Analysis of data from 1871 to 1925 finds that low-risk industries earned higher risk-adjusted returns and that the low-risk industry effect is not a result of data mining in earlier studies. Behavioral biases and liquidity may be key factors.
Abstract
Recently, there has been discussion of a “replication crisis” in Finance, where many empirical results in financial research are said not to be replicable. Previous research finds that low-risk stocks have higher returns than higher-risk stocks on a risk-adjusted basis. We reexamine the low-risk effect using a unique dataset for U.S. industries from 1871 to 1925. We confirm the presence of the effect for portfolios of U.S. industries, indicating that the low-risk effect is not due to data mining in previous studies. Comparing the results to that for more recent data, we find that the overall effect is at least as strong in the earlier data. Given that some market frictions were fewer in the earlier period, the results suggest that implicit trading costs, illiquidity, and/or behavioral biases may play an important role in the low-risk effect.