Stability/Curve-Fit Analysis For Low Trade Count Strategies
- Thread starter sle
- Start date
Too much jargon for me to follow it all ... but perhaps that's your objective; to weed out those who can't understand your point!
What's an example of an "ex condition" test?
Test with all possible combinations of filters? Hmmm ... Now, THAT sounds like curve-fitting to me.
How would this work? Please explain.
What's an example of an "ex condition" test?
Test with all possible combinations of filters? Hmmm ... Now, THAT sounds like curve-fitting to me.
How would this work? Please explain.
From the context, I suppose "ex-condition" means a separate test of trade exit parameters.
I use fixed bar lengths. Random lengths might have a negative effect on time dependent parameters. I've tried random trade samples for OOS, but found that the results are then still biased, probably because the trades cover the same market situation as the in sample data. Therefore, I think the conventional method of separating a part of the price curve for OOS is still the best.
Nope, the more the merrier.
For example, lets say you are buying S&P futures when 50 day MA crosses 250 day MA. In this case, your "condition" is the cross between two MAs, that is your "signal". Your ex-condition test would be - what is the return of S&P futures on all days, not only when your condition occurs.
No, if I am using say 3 filters and the strategy shows good performance, I would remove one or remove two of them and see how much the strategy suffers. An ideal situation is that each filter (and each combination) has independent alpha.
You "imagine" that you are doing the trade you are doing, e.g. buying S&P futures without any prior conditions. Then you say that you are going to use stop losses in these trades and find stop loss that makes decreases in returns/decreases in losses acceptable to you. This ex-condition stop performance you can now apply to any of the strategies that uses that very same trade.
The trader's equivalent of Einstein's quote that "God does not play dice".
Suppose you test 2 strategies on the same period. A's trade sample size is half of B's. Now operationally, A's total allocation would already be half of that of B because it generates less trades. So you're automatically betting smaller, right?
Quote from NYDreamer:
Wrong - the actual bet on a system, is the max drawdown you will allow the system to get into, before pulling the plug. One component of that is intrinsic to the system (although it is up to you to calculate it, and a key factor is the confidence level you use to that effect), the other part is a linear function of position size.