I am interested in building an index that tracks how accurate economists are at predicting several US economic statistics:
- US Jobless claims number: comes out weekly, ranges from 100k to 400k, usually in eight year cycles
- U Michigan Consumer Confidence: comes out bi-weekly, ranges from 50 to 115.
- Case-Shiller Housing index: comes out once per month, ranges from 60 to 200
Each of these indexes have an economists estimate number and then the actual number and I am specifically interested in the accuracy of the estimates versus the actual. there are usually 30-40 economists that provide an estimate. My questions are as follows:
Should I use the average of the economists estimates in my index or should I use the median estimate? Usually the economists "cluster" their guesses near last months numbers while there is always the odd-ball economist who will throw out an extreme number.
I want to equally weight the accuracy of the estimates on each of the numbers, however the data comes out in different frequencies. If I build a spreadsheet with weekly dates, should I "fill in" the last data observation in the blank cells in my spreadsheet. For instance should I collect four different weekly US Jobless Claims numbers while repeating the same number four times for Case-Shiller (since I have no data for three weeks)?
When I build my index, should I calculate the % deviation from estimate vs actual and then get the z-score of that column or should I calculate the difference between estimate and actual and then get the z-score for the entire column of that difference? In the situation of Case-Shiller the housing index grew from ~60 to 183 so a 3-point difference in estimate was much bigger back then than it was when the index was at its peak.
I am curious to see how you would set this up.
