Prediction and Model Selection in Experiments

This paper compares the predictive performance of several models of risk aversion and time preferences in experimental settings. Models are evaluated on the basis of out‐of‐sample prediction rather than in‐sample fit. For preferences over risk, with the exception of very small sample sizes, allowing the estimation procedure to select between constant relative risk aversion and constant absolute risk aversion improves prediction beyond that of a single model. Moreover, adding a behavioural parameter such as disappointment aversion improves prediction further. This contrasts with time preferences, where adding the present‐bias parameter worsens prediction for all sample sizes.