Economic theory assigns a central role to risk preference in assetallocation. This dissertation includes three papers thatinvestigate this relationship empirically. The first paper usespanel data on hypothetical gambles over lifetime income in theHealth and Retirement Study to quantify changes in risk toleranceover time and differences across individuals. Themaximum-likelihood estimation of a model with correlated randomeffects draws on detailed information from 12,000 respondents inthe 1992-2002 HRS. The results support constant relative riskaversion and career selection based on preferences. Whilerisk tolerance changes with age and macroeconomic conditions,persistent differences across individuals account for 73% of thesystematic variation in preferences. The measure of risk tolerancealso relates to actual stock ownership.The second paper develops a measure of relative risk toleranceusing responses to hypothetical income gambles in the HRS. Incontrast to most survey measures that produce an ordinal metric,this paper shows how to construct a cardinal proxy for the risktolerance of each survey respondent. The paper also shows how toaccount for measurement error in estimating this proxy and how toobtain consistent regression estimates despite the measurementerror. The risk tolerance proxy is shown to explain differences inasset allocation across households.The third paper investigates whether the characteristics ofhousehold labor income can account for the observed heterogeneityin financial portfolios. Households differ substantially in theriskiness of their labor income and in the magnitude of theirlabor income relative to their financial assets; however, theresults of this paper suggest that households do not integratetheir human capital in their financial asset allocation. Thisanalysis uses direct, household-level comparisons between actualstock allocations and predicted allocations in three economicmodels with different assumptions about labor income. When laborincome is excluded from the model, the correlation between actualand predicted stock allocations is 0.16. The inclusion of certainor risky labor income in the model leads to negative correlationsof -0.12 and -0.06 respectively. There is no evidence thathouseholds view their wealth broadly and diversify risks acrosstheir financial assets and human capital.