By Eric Ghysels, Edward M. Bernstein Distinguished Professor of Economics, UNC-Chapel Hill and professor of finance, UNC Kenan-Flagler Business School
Artificial intelligence enhancements are increasingly shaping our financial decision-making. But with what result? As part of research my colleagues and I conduct, we explore strategies commonly used in the robo-advising industry, including some involving advanced machine learning methods, to assess the potential benefits of robo-investing over a long period of time for a heterogeneous panel of individual investors. This man-versus-machine comparison has shed some light on potential benefits the emerging robo-advising industry may provide to certain targeted segments of the population, such as low-income investors and/or investors with relatively little financial literacy.
The research, conducted jointly with Catherine D’Hondt and Rudy De Winne of the Université Catholique de Louvain and Steve Raymond of UNC-Chapel Hill, introduces the notion of artificial intelligence alter egos – or AI alter egos – to study the benefits of robo-advising. To illustrate, let’s look at machine learning (ML) advances in other fields, such as literature and music. Today, an ML text mining algorithm can analyze the writings of a famous author and compose entirely new literature in the style of the writer it was exposed to and trained on. The same can be done with music. Franz Schubert started his Unfinished Symphony in B minor in 1882, but wrote only two complete movements, though he lived another six years. Now, deep-learning ML has produced a completed version of the entire symphony. We can characterize this as Schubert’s AI alter ego composing new scores.
Would Schubert have done better than his AI alter ego? I’ll leave that debate to the musicologists, but it’s fair to say it would probably be hard to address the question. On the other hand, it’s much easier to apply the notion of AI alter egos in a setting where comparing the outcomes of human and AI alternatives is more straightforward – such as in financial investments.
In our research, my colleagues and I use a unique data set covering brokerage accounts for a large cross-section of 22,972 individual investors from January 2003 to March 2012 (which included the 2008 financial crisis), to assess the benefits of robo-investing. The study includes detailed data on investor characteristics and records of all trades, and explores robo-investing strategies commonly used in the industry, including some involving advanced machine learning methods.
The stocks and ETFs traded by the study’s investors have international coverage. Although the study focuses on Belgian individual investors, most of their trading activities pertain to foreign (roughly a quarter U.S.-based) stocks.
Many investment brokerage firms are now targeting individuals with modest savings, as general consensus holds that smaller investors don’t get the investment advice they need. In fact, 71 percent, or almost 90 million American families, have investment account balances worth less than $100,000. The growth of the automated investment advisory niche fills a need for such investors.
In terms of annual net income, about 70 percent of the investors in the study declare an income between 20,000 and 75,000 euros. Only 3.36 percent earn more than 150,000 euros per year. The mean portfolio value in our sample is 29,243.88 euros and the average investor is about 48 years old. Consistent with the literature, investors in our sample are under-diversified: the average investor holds a five-stock portfolio, and the median investor a three-stock portfolio. Thus, the data set consists of the types of investors typically targeted by robo-advising.
The study considers three investment strategies. Two are based on a Markowitz mean-variance (MV) scheme, and the third is based on an equally weighted portfolio scheme over a two-year trailing sample.
The two MV strategies distinguish themselves via the sophistication of the conditional mean and variance estimates. The first involves two-year rolling sample estimates for both the mean and variance. For the second, we rev up the robot engines and replace the rolling sample estimators in the Markowitz portfolio allocation by respectively expected return predictions using machine learning algorithms and more sophisticated conditional covariance estimators.
The overall findings suggest that the AI alter ego robo-investors involving equal weighting or rolling sample mean and variance perform poorly and are of little value to any of the study’s investors. In contrast, the machine learning MV AI alter egos result in significant investment portfolio performance improvements for certain types of investors – in particular, low-income (low-education) investors and those featuring high-risk aversion. These results confirm the claims made by industry practitioners, and the promise AI holds for the future of the fintech industry.
More intriguing, and somewhat unexpected, are the results pertaining to performance during the financial crisis, which show that robo-investors outperformed a large swath of investors. The median robo-investor moved assets into cash because of negative expected returns using AI during the crisis. On the other hand, individual investors’ behavioral biases, in particular with respect to disposition, often resulted in unfortunate consequences during the crisis’ onset.
Download the paper in full, here.