Better Than Blume: An Approach to Forecasting Future Stock Betas with Linear Bayesian Models and Sample-Wide Priors