I'm looking at two time series and I would like to determine how they move together. They both are however stationary. Would it make sense to test for cointegration? Linear relationship between both series has a very low R^2 and correlation but I think that there might be a lag or something and that they move together.


No. Cointegration is a property of nonstationary series. Let's call the variables $y_t$ and $x_t$. Your best bet is to fit models with both lags of $y_t$ and $x_t$:

$$y_t = b_0 + b_1 y_{t-1} + ... + b_n y_{t-n} + b_{n+1} x_{t-1} + ... + b_{2n} x_{t-n} + \varepsilon_t$$

and then do likelihood ratio test on the various models, or look at adjusted $R^2$, AIC or hold out forecast performance to pick a set of plausible models. You could then pick the 'best' model, or just average the results from the top models.

| cite | improve this answer | |
  • $\begingroup$ If you also write an analogous equation for $x_t$ and consider the two equations together, you will have a VAR model. There is a large literature on VAR models out there. $\endgroup$ – Richard Hardy Jan 30 '15 at 10:55

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.