Measuring risk attitudes among the urban poor in Kolkata, India: an experimental approach by Joseph Cook, Susmita Chatterjee, Dipika Sur, John Clemens, and Dale Whittington
Evans School Working Paper No. 2009-01 (316 KB PDF)
Abstract: An important question in development economics is how the poor make choices under uncertainty, with implications ranging from private demand for “next generation” vaccines to investments in new agricultural technologies to the design of financial instruments in micro lending. Our study is the first to examine risk preferences in an urban setting in a low-income developing country (Kolkata, India). Rather than rely on simple coin-toss experiments, we adapted a multiple price list experiment often used in developed countries. We conducted 22 group experiments with a total of 404 participants and used real payoffs to ensure incentive compatibility. The raw data indicate that the majority of respondents showed relatively mild risk aversion (0 < r < 0.65, where r is the coefficient on a utility function with constant relative risk aversion), similar to preferences found in developed countries but generally less risk averse than the results from studies in rural populations of developing countries. We also find a higher fraction of risk-loving participants than the results from studies in rural populations of developing countries. Using a maximum likelihood approach to directly estimate the constant relative risk aversion coefficient, we find the 95% confidence intervals for r are 1.02 - 1.33 under an expected utility specification and 1.30 - 1.66 under a prospect theory specification that uses a probability weighting function. These results suggest that participants in Kolkata are considerably more risk averse than rural participants in a recent study in Ethiopia, Uganda, and India that uses a similar modeling framework (Harrison et al 2009). We find little evidence that socioeconomic characteristics are correlated with risk preferences. A sizeable minority (20-42%) had difficulty understanding the experiment, and participants were influenced by the context in which the experiments occurred, although these problems are not unique to our study.