388. Machine Learning in Finance: Supply Chain Asset Pricing and Data-Driven Dynamic Risk Management
Invited abstract in session MC-1: Agostino Capponi, stream Keynotes.
Monday, 12:30-14:00Room: Great Hall
Authors (first author is the speaker)
| 1. | Agostino Capponi
|
| Columbia University |
Abstract
We develop new methodologies that integrate machine learning and AI techniques to address the two most challenging problems in financial engineering, specifically asset pricing and risk management.
Our asset pricing approach introduces a novel data-driven model that captures firm-level characteristics across local sub-graphs of the supply chain. By traversing the supply chain, the model learns how neighboring firm characteristics predict outcomes for a target firm, combining these characteristics non-linearly to create interpretable, tradeable asset pricing factors. These supply chain-derived factors are uncorrelated with those from traditional models and demonstrate positive out-of-sample Sharpe ratios, with most contributions coming from the top five factors. Our findings suggest that combining supply chain and firm characteristics systematically captures unique risks beyond firm characteristics alone.
For risk management, we construct a dynamic framework for high-dimensional factor models, using diffusion maps to uncover the underlying dynamics of common factors in a fully data-driven, nonparametric way. This framework generates low-dimensional embeddings that retain most explanatory power for time series of risk factors. Applied to equity portfolio stress testing across three major financial crises, we demonstrate that our approach outperforms standard methods, which often assume a zero conditional expectation for unstressed factors.
Keywords
- Machine Learning
- Supply Chain Management
- Risk Analysis and Management
Status: accepted
Back to the list of papers