We emphasized that data is a necessary driver of successful ML applications, but that domain expertise is also crucial to inform strategic direction, feature engineering and data selection, and model design.

In any domain, practitioners have theories about the drivers of key outcomes and relationships among them. Finance stands out by the amount of relevant quantitative research, both theoretical and empirical. Marcos López de Prado and others (see GitHub for references https://github.com/PacktPublishing/Hands-On-Machine-Learning-for-Trading) criticize most empirical results given pervasive data mining that may invalidate the findings. Nonetheless, a robust understanding of how financial markets work exists and should inform the selection and use of data as well as the justification of strategies that rely on machine learning. We outlined key ideas in Chapter 4, Alpha Factor Research and Chapter 5, Strategy Evaluation.

On the other hand, novel ML techniques will likely uncover new hypotheses about drivers of financial outcomes that will inform ML theory and should then be independently tested.