Robust Econometric Inference for Stock Return Predictability

Abstract

This study examines stock return predictability via lagged …nancial variables with unknown stochastic properties. We propose a novel testing procedure that (1) robusti…es inference to regressors' degree of persistence, (2) accommodates testing the joint predictive ability of …nancial variables in multiple regression, (3) is easy to implement as it is based on a linear estimation procedure, and (4) can be used for long-horizon predictability tests. We provide some evidence in favor of short-horizon predictability during the 1927-2012 period. Nevertheless, this evidence almost entirely disappears in the post-1952 period. Moreover, predictability becomes weaker, not stronger, as the predictive horizon increases.

On Confidence Intervals for Autoregressive Roots and Predictive Regression

“…Jansson and Moreira (2006) developed a conditional likelihood approach with certain optimal asymptotic properties that, in principle, extends to multiple regressors. Instrumental variable techniques are also available for use in predictive regression, such as the IVX method of Magdalinos and Phillips (2009) which applies to stationary, mildly integrated, and local to unity regressors (Kostakis, Magdalinos, and Stamatogiannis (2012)). The Jansson-Moreira test has not yet been used in empirical research and simulations in Kasparis, Andreou, and Phillips (2012) indicate that the test encounters difficulties in numerical implementation and has size distortion in cases of strong endogeneity.…”

“…They found in simulations that the Campbell-Yogo test had superior power against most alternatives. That method has good size and power properties in simulations (Kostakis, Magdalinos, and Stamatogiannis (2012)), accommodates multiple regressors easily, and allows for varying degrees of persistence (as often occurs in empirical work) as well as mildly explosive roots. In other recent work, Elliott, Müller, and Watson (2012) developed a "nearly optimal" test, treating the localizing coefficient c as a nuisance parameter and using a likelihood ratio test that optimizes weighted average power over c ∈ [−40 5] and switches to the standard t test when the maximum likelihood estimateĉ < −35.…”

Using Market Expectations to Test for Speculative Bubbles in the Crude Oil Market

“…Several methods have been proposed in the literature to draw valid statistical inference in this setting, such as the efficient Q ‐test of Campbell and Yogo (), the conditional likelihood approach of Jansson and Moreira (), and the nearly optimal test of Elliott, Müller, and Watson (). We choose to work with the IVX instrumentation method of Phillips and Magdalinos (), Phillips and Lee (), and Kostakis, Magdalinos, and Stamatogiannis ().…”

“…and the corresponding IVX test statistic follows a standard normal distribution (Phillips and Magdalinos , Phillips and Lee , Kostakis, Magdalinos, and Stamatogiannis ). In addition to allowing standard hypothesis testing, the IVX has several advantages over alternative tests: first, it is valid for autoregressive processes that cover the whole spectrum from stationary to mildly explosive; second, it displays good finite‐sample properties (Kostakis, Magdalinos, and Stamatogiannis , Pavlidis, Paya, and Peel ); third, it is easy to implement since it is based on simple filtering techniques and linear regression; and, finally, it can be extended to long‐horizon predictive regressions (see Phillips and Lee ).…”

“…Monte Carlo simulation results in Kostakis, Magdalinos, and Stamatogiannis () suggest setting and .…”