Working Paper: CEPR ID: DP4645
Authors: Massimo Guidolin; Allan G. Timmermann
Abstract: This Paper characterizes the term structure of risk measures such as Value at Risk (VaR) and expected shortfall under different econometric approaches including multivariate regime switching, GARCH-in-mean models with student-t errors, two-component GARCH models and a non-parametric bootstrap. We show how to derive the risk measures for each of these models and document large variations in term structures across econometric specifications. An out-of-sample forecasting experiment applied to stock, bond and cash portfolios suggests that the best model is asset- and horizon specific but that the bootstrap and regime switching model are best overall for VaR levels of 5% and 1%, respectively.
Keywords: nonlinear econometric models; simulation models; term structure of risk
JEL Codes: G12
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| choice of econometric model (C51) | estimated levels of risk (C13) |
| multivariate regime switching model (C32) | VaR levels at 5% (C29) |
| bootstrap method (Y20) | VaR levels at 1% (C58) |
| predictive performance of models (C52) | asset and horizon-specific (G19) |
| GARCH models (C58) | varying estimates of risk (D81) |