Out-of-Sample Performance of Mutual Fund Predictors

We analyze the out-of-sample performance of variables shown to forecast future mutual fund alphas. The degree of predictability, as measured by alpha spreads from quintile sorts or by cross-sectional regression slopes, falls by at least half post-sample. These declines appear to be primarily the result of changes in the level of arbitrage activity in the market, with mutual fund competition appearing to play a secondary role. We find no evidence that the declines are the result of data snooping or learning. Finally, we show that corporate bond fund performance exhibits similar dependence on measures of bond market arbitrage activity