Source: Operations Engineering and Management: Concepts, Analytics, and Principles for Improvement, 1st Edition
ISBN: 9781260461831
Authors: Seyed M. R. Iravani
3.3. Principles of Forecasting
https://accessengineeringlibrary.com/content/book/9781260461831/toc-chapter/chapter3/section/section8
© McGraw-Hill Education. All rights reserved. Any use is subject to the Terms of Use, Privacy Notice and copyright information.
Forecasting is difficult. The main reason is that there are many factors that affect the demand for a product. Consider the
demand for a large screen TV, as an example. Its demand depends on several factors such as the quality of the TV, its features,
its selling price, the competition price, state of economy, and other factors that are unknown. Forecasting models are designed
to capture the effects of some of these factors and develop a reasonably accurate prediction of future demand. These models
are different in their complexity, the type of data they need, and in the way they develop forecasts. However, they all share the
following principles. These principles give a better understanding of the benefits and limitation of forecasting methods and
prevent misinterpretation of the forecasts developed by these methods.
1. Forecasts are usually wrong, but they are useful: It is important to point out that even the forecasts made by the most
sophisticated forecasting methods are wrong more than being right. As mentioned before, there are many known and
unknown factors that affect the demand of a product. Hence, it is almost impossible to predict the exact demand. Managers
should, therefore, not consider a forecast as a known and 100 percent accurate information. The goal of forecasting
methods is to find a close estimate of the demand, which is then used in making planning decisions. But, the planning
process should take into account the possibility of errors in the forecast.
2. A good forecast also includes a measure of error: Because forecasts are usually wrong, a good forecast should also
provide some measure for the accuracy of the forecast. The measure should correspond to the magnitude of the error that
the forecaster should expect. Examples are sum of squared errors, or variance of errors, see Sec. 3.8.3.
3. A good forecast is more than just a single number: A forecast for demand in the next period is a single number that
provides an estimate for what the demand in the next period is expected to be. This single number is called Point Forecast
for demand. However, because there is variability in demand, we know that the probability that next week's demand will be
exactly what we forecasted is very low. So, it is useful to also forecast a range that contains next week's demand with high
probability (e.g., 95 percent). This is called Prediction Interval Forecast. A good forecast should include a point forecast for
the average demand in the next period, as well as a prediction interval forecast that contains the actual demand with high
probability (see Sec. 3.8.5).
4. Forecast for a group of products is more accurate than the forecast for each individual product: For example, forecast for
demand of blue-color two-door car of a particular model with leather seats in a year is less accurate than the forecast for
the total demand of all car models of different colors with different options. The reason is that the demand for a specific
item is often affected by more factors. The demand for the blue color and leather seat car in a year, for example, is affected
by the popularity of the blue color and leather seat in that year. However, the impacts of these factors—color and seat
options—disappear when the demand for all cars are forecasted. When total number of cars are considered, the impact of
the popularity of a certain color is canceled out by the unpopularity of other colors. Hence, the data for a group of products
is often more stable, even when each individual product has very variable and unstable demand.
5. Forecasts for shorter time horizons are more accurate: This is quite clear. It is easier to forecast next month's demand for a
DVD player than the next year's demand for the DVD player. The reason is that, as we go further into the future, there is a
higher chance that some factors affecting the demand change, but this is less likely in shorter time horizons. For example, in
longer time horizons, there is a higher possibility that new DVD players with better features are introduced to the market.
6. Forecasts are not substitute for derived values: There are cases in practice that the demand for an item can be calculated
using some available information. In such cases, forecasting methods should not be used. One common case is predicting
demand for a part used in assembly of a product. As an example, suppose that the demand for cars is forecasted and a
production schedule is set to assemble 500 cars per day. Considering that each car requires four tires, the demand for tires
would be 4 × 500 = 2000 tires a day. In this case, there is no need for forecasting demand for tires.
https://accessengineeringlibrary.com/content/book/9781260461831/toc-chapter/chapter3/section/section8
© McGraw-Hill Education. All rights reserved. Any use is subject to the Terms of Use, Privacy Notice and copyright information.
