This bactest
was done on — 03/11/2022
And the same code produced
Im running my code on the US Equities dataset.
Why am I experiencing such terrible results now?
This is interesting report. The best way to resolve this for you to share the two backtest results. You can get the report from the backtest page download option. Note: reports are available for backtests only, not quick backtests.
The date you mentioned is not very clear - do you mean 2021? And is it 11 March or 3rd Nov? What has changed recently is the asset universe and other things since we upgraded to Blueshift 2.0 - released in the first week of Nov 2021 (6th I think).
So if you are experienceing different results BEFORE and AFTER Blueshift 2.0, this might be because of this switchover. Everytime we improve our platform, your backtest results become more realistic. Unfortunately, that also means it will look less impressive almost axiomatically. I have no idea about your strategy and other details. But to hazard a guess, it looks like more than changes in the asset universe, the major reason you see a difference is the earlier unrealistic assumption of infinite margin trading that was built into zipline and fixed in Blueshift 2.0. Zipline was happy to trade even if you have 0 or negative balance in your account (pretty unrealistic), and Blueshift 2.0 refuses to trade without enough margin. This leads to excess leverage - which is what the first backtest equity curve looks like. May be your original strategy was run with a lower capital compared to number of assets you were trading? There were also some changes in the cost assumption, but I would guess that is not a major contributor here.
Finally, if you are experiencing different results between two dates AFTER the Blueshift 2.0, then please double check your strategy, they must be different in some aspects.
The first test was done on the 11th of March 2022and the second test was done today. And it works the same on other assets (crypto) I have saved performance from then and now also and they are identical. Was there anything what was done to the assets between that time?
No there were no changes in assets universe logic. If you have saved the backtest reports, you can share them (both) with us and we can further investigate your observation.
I figured it out. I includd the SPY in the backtest!
Good to see you have figured it out yourself.
Thanks but its still somewhat off and its not the only code that performance has changed.