The absorption and multiplication of uncertainty in machine-learning-driven finance

Uncertainty about market developments and their implications characterize financial markets. Increasingly, machine learning is deployed as a tool to absorb this uncertainty and transform it into manageable risk. This article analyses machine-learning-based uncertainty absorption in financial markets by drawing on 182 interviews in the finance industry, including 45 interviews with informants who were actively applying machine-learning techniques to investment management, trading, or risk management problems. We argue that while machine-learning models are deployed to absorb financial uncertainty, they also introduce a new and more profound type of uncertainty, which we call critical model uncertainty. Critical model uncertainty refers to the inability to explain how and why the machine-learning models (particularly neural networks) arrive at their predictions and decisions-their uncertainty-absorbing accomplishments. We suggest that the dialectical relation between machine-learning models' uncertainty absorption and multiplication calls for further research in the field of finance and beyond.