Uncertainty-Aware Lookahead Factor Models for Improved Quantitative Investing

On a periodic basis, publicly traded companies are required to report fundamentals: financial data such as revenue, earnings, debt, etc., providing insight into the company’s financial health. Quantitative finance research has identified several factors—computed features of the reported data—that have been demonstrated in retrospective analysis to outperform market averages. In this paper, we first show through simulation that if we could (clairvoyantly) select stocks using factors calculated on future fundamentals (via oracle), then our portfolios would far outperform a standard factor approach. Motivated by this analysis, we train MLP and LSTM neural networks to forecast future fundamentals based on a trailing window of five years. We propose lookahead factor models to act upon these predictions, plugging the predicted future fundamentals into traditional factors. Finally, we incorporate uncertainty estimates from both neural heteroscedastic regression and a dropout-based heuristic, demonstrating gains from adjusting our portfolios to avert risk. In a retrospective analysis using an industry-grade stock portfolio simulator (backtester), we show simultaneous improvement in annualized return and Sharpe ratio (a common measure of risk-adjusted returns). Specifically, the simulated annualized return for the uncertainty-aware model is 17.7% (vs 14.0% for a standard factor model) and the Sharpe ratio is 0.84 (vs 0.52).