Time‐Series Momentum in Credit: Machine Learning Approach

Artificial Intelligence (AI) and Machine Learning (ML) has seen unprecedented growth over the past decade with applications across healthcare, robotics, data security and automotive industries to name a few. Penetration of AI in finance has been slower; primarily due to the intrinsic nature of markets – non‐stationarity of financial processes and lack of sufficient training data. We make a foray into applying ML to finance by training a range of models to trade Momentum – a systematic strategy that benefits from persistence of trends in markets. Our findings suggest that a class of ML algorithms like Random Forests and Neural Nets produce Sharpe ratios close to optimal vanilla time‐series momentum (TSMOM). However, correlation between ML and vanilla TSMOM signals is low allowing us to harvest fruits of diversification by creating hybrid vanilla – ML TSMOM strategies. We find such hybrid portfolio Sharpe ratios to be twice as high as vanilla TSMOM.