I just updated WL7 from build 41 to build 42, and also updated the Binance extension from build 9 to build 10. Now when I run the exact same backtests as before, I am seeing Profit % about 1/10th of what it was previously. Nothing has changed in terms of the date range, strategy parameters, or datasets.
This has affected every strategy I've tested with so far. I tried truncating the datasets and redownloading them, and that has not seemed to make a difference. Could this be a side effect of some change that was included in build 42?
This has affected every strategy I've tested with so far. I tried truncating the datasets and redownloading them, and that has not seemed to make a difference. Could this be a side effect of some change that was included in build 42?
Rename
No, nothing was even done around the backtesting area, and i haven’t been able to reproduce this with my strategies. we’d really need some more specific details to go on to attempt to help you.
I actually just did remember something, we fixed this Building Block related bug reported here:
https://www.wealth-lab.com/Discussion/Blocks-Long-positions-get-closed-by-Short-exit-condition-7320
If your Strategy uses both long and short positions then indeed it could account for the change.
https://www.wealth-lab.com/Discussion/Blocks-Long-positions-get-closed-by-Short-exit-condition-7320
If your Strategy uses both long and short positions then indeed it could account for the change.
My strategies do not include shorting. However, I do have multiple exit blocks, so perhaps it could still be related in some way. My strategies are generally set up with building blocks like this:
[Buy at Market] ..conditions...
[Sell after N bars]
[Sell at Market] ...conditions...
[Sell at Stop Loss]
Anyway, in case that's unrelated, here is a screenshot of a strategy backtest that I just ran on my Remote Desktop, which has *not* been updated to build 42 yet:
And here is the result of running the same strategy with the same settings on my local desktop, which was just recently updated to build 42:
Note that the backtest on build 41 yields 4365% profit, while the backtest on build 42 yields 157%.
Just to double check, I tried running the strategy on a separate remote desktop that was also on build 41, to make sure that the strategy successfully reproduced the desired results on a separate machine. It again returned 4365%, as expected.
[Buy at Market] ..conditions...
[Sell after N bars]
[Sell at Market] ...conditions...
[Sell at Stop Loss]
Anyway, in case that's unrelated, here is a screenshot of a strategy backtest that I just ran on my Remote Desktop, which has *not* been updated to build 42 yet:
And here is the result of running the same strategy with the same settings on my local desktop, which was just recently updated to build 42:
Note that the backtest on build 41 yields 4365% profit, while the backtest on build 42 yields 157%.
Just to double check, I tried running the strategy on a separate remote desktop that was also on build 41, to make sure that the strategy successfully reproduced the desired results on a separate machine. It again returned 4365%, as expected.
My guess would be a data issue. Check the Data Manager and make sure you have the same amount of 5 minute data for both computers. Also, comparing the equity curves might prove fruitful.
I'd copy over the data from machine #1 to machine #2 to ensure an apples to apples comparison.
Good idea, thank you Eugene. I tried copying the data from the B41 desktop to the B42 desktop and running a backtest there; the profit % increases slightly to 242%, but still no where near the expected 4365%.
I then reset everything back to how it was, and tried the inverse: copying the data from the B42 desktop to the B41 desktop and running a backtest. The backtest results are 4300%, so basically what I would expect. This says to me that the data being downloaded by build 42 is correct, but something about the way the strategy is applying it has changed from build 41 to build 42.
I updated three things as part of this: WL7, Binance extension, and DataExtensions. So theoretically it can be any one of those.
Glitch, I took your advice and compared the equity curves, but I don't notice anything potentially useful:
I then reset everything back to how it was, and tried the inverse: copying the data from the B42 desktop to the B41 desktop and running a backtest. The backtest results are 4300%, so basically what I would expect. This says to me that the data being downloaded by build 42 is correct, but something about the way the strategy is applying it has changed from build 41 to build 42.
I updated three things as part of this: WL7, Binance extension, and DataExtensions. So theoretically it can be any one of those.
Glitch, I took your advice and compared the equity curves, but I don't notice anything potentially useful:
QUOTE:
Now when I run the exact same backtests as before, I am seeing Profit % about 1/10th of what it was previously
Same problem here, and even worse: I get 44 signals on one symbol.
Other open symbols should at least have given my security stop loss (15%).
Backtesting profit was slightly below 60% the last 24 months, beating the benchmark (39% percent, see the screenshot). Today the backtesting is giving 6%!
I upated ALL extensions, and then WL7 to build 42.
Strategy: EOD.
My data provider: IQFeed
I trade the strategy currently live @Interactive Brokers.
The signals are generated after the close, I do trade the next day at the open. My orders are placed manually.
The strategy is LOCKED, since the beginning of my livetrading (10 days ago).
Something strange is happening under the hood... It seems as if only one symbol (CTLT, which I do hold at the moment) is getting all signals.
OK, I do see the issue with the Signals in a case like this, investigating!
We have it corrected, coordinating with Eugene about getting a patch out for this, thanks for the quick report!
Working on the build, and we should be able to push it out the door before the market open!
Have just installed Build 43.
Both the signals and the profit calculation work properly again.
Thanks!
Both the signals and the profit calculation work properly again.
Thanks!
Great, thanks for letting us know!
Your Response
Post
Edit Post
Login is required