Forecast Studio vs TSFS in an ARIMA model with dynamic regressor

[andreas_zaras]

Hello,

I have a dependent and an independent variable. I let Forecast Studio do the modelling work and it concludes to an ARIMA(0,1,2) model with a dynamic regression that transfers the independent's variable influence to the dependent. The independent variable is differenced once and the transfer function is of order 2 in the numerator and order two in the denominator. The work is done throgh the HPFARIMASPEC of HPF.

I am trying to fit the same model in TSFS but i get the message that the model fitting failed. Why is that? Is it because behind TSFS the ETS proc does the work and uses different estimation methods than the HPFARIMASPEC?

Finally the GUI of Forecast Studio covers every modelling cabability of the TSFS GUI? Is there any extra modelling cabablility that one of them has and the other does not have? 

Thanks in adavnce,

Andreas

[chwell · Accepted solution]

Hi Andreas.

The parameter estimation algorithms in Forecast Studio and ETS (TSFS) are very similar; a model that is successfully fit in FStudio should not have problems being fit in ETS and vice versa. The first thing I'd do is double check to make sure the specifications are identical. That is, is the dependent variable first differenced in your ETS specification, ... .

Something else that can be tricky is that pure delays in independent variables in a transfer function are called 'shifts' in ETS and are denoted, e.g.; in PROC ARIMA (1 $ input1, ..). Pure delays are called 'delay' in Procs that run under the hood in Forecast Studio, e.g. PROC ARIMASPEC.

The TSFS system is a handy tool, but it's functionality is very limited relative to Forecast Studio. The TSFS system allows users to fit ARIMA and ESM models from prespecified lists and specify their own custom ARIMAX models. The procedures that run 'under the hood' in Forecast Studio can do this plus: automatically identify or build a transfer funciton and error component specificaiton from scratch based on patterns in the data, identify UCM specifications, do automatic model selection, create, fit estimate and select the best models for hunderds of thousands of series .... The list goes on. If you would like to learn more about Forecast Studio, let me know.

Best, Chip

[Ania14]

Hello Chip,

I am currently researching the SAS Forecast Studio and was wondering if you could give me some pointers. My main concern is: Can SAS FS run all possible combinations and  permutations from the list of my independent variables (with their lags), to arrive with best fit models automatically?

Thank you

Ania