Assistant Professor
NC State University
The Level, Slope and Curve Factor Model for Stocks, 2022
Journal of Financial Economics 143, 159-187.
I develop a method to extract only the priced factors from stock returns. The first step estimates expected returns based on firm characteristics. The second step uses the estimated expected returns to form portfolios. The last step uses principal components analysis to extract factors from the portfolio returns. The procedure isolates and emphasizes the comovement across assets that is related to expected returns as opposed to firm characteristics. It produces three factors--level, slope and curve--which perform as well or better than other leading models. Popular factors contribute little additional explanatory power. The new factors have macroeconomic risk interpretations.
[Discussion by John Cochrane]
Characteristics and the Cross-Section of Covariances (with Matthew Linn)
American Finance Association Annual Meeting (2023)
We model firm-level, stock return covariances as a function of firm characteristics. Flexible panel regressions allow us to estimate the marginal predictive power associated with characteristics in a multi-dimensional setting where portfolio sorts are infeasible. We use the model to identify characteristics that proxy for priced factor exposures, unpriced factor exposures, and near-arbitrages while circumventing the need to identify underlying risk factors. Cyclical variation in how characteristics are related to covariances shows that many well-known characteristics proxy for exposure to business cycle risk while few proxy for market sentiment.
Testing Asset Pricing Models with Individual Stocks (with Morteza Momeni)
American Finance Association Annual Meeting (2023)
This paper tests asset pricing models using individual stocks as test assets, rather than sorted portfolios. Sorted portfolios have the severe limitation that the researcher must know, in advance, reliable predictors of expected returns. We show how to generate appropriately sized tests and verify that our tests have considerable test power. We apply our tests to seven leading factor models. We reject six of the seven leading models we test. The instrumented factor model of Kelly et al. (2019) stands out as the most successful. We show a natural extension of our approach incorporates characteristic driven time-variation in factor loadings.
Biased Expectations and the Time-Series of Anomaly Returns (with Russel Jame)
Recent evidence indicates that the returns on anomaly strategies are diminishing over time. We argue that time-varying biases in cash-flow expectations explain much of this decline. Specifically, analyst optimism for stocks in the short leg of anomalies relative to the long leg has fallen by more than 50% in the last third of the sample, and this trend accounts for nearly 50% of the decline in anomaly returns. We develop an ex-ante measure of biased expectations and show that this measure explains most of the time-series variation in anomaly valuation ratios and strongly predicts future anomaly returns.
Unexpected Consumption Growth and Stock Returns
This paper revisits the low correlation between consumption growth and aggregate stock market returns. I show that professional forecasters are able to predict consumption growth both quarterly and annually. After accounting for this predictability, a 1% surprise in contemporaneous consumption growth is associated with 6.26% (t-stat 5.02) higher stock returns, while a 1% surprise in future growth is associated with an 8.28% (t-stat 3.79) increase. This approach explains 21% of return variation, a stark contrast with the 0% in the traditional approach. The puzzling lack of correlation between the market portfolio and consumption data largely disappears after incorporating forecaster expectations.