ABSTRACT
In this study, we forecast the realized volatility (RV) of the Chinese stock market using the heterogeneous autoregressive (HAR) model and various extended models. To extract the volatility information from the G7 stock markets, we employ a newly proposed approach, the scaled principal component analysis (SPCA), to produce a diffusion index and extend the HAR benchmark (HAR-SPCA). To validate the effectiveness of the SPCA approach, we employ three other dimension reduction approaches, the kitchen sink model, and five popular forecast combinations to deal with multivariate information and make competing forecasts. The results suggest that the combined volatility information from the G7 stock markets significantly predicts Chinese stock market volatility. More importantly, the forecasts from the HAR-SPCA model are steadily more accurate than the benchmark and other competing models under various evaluation criteria. Finally, our results are persistent to various robustness checks and the evaluation of portfolio performance.
Acknowledgments
We are grateful to Paresh Narayan (the editor), a subject editor, and two anonymous referees for insightful comments that significantly improved the paper. This work is supported by the National Natural Science Foundation of China [72001110, 72001109, 72171139, and 71701118], the Fundamental Research Funds for the Central Universities [30919013232], the Research Fund for Young Teachers of School of Economics and Management, NJUST [JGQN2009].
Disclosure Statement
No potential conflict of interest was reported by the author(s).
Supplementary Material
Supplemental data for this article can be accessed on the publisher’s website
Notes
1. We are grateful to the Subject Editor for providing a valuable suggestion of analyzing whether the G7 markets possess higher predictive content than the BRIS (Brazil, Russia, India, and South Africa) markets. However, we cannot obtain the high-frequency data of South Africa and Russia. Nevertheless, we will conduct such empirical work in future research if we can obtain these data.
2. To save space, we provide the econometric specifications of other models in the Internet Appendix.
3. It is important to note that June 1, 2009 is the first day on which we can obtain the Italian stock market index.
4. The data are available at https://realized.oxford-man.ox.ac.uk/. The construction of the data does not suffer from look-ahead bias and thus we can use the data to make real-time forecasts.
5. We do not report in-sample estimation results for combination approaches because all these approaches are implemented based on out-of-sample forecasts generated from multiple individual models.
6. The sample split for constructing the in-sample estimation window and the out-of-sample forecasting window needs to be rational. The size of the in-sample window should be large enough for making accurate parameter estimation and volatility forecasts. However, the out-of-sample period should also be long enough for obtaining valid evaluation results. In the robustness checks, we further use the last 1500 observations to evaluate the predictive performance in avoiding contingency.
7. For each period, the construction of the four diffusion indices is based on the recursive in-sample observations in eliminating look-ahead bias.
8. All of the MCS p-values reported in this study are generated based on the range statistic. Similar results can be obtained when we use the semi-quadratic statistic. Thus, the results based on semi-quadratic statistic are omitted for brevity, but are available upon request. The evaluation results are consistent when we use the significance (confidence) level of 10% (90%).
9. Specific empirical results and their corresponding discussions are provided in the Internet Appendix because of space limitation.