FORECASTING:FORECASTING IN PRACTICE

FORECASTING IN PRACTICE

You now have seen the major forecasting methods used in practice. We conclude with a brief look at how widely the various methods are used.

Every company needs to do at least some forecasting, but their methods often are not as sophisticated as with these major projects. Some insight into their general approach was provided by a survey conducted some years ago2 of sales forecasting practices at 500 U.S. corporations. Although this survey published in 1994 now is somewhat out of date, we believe that its results are still somewhat reflective of current forecasting practices.

This survey indicates that at that time, judgmental forecasting methods were some- what more widely used than statistical methods. The main reasons given for using judgmental methods were accuracy and difficulty in obtaining the data required for statistical methods. Comments also were made that upper management is not familiar with quantitative techniques, that judgmental methods create a sense of ownership, and that these methods add a commonsense element to the forecast.

Among the judgmental methods, the most popular was a jury of executive opinion. This was especially true for companywide or industry sales forecasts but also holds true by a small margin over manager’s opinion when forecasting sales of individual products or families of products.

Statistical forecasting methods also are fairly widely used, especially in companies with high sales. Compared to earlier surveys, familiarity with such methods is increasing. (Given that statistical forecasting methods now have been regularly taught in business schools and management seminars for many years, we anticipate that this trend of in- creasing familiarity with such methods has continued since the time of the survey.) How- ever, many survey respondents cited better data availability as the improvement they most wanted to see in their organizations. The availability of good data is crucial for the use of these methods. (Fortunately, the rapid advances in information technology since the survey was conducted has led to much better data availability in many companies.)

The survey indicates that the moving-average method and linear regression were the most widely used statistical forecasting methods. The moving-average method was more popular for short- and medium-range forecasts (less than a year), as well as for forecasting sales of individual products and families of products. Linear regression was more popular for longer-range forecasts and for forecasting either companywide or industry sales. Both exponential smoothing and the last-value method also received considerable use. However, the highest dissatisfaction is with the last-value method, and its popularity was decreasing compared to earlier surveys.

When statistical forecasting methods were used, it was fairly common to also use judgmental methods to adjust the forecasts. (This continues to be fairly common practice.)

As managers have continued to become more familiar with statistical methods, and more used to using the computer to compile data and implement OR techniques, we anticipate that the usage of statistical forecasting methods is continuing to grow. However, there always will be an important role for judgmental methods, both alone and in combination with statistical methods.

Another important trend in recent years has been an increasing availability and usage of sophisticated software packages for applying statistical forecasting methods. (See Selected Reference 11 for a survey of these packages.) Selected Reference 10 also provides a survey of the use, satisfaction, and performance of forecasting software in practice. Most of the U.S. corporations responding to the latter survey reported using software for their forecasts, although this sometimes involved using only spreadsheets or internally developed forecast- ing software. Those using commercially available software packages reported both the best forecasting performance and the greatest satisfaction with the features of the software.