Projects
I am investigating a few interesting ideas with the help of Codex and GPT-6 Astra. The work presented here is exploratory. It is in order of preference.
Parallel portfolio backtests by Newton iteration
September 2026
Portfolio backtests are typically sequential: each day's trades depend on the holdings left by the previous day. We explore a parallel approach using Newton's method to reconcile the full portfolio trajectory while evaluating daily investment decisions concurrently. Experiments with transaction costs and realistic portfolio constraints show close agreement with sequential backtests for long-only portfolios, while long-short portfolios remain more challenging.
Learning alpha maps for convex portfolio optimization
October 2026
We learn linear combinations of alpha signals by differentiating through convex portfolio optimization, and test whether they improve out-of-sample Sharpe ratios over conditional-mean forecasts under trading costs and portfolio constraints.
Identifying a trading strategy from its return sequence
September 2026
Can a trading strategy be identified from its returns alone? We compare three methods—-squared-error matching, covariance analysis, and variance-weighted PCA—-to identify momentum, mean-reversion, risk-parity, and equal-weight strategies from daily return histories.