I am forecasting a series Y of monthly frequency and there is a variable X of annual frequency that would help a lot in the forecast if it had the same frequency.
I decided to apply the method proposed in Denton (1971), published in the Journal of the American Statistical Association, which transforms the annual series monthly by matrix operations.
This new monthly frequency X-hat variable helps a lot in reducing the prediction error with as
xreg in the
forecast package (
nnetar, for example) methods, including reducing the RMSE during
tsCV (time serie cross validation).
Is there any criticism of this process? Is it wrong to turn annual variable into monthly to forecast?