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

Area of Science:

• Finance
• Computer Science
• Artificial Intelligence

Background:

• Financial markets are characterized by inherent uncertainty.
• Machine learning (ML) is increasingly used to manage financial uncertainty and risk.
• Existing research often focuses on ML's benefits without fully exploring its generated uncertainties.

Purpose of the Study:

• To analyze how machine learning absorbs uncertainty in financial markets.
• To investigate the novel uncertainties introduced by ML models in finance.
• To introduce and define the concept of "critical model uncertainty".

Main Methods:

• Qualitative analysis of 182 interviews within the finance industry.
• Specific focus on 45 interviews with professionals actively using ML in investment, trading, or risk management.
• Thematic analysis of interview data to identify patterns in ML application and uncertainty.

Main Results:

• Machine learning models effectively absorb certain types of financial uncertainty.
• The application of ML introduces a new, profound uncertainty termed "critical model uncertainty."
• Critical model uncertainty stems from the 'black box' nature of ML models, particularly neural networks, hindering explainability.

Conclusions:

• A dialectical relationship exists between ML's uncertainty absorption and its generation of critical model uncertainty.
• The explainability of ML models in finance is a critical area requiring further investigation.
• Findings call for expanded research into the implications of critical model uncertainty in finance and other fields.