I have 365 days of hourly data (24points each) of a prediction error (realised -pred_day_before).
I want to model the evolution of the prediction error as an ARMA process.
Matlab System Identification Toolbox offers an ARMA estimation PER time series, whilst I have 365 time series (with supposedly the same characteristics).
How to estimate just one ARMA model for all of the 365 rows of 24 values?