renaissance_skill

Overview

This skill encodes Renaissance-style statistical arbitrage practices for building disciplined, scientific trading systems in Python. It emphasizes rigorous hypothesis testing, walk-forward validation, ensemble signal construction, decay tracking, and regime-aware adjustments. Use it to develop reproducible alpha research, robust backtests, and operational signal pipelines.

How this skill works

The skill provides templates and patterns for paranoid backtesting (point-in-time data, embargoed walk-forward validation), large-scale signal research with multiple-hypothesis correction, and ensemble signal management with decay weights. It also includes regime detection (HMM) to adjust position sizing and statistical tools to evaluate significance and hit-rate dynamics. These components combine to produce signals that are tested, monitored, and retired when they decay.

When to use it

• Developing new alpha signals and testing them across many instruments and periods
• Building a backtesting pipeline that avoids lookahead, survivorship, and selection biases
• Combining thousands of weak predictors into a stable ensemble with decay tracking
• Implementing regime-aware position sizing using Hidden Markov Models
• Running large-scale hypothesis tests with Bonferroni or FDR correction

Best practices

• Always use point-in-time data and an embargoed walk-forward framework for validation
• Correct for multiple testing when screening many candidate signals (FDR or Bonferroni)
• Track signal decay with rolling performance; retire signals below a predetermined hit-rate
• Favor ensemble and orthogonal signals over single-model reliance
• Model transaction costs, slippage, and market impact in every backtest

Example use cases

• Implementing a Renaissance-style backtester with train/test/embargo windows and PIT snapshots
• Building a SignalEnsemble that weights signals by recent hit rate and applies exponential decay
• Running thousands of candidate IC tests and selecting survivors with Benjamini-Hochberg FDR
• Fitting an HMM to returns/volume/volatility to detect regimes and scale risk-adjusted exposure
• Automating daily research workflows to detect and retire decayed signals