|
Abstract:
|
Machine learning (ML) methods have become popular in predicting stock returns and forming hedging portfolios in finance. However, the majority of literature focuses on conditional mean predictions only. In this paper, we propose to estimate quantile regression (QR) through ML methods with hyperparameters tuned automatically through Gaussian process (GP). As a by-product of QR estimation, our proposed methods can predict the whole conditional density of stock returns in panel data. In the simulation studies, we examine the performances of prediction using modes of the predictive conditional densities, and hedging portfolios constructed using quantiles of the predictive conditional densities. We find that, in presence of noisy and skewed data, mode prediction enjoys a significant improvement over mean prediction, and hedging portfolio formation based on quantiles is more profitable than the traditional approach based on regression means. A real data application in the US stock market is implemented.
|
