Chapter 03 - Forecasting
CHAPTER 03
FORECASTING
Forecasting is placed early in the text mainly because it is a point of departure. Some instructors like to
emphasize the operations part of operations management and de-emphasize the design part. Other
instructors prefer to blend the two. However, forecasting is an important input for both, and for that
reason, it is presented as early as possible.
Teaching Notes
This is a long chapter, so you may want to be selective about the topics covered to shorten the time
devoted to it. I tend to devote more time to the time series methods than I do to regression analysis, for
several reasons. One is that students often are exposed to regression in their stat course(s). Another is that
time series models are used more than associative models are. Other optional materials that can be
mentioned briefly, but not explored in detail, include trend-adjusted exponential smoothing (mentioned so
that students will realize that exponential smoothing does not work well if there is trend present) and
computation of seasonal relatives (you may want to explain how relatives are used without getting into
how they are derived).
I try to emphasize an intuitive approach to forecasting, with frequent reference to the importance of
plotting the data to assist the decision-maker in determining which forecasting technique may be more
appropriate to use.
In operations management, we forecast a wide range of future events, which could significantly affect the
long-term success of the firm. Most often, the basic need for forecasting arises in estimating customer
demand for a firm’s products and services. However, we may need aggregate estimates of demand as well
as estimates for individual products. In most cases, a firm will need a long-term estimate of overall
demand as well as a shorter-run estimate of demand for each individual product or service. Short-term
demand estimates for individual products are necessary to determine daily or weekly management of the
firm’s activities such as scheduling personnel and ordering materials. On the other hand, long-term
estimates of overall or aggregate demand can be used in determining company strategy, planning long-
term capacity and establishing facility location needs of the firm.
Finally, it is important to point out the difference between forecasting and planning. Planning is often in
response to a forecast. A passive response would be to reduce output because of a predicted decrease in
demand, while an active response would be to advertise in an effort to offset the predicted decrease in
demand.
Reading: Gazing at the Crystal Ball
1. Demand forecasting (DF) is part science and part art (intuition) for estimating what future
demand for a product or service will be. The science part uses information technology to generate
demand forecasts using existing data from a variety of sources, e.g., distribution channels, factory
outlets, value-added resellers, historical sales data, and macroeconomic data. The art/intuition
part involves subject matter experts (SMEs) making educated guesses about future demand.
2. A company executive might make bold predictions about future demand to Wall Street analysts to
maintain the company’s stock price.
3-1
,Chapter 03 - Forecasting
3. An executive’s comments to Wall Street analysts may result in the company changing its demand
forecast to reflect the comments made by the executive. The result often is excessive inventory
build-up starting at the distribution channels to the upstream suppliers.
Answers to Discussion and Review Questions
1. It depends on the situation at hand. In certain situations, one approach will be superior to the
other.
Quantitative techniques lend themselves to computerization, they are less subject to personal
biases, and they may force managers to quantify information. On the other hand, the results are
only as good as the data, and in many cases, insufficient data exist to use a quantitative technique.
In addition, use of the computer sometimes creates an impression of “preciseness,” which is
misleading.
2. Poor forecasting leads to poor planning. This could result in offering products and services that
customers do not want. Poor forecasting and planning would negatively affect budgeting and
planning for capacity, sales, production and inventory, labor, purchasing, energy requirements,
capital requirements, and materials requirements.
3. a. Consumer surveys may be invalid if they are not carefully constructed, administered, and
interpreted. Moreover, respondents may be ill informed or otherwise formulate answers that
do not correctly reflect their future actions.
b. Salespeople often tend to be overly optimistic or pessimistic. They may attempt to use
estimates to influence quotas.
c. Committees of managers or executives can be expensive, diffuse responsibility for a forecast,
and reflect the opinions of a few dominant members.
4. Forecasts generally are wrong due to the use of an incorrect model to forecast, random variation,
or unforeseen events.
5. Control limits reveal the bounds of random errors; they enable managers to judge if a forecasting
technique is performing as well as it might (and hence, when a technique should be reevaluated).
6. The relative costs of reevaluating a forecast when nothing is wrong versus not reevaluating it
when something is wrong. (Can also explain in terms of relative costs of Type I and Type II
errors.)
7. MAD focuses on average error while MSE focuses on squared errors. (MSE gives considerably
more weight to large forecast errors.)
8. Exponential smoothing: requires less data storage, gives more weight to recent data, and is easier
to change to make it more responsive to changes in demand.
9. The fewer the periods in a moving average, the greater the responsiveness.
10. The choice of alpha in exponential smoothing depends on how responsive a forecast the manager
desires. This, in turn, relates to the cost of not responding to a real change relative to the cost of
responding to what are merely random variations in the data.
3-2
,Chapter 03 - Forecasting
11. Of course, the accuracy of your five-day weather forecast will depend on a number of variables
such as time of year, where you live, etc. However, there is one trend that will establish itself and
that is as time passes from the first day to the fifth day, the accuracy of the forecast will decline.
Note: Students answers will vary depending on what actual data they obtain.
12. For example, if each average is based on 12 months, as each new data point is added to the
moving average, its counterpart is removed from the other end of the series.
13. Sales indicate how much customers bought, while demand indicates how much they wanted. The
distinction is important when demand exceeds supply, because supply places an upper bound on
the data.
14. A reactive approach takes the forecast as a “given” while a proactive approach takes an
unacceptable forecast and attempts to alter demand. An example of the reactive approach is a
highway department preparing snow removal equipment for a predicted storm. Another is
evacuation of residents due to hurricane warnings. An example of the proactive approach is
colleges adopting a more aggressive stance towards applicants due to a forecast of a declining
college-age population base. Generally, firms that use advertising, promotions, discounts, and so
on tend to be proactive in dealing with forecasts.
15. There is always going to be a certain amount of random variation about the forecast. The amount
of this random variation about the forecast (actual vs. forecast) will increase as the forecasting
horizon is extended. In other words, forecasting accuracy tends to decline over time.
Consequently, one of two approaches might be employed to handle the problem. One would be to
pick out some reasonable future point in time; then, based on past forecasting data, estimate the
amount of random variation that occurs over this period. The next step would be to build or
develop enough flexibility into your production system to be able to adjust to the extremes of the
random variation.
The other approach would be to estimate the flexibility of your production system and then see
how far into the future your production system would be able to handle the random variation that
is inherent in your forecasting approach. This would give you an indication of the amount of
change your system could handle over a period of time. Then adjustments could be made to
increase your capabilities to respond to change or you could try another forecasting approach to
see if its inherent variation is less.
3-3
, Chapter 03 - Forecasting
16. Forecasting in the context of supply chain involves connection and communication between the
supply chain databases. For example, assume that Company X is a durable goods manufacturer.
Based on the market and historical sales information, Company X determines short and
intermediate term multi-period forecasts for its products and provides this forecast information to
its suppliers’ databases. Let us also assume that Company Y supplies Company X with parts and
components. Company Y uses the forecasting information from Company X, as well as from
other companies it supplies, and develops its own forecasts and provides all the forecasting
information it possesses to its suppliers. This type of cooperation and communication among the
supply chain databases provides all of the companies on the supply chain with additional
information to generate better forecasts. Potential difficulties in doing supply chain forecasting
include creating the ability for sharing of data between different information systems and
establishing trust between supply chain partners so that they are willing to share data.
17. It depends on the situation. Sometimes one approach is better, sometimes the other is better, and
sometimes both are used. Considerations include the importance of the forecasts, how quickly the
forecasts are needed, the cost of obtaining the forecasts, the availability of resources, and the
availability of data. The qualitative approach is generally more popular with smaller companies
because they generally cannot afford to install a sophisticated quantitative technique. Larger
companies tend to utilize more sophisticated quantitative techniques due to the availability of
resources and the need to generate a large number of forecasts.
18. In forecasting initial sales for the new version of its software, the software producer should
consider:
a. The historical demand information for the old version.
b. The features of the new version of the software in comparison to the features of the old
version.
c. The price of the new version of the software in comparison to the price of the old version.
d. Market/consumer information and response about the new version of the software based on
the results of a market survey and the beta testing of the new version of the software.
e. The features of competitors’ similar software packages.
f. The price of competitors’ software packages.
19. a. Demand for Mother’s Day greeting cards: Naïve using last year’s demand. Alternatively,
because greeting cards have seasonal demand, we could use a seasonal model where the
season begins a few weeks before Mother’s Day and ends just after Mother’s Day.
b. Popularity of a new TV series: Delphi or associative based on features of existing series.
c. Demand for vacations on the moon: Delphi.
d. The impact of a price increase: Associative.
e. Demand for toothpaste: Naïve or averaging.
3-4
CHAPTER 03
FORECASTING
Forecasting is placed early in the text mainly because it is a point of departure. Some instructors like to
emphasize the operations part of operations management and de-emphasize the design part. Other
instructors prefer to blend the two. However, forecasting is an important input for both, and for that
reason, it is presented as early as possible.
Teaching Notes
This is a long chapter, so you may want to be selective about the topics covered to shorten the time
devoted to it. I tend to devote more time to the time series methods than I do to regression analysis, for
several reasons. One is that students often are exposed to regression in their stat course(s). Another is that
time series models are used more than associative models are. Other optional materials that can be
mentioned briefly, but not explored in detail, include trend-adjusted exponential smoothing (mentioned so
that students will realize that exponential smoothing does not work well if there is trend present) and
computation of seasonal relatives (you may want to explain how relatives are used without getting into
how they are derived).
I try to emphasize an intuitive approach to forecasting, with frequent reference to the importance of
plotting the data to assist the decision-maker in determining which forecasting technique may be more
appropriate to use.
In operations management, we forecast a wide range of future events, which could significantly affect the
long-term success of the firm. Most often, the basic need for forecasting arises in estimating customer
demand for a firm’s products and services. However, we may need aggregate estimates of demand as well
as estimates for individual products. In most cases, a firm will need a long-term estimate of overall
demand as well as a shorter-run estimate of demand for each individual product or service. Short-term
demand estimates for individual products are necessary to determine daily or weekly management of the
firm’s activities such as scheduling personnel and ordering materials. On the other hand, long-term
estimates of overall or aggregate demand can be used in determining company strategy, planning long-
term capacity and establishing facility location needs of the firm.
Finally, it is important to point out the difference between forecasting and planning. Planning is often in
response to a forecast. A passive response would be to reduce output because of a predicted decrease in
demand, while an active response would be to advertise in an effort to offset the predicted decrease in
demand.
Reading: Gazing at the Crystal Ball
1. Demand forecasting (DF) is part science and part art (intuition) for estimating what future
demand for a product or service will be. The science part uses information technology to generate
demand forecasts using existing data from a variety of sources, e.g., distribution channels, factory
outlets, value-added resellers, historical sales data, and macroeconomic data. The art/intuition
part involves subject matter experts (SMEs) making educated guesses about future demand.
2. A company executive might make bold predictions about future demand to Wall Street analysts to
maintain the company’s stock price.
3-1
,Chapter 03 - Forecasting
3. An executive’s comments to Wall Street analysts may result in the company changing its demand
forecast to reflect the comments made by the executive. The result often is excessive inventory
build-up starting at the distribution channels to the upstream suppliers.
Answers to Discussion and Review Questions
1. It depends on the situation at hand. In certain situations, one approach will be superior to the
other.
Quantitative techniques lend themselves to computerization, they are less subject to personal
biases, and they may force managers to quantify information. On the other hand, the results are
only as good as the data, and in many cases, insufficient data exist to use a quantitative technique.
In addition, use of the computer sometimes creates an impression of “preciseness,” which is
misleading.
2. Poor forecasting leads to poor planning. This could result in offering products and services that
customers do not want. Poor forecasting and planning would negatively affect budgeting and
planning for capacity, sales, production and inventory, labor, purchasing, energy requirements,
capital requirements, and materials requirements.
3. a. Consumer surveys may be invalid if they are not carefully constructed, administered, and
interpreted. Moreover, respondents may be ill informed or otherwise formulate answers that
do not correctly reflect their future actions.
b. Salespeople often tend to be overly optimistic or pessimistic. They may attempt to use
estimates to influence quotas.
c. Committees of managers or executives can be expensive, diffuse responsibility for a forecast,
and reflect the opinions of a few dominant members.
4. Forecasts generally are wrong due to the use of an incorrect model to forecast, random variation,
or unforeseen events.
5. Control limits reveal the bounds of random errors; they enable managers to judge if a forecasting
technique is performing as well as it might (and hence, when a technique should be reevaluated).
6. The relative costs of reevaluating a forecast when nothing is wrong versus not reevaluating it
when something is wrong. (Can also explain in terms of relative costs of Type I and Type II
errors.)
7. MAD focuses on average error while MSE focuses on squared errors. (MSE gives considerably
more weight to large forecast errors.)
8. Exponential smoothing: requires less data storage, gives more weight to recent data, and is easier
to change to make it more responsive to changes in demand.
9. The fewer the periods in a moving average, the greater the responsiveness.
10. The choice of alpha in exponential smoothing depends on how responsive a forecast the manager
desires. This, in turn, relates to the cost of not responding to a real change relative to the cost of
responding to what are merely random variations in the data.
3-2
,Chapter 03 - Forecasting
11. Of course, the accuracy of your five-day weather forecast will depend on a number of variables
such as time of year, where you live, etc. However, there is one trend that will establish itself and
that is as time passes from the first day to the fifth day, the accuracy of the forecast will decline.
Note: Students answers will vary depending on what actual data they obtain.
12. For example, if each average is based on 12 months, as each new data point is added to the
moving average, its counterpart is removed from the other end of the series.
13. Sales indicate how much customers bought, while demand indicates how much they wanted. The
distinction is important when demand exceeds supply, because supply places an upper bound on
the data.
14. A reactive approach takes the forecast as a “given” while a proactive approach takes an
unacceptable forecast and attempts to alter demand. An example of the reactive approach is a
highway department preparing snow removal equipment for a predicted storm. Another is
evacuation of residents due to hurricane warnings. An example of the proactive approach is
colleges adopting a more aggressive stance towards applicants due to a forecast of a declining
college-age population base. Generally, firms that use advertising, promotions, discounts, and so
on tend to be proactive in dealing with forecasts.
15. There is always going to be a certain amount of random variation about the forecast. The amount
of this random variation about the forecast (actual vs. forecast) will increase as the forecasting
horizon is extended. In other words, forecasting accuracy tends to decline over time.
Consequently, one of two approaches might be employed to handle the problem. One would be to
pick out some reasonable future point in time; then, based on past forecasting data, estimate the
amount of random variation that occurs over this period. The next step would be to build or
develop enough flexibility into your production system to be able to adjust to the extremes of the
random variation.
The other approach would be to estimate the flexibility of your production system and then see
how far into the future your production system would be able to handle the random variation that
is inherent in your forecasting approach. This would give you an indication of the amount of
change your system could handle over a period of time. Then adjustments could be made to
increase your capabilities to respond to change or you could try another forecasting approach to
see if its inherent variation is less.
3-3
, Chapter 03 - Forecasting
16. Forecasting in the context of supply chain involves connection and communication between the
supply chain databases. For example, assume that Company X is a durable goods manufacturer.
Based on the market and historical sales information, Company X determines short and
intermediate term multi-period forecasts for its products and provides this forecast information to
its suppliers’ databases. Let us also assume that Company Y supplies Company X with parts and
components. Company Y uses the forecasting information from Company X, as well as from
other companies it supplies, and develops its own forecasts and provides all the forecasting
information it possesses to its suppliers. This type of cooperation and communication among the
supply chain databases provides all of the companies on the supply chain with additional
information to generate better forecasts. Potential difficulties in doing supply chain forecasting
include creating the ability for sharing of data between different information systems and
establishing trust between supply chain partners so that they are willing to share data.
17. It depends on the situation. Sometimes one approach is better, sometimes the other is better, and
sometimes both are used. Considerations include the importance of the forecasts, how quickly the
forecasts are needed, the cost of obtaining the forecasts, the availability of resources, and the
availability of data. The qualitative approach is generally more popular with smaller companies
because they generally cannot afford to install a sophisticated quantitative technique. Larger
companies tend to utilize more sophisticated quantitative techniques due to the availability of
resources and the need to generate a large number of forecasts.
18. In forecasting initial sales for the new version of its software, the software producer should
consider:
a. The historical demand information for the old version.
b. The features of the new version of the software in comparison to the features of the old
version.
c. The price of the new version of the software in comparison to the price of the old version.
d. Market/consumer information and response about the new version of the software based on
the results of a market survey and the beta testing of the new version of the software.
e. The features of competitors’ similar software packages.
f. The price of competitors’ software packages.
19. a. Demand for Mother’s Day greeting cards: Naïve using last year’s demand. Alternatively,
because greeting cards have seasonal demand, we could use a seasonal model where the
season begins a few weeks before Mother’s Day and ends just after Mother’s Day.
b. Popularity of a new TV series: Delphi or associative based on features of existing series.
c. Demand for vacations on the moon: Delphi.
d. The impact of a price increase: Associative.
e. Demand for toothpaste: Naïve or averaging.
3-4
