Crisis Period Forecast Evaluation of the DCC-GARCH Model
Yang Ding
Andrew Schwert

Dr. Emma Rasiel & Professor Aino Levonmaa, Faculty Advisors

Honors thesis submitted in partial fulfillment of the requirements for Graduation with Distinction in Economics in Trinity College of Duke University

Duke University
Durham, North Carolina
2010

Acknowledgements
We would like to thank Dr. Emma Rasiel and Professor Aino Levonmaa for their invaluable direction, patience, and guidance throughout this entire process.

Abstract
The goal of this paper is to investigate the forecasting ability of the Dynamic Conditional Correlation Generalized Autoregressive Conditional Heteroskedasticity (DCC-GARCH). We estimate the DCC’s forecasting ability relative to unconditional volatility in three equity-based crashes: the S&L Crisis, the Dot-Com Boom/Crash, and the recent Credit Crisis. The assets we use are the S&P 500 index, 10-Year US Treasury bonds, Moody’s A Industrial bonds, and the Dollar/Yen exchange rate. Our results suggest that the choice of asset pair may be a determining factor in the forecasting ability of the DCC-GARCH model.

I. Introduction
Many of today’s key financial applications, including asset pricing, capital allocation, risk management, and portfolio hedging, are heavily dependent on accurate estimates and well-founded forecasts of asset return volatility and correlation between assets. Although volatility and correlation forecasting are both important, however, existing literature has dealt more closely with the performance of volatility models – only very recently has the issue of correlation estimation and forecasting begun to receive extensive investigation and analysis. The goal of this paper is to extend research that has been undertaken regarding the forecasting ability of one specific correlation model, the Dynamic