to serve as a proxy of financial access. This data gives us information about all types of bank branches in provinces. We choose active full- and limited-service bank branches as these are directly related to financial services provided to households. In addition, we use data on Micro Finance Institute (MFI) branches from Microfinance Investment Facility Support for Afghanistan (MISFA). Finally, we obtain dataset on the population in provinces in Afghanistan from the Central Statistics Organization (CSO) to adjust the branch availability data by 10

11 population residing in the provinces. We estimate the financial penetration of a provinces as the number of banks branches and MFI branches available per 10 thousand people. Figure A in the appendix shows the financial penetration considering banks branches and MFI branches separately. In Figure 5 shows the combined map of bank and MFI branches per 10,000 people across the provinces in Not surprisingly, Kabul is on top of the list with the with highest number of bank and MFI branches. Balkh and Nimroz are other provinces with higher combined financial penetration. Penetration, on the other hand, is much lower in Daykundi, Ghor and Paktika provinces. Yet, the case of Nuristan is still extreme with no branches, either bank of MFI, being available in the province. Figure 5 Number of Bank and MFI Branches per 10,000 People in 2014 Nuristan Ghor D aykundi Paktia (4,5] (3,4] (2,3] (1,2] (.5,1] [0,.5] 3 Econometric Specification 3.1 Evidence of Precautionary Wealth Accumulation We apply cross-sectional analysis to investigate the existence of precautionary wealth in Afghanistan. Employing the OLS estimators, we use the following specification to investigate whether household wealth accumulation differs systematically with the labor income uncertainty. As discussed previously, the primary explanatory variable of our analysis is the income uncertainty that we estimate using households income sources. These income sources map to the three principal occupations through which a household derives its labor income. We start with the following regression specification to identify the relationship of income uncertainty with wealth accumulation: 11

12 log(y igp ) = α + ɸ p + b 1 log(income_variance g ) + b 2 log(income i ) + c 1 X ig + θ p + ε igp (1) In the above regression equation, Y represents the household wealth held by household i, in occupation group g located in province p. Our preferred measure of wealth is what we term the Non-Housing Wealth which is the aggregate value of gold, silver, livestock and the estimated saving of the previous year net of any debt obligation. We use two more measures for wealth that excludes estimate of previous year saving the total value of gold, silver, and livestock (GSL), and the value of gold, silver and livestock net of debt (GSL-Debt). We also look into the effect of income uncertainty on household expenditure to alleviate concerns of measurement errors associated with wealth. A negative association between household expenditure and income uncertainty (translating to positive association between saving and uncertainty) will act as a secondary test confirming the existence of precautionary motive in the country. Note that a negative relationship between period saving and income uncertainty is a very strict test of precautionary motive. The buffer-stock models of saving predict a negative relationship between wealth and uncertainty, with the negative relationship between period saving and uncertainty vanishing once households reach a threshold aggregate wealth. 12 In the context of our analysis, an underlying assumption is that the households in our sample have not yet accumulated the threshold level of wealth for the relationship between period saving and uncertainty to break down. The primary independent variable is the income_variance that is the variance of income associated with the income group g that a household belongs to. We are aware that not all the income variance within a group is driven by fundamental nature of income sources that are independent of worker specific factors, including differences in within group human capital. Thus, we first regress the household income on a bunch of individual level controls such as age, education, and gender for each group g. We then use the logarithm of the variance of the collected error terms to be our measure of income uncertainty. Given the cross-sectional nature of our data and following the approach by Dardanoni (1991) we assume that the current variability of within group income is a proxy for the future income variability. In all regressions, we directly control for the household income to net out the nonprecautionary wealth accumulation associated with higher income. In addition, we add other 12 In the buffer stock models, the households make positive period saving till they reach a threshold level of wealth. Once this threshold is reached, the household savings behavior is independent of any further precautionary accumulation. 12

13 controls that are considered important for wealth accumulation. The controls related to both characteristics of the households and the household head. Specifically, X ig is the vector of controls including the size of the households, the number of male members in each household, type of the dwelling, age, education, marital status, gender of the heads. There are two additional controls that we introduce at the province level φ p captures the number of terrorist events in province p whereas θ p are the province-level dummies. The coefficient of interest in this specification is b 1. A positive estimate of b 1 will suggest an existence of precautionary wealth accumulation in Afghanistan where households facing higher income uncertainty build up larger reserves to tide over future income shocks. 3.2 Impact of Access to Banking Services on Precautionary Wealth The next set of analysis is to test whether presence of financial institutions is associated with a decline in precautionary wealth accumulation. To make progress, we adopt the provincial level variation in presence of financial institutions. Specifically, we use the number of bank branches and MFI presence adjusted by provincial population to measure financial penetration. We estimate the coefficients in the following regression: log(y igp ) = α + ɸ p + b 1 log(income_variance g ) + b 2 log(income i ) + b 3 FA p + c 1 X ig + θ p + ε igp (2) Like in the previous regression specification, b 1 captures the relationship between wealth and income uncertainty which we expect to be positive if precautionary motive is active in the economy. In addition, b 3 is the coefficient on FA which is a dummy that takes the value of one for provinces p in which bank branches per person is lower than the mean across the provinces. Hence, FA is a dummy capturing low financial access. When using the MFI data, the FA takes the value of one if there are no MFIs present in the province. 13 A positive estimated value of b 3 will imply that households in province with low financial access end up accumulation more wealth. Apropos, an argument can be made that the presence of financial institutions aid households in alleviating the impact of future income uncertainty as they may 13 We don t look for the marginal impact of MFIs versus the banks. The reason is that there is a very high correlation between the two and there are only provinces that have the banks but not the MFIs. Considering that we have controls for fixed effects at the provincial level, this specification will not yield any useful information. 13

14 rely on borrowing instead of own savings to smoothen out consumption in the event they receive a negative income shock. Yet, it is possible that factors orthogonal to precautionary motive lead to a lower wealth accumulation in provinces with better financial access. To confirm whether any reduction in wealth accumulation seen in provinces with better access is indeed driven by precautionary channel, we estimate one final equation by interacting the income uncertainty term with low financial access term as in the following regression: log(y igp ) = α + ɸ p + b 1 log(income_variance g ) FA p + b 2 log(income i ) + c 1 X ig + θ p + ε igp (3) This regression can shed some light on whether any observed decline in wealth accumulation in provinces with better access is potentially due to diminishing precautionary needs of households. A higher positive value of b 1 (FA = 1) compared to b 1 (FA = 0) will imply that wealth accumulation by households in provinces responds more elastically to changes in income uncertainty compared to their counterparts in other provinces. Having outlined our empirical methodology, we present the main results of our analysis in the next section. 4 Empirical Results 4.1 Relationship between Income Uncertainty and Wealth Accumulation We begin by investigating the existence of the precautionary wealth in Afghanistan using regression (1). Table 2-A lists the results of this specification. We report the results for four dependent variables. The first three (columns (1) to (12)) correspond to wealth accumulated by the households and the last dependent variable (columns (13) to (16)) relates to annual expenditure in the year prior to survey. We find that income variance is positively correlated with Non-Housing Wealth (columns (1) to (4)) and the relationship is highly significant. The coefficient remains positive and highly significant even after we add controls for household and household head characteristics, conflict related events at province level and province-level fixed effects as seen in columns (2) to (4). This result implies that households that face higher income uncertainty end up accumulating a higher amount of non-housing wealth. We again note that though the wealth measure does not contain the value of housing wealth, we control for the nature of house ownership when we add household level controls in the regression. The point estimates show that a one percent increase in income variance is associated with a 17 percent increase in household wealth accumulation on average. In columns 14

15 (5) to (12), we repeat the exercise for the two other wealth measures that exclude last year saving - GSL and GSL-debt. Although sometimes noisy, the positive coefficients of income variance on the two alternative measures of wealth confirm our earlier findings. There exists a challenge when constructing any measure of wealth. It is possible that some forms of wealth which should have been included in the measure go unaccounted for due to lack of information or due to the definition of wealth imposed. To address this, we also look at the relationship between annual household expenditure and labor income uncertainty. The first four columns of table 2-B report the nature of the association between annual expenditure and income uncertainty. We find an adverse and significant effect of income-variance on spending implying that households who face more uncertainty about their income spend less (and save more) compared to other similar households with a lower level of uncertainty. Households on average spend 0.85 percent less when their income-variance increases by 1 percent. This negative association between expenditure and income uncertainty provides further evidence in favor of existence of precautionary wealth accumulation in Afghanistan. We further build a measure of stream of savings, which is the income net non-durable expenditures. We show the results for the measure in the last four columns of table 2-B. The results suggest that households save 2.86 percent more when their variance of income rises by 1 percent. 4.2 Relationship between Financial Access and Wealth Accumulation Our second inquiry relates to the state of financial development in Afghanistan. We are interested in finding whether the presence of financial institutions has an effect on the precautionary wealth accumulation. To answer this query, we employ the regression equation (2). Table 3 shows the results of this regression for the same four dependent variables. For brevity, we report results after including all controls and province-level fixed effects. Considering financial penetration using the number of bank branches per 10,000 people, we find that wealth accumulation is significantly higher for households located in provinces with low financial access. This is captured by the positive and highly significant coefficient on Low- Financial-Access in columns (1) (3). Households in provinces with low financial access on average accumulate almost four times as much non-housing wealth as households in other provinces. The total value of gold, silver, and livestock held by households in provinces with low financial access is more than twice as much as accumulated by households elsewhere. The relationship is statistically significant even after we net out the value of jewelry and livestock by debt obligation. 15

16 Interpreting this negative association between wealth and financial access as evidence of financial access weakening precautionary savings motive is not straightforward. The measure of wealth used in our analysis does not include financial wealth of households. Hence, it is reasonable to argue that households in provinces with better financial access simply substitute their wealth accumulation to financial wealth with no changes happening at the aggregate. To test this claim, we again rely on annual expenditure of the households. Column (4) of the table 3 shows that households with lower access to banking services spend around 70 percent less. This translates to more saving than similar households in the locations with better financial access. In the second panel of table 3, we report the results of the same exercise when we proxy financial access by presence of MFIs in a province. Household accumulation of non-housing wealth is around 2.5 times as large in provinces with no presence of MFIs compared to households in provinces with at least one MFI branch. The difference in wealth accumulation is higher for the same households when other measures of wealth are considered and the difference in wealth accumulation is highly significant as well. Like in the previous case, we again find no evidence of wealth substitution with expenditure for households in provinces with no MFI being 42 percent lower compared to households in other provinces. The coefficients on income variance remain positive for all wealth measures, though they are only significant when wealth includes saving from last year. This suggests that though financial access is positively correlated with lower wealth accumulation, it is still not enough to break the link between income uncertainty and wealth. As such, financial access might help households insure against labor income shocks, their success is somewhat limited. We perform a simple counterfactual exercise is to estimate the change in aggregate wealth when households in provinces with low financial access are provided with better access. To do this, we measure the change in predicted aggregate wealth in the base regression from the predicted aggregate wealth in which we do not include the impact of the low financial access (FA) dummy. Our most conservative estimate shows that aggregate non-housing wealth declines by 34 percent. Finally, we investigate whether the lower wealth accumulation in provinces with better financial access is driven via the precautionary channel or not. It is possible that factors other than precautionary motive lead to a lower wealth accumulation in provinces with better access. To see if this is the case, we interact income uncertainty with the financial access variables as shown in equation (3). Table 4 reports the results of the exercise. The coefficients on the 16

17 interacted terms are all positive and significant when we consider our preferred measure of household wealth. More importantly though, the coefficient on the interacted term shows a larger wealth accumulation in provinces with low financial access at the same level of income uncertainty. The bottom section of the table reports the results of the test of equality of the coefficients across the two province types. We find that the difference between the two coefficients are always highly significant. The coefficient associated with lower financial access is still higher when we use MFI presence as the measure of financial access. The point estimates on the interacted terms when alternative measures wealth measures (GSL and GSL-Debt) and expenditure are considered are noisy and often not significant. Yet, we always find that the effect is larger in magnitude for households residing in provinces with worse access and difference between the coefficients always highly significant. 5 Robustness Measurements errors are likely to be a concern with any wealth measure. To address the challenge arising from incorrectly measuring wealth used to smooth shocks, we construct three different measures of wealth given the available household data. Our main findings are robust to using all wealth measures, though we find the point estimates to be noisier in certain specifications. A systematic inverse relationship between expenditure and income uncertainty lent further support to the existence of precautionary wealth accumulation in Afghanistan. In this section, we further perform other robustness tests on the main results presented in the earlier section. In this section, we briefly discuss other robustness checks that we perform on the results presented earlier. First, we test whether the main results of the paper are driven by Kabul. Kabul is the most populated province of the country and a considerable number of our sample households reside in Kabul. Kabul also is an outlier in regards to the presence of financial institutions. To check if including Kabul is indeed critical to deliver the main results presented earlier, we perform our analysis by dropping all households residing in Kabul. Table 6 reports the estimates of regression equation (1) obtained using the restricted data. The positive association between the wealth measures and income uncertainty is preserved. The point estimate of b 1 when Non-Housing Wealth is considered is very similar to the estimated value obtained using the benchmark data. The estimate also remains highly significant. Like in the benchmark, the estimate is noisy but positive for other wealth measures. In terms of magnitude it is somewhat smaller compared to what we estimate including the Kabul households. Though 17

18 still negative, the relationship between expenditure and income uncertainty is no longer significant. Table 7 we report the results after controlling for financial access. We still find low financial access to be positively and negatively correlated with wealth and annual expenditure respectively. The relationship is highly significant across all specifications of wealth and expenditure when bank branches per 10,000 people serve as a proxy of access. However, the point estimates are lower than the benchmark estimates. The association is much weaker compared to the benchmark when MFI presence is used as a proxy of access. Finally, we reestimate the regression equation (3) and report the results in Table 8. Except when GSL-Debt is used as a measure of debt, we find higher wealth accumulation and lower expenditure in households located in provinces with lower bank branches per 10,000 people compared to households elsewhere at the same level of income uncertainty. The robustness checks for MFI presence, like in the case of regression (2), are much less stable. In another robustness check, we use an alternative definition of provinces associated with low financial access. Specifically, we now characterize provinces in the bottom 30 percentile of financial access as having a low access staus. We report the re-estimated coefficients of regression equations (2) and (3) in Table 9 and Table 10 respectively. The results in Table 9 reconfirm the positive and negative association of low financial access with wealth and expenditure respectively. In addition, the relationship is highly significant across all variables. Table 10 shows that though expenditure is not higher for households in low access provinces at the same level of income uncertainty, the earlier findings of higher wealth accumulation in low access provinces is still robust to this alternative specification of low access. 6 Conclusion We exploit a unique dataset from the ALSC survey to examine the association between income uncertainty and wealth accumulation by households in Afghanistan. Our analysis finds 18

19 evidence of the existence of precautionary saving motive in the country. Quantitatively, a one percent increase in household income variance is associated with a 17 percent increase in nonhousing wealth of the household. To address the challenge posed by measurement errors associated with wealth measures, we also conduct a stricter test of existence. We estimate the nature of the relationship between period saving and income uncertainty by analysing household annual expenditure in the year previous to the survey year. We find that household expenditure decreases by 85 basis points on average with a one percent increase in income variance. Additionally, we also try and gauge the success of financial institutions in palliating the precautionary savings behavior of the households. If financial institutions are able to provide relief against negative shocks, then households will have lower needs to self-insure via accumulated wealth. Consequently, we utilize the heterogeneity in presence of financial institutions across provinces in Afghanistan to isolate the relationship between financial access and wealth accumulation. We find that wealth is on average two to four times larger for households that are residing in provinces with worse access to finance. We are also able to provide some evidence that increase in wealth accumulation in low financial access provinces operates via diminishing precautionary motive as the households in low access regions accumulate higher wealth at the same level of income variance. Given the current low levels of financial access in Afghanistan, our analysis provides support to place financial market interventions among the development priorities. Policymakers should also be cognizant of changing the landscape of financial markets with leaps in technology creating newer instruments. Apropos to expanding formal banking, mobile technology appears to have certain advantages over other policy interventions. The digital nature of the service can be effective in countering risks associated with more traditional banking interventions that require physical exposure in a violence-ridden environment characterizing FCV economies. Lack of infrastructure in the country spell cost-savings for mobile banking vis-a-vis other options as well. The success of M-Pesa mobile transfer program in Kenya in increasing formal banking access and encouraging price competition among rival firms is a model worth exploring for Afghanistan (Mbiti and Weil (2014)). Furthermore, digital banking can potentially create positive externalities outside banking access by creating opportunities for innovation and giving increased access to women in labor markets, among many other development goals (World Bank (2016)). 19

20 References Alderman, Harold. "Saving and Economic Shocks in Rural Pakistan." Journal of Development Economics 51.2 (1996): Beck, Thorsten, and Asli Demirgüç-Kunt. "Access to Finance: An Unfinished Agenda." World Bank Economic Review 22.3 (2008): Browning, Martin, and Annamaria Lusardi. "Household Saving: Micro Theories and Micro Facts." Journal of Economic Literature 34.4 (1996): Carroll, Christopher D., and Andrew A. Samwick. "How Important is Precautionary Saving?" Review of Economics and Statistics 80.3 (1998): Carroll, Christopher D., and Andrew A. Samwick. "The Nature of Precautionary Wealth." Journal of Monetary Economics 40.1 (1997): Central Statistics Organization. Afghanistan Living Conditions Survey , National Risk and Vulnerability Assessment, Kabul, CSO (2016). Chipchase, Jan, Mark Rolston, Cara Silver, and Joshua Blumenstock. "In the Hands of God: A Study of Risk and Savings in Afghanistan." Institute for Money, Technology & Financial Inclusion (2013). Dardanoni, Valentino. "Precautionary Savings Under Income Uncertainty: A Cross-Sectional Analysis." Applied Economics 23.1 (1991): Deaton, Angus. "Saving in Developing Countries: Theory and Review." World Bank Economic Review 3, (1989): Deaton, Angus. Saving and Liquidity Constraints. Econometrica 69, (1991): Demirgüç-Kunt, Asli, Leora F. Klapper, Dorothe Singer, and Peter Van Oudheusden. "The Global Findex Database 2014: Measuring Financial Inclusion Around the World." (2015). Dynan, K.E. How prudent are consumers? Journal of Political Economy 101 (1993): Fafchamps, Marcel, Christopher Udry, and Katherine Czukas. "Drought and Saving in West Africa: Are Livestock a Buffer Stock?" Journal of Development Economics 55.2 (1998): Guiso, Luigi, Tullio Jappelli, and Daniele Terlizzese. "Earnings Uncertainty and Precautionary Saving." Journal of Monetary Economics 30.2 (1992): Keynes, John Maynard. General Theory of Employment, Interest and Money. London: MacMillan, (1936). Loayza, Norman, Klaus Schmidt-Hebbel, and Luis Servén. What drives private saving across the world? Review of Economics and Statistics (2000): 82(2),