WFA – Am I doing it right?
- This topic has 8 replies, 3 voices, and was last updated 7 years ago by
.
-
I have read quite a lot about WFA and thought I have understood how one should use it. Anyway I have grown a little insecure if I do it right because I see most people are uploading backtest that are obviously optimized differently than I would do it.
I am doing like this:
Let’s say we have 5 years of historical data. Then we are lucky and have found a strategy that performs well during WFA. Let’s assume we optimized with an in-sample period of 12 month and an out-of-sample period of 3 months. The strategy produces good results and has in every OOS period 60-80%. When I have come so far and decide to set the strategy live I would optimize again but this time a regular optimization (no WFA) over the last 12 months. This would give me the parameters to trade with for the next 3 months.
This is what I’m doing. So the parameters that I run live are the optimization results (in this example) from the last 12 months (no longer). If I then decide to run a standard backtest over the whole 5 year test period the result if of course very good during the last year but not so good for the time before. Anyway I think the is the way to go though one can not present a good looking 200K bar backtest with the last parameters.
Now the question, am I doing it right? I think I do but I get insecure because here people upload 200K standard backtests that are obviously optimized over the whole period and are proud to present a nice equity curve.
Another caveat for everybody using WFA (I made this mistake in the beginning). If you run a WFA it is important that there is no serial correlation between the trades otherwise the results become distorted. For example if you use some sort of position sizing, always deactivate it for the WFA.
1 user thanked author for this post.
Now the question, am I doing it right? I think I do but I get insecure because here people upload 200K standard backtests that are obviously optimized over the whole period and are proud to present a nice equity curve.
You explained it right, this is how everyone should use the Walk Forward tool!
For example if you use some sort of position sizing, always deactivate it for the WFA.
Right again, each strategy should be tested with the same position sizing over the whole backtest. If it performs good then any position sizing code could be added.
1 user thanked author for this post.
The current version of the WF module is based on best gain for the whole analysis (like the classic optimiser). I know it is planned to add more sorting criteria in a next version!
1 user thanked author for this post.
Find exclusive trading pro-tools on 
