Two essays on time series patterns in security returns.

Abstract

This dissertation consists of two essays that examine the time-series behavior of returns on portfolios of securities. The first essay, Time-series restrictions on factor-mimicking portfolios, demonstrates how a multifactor asset pricing model with rationally or irrationally time-varying expected returns reduces the set of viable factors by imposing time-series restrictions on the factor realizations. Regardless of the number of factors and the contemporaneous correlation among them, returns on each factor-mimicking portfolio should Granger-cause returns on any randomly or systematically chosen portfolio of securities, while the reverse is not true. These time-series restrictions are used to test the appropriateness of the three factor-mimicking portfolios employed by Fama and French (1993), which are increasingly used as the factors in most statistical models of stock returns. While the evidence for the equally-weighted market portfolio of NYSE/AMEX stocks is ambiguous in the sense that causality appears to be bidirectional, the Fama-French size and book-to-market factor-mimicking portfolios do not satisfy these time-series restrictions. Returns to portfolios of assets Granger-cause returns to the size and book-to-market factor-mimicking portfolios, while the reverse is not true. The second essay, Short-term autocorrelation in security returns, examines the determinants of short-term autocorrelation in security returns. Short-horizon returns to portfolios of securities have been shown to be significantly positively autocorrelated. This phenomenon has two broad classes of explanations: (a) market frictions related to nonsynchronous trading, and (b) time-variation in expected returns from either rational or irrational sources. Two sets of restrictions are developed to discriminate between these explanations under the assumption that the only source of return persistence is nonsynchronous trading. First, given a fixed level of nonsynchronous trading, there exists a structural relation between daily and weekly autocorrelations. Second, returns to a portfolio of synchronously traded securities should lead, both unconditionally and in a Granger sense, returns to a portfolio of securities with nonsynchronous trading, while the reverse is not true. The two sets of restrictions are tested using portfolios of stocks that are created based on differing levels of nonsynchronous trading. Security autocorrelations are demonstrated to be largely attributed to time variation in expected returns rather than market microstructure effects.