eThesis - Kauppakorkeakoulun sähköiset gradut

The objective of my thesis is to study the cause for the low beta anomaly, which is an observation that the high beta stocks perform poorly relative to the low beta stocks. Based on earlier findings, I hypothesize that if a stock has high investor attention, its price overreacts to market-wide shocks, which results in a positive measurement error in its beta. Simultaneously, high attention causes overpricing, because the stock overreacts more often to positive shocks than to negative shocks as the average increment in the market price is positive. Since the overvaluation is corrected in the future periods, the stock has poor expected return. The hypothesized effects are inversed for low attention stocks.

In combination, these effects predict negative correlation between the measurement errors in betas and the expected returns, which is observed as the low beta anomaly. I formalize the hypothesis by presenting a theoretical model that allows variation in investor attention in an otherwise standard capital asset pricing model (CAPM) framework. The model yields several testable predictions that I study in the empirical sections of the thesis.

Using data on US stock markets from January 1926 to December 2013, I find empirical evidence consistent with the four key predictions of my model. First, the change in investor attention is positively correlated with the change in the beta estimate. Second, portfolios that consist of high (low) beta stocks perform well (poorly) during the period when the betas are estimated and subsequently perform poorly (well). Third, the abovementioned return patterns are reversed after periods with negative market returns. Fourth, the empirical relation between beta and return is positive if lagged instead of contemporaneous beta estimates are used as the proxies for the true betas. The intuition behind the last prediction is that when lagged betas are used, the information their errors carry on mispricing is negligible by the time the estimates are used because the mispricing has already been corrected. The last finding is robust to variation in the beta estimation period length and the number of lags in the beta estimate. The finding also persists if the sample is divided into multiple subperiods, or if the Fama-French pricing factors are introduced to the analysis.