This dissertation studies retirement savings, weather insurance take-up and reference-dependent theory in the literature of development economics and behavioral economics. It consists of two field experiments and one laboratory experiment.

In Chapter one, I uses a field experiment to study the relationship between financial literacy and retirement savings in China. When the Chinese government launched a highly subsidized pension system in rural areas in 2009, 73% of households chose to save at a level that is lower than that implied by a benchmark life-cycle model. We test to what extent the low contribution level is due to a fundamental misunderstanding of the nature of compound interest. In a field experiment with more than 1000 Chinese households, we randomly assigned some households to a financial education treatment, emphasizing the concept of compound interest. This treatment increased the pension contribution by roughly 40%. The increase accounts for 51% of the gap between contribution levels in the Control group and those implied by the benchmark model. To pinpoint mechanisms, we elicited financial literacy after the intervention, and added a third group in which we explain the pension benefit in general. We find that the neglect of compound interest is correlated with low contributions to the pension plans in the control group, and that financial education about compound interest does help households partially correct their erroneous understanding of compound interest. Moreover, explaining compound interest increases their ability to translate benefits into their own situation. Welfare analysis suggests that financial education increases total welfare, although the fact that the treatment effects are heterogeneous implies that some households end up saving more than the level implied by the benchmark model.

In Chapter two (coauthored with Jing Cai), we use a novel experimental design to test the role of experience and information in insurance take-up in rural China, where weather insurance is a new and highly subsidized product. We randomly selected a group of poor households to play insurance games and find that it increases the actual insurance take-up by roughly 48%. To pinpoint mechanisms, we test whether the result is due to: (1) changes in risk attitudes, (2) changes in the perceived probability of future disasters, (3) learning the objective benefits of insurance, or (4) the experience of hypothetical disaster. We show that the overall effect is unlikely to be fully explained by mechanisms (1) to (3), and that the experience acquired in playing the insurance game matters. To explain these findings, we develop a descriptive model in which agents give less weight to disasters and benefits which they experienced infrequently. Our estimation also suggests that experience acquired in the recent insurance game has a stronger effect on the actual insurance take-up than that of real disasters in the previous year, implying that learning from experience displays a strong recency effect.

In Chapter three, I conducted a controlled lab experiment to test to what extent expectations and the status quo determine the reference point. In the experiment, I explicitly manipulated stochastic expectations and exogenously varied expectations in different groups. In addition, I exogenously varied the time of receiving new information and tested whether individuals adjust their reference points to new information, and the speed of the adjustment. With this design, I jointly estimated the reference points and the preferences based on the reference points. I find that both expectations and the status quo influence the reference point but that expectations play a more important role. Structural estimation suggests that the model of the stochastic reference point fits my data better than that with expected utility certainty equivalent as the reference point. The result also suggests that subjects adjust reference points quickly, which further confirms the role of expectation as reference point