Backtesting results much better than live market results

Alun · 2019-05-22T13:37:08+00:00

I have found, in doing many strategies for several currency pairs, that back testing after optimisation show very good and promising results (with capital gains plus walk forward testing) but when running the strategy in demo mode for the live market seems to give poor results. I understand that market performance can change from one test time period to the next (daily, weekly or monthly) but what are the other reasons. Is there a list of possible reasons that would help me understand. Is there some reference material that discusses this in detail. Many Thanks

Viewing 4 posts - 1 through 4 (of 4 total)

Author

Posts
05/22/2019 at 2:37 PM #99172 quote

Alun
Participant

Average

I have found, in doing many strategies for several currency pairs, that back testing after optimisation show very good and promising results (with capital gains plus walk forward testing) but when running the strategy in demo mode for the live market seems to give poor results.

I understand that market performance can change from one test time period to the next (daily, weekly or monthly) but what are the other reasons. Is there a list of possible reasons that would help me understand. Is there some reference material that discusses this in detail.

Many Thanks

05/22/2019 at 2:44 PM #99173 quote

GraHal
Participant

Master

Main reason is that you / we are choosing variable values in our strategies to fit historical data … we are making our strategies fit like a glove when what we need is for them to fit like a mitten and have more wiggle room! 🙂

History does not repeat itself (exactly the same) and so future performance does not match historical performance.

Hope that makes sense, if not just say.

05/22/2019 at 3:32 PM #99177 quote

Vonasi
Moderator

Master

We assume that you are applying a spread in your back testing? Without this most strategies look fantastic in back testing but as soon as spread is applied in real life they fail badly. Also the spread that you test with has to be realistic otherwise you are just kidding yourself. Spread varies throughout the day and so if for example you have an end of day strategy then if it is trading at midnight then you need to apply a big spread that compares with the reality of trading at this time of the day.

05/22/2019 at 5:12 PM #99194 quote

Nicolas
Keymaster

Legend

Optimizing variables is like using a time machine. By knowing the future, you can fit anything to be as good as possible. That’s why WFA should be used in order to make the strategies more robust, but it doesn’t even means that the best will come, it could be already behind you..
Author

Posts