My work has stuck me on a forecasting project because of what I learned in school, even though my applied work is zero. So I'm having trouble.
I am applying an ARMAX model to a stationary variable (customer savings outflows). I collected together explanatory variables. I checked for the usual things, uncorrelated variables, stationarity, unit root etc. I started with EViews as it is a bit easier to use, with the intention of using R for more exhaustive analysis once I was happy with results.
I found promising results, however, when I try to transfer my results to R to create the forecasts, the model spits out completely different coefficients. I am using the FinTS package, with its ARIMA function, and inputting a matrix made up of my xreg variables.
The thing is if I put variables in one or two at a time, they are all significant and it will be identical to EViews, however, I put in any more variables into my xreg matrix, all variables turn insignificant and explanatory power comes down solely to the AR(1) and constant terms. I know this is incorrect, I do not know why it does this.
Please can anyone tell me some reasons why this occurs and how I could reconcile my results with EViews, I'm in the dark here.
