American Economic Review: Papers & Proceedings 2013, 103(3): 362–368 92931 http://dx.doi.org/10.1257/aer.103.3.362 Behavioral Biases and Firm Behavior: Evidence from Kenyan Retail Shops † By Michael Kremer, Jean Lee, Jonathan Robinson, and Olga Rostapshova* Many subjects in lab experiments show con- ­usiness owners in developing countries often b siderable risk aversion in small-stakes gambles.1 leave profitable investments unexploited.2 This This is counter to the predictions of expected seems to be true not only for large, lumpy invest- utility theory for any reasonable degree of risk ments, such as investments in machinery that aversion (Rabin 2000) but is consistent with loss might expose firm owners to substantial risk, aversion in prospect theory. Benjamin, Brown, but also for divisible investments where standard and Shapiro (forthcoming) show that math skills dynamic optimization theory would not predict reduce small-stakes risk aversion, consistent with poverty traps, and expected utility theory would broader evidence that mathematical skills can suggest that risk should play a relatively small help debias decision making (Burks et al. 2007). role in investment decisions. For example, many In this paper, we show that acceptance of farmers fail to invest in any fertilizer despite small risky gambles and scores on math tests is apparently large returns and despite the availabil- associated with inventory accumulation among ity of fertilizer in small quantities (Duflo, Kremer, Kenyan shopkeepers. More broadly, we argue and Robinson 2011).3 that loss aversion may be one factor helping In Kremer et al. (2013) we show that many explain the broader puzzle of why high rates of Kenyan shopkeepers fail to make small inven- return on capital among small firms in develop- tory investments with high expected returns. In ing countries are not arbitraged away and do not this paper, we examine the determinants of inven- lead to the high growth rates of consumption tory investments and show that shopkeepers who that the Euler equation would predict. invest one standard deviation more into a risky The development literature has documented asset in a laboratory-style game have 10–16 per- that, across a wide range of contexts, small cent larger inventories. Consistent with the view that math skills may be useful in debiasing, those with one–standard deviation higher math scores * Kremer: Harvard University, 1050 Massachusetts have 14  –18 percent larger inventory levels. Ave., Cambridge, MA 02139 and NBER, Littauer Center, Section I provides background, while Section II Cambridge, MA 02138 (e-mail: mkremer@fas.harvard. edu); Lee: The World Bank, 1818 H St. NW, Washington, reports on credit constraints and departures from DC 20433 (e-mail: jlee20@worldbank.org); Robinson: standard models of entrepreneurial decision mak- University of California, Santa Cruz, 457 Engineering ing as determinants of inventory investment. 2, Santa Cruz, CA 95064 (e-mail: jmrtwo@ucsc.edu); Section III discusses broader implications, argu- Rostapshova: Harvard Kennedy School, 1050 Massachusetts ing that loss aversion may be an important part Ave., Cambridge, MA 02139 (e-mail: olga_rostapshova@ hksphd.harvard.edu). We thank Isaac Mbiti and David of the puzzle of why many small businesses in Laibson for helpful comments. Abdulla Al-Sabah, Kenzo developing countries do not take advantage of Asahi, Pia Basurto, Dan Bjorkegren, Conner Brannen, high expected return investment opportunities. Elliott Collins, Sefira Fialkoff, Katie Hubner, Eva Kaplan, In particular, we argue that loss aversion may Anthony Keats, Jamie McCasland, and Russell Weinstein provided excellent research assistance. We gratefully acknowledge funding from the SEVEN Foundation, the 2 Kauffman Foundation, and the NBER Africa program. We See, for example, de Mel, McKenzie, and Woodruff thank IPA-Kenya for administrative support. (2008), McKenzie and Woodruff (2008), Fafchamps et al. † To view additional materials, and author disclosure (2011), Udry and Anagol (2006), and Banerjee and Duflo statement(s),visit the article page at (2012). 3 http://dx.doi.org/10.1257/aer.103.3.362. Of course, in this example, returns to fertilizer are posi- 1 See, for example, Tversky and Kahneman (1991), Kahne- tively correlated with returns on overall agricultural output, man, Knetsch, and Thaler (1990), and Thaler et al. (1997). so this could be considered a relatively high beta investment. 362 VOL. 103 NO. 3 behavioral biases and firm behavior: evidence from Kenyan Retail shops 363 help explain: (i) why capital injections to small bulk purchase discounts for purchases above businesses yield high estimated returns, and why a set of thresholds. While many purchases are they are neither spent down over time nor gen- just above the thresholds, there is a consider- erate further capital accumulation (as some pov- able mass below each threshold as well, and we erty trap stories would suggest); (ii) why many find that the median shop misses at least some firm owners do not take advantage of microcredit opportunities to earn rates of return in excess of despite high unrealized returns to investment; 100 percent (e.g., by increasing one purchase (iii) why many of those who do borrow do not slightly to meet the bulk discount threshold and invest in their businesses; (iv) why weather insur- correspondingly reducing the next purchase). ance programs stimulate investment by farmers; Returns are highly heterogeneous across firms. and (v) why business training based on simple That paper also calculates bounds on rates of heuristics can induce firm owners to change their return to holding marginally greater inventories behavior. of phone cards based on a survey of lost sales due to stockouts. This method likely yields I.  Background and Data much looser bounds since it does not factor in the long-run costs of losing customers due to We study small-scale owner-operated rural stockouts, but nonetheless it still suggests that Kenyan retail shops selling fast moving con- more than 18 percent of firms still have rates of sumer goods (FMCG). The FMCG manufac- return over 50 percent per year to an additional turing industry is highly concentrated, with unit of phone card inventory, based only on the manufacturers setting retail prices and a single short-run sales lost to stockouts. The correlation supplier holding a very high market share in of bounds on returns across products is low, sug- most goods. These shops are typically located in gesting that shopkeepers may not be equating clusters in market centers serving the surround- the marginal return to inventories across items. ing rural population, with several competing This paper examines the determinants of shops in proximity. inventory investment based on a survey of Three features of the industry and the envi- shop owners served by the distributor within a ronment restrict the ability of firm owners who particular geographic area in Western Kenya. maximize expected profits to displace com- Background surveys included standard demo- petitors whose decision making departs from graphic questions, as well as questions on the expected profit maximization. First, rural shops shop owner’s access to savings and credit; own- face competition from only a limited number ership of land, durable goods, and other assets; of competitors, since customers will walk only transfers given and received, other sources of a limited distance. Second, manufacturers fix income, and self-reported credit constraints. The retail prices, precluding price competition that survey included vocabulary and reading tests in could allow well-stocked, high-traffic shops to English and Swahili, a math problem solving lower prices, driving out competitors. Third, test, a digit recall memory test, Raven progres- whether due to labor market imperfections, sive matrices, and a maze completion speed test. moral hazard with employees, regulatory issues, Respondents were also asked to divide a portfo- or other factors, owners who manage inventory lio of 100 Ksh (approximately $1.33) between optimally are not able to manage multiple shops a safe asset and a risky asset that paid zero with in different locations. A potential policy impli- 50 percent probability and 2.5 times the amount cation of these factors is that prohibiting retail invested with 50 percent probability. price maintenance could allow efficient retailers We collected inventory data for a subset of to displace less efficient ones. With a smaller 380 of these shops, approximately 1.5–2 years number of larger retailers, it might also be eas- after the background surveys. We asked respon- ier for new manufacturers to enter the market dents to estimate the total value of their inven- because building a distribution network would tory (at both wholesale and retail prices) with be a less daunting barrier to entry. the enumerator’s assistance. In addition, the In a companion paper (Kremer et al. 2013) respondent, together with the enumerator, cal- we use administrative data from a distribu- culated the value of the 13 most common items tor to calculate bounds on returns to inventory. stocked by shops. We follow de Mel, McKenzie, The distributor offers progressively greater and Woodruff (2008) in measuring profits by 364 AEA PAPERS AND PROCEEDINGS MAY 2013 Table 1—Summary Statistics Quantiles Mean SD 25th 50th 75th 90th (1) (2) (3) (4) (5) (6) Inventories, profits, and credit to customers Total inventory 26.52 29.67 8.00 15.00 30.00 70.00 Inventory in top 13 items 9.42 11.34 2.80 5.16 10.81 24.93 Profits in past month 2.39 2.70 0.80 1.20 3.00 5.60 Gives out credit to customers 0.9 — — — — — Amount given out in credit in past month 1.12 5.75 0.10 0.20 0.50 1.50 Background characteristics Years of education 10.80 3.33 8.00 11.50 12.00 16.00 Years shop open 7.47 5.60 3.16 6.41 9.90 14.56 Male 0.56 — — — — — Married 0.79 — — — — — Age 33.36 9.48 27.00 32.00 38.00 46.00 Can read and write (Swahili) 0.97 Small stakes risk aversion Percentage invested in risky asset 0.56 0.20 0.50 0.50 0.70 0.80 Asset ownership and formal sector income Owner or spouse has a formal sector job 0.13 — — — — — Acres land owned 1.95 2.64 0.00 1.00 2.50 4.50 Value of durable goods and animals owned 11.70 15.32 4.20 7.00 11.91 25.50 Financial access Has bank account 0.83 — — — — — Participates in ROSCA 0.42 — — — — — Would you like to borrow more money but 0.37 — — — — —   are unable to get it (percentage “yes”) Notes: All monetary values in 10,000 Kenyan shillings. Exchange rate was roughly 75 Ksh to $1. There are 380 shops in the sample. asking respondents to report their income less incomes and wealth in an area where typical expenses and wages to other employees over the agricultural wages are approximately $1 a day. previous 30 days.4 This question was included On average, shopkeepers invested just over only in the latter part of the data collection, so half of the portfolio in the risky asset. One- we have profit data for 188 firms. third invested exactly one-half of the assets in Summary statistics are shown in Table 1. the portfolio, consistent with the behavior of Shopkeepers are substantially more educated many US workers who have retirement plans than the typical rural resident in the area. About that allow them to divide their assets between 13 percent of owners or their spouses have multiple investments and follow the “1/n rule” formal sector jobs. In addition, 82 percent of heuristic of dividing assets evenly across all shopkeepers in our sample have bank accounts, options (Benartzi and Thaler 2001, 2007). The 73 percent own land, and 42 percent participate mass at the 50-50 division suggests that not all in a merry-go-round cooperative (ROSCA). subjects are expected utility maximizers. Inventory value and income distributions among shopkeepers are skewed, but even at the II.  Credit Constraints, Entrepreneurial Decision twenty-fifth percentile shopkeepers have high Making, and Inventory Investment Table 2 reports regressions of inventories on 4 This measure of profits thus includes returns to owner indicators of credit constraints as well as fac- labor and family labor. tors that could cause entrepreneurial ­ decision VOL. 103 NO. 3 behavioral biases and firm behavior: evidence from Kenyan Retail shops 365 Table 2—Correlates of Inventories and Profits log total log inventory on inventory top 13 products log profits in past month (1) (2) (3) (4) (5) (6) (7) (8) Background characteristics Years of education −0.06 −0.09 0.04 −0.01 0.17 0.12 0.29 0.23  (tens of years) (0.20) (0.20) (0.21) (0.21) (0.27) (0.28) (0.21) (0.22) Years shop open 0.16 0.12 0.20 0.17 0.06 0.08 −0.02 −0.05  (tens of years) (0.10) (0.10) (0.11)* (0.11) (0.14) (0.14) (0.11) (0.12) Age −0.01 −0.02 −0.02 −0.02 −0.02 −0.02 −0.01 −0.01 (0.01)** (0.01)*** (0.01)*** (0.01)*** (0.01)** (0.01)* (0.01) (0.01) Cognitive measures Math score 0.18 0.18 0.14 0.14 0.32 0.32 0.15 0.22  (standardized) (0.06)*** (0.06)*** (0.06)** (0.06)** (0.09)*** (0.09)*** (0.07)** (0.07)*** Raven’s matrix 0.07 0.05 0.15 0.13 −0.13 −0.13 −0.07 −0.16   (standardized) (0.07) (0.06) (0.07)** (0.07)** (0.09) (0.09) (0.07) (0.07)** Digit recall −0.02 −0.03 −0.04 −0.05 0.13 0.11 0.00 −0.02  (standardized) (0.07) (0.07) (0.07) (0.07) (0.09) (0.09) (0.07) (0.07) Seconds to finish mazes 0.03 0.03 0.03 0.03 0.15 0.15 0.10 0.07   (standardized) (0.07) (0.07) (0.07) (0.07) (0.09) (0.09) (0.07) (0.07) Combined language 0.02 0.04 0.01 0.03 0.08 0.11 0.06 0.09  score (standardized) (0.07) (0.07) (0.07) (0.07) (0.09) (0.09) (0.07) (0.07) Small-stakes risk aversion Percentage invested in 0.79 0.81 0.51 0.54 0.79 0.80 0.21 0.29   risky asset (0.23)*** (0.23)*** (0.24)** (0.24)** (0.32)** (0.32)** (0.24) (0.26)   (out of 100 Ksh) Asset ownership, and formal sector income Owner or spouse has −0.04 0.00 −0.15 −0.12 0.17 0.18 0.15 0.27   a formal sector job (0.18) (0.17) (0.18) (0.18) (0.24) (0.24) (0.18) (0.19) log (acres land 0.03 0.07 0.02 0.05 0.07 0.10 0.07 0.06   owned + 1) (0.08) (0.08) (0.09) (0.09) (0.11) (0.11) (0.08) (0.09) log (value of durable 0.25 0.22 0.21 0.17 0.16 0.11 0.01 0.05   goods and animals (0.13)* (0.13)* (0.13)* (0.13) (0.16) (0.17) (0.13) (0.13)  owned + 1)   (in 10,000 Ksh) Financial access Has bank account 0.19 0.27 0.25 −0.03 −0.10 (0.13) (0.14)** (0.18) (0.14) (0.15) Participates in ROSCA −0.48 −0.50 −0.20 0.08 0.01 (0.12)*** (0.12)*** (0.16) (0.12) (0.13) Would like to borrow 0.00 −0.02 −0.13 −0.13 −0.09   more money but is (0.11) (0.12) (0.15) (0.11) (0.12)   unable to get it log total inventory 0.61 (0.06)*** log inventory on 0.56   top 13 items (0.06)*** Mean of dependent 11.97 11.97 10.89 10.89 9.62 9.62 9.61 9.61   variable SD of dependent 1.06 1.06 1.09 1.09 1.00 1.00 0.97 0.97  variable Observations 380 380 380 380 188 188 184 184 R2 0.10 0.14 0.09 0.14 0.17 0.19 0.53 0.47 Notes: Dependent variables in (log) Kenyan shillings. To avoid dropping observations, we create dummy variables for having missing information for a given variable and code the underlying variable as a 0 when it is missing. Regressions also include controls for gender, marital status, and literacy. Standard errors in parentheses. *** Significant at the 1 percent level.  ** Significant at the 5 percent level.   * Significant at the 10 percent level. 366 AEA PAPERS AND PROCEEDINGS MAY 2013 m ­ aking to depart from the predictions of stan- Table 2). Raven’s matrix scores are significant in dard economic models. The dependent variable some specifications. in columns 1 and 2 of Table 2 is log total inven- Of course, it is possible that the decision tory, while the dependent variable in columns 3 of how much to allocate to a risky portfolio is and 4 is log inventory of the top 13 items (both in endogenous to business performance, and that Kenyan shillings). We start with a sparse speci- difficulty in their business causes shopkeepers fication, including only covariates that are most to invest less in the risky asset. However, since plausibly exogenous to business performance. stakes are small, expected utility maximizers In the second specification, we add various mea- should still take a positive expected value, zero- sures of financial access, asset ownership, and beta gamble. Moreover, reverse causality would formal sector income. The general pattern of not explain the math results, since math scores results is robust to the specific list of included are largely determined by education that ante- covariates. dates establishment of the business. There is some evidence that could be inter- Shopkeepers who invest more in the risky preted as indicating that credit constraints affect asset have higher profits. The data are consistent inventories. Shopkeepers with higher levels of with this working through the inventory chan- other assets have bigger inventories. However, nel (see columns 5–8, Table 2). A one–standard there is no significant correlation between inven- deviation increase in the math score is associated tories and self-reported credit constraints, land with 32 percent higher profits, unconditional on ownership, or formal sector employment. There inventories. Columns 7 and 8 suggest that much is some evidence that those with bank accounts of the math score effect on profits works through have greater inventories, while members of inventories, but that other channels also play a Rotating Savings and Credit Associations, or role.6 ROSCAs, have smaller inventories. It is difficult Several factors suggest entrepreneurial deci- to interpret this as a causal effect of ROSCA sion making is not an immutable inherent char- membership, but this could indicate lower acteristic. Math scores are highly correlated with inventories among those who have more trouble educational attainment. Kremer et al. (2013) saving or are more subject to “taxes” on sav- find that the more shopkeepers are interviewed ing from family members and therefore turn to about stockouts, the less they stockout, and pres- ROSCAs. ent some evidence that shopkeepers increase With or without the credit constraint vari- purchase size after receiving information on ables in the regression, there is evidence that profits lost by missing bulk discount thresholds. small-stakes loss aversion is associated with Similarly, Beaman, Magruder, and Robinson lower inventories. Shopkeepers who invested (2012) report evidence that regularly survey- 10 percent less of a 100 Ksh portfolio in an asset ing small business owners about lost sales from yielding a 0 percent return for sure and 10 per- inadequate supplies of small change and provid- cent more in the risky asset had 7.9–8.1 percent ing information on the lost sales costs reduces greater inventories. Note that a movement of lost sales as they increase their supply of change. 10 Ksh is equivalent to about 0.014 percent of Firm owners’ willingness to change their behav- the value of the median respondent’s stock of ior in response to interventions implies they animal and consumer durables, or to 0.007 per- were not perfectly optimizing initially. cent of the median respondent’s inventory. A one–standard deviation increase in the math score is robustly associated with 14–18 percent higher inventories (columns 1 and 3, Table 2).5 The estimated effect is not affected by including 6 measures of credit constraints (columns 2 and 4, Note that unlike Benjamin, Brown, and Shapiro (forthcoming) we do not find that higher math scores are associated with lower small-stakes risk aversion in a lab- experiment type game, but we do find that higher math scores are associated with higher inventories. This may be 5 The math test was adapted from standard psychomet- because the small-stakes lab experiment gambles are highly ric and personnel IQ tests, including the Wonderlic Test and artificial in our context, while inventory investment is a deci- Cognitive Reflection Test. All modules were refined for the sion that shopkeepers confront every week and have had local context and were extensively pretested. more opportunity and incentive to think through. VOL. 103 NO. 3 behavioral biases and firm behavior: evidence from Kenyan Retail shops 367 III. Discussion Evidence from a Randomised Field Experi- ment in Mongolia.” Institute for Fiscal Studies, Loss aversion can potentially help explain a IFS Working Paper W11/20. series of puzzles related to the persistence of Banerjee, Abhijit, Esther Duflo, Rachel Glenner- unrealized high-return investment opportunities. ster, and Cynthia Kinnan. 2010. “The Miracle Since a loss-averse firm owner may turn down of Microfinance? Evidence from a Random- small, highly positive expected return invest- ized Evaluation.” Unpublished. ments if they carry risk, loss aversion offers a Banerjee, Abhijit V., and Esther Duflo. 2012. “Do potential explanation for several puzzles and Firms Want to Borrow More? Testing Credit recent empirical findings: (i) why shop owners Constraints Using a Directed Lending Pro- with high unrealized returns to divisible invest- gram.” Unpublished. ments do not have the high growth rates of Beaman, Lori, Jeremy Magruder, and Jonathan consumption and assets predicted by the Euler Robinson. 2012. “Minding Small Change: equation even for credit-constrained agents; Limited Attention among Small Firms in (ii)  why many small business owners do not Kenya.” Unpublished. take up microcredit and why many of those Benartzi, Shlomo, and Richard H. Thaler. 2007. who do borrow do not use the loans for busi- “Heuristics and Biases in Retirement Savings ness investment;7 (iii)  the finding in de Mel, Behavior.” Journal of Economic Perspectives McKenzie, and Woodruff (2008) that when 21 (3): 81–104. small business owners are given an infusion of Benartzi, Shlomo, and Richard H. Thaler. 2001. new capital, they neither spend it down (as they “Naive Diversification Strategies in Defined would if they were simply impatient) nor do they Contribution Saving Plans.” American Eco- break out of a poverty trap and accumulate more nomic Review 91 (1): 79–98. assets; (iv) recent findings that insuring farm- Benjamin, Daniel J., Sebastian A. Brown, and Jesse ers against adverse weather shocks can increase M. Shapiro. Forthcoming. “Who is ‘Behav- their willingness to invest (i.e., Karlan et al. ioral’? Cognitive Ability and Anomalous Pref- 2012; Cole, Giné, and Vickery 2011; Mobarak erences.” Journal of the European Economic and Rosenzweig 2012); and (v) recent literature Association. on business training that suggests that training Burks, Stephen V., Jeffrey Carpenter, Lorenz based on “rules of thumb” or simple heuristics Götte, Kristen Monaco, Kay Porter, and Aldo can induce firm owners to change their behavior Rustichini. 2007. “Using Behavioral Eco- (e.g., Drexler, Fischer, and Schoar 2012). nomic Field Experiments at a Firm: the Con- Our findings that small business owners text and Design of the Truckers and Turnover behave as if they are loss averse raise the pos- Project.” In The Analysis of Firms and Employ- sibility that social safety nets might increase ees: Quantitative and Qualitative Approaches, investment among small business owners more edited by Stefan Bender, Julia Lane, Kath- generally. Our work also suggests that at least ryn Shaw, Fredrik Andersson, and Till von some of the heterogeneity in returns to capital Wachter, 45–106. Chicago: University of Chi- identified by Hsieh and Klenow (2009) may be cago Press. due to differences in management quality across Cole, Shawn, Xavier Giné, and James Vickery. firms, as opposed to the impact of tax and regu- 2011. “How Does Risk Management Influence latory distortion across firms. Production Decisions? Evidence from a Field Experiment.” Unpublished. References Crépon, Bruno, Florencia Devoto, Esther Duflo, and William Parienté. 2011. “Impact of Attanasio, Orazio, Britta Augsburg, Ralph De Microcredit in Rural Areas of Morocco: Haas, Emla Fitzsimons, and Heike Harmgart. Evidence from a Randomized Evaluation.” 2011. “Group Lending or Individual Lending? Unpublished. de Mel, Suresh, David McKenzie, and Christopher Woodruff. 2008. “Returns to Capital in Micro- 7 See Banerjee et al. (2010), Karlan and Zinman (2010), Crépon et al. (2011), and Attanasio et al. (2011) for some enterprises: Evidence from a Field Experi- recent evidence on the effects of expanding microcredit ment.” Quarterly Journal of Economics 123 access. (4): 1329–72. 368 AEA PAPERS AND PROCEEDINGS MAY 2013 Drexler, Alejandro, Greg Fischer, and Antoinette “Expanding Microenterprise Credit Access: Schoar. 2012. “Keeping it Simple: Financial Using Randomized Supply Decisions to Esti- Literacy and Rules of Thumb.” Unpublished. mate the Impacts in Manila.” Unpublished. Duflo, Esther, Michael Kremer, and Jonathan Kremer, Michael, Jean Lee, Jonathan Robinson, Robinson. 2011. “Nudging Farmers to Use and Olga Rostapshova. 2013. “The Return to Fertilizer: Theory and Experimental Evidence Capital for Small Retailers in Kenya: Evidence from Kenya.” American Economic Review 101 from Inventories.” Unpublished. (6): 2350–90. McKenzie, David, and Christopher Woodruff. Fafchamps, Marcel, David McKenzie, Simon 2008. “Experimental Evidence on Returns to R. Quinn, and Christopher Woodruff. 2011. Capital and Access to Finance in Mexico.” “When is Capital Enough to get Female World Bank Economic Review 22 (3): 457–82. Microenterprises Growing? Evidence from a Mobarak, A. Mushfiq, and Mark Rosenzweig. Randomized Experiment in Ghana.” National 2012. “Selling Formal Insurance to the Infor- Bureau of Economic Research Working Paper mally Insured.” Unpublished. 17207. Rabin, Matthew. 2000. “Risk Aversion and Hsieh, Chang-Tai, and Peter J. Klenow. 2009. Expected-Utility Theory: A Calibration Theo- “Misallocation and Manufacturing TFP in rem.” Econometrica 68 (5): 1281–92. China and India.” Quarterly Journal of Eco- Thaler, Richard H., Amos Tversky, Daniel Kahne- nomics 124 (4): 1403–48. man, and Alan Schwartz. 1997. “The Effect of Kahneman, Daniel, Jack L. Knetsch, and Richard Myopia and Loss Aversion on Risk Taking: An H. Thaler. 1990. “Experimental Tests of the Experimental Test.” Quarterly Journal of Eco- Endowment Effect and the Coase Theorem.” nomics 112 (2): 647–61. Journal of Political Economy 98 (6): 1325–48. Tversky, Amos, and Daniel Kahneman. 1991. Karlan, Dean, Robert Darko Osei, Isaac Osei- “Loss Aversion in Riskless Choice: A Refer- Akoto, and Christopher Udry. 2012. “Agricul- ence-Dependent Model.” Quarterly Journal of tural Decisions after Relaxing Credit and Risk Economics 106 (4): 1039–61. Constraints?” National Bureau of Economic Udry, Christopher and Santosh Anagol. 2006. Research Working Paper 18463. “The Return to Capital in Ghana.” American Karlan, Dean, and Jonathan Zinman. 2010. Economic Review 96 (2): 388–93.