Modern literature departs from time-separable constant relative risk aversion preferences to explain asset pricing facts. This deviation typically implies that wealth shocks generate transitory variations in agents’ relative risk aversion and, possibly, portfolio re-allocations over time. I empirically analyze this relationship using U.S. macroeconomic data and and evidence for time-variation in portfolio shares that is consistent with counter-cyclical risk aversion. These results suggest, therefore, that wealth-dependent, habit-formation or loss and disappointment aversion utility functions are a good description of preferences. Controlling for observed versus expected asset returns, I also show that: (i) wealth effects are significant (although temporary) and there is no evidence of inertia contrary to Brunnermeier and Nagel (2006); and (ii) the consumption-wealth ratio (Lettau and Ludvigson, 2001), the labor income risk (Julliard, 2004) and the labor income-consumption ratio (Santos and Veronesi, 2006) partially explain changes in the risky asset share.
