• Abstract: We introduce a conditional machine learning approach to forecast the stock index return. Our approach is designed to work well for short-horizon forecasts to address the well-documented instability in predicting aggregate stock returns in long panels. We formally characterize the forecast standard errors to assess the uncertainty associated with our cross-sectional neural network predictions, which also enables us to explain the predictability of our model. The explainability covers both correctly and incorrectly assumed forecasting models, and stems from the forecast standard errors and out-ofsample R square. To explain the economic impacts of the economy’s stability on the forecast quality, we introduce a ``CDI" index defined as the correlation between firms’ market value share and sales share, and show that it can well explain the forecast uncertainties, thus provides economic insights of the success and failure of machine learning based forecasting models.

• Paper: pdf file