Data-Snooping, Technical Trading Rule Performance, and the Bootstrap
Read the paperopens doi.org in a new tab
What they found
The authors revisited Brock, Lakonishok, and LeBaron's technical trading results using White's reality check, which accounts for the fact that the best rule out of many tested will look good by chance. Across a universe of 7,846 trading rules on the Dow from 1897 to 1986, the best rule still showed significant profits after the correction, so the earlier finding survived. But in the out-of-sample period from 1987 to 1996, the best rules showed no significant performance, suggesting the historical profits had eroded as markets became more efficient.
What you can use
- When you test many rules and report the best, its apparent performance is inflated; the reality check tells you how much.
- The classic technical rules that worked for ninety years stopped working in the decade after they were published.
- Out-of-sample testing is not optional; a rule that survives data-snooping adjustment in-sample can still fail going forward.
Caveats
Index-level daily data; no transaction costs. The reality check has low power when many poor rules are included in the universe.
Tags: backtesting, data-snooping, technical-analysis, bootstrap
Summaries are our own reading of the paper, not the authors' words. Educational only, not advice. Discuss it in Book Club.