Bayesian perspectives on portfolio allocation and stock return prediction

This dissertation consists of two essays that use Bayesian statistical methods to examine aspects of financial asset pricing and asset price prediction. The first essay examines portfolio choice when an investor lacks full confidence in an asset pricing model. The second essay examines prediction of future asset returns when an investor is uncertain which prediction model is correct.; In the first essay, I examine the portfolio choices of hypothetical mean-variance investors who use the Capital Asset Pricing Model (CAPM) to allocate wealth between a momentum portfolio and the market portfolio. An investor with complete confidence in the CAPM would allocate nothing to the momentum portfolio. In contrast, an investor who lacks full confidence in the CAPM might abandon the model entirely. Rather than take this harsh approach, an investor could instead specify this lack of confidence as a Bayesian prior belief in the ability of the CAPM to price the momentum portfolio. In this way, the investor can make use of the asset pricing model without having to rely on it entirely. I use data for 1926 through 2001 to examine the portfolio allocation choices investors with different prior beliefs would have made. Not surprisingly, investors with less prior confidence in the CAPM eventually allocate more to the momentum portfolio. However, and somewhat surprisingly, an investor with near complete confidence in the CAPM would still have allocated a substantial percentage to the momentum portfolio.; In the second essay, I examine prediction of future stock returns. Previous studies have identified several variables that would have predicted stock returns, though others studies suggest these results may be due to data mining. To guard against data mining, previous researchers have suggested use of Bayesian model averaging to account for the uncertainty about prediction models. In common with other researchers, I find some evidence of predictability when a hypothetical investor uses Bayesian model averaging with no restrictions on use of predictive variables. However, when I limit the hypothetical investor to using only variables whose predictive ability would have been known at the time of the forecast, the predictability disappears.