Optimizations

Hello all,

I am hitting my stride coming up with the proper settings and developing good strategies through all the processes in SQ.  But I still get a snag or two with the WF Optimization and WF Matrix Optimization.  In one forum post, Mark Fric said if you do Simple Optimization it’s not necessary to do WF Optimization. But what I’ve found is that my original strategy often looks fitted, despite good OOS, such as having stop losses too specific- like 33, and profit targets of 213.  So I have been using simple optimization to round numbers out to make them more generic.  Then, I pick the most commonly used generic inputs for the parameters that produce a solid profit in Simple Optimization, but that doesn’t necessarily mean it’s the most profitable.  So if 6 out 9 Simple Optimized versions use an RSI of 18, I will use that version even if that version produces a smaller profit than a version that uses a value all by itself.

By trying to generalize the parameters, and using the strategy with the most commonly shared input parameters, I feel it has a better chance of surviving the longer term.  And in doing so, I have noticed that the success rate in the WF Optimization and Matrix Optimizations come up with a higher pass rate when I use this Simple Optimization method first.  I know over-optimization is a problem but I feel that doing it this way is the opposite, I’m using optimization to try and make the parameters more general, not more specific and “fitted”.

Where am I wrong?

Todd

Good point toddone, too often I fall into looking for best results from optruns, this is keen observation and thought.

That is exactly how i do it. I don’t make steps of 0.1 for Coef and I don’t adjust RSI 40 to RSI 39. I do wide steps. Maybe 5-10 step for that RSI example for 0.5 step for the coef example. Far more robust.

Its funny how after people do over optimization or too little data, they blame their “bad data” when it doesnt work in live trading.

Then, I pick the most commonly used generic inputs for the parameters that produce a solid profit in Simple Optimization, but that doesn’t necessarily mean it’s the most profitable.

 

Does SQ offer a quick way to select the most common set of parameters?  Is there some way of quickly seeing which values appear the most often are are robust values?

Does SQ offer a quick way to select the most common set of parameters?  Is there some way of quickly seeing which values appear the most often are are robust values?

You can use Load settings/Save settings in StrategyQuant to pre-define your set of parameters (selected indicators, money management settings etc)

More complex strategies usually have the tendency to be less robust