Skip to content
All library documents

Testing Whether an Entry Filter Selects Better Trades with Block Permutations

Article MQL5 articles
Read the original

Summary

This article separates two questions about an entry filter: whether enforcing it improves the resulting portfolio metrics, and whether its accepted trades are unusually favorable within the base strategy’s fixed set of trades. For the second question, the analyzer labels each completed base trade as accepted or rejected, measures realized net profit including trading costs, and compares the accepted-minus-rejected mean with fixed-count placebo selections. It uses individual shuffling as a naive benchmark, plus equal-block permutations and circular shifts to retain different amounts of temporal structure.

The verdict requires a positive observed difference, support from both dependent-data tests, and agreement across configured block sizes. The operational comparison separately scores six tester metrics, since occupancy and sizing can alter opportunities when a filter is enforced. The method assumes one labeled entry and one complete exit per position; partial exits and scale-ins are not supported. Its example uses one symbol and fixed volume, so heterogeneous trades need outcome normalization. The author recommends out-of-sample validation and notes that the permutation models cannot capture every form of dependence.

Key ideas

  • Operational portfolio improvement and statistical selection ability are distinct claims.
  • The primary statistic is the mean net-profit difference between accepted and rejected base trades.
  • Placebo masks preserve the accepted-trade count and are tested with shuffling, equal-block permutations, and circular shifts.
  • The automatic selection verdict requires dependent-data test support and consistent block-size sensitivity.
  • The collector assumes a single entry and complete exit per position, and mixed trade sizes require normalization.
  • Out-of-sample validation is recommended before treating a filter as useful.

Tags

StatisticsBacktestingRisk management

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.