# ARIMA model with splitted data

Question: When analyzing a time series of size $N$ assume that you fit models for data in the intervals $[1,N/2]$, $[N/2+1,N]$, $[1,(2/3)N]$ and $[(2/3)N+1,N]$. Discuss what this approach is for.

In this case, all intervals are disjoint and the union of those represents the complete data set.

What I read here Forecasting with Univariate Box - Jenkins Models: Concepts and Cases is that the model may well fit the data as a whole, but the distant past or recent past are poorly adjusted and therefore the predictions can be bad.

But why four models?