Topic: OOS in Studio

I know there's a difference in how the Studio calculates OOS vs FSB but I've noticed a variation I was curious about.  In FSB if I run everything with 40% oos what generally happens is 90% of the strategies have a solid equity curve for the 60% of the curve the strategies were optimized for and then goes off a cliff for the remaining 40% that it wasn't.  That's expected as most strategies are overoptimized and you're looking for a few gems that can stand on their own. 

In studio when I run with 40% OOS I'm not seeing that cliff in most strategies when the equity curve hits the part of the data range the strategy wasn't optimized for.  Is the selection algorithm pre pruning out the bad overoptimized curves or is something else going on?

Re: OOS in Studio

There is a difference on how both programs use OOS.

FSB Pro cut out the OOS part and calculates only on the base data. It pushes to the collection all profitable until the OOS separator strategies. We cannot see the OOS part of the backtest because it doesn't exist. We can see it when we send a strategy on the Editor or if we use the Multi Tester to recalculate the complete collection.

EA Studio calculates against the full data series. However, it shows on the Generator screen the most profitable strategy at the OOS separator. Because the fact EA Studio uses the complete data series, it is able to display the green OOS zone.
Up to here there is no difference between the programs. However, EA Studio applies a very cheap trick. Because it sees the final balance (you remember that it uses the complete data series), it sends to the collection only strategies that have Net Balance not lower than 70% of the currently shown strategy.

This trick have on positive and one negative effect:
- pros: the collections not cluttered with bad strategies.
- cons: if the Generator discovers a strategy with a huge profit shortly after the run, it will be difficult to find other strategies with profit within the 70% margin and will not push to the Collection. The solution is to Stop and Start it from time to time.

Re: OOS in Studio

I guess what I'm trying to figure out is if the strategies presented are only optimized up to the amount of data present at the 60/40 boundary.... in which case that 40% green zone essentially represents a forward test or, if they're optimized for the full timespan but the filtering happens based off the balance at that boundary line (in which case the data might still be over-optimized to curve fit the data in the whole range.)