Researchers Release v2 of LLM Counterfactual Stress Testing Portfolio Risk Simulation Paper

Masoud Soleimani released v2 of a 30-page arXiv paper proposing an auditable large language model macro-financial stress testing pipeline for portfolio risk simulation. The pipeline generates G7 macro scenarios including GDP growth, inflation, and policy rates through structured prompts and optional retrieval of national fundamentals and news, converts them to deviations from the IMF World Economic Outlook baseline, and maps them to portfolio losses via a three-channel PCA factor model; the macro-to-factor mapping is pre-specified rather than estimated, and all tail risk multiples are conditional on established assumptions. The study validates based on 2420 valid scenarios obtained from two model types and three runs, plus an unanchored severity run and 8 risk layer ablation experiments. The authors state that LLM scenario generators are better suited for narrative and industry-level scenario extraction under pre-specified severity paths, rather than as autonomous calibration tools for stress magnitude.

AI-generated from reports · 2 hr agoUpdated

Not enough consecutive observations yet; trend is not drawn.