I want forecast inflation using ARIMA model. My questions are:

  1. Does ARIMA require normally distributed input data? (Because my data—inflation—is not normal.)
  2. If ARIMA require normally distributed input data, how can I transform it to be normal? My data (inflation) contains negative values so I couldn't use the log transform to handle it.

NO. The errors resulting from the ARIMA model must be normally distributed.

| cite | improve this answer | |
  • 3
    $\begingroup$ Please add the following caveat "for parametric tests of sufficiency( t values for the coefficients) and necessity (randomness of the residuals) to be valid" $\endgroup$ – IrishStat Aug 1 '15 at 21:29
  • 2
    $\begingroup$ I would be a bit more careful. It is usually easiest when the errors are normal, but the ARIMA model may still be estimated consistently when they are not. math.unice.fr/~frapetti/CorsoP/Chapitre_4_IMEA_1.pdf $\endgroup$ – Zachary Blumenfeld Aug 1 '15 at 21:30
  • $\begingroup$ @ZacharyBlumenfeld, nice lecture notes under the link! $\endgroup$ – Richard Hardy Aug 2 '15 at 8:27

Not the answer you're looking for? Browse other questions tagged or ask your own question.