Machine-learning algorithms are increasingly reshaping the trading of securities and management of investments in the financial markets. Adopting adaptive algorithms programmed to discover patterns and anomalies in large datasets, financiers hope to gain a profitable edge on competitors or detect undesired risk-exposures. The spread of algorithms in data- and model-driven finance challenges existing practices of model use and creates a demand for practical solutions for how to manage the complexity these devices add to the financial market system. Drawing on interviews with market participants developing and using machine learning algorithms for trading and investment management purposes, this paper identifies the main problems associated with the adoption of machine learning techniques in finance and examine specific solutions proposed by practitioners. The study demonstrates that practitioners use a simple heuristics, Ockham’s razor, to manage model complexity, increase model accuracy, and reduce the risk of errors. It furthermore shows that the razor principle also serves to retain an element of human judgment in an otherwise often largely automated decision-making process. The study contributes to the burgeoning scholarship on the role of automation, big data, and adaptive models in the present and future of work.
Kristian Bondo Hansen (Tue,) studied this question.