Forecasting 101: Going Beyond Automatic Forecasting
Part 1: An overview of automated forecasting

A time series forecasting method is a forecasting technique that bases the forecast solely on the history of the item being forecasted. Many organizations utilize automated time series algorithms to generate forecasts. These time series forecasts are typically either used directly or adjusted judgmentally to establish the final forecasts.

In a series of three articles, Forecasting 101 will discuss the pros and cons of automatic time series forecasting and examine ways to improve your forecasts using more sophisticated approaches. This first article overviews automatic time series approaches, examines cases where there may be better approaches and discusses your options for generating alternative forecasts.

How Prevalent is the Use of Automated Time Series Methods?
The majority of forecasting software packages offer both automatic and user-specified forecasting options. For example, Forecast Pro includes an expert selection mode whereby the program automatically analyzes your data, selects an appropriate forecasting technique and generates the forecasts. Alternatively you can specify that a specific forecasting approach be used. 43% of Forecast Pro users surveyed replied that they use expert selection "always" and another 44% replied that they rely on it "most of the time."

Are All Automatic Approaches the Same?
An automatic time series forecast system might be as simple as an Excel spreadsheet that adds a percentage to last year’s sales or as sophisticated as a dedicated forecasting package that automatically selects among different statistical forecasting models for each individual item being forecasted.

The accuracy of the forecasts generated by automatic systems can vary dramatically. Excel-based systems rarely perform as well as dedicated forecasting systems that utilize proven statistically-based time series methods such as exponential smoothing models, Box-Jenkins, Croston’s and others. Empirical studies have even shown substantial differences in accuracy between commercial forecasting packages utilizing the same types of models. These differences stem from many sources including how the selection algorithms work, how the model parameters are optimized and how the models are initialized.

Ideally, when selecting an automated forecasting approach to implement, you would benchmark the forecast accuracy against your current system and against all software systems under consideration.

When Should an Automated Time Series Approach Not be Used?
In many cases automated time series approaches work quite well and, as stated previously, it is common for businesspeople to rely upon them. However, you must keep in mind that an automatic algorithm views your data as a series of numbers and takes a purely statistical approach to generating the forecasts. At times your knowledge of your products and future events may lead you to either adjust the automatically generated forecast judgmentally or reject it completely and use an alternative forecasting method. Let’s examine several cases where overriding an automatically generated forecast should be considered.

1. When a non-time series method is called for.
Time series methods work by capturing patterns in the historical data and extrapolating those patterns into the future. Time series methods are appropriate when you can assume a reasonable amount of continuity between the past and the future. They are best suited to shorter-term forecasting (say 18 months or less). This is due to the underlying assumption that future patterns and trends will resemble current patterns and trends. This is a reasonable assumption in the short term but becomes more tenuous the further out you forecast.

There are many situations where time series methods should not be considered. These include new product forecasting, forecasting products subject to promotions and/or business interruptions and forecasting products which are being driven by explanatory variables such as price, economic indicators, etc.

2. When you feel that the system has selected the wrong forecasting method.
There may be times a forecasting method is selected that you feel is inappropriate. For instance, the system may elect to use a non-seasonal model to forecast data that you know are seasonal. In these cases you will want to override the automated system and use a technique you feel is better suited. Often times, electing to use a specific form of exponential smoothing to reflect the desired trend and seasonal pattern is a good solution.

These cases tend to occur more frequently when working with short data sets where the ability to test the data statistically is limited--so keep a close eye on your automated system when forecasting short data sets.

3. When your knowledge of future events is not captured in the statistical model.
At times you may have knowledge of future events that are not captured in the forecasting model. For instance, there may be a planned promotion, a competitor entering the market or a pre-booked one-time sale. In these instances, you’ll want to treat the automatic forecast as a baseline and judgmentally adjust the forecast to reflect the future event.

Coming Next: Improving Your Forecasting via Event Models.
In the next installment of Forecasting 101 we will explore how event models can be used to improve upon automatic time series forecasts. These models provide adjustments for special events like promotions, strikes, moveable holidays and other irregular occurrences.

About the author:
Eric Stellwagen is Vice President and co-founder of Business Forecast Systems, Inc. (BFS) and co-author of the Forecast Pro software product line. He consults widely in the area of practical business forecasting—spending 20-30 days a year presenting workshops on the subject—and frequently addresses professional groups such as the University of Tennessee’s Sales Forecasting Management Forum, APICS and the Institute for Business Forecasting. Recognized as a leading expert in the field, he has worked with numerous firms including Coca-Cola, Procter & Gamble, Merck, Blue Cross Blue Shield, Nabisco, Owens-Corning and Verizon, and has served on the board of directors of the International Institute of Forecasters (IIF).