Mechanical research can be thought of as a process of self-fortification.  It's a way of building resolve that what you have in the present is something that will probably work in the future based on how you've seen it perform in the past.  To capitalize, you have to follow your model exactly as it traded over the history of your data field.  Picking and choosing signals you agree with or haphazardly changing parameters to “reflect what's happening now” are all ways of ensuring your real-time trading approach won't resemble your projections.  People think such moves will be an improvement.  It's almost inevitably the opposite case, quite frequently disastrously so.

Optimization is the process of determining which variables worked best within a given data field -- a crucial part of the fortification process.  Why would you want to buy a 10-day high if research demonstrated that a 20-day high worked much better?  It will bolster confidence to trade the best possibilities.  Of course, if you over-do the weaning out-filtering process, you're pretty much guaranteeing that your future will look nothing like your past.  It's a fine line we're forced to walk.  Great results won't just pop up out of a totally untilled field.

Popular software offers several optimization variations.  The common assumption is that you want to optimize for your best net return.  Sounds reasonable, but it's actually not the best function to use.  Why?  Consider a system made $100,000 profit over a given time period but had a worst drawdown of $80,000.  Another way of saying that is that at some time within the period, $80,000 was an investment necessary to make the $100,000. 

Compare that to a system that over the same period netted $40,000 with a $7,000 drawdown.  Which system is better?  All other things being close to equal, system two wins hands down.  By tripling your trading size, you could not only make more money ($120,000 verses $100,000), but you could do so with appreciably less risk ($21,000 verses $80,000).

The function that describes that type of analysis is the “Return on Account.”  (At least that's what TradeStation calls it).  It is computed by dividing the net profit by the worst drawdown.  Your result is a percentage gain which assumes that your initial investment was your worst drawdown.  In the hypothetical second example, you invested $7000, were unlucky enough to start trading at the very start of the worst losing streak, stuck it out until the end, so in effect, invested $7,000 to make $40,000.  Your Return on Account was the latter divided by the former, or a 571% return.

Return on Account is my usual optimization tool.  There are occasionally, however, reasons to prefer some of the other alternatives. We'll look at another one in our next episode.  Stay tuned.

P.S. I'd love to receive feedback from you. Please leave a comment or discuss the article by clicking on "Make a comment on this article" below.

Art Collins is the author of Market Beaters, a collection of interviews with renowned mechanical traders. He is currently working on a second volume. E-mail Art at artcollins@ameritech.net.