I want to test whether a time series $\{x_1,x_2,\dots,x_t\}$ is stationary (weak stochastic stationarity/WSS). I first want to show that the time series has no deterministic trend and then use an ADF (augmented Dickey-Fuller) test to show that it is not a unit root process. From eyeballing, I know that the time series has an at most linear deterministic trend, but there may be several periodic signals of different frequencies present.

I would like to use the following approach to test whether my data is stationary:

Step 1A: I make an OLS fit of my time series with the model $x_t = a+bx_t+e_t$, where $e_t$ is the error term.

Step 1B: I show that there are no endogenous variables by showing that there are $e_t$ has no serial correlation. (I can use a Durbin-Watson, Breusch-Godfrey, Ljung-Box, or similar test for this.) This shows that I do not need to include periodic deterministic trends in my model.

Step 1C: I use a t-test to show that b=0. This shows that there is no linear deterministic trend either.

Step 2: I use an ADF test to show that the time series is not a realisation of a unit-root process.

Is this a valid approach to test my data for stationarity? Are there approaches that would be easier, more robust, or more common?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.