Saltar a contenido

Backtest frameworks + pitfalls — synthesis

Síntesis de frameworks Python populares (VectorBT, Backtrader, Zipline, QuantConnect, NautilusTrader) + pitfalls clásicos del backtesting (López de Prado: survivorship, look-ahead, overfitting). Raw en raw/backtest-frameworks-and-pitfalls.md.

Key data usado en el wiki

  • Tradeoffs frameworks → concepts/backtest-methodology.
  • Purged K-Fold CV, Combinatorial Purged CV (López de Prado 2017).
  • Overstated returns por survivorship: 1-4%.
  • Cita López de Prado: "7 configs → 1 con Sharpe >1 OOS=0".
  • Principle: "Backtesting is not a research tool. Feature importance is."

Fuentes

  • Medium reviews multi-framework (Trading Dude, Pham, autotradelab).
  • Bailey/López de Prado — Probability of Backtest Overfitting paper.
  • López de Prado "Advances in Financial Machine Learning" notes.
  • GARP whitepaper "10 reasons ML funds fail".

Relaciones