Flexible Time-Varying Betas in a Novel Mixture Innovation Factor Model with Latent Threshold

This paper introduces a new methodology to estimate time-varying alphas and betas in conditional factor models, which allows substantial flexibility framework. To circumvent problems associated with the previous approaches, we introduce Bayesian parameter model where innovations of state equation have spike-and-slab mixture distribution. The distribution specifies two states specific probability. In first state, innovation variance is set close zero certain probability parameters stay relatively constant. second large change normally distributed mean given variance. latent specified threshold that governs change. We allow separate for each parameter; thus, may shift an unsynchronized manner such moves from one another when exceeds vice versa. approach offers great nests plethora other specifications, allowing us assess whether models evolve gradually over time or display infrequent, but large, shifts. apply proposed industry portfolios within five-factor setting show Capital Asset Pricing Model (CAPM) provides robust beta estimates coupled smaller pricing errors compared alternative approaches. results significant implications implementation smart strategies rely heavily on accuracy stability yields.