Investor sentiment based on scaled PCA method: A powerful predictor of realized volatility in the Chinese stock market

Abstract

In this study, we construct an investor sentiment indicator (SsPCA) to predict stock volatility in the Chinese stock market by applying the scaled principal component analysis (sPCA). As a new dimension reduction technique for supervised learning, sPCA is employed to extract useful information from six individual sentiment proxies and obtain the common variations to characterize the investor sentiment (SsPCA). The empirical results indicate that SsPCA is a significant and powerful volatility predictor both in and out of sample. We also employ the partial least squares (PLS)-based investor sentiment index, three extra sentiment measures in past studies, and six individual sentiment proxies for comparison, and find SsPCA outperforms them on predicting stock volatility in the Chinese stock market. More importantly, the predictability of SsPCA remains significant before and after the famous financial crises (the sub-prime mortgage crisis and Chinese stock market turbulence) and the spread of the pandemic (COVID-19). Additionally, our findings imply that SsPCA still plays an essential role in predicting sock volatility after considering the leverage effect. The robustness of SsPCA in volatility forecasting is further verified in various industry indices of the Chinese stock market. Finally, we state that the strong predictability of SsPCA is highly related to its dimensionality reduction. Our results indicate that SsPCA is a robust volatility predictor from various aspects and performs better compared with existing sentiment indicators.