The desired outcome is a smoothly increasing equity curve. Exists there a way to characterize that I wonder? The increase part is obvious, the smoothness could be characterized by a linear regressive channel width or something...
There is always a curve being fitted. Posteriori it is the one you got. Think about it: you start optimizing with no target curve and you get curve C1. If you optimized for curve C2=C1+dC, where dC is some small change about C1, you will get close to C1. At the end, the result is the same. Optimization of metrics = curve-fitting.
No, there isn't. Curve fitting is finding a function that best approximates a set of data. We aren't trying to approximate a set of data, we are trying to maximize a given outcome. Entirely different procedures.
Thanks for the advice guys. I also did a monte carlo simulation and it showed pretty good stuff. PF for the optimization for 2011 was around 8.0, the unoptimized years 2010, 2012, 2013, 2009 were around 3.0-4.0 PF per year for an intraday sytem. Perhaps that would imply that my system is only somewhat curve fitted but this definitely passes the out of sample test. My concern about testing against unrelated instruments is that my system incorporates certain variables that are directly related to the S & P itself so they wouldn't be relevant to other nonrelated instruments unless it were something like NQ or some other market beta 1 instrument.
Hi EliteTrader, Good PF numbers tend to be in the 1.5 - 2.5 range for a lot of strategies I've seen and your PF of 8.0 is unusual. I normally calculate Profit Factor as gross profits divided by gross losses, but it looked like your definition in the 1st post was gross profits minus gross losses unless I misunderstood. Is that how you calculated it?
Yes, thats the right calculation, but what I suspect will happen is that slippage in live trading will knock it down to 2.5 or 3.0PF so 8.0 is a pipe dream.
You should recalculate your profit factor. It's not very meaningful as you defined it by subtracting the gross loss from the gross profit. Instead, it should be the ratio of the two. If your profit factor is 8 by your calculations, it's pretty much a break-even system. For example: Gross proft: 10,000 Gross loss: 9,992 Your profit factor as you defined it: 10,000 - 9,992 = 8 Real profit factor: 10,000 / 9,992 = 1.0008
That's confusing because you mentioned "gross profits minus gross losses" in your original post. It would help if you posted a more complete report which should include the number of trades, the time period covered, max DD, etc.
Wouldn't the most important be the easiest to test? If it takes a large drawdown before you know the strategy works, then you'll never know if the strategy works. So it has to be a strategy that can be tested quickly & cheaply. That would rule out everything but profit factor.