Most classical tests of constant relative risk aversion (CRRA) based on individual portfolio composition use cross sectional data. Such tests must assume that the distributions of wealth and preferences are independent. We use panel data to analyze how individuals’ portfolio allocation between risky and riskless assets varies in response to changes in total financial wealth. We find the elasticity of the risky asset share to wealth to be small and statistically insignificant, supporting the CRRA assumption; this finding is robust when the sample is restricted to households experiencing ‘large’ income variations. Various extensions are discussed.