Augmented Dickey Fuller

[mscarlatos]

I am testing to see whether series is stationary via ADF. I get a test statistic of -2.8 vs Critical Value of -3.5 at 1%, -2.9 at 5% CV and -2.5 at 10% CV. Does this mean that my series is nonstationary at the 10% CV but statioanry at the 1% and 5% CVs?

[glu]

No. While dealing with randomness, you have to be very careful in interpretation. In fact, there is no way to draw a clear-cut conclusion whether your time series is "indeed" stationary or not. All you can is to draw such a conclusion with one or other degree of reliability. The zero hypothesis tested by the ADF test, H0, is that a series is non-stationary. Rejecting it at the 1-percent level implies that out of 100 series, you might one time reject the zero hypothesis when it is true. With the 5-percent level, the chance is 5 to 100, with the 10-percent level the chance is 10 to 100. However, these are mere probabilities; you might make mistake in the rejection or not, nobody never knows.