Overview

In this talk, we consider forecasting in big data environments. We develop a category of high-dimensional nonlinear forecasting to benefit from many potential predictors while accounting for any possible nonlinear dynamics within the environment. Combining these two elements is a developing area of research. To overcome the curse of dimensionality and to manage data and model complexity, we suggest a shrinkage estimation of a backpropagation algorithm of a neural network with skip-layer connections including both linear and nonlinear structure. We apply this approach to forecast equity returns, and show that capturing nonlinear dynamics between equities significantly improves the quality of forecasts over current univariate and multivariate factor models.