Aditya Shahquant research at the intersectionof RL, LLMs & derivatives.

Backtests Black–Scholes, GBM Monte Carlo, Heston (stochastic volatility), and Merton Jump-Diffusion against bid/ask/mark/IV quotes from a historical options-chain dataset (2008–2025). Calibrates Heston and Merton once at t₀, then evaluates out-of-sample on subsequent dates. Discount factors come from a Vasicek short-rate model fit to FRED 6-month Treasury yields. Scoring uses RMSE, vega-scaled RMSE, and percentage of predictions within the bid–ask spread.

• Real SPY option market quotes (2008–2025)
• Vasicek stochastic discounting from Treasury yields
• Strict t₀ calibration → out-of-sample evaluation
• Vega-scaled RMSE + bid-ask hit-rate metrics

End-to-end VaR/CVaR system over a 3,536-day multi-asset book (SPY, QQQ, IWM, EFA, EEM, IEF, TLT, LQD, HYG, GLD) at 95% and 99% confidence. Uses Ledoit–Wolf shrinkage for Monte Carlo covariance. Formal Kupiec and Christoffersen tests reveal that Parametric and MC VaR materially underestimate 99% tail risk during regime shifts, with breach clustering during COVID-19. Stress scenarios (equity −20% / rates +100bp, risk-off −10% equity / −8% credit) quantify non-linear losses VaR alone misses.

• 3 VaR methodologies + CVaR @ 95% & 99%
• Kupiec (UC) + Christoffersen (CC) backtests
• Macro stress scenarios with non-linear loss attribution
• 504-day rolling lookback windows

Reconstructs AAPL option prices and Greeks via Black–Scholes with a time-varying implied-vol proxy derived from realized volatility. Holds short-dated ATM options, continuously delta-hedges with the underlying, and scales position size to a target portfolio volatility. Evaluates P&L under realistic trading and financing assumptions, decomposing returns into gamma, vega, and hedging-cost components.

• Systematic delta hedging with daily rebalancing
• Volatility-targeted position sizing
• Realized-vol-based IV proxy (chain-data-free)
• Greeks-decomposed P&L attribution

Classifies SPY into volatility regimes and scales exposure accordingly. Reduces drawdowns during high-vol periods, maintains exposure during stable regimes, and minimizes turnover. Evaluated with realistic transaction costs across the full sample, achieving superior Sharpe and Calmar versus passive SPY.

• Regime-conditional position sizing
• Lower drawdowns vs SPY buy-and-hold
• Low-turnover, transaction-cost realistic

Risk-analyst framework that attributes portfolio risk to underlying factors and computes Component VaR - identifying which assets drive tail risk. Built on a 10-asset multi-asset ETF universe with 504-day rolling estimation. Outputs include factor-contribution charts and component-VaR percentages at 99%.

• Systematic vs idiosyncratic risk decomposition
• Component VaR contribution per holding
• Fama–French 3-factor regression attribution

End-to-end retrieval-augmented generation for financial filings. Ingests 10-K / 10-Q / earnings transcripts, chunks and embeds them, runs FAISS semantic retrieval, and generates answers with explicit source citations. Designed to minimize hallucination and provide an auditable evidence trail.

• Modular RAG pipeline (ingest → chunk → embed → retrieve → generate)
• Citation-backed answers, hallucination guards
• Streamlit UI for real-time querying