WPS8453 Policy Research Working Paper 8453 Conflict and the Nature of Precautionary Wealth Leila Aghabarari Ahmed Rostom Rishabh Sinha Finance, Competitiveness and Innovation Global Practice & Development Research Group May 2018 Policy Research Working Paper 8453 Abstract The paper uses a detailed household survey to document However, the constituents of precautionary wealth vary precautionary wealth accumulation in Afghanistan, with drastically. While households in the low-conflict regions wealth being significantly higher for households facing rely almost exclusively on livestock to iron out uncer- higher income uncertainty. Annual household expenditure tainty, households in the high-conflict areas also build on nondurable goods is also lower for these households. up a reserve of gold and silver. This shift in household There is no significant difference in the wealth response to portfolio suggests a more substantial decline in real returns income uncertainty across high- and low-conflict provinces. of livestock relative to jewelry in high-conflict provinces. This paper is a joint product of the Finance, Competitiveness and Innovation Global Practice and the Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http:// www.worldbank.org/research. The authors may be contacted at arostom@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Conflict and the Nature of Precautionary Wealth∗ Leila Aghabarari Ahmed Rostom Rishabh Sinha May 2018 JEL C : D14, D16, D31 K : Precautionary Wealth, Conflict, Livestock, Jewelry, Afghanistan ∗ Authors in alphabetical order of last names; 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. Leila Aghabarari is a Consultant, Ahmed Rostom is a Senior Financial Sector Specialist – (both Finance, Competitiveness and Innovation Global Practice (FCIGP); Middle East and North Africa-South Asia Region) and Rishabh Sinha is an Economist in the Development Economics Research Group (Macroeconomics and Growth). Authors would like to thank Shubham Chaudhuri (Country Director, SACKB), Niraj Verma (Practice Manager, FCIGP), Norman Loayza (Lead Economist, DECMG), Claudia Nassif (Lead Economist, MTI06), Christina Wieser (Economist, GPV06), Subika Farazi (Financial Sector Specialist, FCIGP), and Aminata Ndiaye (Financial Sector Specialist, FCIGP) for comments and suggestions. Authors also benefited from comments of participants in the World Bank Seminar Series – Navigating Risk and Uncertainty in Afghanistan: Promoting Savings and Investment under Uncertainty (Kabul). Corresponding author: Ahmed Rostom (arostom@worldbank.org). 1 Introduction Households often accumulate wealth as a self-insurance mechanism against future shocks. 1 In the absence of complete markets, households build up a buffer of wealth to smooth consumption over the lifetime (Deaton, 1991). While a vast literature has analyzed the implications of the pre- cautionary motive in different settings, including many that have focused on developing economies, not much is known about the precautionary behavior in economies trapped in fragility, conflict, and violence (FCV). 2 This gap in research is unfortunate as the volatile situation in FCV economies arguably makes the precautionary motive much stronger. If true, welfare costs associated with precautionary behavior are likely to be much higher under such conditions. 3 However, an analysis of precautionary behavior requires deep household-level data which are difficult to collect in such economies. In this paper, we use data from a detailed household survey from Afghanistan to shed light on the precautionary behavior of households in a country where a range of factors packed under the broad umbrella of FCV has created an environment of utmost uncertainty. There are two principal components to our analysis. First, we want to ascertain whether the precautionary motive is a significant driver of wealth accumulation in Afghanistan and second, whether the presence of conflict affects the nature of wealth accumulation behavior of households. To make progress with these tasks, we analyze the relationship between household wealth and income uncertainty faced by households (Guiso et al. (1992), Carroll & Samwick (1997), Carroll & Samwick (1998) etc.). Critical to our analysis, the survey we use was designed with a special focus to include important sources of household wealth. It contains information on households ownership of gold and silver together with livestock holding which we use to construct our measure of wealth. The inclusion of these asset-types is essential as it has been well-documented that jewelry and livestock form a vital share of household wealth in poor economies. 4 These assets can be particularly handy in protecting against future shocks. On the one hand, the lack of financial access creates barriers to accumulating wealth using financial assets, but it is also likely that households consider jewelry and livestock to be a better store of value than cash and other financial assets based on their low trust in the formal financial institutions. 1 See Marshall (1920) and Keynes (1936) for the foundations of the theory. 2 In a large cross-country study consisting of both industrial and developing countries, Loayza et al. (2000) found widespread evidence of the precautionary savings motive. 3 To be clear, self-insurance via precautionary wealth accumulation is still preferable to a no insurance scenario. In fact, they can be more effective than other insurance instruments in combating mild idiosyncratic shocks. However, there are both private and social losses when precautionary wealth is used to protect against large idiosyncratic risks or systemic risks. For a detailed discussion, see the 2014 World Development Report (World Bank, 2014). 4 Studies going back to Jodha (1978), Watts (1983), Binswanger & McIntire (1987), Swinton (1988) etc. highlight the importance of livestock sales in smoothing shocks. Precious metals like gold and silver are used less frequently but is an important component of household wealth (Fafchamps et al. , 1998). 1 Cross-sectional analyses of precautionary motive have employed the occupational information to measure income uncertainty (Skinner (1988), Dardanoni (1991) etc.). We follow this literature and posit that different occupations are associated with different levels of income uncertainty. To this end, we classify households into different occupational groups. Following, we measure the variance in the residual of household income within each group after the effects of human capital variables and a host of fixed factors have been accounted for. This variance of the residual becomes our measure of income uncertainty which we use to analyze precautionary wealth accumulation in the country. Using two measures of wealth – one narrow and the other broad, we find a strong and highly significant positive relationship between income uncertainty and household wealth. The coefficients estimated using the narrower definition are five to six times as large as for the broader measure. On the other hand, the estimated coefficients for the broader wealth measure are similar in magnitude to what is obtained in some developed countries. 5 A challenge associated with any wealth measure is that it might fail to account for some sources of wealth that are held for precautionary needs. This omission can occur due to gaps in data or due to the definition of wealth used. We address this concern by looking at annual household expenditures on non-durables. We find that households facing higher income uncertainty not only hold higher stock of wealth but also spend less compared to households with lower variance at similar levels of income. However, period adjustment in aggregate wealth via changes in expenditures is not a necessary condition for the existence of the precautionary motive under the buffer-stock model. In this light, we recognize this relationship as a secondary test of the precautionary channel. We also consider how the precautionary wealth accumulation varies across the income distribution. Poor households typically have lower access to formal and informal borrowing channels which will make the precautionary channel stronger. We compare wealth accumulation across income quantiles and find that the strength of the precautionary motive decreases systematically with rising income. The precautionary response is specifically higher for the households in the bottom 40 percent of the income distribution compared to the relatively richer households. Having found a systematic association between income uncertainty and wealth accumulation, we next investigate how conflict interacts with the precautionary behavior of the households. What motivates this line of inquiry is the possibility that some instruments of wealth accumulation that are available in the relatively low-conflict regions may not be available in the high-conflict regions. If the sale of certain assets occurs at centralized markets, then it is a dangerous undertaking to sell them in a high-conflict region. Additionally, it is possible that certain assets like livestock are 5 For example, Carroll & Samwick (1998) estimate these coefficients using US data and Fuchs-Schündeln & Schündeln (2005) study precautionary wealth accumulation in Germany. However, we note that both differences in data and definitions do not allow for a direct comparison of these coefficients across countries. 2 more prone to losing value even at home relative to jewelry in high-conflict regions. Livestock needs upkeep in terms of feed and care, both of which are relatively scarce in high-conflict regions. Finally, more direct effects of conflict, like animal raiding in case of livestock, might also tilt the scales against certain assets. 6 We bring conflict under analysis by exploiting the variation in the intensity of conflict events across the provinces. We classify the provinces into two groups depending on whether they encounter more or fewer events compared to the median province. Consistent with the previous finding, wealth accumulation in low-conflict provinces increases with income uncertainty. The relationship remains highly significant and is robust to the choice of narrow and broad measure of wealth. The wealth accumulation response to rising income uncertainty in high-conflict provinces is marginally higher but the difference relative to low-conflict provinces is not statistically significant. This invariable nature of precautionary behavior across the low and high conflict provinces is surprising. It is very likely that our measure of income uncertainty does not capture some uncertainty associated with conflict. To the extent that the unobserved conflict-specific uncertainty is positively correlated with income uncertainty, it is expected that households in high conflict provinces accumulate more wealth at the same level of uncertainty. There are three reasons why we do not observe higher precautionary wealth accumulation in high-conflict provinces. First, it is possible that households in high-conflict regions accumulate wealth in certain forms that are not included in the measure of wealth in our study. Hence, actual wealth in these provinces is potentially higher than what is observed. Second, the price of assets relative to consumption goods might be significantly lower in these provinces owing to a general scarcity. This relative scarcity worsens the terms of trade of wealth and reduces the real return which leads to lower precautionary wealth accumulation. Finally, this worsening of terms of trade might be asset-specific with real returns of certain assets being farther from real returns observed in low-conflict provinces. If the terms of trade deterioration is more acuter for assets that have higher real returns, then households substitute towards assets with lower returns further dampening the wealth accumulation. We explore this last possibility by focusing on jewelry and livestock holding patterns of households across provinces. We find that households in low-conflict provinces increase their ownership of livestock with an increase in income uncertainty. This is in stark contrast to the ownership of jewelry, the stock of which reduces with a rise in uncertainty. This suggests that households employ livestock as an instrument of self-insurance in low-conflict settings but not jewelry. Rather, the ownership of jewelry in these provinces seems to be driven by non-precautionary needs. 7 Looking at households 6 See Bukurura (1995), Fleisher (1998), Gray et al. (2003), Bundervoet (2009) for a discussion on livestock loss in some East-African countries during conflict. 7 See Chipchase et al. (2013) for a qualitative discussion. 3 in the high conflict provinces, we find a higher accumulation of livestock at the same level of uncertainty. Crucially though, we also find that the ownership of jewelry responds very strongly to any rise in uncertainty. This points to a marked difference in the available coping strategies across provinces. In the context of the previous discussion, this can be mapped to a larger decline in real returns of livestock which makes accumulating in the form of jewelry attractive in high-conflict provinces. The welfare cost of low real returns can in part be attributed to wealth being substituted away from productive wealth like livestock which contributes directly to income and consumption of households. There exists a sizable literature that has looked at the importance of livestock as an instrument of self-insurance. Many studies have found the sale of livestock as a means to smoothing consumption (Corbett (1988), Kinsey et al. (1998), Rosenzweig & Wolpin (1993), McPeak (2004)). Nonetheless, there are many instances where livestock is either not employed or performs only a secondary role (Udry (1995), Fafchamps et al. (1998), Kazianga & Udry (2006)). The differences in the use of livestock as an instrument of self-insurance highlight the differences in settings in which these studies are conducted. The availability of other instruments is an important factor determining household decisions, and conflict reduces the set of available resources drastically. Verpoorten (2009) compares livestock sales during the period of peace with sales during the period of civil war in Rwanda and finds that it was during the conflict when livestock sales were used to obtain food. It is worth noting that the household response in the Rwandan context is tied to an unanticipated conflict situation. Households made adjustments after they had made decisions of wealth accumulation assuming a peacetime scenario. In contrast, we contribute by looking at long-term wealth accumulation behavior when conflict has become a part of everyday reality, and households understand the cushion provided by various assets under conflict. While we find an active role of livestock across high- and low-conflict provinces, we also provide evidence that households precautionary wealth accumulation extends over to jewelry as the conflict situation worsens. We conclude our analysis by performing a battery of robustness checks on our main findings. In the first set of tests, we consider different definitions of the low- and high-conflict provinces. In addition, our estimates remain robust to dropping provinces with very high conflict intensity and when we employ conflict as a continuous variable instead of an indicator variable. Second, we address the concern that our findings are driven by the inclusion of households in livestock occupations which require livestock ownership. The livestock ownership is prevalent independent of occupations practiced by members and we find our results to be robust when we consider only those households who report not being involved in livestock occupations. The rest of the paper is organized as follows. We begin by giving a brief description of the 4 data and details on the construction of the main variables. Thereafter, we outline our econometric specifications, report our findings, and test for the robustness of our main results. We conclude the paper with a brief discussion on the policy implications of our study. 2 Data We use data from two primary sources to conduct our analysis. The essential data come from the Afghanistan Living Conditions Survey (ALCS) (Central Statistics Organization (2016)) household survey, which we use to construct measures of labor income uncertainty and household wealth. The ALCS is a large sample survey conducted by the Central Statistics Organization (CSO) and is representative at the province level. The detailed nature of the ALCS data allows us to control for many household-level characteristics which are relevant to the saving decisions. We use conflict data from the Global Terrorism Database (GTD) (START (2017)) to measure the conflict intensity at the province level. We discuss some details of these data in this section. 2.1 Wealth The ALCS has been conducted sporadically for more than a decade. Yet, the 2013-14 cross- section is of particular importance because the survey instrument for that year was designed to capture household wealth holistically. This survey round contains information on household ownership of gold and silver as well as livestock, which are essential components of household wealth as evidenced by the previous literature. We begin by constructing a narrow measure of household wealth that corresponds to the total value of gold, silver, and livestock owned by the household. The survey contains detailed information in regards to the livestock ownership. Households are asked to report the ownership of livestock across nine categories ranging from cattle and camels to chicken and small birds. The price data for this diverse set of livestock category are not available for Afghanistan and we rely on price data from different sources to aggregate the total value of livestock. 8 A first observation from the data is the widespread prevalence of livestock ownership across the Afghan households. Around two-thirds of the households in our sample report owning any livestock. Nonetheless, there exists a marked heterogeneity across the provinces. The share of households with any livestock ownership varies from a high of 90 percent in Paktika and Khost to 25 percent in the more urban Kabul province. The nature of ownership is diverse as well with households not only keeping animals 8 See appendix section on livestock for more details on this aggregation. 5 necessary for farm work but also owning smaller animals and birds that are better associated with consumption. In fact, chicken and sheep are the most common types of livestock. With regards to ownership of jewelry, the survey reports the gram weight of gold and silver held by the households. In contrast to livestock ownership, only 14 percent of households reported owning either gold or silver. The ownership of gold or silver is higher in Kunarha and Khost where around 40 percent of households report owning some amount of gold and/or silver. Interestingly, livestock ownership is also higher in these provinces. Yet, in many provinces, there is almost no ownership of gold and silver. Less than 1 percent of households in Baghlan, Ghor, and Zabul have some wealth in the form of gold and silver. No household in our sample from Nooristan and Daykundi had any ownership of gold and silver either. Figure 1 shows the average value of gold and silver and livestock across the provinces. Not only do more households own livestock, we find that livestock accounts for a large share of wealth in each province. In addition to our narrow measure of wealth, we use information from the survey to construct another broader measure of wealth. Specifically, we augment the narrow measure to include any savings from the previous year less any debt that a household owes. The savings from the previous year is defined as annual income in excess of annual non-durable expenditure. Our motivation for including savings from the previous year is based on the notion that there may be a lag in converting such savings to assets which we observe in the narrow measure. Around half of the households in the sample reported having any debt. As is the case with asset ownership, there are huge variations in debt across provinces. Almost every household in Ghor in our sample reported having a debt obligation. The debt was also high in the provinces of Kunarha and Urozgan. In sharp contrast, households in Paktika and Faryab had minimal debt obligations. In figure 2, we show the average annual income and expenditure across the provinces. Average annual income ranges from more than AFN 200,000 in Kabul and Khost to just around AFN 50,000 in Ghor and Daykundi. In line with income, Kabul also has the highest level of annual expenditure. Average expenditure is also high in Helmand, almost as high as the average annual income for the province, resulting in a low savings rate. Daykundi and Ghor have the lowest levels of average expenditure in line with their lower average incomes. We note that our measure of wealth does not include housing wealth. To be effective as an instrument of self-insurance, an asset needs to be liquid so that it can be readily exchanged when needed. Housing wealth, however, tends to be relatively much less liquid than other assets and we do not view this exclusion as a severe concern. Nonetheless, we attempt to account for housing wealth by adding controls for house ownership. Specifically, the ALCS data allow us to identify whether a particular household owns a house or not. Furthermore, we can also identify the broad characteristics of the house which we control for. Houses in Afghanistan range from temporary 6 shacks and tents to traditional, more durable buildings. 9 2.2 Labor Income Uncertainty To measure uncertainty linked with the household income, we follow previous literature (Skin- ner (1988), Dardanoni (1991) etc.) by postulating that occupations are associated with different levels of income uncertainty. The ALCS contains information on households’ three principal sources of income. Most of these sources can be mapped to an occupation, like farming, teaching etc. 10 Household income is derived from multiple sources not only because more than one member contributes to the aggregate but also because sometimes a single household member derives in- come from multiple sources. We use all three sources of income to classify households into various groups. We put households that report the same first, second, and third sources in the same order into a single occupational group. We use a more restrictive classification to separate households that may have had the same sources of income but may have differed in terms of how important they are to the household income. To ensure we have enough households in a single group so that the corresponding variance is well measured, we drop those occupational groups that contain fewer than 40 observations. We end up with 83 such occupational groups. The quantitative analysis of precautionary behavior hinges on the extent of variation of income uncertainty that is observed in the data. Our classification of households into many occupational groups instead of a few is thus essential to tease out the relationship between income uncertainty and wealth accumulation. In regards to the sample selection, we exclude all households that do not report their income and those with heads aged less than 18 or over 64 years old. We also drop those households with retired or unemployed heads as well as households for which any of the three primary sources is a non-labor source such as pension earnings, rental income, and Zakat (borrowing). This leaves us with 13,156 households which constitute our benchmark sample. Not surprisingly, agriculture is the dominant activity from which household income is derived. The production and sale of field crops (non-opium) is the most common source of household income and is the primary source of income for 21 percent of the households in the sample. At the other extreme, road and building construction is reported to be the primary income source by only 0.01 percent of the households. The average annual income in our sample is AFN 124,000 while the median household earned income is AFN 93,000. There exists a considerable variation in aggregate household income across provinces. Households in Kabul and Khost have the highest 9 We list the distribution of house types in the table A.3 of the appendix. 10 Table A.2 in the appendix lists the 28 different sources of household income. 7 average levels of income (AFN 235,000 and AFN 211,000 respectively), while households in Daykundi and Ghor have the lowest average incomes of around AFN 50,000. We now provide details on how we measure uncertainty associated with the household income. Our classification of households into occupational groups implies that we are able to measure the variability in household income that stems from the nature of occupations from which households derive their income. Nonetheless, we are aware that not all of the income variance within a group is driven by the fundamental nature of occupations. Worker-specific factors, including differences in the within-group human capital, might also affect the within-group variation in incomes. Thus, we first regress the log of household income using a bunch of individual-level controls such as age, education, and gender for each group g as specified in equation (1). We then use the log of the variance of the collected error terms for each occupational group as our measure of income uncertainty. log( I NCi ) = α + β1 E DUi + β2 AGEi + β3 AGEi2 + β4 GE N DE Ri + i (1) Figure 3 shows how the narrow measure of wealth at the provincial level varies with income uncertainty. The horizontal axis corresponds to the mean of the variance of the household income. The income variance of the households is normalized by household income to control for wealth accumulation driven by income changes. We find evidence of a significant positive association between the two measures. Wealth accumulation is lower in provinces where the income adjusted level of uncertainty is lower which suggests that the precautionary margin of wealth accumulation is active in the country. We also find some outliers at the either extreme. Households in Khost, for example, accumulate much higher wealth given the low variability of income in the province. On the other hand, households in Daykundi and Urozgan accumulate much lower wealth considering the high uncertainty associated with their incomes. 2.3 Conflict The third important element of our analysis relates to the existence of conflict which makes FCV economies like Afghanistan different from the other developing economies. The country has been affected by conflict for many decades, though the conflict situation has ebbed and flowed over time. Crucial to our analysis though is the notion that the presence of conflict fundamentally affects the precautionary behavior of households. There are three important reasons why we think this might happen. First, there is a severe deterioration in the efficiency with which markets operate under a conflict situation. In extreme circumstances, markets might cease to exist. Yet, some markets are 8 affected more than others with large, central markets being more at risk. If certain assets can only be exchanged in such central markets, then they lose their purpose to serve as an efficient instrument of self-insurance. Second, certain assets are also prone to lose value at home compared to others. For example, livestock requires constant upkeep which not only includes fodder requirement which gets pressurized during a conflict but other specialized care like a veterinarian visit. In contrast, some assets like jewelry are more dormant and are more resistant to losing value at home. Finally, some assets might bear particular interest to conflict groups who might target these assets specifically. For instance, cattle raiding is a strategy often employed by many groups. All these reasons suggest that conflict drives a wedge in the decision-making behavior of households living in high-conflict provinces relative to their counterparts in low-conflict regions. Our approach is to utilize the variation in conflict intensity across the Afghan provinces to consider whether the nature of precautionary wealth is affected by conflict. We measure conflict intensity using the data on the number of terrorist attacks which we source from the GTD. The database allows us to construct the total number of such events for each province from 1973 – 2014. Figure 4a shows the variation of conflict intensity across the provinces. The conflict situation is particularly grave in the southern provinces with more than 800 terrorist attacks being reported for Helmand and Kandahar over the entire period. The variation in intensity across provinces is stark, and many central and some northeastern provinces witness much less conflict activity. For instance, a total of three attacks have been reported in the northeastern province of Panjsher and a total of eight attacks in the central province of Daykundi over the four decades. An alternative measure of conflict intensity will be to consider a period close to the survey year. We consider many smaller periods, not only limited to periods near the survey year, and find that the cross-sectional variation of conflict intensity remains stable across all these periods. For instance, in figure 4b we show the cross-sectional variation in attacks for the year 2012 which is the year preceding the survey period. We note that the variation of intensity remains similar to the figure on the left which is also confirmed with a high correlation coefficient of 0.92. In this light, we rely on the broad horizon and aggregate the total attacks over the four decades. This concludes our outline of the data, and we detail the elements of our formal analysis of these data in the next section. 3 Econometric Specification In the previous section, we discussed some details of the data that we use in our analysis. We also highlighted the heterogeneity in wealth accumulation, incomes, income uncertainty and 9 conflict intensity across the provinces which suggests the presence of a precautionary motive in the household wealth accumulation behavior. Taking our investigation further, we now formally investigate the presence and the strength of this precautionary motive in the country. We source our econometric specification from previous studies which we discuss in more detail below. We note again that in our investigation of the household wealth accumulation behavior, we are also concerned with whether conflict interacts with this behavior. For this purpose, our specification needs an extension of the standard specification followed in the past literature. The data that we use in our analysis are of a cross-sectional nature. Hence, our strategy is to exploit the cross-sectional variation to test whether higher wealth accumulation is associated with higher income uncertainty. To this end, we use the following regression specification: log(Wi g p ) = α + β1 log(V ARLYg ) + β2CON Fp + β3 log(V ARLYg ) × CON Fp (2) + β4 log( I NCi ) + γ Xi + θ p + ig p where Wi g p denotes the wealth accumulated by a household i belonging to an occupational group g living in a province p. The precautionary motive dictates that this accumulated wealth rises with an increase in income uncertainty which is captured by V ARLYg . The variable V ARLYg is the variance of the log of income which varies with the occupational group of a household. Using a Deaton (1991) type buffer-stock model of wealth accumulation and data from the US, Carroll & Samwick (1998) show that this theoretical measure of uncertainty fits the data relatively well compared to a theoretical measure of relative uncertainty. We follow the same specification. The coefficient β1 is the above specification and hence becomes the first object of interest in our study. To bring variation in conflict under analysis, we extend the standard specification by introducing a dummy for conflict intensity CON Fp . This dummy classifies the Afghan provinces into two groups – high- and low-conflict. Our objective is to check how the precautionary behavior changes with conflict and for this purpose we interact the conflict dummy with the measure of income uncertainty V ARLYg . The parameter β3 captures the variation in wealth accumulation across high- and low- conflict provinces at the same level of income uncertainty. In our benchmark specification, we define a province to be a high-conflict province if the total number of terrorist attacks in the province exceeds that of the median province (189 attacks). We perform several alternative variations around this classification to ensure robustness. We note that it is straightforward to use conflict intensity as a continuous variable compared to constructing a dummy. We prefer the dummy classification because it allows for a simpler interpretation of the coefficient β3 . We complete our specification by controlling for a host of factors that are considered important 10 differences in household wealth accumulation behavior. Most importantly, we directly control for household income I NCi to account for any non-precautionary wealth accumulation associated with higher income. In addition, we add controls at the household-level which includes capturing char- acteristics of household-heads. Specifically, Xi represents this vector of household-level controls which includes the size of the households, the number of male members in each household, the type of the dwelling, and the age, education, marital status, and gender of the heads. Finally, we control for non-conflict variation across provinces by adding province-level fixed effects θ p . This concludes our econometric specification. We make one final point before discussing our empirical findings. We consider two measures of wealth – one being broader than the other, which serves as the left-hand side variable in the econometric specification. The positive relationship between household wealth and income uncertainty like one being explored using these wealth measures follows from a buffer-stock model. Nonetheless, there are concerns with any wealth measure that it fails to account for some sources of wealth that households might use to smooth income shocks. To alleviate such concerns, we consider two supplementary variables – annual household expenditure on nondurable goods and excess of annual income over this non-durable expenditure. The idea behind using these measures is to whether households who face higher income uncertainty also spend less on non-durable goods, or add to their stock of buffer on a periodic basis. If there exist errors with measuring wealth that drive false evidence of precautionary accumulation, it might be indicated by no corresponding differences in the period expenditure and the period addition to the buffer stock across households with different levels of income uncertainty. However, a period adjustment is required only if the households have not yet reached the optimum level of the buffer stock. Hence, these variables require a somewhat stricter test of the precautionary motive. 4 Empirical Results We now discuss the results of our analysis which we present in the following three subsections. In the first subsection, we document the existence of precautionary wealth in Afghanistan and compare the responsiveness of the two wealth measures to changes in income uncertainty. In the second subsection, we highlight the differences in wealth accumulation behavior of households across the high- and low-conflict provinces. We conclude this section by performing robustness checks on our findings. 11 4.1 Income Uncertainty and Wealth Accumulation In the first set of regressions, we use a restricted version of the estimating equation (2) to focus on the precautionary wealth accumulation in the country independent of its interaction with conflict, that is, we set β2 = β3 = 0. Table 2 shows the results of this specification for the two wealth measures. First, consider the narrow measure of wealth which corresponds to the ownership of gold, silver, and livestock (GSL). The estimated β1 for the narrow measure are shown in columns (1)–(4). Column (1) corresponds to the case when no controls other than household income are added to the specification. We find that the narrow measure of wealth is positively correlated with the variance of income associated with a household’s occupational classification and the relationship between the two variables is highly significant. The estimates in the next columns correspond to specifications when we add additional controls beginning with household- level controls discussed earlier in column (2). In column (3), we further add province-fixed effects whereas in column (4) we add province-level control for the conflict intensity which we proxy by the aggregate number of conflict events that occurred in the province during 1973–2014. We find that the positive correlation between household wealth and variance of income is robust to the inclusion of household- and province-level controls. The relationship also remains highly significant in all cases. Our point estimates indicate that a 1 percent increase in the variance of income is associated with a 2.3 – 3.4 percent increase in the household wealth accumulated in the form of gold, silver, and livestock. Next, we consider the relationship between income uncertainty and the broader measure of wealth – Non-Housing Wealth, which accounts for saving from the previous year. The specifications in columns (5)–(8) match the specifications in columns (1)–(4). The broader wealth measure endorses the positive and highly significant relationship found earlier. The point estimates when using the broader wealth measure are five to six times smaller compared to the estimates obtained when the narrower measure of wealth is used. A 1 percent increase in the variance of income is associated with an increase of approximately 0.5 percent in household wealth when accounting for savings from the previous year. While differences in data and definitions do not allow for a direct comparison of the coefficients, we note that our estimated coefficients for the broader wealth measure are close to some earlier studies in developed countries (Carroll & Samwick (1998), Fuchs-Schündeln & Schündeln (2005)). Driven by the fact that we observe large adjustments in the narrow measure of wealth to changes in income uncertainty, we go a step deeper into the constituents of the narrow wealth measure. We compare the responsiveness of wealth accumulated in the form of gold and silver, and livestock separately to changes in income uncertainty. Both these assets can be effective in coping with shocks 12 but there exist fundamental differences between them which might be important when considering the precautionary behavior. For example, livestock not only can be used to smooth consumption in the event of experiencing a shock to income, it also contributes to income (or consumption) every period. Similarly, the demand for jewelry can also stem from non-precautionary motives. There is also a marked difference in how costly it is to preserve the two assets. On the one hand, there are costs associated with the general upkeep of livestock which are not incurred when wealth is accumulated in the form of gold and silver. But, gold and silver are more prone to losses due to theft. Table 3 reports the results for the two assets separately. Columns (1) – (4) refer to the ownership of gold and silver and columns (5) – (8) show the results when we restrict our attention to wealth held as livestock. We find that the accumulation of both assets increases with an increase in the variance of income associated with a household’s occupational group. Though the relationship is not statistically significant when we only consider wealth accumulated in the form of gold and jewelry. The magnitude of variation in wealth accumulation due to variation in the variance of income is significantly higher for livestock. A 1 percent increase in the variance of income is correlated with 3 – 4 percent increase in the value of livestock owned. In sharp contrast, the ownership of gold and silver only increases by 0.4 – 0.6 percent which is close to the adjustments made for the broader wealth measure. This suggests a relatively higher importance of livestock as a precautionary asset compared to gold and silver. 4.1.1 Precautionary Wealth Accumulation across Income Quantiles Before moving to analyzing how conflict interacts with income uncertainty, we ask whether the strength of the precautionary motive changes across the income distribution. There are valid reasons why the precautionary motive might vary systematically across the income distribution. For one, the borrowing constraints are likely to be more severe for the low-income households who generally have poor access to both formal and informal credit. As such, it is expected that these households will accumulate larger buffer-stock when faced with similar levels of uncertainty. In order to test this hypothesis, we divide our sample into five income quantiles and estimate the coefficients for each income group for the two wealth measures. For brevity, we discuss only the results when all household-level controls are added jointly with province-level dummies. Figure 5a shows the estimated coefficients for the five income quantiles when we use the narrow wealth measure as the dependent variable. Income quantile 1 represents the households in the bottom 20 percent of the sample income distribution and income quantile 5 represents the top 20 percent of the households. We find a systematic downward trend in the strength of precautionary 13 wealth accumulation as we move up the income distribution. The coefficients for the bottom two quantiles are not very different from each other and are significantly higher than coefficients obtained at other income quantiles. The point estimates for the upper-income quantiles are away from zero signifying prevalence of precautionary behavior but the estimates are much less precisely estimated. Figure 5b shows the results of the exercise when we consider the broader wealth measure. Like before, we find a similar downward trend in the estimated coefficients with rising income. The point estimates are again not distinct from each other for the bottom two quantiles and are roughly twice in magnitude relative to the aggregate. A point of departure in the case of the broader wealth measure is that the estimated coefficients for the top three income quantiles are close to zero implying no adjustment to household wealth with rising income uncertainty. The findings from this exercise hint that the burden of self-insurance is felt more by the low-income households perhaps because they have worse access to credit channels that they can tap into when exposed to shocks. 4.2 Wealth Accumulation and Conflict We now proceed with our second line of inquiry by bringing conflict under analysis. Afghanistan is a region heavily influenced by conflict. In the previous section, we find evidence of the precau- tionary motive in households wealth accumulation. However, the instruments of self-insurance might change with the incidence of war and conflict. To entertain this possibility, we analyze the variation of households’ precautionary wealth in provinces that are more prone to conflict compared to relatively safer areas. Table 4 shows the estimates of coefficients in equation (2) for the two measures of wealth together with constituents of the narrow wealth measure. For brevity, the table presents our results after including all controls at the household and province level. The first row of table 4 shows the result for the low-conflict provinces. The estimates in columns (1) and (4) are comparable to the estimates in columns (3) and (7) of table 2. A 1 percent increase in income variance is associated with 2.3 percent higher accumulation of the narrow wealth measure (GSL) and 0.4 percent higher accumulation of the broader wealth (Non-Housing Wealth). The second row of table 4 shows the marginal impact of conflict on precautionary wealth. The effect of conflict on precautionary wealth accumulation is marginally positive but is not significant. However, we take our analysis one step further. As conflict potentially distorts the availability and relative benefits of assets which are used for self-insurance, we take a deeper look at households’ choices. The central question is whether households in high-conflict provinces use the same instruments for their self-insurance that is used by households in low-conflict provinces. Hence, we decompose the narrow measure into its two constituents – jewelry and livestock. In our analysis of the conflict, we use a dummy for high- and low-conflict regions. This dummy 14 equals one when the aggregate number of terrorist attacks from the year 1973 to the year 2014 in a province exceeds the number of attacks experienced by the median province. 11 Figure 1 shows the average level of livestock- and jewelry-ownership in all the provinces. In almost all high-conflict provinces, households own at least some amount of jewelry. In addition, the average value of jewelry in Kabul, Herat, Ghazni, Khost and, Paktika is higher than other provinces. To have a more formal estimate of the variation of livestock- and jewelry-ownership of households in provinces with different levels of conflict, we estimate regression (2) for livestock and jewelry separately. Columns (2) and (3) of table 4 correspond to the accumulation of jewelry and livestock respectively. A 1 percent increase in the variance of income is associated with a 2.8 percent higher accumulation of livestock in the low-conflict provinces. The accumulation of livestock in high-conflict provinces for a matched change in the variance of income is marginally higher but the difference is not statistically significant. However, there exists a stark difference in the nature of jewelry accumulation across the two province types. While the accumulation of jewelry is negatively correlated with the variance of income in the low-conflict provinces, a 1 percent increase in the variance of income is connected with a 1.5 percent increase in the value of jewelry accumulated in the high-conflict provinces. This outcome translates into the non-precautionary use of gold and silver in low-conflict provinces, as opposed to the high-conflict areas where jewelry is also used as an instrument of self-insurance in addition to livestock. We speculate on three reasons why jewelry is used as an instrument of self-insurance in the high-conflict regions. A first possibility is that households may prefer keeping gold and silver to livestock due to the risks associated with selling livestock in the centralized markets when the market for jewelry exchange is relatively decentralized and safer. Second, to keep livestock one needs to feed them and provide some care. This livestock care might be relatively more challenging for households in high-conflict regions since both food and health services are scarce. Third, as human lives are more in danger in high-conflict regions, so are the lives of livestock. It is riskier to keep animals compared to jewelry with directed attacks on livestock as evidenced by animal raiding in many countries. The outcomes after taking conflict into account are significant. In light of this, it is prudent to employ a different definition for the conflict dummy. We repeat the previous analysis using a different threshold for the conflict dummy. In table 5 we show the results when the conflict dummy takes the value of one for a province which experienced more attacks than the mean of attacks across provinces. 12 The results of this exercise reinforce the results we report in table 4 where the 11 Provinces that are considered high-conflict using the median definition are Farah (309), Faryab (220), Ghazni (474), Helmand (862), Herat (408), Kabul (536), Kandahar (537), Khost (380), Kunar (302), Kunduz (255), Nangarhar (537), Paktika (206), Urozgan (294) and, Zabul (230). 12 Using this alternative measure for the conflict dummy, Farah, Ghazni, Helmand, Herat, Kabul, Kandahar, Khost, 15 median province provided the cut-off for the conflict dummy. Another concern is that these results might be driven by the provinces with very high levels of conflict. Helmand and Kandahar are the two provinces with highest numbers of terrorist attacks. Therefore, in the next step, we show the results in Table 6 where we run the regressions for the sample after dropping observations from Helmand and Kandahar. The outcome of this table is very similar to the results in Table 4. After omitting the two highest conflict provinces from our samples, we can still confirm our previous results. In our sample, we have households’ whole primary source of income is from selling livestock or livestock products. Including these households in our final sample, we may be misled by the income motive of these families, they might hold livestock as a source of income and not for their precautionary need. In our final primary analysis, we run the regressions for those households whose source of income is independent of the sale of livestock or its products. In Table 7 we show the results of this set of regressions. We still find that households in relatively low-conflict provinces rely on livestock for their precautionary needs but in the presence of high-conflict, they also build-up a stock of jewelry in addition to livestock as a precaution against income uncertainty. 4.3 Robustness Checks In this section, we perform some robustness checks on the findings of the previous section. The first concern that we address is related to a fundamental challenge associated with adopting any measure of wealth to study precautionary behavior. It is likely that some forms of wealth that should be included in the measure may be left unaccounted for because of a lack of information or because of the definition of wealth chosen. To address this, we look at the relationship between household non-durable expenditure and income uncertainty. The first four columns of table 8 show the nature of the association between annual non-durable expenditure and income uncertainty. We find that non-durable expenditure of households is negatively and significantly correlated with the variance of income. This essentially confirms that households with higher income uncertainty not only accumulate higher wealth but also spend less on non-durable goods and services. We find that, on average, non-durable households expenditure decreases by 10 basis points with a 1 percent increase in the variance of income. We also consider a measure of saving which we define as the excess of household income over the non-durable expenditures. Consistent with our previous findings, this measure of saving is positively and significantly correlated with the variance of income as seen in columns (5) – (8) of table 8. The point estimates suggest that a 1 percent increase in income Kunar, Nangarhar and, Urozgan are provinces with high-conflict. 16 variance is connected to 12 basis points increase in household saving. Next, we go back to the concern of livestock ownership being associated with income needs and not necessarily with precautionary motive. Like before, we exclude all households that report some direct livestock activity as one of their three main sources of income and focus on the smaller sample of households. Table 9 shows the results of the exercise with the smaller sample and we find that our results are robust to the exclusion. In fact, the point estimates that we obtain for the smaller sample are higher compared to what we find for the benchmark sample. A 1 percent increase in the variance of income is associated with a 2.8 – 3.8 percent increase in the household wealth accumulated in the form of gold, silver, and livestock compared to a 2.3 – 3.4 percent increase seen in the baseline sample. We also detect an increase in the point estimates for the broader measure of wealth. Table 10 further shows that the previous results on the jewelry and the livestock accumulation separately are also preserved under this specification. As the last robustness check for the results on precautionary wealth, we exclude households who reside in Helmand and Kandahar which are the two most conflict-prone provinces in the country. Our motivation behind this exercise is to check if the results are driven by the extreme nature of the conflict seen in the two provinces or if the behavior is prevalent in other provinces also. Table 11 reports the outcome after we drop Helmand and Kandahar. The wealth measures still respond to changes in income uncertainty. The point estimates also survive a major correction and remain close to the benchmark case. A 1 percent increase in the variance of income is associated with a 2.4–3.1 percent increase in the accumulation of the narrow measure and a 0.4 – 0.5 percent higher accumulation of the broader measure of wealth. We also find that further decomposing the narrow measure to its constituents yields similar results as those obtained earlier as seen in table 12. 5 Conclusion Using data from a comprehensive household level survey in Afghanistan, we examine the association between income uncertainty and wealth accumulation. Our analysis yields many expected results. First, we find that the stock of household wealth increases systematically with increasing income uncertainty after controlling for other factors including the level of household income. The elasticity of jewelry and livestock accumulation is particularly high. Second, we also find that households with higher uncertainty also spend less at similar levels of income. Third, the precautionary behavior is most active for the poorest households and decreases as income rises. This last finding suggests that poor households have much lower access to borrowing channels, either formal or informal, which makes self-insurance the dominant instrument of coping with 17 risks. Taking the analysis further, we ask whether the precautionary behavior of households varies with respect to how much conflict intensity they are exposed to. Conflict can alter household decisions because real returns on assets are expected to be lower in highly unsafe provinces. To the extent that this reduction in real returns is variant across different asset types, the conflict might influence the nature of wealth accumulation. We find evidence of this phenomenon in Afghanistan. Households in relatively low-conflict provinces rely heavily on livestock in order to iron out shocks and abstain from using jewelry. On the other hand, we find that while livestock remains an instrument of self-insurance in high-conflict regions, households also accumulate a large stock of jewelry as their uncertainty rises. Given a choice to alter their portfolio, households in high-conflict provinces allocate a higher weight to jewelry for which the real returns are preserved better compared to livestock. The pertinent question though is what policy intervention can aid households to cope with risks in conflict settings. This is a difficult area to explore because of the systemic nature of conflict- related shocks at the local level. Nonetheless, if conflict shocks are independent across localities in the high-conflict provinces, then it might be feasible to design insurance programs that can diversify conflict shocks over such localities. The use of technology can also help to reduce frictions in inter- and intra-household risk-sharing mechanisms. If conflict creates barriers to timely remittances from a member or relatives living far when the household experiences adverse shocks, then use of mobile banking can provide an alternative as such transactions are arguably less exposed to conflict shocks. Batista & Vicente (2013) argue that the introduction of mobile money in Mozambique led to a reduction in the vulnerability of households to weather and other shocks in light of increased remittances. Similarly, Blumenstock et al. (2016) document evidence of mobile money transfers amongst households separated by large distances for risk-sharing purposes in Rwanda. Of course, the context in Afghanistan in markedly different and extreme care needs to be exercised. For instance, a basic requirement for mobile money to succeed is that the infrastructure surrounding its operation remains robust in times of shocks. The dynamic nature of the conflict also presents some interesting avenues. Though still nascent, a growing body of literature argues that in certain cases conflict wanes when interventions lead to economic development (Miguel et al. (2004), Beath et al. (2012), Dube & Vargas (2013)). In this regard, an insurance-based intervention that is infeasible in the short-run may become feasible when dynamic benefits of declining conflict intensity are taken into account. Finally, given the importance of livestock ownership in household precautionary behavior, an insurance program that protects against non-conflict shocks that are idiosyncratic like a predator attack or animal disease 18 will also aid households. 13 References Batista, Catia, & Vicente, Pedro. 2013. Introducing Mobile Money in Rural Mozambique: Initial Evidence from a Field Experiment. Nova Africa Center for Business and Economic Development, Working Paper No. 1301. Beath, Andrew, Fotini, Christia, & Enikolopov, Ruben. 2012. Winning Hearts and Minds through Development? Evidence from a Field Experiment in Afghanistan. Policy Research Working Paper 6129, World Bank: Washington, DC. Binswanger, Hans, & McIntire, John. 1987. Behavioral and Material Determinants of Production Relations in Land-Abundant Tropical Agriculture. Economic Development and Cultural Change, 36(1), 73–99. Blumenstock, Joshua E, Eagle, Nathan, & Fafchamps, Marcel. 2016. Airtime Transfers and Mobile Communications: Evidence in the Aftermath of Natural Disasters. Journal of Development Economics, 120, 157–181. Bukurura, Sufian Hemed. 1995. Combating Crime Among the Sukuma and Nyamwezi of West- Central Tanzania. Crime, Law and Social Change, 24(3), 257–266. Bundervoet, Tom. 2009. Livestock, Land and Political Power: The 1993 Killings in Burundi. Journal of Peace Research, 46(3), 357–376. Carroll, Christopher, & Samwick, Andrew. 1997. The Nature of Precautionary Wealth. Journal of Monetary Economics, 40(1), 41–71. Carroll, Christopher, & Samwick, Andrew. 1998. How Important is Precautionary Saving? Review of Economics and Statistics, 80(3), 410–419. Central Statistics Organization. 2016. Afghanistan Living Conditions Survey 2013-14, National Risk and Vulnerability Assessment. Kabul, Afghanistan. Chipchase, Jan, Rolston, Mark, Silver, Cara, & Blumenstock, Joshua. 2013. In the Hands of God: A Study of Risks and Savings in Afghanistan. Institute for Money, Technology and Financial Inclusion. 13Wakhan Livestock Depredation Insurance Program is one such pilot intervention in the relatively low-conflict Badakhshan province. 19 Corbett, Jane. 1988. Famine and Household Coping Strategies. World Development, 16(9), 1099– 1112. Dardanoni, Valentino. 1991. Precautionary Savings Under Income Uncertainty: A Cross-Sectional Analysis. Applied Economics, 23(1), 153–160. Deaton, Angus. 1991. Saving and Liquidity Constraints. Ecomometrica, 59(5), 221–248. Dube, Oeindrila, & Vargas, Juan F. 2013. Commodity Price Shocks and Civil Conflict: Evidence from Colombia. Review of Economic Studies, 80(4), 1384–1421. Fafchamps, Marcel, Udry, Christopher, & Czukas, Katherine. 1998. Drought and Saving in West Africa: Are Livestock a Buffer Stock? Journal of Development Economics, 55(2), 273–305. Fleisher, Michael L. 1998. Cattle Raiding and its Correlates: The Cultural-Ecological Conse- quences of Market-Oriented Cattle Raiding Among the Kuria of Tanzania. Human Ecology, 26(4), 547–572. Fuchs-Schündeln, Nicola, & Schündeln, Matthias. 2005. Precautionary Savings and Self-Selection: Evidence from the German Reunification ‘Experiment”. Quarterly Journal of Economics, 120(3), 1085–1120. Gray, Sandra, Sundal, Mary, Wiebusch, Brandi, Little, Michael, Leslie, Paul, & Pike, Ivy. 2003. Cattle Raiding, Cultural Survival, and Adaptability of East African Pastoralists. Current Anthro- pology, 44(S5), S3–S30. Guiso, Luigi, Jappelli, Tullio, & Terlizzese, Daniele. 1992. Earnings Uncertainty and Precautionary Saving. Journal of Monetary Economics, 30(2), 307–337. Jodha, Narpat Singh. 1978. Effectiveness of Farmers’ Adjustments to Risk. Economic and Political Weekly, A38–A48. Kazianga, Harounan, & Udry, Christopher. 2006. Consumption Smoothing? Livestock, Insurance and Drought in Rural Burkina Faso. Journal of Development Economics, 79(2), 413–446. Keynes, John Maynard. 1936. General Theory of Employment, Interest and Money. London: MacMillan. Kinsey, Bill, Burger, Kees, & Gunning, Jan Willem. 1998. Coping With Drought in Zimbabwe: Survey Evidence on Responses of Rural Households to Risk. World Development, 26(1), 89–110. Loayza, Norman, Schmidt-Hebbel, Klaus, & Servén, Luis. 2000. What Drives Private Saving Across the World? Review of Economics and Statistics, 82(2), 165–181. 20 Marshall, Alfred. 1920. Principles of Economics. London: MacMillan. McPeak, John. 2004. Contrasting Income Shocks With Asset Shocks: Livestock Sales in Northern Kenya. Oxford Economic Papers, 56(2), 263–284. Miguel, Edward, Satyanath, Shanker, & Sergenti, Ernest. 2004. Economic Shocks and Civil Conflict: An Instrumental Variables Approach. Journal of Political Economy, 112(4), 725–753. Rosenzweig, Mark, & Wolpin, Kenneth. 1993. Credit Market Constraints, Consumption Smoothing, and the Accumulation of Durable Production Assets in Low-Income Countries: Investments in Bullocks in India. Journal of Political Economy, 101(2), 223–244. Skinner, Jonathan. 1988. Risky Income, Life Cycle Consumption, and Precautionary Savings. Journal of Monetary Economics, 22(2), 237–255. START. 2017. Global Terrorism Database [Data File]. National Consortium for the Study of Terrorism and Responses to Terrorism (START), Retrieved from https://www.start.umd.edu/gtd. Swinton, Scott M. 1988. Drought Survival Tactics of Subsistence Farmers in Niger. Human Ecology, 16(2), 123–144. Udry, Christopher. 1995. Risk and Saving in Northern Nigeria. American Economic Review, 85(5), 1287–1300. Verpoorten, Marijke. 2009. Household Coping in War-and Peacetime: Cattle Sales in Rwanda, 1991–2001. Journal of Development Economics, 88(1), 67–86. Watts, Michael J. 1983. Silent Violence: Food, Famine, and Peasantry in Northern Nigeria. University of Georgia Press. World Bank. 2014. World Development Report 2014: Risk and Opportunity – Managing Risk for Development. World Bank, Washington D.C. 21 Table 1: Summary Statistics Variable Observations Mean SD Min Max Source Definition Main Household Variables GSL (log) 13156 1.35 3.62 0 14.57 ALCS Value of gold & silver & livestock (in log). Non-Housing Wealth (log) 8230 11.26 1.39 0 15.74 ALCS Income+Expenditure+Value of Gold & Silver + Value of Livestock 22 - Debt (in log). Expenditure (log) 13156 11.36 0.66 0 14.01 ALCS Annual expenditure (in log). Income (log) 13156 11.43 0.75 7.6 15.76 ALCS Annual household income (in log). Income Variance (log) 84 0.81 0.30 0.70 2.50 ALCS Variance of income of occupational group (in log). Conflict 34 226.12 214.53 3 862 GTD Total number of incidents per province from 1973 to 2014 (continued) All figures are reported in the local currency (Afghani). Summary Statistics (continued) Variable Observations Mean SD Min Max Source Definition Supplementary Controls (Household, Household-Head Level) Age 13156 39.15 11.02 18 64 ALCS Age of the household head. Education 13156 0.74 1.37 0 7 ALCS Education level of the household head. Gender 13156 0.99 0.07 0 1 ALCS Dummy for the gender of household head (male = 1). 23 Marital Status 13156 1.09 0.55 1 5 ALCS Marital Status of the household s head. Type of Dwelling 13156 1.41 0.85 1 6 ALCS Type of dwelling of the household. Household Size (log) 13156 2.06 0.38 0.69 3.61 ALCS Number of members in a household (in log). Males per Household (log) 13156 1.49 0.40 0 3.04 ALCS Number of male members in a household (in log). All figures are reported in the local currency (Afghani). Table 2: Income Uncertainty and Wealth Accumulation Variables GSL Non-Housing Wealth (1) (2) (3) (4) (5) (6) (7) (8) Variance of Income (log) (-3.16∗∗∗ ) (-3.35∗∗∗ ) (-2.33∗∗∗ ) (-3.29∗∗∗ ) (-0.42∗∗∗ ) (-0.55∗∗∗ ) (-0.40∗∗∗ ) (-0.53∗∗∗ ) -(0.31)∗∗∗ -(0.38)∗∗∗ -(0.40)∗∗∗ -(0.38)∗∗∗ -(0.13)∗∗∗ -(0.13)∗∗∗ -(0.13)∗∗∗ -(0.13)∗∗∗ Income (log) (-0.32∗∗∗ ) (-0.48∗∗∗ ) (-0.47∗∗∗ ) (-0.49∗∗∗ ) (-0.60∗∗∗ ) (-0.62∗∗∗ ) (-0.92∗∗∗ ) (-0.62∗∗∗ ) -(0.07)∗∗∗ -(0.07)∗∗∗ -(0.08)∗∗∗ -(0.07)∗∗∗ -(0.02)∗∗∗ -(0.02)∗∗∗ -(0.03)∗∗∗ -(0.02)∗∗∗ 24 Household Controls (-No∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) Province Fixed Effects (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) Conflict (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) Observations -8,230∗∗∗ -8,230∗∗∗ -8,230∗∗∗ -8,230∗∗∗ -13,156∗∗∗ -13,156∗∗∗ -13,156∗∗∗ -13,156∗∗∗ Adjusted R2 -0.10∗∗∗ -0.26∗∗∗ -0.34∗∗∗ -0.27∗∗∗ -0.00∗∗∗ -0.11∗∗∗ -0.22∗∗∗ -0.12∗∗∗ The table shows the estimated coefficients of equation (2) after setting β2 = β3 = 0. Robust standard errors are in parentheses. The two dependent variables are GSL (gold+silver+livestock) and Non-Housing Wealth (income-non-durable expenditure+gold+silver+livestock-debt). Household/Household-head controls include gender, age, marital status, education, household size, number of males per household, and type of dwelling. Significance levels: ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1. Table 3: Income Uncertainty and Jewelry and Livestock Accumulation Variables Gold + Silver Livestock (1) (2) (3) (4) (5) (6) (7) (8) Variance of Income (log) (-0.65∗∗∗ ) (-0.51∗∗∗ ) (-0.39∗∗∗ ) (-0.50∗∗∗ ) (-3.65∗∗∗ ) (-3.95∗∗∗ ) (-2.91∗∗∗ ) (-3.95∗∗∗ ) -(0.53)∗∗∗ -(0.48)∗∗∗ -(0.48)∗∗∗ -(0.48)∗∗∗ -(0.32)∗∗∗ -(0.41)∗∗∗ -(0.43)∗∗∗ -(0.41)∗∗∗ Income (log) (-1.11∗∗∗ ) (-0.79∗∗∗ ) (-0.90∗∗∗ ) (-0.79∗∗∗ ) (-1.12∗∗∗ ) (-1.15∗∗∗ ) (-0.16∗∗∗ ) (-1.15∗∗∗ ) -(0.06)∗∗∗ -(0.06)∗∗∗ -(0.07)∗∗∗ -(0.06)∗∗∗ -(0.07)∗∗∗ -(0.07)∗∗∗ -(0.08)∗∗∗ -(0.07)∗∗∗ 25 Household Controls (-No∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) Province Fixed Effects (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) Conflict (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) Observations -13156∗∗∗ -13156∗∗∗ -13156∗∗∗ -13156∗∗∗ -13156∗∗∗ -13156∗∗∗ -13156∗∗∗ -13156∗∗∗ R-squared -0.05∗∗∗ -0.10∗∗∗ -0.18∗∗∗ -0.10∗∗∗ -0.03∗∗∗ -0.20∗∗∗ -0.31∗∗∗ -0.20∗∗∗ The table shows the estimated coefficients of equation (2) after setting β2 = β3 = 0. Robust standard errors are in parentheses. Household/Household-head controls include gender, age, marital status, education, household size, number of males per household, and type of dwelling. Significance levels: ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1. Table 4: Conflict and Wealth Accumulation Variables GSL Gold & Livestock Non-Housing Silver Wealth (1) (2) (3) (4) Variance of Income (log) (-2.27∗∗∗ ) (-0.45∗∗∗ ) (-2.77∗∗∗ ) (-0.37∗∗∗ ) -(0.48)∗∗∗ -(0.22)∗∗∗ -(0.50)∗∗∗ -(0.17)∗∗∗ Conflict × Variance of Income (log) (-0.09∗∗∗ ) (-1.97∗∗∗ ) (-0.32∗∗∗ ) (-0.07∗∗∗ ) 26 -(0.83)∗∗∗ -(0.89)∗∗∗ -(0.87)∗∗∗ -(0.26)∗∗∗ Conflict (-1.09∗∗∗ ) (-0.04∗∗∗ ) (-2.61∗∗∗ ) (-0.70∗∗∗ ) -(0.79)∗∗∗ -(0.67)∗∗∗ -(0.80)∗∗∗ -(0.21)∗∗∗ Income (log) (-0.47∗∗∗ ) (-0.90∗∗∗ ) (-0.16∗∗∗ ) (-0.92∗∗∗ ) -(0.08)∗∗∗ -(0.07)∗∗∗ -(0.08)∗∗∗ -(0.03)∗∗∗ Observations -13156∗∗∗ -13156∗∗∗ -13156∗∗∗ -8230∗∗∗ R-squared -0.22∗∗∗ -0.18∗∗∗ -0.31∗∗∗ -0.34∗∗∗ The table shows the estimated coefficients of equation (2). Robust standard errors are in parentheses. Household/Household-head controls include gender, age, marital status, education, household size, number of males per household, and type of dwelling. Significance levels: ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1. Table 5: Conflict and Wealth Accumulation: & Mean Cut-Off for Conflict Dummy Variables GSL Gold & Livestock Non-Housing Silver Wealth (1) (2) (3) (4) Variance of Income (log) (-2.25∗∗∗ ) (-0.42∗∗∗ ) (-2.73∗∗∗ ) (-0.38∗∗∗ ) -(0.44)∗∗∗ -(0.19)∗∗∗ -(0.46)∗∗∗ -(0.15)∗∗∗ Conflict × Variance of Income (log) (-0.22∗∗∗ ) (-3.16∗∗∗ ) (-0.66∗∗∗ ) (-0.09∗∗∗ ) -(0.96)∗∗∗ -(0.97)∗∗∗ -(0.99)∗∗∗ -(0.29)∗∗∗ 27 Conflict (-1.19∗∗∗ ) (-0.80∗∗∗ ) (-2.85∗∗∗ ) (-0.72∗∗∗ ) -(0.86)∗∗∗ -(0.73)∗∗∗ -(0.86)∗∗∗ -(0.23)∗∗∗ Income (log) (-0.47∗∗∗ ) (-0.89∗∗∗ ) (-0.16∗∗∗ ) (-0.92∗∗∗ ) -(0.08)∗∗∗ -(0.07)∗∗∗ -(0.08)∗∗∗ -(0.03)∗∗∗ Observations -13156∗∗∗ -13156∗∗∗ -13156∗∗∗ -8230∗∗∗ R-squared -0.22∗∗∗ -0.18∗∗∗ -0.31∗∗∗ -0.34∗∗∗ The table shows the estimated coefficients of equation (2). A conflict is classified high-conflict if it experienced more terrorist attacks from 1973 to 2014 compared to the mean of all provinces. Robust standard errors are in parentheses. Household/Household-head controls include gender, age, marital status, education, household size, number of males per household, and type of dwelling. Significance levels: ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1. Table 6: Conflict and Wealth Accumulation: Excluding Helmand & Kandahar, Mean Cut-Off for Conflict Dummy Variables GSL Gold & Livestock Non-Housing Silver Wealth (1) (2) (3) (4) Variance of Income (log) (-2.41∗∗∗ ) (-0.56∗∗∗ ) (-2.89∗∗∗ ) (-0.38∗∗∗ ) -(0.46)∗∗∗ -(0.20)∗∗∗ -(0.47)∗∗∗ -(0.16)∗∗∗ Conflict × Variance of Income (log) (-0.08∗∗∗ ) (-2.59∗∗∗ ) (-0.41∗∗∗ ) (-0.07∗∗∗ ) -(0.88)∗∗∗ -(0.95)∗∗∗ -(0.93)∗∗∗ -(0.27)∗∗∗ 28 Conflict (-1.09∗∗∗ ) (-0.43∗∗∗ ) (-2.65∗∗∗ ) (-0.69∗∗∗ ) -(0.81)∗∗∗ -(0.71)∗∗∗ -(0.83)∗∗∗ -(0.22)∗∗∗ Income (log) (-0.48∗∗∗ ) (-0.96∗∗∗ ) (-0.21∗∗∗ ) (-0.89∗∗∗ ) -(0.09)∗∗∗ -(0.08)∗∗∗ -(0.08)∗∗∗ -(0.03)∗∗∗ Observations -12060∗∗∗ -12060∗∗∗ -12060∗∗∗ -7804∗∗∗ R-squared -0.21∗∗∗ -0.18∗∗∗ -0.32∗∗∗ -0.34∗∗∗ The table shows the estimated coefficients of equation (2). A conflict is classified high-conflict if it experienced more terrorist attacks from 1973 to 2014 compared to the mean of all provinces. The sample excludes households residing in Helmand and Kandahar which are the two most conflict-ridden provinces. Robust standard errors are in parentheses. Household/Household-head controls include gender, age, marital status, education, household size, number of males per household, and type of dwelling. Significance levels: ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1. Table 7: Conflict and Wealth Accumulation: Excluding Livestock Occupations, Mean Cut-Off for Conflict Dummy Variables GSL Gold & Livestock Non-Housing Silver Wealth (1) (2) (3) (4) Variance of Income (log) (-2.82∗∗∗ ) (-0.59∗∗∗ ) (-3.32∗∗∗ ) (-0.61∗∗∗ ) -(0.46)∗∗∗ -(0.20)∗∗∗ -(0.47)∗∗∗ -(0.15)∗∗∗ Conflict × Variance of Income (log) (-0.11∗∗∗ ) (-2.46∗∗∗ ) (-0.29∗∗∗ ) (-0.04∗∗∗ ) -(0.82)∗∗∗ -(0.95)∗∗∗ -(0.84)∗∗∗ -(0.26)∗∗∗ 29 Conflict (-1.21∗∗∗ ) (-0.23∗∗∗ ) (-2.91∗∗∗ ) (-0.75∗∗∗ ) -(0.78)∗∗∗ -(0.72)∗∗∗ -(0.78)∗∗∗ -(0.21)∗∗∗ Income (log) (-0.43∗∗∗ ) (-0.94∗∗∗ ) (-0.24∗∗∗ ) (-0.96∗∗∗ ) -(0.08)∗∗∗ -(0.08)∗∗∗ -(0.08)∗∗∗ -(0.03)∗∗∗ Observations -12364∗∗∗ -12364∗∗∗ -12364∗∗∗ -7484∗∗∗ R-squared -0.19∗∗∗ -0.17∗∗∗ -0.28∗∗∗ -0.29∗∗∗ The table shows the estimated coefficients of equation (2). A conflict is classified high-conflict if it experienced more terrorist attacks from 1973 to 2014 compared to the mean of all provinces. The sample excludes households who report an occupation with direct links to some livestock activity as one of their three main sources of income. Robust standard errors are in parentheses. Household/Household-head controls include gender, age, marital status, education, household size, number of males per household, and type of dwelling. Significance levels: ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1. Table 8: Income Uncertainty and Non-Durable Expenditures Variables Non-Durable Expenditures Income less Non-Durable Expenditures (1) (2) (3) (4) (5) (6) (7) (8) Variance of Income (log) (-0.13∗∗∗ ) (-0.14∗∗∗ ) (--0.04∗∗∗ ) (-0.14∗∗∗ ) (-0.12∗∗∗ ) (-0.15∗∗∗ ) (-0.10∗∗∗ ) (-0.14∗∗∗ ) -(0.05)∗∗∗ -(0.04)∗∗∗ -(0.04)∗∗∗ -(0.04)∗∗∗ -(0.07)∗∗∗ -(0.07)∗∗∗ -(0.08)∗∗∗ -(0.07)∗∗∗ Income (log) (-0.51∗∗∗ ) (-0.40∗∗∗ ) (-0.33∗∗∗ ) (-0.40∗∗∗ ) (-1.37∗∗∗ ) (-1.50∗∗∗ ) (-1.64∗∗∗ ) (-1.50∗∗∗ ) 30 -(0.01)∗∗∗ -(0.01)∗∗∗ -(0.01)∗∗∗ -(0.01)∗∗∗ -(0.01)∗∗∗ -(0.02)∗∗∗ -(0.02)∗∗∗ -(0.02)∗∗∗ Household Controls -No∗∗∗ -Yes∗∗∗ -Yes∗∗∗ -Yes∗∗∗ -No∗∗∗ -Yes∗∗∗ -Yes∗∗∗ -Yes∗∗∗ Province FE -No∗∗∗ -No∗∗∗ -Yes∗∗∗ -No∗∗∗ -No∗∗∗ -No∗∗∗ -Yes∗∗∗ -No∗∗∗ Conflict -No∗∗∗ -No∗∗∗ -No∗∗∗ -Yes∗∗∗ -No∗∗∗ -No∗∗∗ -No∗∗∗ -Yes∗∗∗ Observations -13156∗∗∗ -13156∗∗∗ -13156∗∗∗ -13156∗∗∗ -9123∗∗∗ -9123∗∗∗ -9123∗∗∗ -9123∗∗∗ The table shows the estimated coefficients of equation (2). Robust standard errors are in parentheses. The two dependent variables are annual expenditure and Income less expenditure on non-durable goods. Household/Household-head controls include gender, age, marital status, education, household size, number of males per household, and type of dwelling. Significance levels: ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1. Table 9: Income Uncertainty and Wealth Accumulation: Excluding Livestock Occupations Variables GSL Non-Housing Wealth (1) (2) (3) (4) (5) (6) (7) (8) Variance of Income (log) (-3.85∗∗∗ ) (-3.76∗∗∗ ) (-2.78∗∗∗ ) (-3.76∗∗∗ ) (-0.77∗∗∗ ) (-0.75∗∗∗ ) (-0.62∗∗∗ ) (-0.75∗∗∗ ) -(0.34)∗∗∗ -(0.37)∗∗∗ -(0.38)∗∗∗ -(0.37)∗∗∗ -(0.12)∗∗∗ -(0.13)∗∗∗ -(0.12)∗∗∗ -(0.13)∗∗∗ Income (log) (-0.30∗∗∗ ) (-0.49∗∗∗ ) (-0.43∗∗∗ ) (-0.49∗∗∗ ) (-0.67∗∗∗ ) (-0.68∗∗∗ ) (-0.96∗∗∗ ) (-0.67∗∗∗ ) -(0.07)∗∗∗ -(0.08)∗∗∗ -(0.08)∗∗∗ -(0.08)∗∗∗ -(0.02)∗∗∗ -(0.03)∗∗∗ -(0.03)∗∗∗ -(0.03)∗∗∗ 31 Household Controls (-No∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) Province Fixed Effects (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) Conflict (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) Observations -12364∗∗∗ -12364∗∗∗ -12364∗∗∗ -12364∗∗∗ -7484∗∗∗ -7484∗∗∗ -7484∗∗∗ -7484∗∗∗ R-squared -0.01∗∗∗ -0.08∗∗∗ -0.19∗∗∗ -0.08∗∗∗ -0.14∗∗∗ -0.19∗∗∗ -0.29∗∗∗ -0.20∗∗∗ The table shows the estimated coefficients of equation (2) after setting β2 = β3 = 0. The two dependent variables are GSL (gold+silver+livestock) and Non-Housing Wealth (income-non-durable expenditure+gold+silver+livestock-debt). The sample excludes households who report an occupation with direct links to some livestock activity as one of their three main sources of income. Robust standard errors are in parentheses. Household/Household-head controls include gender, age, marital status, education, household size, number of males per household, and type of dwelling. Significance levels: ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1. Table 10: Income Uncertainty and Jewelry and Livestock Accumulation: Excluding Livestock Occupations Variables Gold & Silver Livestock (1) (2) (3) (4) (5) (6) (7) (8) Variance of Income (log) (-0.48∗∗∗ ) (-0.43∗∗∗ ) (-0.34∗∗∗ ) (-0.42∗∗∗ ) (-4.47∗∗∗ ) (-4.40∗∗∗ ) (-3.43∗∗∗ ) (-4.41∗∗∗ ) -(0.53)∗∗∗ -(0.48)∗∗∗ -(0.48)∗∗∗ -(0.48)∗∗∗ -(0.35)∗∗∗ -(0.39)∗∗∗ -(0.40)∗∗∗ -(0.39)∗∗∗ Income (log) (-1.15∗∗∗ ) (-0.83∗∗∗ ) (-0.95∗∗∗ ) (-0.83∗∗∗ ) (-1.15∗∗∗ ) (-1.20∗∗∗ ) (-0.24∗∗∗ ) (-1.20∗∗∗ ) -(0.06)∗∗∗ -(0.07)∗∗∗ -(0.08)∗∗∗ -(0.07)∗∗∗ -(0.07)∗∗∗ -(0.07)∗∗∗ -(0.08)∗∗∗ -(0.07)∗∗∗ 32 Household Controls (-No∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) Province Fixed Effects (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) Conflict (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) Observations -12364∗∗∗ -12364∗∗∗ -12364∗∗∗ -12364∗∗∗ -12364∗∗∗ -12364∗∗∗ -12364∗∗∗ -12364∗∗∗ R-squared -0.05∗∗∗ -0.10∗∗∗ -0.17∗∗∗ -0.10∗∗∗ -0.03∗∗∗ -0.16∗∗∗ -0.28∗∗∗ -0.16∗∗∗ The table shows the estimated coefficients of equation (2) after setting β2 = β3 = 0. The sample excludes households who report an occupation with direct links to some livestock activity as one of their three main sources of income. Robust standard errors are in parentheses. Household/Household-head controls include gender, age, marital status, education, household size, number of males per household, and type of dwelling. Significance levels: ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1. Table 11: Income Uncertainty and Wealth Accumulation: Excluding Helmand & Kandahar Variables GSL Non-Housing Wealth (1) (2) (3) (4) (5) (6) (7) (8) Variance of Income (log) (-3.00∗∗∗ ) (-3.10∗∗∗ ) (-2.44∗∗∗ ) (-3.11∗∗∗ ) (-0.37∗∗∗ ) (-0.49∗∗∗ ) (-0.41∗∗∗ ) (-0.49∗∗∗ ) -(0.30)∗∗∗ -(0.38)∗∗∗ -(0.40)∗∗∗ -(0.38)∗∗∗ -(0.13)∗∗∗ -(0.13)∗∗∗ -(0.13)∗∗∗ -(0.13)∗∗∗ Income (log) (-0.28∗∗∗ ) (-0.36∗∗∗ ) (-0.48∗∗∗ ) (-0.37∗∗∗ ) (-0.61∗∗∗ ) (-0.65∗∗∗ ) (-0.89∗∗∗ ) (-0.64∗∗∗ ) -(0.08)∗∗∗ -(0.08)∗∗∗ -(0.09)∗∗∗ -(0.08)∗∗∗ -(0.02)∗∗∗ -(0.02)∗∗∗ -(0.03)∗∗∗ -(0.03)∗∗∗ 33 Household Controls (-No∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) Province Fixed Effects (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) Conflict (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) Observations -12060∗∗∗ -12060∗∗∗ -12060∗∗∗ -12060∗∗∗ -7804∗∗∗ -7804∗∗∗ -7804∗∗∗ -7804∗∗∗ R-squared -0.00∗∗∗ -0.12∗∗∗ -0.21∗∗∗ -0.13∗∗∗ -0.10∗∗∗ -0.26∗∗∗ -0.34∗∗∗ -0.27∗∗∗ The table shows the estimated coefficients of equation (2) after setting β2 = β3 = 0. The two dependent variables are GSL (gold+silver+livestock) and Non-Housing Wealth (income-non-durable expenditure+gold+silver+livestock-debt). The sample excludes households residing in Helmand and Kandahar which are the two most conflict-ridden provinces. Robust standard errors are in parentheses. Household/Household-head controls include gender, age, marital status, education, household size, number of males per household, and type of dwelling. Significance levels: ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1. Table 12: Income Uncertainty and Jewelry and Livestock Accumulation: Excluding Helmand & Kandahar Variables Gold & Silver Livestock (1) (2) (3) (4) (5) (6) (7) (8) Variance of Income (log) (-0.45∗∗∗ ) (-0.37∗∗∗ ) (-0.41∗∗∗ ) (-0.37∗∗∗ ) (-3.64∗∗∗ ) (-3.79∗∗∗ ) (-3.04∗∗∗ ) (-3.80∗∗∗ ) -(0.52)∗∗∗ -(0.47)∗∗∗ -(0.47)∗∗∗ -(0.47)∗∗∗ -(0.32)∗∗∗ -(0.41)∗∗∗ -(0.43)∗∗∗ -(0.41)∗∗∗ Income (log) (-1.25∗∗∗ ) (-0.90∗∗∗ ) (-0.97∗∗∗ ) (-0.90∗∗∗ ) (-1.18∗∗∗ ) (-1.11∗∗∗ ) (-0.21∗∗∗ ) (-1.11∗∗∗ ) -(0.07)∗∗∗ -(0.07)∗∗∗ -(0.08)∗∗∗ -(0.07)∗∗∗ -(0.07)∗∗∗ -(0.07)∗∗∗ -(0.08)∗∗∗ -(0.07)∗∗∗ 34 Household Controls (-No∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) (-Yes∗∗∗ ) Province Fixed Effects (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) Conflict (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-No∗∗∗ ) (-Yes∗∗∗ ) Observations -12060∗∗∗ -12060∗∗∗ -12060∗∗∗ -12060∗∗∗ -12060∗∗∗ -12060∗∗∗ -12060∗∗∗ -12060∗∗∗ R-squared -0.06∗∗∗ -0.10∗∗∗ -0.17∗∗∗ -0.10∗∗∗ -0.03∗∗∗ -0.22∗∗∗ -0.32∗∗∗ -0.23∗∗∗ The table shows the estimated coefficients of equation (2) after setting β2 = β3 = 0. The sample excludes households residing in Helmand and Kandahar which are the two most conflict-ridden provinces. Robust standard errors are in parentheses. Household/Household-head controls include gender, age, marital status, education, household size, number of males per household, and type of dwelling. Significance levels: ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1. 0 100000 200000 300000 400000 0 100000 200000 300000 Daykundi Kandahar Ghor Kapisa Wardak Helmand Farah Daykundi Baghlan Urozgan Zabul Paktya Samangan Kabul Baghdis Panjsher Urozgan Baghlan Takhar Parwan Kapisa Takhar Sar−e−Pul Jawzjan Paktika Kunar Laghman Herat Livestock (Afghani) Kunar Ghazni Annual Income (Afghani) Bamyan Balkh 35 Parwan Laghman Jawzjan Logar Nangahar Nimroz Panjsher Kunduz Balkh Khost Nooristan Wardak Kunduz Farah Paktya Samangan Nimroz Zabul Badakhshan Paktika Logar Bamyan Kandahar Ghor Herat Badakhshan Faryab Sar−e−Pul Ghazni Nooristan Gold and Silver (Afghani) Helmand Baghdis Khost Nangahar Annual Expenditure (Afghani) Kabul Faryab Figure 1: Livestock and Jewelry Ownership Across Provinces Figure 2: Annual Household Income and Expenditure Across Provinces Figure 3: Income Uncertainty and Narrow Wealth 12 Paktika Khost Kunarha 10 Bamyan Baghdis Zabul Wardak Sar−e−Pul Ghor Nangarhar Farah Samangan Paktya Badakhshan Faryab Laghman Ghazni 8 Kapisa Urozgan Daykundi Panjsher 36 Takhar Balkh Parwan Kunduz Baghlan Helmand Nooristan 6 Jawzjan Nimroz Herat Gold, Silver and Livestock (log) Logar Kabul Kandahar 4 .058 .06 .062 .064 .066 .068 Variance of Log Income (log)/Income (log) Figure 4: Number of Terrorist Attacks across Provinces (a) 1973 – 2014 (b) 2012 37 (67,227] (380,862] (35,67] (206,380] (23,35] (109,206] (11,23] (67,109] [0,11] [3,67] The above maps show the variation in the total number of terrorist attacks across the Afghan provinces. The data on attacks comes from the Global Terrorism Dataset (GTD) and the shape-file used to create the maps is taken from the DIVA-GIS project. The maps are cleared by World Bank’s Map Design Cartography as of May 23rd, 2018, these maps meet the Bank’s minimum standards for the depiction of international boundaries and formal country names. Figure 5: Wealth Accumulation by Income Quantiles (a) GSL (b) Non-Housing Wealth 6 2 4 1 38 2 0 Estimate Estimate 0 −1 −2 −2 1 2 3 4 5 1 2 3 4 5 Quantiles Quantiles A Appendices A.1 Estimation of Livestock Value The price data for livestock are neither available at the provincial nor available at the national level. While it is possible to get some estimates for a couple of larger livestock types, we are not able to obtain reasonable prices for most other types. As such, we collect prices of livestock from three different sources to aggregate the wealth accumulated in the form of livestock. Table A.1 lists these prices. Table A.1: Prices of Livestock Used In Estimating Livestock Wealth (2014) Income Source US Dollars (1) Cattle 140 Oxen 150 Horses 1,500 Donkeys 500 Camels 1,000 Goats 100 Sheep 150 Chicken 20 Other Birds 10 Sources: Cattle.com, BusinessInsider.Com, Ino.Com. We note two concerns with the price data. First, we find the price of donkeys to be particularly high – more than three times that of price of cattle and oxen. Second, we find the price of chicken and other birds to be unreasonably higher as well. To check if these prices quantitatively affect our findings, we consider several specifications in which we reduce the price of these three livestock types by a factor of 3 – 5. We observe no major changes in our findings. A.2 Household Income and Occupations To estimate the income uncertainty of households, we categorize them according to the occu- pations from which household income is derived. Table A.2 shows the complete list of occupations in our sample from the 2013-2014 round of the ALCS. As many of the household controls that we use are at the household head level, We exclude those households with heads who were under 18 or over 64 years old, or the household head was retired or unemployed. We also drop those 39 households for which any of the three main sources of income corresponds to non-labor income earning such as pension earnings, rental income, Zakat, or borrowing. Table A.2: Sources of Income Income Source Frequency Percent (1) (2) Production and sale of field crops (non-opium) 2780 21.13 Production and sale of opium 119 0.90 Production and sale of orchard products 371 2.82 Agricultural wage labor (non opium) 319 2.42 Production and sale of livestock 792 6.02 Shepherding wage labor 157 1.19 Sewing, embroidery etc. 127 0.97 Handicraft work, not classified elsewhere 51 0.39 Food production and processing 124 0.94 Mechanics 121 0.92 Road/building construction 17 0.13 Production work, not classified elsewhere 48 0.36 Teachers 417 3.17 Medicine (doctors, nurses, and others) 108 0.82 Military 294 2.23 Police 501 3.81 Office work, government 365 2.77 Office work, non-government 222 1.69 NGO, UN and government work not classified elsewhere 42 0.32 Taxi/transport 827 6.28 Security 42 0.32 Service work, not classified elsewhere 268 2.04 Shop keeping and small business 1419 10.78 Street and market sales 144 1.09 Trade, not classified elsewhere 180 1.37 Wage labor, not classified elsewhere 1764 13.41 Day labor, not classified elsewhere 1238 9.41 Remittances from migrants 302 2.30 A.3 Dwelling Characteristics Though we do not directly control for household wealth in our analysis, we add many dummies for dwelling characteristics which are reported in the survey. Table A.3 reports these dwelling types together with the distribution of households across the dwelling types. Approximately 72 percent 40 of the households report living in a single family home. Yet, more than 5 percent of the households live in fluid temporary settings like tents and shacks. Table A.3: Dwelling Characteristics and Distribution Frequency Percent (1) (2) Single Family Home 9,471 72.0 Part of a Shared House 2,974 22.6 Apartment 18 0.1 Tent 374 2.8 Temporary Shelter/Shack 300 2.3 Other 19 0.1 41