Weird strategy results difference on altered data
i have tested the strategy that Marc published a while ago (EURUSD_H1_bothStrategy 15.102), everything with 0.1 lot fixed position size. Tested the strategy with Dukascopy tick data (downloaded through TDD) and exported two different settings:
- UTC +0
- UTC +2 with removed weekend data
On non-altered data (UTC +0) everything looks ok.
But when i use the UTC +2 with removed weekend data, i´ll end up with something very unusual. The strategy now trades more frequently and loses pretty much all of its performance. The weirdest thing is the robustness analysis. While it looks usual on the UTC +0 data, it looks completely screwed on the other data. The original strategies performance is way lower than the 95% or 100% confidence level.
From the first look the strategy suffers a lot with altered data. On the second look it is a lot better on it (judged by the performance metrics of the 95% confidence level). Can anybody explain to me what happened?
Strategy references price data from points at specific time so when you use data shifted by 2 hours its quite common you get different and usually worse results. this makes sense to me
There was a hickup in the data set. It now matches and makes more sense (not falling apart completely after shifting the data 2 hours). Additionally the problem on the monte carlo analysis was a result of wrongly set options. You must not alter history data when you test on higher resolution (like tick or M1 data). Otherwise the robustness tests will look like the ones i have posted.
Viewing 3 posts - 1 through 3 (of 3 total)