Composite Uncertainty Indicators and Stock Market Returns: Based on Supervised Dimension Reduction Techniques

Abstract

This study develops two composite uncertainty indicators for China (U^SPCA and U^PLS) by employing scaled principal component analysis (SPCA) and partial least squares regression (PLS) as dimensionality reduction techniques to synthesise critical information from multiple uncertainty indices. This study subsequently evaluates the effectiveness of these techniques in predicting returns in China's stock markets. Empirical findings demonstrate that the SPCA and PLS methodology substantially enhances stock return forecasting across in-sample and out-of-sample tests while generating meaningful economic benefits for mean-variance optimised portfolios. Furthermore, both U^SPCA and U^PLS outperform individual uncertainty indices and conventional economic predictors in predictive capability. In particular, the forecasting power of the two composite indicators is stronger during bear market phases than under bull market conditions. The analysis also reveals that geopolitical events such as the Russia–Ukraine conflict can temporarily reduce the predictive efficacy of uncertainty-based indicators for stock returns.