The process of optimization appears to be both critical and mind boggling complicated. I would like some input from anyone with a successful approach. Here are some of my concerns.

1. I read a lot of warnings about avoiding "curve fitting." How do I know when I am curve fitting? What are the symptoms?

2. I have been experimenting with optimization over 6, 12, 18, and 24 month time frames. So far, 12 months seems to give the best results. Is this common for a strategy that uses a 30 min. scale and average hold times of 10-30 bars?

3. I have been using Select net profit and % winners as my indicators of a 'optimized" strategy. Does anyone have a better idea.

4. These items worked well in a poorly organized walk backward approach. When I conducted a more statistically valid analysis in a walk forward approach, the four mot powerful predictors were bars held-(33%), %win -(26%), avg profit-(11%), and select net profit (7%). The number in parentheses is the % of variation in profit that can be explained by the parameter. When I did a multiple linear regression, I could only justify including the first two variables (37%). Adding more paramters did not materially move the needle. Does this seam reasonable?

Does anyone know of a good reference on this topic?

1. If you arbitrarily pick the "best values" without checking the optimization space, you could be curve fitting. You want to make sure that you don't pick "peaks" with big drop-offs on either side; in other words, choose values in the area of space that produce good results for many different values.

We recommend using the Genetic Optimizer to find an optimized set of values around which you can run an exhaustive optimization (hence, with fewer iterations) so that you can visualize the optimized space.

2. There's no telling what's common. Although it requires some work, you can experiment with walking forward an optimization by optimizing over different ranges of data. Then apply the results to a simulation on the non-optimized set for say, 6 months. Then, optimize through those 6 months, and test out-of-sample again.

3. % winners aren't that important, unless it is to you. Drawdown is far more important in my opinion.

4. I should have read this item first.. looks like you're familiar with walk-forward testing.

QUOTE:
Does anyone know of a good reference on this topic?

Check Amazon.com and support the U.S. economy - it needs it today!

FWIW, this one's a classic on optimization: The Evaluation and Optimization of Trading Strategies

Thanks for the help. I ordered the book. Hopefully it will help.

Does Fidelity ever hold a seminar on optimiization?

Thanks again.

Eugene, I hope I posted to the right thread. Is there anyway we can determine that wlp is running in the optimization mode? When my strategy runs, if it is being optimized by wlp, I need to know to set a few flags. I am looking for a function call similar to IsStreaming, which tells us whether we are in the streaming mode.

-thx j

See the IsPricePane property's example code in the QuickRef. Since chart panes are N/A in optimizations, the Strategy Monitor case also applies to optimizations.

To all readers: let's not continue with purely technical queries here. We're in "Trading Strategies", and this thread is purely for discussing optimization in general, various optimization techniques, best practices etc.