Getting More Insight into Your Forecast Errors with the GLMSELECT and QUANTSELECT Procedures

Authors

Gerhard Svolba, SAS Institute Inc. Austria

Paper SAS1673-2018 - SAS Global Forum in Denver, CO

Abstract

Is it sufficient just to monitor the quality of your forecast models over time? Can data science methods identify the drivers for large forecast errors and provide more insights than descriptive statistics? Do demand planners really improve forecast accuracy with their manual overwrites? Using a real-life case study, this paper answers these questions. It shows how you can study the impact of factors like product group, forecast horizons, seasonality, or the forecast model type on forecast accuracy and convert them into actionable results. You learn how univariate methods provide first insights into the structure and relationships of your forecast data. You gain insight into how manual overwrites of the statistical forecast change forecast accuracy in both directions and how you use analytical and graphical methods to illustrate these findings. You see how multivariate analytical methods like linear and quantile regression provide additional relevant insight. You learn how to use the GLMSELECT, QUANTSELECT, and QUANTREG procedures to identify the most important influential factors on the forecast error. You see how you can enhance and interpret the output of these procedures to quantify the effects of the influential factors. You learn how to convert the results from the SAS® procedures into actions to improve your forecasting process. The paper shows an outline of how to use the REGSELECT and QTRSELECT procedures to apply these methods in SAS® Viya®.

Watch the presentation

Download the full paper

https://www.sas.com/content/dam/SAS/support/en/sas-global-forum-proceedings/2018/1673-2018.pdf

Download the SLIDES

Navigate to https://github.com/gerhard1050/DataScience-Presentations-By-Gerhard and download presentation #111.

Sample Graphs

Conclusion

You have seen that the application of analytical methods provides many relevant insights to help you make better business decisions. This is not only the case for the analysis of the forecast error as presented in this paper, but also for many other business questions. Svolba 2016 presents a collection of examples and SAS code where relevant business questions are analyzed with analytical methods.
In the example presented here, you have seen that the descriptive method also provides a lot of insight. Using linear regression enables you to better quantify the importance of different influential factors and to assess the strength and the direction of different categories. You see that the multivariate analysis provides a more comprehensive picture than the isolated univariate analysis of influential factors.
Quantile regression enables you get a clearer picture about the extremes of your distribution. You learn which influential factors trigger the fact that forecast errors do not exceed a certain limit. In the above example you have seen that some variables are important to explain the higher quantiles of the outcome but not the lower quantiles of the outcome.
The SAS platform with SAS9 and SAS Viya procedures provides a comprehensive set of analytical methods that enable you gain more insight in the relationships between your data and your business processes.

References

Svolba, Gerhard. 2017. Applying Data Science: Business Case Studies Using SAS®. Cary, NC: SAS Institute Inc.
Fildes R., P. Goodwin, M. Lawrence, and K. Nikolopoulos K. 2009. “Effective Forecasting and Judgmental Adjustments: An Empirical Evaluation and Strategies for Improvement in Supply-Chain Planning.” International Journal of Forecasting 25(1): 3–23 (DOI: 10.1016/j.ijforecast.2008.11.010).
Gilliland, M. 2010. The Business Forecasting Deal. Hoboken, NJ: Wiley.
Svolba, Gerhard. 2006. Data Preparation for Analytics Using SAS®. Cary, NC: SAS Institute Inc. Link
Svolba, Gerhard. 2012. Data Quality for Analytics Using SAS®. Cary, NC: SAS Institute Inc. Link

Acknowledgements

Many people have helped and inspired me to write and to complete this paper: Bob Rodriguez, Paul Goodwin, Mike Gilliland, Mihai Paunescu, Albert Tösch, and Robin Langford.