"Preponderance of metrics" approach to investment forecasting - A simple approach

 

A different way of improving forecasts that is also model agnostic is looking at data that have similarity that can be bundled as an index. It can take the form of a count indicator that can be described as the "preponderance of metrics". 

I first heard this term in a NY Fed research piece "Measuring Corporate Bond Market dislocations" but have not seen any other reference from a literature search. The researcher's idea is that several metrics that focus on market dislocations can be bundled to form a better index than using the metrics individually.

There, of course, is the term "preponderance of evidence" from a legal perspective but nothing on metrics. The preponderance of evidence is elusive because it does not tell us exactly how the idea is formed or reached. One man's preponderance is another man's uncertainty. This can also be called the weight of the evidence, but this idea of weighting does not focus on forming metrics. 

Nevertheless, there is a strong body of work on preponderance of metrics with diffusion indices. A diffusion index bundles a set of like factors and then measure the count of ups and downs. If the number of up is high relative to the total or number of downs, there is a preponderance. We see this is the PMI data which is an effective tool for measuring macro risks.

A simple monetary metric is the count of central banks around the world who are tightening or loosening policy. The Fed uses a preponderance metric with their financial stress indices (St. Louis and Kansas City Fed) or the financial conditions indices (Chicago Fed). 

The advantage of this forecast approach is that it is model agnostic. There is a creation of a metric but no fitting of the data to a dependent variable. The metric can be simply compared with a return series to measure forecasting success. Does the metric do better than a flip of the coin?

A trend-follower that bundles trend models that have different look-back periods creates a preponderance metric. If the multiple trends all point higher, then the forecast is for higher prices. 

Preponderance metrics can be created simply and without much technical expertise and easily tested if the bundle of metrics have some commonality.