Insidiousness of "good" indicators

written by Sergey Tarassov

 

During this year the Timing Solution research group did a lot of back tests for different indicators. Usually people who invent these indicators tell about their benefits only and do not mention the side effects that these indicators might have. We looked at the indicators that are used in Timing Solution software and the most popular indicators used in other programs as well. Our main focus was to understand where and why possible "future leaks" may occur. Definitely, we used different data sets for training and testing intervals. However, it was quite possible that sometimes indicators themselves may cause "future leaks". The results were very surprising: I did not expect that this effect could be so significant. Now I would like to share with you our findings and some thoughts. I need to mention that the tools suggested in Timing Solution software (as well as in other packages of other groups) do what they are supposed to do. My purpose now is to discuss a problem and bring your attention to it.

There are some indicators used in Technical Analysis and software for traders (including Timing Solution as well) that may lead to "future leaks" effect. The best example are Moving Averages (MA). Let us discuss them first.

The biggest problem with regular moving averages (MA) is the delayed lag. See below the price chart together with 100-days moving average (the red line):

The major turning points on the moving average are shifted to the right in respect to their real position on the price chart. And this is the real problem! This moving average can be compared to the dog that barks too late, when robbers are already in the house. Unfortunately, this is unavoidable price for the smoothed view of the moving average. (The opposite to MA is momentum; it is a leading indicator, and it does not provide future leaks either. This indicator feels the future price movements well, but there is another problem there: it is too noisy. This dog barks too much, so much that we are not able to pay attention to all its signals.)

Is there any way to deal with MA problem (i.e., delayed lag)? The answer is YES, but a BIG NO appears here as well. We can use a special type of MA that avoids the delayed lag. It is the Symmetric oscillator. There are two variants of it in Timing Solution. Look at the example of one of them:

As you see, this moving average (a purple curve) exactly reflects all turning points. But what is the price for this beauty?  The price is that this indicator has future leaks.

In other words, to calculate the value of symmetric moving average for point A, we use the price bars located AFTER this point. There is nothing bad with that - till we come close to the last available price bar. There is simply no data to calculate Symmetric MA further, and the program switches to a regular (exponential) MA - bringing back a delayed lag problem:

This delayed lag effect is not so big for MA used in Timing Solution software. However, it still exists. We always should keep this fact in our minds. As to my knowledge, other algorithms of calculating MA (like Mark Juric's MA) do not change this situation much.

This fact we have revealed doing the massive Back Testing. Practically it is impossible to discover this effect in any other way.

 

March 10, 2007

Toronto Ontario

Sergey Tarassov