Hi everyone,

I have a variable with daily observations that is stationary according to dickey fuller tests. I am collapsing the data into yearly observations using the average per year. Now the dickey fuller test states that the data is non-stationary, thus has a unit root. Accordingly, the results of my tests could be spurious.

Specifically, I am using panel vector autoregressions (pvar) and the stability tests imply that the results are not stable (Eigenvalues > 1). I assume its due to the now non-stationary variables.
How to proceed? Applying first difference creates very ambiguous results. Interestingly, the results with the variable in levels are stable when i include enough lags of the variable.

Any help would be most appreciated

Best,
Micha