Topic: Stats for Sidekick-style trade filtering

I know I'm years late to the party but finally I have arrived!

Back in the day when Steve introduced Sidekick I was quite curious of the actual performance. You know, it was mind tickling to read about different "policies", which filter trades. But how to obtain the stats, which at least try to be intentionally meaningful? Demo trade? Ok, how about for 30k hourly bars? That's roughly 1250 trading days... Choosing that route I would be still playing on a demo. Resorting to accept stats for 2 weeks of demo would be moronic to say the least, just like backtesting strats over multiple platforms over the same set of data and expecting a significant raise in its future ability to generate profit. Better to stick to cold hard numbers.

I just finished introducing the consecutive win policy to FSB's account statistics. This is the first version, the most primitive one. Only thing to look at are balance numbers, which account all trades, which happen after 2 and 3 consecutive wins respectively.

I'll show the stats for a strat done on a 30k H1 bars. The sequence from left to right is initial strat, then the middle one is optimized for highest win/loss ratio, and the last one is optimized for highest net balance.

Doesn't look good for optimized versions. I think I'll go through many more and much different type of strats before I commit developing the policies further.

Post's attachments

bal.JPG 117.65 kb, file has never been downloaded. 

You don't have the permssions to download the attachments of this post.

Re: Stats for Sidekick-style trade filtering

A little update. There was another policy in the late Sidekick, which took into account the win/loss ratio for the last 10 trades and if it exceeded a set threshold, it would allow trades to be opened. I managed to incorporate that into Account Statistics as well. As the stat file is "static", I made it to present 3 threshold levels (0.55, 0.65, 0.75).
For illustration purposes I have attached a round-up of the same strat following the same procedure (standard parameters, optimized for best W/L ratio, optimized for best net balance). I'll now include the file as well so everyone interested can have a look, make sure though you keep the copy of your current AccountStatistics file.

What I've learnt so far: the consecutive win policy is rubbish. It is hard to make it work purely on a theoretical basis and the results show for themselves. It might work for a very specific type of strats on lowest TFs, but that's easier said than done. And as one would expect, it is a killer of trend-following strats.
Winrate policy is worth to be looked into a bit more. One thing though is certain - for an outright backtest performance nothing beats the original strat. But now the question is whether it is possible to eradicate more losses than wins with this policy. That's the only research angle I'm contemplating at the moment.

Post's attachments

AccountStatistics.cs 32.15 kb, 12 downloads since 2020-02-05 

bal.JPG 125.78 kb, file has never been downloaded. 

You don't have the permssions to download the attachments of this post.

3 (edited by Lagoons 2020-02-08 10:42:23)

Re: Stats for Sidekick-style trade filtering

Hi footon,

really great stuff. Thanks for sharing it.

About the consecutive win policy; I’ve told you so. ;-)
I’ve also played around for a year with it, I think it makes no real sense to use it.

First of all, as you already mentioned, profitable trades will be left out but you’ll most likely would have needed them to cover future losses. Therefore, your strategy would need a really high w/l ratio trying to cover this.
Additionally, you’d need a proper risk/reward ratio to limit the impact of losing trades.
All this combined is a really tough weight to carry and in the end it just isn’t worth it.

What you may also consider when you set a new strategy live, in general the best performance is achieved when it starts trading right after your sample size (for creation/optimization). If time goes by the performance usually drops, so you’d might lose the most profitable time of your strategy on top.

The only thing it might work here is for a strategy which is not really working on its own, then you might get lucky and achieve better result than then original strategy.
But I don’t know if that’s really a proper way to go.

The w/l ratio policy might be better, maybe you’d able to cut losses when the strategy performance starts to decline.
But I assume similar to the consecutive win example, this approach might work with poorly performing strategies better than which carefully selected ones.