I have quarterly data on inflation from 1990 Quartal 1 to 2016 Quartal 3.

If I want to perform the pseudo out-of-sample forecasting one quarter ahead with an autoregressive function, do I have to estimate the model until 2016 Quartal 2?

When choosing how many lags to include in the AR model (by looking at BIC and AIC) do I have to consider the whole sample or just until 2016:2?


You should estimate the model with sample until 2016:2 and make a forecast for 2016:3. It is better explained here.

Have you considered to make the pseudo-out-of sample forecasts for longer period? Let's say for all quarters in 2016?

  • $\begingroup$ I considered it, yes. Let's say i want to forecasting all the four quarter of 2016, in that case should I use data up until 2015:4? And could I use the lag order determined before when forecasting 2016:3, or should I reengage in model selection considering the sample 1990:1-2015:4? $\endgroup$ – Federica Dec 13 '16 at 15:50
  • $\begingroup$ Usually it makes sense to use ARMA(p,q) models for forecasting not further than p or q periods forward. So, if you have ARMA(1,1), you should make a forecast for the next period only. $\endgroup$ – damian Dec 13 '16 at 16:16
  • $\begingroup$ Then, there is a question: how do you want to use your model? If you want to estimate model everytime when new data are available, then I would reestimate model 3-times and make one-step forward predictions for 2016:1-2016:3. If you want to use the same model with the same parameters, then you can estimate the model with data until 2015:4 and make predictions for 2016:1-2016:3. However, prediction in time 2016:2 will be based on 2016:1 and I recommend using actual value, not predicted. You can read about this here: davegiles.blogspot.sk/2013/11/… $\endgroup$ – damian Dec 13 '16 at 16:29
  • $\begingroup$ Thank you for your explanation, I've succeeded in making the one-quarter and three quarters ahead forecasts. However, I have a last question: I would like to make also the 2 quarters ahed forecast, and I begin by estimating the model up until 2015:4 and make a forecast up to 2016:2. Since my forecasting period is 2016:1-2016:3, is the forecasting exercise over or should I estimate the model using observation until 2016:2 and make the forecast fore 2016:3? In this case wouldn't it be just a one period ahead forecast, since i don't have observation for 2016:4? $\endgroup$ – Federica Dec 24 '16 at 9:11

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.