What would be really cool is if you took say 5 good strategies that are stable over time and then 5 strategies which die/fail at some point in the backtest, then see if the controller can catch on to this and keep the portfolio afloat when half of its strategy’s are decaying.
Then in live trading you can rest assured that if a strategy loses it edge your portfolio won’t suffer.
Yea I think it would be super powerful. Have you considered a noise buffer where allocations don’t change to ignore noise? May make it less likely to whipsaw
Interesting… how would you define noise? Some of the systems I wrote about like OLMAR and LOAD use the passive aggressive update for this reason. It tries to stay passive to avoid transaction fees and whipsaw but then aggressively updates if needed.
That’s probably more sophisticated than I was thinking. I was thinking a simple % buffer where you don’t update unless the change in allocation is more than x% different. But I’m sure there are more intricate methods
Yeah maybe you look at directional change so you only make a decision after an asset has moved X% since the last time you updated its allocation.
You could also probably use some sort of digital filters to help. There was some StackOverflow thread I came across a long time ago where someone made a novel online system that ignored outliers but then “caught on” if the outlier persisted after a few time step. I’ll try to find it.
But yeah something like this would be super helpful I imagine.
Also tuning the target and maybe changing that into some sort of loss function or regret function (amount of profit lost by not being allocated optimally or in the current winning strategy).
What would be really cool is if you took say 5 good strategies that are stable over time and then 5 strategies which die/fail at some point in the backtest, then see if the controller can catch on to this and keep the portfolio afloat when half of its strategy’s are decaying.
Then in live trading you can rest assured that if a strategy loses it edge your portfolio won’t suffer.
That’s a cool idea… Maybe I’ll make a mock example and see how it goes. In theory, it should work.
I found it kind of difficult to allocate across systems that oscillate between bad and good in my preliminary work continuously.
Yea I think it would be super powerful. Have you considered a noise buffer where allocations don’t change to ignore noise? May make it less likely to whipsaw
Interesting… how would you define noise? Some of the systems I wrote about like OLMAR and LOAD use the passive aggressive update for this reason. It tries to stay passive to avoid transaction fees and whipsaw but then aggressively updates if needed.
That’s probably more sophisticated than I was thinking. I was thinking a simple % buffer where you don’t update unless the change in allocation is more than x% different. But I’m sure there are more intricate methods
Yeah maybe you look at directional change so you only make a decision after an asset has moved X% since the last time you updated its allocation.
You could also probably use some sort of digital filters to help. There was some StackOverflow thread I came across a long time ago where someone made a novel online system that ignored outliers but then “caught on” if the outlier persisted after a few time step. I’ll try to find it.
But yeah something like this would be super helpful I imagine.
Also tuning the target and maybe changing that into some sort of loss function or regret function (amount of profit lost by not being allocated optimally or in the current winning strategy).
Very good content.
I’m skeptical