Research • Knowledge • Skills
Publications · external

Machine Learning and Artificial Intelligence in Market Risk Management: A Review of Contemporary Approaches

Abstract

Read abstract

<span>Market risk management—the practice of quantifying and mitigating potential losses from adverse movements in financial markets—has been fundamentally transformed by artificial intelligence and machine learning techniques. Drawing on the author's experience implementing risk frameworks at </span><strong><span>HSBC, Bank of America, and Citibank</span></strong><span>, this article examines how AI is reshaping the risk function at the world's largest banks. Traditional approaches, including Value-at-Risk (VaR) and historical simulation, rely on linear assumptions and backward-looking data that often fail to capture the complexity, velocity, and interconnectedness of modern financial markets. This article synthesizes advances across four complementary streams: (1) deep reinforcement learning for dynamic portfolio optimization under uncertainty; (2) large language models and their implications for systemic risk and market fragility; (3) machine learning approaches for forecasting market stress and volatility; and (4) the critical role of data infrastructure in enabling AI-driven risk management. We present quantitative comparisons of model performance and discuss the emerging challenge of AI-induced systemic risk—where coordinated actions from multiple AI systems could trigger simultaneous "buy" or "sell" signals, amplifying market crashes or bubbles. </span><strong><span>The bottom line in plain English:</span></strong><span> AI is making market risk management faster and smarter, but it's also creating a new danger—when every bank uses the same AI models trained on the same data, they all end up making the same mistakes at the same time. The very technology designed to protect banks could end up blowing them up together.</span><span></span>

Record information

Authors
Shweta Khosla
Publisher
Elsevier BV
DOI
10.2139/ssrn.6423478

Publication: