turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Forecasting
- /
- How to forecast a mean revert process

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

03-10-2014 04:25 PM

HI,

I Want to forecast a mean revert process. The process is as following, y(t) = y(t-1) + a(b-y(t-1)). Here y(t-1) is lagged value for y(t). This is actually an AR(1) process. I have a data set with y value from 1960/01 to 2000/12. I want to use these data to estimate a mean revert (AR(1)) process, then I want to forecast y from 2001/01 for 10 months. How to do this? I searched and found that I can use proc model to estimate this model, but don't know how to forecast. Also I found that I may use proc arima to estimate and forecast, but don't quite know how to.

thanks.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to SeanZ

03-10-2014 05:12 PM

Hello -

If you have access to SAS/ETS software you may want to use the ARIMA procedure to model this process.

Here is an example using SASHELP.AIR:

proc arima data=sashelp.air;

identify var=air;

estimate p=1;

forecast id=DATE interval=MONTH lead=12 out=work.forecast;

quit;

Of course AR(1) is not a good model for this data, but you'll get the idea.

You will find many more examples here: http://support.sas.com/rnd/app/examples/ETSexamples.html

Thanks,

Udo

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to udo_sas

03-10-2014 05:45 PM

Thank you for your help. In my dataset, I have one variable date showing the date of observations. These dates may not continue. For example, 2001/01, 2001/02, 2001/04, March is missing. But it's ok for me since I will use 2001/02 as the lag of 2001/04. There is no variable month. Do I have to include 'interval=month' in the codes? What is this for? Also, I found some examples using crosscorr= in the code. What is the difference if I change identify part into

identify var=air crosscorr = (date);

Thanks.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to SeanZ

03-11-2014 11:38 AM

Hello -

In order to use PROC ARIMA (or any other forecasting procedure) you will make to make sure that your data is in time series format - with other words no gaps in the time dimension are allowed. It is ok to have missing values for the analysis variable for not observed periods (such as no value for March in your example), but your variable DATE needs to be modified. You could use the TIMESERIES procedure for example - see for example: http://support.sas.com/documentation/cdl/en/etsug/66840/HTML/default/viewer.htm#etsug_timeseries_exa....

Yes, you will need to tell the procedure about the periodicity of your data - in the case of the airline data (sashelp.air) the periodicity is monthly. See: http://support.sas.com/documentation/cdl/en/etsug/66840/HTML/default/viewer.htm#etsug_tsdata_sect029....

In your case ID will need to point to your column DATE and INTERVAL will be month.

CROSSCOR allows you to specify and cross-correlate input series or ARIMAX models. See: http://support.sas.com/documentation/cdl/en/etsug/66840/HTML/default/viewer.htm#etsug_arima_gettings... for more details.

Thanks,

Udo