I have in my possession price and time of different trade from an auction. The price series isn't stationnary so I work with the log return series. I'd like to forecast the evolution of the log return series thanks to models such as AR, MA and ARMA. To identify the parameters of AR and MA models, I used the ACF and PACF plots. However, I don't understand their signification because they're almost identical. I know how to find the p and q parameters from these plots but in this case, I don't know how to do it and which of the models can be use.

Here is a picture of the ACF and PACF plot https://i.stack.imgur.com/PhbFj.jpg

  • $\begingroup$ "The price series isn't stationnary so I work with the log return series." Whether the price is stationary or not has nothing to do with whether you are interested in log(price) or not. $\endgroup$ – Alexis Feb 9 at 19:04
  • $\begingroup$ There doesn't seem to be enough autocorrelation in your series. Did you try a Durbin-Watson or Ljung-Box test to see whether it's possible to use a time series method at all? $\endgroup$ – Digio Feb 9 at 22:10
  • $\begingroup$ @Digio It is not only possible to use time series models (e.g., error correction models) with stationary data, it can actually be desirable to do so. See Boef, S. D., & Keele, L. (2008). Taking Time Seriously. American Journal of Political Science, 52(1), 184–200. $\endgroup$ – Alexis Feb 10 at 5:04
  • 1
    $\begingroup$ @Alexis, obviously. But the Ljung-Box test is for autocorrelation, not stationarity, two different things. How would you go about modelling time information when there's not enough of it? $\endgroup$ – Digio Feb 10 at 7:42
  • 2
    $\begingroup$ @Alexis, regarding your first comment, log indeed is irrelevant; the key word is returns, i.e. not the level of price but its first difference. $\endgroup$ – Richard Hardy Feb 10 at 15:29

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.