Stationary Test issue

问题

I am working with air miles data set and i conducted three different tests to check for stationary in the time series data set

Test 1: Using acf and pacf

acf(airmiles)

pacf(airmiles)

After differentiating its seems most of the values lies in significance level now

acf(diff(airmiles))
pacf(diff(airmiles))

Test 2: Using adf.test

adf.test(airmiles,k=0,alternative = "stationary")


    Augmented Dickey-Fuller Test

data:  airmiles
Dickey-Fuller = -1.1415, Lag order = 0, p-value = 0.8994
alternative hypothesis: stationary

p-value seems to be greater than 0.05 so i differentiate and then conduct same test

adf.test(diff(airmiles),k=0,alternative = "stationary")


    Augmented Dickey-Fuller Test

data:  diff(airmiles)
Dickey-Fuller = -5.4406, Lag order = 0, p-value = 0.01
alternative hypothesis: stationary

and so value is less now but in case of kpss.test

kpss.test(diff(airmiles))   KPSS Test for Level Stationarity

data:  diff(airmiles) KPSS Level = 0.83442, Truncation lag parameter = 1, p-value = 0.01

The p-value is already less than 0.05 and i am concerned about which tests should i actually work with and which one leads to a better model at the end.

回答1:

A differencing model with an AR1 and 2 outliers dealing with period 10 and 22 would be a good model. Notice there is no constant.

来源：https://stackoverflow.com/questions/44406875/stationary-test-issue