Three essays on statistical inference for stock return predictions and capital asset pricing models

Abstract

In this dissertation, I focus on econometric issues arising in the fields of Financial Economics. In the first chapter, I study return predictability in international equity markets focusing on the effects of the bias and spurious regression problems for statistical inference. The slope coefficient estimator in predictive regressions for stock returns is biased in the presence of a lagged stochastic regressor. Spurious regression may also occur if the underlying expected return is highly persistent. I consider the effect of these biases in the presence of data mining for the predictive variables. I find that the two biases can reinforce or offset each other, depending on the parameters of the model. I present a new bias expression valid with an unobserved true expected returns and re-evaluate return predictability in international equity markets adjusting for data mining associated with both effects. The second chapter studies tests for structural changes in the trend function of a univariate time series that are robust to whether the noise component is stationary (I (0)) or contains an autoregressive unit root (I (1)). The tests of interest are the robust procedures recently proposed by Perron and Yabu (2009) and Harvey, Leybourne and Taylor (2009), both of which attain the same limit distribution under I (0) and I (1) errors. We compare their finite sample size and power under different data-generating processes for the noise components. We apply the tests to a large historical panel of real exchange rates with respect to the U.S. dollar for 19 countries and document simultaneous shifts in level and trend for many series. The third chapter studies the sampling interval effect in estimating capital asset pricing models. In past empirical studies, the beta coefficient estimates are documented to be sensitive to the sampling interval used for returns. We provide a theoretical framework to explain this sampling interval effect. We show that it can be attributable to the existence of transitory components in stock prices, and provide empirical evidence supporting its presence.