Forecasting Stock Return Volatility Based On Data-rich Environment

The prediction of stock return volatility plays a crucial role in many financial fields,such as portfolio allocation,risk management and asset pricing.Therefore,forecasting financial volatility in the financial market has received more and more attention.Understanding the robustness and importance of the relationship between a set of variables and volatility is an important empirical problem in finance.We explore new forecasting methods and new forecasting factors to improve the accuracy of forecasting stock return volatility and get more accurate stock return volatility prediction.First,the paper introduces the explanatory variables including macroeconomic variables,technical indicators,market and investor sentiment indexes and oil volatility indexes,and then expounds the basic autoregressive model,popular shrinkage method,partial least squares method.The in-sample test was carried out according to the heteroscedastic consistency statistics of regression parameters,and the out-of-sample test was performed using statistics and corresponding P-values.In addition,the economic utility benefits of these models are compared from an economic point of view.Secondly,using the idea of dimensionality reduction,we forecast stock return volatility in a data-rich environment based on three popular shrinkage methods,namely elastic net,lasso and ridge regression.The shrinkage method uses the absolute shrinkage operator and the selection operator to solve the problem of parameter instability and model uncertainty in the data-rich environment,to provide a more accurate forecast of stock return volatility.Out-of-sample results show that the shrinkage method exhibits superior performance in terms of Roos2 statistics relative to the autoregressive model,univariate regression and the five combination methods,and mean-variance investors can achieve significant economic gains,and our results are robust to various checks.Thirdly,from the perspective of creating new predictors,the idea of dimensionality reduction is also used to improve the forecasting ability of stock return volatility.Therefore,we use the partial least squares method to construct a real but unobservable factor to connect a series of variables and reasonably explain the stock return volatility.Empirical results show that the new indicators have strong in-sample predictive performance,while out-of-sample predictive power is better than existing indicators,consistent with the in-sample results.Specifically,extending the application of the new index to portfolio allocation,mean-variance investors reap substantial economic gains.Furthermore,our results are robust to different economic cycles,various risk aversion coefficients and transaction costs.Finally,the new prediction methods and new predictors are summarized.