From market flags to measured investments.
Within these projects, verifying value drivers means decomposing equity valuations into cash-flow and discount-rate expectations, separating common return factors from idiosyncratic noise, modeling how volatility changes through time, estimating the market-implied risk of corporate default, and testing how hedging strategies behave when volatility and transaction costs depart from idealized assumptions. These work samples range from fundamental company ratings to gradient descents across thousands of price paths, showcasing an ability to find value across the market.
Optimizers can amplify the noise until you use the right constraints
Factor-informed clean energy portfolio
A constrained mean-variance optimization across five clean-energy ETFs, disciplined by return shrinkage and validated against a Fama-French six-factor regression — Sharpe 0.564 vs. a 0.498 equal-weight benchmark.
VaR backtesting, GARCH & Extreme Value Theory
Backtested historical, exponentially-weighted, and GARCH-based VaR estimates against realized exceptions, then fit an Extreme Value Theory tail model to compare coverage accuracy across methods.
View notebook ↗Curve mechanics and credit fundamentals, tested against real filings and market data.
Colombian private-company credit screening
A fundamentals-based screening and ranking pipeline built on Colombia's Supersociedades XBRL filings — cleaning raw balance-sheet and income-statement data, then scoring companies on quality, safety, and value to construct a risk-capped candidate portfolio by position and sector.
Black-Derman-Toy interest rate tree calibration
Calibrated a BDT short-rate tree to observed discount factors and a market volatility curve via sequential bisection, then used backward induction to value a bond and an embedded option under both European and American exercise.
View notebook ↗From order book to stochastic modelling, know what's about to happen before the tape does.
A market-neutral trading signal built on order-flow imbalance, and a neural-network hedging policy tested against classical Black-Scholes delta hedging.
AI lead-lag microstructure trading
A market-neutral repricing signal on the AI hardware/software order book, built on Stoikov microprice dynamics and walk-the-book execution — Sharpe 22.37 out-of-sample with a market beta indistinguishable from zero.
Adaptive deep hedging
Trained a neural-network hedging policy end-to-end on simulated tail risk, comparing it against closed-form Black-Scholes delta hedging under transaction costs and both geometric Brownian motion and Heston stochastic volatility.
Open to full-time and internship opportunities in credit, risk, impact, and portfolio analysis.
Reach out or browse the project archive.