Don't optimize your trading systems!  What?!  Why?  Because optimization picks the single "best" set of trading system parameters, perching them on a single peak equity over a history.  Chances are the market will shift and the equity peak will crumble beneath you.  The more you tighten your optimization to make it look better in the past, the more you are pushing bad performance into the future, when real money is on the table.  Ouch!

A good trading system will have a region of parameter values that delivers good equity performance.  Since the markets are "non-stationary" (shift), the optimal parameter values move day-to-day.  You need a trading tool that will sense and track along with it.  You need something like BioComp Dakota and I'll show you why...

To give an example, let's look at the graph below for MCHP traded with an Advance/Decline trading system (available with Dakota).  The vertical axis is money gained over 5 years on 100 shares.  The horizontal axis is the look-back period of the trading system.  Investigation shows us a "fruitful range" of look-back periods from 2 to 10 bars (red circle).  Optimization finds that 7 bars gives the "best" trading result: $1,013/100 shares.  We can see that higher look-back values yield spotty results which we can avoid altogether ("No Man's Land").

We can also see that if we merely shifted one bar to the left from the "optimal" 7 bars, our $1,013 "optimal" result collapses to $400.  That's a 60% drop for a mere 1 bar shift.  Sounds fairly "brittle" to me.  Also, the 7 bar period was an average over the 5 years.  Was the best value really exactly 7 bars every day for the entire 5 years?  Probably not.  Maybe we would have made more money had we shifted that look-back adaptively as the 5 years unfolded.

To test out this theory, I used BioComp Dakota to adapt this trading systems to capitalize upon the equity dynamics of this Advance/Decline trading system upon MCHP.  Since Dakota purposely injects a bit of noise into the trading system to hover around good equity, I ran the swarm of Advance/Decline trading bots 30 times and captured the results (a feature of Dakota).  What did I find?

I found that the best adapted swarm system made 86% more money than the best-of-the-best "hand-optimized" solution and I found on average, the adaptive swarms made 129% more money than the average of the "fruitful region" and the average swarm even beat the "best" hand optimized result.  Why?  Because Dakota's adaptive swarm technology was able to automatically track the shifting dynamics of MCHP over the 5 years and captured much higher equity results.

So, the optimized result of 7 bars was not only brittle with a neighboring precipitous drop in equity, but it really wasn't "optimal" at all.  A dynamic system like BioComp Dakota delivered substantially more money because it could shift along with the changing nature of the security, MCHP.  Now, every security is its own animal, as is each trading system, so results do vary, but I am sure you can see the advantages of Dakota's adaptive technologies.

Please send questions and comments to cmcook@biocompsystems.com.

(graphics shown were created in Excel and Snag-it)