Abstract

This paper studies the design of investment policies in defined contribution retirement systems. I estimate a dynamic system of correlated equations of lifecycle behavior that fully models the individual's decision-making process to account for estimation biases. In the model, individuals make decisions that impact their retirement wealth within the Chilean retirement system. Behaviors are allowed to depend on risk preferences while modeling other sources of nonlinear unobserved heterogeneity. The estimated decision-making process allows us to simulate the effects of policy experiments (ex ante), such as defaulting individuals into riskier investment strategies or increasing contribution rates. The results indicate that individuals react by opting into safer plans despite their observed inertia and that increases in mandatory contributions generate little crowding out of other behaviors. Not modeling risk aversion and its endogeneity with behavior leads to substantial simulation biases.