Empirical Asset Pricing

This project is primarily concerned with the econometric testing of asset pricing theories using data on equities, currencies, and bonds. The first part develops an empirical analysis of several competing equity pricing models. Currently, there is no generally accepted asset-pricing model. The most important model in the literature is the CAPM. Unfortunately, it is now well known that the CAPM fails along a number of dimensions. Its primary failures are the size and book-to-market anomalies: the CAPM under-predicts the average returns for small firms and high book-to-market firms. Several new models have been offered as replacements for the CAPM. All of the models have distinct business-cycle attributes, and the proposed research should enhance our understanding of the links between observed cyclical fluctuations and the required rates of return on equities. I propose to compare the new models and the CAPM on a common data set using the methodology of Hansen and Jagannathan. I also plan to draw out the implications of the analysis for international pricing of equities. The analysis is important because the expected return on equity is one aspect of the cost of capital for corporations. Without an understanding of the determinants of required rates of return on equity, the profession cannot advise corporations how to perform capital budgeting analyses.

The second part of the proposal is a comprehensive econometric examination of the expectations hypotheses of the foreign exchange market and the bond market. Econometric inference using standard asymptotic distribution theory indicates strong evidence against the expectations hypotheses. In fact, the evidence is so strange that development of alternative models with time-varying risk premiums has proved difficult. However, there is accumulating evidence that the small-sample distributions of the usual test statistics do not conform well to the asymptotic theory. In such a situation, it is prudent to take a close examination of this issue by developing a

well-structured Monte Carlo analysis in which the null hypotheses are true by construction and the data-generating process conforms as closely as possible to the actual data. The proposed research will estimate vector autoregressions that satisfy the null hypotheses, and it will use modern bootstrap methods to generate the simulated data. The estimation strategy is something that should be emulated widely as it involves only matrix manipulations and does not involve non-linear search routines which are often difficult to implement and time consuming to estimate.