Latent variable tests of asset pricing models make assumptions about the joint distribution of observable returns and unobservable benchmark returns. These tests can falsely accept models when a mean-variance efficient portfolio other than the benchmark satisfies the distributional assumptions imposed on the benchmark portfolio. Also, because the assumptions are untestable, there is no way to discover whether a model is being rejected because the assumptions are false. Without these assumptions, however, latent variable tests can be viewed only as tests of distributional hypotheses about mean-variance efficient portfolios of unknown composition.

It has long been market folklore that the best time to buy stocks is when individual investors are bearish, and the best time to sell is when individual investors are bullish. We examine the forecast power of three popular measures of individual investor sentiment: the level of discounts on closed-end funds, the ratio of odd-lot sales to purchases, and net mutual fund redemptions. Using data from 1933 to 1993, we find that fund discounts and net redemptions predict the size premium, the difference between small and large firm returns, but little evidence that the odd-lot ratio predicts returns.