Ambiguity and Insurance

Why don’t people in developing countries insure themselves? I propose that some households are ambiguity averse, and that these households fear any insurance contract they purchase will payout when it is not needed and will not payout when it is. I study two applications of this observation. I first study formal crop insurance, in particular index insurance and the insurance element of limited liability credit. I provide a formal model of ambiguity aversion and derive four implications. First, households that are sufficiently ambiguity averse will not value any actuarially fair insurance contract and will have a lower willingness to pay for any specific contract. Second, because learning diminishes the impact of ambiguity, increased experience with the insured crop will mitigate the effect of ambiguity aversion. Third, increasing risk aversion makes index insurance more valuable for the ambiguity neutral, and relatively less valuable for the ambiguity averse. Fourth, increasing risk aversion has no effect on the demand for limited liability credit (ie, a state contingent transfer) for ambiguity neutral individuals, but increases demand for limited liability credit for the ambiguity averse. I test these predictions using two experiments, one on rainfall insurance in Malawi and one on limited liability credit in Kenya. Both data sets contain baseline measures of ambiguity aversion and risk aversion, and strongly support the theoretical claims. As a second application I study informal risk sharing. I provide a model showing that the same preferences used to study crop insurance can explain the common finding that consumption is sensitive to idiosyncratic income shocks. The model I use assumes that households cannot make accurate probability assessments and entertain a set of priors. I assume these households assess uncertain prospects using a prior that minimizes the gain from leaving the endowment. The model implies that households have different beliefs at the optimal contract, creating a wedge between the consumption levels of households giving and receiving transfers in any state. I derive testable implications from the model, which I evaluate using both the ICRISAT village level data from India and the Townsend Thai monthly panel. I am unable to reject the model using either data set. 1 Ambiguity Aversion and Missing Insurance Markets