Description:
Measuring, modeling, and forecasting volatility are of great importance in financial applications such as asset pricing, portfolio management, and risk management. In this paper, we investigate predictability of stock market volatility by macrofinance variables in a dynamic regression framework using latent thresholding. The latent threshold models allow data-driven shrinkage of regression coefficients by collapsing them to zero for irrelevant predictor variables and allowing for time-varying nonzero coefficients when supported by the data. This is a parsimonious framework which selects what potential predictor variables should be included in the regressions and when. We extend this model to allow for stochastic volatility for realized volatility innovations and discuss Bayesian estimation methods. We apply the models to monthly S&P 500 volatility and find that using macro-finance variables in volatility forecasts enhances model performance statistically and economically, particularly when we allow for dynamic inclusion/exclusion of these variables