Whether dimensionality reduction techniques can improve the ability of sentiment proxies to predict stock market returns

In this study, we construct China's aggregate sentiment indicator (SsPCA) based on the method of Huang et al. (2021a), which employs a new dimension reduction method of scaled principal component analysis (PCA), to aggregate useful information from individual sentiment proxies, and further examine its return predictability for the Chinese stock market. The empirical evidence suggests that SsPCA significantly improves the prediction accuracy for stock market returns both in and out of the sample, and also obtains considerable economic gain for a mean-variance investor. Additionally, the forecasting effect of SsPCA is superior to that of SPCA and SPLS, evaluated using the traditional PCA and partial least square methods, respectively. Moreover, relative to the period of the bull market, SsPCA exhibits better ability in forecasting stock market returns during the bear market. Finally, special events, such as the outbreak of coronavirus disease 2019 (COVID-19), also affect the predictive performance of the sentiment indicator.

Automated summary

In this study, a China's aggregate sentiment indicator (SsPCA) is constructed using a new dimension reduction method. The empirical evidence shows that SsPCA significantly improves the prediction accuracy of stock market returns in and out of the sample, and also obtains considerable economic gain for investors. Compared to traditional methods, SsPCA outperforms in forecasting and performs better during bear markets. Moreover, special events like the outbreak of COVID-19 also impact the predictive performance of the sentiment indicator.