Statistical Arbitrage / Daily Equities / Pipeline Engineering
Daily Residual Statistical Arbitrage Pipeline
Implementation in progress
A daily US-equity stat-arb pipeline — universe construction, Fama-French factor mapping, rolling beta, residual z-score signals, daily target weights, and a full backtest — validated walk-forward with robustness gates on costs, borrow, gross exposure, and capacity.
Research Question
Can a disciplined, reproducible pipeline extract a daily residual mean-reversion signal from US equities — one whose performance is attributable to the signal itself, not to data, universe, or portfolio-construction artifacts?
Pipeline Scope
The system is built phase by phase:
- An IBKR historical daily-bar provider (adjusted and trades bars) with Parquet-based incremental download and caching.
- A universe builder (liquid stocks, daily snapshots) and a returns builder.
- Fama-French factor mapping and rolling factor-beta estimation.
- Residual z-score signals turned into daily target weights.
- A full backtest simulating costs, positions, and equity, followed by walk-forward out-of-sample validation.
- Robustness gates: cost sensitivity, borrow fees, gross-exposure limits, and capacity checks.
Current Focus
Optimization and parameter search are intentionally deferred. The first objective is a reliable research harness whose every stage produces inspectable outputs and explicit assumptions, before any more complex signal variants are introduced.
Validation Standard
A candidate signal does not advance on a single favorable report. It must pass reproducibility, diagnostics, and stability checks across clearly separated evaluation windows, with cost and capacity assumptions made explicit up front.