Okay,

I realize that it's generally accepted that curve fitting is bad. There have been several discussions about how to fight curve fitting (using out of sample testing historically, out of sample forward tests, etc)....

From what I've read, the worse type of curve fitting is signal based. I.e. your strategy involves an entry signal that's variable (rather than binomial) and you optimize the entry signal....

Assuming that your entry signal(s) are binomial and/or fixed (there's no way to alter your entries, they are where they are).....

Is it a bad thing to optimize exits and money management strategies?

For instance....if your strategy has an entry signal based on price action.....and also has an exit based upon price action.....and it's profitable, but maybe features an equity curve that's not as smooth as you'd like or more drawdown than you'd like....

By adding exits, fixed stops, trailing stops, etc....

I've been successfully trading an automated system (with marginal success), but I'm working on a couple of other setups that are MUCh more profitable...and all I read about are horror stories surrounding over-optimized systems that not only didn't perform as well, but actually lost money (beyond an excessive drawdown condition).

Now, I've also tested (manually) for "robustness" and the systems still make money if you adjust the exits (P/L amounts, trail amount, trail activation level, etc) and outside of deliberately trying to sabotage the system into turning negative, it almost always makes money, just not nearly as much under "optimized" exit criteria.

Thoughts?

I guess what I'm saying is that I know optimization and curve fitting is bad....but what else are you supposed to do? Throw a dart against a wall and go with it?

Wouldn't you want at least SOME methodology for profits and stop limits based upon some historical "average" of swing lengths, trend lengths, cycles, etc. (even if that historical information is extremely local...as in the previous several bars, the previous 2 swings, previous couple of trends....or maybe it's an average of the previous 20 swings, etc).?

I guess what I'm saying, if curve fitting exits and money management is bad.....then what's good?

I find defined price based stops and targets work best in my system. I essentially optimize to get the best ratio based on the interplay of stop to target when developing a strategy. They are based on MAE and MFE data. At one time I had a signal based exit but was not consistent. It had times of high profitability but unpredictable large draw downs. I found defining my stops and targets based on price created a much more stable system with much less draw down. It also gave me much more confidence in my expectancy allowing me to really focusing on position sizing. I find 80% to 90% of my profits in a given year are a result of position sizing. I think optimizing stops and targets is very important. To optimize does not necessarily mean you are curve fitting. Curve fitting can be a real problem. I think one way to avoid that is to stay simple enough that when you backtest it gives you a reasonable degree of accuracy. Then forward test to see how things line up with your backtest. Look at the data as it is not what you want to see and you can avoid curve fitting. Also try to really find any point in your testing that could skew your results.