BookmarkSubscribeRSS Feed
lizzy28
Quartz | Level 8

Hi,

 

I am trying to estimate a VAR model with an exogenous variable. Y1 and Y2, for instance, auto sales and gas price, are endogenous, which I want to predict 4 periods ahead for. X is the exogenous varaible, for example, auto retail events, and 1 means there is an event that time period. The events have been planned out at the beginning of each year. 

 

Is it possible to make forecast for the sales of the next year with the future event indicated?

 

Please see my example data as attached.

 

Thank you!

7 REPLIES 7
SteveDenham
Jade | Level 19

Try posting this in the Forecasting and Econometrics community.

 

Steve Denham

ChrisHemedinger
Community Manager
Good idea -- I moved it for you.
It's time to register for SAS Innovate! Join your SAS user peers in Las Vegas on April 16-19 2024.
lizzy28
Quartz | Level 8
Thank you, Chris and Steve!
JohndeKroon
Obsidian | Level 7

Hello Liziwu,

 

What do you want to do exactly? I sense a kind of two step approach:

  • VAR model to predict Y1 and Y2 into the future
  • Use these future estimates of Y1 and Y2 determine the value of X (through simple regression?)

Is this correct? Or do you want to do something else? If so, please state...

 

John 

udo_sas
SAS Employee

Hello -

This example might get you started: http://support.sas.com/rnd/app/examples/ets/tourism/index.htm

 

Having said this please allow me to propose an alternative for you to consider. First I'd like to point you to some research on usefulness of complexity in forecasting models, which is worth reading: http://www.kestencgreen.com/simplefor.pdf


I'm assuming that your goal is to forecast sales - and take gas price and events into account as dependent variables.

If that is the case than the following approach might work:

Build an ARIMAX or UCM model which uses gas price and event as inputs and sales as output. These models will be able to incorporate cross-correlation effects (between price and sales for example) in a dynamic fashion. Some people refer to these models as dynamic regression. The more tricky part is too figure out if price and sales are cross-correlated indeed.

Of course you will also need to forecast the gas price - either by a statistical model such as exponential smoothing or by assuming future values yourself.

After successfully building your model - and assuming price was significant - you can run some what-if scenarios (change the future values of price) to see how your model is behaving.

There are several ways to do such analysis in SAS - SAS Visual Analytics provides this functionality to some extend. SAS Forecast Server (or SAS Forecasting for Desktop) would be my tool of choice - but I'm biased. SAS/ETS provides you to a couple of procedures which will allow you to implement the approach above yourself - check out the following procedures: TIMESERIES, ARIMA, UCM, and ESM.

Of course if you don't agree with my statements above you can also revert back to VARMAX and start with the example I shared earlier.

Happy Friday,

Udo

lizzy28
Quartz | Level 8

Thanks, John!

 

What I want to do is to use the existing data of y1, y2 and X to predict future values of y1 and y2. I also want to use the future values of X which is known already to make better prediction of y1 and y2. Is it possible to do the second part?

SteveDenham
Jade | Level 19

See @udo_sas's response, especially regarding UCM forecasting.  You are still going to have to do something to forecast future gas prices  (NOTE: If you are successful in that alone, you have accomplished something that has continually eluded many of the best economists and commodity traders in the world.)

 

Steve Denham

sas-innovate-2024.png

Join us for SAS Innovate April 16-19 at the Aria in Las Vegas. Bring the team and save big with our group pricing for a limited time only.

Pre-conference courses and tutorials are filling up fast and are always a sellout. Register today to reserve your seat.

 

Register now!

Multiple Linear Regression in SAS

Learn how to run multiple linear regression models with and without interactions, presented by SAS user Alex Chaplin.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 7 replies
  • 1708 views
  • 4 likes
  • 5 in conversation