I am creating and using OLS Regression models using historical data to forecast quarterly balances for banking products (loans, mortgages, deposits, etc) for the Dodd-Frank/CCAR exercises. One problem we have run into is that sometimes the last historical time period's value (jump-off point which is used to start forecasting from) can be unexpectedly high or low (possibly due to a business action like a temporary interest rate change for marketing purposes or maybe an unusual event in the marketplace). This creates a problem in that the forecast generated is unusually high or low due to the out of the ordinary jump-off point. Any ideas on how to adjust for this in the model? An initial thought is to forecast from a previous data point where the growth rate of the data point is within a certain acceptable range. Any thoughts, ideas, or references to scholarly papers on this topic would be helpful. Thanks.
I don't have time to figure out what a "Dodd-Frank/CCAR exercise" is, but identifying outlier points in OLS regression is best done with proc robustreg. Check the OUTLIER= option in the OUTPUT statement.
Thanks for the reply PG. Actually Dodd-Frank/CCAR can be put aside. Basically I have a historical time series of quarterly data that I used to develop an OLS regression equation, and I am using the OLS regression equation to forecast future time points. The problem is that the last historical time point IS an outlier. So my question is if I still want to use that last (outlier) point as my jump off point to start forecasting, is there a way to adjust either that point or the future forecast to take that issue into account? Currently my forecast is unreasonable since the jump off (last historical) point was unreasonable to start with.
The idea would be to reduce the importance of that last point in your forecasting equation, either by removing it, downweighting it, or diluting it by using more terms in your model.
You could use, say, the 4-quarter trailing average for the jump-off date.
Are you ready for the spotlight? We're accepting content ideas for SAS Innovate 2025 to be held May 6-9 in Orlando, FL. The call is open until September 16. Read more here about why you should contribute and what is in it for you!
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.