Printer Friendly

The forecasting process: guidelines for the industrial engineer.

Many industrial engineers feel that they cannot afford the luxury of planning because they are too busy handling one crisis after another. Although forecasting is the cornerstone of planning, engineers may even spend less time forecasting future events that relate to their managerial responsibilities than they do in planning such projects as the design of products or the development of production processes. An unplanned project may take four times longer than expected (if it gets done at all), whereas a carefully planned project may only take twice as long as was planned. Moreover, by adjusting a project's schedule to work around a variety of unforeseen events, analyses will be performed and improvements will be implemented, however, goals such as deadlines met at minimum costs are unlikely to result.

IEs must be involved in the forecasting process even though they realize that their forecasts may not be correct. Like planning, it is the process of forecasting that is beneficial. Forecasting usually incorporates both subjective and objective approaches to predicting future values. Objective procedures often focus on using statistically-based techniques to project past values into the future. The subjective approach incorporates the opinions or judgments of one or more persons who are experienced or considered knowledgeable. Both approaches need to be integrated into a comprehensive process for estimating future values or ascertaining the occurrence of future events.

The forecasting process

Predicting the future is the essence of forecasting. Estimating when future events will occur, or the values of operating characteristics or performance measures over some planning horizon are representative outputs of the forecasting process. Few individuals are able to develop accurate forecasts of the future in a consistent manner without utilizing a systematic procedure. A comprehensive forecasting process is shown in Figure 1. Engineering managers who engage in this sequential procedure should achieve benefits associated with increasingly effective decision making relative to their areas of responsibility.

Step 1. As is true with most management activities, it is critical to begin with a specific statement of the purpose for the forecast. Organizations use forecasts for a variety of purposes. For example, forecasted annual revenues could be used by the marketing manager to establish quotas for sales representatives or by the engineering manager to make capacity decisions related to anticipated staffing needs or production levels. While appropriate as a motivational technique in marketing, the optimistic orientation of an annual sales forecast might be totally inappropriate for the development of a more realistic cost-minimizing master production plan.

Most functional managers will want to develop their own set of forecasts that relate to their needs and managerial style. Thus, engineering managers should be prepared to modify existing forecasts that have been developed elsewhere in the organization in order to reflect their needs and aid in making decisions to achieve their goals. Often, this modification process just requires an adjustment for any biases inherent in the purpose of the original forecast.

Step 2. Collecting related historical data provides the opportunity to examine similar situations from an experimental perspective. Since the intent is to identify trends, seasons or cycles associated with the past that could be projected into the future, historical performance data should be purged of any onetime occurrences that probably will not reoccur in the future. For example, if a previous design effort required the staff to learn how to effectively utilize newly purchased workstations, the man-hours required for a similar design project should reflect this newly acquired proficiency. Thus, it would be very appropriate to exclude man-hours, which would reflect basic training and initial skill development, from the historical database.

Data are commonly obtained from organizational records, industrial trade associations and governmental documents. In the case where new industrial engineering capabilities are being developed, expected usage of these services needs to be estimated without the benefit of historical data. For these situations, it is frequently necessary to use historical demand or performance data from similar services that have been introduced recently.

Step 3. Displaying relevant historical data to determine if any patterns exist that would help in predicting future values is the goal of this step. Since graphical displays often reveal underlying regularities in past demands or performance levels, plotting historical data over time may produce a better understanding of previous behavior, which should improve forecast accuracy. Most engineering managers realize that visual displays of past performance measures facilitate the identification of underlying time-dependent patterns of resource consumption.

Step 4. Careful consideration is given to using one or more of the following three different types of forecasting models:

* Time Series Models;

* Causal Models; or

* Subjective Models.

These models can be used for a variety of decision making situations. Table 1 summarizes the unique characteristics of each forecasting model relative to illustrative methods, typical applications costs, required data, technology considerations and relevant planning horizons. The engineering manager should employ those methods or techniques that are most suitable for addressing their current needs. [Tabular Data Omitted]

Time series models use an item's own historical data to predict future demand values or performance levels. These methods examine past behavior of the dependent variable over time and extrapolate those values into the future. Resultant forecasts are based on projecting any observed historical patterns. Time series models assume that time is a surrogate measure for one or more independent variables that are related to the dependent variable but may be difficult to measure.

Causal models also use historical data on independent variables that typically represent the usage of related goods or services to develop a forecast for the value of interest, that is, the dependent variable. This type of forecasting model uses a variety of statistically-based methods to establish a functional relationship that depicts prevalent patterns in the data. The causal model assumes that as the system of relevant independent variables changes, a corresponding change will occur in the dependent variable. Thus, by knowing the values of the independent variables, an appropriate prediction of the dependent variable can be determined.

Specific curve-fitting methodologies within both types of models range from simple approaches such as an arithmetic average to very complex statistical models like multi-linear regression. Both time series and causal models use analytical procedures to determine underlying patterns, and thus, their validity rests on the assumption that the future will continue to behave like the past. In other words, the major effects governing past performance are assumed to remain unchanged over the current planning horizon. Typically, this assumption reflects a situation that is characterized statistically as a stationary process.

In contrast, subjective models are more qualitative and judgmental in nature. They bring together experts to develop agreements on forthcoming events or dependent variable values through brainstorming, drawing analogies and writing scenarios that may relate to the unusual, non-extractable, distant occurrences. These methodologies focus on predicting major happenings, turning points or important changes and are considered to be more speculative than the two previously mentioned models.

The engineering manager is responsible for identifying the type of forecast model that is most appropriate relative to the stated forecast purpose. Since a variety of specific forecasting techniques are available within each forecast model type, the manager needs to select one or more of these techniques that will yield an acceptable level of forecast error.

Step 5. Testing the accuracy of various forecasting techniques by having each predict recent actual performance or resources usage is the major focus of this step. This approach of ex-post testing pretends that actual data for immediate past periods are not available and uses older historical data to forecast events or values for these recent time periods. Typically, the manager selects the technique that produces the smallest average forecast error and uses it to forecast over the planning horizon. These procedures constitute a logical approach for the short term (1 to 3 months) or the intermediate term (3 to 24 months) forecast. Long-term forecasts often require agreements between members of an expert panel to "identify the best approach." The specific technique selected should be sufficiently accurate and reliable for the purpose stated in Step 1. In addition, it should be cost-effective, and perhaps most importantly, readily understood by those managers using the results.

There are several common approaches to calculating the error associated with a forecast that are regularly used with time series or causal models. Four particularly popular measures of forecast error are:

* Residual Error

* Mean Absolute Deviation Error

* Mean Absolute Percentage Error; and,

* Mean Squared Error

Simple forecast error or the residual is computed by subtracting the forecasted value from the actual value. Individual error values show both magnitude and direction while the error average incorporates the cancelling effort of positive and negative deviations. A second error measure is the absolute value of simple forecast error which reveals only the magnitude of error. The mean absolute deviation (MAD) records the average of the absolute values of differences between the actual and forecasted values. The absolute value of forecast error may also be recorded in percentage terms by dividing the absolute value of the error by its actual value and multiplying by 100 percent. This third approach to measuring forecast error places the resulting error in perspective to the size of the value being forecasted. The mean absolute percentage error (MAPE) also allows meaningful comparisons to be made between several forecasted items or events having significantly different magnitudes. The final error measure is computed by squaring the simple forecast error. The mean squared error (MSE) is the average of all error values squared. Squaring the error places a disproportional penalty on large differences between forecasted and actual values that result from data points outside the normal pattern. This type of error measure should be used when significant discrepancies between actual and forecasted values can have a major impact on decision making.

The appropriate measure of forecast error depends upon the situation and the purpose underlying the forecast. It is important for engineering managers to realize that multiple measures of forecast error exist. Often, it is meaningful to evaluate a forecasting technique by using several different error measures. A great deal of computer-assisted computation and analysis of different alternatives is involved in identifying a suitable forecasting technique needed to conclude this step.

Step 6. In this step, the selected technique is used to forecast the value of the dependent variable or the occurrence of an event during the desired planning time frame. It is often prudent to develop a range over which the forecast value may vary. If historical data are being used to predict the future, statistical techniques can be applied to establish an analytically-based confidence level that the actual future value will reside within a prescribed interval. The use of a forecast range provides an overt indication of the uncertainty associated with forecasting the future.

Step 7. The purpose of this step is to assure that the impact of relevant internal and external factors on the results generated by a specific forecasting technique are incorporated. Quantitative adjustments to a specified forecast value are required for significant internal changes such as the introduction of new engineering services, production process redesigns, or facility/workstaff expansion plans. Additionally, expected changes in the business cycle, new government regulations, anticipated technological upgrades and other environmental activities must be evaluated to assess their impact on the forecast. Informed judgment on the part of responsible engineering managers is required to determine the influence of these factors on projected future values or events.

Step 8. The importance of documenting all assumptions involved in developing the forecast is one of the important dimensions of this step. In addition, the procedures which have been utilized to adjust the historical data, to test viable forecasting techniques and to analyze resulting forecast errors need to be recorded. The written report will serve as a record of what was done and why certain choices were made. However, the real benefit from this activity occurs when the forecasting process is successful and documentation is readily available to replicate the forecasting procedures for the next planning period. Even if the process produced a poor forecast, management can learn from this documented experience and hopefully improve both the process and the results in the future.

Step 9. This step involves evaluating the actual performance relative to forecasted performance. Monitoring a relevant measure of forecast error allows the effectiveness of the forecast to be continually evaluated. The manager must periodically decide whether the current forecast process is producing a forecast that is acceptable for planning purposes. If the rate is tolerable, then the forecast should be updated for the next planning period. It not, a new forecast model or the same technique with different parameter values may be needed.

Step 10. If the forecast results are not being used for their stated purpose, then corrective action needs to be initiated. Often this occurs when the forecast error exceeds an unspecified tolerance limit. Significant forecast errors require cycling back to Step 3 where, for example, historical data associated with workforce staff utilization or project performance are once again visually displayed to aid in identifying past performance patterns. The engineering manager needs to be prepared to assess forecast errors relative to acceptable tolerances and evaluate the costs and benefits of utilizing alternative forecast models or techniques.


Personal involvement with the forecasting process does not guarantee improved forecast accuracy; however, it will provide additional insights into departmental activities and intraorganizational needs which will increase decision making effectiveness. The benefits of improved planning, greater understanding and increased sensitivity to their managerial responsibilities accrue to those engineering managers who are actively engaged in the forecasting process.

The fact that a particular forecast "may be wrong" should not discourage the industrial engineer from addressing these future uncertainties by becoming directly involved in selected aspects of the forecasting process. At a minimum, serious consideration should be given to (1) specifying the forecast purpose, (2) reflecting on past data patterns, (3) modifying quantitatively determined forecast values with relevant qualitative information, (4) assuring that the forecast approach and results are documented for future use, and (5) determining if the resultant forecast error is acceptable. Effective and efficient goal achievement can be enhanced from insight and understanding gained through direct participation in the forecasting process.

Richard A. Reid, Ph. D. is a professor of operations management at the University of New Mexico, Albuquerque.
COPYRIGHT 1992 Institute of Industrial Engineers, Inc. (IIE)
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1992 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Reid, Richard A.
Publication:Industrial Management
Date:Jan 1, 1992
Previous Article:Revisiting the original management primer: defending a great productivity innovator.
Next Article:The environment and its effect on today's management.

Related Articles
Society for the Advancement of Material and Process Engineering.
Ohio State University's.
Budget retreats for financial planning: the City of Cape Coral experience.
Environmental legislation important factor in US activated carbon growth.
Ergonomics seminars target improving productivity, reducing injuries.

Terms of use | Copyright © 2016 Farlex, Inc. | Feedback | For webmasters