How NOT to Trade with Any Indicator

Generally, I avoid criticizing my fellow traders’ styles.  Variety is the spice of life . . . no one knows everything . . .  if you can’t say something nice, don’t say anything at all . . .  etc.

But I’m going to make an exception in this case, since the material presented can be downright dangerous if applied as Mr. Jepsen lays it out.  Why?  Because, while his initial premise – that the ‘standard’ settings for any given indicator may not produce the best signals – is fine, his method of finding superior settings consists of the most blatant (and sloppy) curve fitting I’ve come across in some time.

You can get a hint of the danger at about the 2:40 mark in the video.  He’s already tinkered with the settings for the slow stochastic indicator so that we can all see a nice correspondence between the SPY price activity and the oversold/overbought triggers.  Looks pretty good, right?  But that’s when he mentions the indicator at the very beginning of the chart ‘got a bit rockety’ (meaning the signals weren't giving good direction).

This is a classic trap for technical traders.   The underlying thought makes sense: if one can fine tune an indicator's settings to the point that it exactly catches the 'personality' of a given instrument (the SPYders, in this case), catching good moves becomes child's play.  The trap is that you can always (especially with the help of number crunching computers) find settings which accurately DESCRIBE the past, but have no PREDICTIVE power at all.   As Mr. Jepsen hints, the settings he settled on for that date range of SPY behavior did NOT do a great job of describing the ETF's movement at the beginning of his chart - indicating that an earlier date range would likely have dictated a different set of stochastic configurations.  Consequently, all the great buy/sell points he highlights in this video would not have been there.

Unfortunately, I'm speaking from experience here.  Early in my trading career I fell in love with a very slick program which claimed to identify the 'personality' of any given stock by automating the same sort of optimization we witness Jepsen presenting here.  It would do it for any number of stocks, using just about every technical indicator the budding trader might want to use.  Every night after the close of trading I would watch raptly as it cranked through chart after chart.  When a signal with a high level of confidence was produced, I pounced!

So you can imagine my panic when, on more than one occasion, I had purchased a stock on the basis of such a signal and a few days later the signal DISAPPEARED.  Apparently, after re-optimizing for that stock, the program decided that earlier signal hadn't really been a great idea.  I was left flailing, wondering how to proceed.

It probably goes without saying, but I didn't stick with that program - or that approach to scanning for trades - for very long.

All of which is not to say that optimizing signals as part of your trading plan is bad.  But if you are going follow that path, a critical element is testing.  Again, most easily done with a sophisticated trading program, you would identify a time frame containing large amounts of data points.  After optimizing against that data set, you test your results against a different set of data to see how robust the results are.  So, if you optimize indicators within a time frame spanning the years 2004 - 2008, you could then test those optimized indicator settings against price data spanning the years of 2009 - 2010.  If the results are similar, you have probably identified a nice set of configurations.   If not, you know the 2004 - 2008 results were a statistical fluke, and can be ignored.

And THAT is how you can PROFITABLY monkey around with the settings of your charting program's indicators.

Above all - stay timid.

Timothy "Timorous" McCready