Multi-Horizon Echo State Network Prediction of Intraday Stock Returns

Abstract

Stock return prediction is a problem that has received much attention in the finance literature. In recent years, sophisticated machine learning procedures have been shown to perform significantly better than conventional prediction techniques. One downside of these approaches is that they are often numerically expensive to implement, for both training and inference, because of their high complexity. We propose a return prediction framework for intraday returns at multiple horizons based on Echo State Network (ESN) models, wherein a large subset of parameters is drawn at random and never trained. The coefficients that do require training can be fitted by explicitly solving a rolling-window, ridge-regularized least squares problem. Our approach enjoys the flexibility of machine learning methods, has an inherently efficient implementation, and shows strong forecasting performance.