topic SAS version of R-DLNM package in SAS Forecasting and Econometrics

SAS version of R-DLNM package

PamG — Sun, 17 Sep 2023 07:14:40 GMT

I wasn't able to find any information online about using distributed lag models in SAS. This is different from what PROC PDLREG does and is used extensively to to model ambient temperature and pollution data. I was wondering if SAS has any plans to introduce a procedure for this.

Re: SAS version of R-DLNM package

sbxkoenk — Sun, 17 Sep 2023 08:10:14 GMT

Hello,

[EDIT : Have moved this to "SAS Econometrics and Forecasting" - board by the way ]

DLNM = Distributed lag non-linear models
(it's like PROC PDLREG , but allowing for more flexibility in the relationship between regressors and target time series)

Maybe @SASCom1 or @TammyJackson or @StatsMan can answer that question (on having R-DLNM in SAS).

DLNM is a modeling framework to flexibly describe associations showing potentially non-linear and delayed effects in time series data, thus a framework for estimating dynamic relationships.

Just note that this can be done in several other ways within SAS !!

Koen

Re: SAS version of R-DLNM package

PamG — Sun, 17 Sep 2023 09:28:58 GMT

Thanks for the response Koen. Could you elaborate the ways one can do similar analysis in SAS? How does one go about creating the crossbasis functions that can incorporate the temperature along the non-linear and lagged dimension ? I need to do conditional logit on individual level data. It is a case crossover design with 1:5 matching.

Re: SAS version of R-DLNM package

sbxkoenk — Sun, 17 Sep 2023 10:31:52 GMT

Hello,

Are you studying / examining the conditional logit estimator for binary panel data models with unobserved heterogeneity and serial correlation?

What is your study trying to explain / trying to predict? Natural disasters?

Thanks,

Koen

Re: SAS version of R-DLNM package

PamG — Sun, 17 Sep 2023 11:52:24 GMT

This is an ecological study where we are trying to study mortality risk due to exposure to increased ambient temperature. Most studies are conducted at an aggregate level and we are studying at the individual level.

Re: SAS version of R-DLNM package

PamG — Sun, 17 Sep 2023 12:32:24 GMT

Tagging @Rick_SAS as well in case he has any information.

Re: SAS version of R-DLNM package

TammyJackson — Mon, 18 Sep 2023 04:59:42 GMT

There is some support for identifying delayed effects in HPFDIAGNOSE (SAS Foundation) and the DIAGNOSE Object of the ATSM Package (SAS Viya). Look for DELAYINPUT and DELAYEVENT.
https://go.documentation.sas.com/api/collections/pgmsascdc/v_042/docsets/castsp/content/castsp.pdf?locale=en#nameddest=castsp_atsm_sect030

'DELAYEVENT' takes a nonnegative numeric Value that specifies the event variable lag.
'DELAYINPUT' takes a nonnegative numeric Value that specifies the input variable lag. If not specified,
the delay lags for the inputs are automatically chosen.

Another option would be to call R from SAS. However, if there is support for this method in Python, that would be easier to integrate with SAS.

Here is a blog that shows using Python from SAS.

https://communities.sas.com/t5/SAS-Communities-Library/Improve-Your-Nowcasting-in-SAS-Visual-Forecasting-Using-Real/ta-p/849697

and the original slides from the Federal Forecaster's Conference:

https://www2.gwu.edu/~forcpgm/Jackson_slides.pptx