Dear All,
We are still working on the best way to use MACRO data with AI Factors. Just plopping some of the predefined MACRO features we created has two major problems:
- They get normalized like any other feature
- The predefined macros we created are NOT point in time
For 1) we will have a solution next week. At the very least you will be able to skip the preprocessor.
For 2) we used the "non-PIT" formula. To use them correctly historically you have to use the PITSeries() function. It's an extremely confusing solution we introduced a while back. So confusing in fact that we completely forgot about it.
The chart below shows you what happens if you don't use PITSeries(). As you can see the GNP recovery was announced around Nov 2020, but if you used the wrong formula you would see the recovery around June 2020; a full 5 months before. Not Good!
Instead of correcting the predefined features in AI Factor, we will be deprecating PITSeries() altogether. Using a macro series in the past will simply just do the right thing instead of you having to use a "special function for backtest".
The fix will go in sometime next week. Very sorry about this. Not sure how we came up with this awful solution.
Formulas used in the chart
Revised ##GNP
Close(0,##GNP)
%(Close(0, ##GNP), Close(3, ##GNP))
PIT ##GNP
Close(0,PITseries(##gnp))
%(Close(0, PITSeries(##GNP)), Close(3, PITSeries(##GNP)))