Using data on the portfolio holdings and income of millions of U.S. retirement investors, I find that positive and persistent shocks to income lead to a significant increase in the equity share of investor portfolios, while increases in financial wealth due to realized returns lead to a small decline in the equity share. In a standard homothetic life-cycle model with human capital and constant risk aversion, the portfolio responses to these two wealth shocks should be of equal magnitude and opposite sign. The positive net effect in the data is evidence for risk aversion that decreases in total wealth. To quantify the implications for risk preferences, I estimate a structural life-cycle consumption and portfolio choice model that accounts for inertia in portfolio rebalancing. The model matches the reduced-form estimates with a significant degree of non-homotheticity in risk preferences, such that a 10% permanent income growth leads to an average decrease in risk aversion by 1.7%. I find that decreasing relative risk aversion in the model doubles the share of wealth at the top, as equity is concentrated in the hands of the wealthy. The model also implies that rising income inequality in the U.S. over the past three decades has led to increased demand for equity that is equivalent to a 15% decline in the equity premium.
[Online Appendix] [NBER Working Paper w25108] [Video summary]
Following the US national election of 2016, Republicans report becoming more optimistic about future US economic performance, and Democrats more pessimistic. Using a proprietary dataset containing the portfolio holdings of millions of US households and inferring political affiliation probabilistically from zip codes, we show that these reported beliefs are informative and that there are real economic effects of subjective disagreement: (likely) Republicans increase the equity share and market beta of their portfolios following the election, while (likely) Democrats rebalance into safe assets. Detailed demographic controls for age, wealth, income, location, and even county-employer-period controls allow us to rule out many non-belief-based channels for this effect, such as hedging needs, preferences, or economic exposure. We conclude that people interpret public information using different models of the world and show this has real effects on portfolios and the volume of trade in assets.