ARIMA Model - AR(4)AR(364) with MA(7) - too big for MATLAB - how to handle?

I have daily data which has a yearly seasonal component AR(364), but when I try to enter the orders AR 348 and MA 7, I get an error. Does anyone know how to accomodate large AR models.

PLUS, I really only want the parameters AR 1, 2, 3 and 4 and 364 calculated NOT all the intemediate values. How can I specify JUST these AR paramters in the ARMAX function?

I thought MatLab 2011a supported this possibility but I cannot find any doc.

Thanks for any help. Dave

Answer 1

[Rajiv Singh]

Try nlarx. Choose orders [0 0 0], and specify all regressors using "custom regressors". For example:

model = idnlarx([0 0 0],[],'CustomReg',{'y1(t1-1)','y1(t-2)','y1(t-3)','y1(t-4)', 'y1(t-364)'});

model_estimated = nlarx(data,model)

The estimated coefficients are located under model_estimated.Nonlinearity

Limitation: This only lets you create AR models with arbitrary delays.

For handling seasonality (ARIMA), see the "IntegrateNoise" feature of ARMAX command in R2012a release.