Working Papers

[4] "The Subjective Risk and Return Expectations of Institutional Investors" (with Spencer Couts and Andrei Goncalves)

We use the long-term Capital Market Assumptions of major asset managers and investment consultants from 1987 to 2022 to study their subjective risk and return expectations across 19 asset classes. We find a strong and positive subjective risk-return tradeoff, with most of the level and variation in subjective expected returns arising from risk premia (beta compensation) rather than alphas. Subjective expected returns predict future realized returns both across asset classes and over time, with subjective risk premia driving most of this predictability. Subjective risk also predicts future realized risk, with stronger predictability across asset classes than over time.

[3] "Subjective Beliefs and the Portfolio Allocations of Institutional Investors" (with Aleksandar Andonov, Andrei Goncalves,   Spencer Couts and Andrea Rossi)

We study how subjective beliefs shape the portfolio allocations of institutional investors. Linking the multi-asset allocations of U.S. public pension funds to the long-term capital market assumptions of their consultants, we examine the extent to which differences in subjective expected returns, volatilities, and correlations map into differences in portfolio weights. We embed these belief inputs in a mean-variance framework that incorporates fund-consultant belief wedges, heterogeneous risk aversion, non-negative weight constraints, and a benchmarking incentive due to frictions. We find that pension fund allocations are significantly linked to belief-implied mean-variance efficient allocations across pension funds, across asset classes, and over time. Accounting for frictions is essential: it dramatically increases the pass through and explanatory power of beliefs to portfolio allocations. Overall, our results show that beliefs play a central role in institutional portfolio decisions and that frictions critically shape their transmission into observed allocations.

[2] "Institutional Investors' Subjective Risk Premia: Time Variation and Disagreement" (with Spencer Couts, Andrei Goncalves, and Yicheng Liu)

We study how institutional investors' subjective risk premia shape variation in their expected returns over time and across institutions. Our analysis uses long-term Capital Market Assumptions from asset managers and investment consultants from 1987 to 2022. Perceived market risk premia explain most of the countercyclicality and overall time variation in expected returns, with alphas accounting for the remainder. This risk premium effect is driven almost entirely by variation in perceived risk quantities rather than variation in the price of risk (risk aversion). Expected return disagreement across institutions rises with macro-financial uncertainty and is also explained primarily by disagreement about market risk premia. However, in contrast to the time-series results, alphas account for a quantitatively important share of disagreement, and the price and quantity of risk contribute roughly equally to the risk premium effect. These findings provide benchmark moments that asset pricing models should match to be consistent with institutional investors' beliefs.

[1] "Out-of-Sample Alphas Post-Publication"  (with Andrei Goncalves and Richard Ogden)

Anomaly strategies generate positive and significant CAPM alphas post-publication. Existing explanations include non-market risks, trading costs, and investment frictions. This paper introduces a complementary and novel channel: when a new anomaly strategy is published, investors face uncertainty in identifying the optimal weight to allocate to the anomaly in order to achieve a positive alpha post-publication, making the strategy less appealing. Empirically, we find that the average post-publication alpha of anomaly strategies is close to zero when optimal weights are estimated out-of-sample using pre-publication data. This finding is robust across specifications, including those using empirical Bayesian shrinkage and machine learning to estimate weights. Conceptually, this suggests investors have little incentive to add a new anomaly strategy to their portfolios. While investors can generate positive out-of-sample alphas by combining multiple anomaly strategies via shrinkage methods, we show the demand from such investors is insufficient to eliminate alphas in equilibrium.