Machine learning models predicting returns: Why most popular performance metrics are misleading and proposal for an efficient metric

Numerous machine learning models have been developed to achieve the ‘real-life’ financial objective of optimising risk/return profile investment strategies. In current article: (a) we present and classify most popular performance metrics used in 190 articles analysed. We noticed that, articles, no attention is devoted criteria compare algorithms. (b) evaluate ability literature assess efficiency algorithms improve investments results. demonstrate that many metrics, like mean squared error (MSE) or root (RMSE), are inappropriate for this purpose while others, accuracy F1, just weak. explain why risk-adjusted return-based best-in-class, although they suffer from statistical limitations do not allow easy comparison across assets over time. (c) propose a new discriminant metric measures AI optimize return, which statistically more robust, can test effectiveness stability time assets.