/ October 24, 2019
Two disciplines familiar to econometricians, factor analysis of equities returns and machine learning, have grown up alongside each other. Used in tandem, these fields of study can build effective investment-management tools, according to City University of Hong Kong’s Guanho Feng (a graduate of Chicago Booth’s PhD Program), Booth’s Nicholas Polson, and Booth PhD candidate Jianeng Xu.
The researchers set out to determine whether they could create a deep-learning model to automate the management of a portfolio built on buying stocks that are expected to rise and short selling those that are expected to fall, known as a long-short strategy. They created a machine-learning algorithm that built a long-short equity portfolio from the top and bottom 20 percent of a 3,000-stock universe.
They ranked the equities using the five-factor model of Chicago Booth’s Eugene F. Fama and Dartmouth’s Kenneth R. French. Fama and French break down the components of stock returns over time into five factors: market risk, in which stocks with less risk relative to their benchmark outperform those with more risk; size, in which companies with small market capitalizations outperform larger companies; value, where a low price-to-book ratio outperforms high; profitability, where higher operating profits outperform; and reinvestment, in which companies that reinvest outperform those that don’t.
Deep learning is a form of machine learning that is typically based on an artificial neural network, so it mimics human thinking and relies on sensing and analyzing patterns and conditions rather than following task-based rules. Although the researchers’ deep-learning system started with the five factors, it quickly found others that were relevant to its task. So the system added characteristics such as dividend yield, leverage, liquidity, bid-ask spread, and even macroeconomic conditions. Using these deep-learning factors, the researchers’ portfolio outperformed a cap-weighted portfolio drawn from 3,000 stocks between the years 2011 and 2017.
“Our method is the first one that unifies all procedures of the characteristics-based asset pricing models with a clear optimization objective,” the researchers write.
The research further validates factor analysis of stock returns and offers the promise that if deep learning is set on the right path and given enough computational power, it can productively manage portfolios and gain new insights as conditions change. In this study, Feng, Polson, and Xu’s framework was able to generate superior returns not just to the market but also to more static approaches that relied on either the capital asset pricing model or the five Fama and French factors alone.
Implications of the research may not be limited to portfolio management. If deep learning can identify factors that are relevant to returns under certain market conditions, it may be able to assist in capital budgeting and product management. Such a system could inform corporate executives when it is time to pay down debt or increase spending on research and development.
Deep learning may have a lot to say about how important decisions, now left to humans, wind up getting made.
Eugene F. Fama and Kenneth R. French, “A Five-Factor Asset Pricing Model,” Journal of Financial Economics, April 2015.
Guanho Feng, Nicholas G. Polson, and Jianeng Xu, “Deep Learning in Asset Pricing,” Working paper, February 2019.
This article was originally published on ChicagoBoothReview