Can the Representativeness Heuristic Explain the Asset Growth Anomaly?

Abstract

We find that the low average returns to firms with high asset growth are consistent with two key implications of models of diagnostic investor expectations (e.g., Bordalo, Gennaioli, La Porta, and Shleifer, 2019) that formalize the representativeness heuristic of Kahneman and Tversky (1972). These models predict that investors overestimate the subjective probability of states that are more representative of a firm’s type and also neglect risk after a string of good news. We construct a measure of how representative the stereotype the ‘next Google’ is of high asset growth in the recent past. We show that this measure predicts the returns of CMA, the asset growth factor in the five factor model of Fama and French (2015). Returns to CMA are 17.5% over the 3 years after months with high representativeness and only 5.4% following low representativeness months. In the cross-section, we find evidence consistent with investors neglecting the risk of high asset growth firms. The asset growth effect is not present in portfolios with low distress risk and the interaction between distress and asset growth, rather than asset growth by itself, predicts low returns in Fama and MacBeth (1973) regressions and portfolio sorts. We also find that analysts and markets do not appreciate the importance of the interaction between asset growth and distress and are sluggish in responding to news for the ‘interaction’ portfolio. Finally, we show that our measure of representativeness predicts the returns of the interaction portfolio.