Machine learning techniques are now very common in many spheres, and there is a growing popularity of these approaches in macroeconomic forecasting as well. Are these techniques really useful in the prediction of macroeconomic outcomes? Are they superior in performance compared to their traditional counterparts? We carry out a meta-analysis of the existing literature in order to answer these questions. Our analysis suggests that the answers to most of these questions are nuanced and conditional on a number of factors identified in the study.
