English
Auf die Warteliste
Die Bücher... aber nicht gratis
10 Strategier Som Slår BörsenSvenska10 Strategier Som Slår Börsen10 Strategies That Beat the MarketEnglish10 Strategies That Beat the MarketProRealTime Coding Made SimpleEnglishProRealTime Coding Made Simple
600+StatistikMarktstatistiken und Edges1,327StrategienBackgetestete Strategien18Research PapersDie Studie hinter den Algos301DokumentationProBuilder-Referenz27Algo-EntwicklungsprogrammKurse und Lektionen24Alles hier fließt in die 24 AlgosDie Algos ansehen

Statistik › Backtesting und Belege › The haircut

A ten year backtest replicates about 30% of its Sharpe ratio, and the parameter count is why

444 parameters|real data, 9% replicatedIn sample: 3.6Out of sample: 0.30.33 parameters|real data, 68% replicatedIn sample: 0.7Out of sample: 0.50.5444 parameters|simulated, 38% replicatedIn sample: 8.2Out of sample: 3.13.13 parameters|simulated, 0.9 to 0.6In sample: 0.9Out of sample: 0.60.6In SampleOut of Sample

Sharpe ratios in sample and out of sample for the same twelve commodity futures and the same 36 signals, fitted two ways. Data May 1998 to December 2023, ten years in sample and three out, with ridge regularisation. The simulation rows are 10,000 realisations of the same setup.

The common rule is to halve an in-sample result. One derivation puts the expected replication ratio of a ten year backtest at 30%. On real commodity data a full regression with 444 parameters replicated 9% of its in-sample Sharpe while a three parameter panel replicated 68%.

About 30%Expected replication ratio of a ten year backtest
9%Replication ratio, 444 parameter model on real commodity data
68%The same data with a three parameter panel model
3.6 against 0.3In-sample Sharpe of the 444 parameter model against its out of sample Sharpe

Nur für Mitglieder

Der restliche Teil dieser Seite ist für Mitglieder

Below this point there are 5 sections and 4 named sources, roughly 750 words of it. Every figure carries the source it came from and the date the data is from.

Sie können weiterhin jede Statistik in der Bibliothek kostenlos durchsehen. Die Einleitung und die Zusammenfassung sind immer offen.