I would have a question concerning the modeling of a non-stationary time series. I do know some of the models for stationary time series such as AR MA ARMA ARCH or GARCH, but what if the time series is non-stationary. I know that it is possible to make it stationary by differencing, but maybe there would be a way to do it differently.

My series is kind of constant (it represents a baseline) but follows three trends/patterns : - First, increasing - Second, kind of constant - And third, decreasing

Would you have an idea of how to represent it elegantly via a time series (or others) ?

Thanks in advance !

  • $\begingroup$ Sometimes State space models, such as the BSM (Basic structural model) are used for time series, including some kinds of nonstationary series; they have a number of nice features and give decompositions that are fairly easily understood. This paper covers the ideas. The book by Harvey on Forecasting, Time Series and the Kalman Filter gives more explanation. $\endgroup$ – Glen_b Jun 5 '13 at 11:46
  • $\begingroup$ Alex, You can use stochastic and deterministic modeling techniques to handle this. It sounds like you could use trend variable, level shift and then another trend variable. Post your data set and we can take a look. $\endgroup$ – Tom Reilly Jun 14 '13 at 14:17

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.