topic Re: Autogerression direction of the AR estimate in SAS Forecasting and Econometrics

Autogerression direction of the AR estimate

Taliah — Wed, 30 Oct 2024 11:23:30 GMT

Hello

I'm running proc autoreg. In the results the estimate of AR1 is negetive (-0.58)

The data is such that I would expect that the higher the value of the dependent varaiable in the previous data line (coresponding to AR1) the higher the value of the dependent vaiable in the current month.

Is it correct to say that a negetive estimate of AR1 means that the higher the value of the dependent varaiable in the previous data line (coresponding to AR1) the higher the value of the dependent vaiable in the current month?

Thank you!

Re: Autogerression direction of the AR estimate

sbxkoenk — Mon, 04 Nov 2024 19:12:30 GMT

Can you show us your PROC AUTOREG code?

Do you know how AR1 (the coefficient) is used in your model?

Are you fitting a (linear) trend-stationary model like the one in :

SAS/ETS User's Guide
The AUTOREG Procedure
Example 9.1 Analysis of Real Output Series
https://go.documentation.sas.com/doc/en/pgmsascdc/9.4_3.5/etsug/etsug_autoreg_examples01.htm

Scroll down to see the formulas.
AR1 is NOT the coefficient for t, it is the coefficient for Nu(t-1) ... with Nu being the Greek alphabet letter.
AR1 is in the ERROR model (2nd equation).

BR, Koen

Re: Autogerression direction of the AR estimate

Taliah — Tue, 05 Nov 2024 10:50:59 GMT

Thank you

The code is -

proc autoreg data= file1 outest=file2 plots;

model y= x1 x2 x3 x4/

model=ML

maxiter=100

NLAG=(1)

backstep slstay=0.0500

dw=1;

output out=file3 lcl=lcl ucl=ucl p =predicted rm=r pm=pm r=residual alphacli=0.05;

run;

Does that enable an answer to what I asked above?

Thank you!

Re: Autogerression direction of the AR estimate

sbxkoenk — Thu, 07 Nov 2024 12:29:25 GMT

@Taliah wrote:

Does that enable an answer to what I asked above?

That's your question in bold on page 1.

I don't think you can make that general conclusion.

Remember you are NOT dealing with an AR(1) Model.
AR(1) model = a first-order autoregressive model.
Your model, speaking in equations, looks different. See my previous post (p.2).
If you want to fit an AR(1) model, you need PROC ARIMA (instead of PROC AUTOREG).
An ARIMA(X) model also allows for independent variables (forecast drivers).

Cheers,

Koen