1

I have data, which I am sure has a downward trend. I am trying to forecast this data using ARIMA and I want ARIMA to consider the trend when it is forecasting.

The first step in ARIMA is to determine the order of differencing for which I am using the KPSS test. The KPSS test in R has an option 'trend' and 'level'. Which one should I choose if I know that my data has a downward trend? I am confused because I dont understand what is the input ARIMA is expecting if trend has to be taken into consideration for forecasting.

Another question: Is there any sure shot way of knowing if the data follows a trend?

Ferdi
  • 4,882
  • 7
  • 42
  • 62
nancy
  • 81
  • 2
  • 3
  • As a complement to the answer given below you may see [this post](http://stats.stackexchange.com/questions/107551/can-a-trend-stationary-series-be-modeled-with-arima/107696#107696). – javlacalle Jul 16 '14 at 17:08

1 Answers1

1

It is about level or trend stationarity, i.e. a deterministic trend. So, if you think your data is stationary around a constant, you should opt for "level" but if you think that your data is stationary around a deterministic trend then select the second alternative, "trend".

DatamineR
  • 1,477
  • 3
  • 18
  • 25
  • how do i know if my data is stationary around a constant or trend? So I understand you answer as if I want trend to be maintained in forecast, i should do the kpss test for trend. Am I correct? – nancy Jul 16 '14 at 17:40