The effect of uncertainty index based on sparse method on volatility prediction of stock market

The stock market’s volatility is intricately linked to the trajectory of the real economy and macroeconomic policies. This study focuses on the US S&P 500 index, utilizing five uncertainty indices: the US Economic Policy Uncertainty Index (EPU), the Global Economic Policy Uncertainty Index (GEPU), the Monetary Policy Uncer- tainty Index (MPU), the Global Geopolitical Risk Index (GPR), and the US Stock Market Volatility Index (EMV). It employs sparse machine learning models, including Lasso regression and Elastic Net regression, to assess the influence of economic policy uncertainty on S&P 500 index volatility, while also comparing out-of-sample predictions with various autoregressive models. The empirical findings indicate that: 1) the predictive efficacy of the sparse machine learning model surpasses that of the autoregressive model overall; 2) in periods of abnormal market volatility, the AR family model is susceptible to extreme values, whereas the predictive out- comes of the Lasso family model exhibit greater robustness; 3) the model confidence set (MCS) test results further substantiate the predictive superiority of the Lasso model following the inclusion of the uncertainty index. This discovery offers a novel approach to enhancing the precision of stock market volatility forecasting, holding significant reference value for policymakers and market stakeholders.