Suppose you come across a financial trading system which starts by checking a security's 200-Period Moving Average, either by itself or in conjunction with the security's price data, perhaps signaling bullish evidence while price is above the 200 MA and signaling bearish evidence while price is below the 200 MA, or whatever other example interpretation you like.

Then the system checks the security's 199 MA for confirming evidence, with respect to the same interpretation as the 200 MA. Then it checks the security's 201 MA for more confirming evidence. Maybe it checks 5 or 7 such clustered MAs before feeling confident if the "evidence" all points in the same direction.

This is obviously an absurd trading system, but my prior my framework for understanding how to quantify its absurdity was in error. I used to call this problem autocorrelation. I believed the term to mean "automatic correlation," describing (at the most extreme end of a continuum) two indicators that would automatically produce identical bullish/bearing evidence, buy/sell signals, etc., just by virtue of their construction, regardless of their explanatory power derived from the security, and as a result, the pair of indicators could be no better than the more predictive of the two. Similarly for trading systems involving more than two indicators.

I have since learned what autocorrelation actually means, but now I don't know how to quantify the above problem, in order to identify it in less obvious contexts. For the special case in which a trading system's overall decision rule is a linear combination of the pieces of evidence output by its indicators, would this be an example of the phenomenon known as multicolinearity? What would it be studied as in the more general, not-necessarily-linear case?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.