I am trying to predict with a VAR model the number of orders of product X in company Y. It uses various indexes to do this. Unfortunately, the problem is that orders are specified in millions and such indexes are usually in the range (10, 100). The model on the test data gives, for example, values like 500k, when the real value is, for example, 2.5 million.
1 Is it because the values of other variables are very small? How to deal with this? I thought something about multiplying the index data by a constant. Only the range of the variable would change, but other statistics like trend etc would remain.
My order data is also seasonal, but VAR doesn't seem to take that into account. Of course, try the SARIMAX and AR model.