Statistical Tests of Indicator Effectiveness

Overview

A discrete indicator is one that can only assume one of a finite set of values, typically only one of two values. As an example, an analyst may set up a moving average (such as a 200 day moving average) and then will judge that the asset is trending up if the current price is above the moving average. That is, the indicator returns only a yes or no answer, the asset is trending or it is not.

Two Sample Test

In the case of discrete indicators, it is common to compare the average returns of the asset after the indicator is found to hold, versus the average returns of the asset after a time when the indicator did not hold. In the trend example, the analyst would separate the sample data points into points where the asset was trending up (according to the indicator) and the points where the asset was not trending up. Next the analyst would take the average return after each data point, and compare.

Of course, these two averages will never be exactly equal. Random fluctuations in the data will show up in the averages. The question will become, is the difference in the averages statistically significant such that it is likely that the difference is due to the presence or absence of the indicator, rather than statistical noise.

The common way to assess this, is to use a two sample hypothesis test. To see how to run this test on an example, see example