Boston College - Assistant Professor of Finance
Stochastic volatility models (hereafter SVOL) are noticeably harder to estimate than standard time-varying volatility models. Recent advances in the literature now allow for efficient estimation and prediction for a basic univariate SVOL model. However, the basic model may be insufficient for numerous economic and finance applications. In this paper, we develop a bivariate model to allow to allow for: the leverage effect through a correlation between observable and variance errors; and fat-tails in the conditional distribution. We provide an algorithm to analyze a multivariate factor model with stochastic volatility. Our methods simulataneously perform finite sample inference, smoothing and prediction. We also provide the researcher with a range of model diagnostics including how to identify outliers for stochastic volatility models and how to select between different model specifications. We apply some of these extensions to financial time series. We find (1) strong evidence of non-normal conditional distributions for all series, and (2) correlated errors for stock returns. We illustrate the sensitivity of parameter inference, e.g. the volatility persistence, and predictive inference to different model specifications. Our result has policy implications on decisions based upon prediction of volatility, such as asset allocation, risk management and option pricing.
This is joint work with Nick Polson and Peter E. Rossi.