Comovement Revisited

Recent evidence of excessive comovement among stocks following index additions (Barberis, Shleifer, and Wurgler, 2005) and stock splits (Green and Hwang, 2009) challenges traditional finance theory. Based on a simple model, we show that the bivariate regressions relied upon in the literature often provide little or no information about the economic magnitude of the phenomenon of interest, and the coefficients in these regressions are very sensitive to time-variation in the characteristics of the return processes that are unrelated to excess comovement. Instead, univariate regressions of the stock return on the returns of the group it is leaving (e.g., non-S&P stocks) and the group it is joining (e.g., S&P stocks) reveal the relevant information. When we reexamine the empirical evidence using control samples matched on past returns and compute Dimson betas, almost all evidence of excess comovement disappears. The results in the literature are consistent with changes in the fundamental factor loadings of the stocks. One key element to understanding these striking results is that, in both the examples we study, the stocks exhibit strong returns prior to the event in question. We document the heretofore unknown empirical regularity that winner stocks exhibit increases in betas. Thus, much of the apparent excess comovement is just a manifestation of momentum.