We demonstrate how a machine learning algorithm can be applied to predict and explain modern market microstructure phenomena. We investigate the efficacy of various microstructure measures and show that they continue to provide insights into price dynamics in current complex markets. Some microstructure features with apparent high explanatory power exhibit low predictive power, and vice versa. We also find that some microstructure-based measures are useful for out-of-sample prediction of various market statistics, leading to questions about the efficiency of markets. Our results are derived using 87 of the most liquid futures contracts across all asset classes.
Read full abstract