Gender differences in risk behavior: An analysis of asset allocation decisions in Ghana

World Development 117 (2019) 127–137

Contents lists available at ScienceDirect

World Development journal homepage: www.elsevier.com/locate/worlddev

Gender differences in risk behavior: An analysis of asset allocation decisions in Ghana Marya Hillesland ⇑ American University, Department of Economics, 4400 Massachusetts Ave, NW, Washington, DC 20016, United States

a r t i c l e

i n f o

Article history: Accepted 4 January 2019

JEL classification: D8 G1 O1 D3 Keywords: Gender difference Risk preferences Ghana Wealth Asset allocation decisions Risk aversion

a b s t r a c t Within a developing country context, little is known about gender differences in risk preferences as reflected in asset allocation decisions. Much of the empirical literature within development studies on risk and investment in assets centers on risk coping and risk management strategies of households. With some exceptions, this literature primarily focuses on household and household-level outcomes. Previous experimental studies have explored gender differences in risk preferences in developing countries with inconclusive results. Using unique national-representative sex-disaggregated data with selfreported asset ownership and wealth information of individuals within households in Ghana from a multi-country project, The Gender Asset Gap Project, this paper explores men and women’s risk preferences as reflected in asset allocation decisions through a decomposition method typically used to explore gender differences in wage employment. The study finds that although women hold significantly fewer risky assets than men in absolute terms and as a proportion of their wealth in the sample, men and women do not have systematically different risk preferences. The results in this paper differ from the results in many empirical studies in developed countries, where women are often found to be more risk averse than men. As the first study to look at gender differences in risk aversion in terms of asset allocation with in the developing country context, the results from this study provide evidence that gender differences in risk attitudes may vary by cultural context as the growing experimental literature in developing countries suggests. Ó 2019 Elsevier Ltd. All rights reserved.

1. Introduction Physical and financial assets serve a number of important functions. Physical assets—such as a household’s residence and vehicles—may provide current and future consumption value and also be a means of production, generating future consumption flows. Financial assets have a monetary value that can be converted into future consumption. Both types of assets may generate profits or losses as well as rent or interest. They also may be held as a form of savings as a way to self-insure against possible future economic hardships. An individual’s propensity for risk influences how his or her assets are allocated between savings and riskier assets that may generate profits or result in losses. Risk averse individuals prefer to invest in secure assets or assets with a constant rate of return over risky assets of the same expected value with a variable rate ⇑ Address: Food and Agricultural Organization of the United Nations, Viale delle Terme di Caracalla, 00153 Rome, Italy. E-mail addresses: [email protected], [email protected] https://doi.org/10.1016/j.worlddev.2019.01.001 0305-750X/Ó 2019 Elsevier Ltd. All rights reserved.

of return. For risk averse individuals, investing in risky assets requires that the expected return of risky assets is greater than the return from risk-free assets. The difference must be large enough to offset the value the individual places on the cost of taking the risk. Within a developing country context, little is known about gender differences in risk preferences as reflected in asset allocation decisions. Much of the literature within development studies on risk and investment in assets focuses on risk coping and risk management strategies of the household as opposed to individuals within households. As a coping strategy, accumulating assets is an important form of self-insurance against consumption loss due to shocks. Poor households with relatively low asset holdings in developing countries often hold a significant share of their assets as precautionary savings in the form of cash or other assets to insure against possible negative shocks (Fafchamps, Udry, & Czukas, 1998; Lee & Sawada, 2010; Park, 2006). Poor households have an even greater propensity to save when faced with liquidity and credit constraints (Lee & Sawada, 2010).

128

M. Hillesland / World Development 117 (2019) 127–137

In many low-income countries, where wage labor opportunities are minimal, households invest their assets into different productive activities. As a form of risk management, poor households with lower resilience and higher sensitivity to shocks often seek to minimize risk by allocating their wealth into low-risk, low-return activities rather than higher-risk, higher-return activities to cushion themselves from the effects from possible shocks (e.g. Dercon, 2005; Zimmerman & Carter, 2003; Dercon & Christiaensen, 2011). The discussion on asset portfolios, risk management, and risk coping strategies in developing countries largely focuses on the household and household level outcomes. There are a few studies that explore gender differences in asset allocation and accumulation (Antonopoulos & Floro, 2005; Quisumbing, 2010; Oduro & Doss, 2018) and gender differences in the effects of negative household shocks within developing countries (Dercon & Krishnan, 2000; Doss, 2001; Doss, McPeak, & Barrett, 2008). None, however, specifically explore gender differences in risk preferences as reflected in asset allocation decisions. This gap is primarily due to the lack of sex disaggregated asset and wealth data. Until recently, most household surveys that collected wealth and asset data did so at the household level with the assumption that wealth held by individuals within the household is pooled. As such, sex-disaggregated wealth data is not readily available, particularly within a developing country context. To begin to address this gap, a nationally representative sexdisaggregated asset and wealth survey was implemented in Ghana in 2010 as part of a multi-country project, The Gender Asset Gap Project. Using this unique sex-disaggregated data containing detailed self-reported information on asset ownership and wealth from two individuals of the opposite sex in each household, this paper explores gender differences in risk preferences in terms of asset allocations in Ghana. Similar empirical studies use household-level data from the United States (e.g. Jianakoplos & Bernasek, 1998; Sundén & Surette, 1998). This is the first study of its kind within a developing country context with individually self-reported assets. Previous experimental studies have explored gender differences in risk preferences in a developing country with inconclusive results. Many of these studies find women are more risk averse than men (Fletschner, Anderson, & Cullen, 2010; Cárdenas, Dreber, Von Essen, & Ranehill, 2012; Khachatryan, Dreber, Von Essen, & Ranehill, 2015). However, some suggest the setting and surrounding environment may play a considerable role in gender differences in preferences (Schubert, Brown, Gysler, & Brachinger, 1999; Henrich et al., 2001; Gneezy, Leonard, & List, 2009; Booth & Nolen, 2012; Booth, CardonaSosa, & Nolen, 2014). Ghana is a fitting place to examine whether women are more risk averse than men as reflected in individual asset allocation decisions because wealth is primarily held individually rather than pooled with other household members (Deere, Oduro, Swaminathan, & Doss, 2013). Legal norms facilitate a strong separation of property even within marriage, meaning individual property acquired before and during marriage—including inheritance and gifts as well as all earnings from wages, salaries, and income activities—remains the property of the individual rather than joint marital property (Deere et al., 2013; Deere & Doss, 2006). Gender differences in risk aversion is an important topic in development. If risk aversion in terms of asset allocation decisions can proxy for risk aversion in other ways, such as in decisions on technology adoption or in types of self-employment activities, gender differences in risk aversion in Ghana means that programs and

interventions that require some amount of risk taking will need to be tailored to fit these differences to ensure gender inclusiveness.

2. Previous literature on gender and risk aversion Many of the empirical studies that have explored gender differences in risk preferences in terms of asset allocation decisions have done so using data from the United States, most of which focus on retirement portfolios with mixed findings (Riley & Chow, 1992; Bajtelsmit & VanDerhei, 1997; Sundén & Surette, 1998; Bernasek & Shwiff, 2001; Arano, Parker, & Terry, 2010). Jianakoplos and Bernasek (1998) is one of the only empirical studies that explores gender differences in risk preferences as reflected in asset allocation beyond pension decisions. Using 1989 Survey of Consumer Finances (SCF) data of 3143 households, Jianakoplos and Bernasek (1998) find that single female-headed and couple households have greater relative risk aversion as reflected in financial asset allocation decisions than single male-headed households. Comparing risk aversion of couple households to single households is problematic, however, as it assumes married couples act as an individual (i.e. act in unison with no conflicting interests). Knowledge of decision-making within the household would help tease out individual risk attitudes. Indeed, Jianakoplos and Bernasek (1998) admit their estimates are ‘‘clouded” by not knowing who makes the decisions about asset allocation within the household. Within the experimental literature many studies focus on university students in the United States, largely finding women are more risk averse than men to varying degrees (see survey by Eckel & Grossman, 2008; Croson & Gneezy, 2009; Charness & Gneezy, 2012) with some criticism (Nelson, 2015, 2016). The few studies that explore gender differences in risk preferences within developing countries also mostly find that men are less risk averse than women (Fletschner et al., 2010; Cárdenas et al., 2012; Khachatryan et al., 2015). There is some evidence, however, that gender differences in risk preferences may vary by country. For example, Cárdenas et al. (2012) find that among a sample of children in Colombia and Sweden, girls are more risk averse than boys; however, the gender differences in risk preferences is much greater in Columbia. Similarly, there is some evidence that gender differences in risk preferences change depending on how the background circumstances are framed (Schubert et al., 1999) and the surrounding environment, such as the sex-composition of classrooms (Booth & Nolen, 2012; Booth et al., 2014). Differences in findings across countries and environment suggest that it may be that differences found in men and women’s risk preferences are likely a consequence of men and women facing different constraints or due to learned societal traits, rather than reflecting actual risk preferences due to biological sex differences. That is, men are not necessarily innately more risk taking than women, but rather differences arrive through cultural pressures, which differ across countries.

3. Conceptualizing risk aversion Portfolio studies in high income countries often gauge risk aversion as measured by the proportion of risky assets held to one’s worth. Assuming that individuals are risk averse and that their utility functions can be characterized by the mean and variance of final wealth, Mossin (1966) shows that in a competitive market—where assets are perfectly divisible, assets can be sold at any point in time, there are no transaction costs, and the expected yield on an asset is a random variable whose distribution is known

129

M. Hillesland / World Development 117 (2019) 127–137

to individuals—there is a general equilibrium for the market price of risk.1 Using this equilibrium assumption, Budd and Litzenberger (1972) extend Lintner’s (1970) examination of the relationship between the size of the market and the market for risky assets. If we assume that an investor has a utility function that is twice differentiable and U 0 ðW Þ >0 and U 00 ðW Þ < 0, where W is the investor’s asset wealth, then the market price of risk is equal to the inverse of the sum of the measure of individual investors’ absolute risk aversion, multiplied by the aggregate market value of all the risky assets (Lintner, 1970; Budd & Litzenberger, 1972). Using this relationship, Friend and Blume (1975) use a Taylor series expansion to show that the ratio of the value of total liquid assets across k investors is a function of individual investors’ risk tolerance. This means that the proportion of wealth held in risky assets to total liquid wealth is inversely related to Pratt’s (1964) measure of relative risk aversion. Pratt’s (1964) measure of relative risk aversion is a single dimensional outcome variable that captures an individual’s propensity to take risks with respect to wealth. In order to estimate gender differences in risk preferences based on the asset allocation decisions of individuals, a simple version of the Friend and Blume’s (1975) measure is adapted slightly to better fit a context like Ghana. Suppose a risk averse individual k, who has a utility function with respect to wealth, W, that is twice differentiable and U 0 ðW Þ >0 and U 00 ðW Þ < 0, must decide what proportion, a, of her worth to allocate her assets in risky, productive investments with a random return, where the expected rate of return is ~r r , and 1  a in secure assets with a non-variable rate of return, r f . The individual chooses a such that the expected value of wealth in 

 some future period, W tþ1 ¼ 1 þ rf W t þ ak W t ~r r  rf , maximizes his or her expected utility:




 max E uk 1 þ r f W t þ ak W t ~r r  r f

ð1Þ

To simplify, let the non-variable rate of return equal zero, r f ¼ 0, so that the difference in the return due to investing in risky assets is ~r r . The first order condition is then


 E ~r r W t u0k W t þ ak W t ~r r ¼ 0

ð2Þ


 Using a first order Taylor series to expand uk W t þ ak W t r r around W t , Eq. (2) is approximately 0

 
 E ~r r W t u0k ðW t Þ þ u00k ðW t Þ ak W t ~r r 0

ð3Þ

This can be rearranged so that

ak ¼

Eð~r r Þ

r2r







u0k ðW t Þ W t u00k ðW t Þ



ð4Þ ~

where ak is the optimal demand for risky assets, Erðr2r Þ is the price of r risk, and r2r is the variance of the additional return to risky assets. Pratt’s (1964) measure of relative risk aversion for the kth individðW t Þ so that Eq. (4) becomes ual is C k ðW t Þ ¼ W t U} U 0 ðW t Þ

ak ¼

Eð~r r Þ

r2r





1 C k ðW t Þ

ð5Þ

1 The Expected Utility Model is often used to represent behavior under conditions of risk. The Mean-Variance Model is a simplification of the Expected Utility Model, where utility can be expressed as the mean and variance of a probably distribution that gives an investor different wealth outcomes at different probabilities. The two models are equivalent when either investors’ utility functions can be represented by a function that has only two moments or the portfolio return distribution is an elliptical distribution. While neither assumption is entirely realistic, the Mean-Variance Model is considered a reasonable approximation to the solutions found in the Expected Utility Model.

~

In market equilibrium, Erðr2r Þ is constant across individuals. With conr

~

stant market price of risk, Erðr2r Þ, and a fixed non-variable rate of r return, rf , Eq. (5) suggests that the proportion of risky assets to all assets is inversely related to relative risk aversion for a given individual. This means that the lower an individual’s relative risk aversion (and thus absolute risk aversion), the greater the optimal demand for risky assets as a proportion of wealth, ak . ~

The market price of risk, Erðr2r Þ, assumes individuals have similar r expectations about the risks of assets. Additionally, only aggregate risks should affect prices. Idiosyncratic risks are assumed to be diversified away, so that the marginal price of risk is constant across individuals. This means individual shocks do not affect the market equilibrium and that no single individual is subject to a random asset price shock that is not shared with everyone else (i.e. individuals are only subject to covariate price shocks). If, for instance, an individual completely lost an asset (so that the return is less than the market price of the asset) due to an idiosyncratic shock, it is assumed the loss is diversified away and does not affect the marginal price of risk of holding this asset in the market. Investments in formal financial assets in markets where individuals are price-takers—meaning that no single individual in the market has the power to change prices—best meet the model’s assumptions. However, in a developing country context, only a small share of individuals access and use formal financial institutions. Individuals are more likely to have access to informal or semi-formal financial mechanisms locally and to save and invest in physical assets. In this context, the equilibrium is more likely to hold if the aggregate market is limited to a local market consisting of individuals with overlapping social networks engaged in risk-sharing and with similar expectations around the price of risk. While households in all countries are vulnerable to shocks, households in developing income countries are more likely to be subject to a lack of or limited access to functioning formal insurance mechanisms, formal credit markets, or other formal institutions to help them avoid consumption shortfalls when a shock occurs. However, studies suggest that in addition to self-insurance through savings, informal forms of insurance such as kinship networks help provide protection against loss of consumption due to idiosyncratic risks. Within the context of a developing country, the consumption risk sharing model states that when the market is complete, idiosyncratic changes in household income that is not smoothed through savings should be absorbed by other members of the risk sharing network. It means that idiosyncratic income shocks should not affect consumption. While full risk sharing is often rejected by the data, many empirical studies find at least partial risk sharing. Building on studies on risk sharing and networks, Ambrus, Mobius, and Szeidl (2014) shows that a full insurance model among households in villages can be nearly reached through loosely overlapping social networks. Among kinship networks, Chiappori, Samphantharak, Schulhofer-Wohl, and Townsend (2014) find nearly complete risk sharing among kin within villages in Thailand. Gifts and transfers among these households reduce the effect of liquidity constraints on household assets. In Vietnam, Sawada, Nakata, and Kotera (2017) take into account selfinsurance in the form of own use production of goods for consumption and cannot reject full risk sharing in Vietnam. Other empirical studies, although they generally reject full risk pooling, find evidence of insurance among households in developing countries to varying degrees (for example, Townsend, 1994, 1995; Grimard, 1997; Dercon & Krishnan, 2000; Ogaki & Zhang, 2001; Ligon Thomas, & Worrall, 2002; Murgai, Winters, Sadoulet, & Janvry, 2002; Fafchamps & Lund, 2003; De Weerdt & Dercon, 2006; Fafchamps & Gubert, 2007; Morduch, 2005; Mazzocco & Saini, 2012; Ambrus et al., 2014).

130

M. Hillesland / World Development 117 (2019) 127–137

Table 1 Risky and secure assets in Ghana. Riskier Assets (higher return)

More Secure Assets

 Stocks  Registered businesses (as owner or part owner) that can be sold (i.e. there’s a market)  Commercial and residential real estate where there is a market

Most of the literature explores risk sharing across households, which assumes that within the household, members share risk. In Ghana, there is some evidence that this may not be the case. Goldstein (1999), for example, finds that men and women within the same households in Ghana do not necessarily share risks and make different choices in their risk pooling groups and arrangements. Strategies used by individuals to reduce risk are often shared with individuals from other households, particularly those of the same family lineage (Goldstein, de Janvry, & Sadoulet, 2004; Vanderpuye-Orgle & Barrett, 2009). Within the Ghanian context, therefore, Eq. (5) holds if it can be limited to a local economy, u; where individuals within households participate in risk sharing arrangements or other formal and informal insurance mechanisms and have similar expectations about riskiness of assets:

ak ðWjuÞ ¼

Eð~rr Þ

r2r

 

1 C k ðW t Þ

ð6Þ

If there is limited risk sharing, and other formal and informal mechanisms do not insure against idiosyncratic risks (i.e. so that most individuals must reduce consumption in the face of an idiosyncratic shock), members of the local economy will have different expecta~

tions about the riskiness of assets and the price of risk, Erðr2r Þ, will not r

be constant across members. As a result, the price of risk of assets cannot be separated from individuals’ risk preferences and empirically it will be more difficult to capture any systematic difference in risk preferences between men and women. While resources are not necessarily pooled in households, we would expect decisions around assets to be influenced by an individual’s role within the household. We would also expect that gender norms and institutions to partly determine the types of assets men and women hold. Women, for instance, may not necessarily have equal access to formal financial products as men, which influences the type of asset portfolios women may hold as compared to men. Women may be more likely to hold safe assets not in formal savings account, but in informal savings accounts or in the form of cash. Because of this, this study uses an expanded definition of financial wealth to include all formal and informal financial assets. Since many men and women in Ghana hold the majority of their wealth in non-financial assets, it would be ideal to analyze portfolio decisions of individuals’ full range of financial and physicalasset wealth. However, it is difficult to classify many nonfinancial assets as risky and risk-free. Consumer durables and many productive assets are valued not just for their monetary rewards, but also for their present and future consumption value. Additionally, data on the returns to many assets are not available or possible to estimate with the data available. The next section describes how risky and non-risky assets are classified following previous literature. 4. Defining wealth and the classification of risky and non-risky assets in Ghana This study uses the same division of risky and non-risky assets as previous studies such as Jianakoplos and Bernasek (1998) and

 Savings in formal and informal accounts and cash  Treasury bills and bonds

Friend and Blume (1975). Risk-free financial assets are informal and formal savings accounts, cash holdings, and treasury bills and bonds. Risky financial assets include stocks.2 Although commercial and residential real estate and formal business enterprises are less divisible than financial assets, following Jianakoplos and Bernasek (1998) and Friend and Blume (1975), commercial and residential real estate that can be sold on the market and formally registered businesses are also included as risky assets (Table 1).3 5. Data and wealth statistics The data is from a sex-disaggregated asset and wealth household survey implemented in Ghana from May to July 2010, which was carried out as part of a multi-country gender and assets project, The Gender Asset Gap Project. In most households, two individuals of the opposite sex, who were best informed about the household’s assets, provided information on the household’s assets and wealth as well as their own assets and wealth. The sampling frame is based on 144 enumeration areas from the national census. The number of enumeration areas selected within each of Ghana’s ten regions was based on the region’s share of the total population.4 Within each enumeration area, 15 households were randomly selected to be surveyed. In all, 2170 households were surveyed.5 The final sample for this analysis is 1341 males and 1690 females from 2080 of these households. Table 2 presents gross wealth and asset allocations by sex and Table 3 presents gross wealth and asset allocation across the wealth distribution. The majority of assets are owned individually.6 The value of any assets owned jointly is the value of the full asset divided by the number of owners, except for some of the registered businesses in which the share of the business owned is reported. While one-fifth of the sample has debt, debt is minimal and the average net wealth is not statistically significantly different from average gross wealth. As such, gross wealth is used in this study. There is a substantial gender wealth gap. Nearly 42 percent of women and 56 percent of men in the sample hold positive wealth. Of the 42 percent and 56 of women and men respectively, women 2 Few individuals in Ghana hold stock, although, Ghana has had a stock exchange since 1990. Risky assets also include investments items such as art work and precious metals, but these assets are rare in Ghana and not found in the dataset. 3 In theory, portfolio investment decisions assume wealth fits closely with Arrow (1965) and Pratt (1964) definition, where assets within a portfolio are infinitely divisible—liquid or non-lumpy—and can be reallocated without cost from one asset to another. Financial assets come closest to this definition and are therefore most frequently used in theoretical studies. Empirical studies, however, often include a broader set of assets, even though they may not be infinitely divisible, under the implicit assumption that wealth as defined by Arrow and Pratt is highly correlated with other measures of wealth (Meyer & Meyer, 2006). 4 There are fewer enumeration areas in the Upper East Region due to conflict in parts of the region. 5 Of the 2,170 households surveyed, both spouses were interviewed in 956 households. For the other 1214 households, the second respondent may be a different family member (e.g. sibling, parent, parent-in-law) even if the first respondent is married and lives with his or her spouse. 6 For assets that are owned jointly, as long as the individual can make the decision to sell the asset alone or in consultation, the value of the proportion of the asset owned is included in the individual’s wealth.

131

M. Hillesland / World Development 117 (2019) 127–137 Table 2 Wealth and asset allocations by sex.

Proportion who hold positive gross value of assets (%) Of those who hold positive gross value of assets: Mean value of risky assets of (GH¢) Mean wealth (GH¢) Mean alpha Debt (formal business, real estate, or other) (GH¢) Proportion who have debt (%)

Women (n = 1690)

Men (n = 1341)

Total (n = 3031)

42

56

48

1204.9 (5838.0) 1426.8 (5970.0) 0.21 (0.39) 39.7 (238.0) 20.1

4417.6 (20638.2) 4881.7 (4881.7) 0.37 (0.46) 108.4 (948.6) 21.3

2866.4 (15464.9) 3213.6 (15666.7) 0.29 (0.43) 72.4 (658.0) 20.6

Note: Statistics are unweighted. Standard deviations are in parenthesis. The debt value is based on a slightly smaller sample due to missing observations.

Table 3 Asset allocation across the wealth distribution. Quintile

1 and 2 (n = 1575)

3 (n = 267)

4 (n = 602)

5 (n = 587)

Percent female (%) Mean value of risky assets (GH¢)

63 0 – 0 – 0 –

63 0.30 (2.76) 18.37 (8.60) 0.01 (0.12)

53 16.24 (55.04) 121.27 (80.65) 0.09 (0.27)

37 7093.10 (23745.34) 7838.20 (23948.23) 0.62 (0.44)

Mean gross wealth (GH¢) Mean alpha

Note: Statistics are unweighted. Standard deviations are in parenthesis.

hold GH¢ 1427 and men hold GH¢ 4882 average wealth.7 The mean measure for relative risk aversion (alpha), the ratio of risky assets to wealth, is 0.29 meaning that on average individuals hold about 29 percent of their wealth in risky assets. Although not entirely comparable given that the countries are very different, this is not dissimilar to the average values for the United States found in Friend and Blume (1975).8 Men on average have a greater alpha than women in the sample. Of those who hold wealth, women hold 21 percent of their wealth in risky assets and men hold 37 percent. A t-test suggests that we can reject the hypothesis that the means are equal at a 0.1 percent significance level. If men and women in Ghana exhibit constant relative risk aversion, this difference would suggest women are more risk averse than men on average with respect to asset allocation decisions. However, many empirical studies find evidence of decreasing relative risk aversion as wealth increases (see, for instance, Friend & Blume, 1975; Jianakoplos & Bernasek, 1998; Ogaki & Zhang, 2001). If we expect individuals in Ghana also to have decreasing relative risk aversion, it would mean that the proportion of risky assets to wealth would increase as wealth increases. Indeed, the mean alphas by wealth quintile in Table 3 and the asset allocation decisions by men and women across the wealth distribution in Fig. 1 provide evidence of decreasing relative risk aversion, suggesting that the difference between the value of men and women’s assets may explain the large difference in men and women’s proportion of risky asset holdings to wealth (i.e. the difference in men and women’s mean alphas).

7 This is a low wealth sample. The GLSS suggests income is also low in Ghana. The mean annual household income in Ghana in 2005 was GH¢1,217 (GLSS5 2008). The mean annual household income in Ghana in 2005 for the bottom quintile was GH¢728 or GH¢116 per capita. For the top quintile, the mean annual household income was GH¢1,544 or GH¢397 per capita (GLSS5 2008). 8 When the authors include additional assets, such as the estimated value for human capital and the market value of a family’s home in the measure of risky assets, the ratios are much higher (see Table 3 in Friend & Blume, 1975).

6. Empirical model Eq. (6) provides the theoretical basis for testing whether there are differences in men and women’s relative risk aversion after controlling for other factors:

c ak ¼ b0 þ b1 ln ðwealthk Þ þ b2 femalek þ b3 femalek  ln ðwealthk Þ þ local economykn um þhousehold v ariableskq cq

þindiv idual v ariableskm wn þ 

where c ak is the relative risk aversion measure. It is the proportion of risky assets to wealth held by the individual.  is the error term and b0 ; b1 ; b2 ; and b3 are parameters to be estimated as are the vector of m coefficients of variables that capture risk sharing and similar expectations about the market, um , the vector of q coefficients of household characteristics, cq , and the vector of n coefficients of variables that capture individual characteristics, wn . Femalek is a dummy variable, and the ln(wealthk ) variable is the natural log of gross wealth. Given that alpha increases as wealth increases (Table 3 and Fig. 1), and that theoretically at subsistence level (or zero wealth) relative risk aversion is infinite and decreases as wealth increases, it is expected that individuals in Ghana have decreasing relative risk aversion, such that b1 is positive. Table 4 presents the descriptive statistics for the variables that capture risk sharing and similar expectations about the market,um , the household characteristics, cq , and the individual characteristics, wn . To control for an absence of risk sharing and other formal and informal insurance mechanisms, whether the individual is part of a household that reduced consumption due to an idiosyncratic shock between 2005 and 2010 is included inum . Less than three percent of individuals are from households that reduced food or non-food consumption in order to cope with an idiosyncratic shock. This suggests that most individuals that faced idiosyncratic shocks in the sample are part of a risk sharing arrangement or have

132

M. Hillesland / World Development 117 (2019) 127–137

Notes: Average asset holdings. Estimates from the author. The darker assets are risky-assets; the lighter are nonrisky assets. Fig. 1. Asset allocation across the wealth distribution by sex.

access to other formal and informal mechanisms to insure themselves. Similar expectations about the market requires that individuals have access to information, which may be provided through faceto-face communication as well as through technology such as cell phones. Cell phones are included as a control. Differences in faceto-face communication differ by proximity to other households; to control of this, a variable for whether the household is located in a rural zone is also included as a control. Districts are included as a proxy for local networks. Household characteristics, cq , include the size of the household and household wealth, which is proxied by agricultural plots. While financial assets are not necessarily pooled in households in Ghana, we would expect asset allocation decisions to be influenced by household wealth. Agricultural plots are an imperfect measure of household wealth, as it does not capture the size and quantity of land and it overlooks many urban households. However, the majority of households in Ghana in 2010 were agricultural households.9 In addition, land is a major asset that makes up a large share of households’ wealth. Individual characteristics, wn , include age, marital status, and whether the individual has children under five, and children ages six to 11. To capture the individual’s position in the household, an individual’s decision-making power is proxied by whether the individual reported he or she could make the decision alone to engage in employment or income earning activities, and whether he or she can decide how the earnings are used. Education is a non-marketable (i.e. cannot be directly traded) asset and is either controlled for or included in the value of risky assets in other studies (see, for example, Friend & Blume, 1975). It is included as a control in this study. Agriculture is an important livelihood for many rural households in Ghana and influences how productive assets are allocated and, as such, is included as a control. Since expectations around inheritance may influence how an individual allocates his or her assets, whether an individual 9 Nearly five million households of 6.6 million were estimated to engage in cropping activities in the 12 month period prior to the survey (based on the GLSS6 which was implemented from October 2012 to October 2013).

expects future inheritance is also included as a control in wn . Similarly, whether the individual holds a pension is included as a control. Debt is trivial, but it is also included as a control.

7. Results There are a number of individuals who do not hold wealth in the form of formal and informal financial assets, registered businesses that can be sold, and commercial real estate the individual owns and holds the right to sell. Alpha is observed only when an individual holds positive wealth (ak is undefined for an individual without wealth). This means the relative risk aversion measure, the proportion of risky assets to wealth, ak , is incidentally truncated and may result in a specification error if not addressed. Studies such as Jianakoplos and Bernasek (1998) do not include households below a certain wealth threshold and, thus, do not need to address truncation.10 This study does not use a threshold; instead, to address truncation, a two-step Heckman selection model is used. The first stage of the Heckman selection model estimates the likelihood an individual has positive wealth. The second stage estimates the relative risk aversion measure, ak , while incorporating information from the first. The estimated coefficients of wealth, b1 , are positive and statistically significant, suggesting as wealth increases, ak , the relative risk aversion measure, increases (Table 5). It provides evidence that individuals in this sample exhibit decreasing relative risk aversion in terms of asset allocation decisions, ceteris paribus. Ideally, instrumental variables would be used to address potential biases of simultaneity of wealth and the risk aversion measure; appropriate instruments were unavailable however. As such, the estimates of wealth should be interpreted with some caution. Even so, the results are consistent with the theoretical expectations and previous studies (e.g. Friend & Blume, 1975; Jianakoplos & Bernasek, 1998). Fig. 2 presents the mean predicted ratios across wealth. 10 Studies in the United States, such as Jianakoplos and Bernesek (1998), often only include households with wealth greater than US$1000 in their analyses.

133

M. Hillesland / World Development 117 (2019) 127–137 Table 4 Descriptive Statistics. Women (n = 1690)

Men (n = 1341)

Total (n = 3031)

43.9 (16.2)

45.3 (16.7)

44.5 (16.5)

Number of children 0 to 5 years old (%) None 1 2 3 or more

62.0 22.5 12.3 3.32

62.8 20.6 12.8 3.9

62.3 21.6 12.5 3.6

Number of children 6 to 11 years old (%) None 1 2 3 or more

62.3 20.8 12.5 3.9

63.5 20.8 12.0 3.65

63.1 20.8 12.3 3.8

Marital Status (%) In a monogamous union In a polygamous union In a consensual union Never married Divorced, widowed, or deserted

45.6 6.7 10.4 6.6 30.7

54.7 8.0 13.4 13.4 10.5

50.6 7.7 12.7 9.4 19.5

60.4 30.4 7.0 2.3 37.0

40.4 40.0 13.8 5.7 40.1

51.5 34.7 10.0 3.8 38.7

8.0

8.1

8.0

Women (n = 1690)

Men (n = 1341)

Total (n = 3031)

5.1 11.8 8.9 33.5 39.6 3.1 65.1 4.2 (2.6)

13.6 29.8 27.2 44.1 61.7 2.2 66.5 4.1 (2.8)

8.9 19.8 17.0 38.2 49.4 2.7 65.8 4.2 (2.7)

70.9 16.8 7.2 2.9 1.4 0.5 0.3

69.0 16.9 8.0 3.7 1.3 0.8 0.4

70.0 16.9 7.6 3.3 1.4 0.6 0.3

a Mean age

Education (%) No education or attended some primary school only Attended some junior secondary school or equivalent Attended at least some senior secondary school or vocational or technical training Attended at least some university, professional training, or other post senior secondary education Made decision whether to be employed or pursue an income-generating activity alone and can make the decision how to spend the earnings (%) Expects an inheritance (%)

b Has a pension or other type of retirement (%) Has rights over agricultural land (%) Owns place of residence and it can be sold (%) Occupation is agriculture, animal husbandry, forestry work, fishing, or hunting (%) Owns a mobile phone (%) Household reduced consumption due to an idiosyncratic shock anytime between 2005 and 2010 (%) Household is in a rural setting Mean household size Average number of agricultural plots the household holds (owned land or family land) None 1 2 3 4 5 6 Notes: Statistics are unweighted. Standard deviations are in parenthesis.

Controlling for men and women’s differing characteristics, men and women do not have statistically different levels of relative risk aversion. b2 and b3 are not individually statistically significant and a Wald-test suggests there is not joint statistical significance for the coefficients (chi-squared = 0.07).11,12 Women are not more relatively risk averse than men; rather it is men and women’s different characteristics that contribute to the considerable difference in average ratios of risky assets to wealth, ak : Since there is evidence of decreasing relative risk aversion, it is possible that the difference in men and women’s average proportion of risky assets to wealth,

11

The difference is also not significant across the wealth distribution. While the dependent variable is shares between 0 and 1, theoretically, the shares could be above and below 1 and 0. As such, a Heckman two step model is used as it fits the data best. However, similar results were found using a fractional probit model which can be used when the dependent variable is a fraction between zero and one. These are available by request. 12

ak , is due primarily to the difference in average wealth. A OaxacaBlinder decomposition technique is used to test this. This technique is typically used in the labor market literature, but it can be used to study differences in any outcome variable. Eq. (7) is estimated separately for men and women with a twostep Heckman selection model for each equation.13 Using the results from these two equations and the pooled equation from above, the difference in men and women’s average ratios of risky  

assets to wealth, E am xm  E a f x f , is divided into explained k

k

k

k

and unexplained components. The explained component is the sum of the product of the estimated coefficients for the pooled equation except for the coefficient for the dummy sex variable and difference of men and women’s expected values. The unexplained 13 I use a pooled regression with group indicator so that unexplained factors due to sex are not transferred to the coefficients in the explained components.

134

M. Hillesland / World Development 117 (2019) 127–137

Table 5 Risk preference model: select variables. Dependent variable: Alpha

Heckman Two-Step Model

Natural log of financial wealth

0.1494** (0.0047) 0.0123 (0.0439) 0.0064 (0.0072) 0.0060** (0.0032) 0.0001 (0.0000) 0.0341 (0.0209) 0.8659 (0.1471)

Female Interaction of Natural log of financial wealth and Female Age Age squared Married or in a consensual union Constant Lambda

0. 1203 (0.0915)

Observations Censored observations Chi-squared

3031 1575 2252.64

coefficients from the coefficients from the pooled equation plus the product of the expected values of women and the difference between the female coefficients from the pooled equation coefficients:

 0  
 m f f f pooled pooled m E am jx a jx  x b  b ¼ ⏟ x  E k k sex k k

þ⏟

Explained component m0 bpooled þx sex



 f 0  bm  bpooled þ x bpooled  b f

Unexplained component

ð8Þ

Notes: Standard errors are in parentheses. *p < .10, **p < .05. Marital status includes all married individual and those in consensual unions. Additional controls are number of children of the individual ages 0–5, number of children of the individual ages 6–11, whether the individual attended at least some senior secondary school or vocational or technical training, whether the individual attended at least some university, professional training, or other post senior secondary education, decision-making power of employment and income, whether the individual holds debt, whether the individual owns agricultural land he or she can sell, whether the individual owns place of residence he or she has the right to sell, whether the individual’s occupation is in agriculture, whether the individual expects an inheritance, whether the individual has a pension or other retirement source, household size, household is in rural setting, number of plots of land the household owns as a proxy for household wealth, whether the household reduced consumption due to an idiosyncratic shock anytime between 2005 and 2010, and district fix effects. The selection equation predicts the probability the individual holds wealth. Variables that would explain whether or not an individual has positive wealth and does not impact allocation of wealth are not available. It is assumed the nonlinearity of the selection equation creates enough exclusion restrictions. The full model is in the Appendix.

where bpooled , bm , and bf are vectors containing the intercepts and slope parameters (um , wn ; and cq ) for the pooled equation, the estimated equation consisting of only men, and the estimated equation consisting of only women. Table 6 presents the results. The explained part of the outcome differential is significant and accounts for nearly the total difference between men and women’s mean proportion of risky assets to wealth. The unexplained part, which is the effect of the unobserved predictors, is also statistically significant but significantly smaller. The greatest contribution to the difference in men and women’s average proportion of risky assets to wealth in both models is the difference in average wealth. The contribution of other predicators are minimal. The results suggest that, in Ghana, women are not more risk averse than men in terms of asset allocation decisions, but because men and women exhibit decreasing relative risk aversion, women’s lower average wealth

Table 6 Oaxaca-Blinder decomposition of men and women’s proportion of risky assets to gross wealth (alpha). Oaxaca-Blinder decomposition

.8

Differential Male Female Difference Decomposition Explained Unexplained Individual contributions to the explained component of the decomposition of some of the variables Natural log of wealth

0

Married or in a consensual union 0

2000

4000

6000

8000

10000

Age

Wealth (Ghanaian cedis)

Notes: Predicted values are based on average characteriscs. Fig. 2. Predicted Measure of Relative Risk Aversion (Risky Assets to Wealth).

component includes the coefficient on the sex dummy variable (indicator of group membership) in the pooled equation plus the product of the expected values of men and the difference between the male

Individual owns agricultural land (where there is a market and individual and has the right to sell) Individual owns the place of residence (where there is a market and individual and has the right to sell) Occupation is agriculture, animal husbandry, forestry work, fishing, or hunting

0. 3673 (0.0168) 0.2085 (0.0148) 0.1588** (0. 0223) 0.1208** (0.0163) 0.0380* (0.0176)

0. 1375** (0.0164) 0.0043* (0.0022) 0.0037* (0.0020) 0.0093* (0.0042) 0.0158** (0.0048) 0.0116** (0. 0115)

Notes: Standard errors are in parentheses. *p < .10, **p < .05. The estimates are based on using a pooled Oaxaca-Blinder approach and two-step Heckman models for the pooled, female, and male equations. The estimations for the individual contributions to the explained component of the decomposition of all predictors are available by request.

135

M. Hillesland / World Development 117 (2019) 127–137

than men’s means women invest less in risky assets proportional to their overall wealth. 8. Discussion and conclusion Without taking into account wealth, there is a statistically significant difference in how men and women allocate their wealth between risky and non-risky assets in households in Ghana, suggesting gender differences in relative risk aversion. However, because men and women exhibit decreasing relative risk aversion, men and women do not have systematically different risk preferences when wealth differences and differences in other characteristics are taken into account. The results of this paper differ from the results in many empirical studies in the United States, where women are found to be more risk averse than men. As one of the first studies to look at gender differences in risk aversion in terms of asset allocation decisions within a developing country, the results provide support for the idea that gender differences in risk preferences may vary by the surrounding environment—including by country—as some of the experimental literature suggests. If risk preferences vary by county, differences found in men and women’s risk preferences are less likely to reflect innate sex differences in risk preferences, but rather differences in constraints, opportunities, or responsibilities men and women face. That is, any systematic difference in risk attitudes between men and women found in studies is likely not based on sex, but rather influenced by gender norms and roles, which vary by country. These norms and roles play out differently in different contexts due to different cultural and environmental pressures. So while women may be more risk averse in terms of asset allocation decisions than men in the United States, this may not necessarily be the case for men and women in Ghana. Funding This work was supported the American University’s Vice Provost for Graduate Studies Doctoral Student Research Award 2011-2012 and the Association for Social Economics William R. Waters Research Grant 2011. I also received support from The Gender Asset Gap Project. Conflicts of interest The author confirms there are no known conflicts of interest associated with this publication and there has been no significant financial support for this work that could have influenced its outcome. Acknowledgements I want to thank Caren Grown, Cheryl Doss, Amos Golan, and Maria Sagrario Floro for their many reviews and helpful suggestions. I would like to extend my gratitude to the Gender Asset Gap team and the Department of Economics at University of Ghana, Legon; I especially want to thank Abena Oduro, Louis Boakye-Yiadom, and Hellen Enyonam at the University of Ghana, Legon, for hosting me in the summer of 2011. I also thank the reviewers for their thoughtful and thorough reviews. Appendix Table 7

Table 7 Risk preferences: Full model. Dependent variable: Alpha

Heckman TwoStep Model

Female

0.0113 (0.0439) 0.1494** (0.0047) 0.0064 (0.0072) 0.0060* (0.0032) 0.0000 (0.0000) 0.0341 (0.0209) 0.0171 (0.0244) 0.0315 (0.0383) 0.0079 (0.0228)

Natural log of mean wealth Female * Natural log of mean wealth Age Age squared Married or in a consensual union Attended at least some senior secondary school or vocational or technical training Attended at least some university, professional training, or other post senior secondary education Made decision whether to be employed or pursue an income-generating activity alone and can make the decision how to spend the earnings Has rights over agricultural land Owns place of residence and it can be sold Expects an inheritance Pension or other type of retirement Holds debt Number of children under five Number of children 6 to 11 years old Occupation is agriculture, animal husbandry, forestry work, fishing, or hunting

0.0654** (0.0261) 0.0740** (0.0240) 0.0523* (0.0268) 0.0213 (0.0539) 0.0803** (0.0194) 0.0230* (0.0118) 0.0233** (0.0116) 0.0687** (0.0272)

Average number of agricultural plots the household holds (owned land or family land) 1 0.0570** (0.0262) 2 0.0666* (0.0351) 3 0.0470 (0.0488) 4 plots or more 0.0286 (0.0670) Number of household members 0.0010 (0.0046) Household reduced consumption due to an idiosyncratic 0.0328 shock anytime between 2005 – 2010 (0.0494) Owns a mobile phone 0.0413 (0.0322) Rural 0.0475 (0.0394) Constant 0.8659** (0.1471) Dependent variable: Has positive wealth

Probit model

Age

0.0233** (0.0091) 0.0003** (0.0001) 0.2045** (0.0611) 0.0440 (0.0939) 0.6792** (0.2015) 0.2987** (0.0602)

Age squared Married or in a consensual union Attended at least some senior secondary school or vocational or technical training Attended at least some university, professional training, or other post senior secondary education Made decision whether to be employed or pursue an income-generating activity alone and can make the decision how to spend the earnings Has rights over agricultural land

0.0884 (0.0830) (continued on next page)

136

M. Hillesland / World Development 117 (2019) 127–137

Table 7 (continued) Dependent variable: Alpha

Heckman TwoStep Model

Owns place of residence and it can be sold

0.1530** (0.0760) 0.0688 (0.0960) 1.3956** (0.1387) 0.1553** (0.0644) 0.0390 (0.0386) 0.0155 (0.0384) 0.2273** (0.0766)

Expects an inheritance Pension or other type of retirement Holds debt Number of children under five Number of children of 6 to11 years old Occupation is agriculture, animal husbandry, forestry work, fishing, or hunting

Average number of agricultural plots the household holds (owned land or family land) 1 0.1046 (0.0811) 2 0.0541 (0.1105) 3 0.3304** (0.1553) 4 plots or more 0.0631 (0.1869) Number of household members 0.0205 (0.0149) Household reduced consumption due to an idiosyncratic 0.0745 shock anytime between 2005 and 2010 (0.1590) Owns a mobile phone 0.4902** (0.0581) Rural 0.1738 (0.1250) Female 0.1888** (0.0572) Constant 0.4725 (0.3584) Lambda

0.1203 (0.0916)

Observations Censored observations Chi-squared

3031 1,575 2252.64

Notes: Standard errors are in parentheses. *p < .10, **p < .05. The model also includes district fix effects. The selection equation predicts the probability the individual holds wealth. Variables that would explain whether or not an individual has positive wealth and does not impact allocation of wealth are not available. While not ideal, it is assumed the nonlinearity of the selection equation creates enough exclusion restrictions.

References Ambrus, A., Mobius, M., & Szeidl, A. (2014). Consumption risk-sharing in social networks. The American Economic Review, 104(1), 149–182. Antonopoulos, R., & Floro, M. S. (2005). Asset Ownership along gender lines: Evidence from Thailand Working Paper No. 418. NY: Levy Economics Institute. Arano, K., Parker, C., & Terry, R. (2010). Gender-based risk aversion and retirement asset allocation. Economic Inquiry, 48(1), 147–155. Arrow, K. J., (1965). Aspects of a theory of risk bearing, Yrjo Jahnsson Lectures, Helsinki. Reprinted in Essays in the theory of risk bearing (1971). Bajtelsmit, V. L., & VanDerhei, J. L. (1997). Risk aversion and pension investment choices. In M. S. Gordon, O. S. Mitchell, & M. M. Twinney (Eds.), Positioning pensions for the twenty-first century. University of Pennsylvania Press. Bernasek, A., & Shwiff, S. (2001). Gender, risk, and retirement. Journal of Economic Issues, 35(2), 345–356. Booth, A., Cardona-Sosa, L., & Nolen, P. (2014). Gender differences in risk aversion: Do single-sex environments affect their development? Journal of Economic Behavior & Organization, 99, 126–154. Booth, A. L., & Nolen, P. (2012). Gender differences in risk behaviour: Does nurture matter? The Economic Journal, 122(558), F56–F78. Budd, A., & Litzenberger, R. H. (1972). The market price of risk, size of market and investors’ risk aversion: A comment. The Review of Economics and Statistics, 54 (2), 204–206.

Cárdenas, J. C., Dreber, A., Von Essen, E., & Ranehill, E. (2012). Gender differences in competitiveness and risk taking: Comparing children in Colombia and Sweden. Journal of Economic Behavior & Organization, 83(1), 11–23. Charness, G., & Gneezy, U. (2012). Strong evidence for gender differences in risk taking. Journal of Economic Behavior & Organization, 83(1), 50–58. Chiappori, P. A., Samphantharak, K., Schulhofer-Wohl, S., & Townsend, R. M. (2014). Heterogeneity and risk sharing in village economies. Quantitative Economics, 5 (1), 1–27. Croson, R., & Gneezy, U. (2009). Gender differences in preferences. Journal of Economic Literature, 47(2), 448–474. Deere, C. D., & Doss, C. (2006). The gender asset gap: What do we know and why does it matter? Feminist Economics, 12(1–2), 1–50. Deere, C. D., Oduro, A. D., Swaminathan, H., & Doss, C. (2013). Property rights and the gender distribution of wealth in Ecuador, Ghana and India. The Journal of Economic Inequality, 11(2), 249–265. Dercon, S. (2005). Risk, insurance, and poverty: A review. In Insurance against poverty (pp. 9–37). Oxford; New York: Oxford University Press. Dercon, S., & Christiaensen, L. (2011). Consumption risk, technology adoption and poverty traps: Evidence from Ethiopia. Journal of Development Economics, 96(2), 159–173. Dercon, S., & Krishnan, P. (2000). In sickness and in health: Risk sharing within households in rural Ethiopia. Journal of Political Economy, 108(4), 688–727. De Weerdt, J., & Dercon, S. (2006). Risk-sharing networks and insurance against illness. Journal of Development Economics, 81(2), 337–356. Doss, C. (2001). Is risk fully pooled within the household? Evidence from Ghana. Economic Development and Cultural Change, 50(1), 101–130. Doss, C., McPeak, J., & Barrett, C. (2008). Interpersonal, intertemporal and spatial variation in risk perceptions: Evidence from East Africa. World Development, 36 (8), 1453–1468. Eckel, C. C., & Grossman, P. J. (2008). Men, women and risk aversion: Experimental evidence. In C. Plott & V. Smith (Eds.), Handbook of experimental economics results, 1 (pp. 1061–1073). Oxford: North-Holland. Fafchamps, M., & Gubert, F. (2007). The formation of risk sharing networks. Journal of Development Economics, 83(2), 326–350. Fafchamps, M., & Lund, S. (2003). Risk-sharing networks in rural Philippines. Journal of Development Economics, 71(2), 261–287. Fafchamps, M., Udry, C., & Czukas, K. (1998). Drought and saving in West Africa: Are livestock a buffer stock? Journal of Development Economics, 55(2), 273–305. Fletschner, D., Anderson, C. L., & Cullen, A. (2010). Are women as likely to take risks and compete? Behavioural findings from Central Vietnam. The Journal of Development Studies, 46(8), 1459–1479. Friend, I., & Blume, M. E. (1975). The demand for risky assets. American Economic Review, 65(5), 900–922. Goldstein, M. (1999). Chop time, no friends: Intrahousehold and individual insurance mechanisms in Southern Ghana. University of California, Berkeley, mimeo. Goldstein, M., de Janvry, A., & Sadoulet, E. (2004). Is a friend in need a friend indeed? Inclusion and exclusion in mutual insurance networks in Southern Ghana. In S. Dercon (Ed.), Insurance against poverty (pp. 9–37). Oxford; New York: Oxford University Press. https://doi.org/10.1093/0199276838.003.0012. Gneezy, U., Leonard, K. L., & List, J. A. (2009). Gender differences in competition: Evidence from a matrilineal and a patriarchal society. Econometrica, 77(5), 1637–1664. Grimard, Franque (1997). Household consumption smoothing through ethnic ties: Evidence f from Cote d’Ivoire. Journal of Development Economics, 53(2), 391–422. Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., Gintis, H., & McElreath, R. (2001). In search of homo economicus: Behavioral experiments in 15 smallscale societies. American Economic Review, 91(2), 73–78. Jianakoplos, N. A., & Bernasek, A. (1998). Are women more risk averse? Economic Inquiry, 36, 620–630. Khachatryan, K., Dreber, A., Von Essen, E., & Ranehill, E. (2015). Gender and preferences at a young age: Evidence from Armenia. Journal of Economic Behavior & Organization, 118, 318–332. Lee, J., & Sawada, Y. (2010). Precautionary saving under liquidity constraints: Evidence from rural Pakistan. Journal of Development Economics, 91(1), 77–86. Ligon, E., Thomas, J. P., & Worrall, T. (2002). Informal insurance arrangements with limited commitment: Theory and evidence from village economies. The Review of Economic Studies, 69(1), 209–244. Lintner, J. (1970). The market price of risk, size of market and investor’s risk aversion. The Review of Economics and Statistics, 52(1), 87–99. Mazzocco, M., & Saini, S. (2012). Testing efficient risk sharing with heterogeneous risk preferences. American Economic Review, 102(1), 428–468. Meyer, D. J., & Meyer, J. (2006). Measuring risk aversion. Foundation and Trends in Microeconomics, 2(2), 107–203. Morduch, J. (2005). Consumption smoothing across space: Testing theories of risksharing in the ICRISAT study region of South India. In S. Dercon (Ed.), Insurance against poverty. Oxford: Oxford University Press. Mossin, J. (1966). Equilibrium in a capital asset market. Econometrica: Journal of the Econometric Society, 768–783. Murgai, R., Winters, P., Sadoulet, E., & De Janvry, A. (2002). Localized and incomplete mutual insurance. Journal of Development Economics, 67(2), 245–274. Nelson, Julie A. (2015). Are women really more risk-averse than men? A re-analysis of the literature using expanded methods. Journal of Economic Surveys, 29(3), 566–585. Nelson, J. (2016). Not-so-strong evidence for gender differences in risk taking. Feminist Economics, 22(2), 114–142. https://doi.org/10.1080/13545701.2015.1057609.

M. Hillesland / World Development 117 (2019) 127–137 Oduro, A., & Doss, C. (2018). Changing patterns of wealth distribution: Evidence from Ghana. The Journal of Development Studies, 54(5), 933–948. https://doi.org/ 10.1080/00220388.2018.1430769. Ogaki, M., & Zhang, Q. (2001). Decreasing relative risk aversion and tests of risk sharing. Econometrica, 69(2), 515–526. Park, Albert (2006). Risk and household grain management in developing countries. The Economic Journal, 116(514), 1088–1115. Pratt, J. W. (1964). Risk aversion in the small and in the large. Econometrica: Journal of the Econometric Society, 122–136. Quisumbing, A. (2010). Do men and women accumulate assets in different ways. Evidence from rural Bangladesh. Background paper prepared for the Food and Agricultural Organization of the United Nations (FAO) State of Food and Agriculture. Riley, W. B., Jr, & Chow, K. V. (1992). Asset allocation and individual risk aversion. Financial Analysts Journal, 32–37. Sawada, Y., Nakata, H., & Kotera, T. (2017). Self-production, friction, and risk sharing against disasters: Evidence from a developing country. World Development, 94, 27–37.

137

Schubert, R., Brown, M., Gysler, M., & Brachinger, H. W. (1999). Financial decisionmaking: Are women really more risk-averse? The American Economic Review, 89 (2), 381–385. Sundén, A., & Surette, B. (1998). Gender differences in the allocation of assets in retirement savings plans. American Economic Review Papers and Proceedings, 88, 207–211. Townsend, R. M. (1994). Risk and insurance in village India. Econometrica: Journal of the Econometric Society, 539–591. Townsend, R. M. (1995). Consumption insurance: An evaluation of risk-bearing systems in low-income economies. The Journal of Economic Perspectives, 9(3), 83–102. Vanderpuye-Orgle, J., & Barrett, C. B. (2009). Risk management and social visibility in Ghana. African Development Review, 21(1), 5–35. Zimmerman, F. J., & Carter, M. R. (2003). Asset smoothing, consumption smoothing and the reproduction of inequality under risk and subsistence constraints. Journal of Development Economics, 71(2), 233–260.