Asymptotic and Bayesian Confidence Intervals for Sharpe Style Weights By

Sharpe style regression has become a widespread analytic tool in the financial community. The style regression allows one to investigate such interesting issues as style composition, style sensitivity, and style change over time. All previous methods to obtain the distribution and confidence intervals of the style coefficients are statistically valid only in the special case in which none of the true style weights are zero or one. In practice it is quite plausible to have zero or one for the values of some style weights. In this paper we apply new results of Andrews (1997a, 1999) and develop a comparable Bayesian method to obtain statistically valid distributions and confidence intervals regardless of the true values of style weights.