Why are all coefficients of my seasonal ARIMA = 0 or NaN?

Hello everyone!

I created a seasonal ARIMA model with the following code:

 % model = arima('Constant',0,'D',1,'Seasonality',24,'ARLags',3,'SARLags',3,'MALags',2,'SMALags',2); 
[fit,VarCov,LogL,info] = estimate(model,Y);

Now matlab gives me the following output:

      Parameter       Value          Error       Statistic 
    -----------   -----------   ------------   -----------
     Constant              0         Fixed          Fixed
        AR{3}     -0.0321142    0.00310401        -10.346
       SAR{3}     -0.0321142    0.00310401        -10.346
        MA{2}      -0.138189    0.00201315        -68.643
       SMA{2}      -0.138189    0.00201315        -68.643
     Variance       0.921722    0.00357711        257.672

When I try to retrieve the coefficients of my polynomials with for example

 % model.SAR

I only get zeros and NaN as coefficients. Does anyone know what the problem could be?

Thanks a lot!! I am very new to matlab, so help is greatly appreciated!!

Answer 1

[Roger Wohlwend]

The variable model specifies the model but it does not contain any values for the coefficients (except the constant). That is why you only get zeros and NaNs. The estimated model is the variable fit . You have to retrieve the coeffiencts from that variable.

SAR = fit.SAR

[MC3105]

Thank you so much for your answer!

When I retrieve my coefficients now, I only get a coefficient for the last polynomial. For example for

 fit.SAR 
i get the output

 [0] [0] [-0.0321142]
meaning that I basically only have a SAR polynomial with one variable (the variable at lag 3).

Normally a seasonal ARIMA that is specified as (3,1,2)x(3,1,2) should have 3 AR and 3 SAR variables as well as 2 MA and 2 SMA variables (one variable for each lag), right?

Is there a different way to tell matlab what I want exactly?

[Roger Wohlwend]

You have to include all lags in the specification, not just the highest one. With the following line of code Matlab will estimate the model you want:

model = arima('Constant', 0, 'D', 1, 'Seasonality', 24, 'ARLags', 1:3, 'SARLags', 1:3, 'MALags', 1:2, 'SMALags', 1:2);