1-day 95% VaR
Calculated for a Treasury portfolio
$291,342
Source: ValueAtRisk.ipynb — README-stated value shown; verify by running the notebook if needed.
Finance Portfolio
Models, backtests, and research across rates, equities, crypto and derivatives.
Notebook-driven portfolio demonstrating yield-curve & fixed-income analysis, comparative VaR methods, and a FinBERT sentiment pipeline. Notebooks are linked below — quick runnable instructions included.
- Elevator: reproducible notebooks for risk & signal research — suitable for quant research or quant engineering teams.
- Why this matters: clear metrics, backtests, and diagnostic tooling to evaluate model reliability and production readiness.
Key results (quick)
News articles processed
FinBERT sentiment pipeline (sample)
6,250+
Source: SentimentAnalysis.ipynb
PCA for yield curve
Dimensionality reduction result
2 factors — 97% variance
Featured projects
Three focused notebooks: fixed income, market risk, and alternative-data sentiment. Each card links directly to the notebook for quick review.
Fixed Income — Nelson-Siegel & VaR
Objective: model US Treasury curves and compute portfolio VaR using factor decomposition and PCA.
- FRED data fetch (notebook contains fetch + modeling cells; FRED API key required to run fresh).
- Nelson-Siegel parameter estimation and factor decomposition (level, slope, curvature).
- VaR computed for a multi-bond portfolio; backtest cells included.
Result (notebook): 1-day 95% VaR = $291,342 (see notebook for backtest details).
Market Risk — VaR methodologies
Objective: implement and compare Historical, Parametric (normal & t), and Monte Carlo VaR methods (example asset: BTC-USD).
- Distributional fits (normal & Student’s t) and tail diagnostics.
- Monte Carlo simulation with configurable iteration count (notebook includes examples).
- Backtesting comparisons across confidence levels (90%, 95%, 99%).
Notebook summary: Historical 95% VaR ≈ −6.0% · Monte Carlo 95% VaR ≈ −3.23% (sample wallet)
Alternative Data — Financial news sentiment
Objective: ingest GDELT news, extract article text, run FinBERT, and correlate weekly sentiment with returns.
- Article extraction (newspaper3k) → FinBERT inference (Hugging Face); batch processing included.
- Weekly aggregation, bootstrap CIs, structural break detection, and robustness checks.
- Resume-capable pipeline with Parquet persistence for large runs.
Notebook notes: 6,250+ articles processed in sample runs — see notebook for sampling & inference cells.
Technical skills (quick)
Quant: yield-curve modelling, VaR, risk attribution · Stats: PCA, bootstrap CIs, structural-break detection · ML/NLP: FinBERT, PyTorch inference · Engineering: Python, pandas, matplotlib, reproducible notebooks
Hiring manager TL;DR
30 days: deliver reproducible backtests and dashboardable metrics for candidate models. 90 days: productionise the best risk pipeline with CI, monitoring and scheduled retraining.