Is the Federal Reserve Causing Funds to Underperform? A causal machine learning analysis

Abstract: Macroeconomic conditions heavily influence financial markets, and the leading corpus of theory in the field lays out some general principles observed by investors and asset managers alike. However, while the theory is sound, it is hard to measure how much effect these conditions have. To better understand this scenario, this study uses a novel technique that combines traditional causal inference with machine learning techniques to measure macroeconomic policy effects in the financial market. Specific to this study, the double machine learning framework measures the average treatment effect of the US Federal Reserve System’s interest rate’s growth rate on fixed income and equity funds returns. The period studied is from January 1986 to December 2021, as it includes data from many relevant events that caused interest rates to change around the world. Furthermore, the data is separated into two main clusters, i.e., passively and actively managed funds, since finance theory indicates that the latter should be less affected by a central bank’s interest rate changes. As the double machine learning framework can use virtually any statistical learning procedure as a learner, two different techniques are tried in this study. Linear regression sets a baseline, and gradient boosting is used to assess what technique would produce better results. The results show some evidence of gradient boosting being a worthy technique for predicting returns and, therefore for the double machine learning procedure. The actively managed funds dataset results indicate that a 1% increase in the US Federal Reserve System’s interest rate led to a -11.97% decrease in actively managed fund returns. However, this effect must be considered alongside all others that might affect financial markets. The results indicate a rich field for future research that can serve investors and managers through data-driven decision-making.