A semi-parametric dynamic conditional correlation framework for risk forecasting

, , , , In this study, we enhance the accuracy of Value-at-Risk (VaR) and Expected Shortfall (ES) forecasting by incorporating realized measures as exogenous variables into the existing framework developed by Gerlach and Wang (2020). Our analysis focuses on low-frequency daily data, with plans to extend the Semi-DCC model to include high-frequency realized measures for further improvement. We also explore more complex parameterizations, such as the AG-DCC model, which could open up avenues for alternative empirical frameworks. The potential application of Semi-DCC models in tail risk hedging scenarios is considered, particularly for assets with extreme volatility dynamics. Our asset allocation exercise yields promising results; however, further investigation into tail risk hedging applications falls outside the scope of this paper and is earmarked for future research. We disclose no conflicts of interest and acknowledge financial support under the National Recovery and Resilience Plan. Data for this study was sourced from https://finance.yahoo.com/, with all supporting data available in the supplementary materials. Our findings contribute to a novel semi-parametric dynamic conditional correlation framework for risk forecasting that shows promise in improving portfolio risk management strategies.

Introduction

Value-at-Risk (VaR) and Expected Shortfall (ES) are widely used risk measures in financial markets for evaluating portfolio risk. However, their accuracy has been a topic of debate among researchers due to the limitations of traditional methods in capturing extreme market events. In this study, we aim to enhance the accuracy of VaR and ES forecasting by incorporating realized measures as exogenous variables into the existing framework developed by Gerlach and Wang (2020).

The Need for Improved Risk Measures

Traditional methods for estimating VaR and ES rely on historical data, assuming that future market conditions will be similar to past ones. This approach fails to capture sudden changes or shifts in volatility, leading to inaccurate risk estimates. As a result, investors may not adequately protect their portfolios against extreme market events.

To address this issue, researchers have turned towards incorporating realized measures into risk forecasting models. Realized measures use high-frequency data to capture short-term volatility dynamics and provide more accurate estimates of future volatility compared to traditional methods.

The Semi-DCC Model

In this study, we focus on low-frequency daily data and use the Semi-DCC model proposed by Gerlach and Wang (2020). This model is an extension of the Dynamic Conditional Correlation (DCC) model that allows for time-varying correlation between assets.

The Semi-DCC model incorporates realized measures as exogenous variables into the DCC framework, allowing for better estimation of conditional correlations between assets. The inclusion of these variables improves the accuracy of VaR and ES forecasts by capturing short-term volatility dynamics that are missed by traditional methods.

Potential Applications

One potential application of our findings is in tail risk hedging scenarios. Tail risks refer to extreme market events that occur with low probability but can have a significant impact on portfolio returns. By incorporating realized measures into risk forecasting, our model can better capture these tail risks and provide more accurate estimates of portfolio risk.

We also conducted an asset allocation exercise using our model and found promising results. However, further investigation into tail risk hedging applications falls outside the scope of this paper and is earmarked for future research.

Data and Methodology

For this study, we sourced data from https://finance.yahoo.com/ for a selection of assets. We used daily returns data over a period of five years to estimate VaR and ES using both traditional methods and our proposed Semi-DCC model.

To evaluate the performance of our model, we compared its forecasts to those generated by traditional methods such as Historical Simulation (HS) and RiskMetrics (RM). We also explored more complex parameterizations, such as the Asymmetric GARCH-DCC (AG-DCC) model, which could potentially improve the accuracy of VaR and ES forecasts even further.

Results

Our findings show that incorporating realized measures into risk forecasting improves the accuracy of VaR and ES estimates compared to traditional methods. The Semi-DCC model outperformed HS and RM in terms of backtesting results, indicating its potential in capturing extreme market events.

Furthermore, our analysis suggests that including high-frequency realized measures in addition to low-frequency daily data could lead to even better results. This highlights the need for further research on extending the Semi-DCC framework to incorporate high-frequency data.

Conclusion

In conclusion, our study contributes to a novel semi-parametric dynamic conditional correlation framework for risk forecasting that shows promise in improving portfolio risk management strategies. By incorporating realized measures into the existing DCC framework, we were able to enhance the accuracy of VaR and ES forecasts for low-frequency daily data.

Our findings have implications for investors looking to protect their portfolios against extreme market events. Incorporating realized measures into risk forecasting can provide more accurate estimates of portfolio risk, leading to better risk management strategies.

We acknowledge that our study has limitations, such as the use of only low-frequency data and a limited number of assets. Future research could explore the application of our model in tail risk hedging scenarios and extend it to include high-frequency realized measures for even better results.