ABSTRACT
In this article, we study the forecasting power of 12 different systemic risk measures on the macroeconomic shocks in China. We employ the FQGLS estimation with structural break. The violation of classical assumptions is detected, and the significant difference between OLS and FQGLS estimations further highlights the importance of model specification. The combined forecasts significantly outperform the historical mean in out-of-sample predictions, although most of the individual forecasts cannot. That is, the macroeconomic shock is predictable by the systemic risk measures, but the noise overwhelms the signal coming from real systemic risk.
Notes
1. We introduce the selected systemic risk measures in detail in appendix.
2. VOLATILITY is constructed on a monthly basis. TED starts in October 2006 and CREDIT starts in December 2007.
3. In the robustness verification, we also consider macroeconomic shocks measured by innovations to some more specific indicators, such as industrial production growth (IP), credit growth, inflation, and real estate investment (REI). Results are shown in Appendix.
4. We also construct the systemic risk measures using the US data as shown in Appendix.
5. We also test the predictive power of systemic risk measures in the United States and the results are shown in appendix.