We propose a new semiparametric observation-driven volatility model where the form of the error density directly influences the volatility dynamics. This feature distinguishes our model from standard semiparametric GARCH models. The link between the estimated error density and the volatility dynamics follows from the application of the generalized autoregressive score framework of Creal, Koopman, and Lucas (2012). We provide simulated evidence for the estimation efficiency and forecast accuracy of the new model, particularly if errors are fat-tailed and possibly skewed. In an application to equity return data we find that the model also does well in density forecasting.
# 12-055/2/DSF35 (2012-05-22)
- Jiangyu Ji, VU University Amsterdam; Andre Lucas, VU University Amsterdam, and Duisenberg school of finance
- volatility clustering, Generalized Autoregressive Score model, kernel density estimation, density forecast evaluation
- JEL codes:
- C10, C14, C22