A1435
Title: Simultaneous inference in possibly incomplete multivariate regressions with application to asset pricing
Authors: Lynda Khalaf - Carleton (Canada) [presenting]
Marie-Claude Beaulieu - Universite Laval (Canada)
Jean-Marie Dufour - McGill University (Canada)
Abstract: The focus is on finite sample tests and confidence procedures for a system of several estimating equations with cross-dependent, possibly non-Gaussian disturbances, and allowing for possibly missing explanatory variables. Standard statistics, based on the roots of a determinantal equation involving the restricted and unrestricted least squares residuals, are considered. First, the exact distribution of these roots are characterized, and pivotal special cases are discussed. Second, an identifiable ex-post parameter is introduced that embeds the impact of left-out explanatory variables, in which case usual tests remain valid from a conservative perspective. Third, exact simultaneous confidence bands are derived for a vector of regression coefficients, or for a combination thereof. Fourth contribution pertains to asset pricing, with a focus on risk-return analysis. The distinction between the alpha (risk-adjusted outperformance) and beta (systematic risk component) of returns is revisited, from the ex-post parametrization perspective. Empirically, the risk-return profile of catastrophe bond mutual funds is analyzed.
