Unit root test results differ for quarterly and monthly data
Posted: Thu Sep 04, 2014 3:42 am
Hello everyone,
I have run some unit root tests (ADF and KPSS) for a few time series and I get different results depending on what frequency I use. For the quarterly data I seem to get a lot more results that indicate stationarity (no unit root).
Here an example with the EONIA-dataset (from the ECB stats-website). I have attached a picture of both timeseries. As you can see the graphs are obviously similar. (The quarterly data is calculated as an average of the monthly values) The quarterly data has 61 observations and the monthly data has 183.
Now, if I run an ADF test, with trend+intercept and SIC criterion I get the following result:
- The monthly data is not significant (p-value of 0.2401). The monthly data seems to have a unit root, it is non-stationary.
- The quarterly data is significant (p-value of 0.0265). The quarterly data has no unit root, it is stationary.
How come those results are so opposite for practically the same data?
-
I have run some unit root tests (ADF and KPSS) for a few time series and I get different results depending on what frequency I use. For the quarterly data I seem to get a lot more results that indicate stationarity (no unit root).
Here an example with the EONIA-dataset (from the ECB stats-website). I have attached a picture of both timeseries. As you can see the graphs are obviously similar. (The quarterly data is calculated as an average of the monthly values) The quarterly data has 61 observations and the monthly data has 183.
Now, if I run an ADF test, with trend+intercept and SIC criterion I get the following result:
- The monthly data is not significant (p-value of 0.2401). The monthly data seems to have a unit root, it is non-stationary.
- The quarterly data is significant (p-value of 0.0265). The quarterly data has no unit root, it is stationary.
How come those results are so opposite for practically the same data?
-