Embedded Files
Systematic Asset Pricing Engine
An institutional-grade, object-oriented framework for active portfolio management rooted in the Grinold & Kahn Fundamental Law. Unlike standard backtesters, this engine enforces strict Point-in-Time (PIT) logic to eliminate look-ahead bias, utilizing a custom ETL pipeline to merge ragged CRSP and Compustat data. It features a Fundamental Factor Risk Model (Fama-MacBeth), Gram-Schmidt alpha purification, and a convex optimizer that solves for maximum Information Ratio while controlling turnover.
Tech Stack: Python, Pandas, NumPy, Statsmodels, SciPy.
Figure 1: Cumulative Active Return of Financial Constraints Signal (1995–2024)
NLP-Driven Alpha: Financial Constraints Factor
NLP-Driven Alpha: Financial Constraints Factor
Extracting latent constraints signals from 10-K filings using Large Language Models.
Extracting latent constraints signals from 10-K filings using Large Language Models.
A next-generation alpha factor designed to outperform traditional accounting-based indices (like Whited-Wu) by analyzing unstructured corporate disclosures. While standard metrics rely on backward-looking financial ratios, this model applies Generative AI and Large Language Models (LLMs) to 10-K/10-Q filings to capture the nuanced "context" of management sentiment.
The methodology utilizes a hybrid two-stage process: first, Generative LLMs create high-fidelity ground-truth labels by analyzing MD&A sections for complex thematic distress signals (e.g., covenant risks, liquidity friction). Second, these labels train a scalable XGBoost classifier to score the entire investment universe. The result is a high-frequency "Financial Constraint Score" that serves as a potent overlay for Value and Quality strategies.
Tech Stack: Python, PyTorch (Transformers), XGBoost, SEC EDGAR API.
Dynamic Hedging & Market Risk Simulator
Dynamic Hedging & Market Risk Simulator
A discrete-event simulation engine designed to stress-test hedging strategies under extreme liquidity constraints. Unlike static "Day-1 Greek" models, this framework analyzes path-dependent risks, specifically the "Convexity Trap"—the structural P&L bleeding that occurs when hedging long-dated liabilities with short-dated assets. The engine models endogenous transaction costs (dynamic bid-ask spreads based on volatility regimes), margin loan funding topologies, and the efficacy of "Lazy Hedging" (risk limits) during market crashes.
Tech Stack: Python, NumPy, SciPy, Matplotlib.
Page updated
Google Sites
Report abuse