Deciding sufficient lags (input variables) is very tedious in modelling non-linear process. Which one i.e., ACF and PACF is best for deciding sufficient lags from the time series data. Can anyone help me to short out this problem from the above-mentioned question.
If you are using ARIMA methodology then you really need to look at both. If the ACF function decays slowing from either direction and the PACF ends abruptly that would show your AR components. The opposite shows the MA components , where the ACF cuts off abruptly and the PACF decays slowly. I would first determine how much memory you need in your forcast, I find ARIMA fairly good for short term things, but with longer memory processes I look at a measure of persistence such as a Hurst Exponent.
- Badges
- Science topic
- Similar topics
- Geoscience
- Atmospheric Sciences
- Atmosphere