Forecasting

Learning Objectives

After completing this chapter, you should be able to:

1. Describe the importance of forecasting in organizations.
2. Explain the two types of forecasting techniques used for
demand forecasting: qualitative and quantitative.
3. Compare moving averages, exponential smoothing, and other
time-series models.

4. Understand Delphi and other qualitative decision making

approaches.
5. Choose the procedures that can be used to select the best
forecasting techniques.
6. Calculate a variety of error measures.

Chapter Outline

The importance of forecasting

7.1
Introduction
7.2
7.3
Steps in forecasting
7.4
Techniques of forecasting
7.5 Choosing the best forecasting technique
7.6 Glossary
7.7
 7.8
Review questions
Exercises

Forecasting is one of the important tools for any organization. This chapter identifies the
decisions that managers should make in designing a forecasting system. The chapter ends with a
discussion of the benefits and limitations of forecasting.
7.1 Introduction

A forecast predicts an event that will use in the future. In a sense, management needed
accurate forecasts to ensure operational success. If the future were determined, there would be no
need for forecasting. The forecasting value is not a deterministic quantity. Meanwhile, it is only
an assessment based on the previous data related to a particular event for a future period. The forecast
is sometimes called a prediction, estimate, guess, assessment, or extrapolation. The term
forecasting is concerned with the determination of the best basis available for the formulation of
efficient managerial expectations and the control of uncertainty about the future day. We know
the implication of decisions becomes very comprehensible. There is a need to make the right
decisions in the present situation because the market is very competitive. The success of many
organizations depends on how well they are forecasting their demand accurately. The demand
becomes important to achieve business objectives. This is done by gathering knowledge
about various aspects of the market. Thus, this is the concept that we use for forecasting.
Forecasting has also been used to predict the improvement of unusual situations.
Predictors perform research that uses realistic results to estimate the effectiveness of definite forecasting
models. However, different models will be discussed in this section.

Forecasts and Forecasting

A forecast is to determine or predict future events usually as a result of the study and analysis
of available relevant data. In addition to being able to see the past and present data, the
forecast will allow you to look at expected future figures by creating forecast items. A perfect
forecast is an exceptional one.
 Forecasting is a technique of making predictions of the future based on the past and
present history of data as estimates in determining by analysis of trends. Organizations apply
forecasting to decide how to allocate their budgets or plan for expected expenses for
upcoming time periods. A common example might be a projection of a certain variable of
interest at a specified coming time period. Nevertheless, the data must be a recent date for the
forecast to be as accurate as possible. One thing is constant, and all organizations depend on
business forecasting.
Business forecasting could be classified into economic forecasting, demand forecasting,
and technology forecasting. These are:

Economic Forecasting: These will be useful to government agencies and other organizations in
predicting future tax revenue, employment rate, inflation rate, interest rate, GDP, etc.
Demand Forecasting: Demand forecasting gives the predicted future demands of existing
products and services. Operational planning is done on the basis of business skills and
previous experiences.
Technology Forecasting: This kind of forecasting is an estimate of the future activities of useful
technological machines, products, procedures, techniques, etc.
7.2 The Importance of Forecasting

Forecasting is an essential activity for planning the input and output of the operations
management system and also affects other managerial activities required of managers. Unfortunately,
many organizations have failed in their proper planning because of defective forecasting.
Therefore, the importance of forecasts to a firm cannot be overemphasized.
The importance of forecasting involves the aforementioned areas are as follows:
i. Forecasting arranges appropriate and consistent figures about past and present
events to do future comprehensive planning.
ii. It gives confidence to the managers for making significant decisions.
iii. It arranges how many units of supplies and materials to acquire.
iv. It provides how many employees to hire and how large a facility to acquire. and
v. It keeps managers attentive and also dynamic to face future challenges.

7.3 Steps in Forecasting

Forecasting is sometimes an ignored part of the business organization. However, predicting
future events can significantly help managers make the best possible decisions. The forecasting
task normally involves the following steps:
1. Identify the objective: The first step in the process is to identify the objective explicitly or
uses for the forecast and identify where the business is currently located in the market.
2. Identify the problem: Defining the problem judiciously needs an understanding of the
technique the forecasts will be used, whom the forecast is concentrating on, how the
forecasting task fits within the organization, how the market works, and what your customer base
and competition are.
3. Collecting and gathering data to be used as measuring devices: The information comes
basically in two ways: the knowledge gathered by experts and actual statistical data.
Collecting and gathering the data and information required for the forecasting
technique selected.
4. Analysis of data: In this step, the data are analyzed in light of one's understanding of
the reason why changes occur, after choosing a suitable tool to run the data through
analysis and conclude the future course of action.
5. Choosing and fitting models: It is common to compare various potential models.
Unfortunately, there is not a single model that works best in all situations. It all depends
on the availability and nature of the data. When all the information is collected and treated, you
may choose and fit the appropriate forecasting model based on your objective. 6. Using
and evaluating a forecasting model: Finally, once a
a select appropriate
model is used to induce forecasts and evaluate it. Estimate not only the performance of
the model, but estimate it in terms of its inherent or potential shortcomings, the entire
process, sources of error, level of reliability, and need for revisions or improvements.
म
7.4 Techniques of Forecasting
The qualitative and quantitative techniques might be two major categories of forecasting
techniques. These techniques are obtained as follows:

Qualitative forecasting techniques: Qualitative techniques consist of subjective inputs, which

challenge accurate numerical descriptions. It is based on the opinion and judgments of consumers
and experts. These are useful when previous data are not available and are suitable for new
products. Most of the long-term forecasting is qualitative. Two methods that are
commonly used in qualitative forecasting are the Delphi method and the Market research.

Quantitative forecasting techniques: Quantitative forecasting techniques are based

on analyzing historical data and its trends. These techniques are more accurate because they use
statistical analysis and other mathematical techniques to forecast future events. These methods are
commonly applied to short-range or medium-range decisions. The quantitative techniques are
classified into two categories: Time Series (i.e., moving average, exponential smoothing) and
Causal (i.e., regression analysis). The classification of the forecasting techniques is in the
following Figure 7.1.

Figure 7.1 Classification of the Forecasting Techniques

Forecasting
Techniques
Delphi

Qualitative
Market Research

Moving
Average

Time series

Exponential
Quantitative
Smoothing

Regression
Causal
Analysis

Simple
weight
ed
The above discussion is about various techniques of qualitative and quantitative forecasting.
Though, no specific technique can be recommended as universally applicable.
Most of all, forecasts are done by involving different techniques. Brief discussions of the most important
forecasting methods are given below:

7.4.1 Delphi Method

The Delphi method was developed by Olaf Helmer and Norman C. Dalkey in 1950
at the Rand Corporation's think tank for the specific military problem. In ancient times, the
name 'Delphi' comes from the Greek town South of Athens, has been successfully used
by major manufacturing organizations for forecasting. Later on, this method was
applied in other areas, such as economic trends, health, and education.
The Delphi method or Delphi technique is popular in many disciplines. It is a
qualitative technique of forecasting. This method is based on the logic that a panel of experts
can arrive at a better solution than a single individual. It is ideally used in forecasting situations
that involve subjective or judgemental forecasting. This method of forecasting is suitable
when historical data is absent.
The Delphi method is usually used to forecast long-range product demand and innovative
product sales forecasts. It can also be used for scientific and technological forecasting. The
advantage of the method is to interrogate a group of knowledgeable persons or expertise in
the field of interest, such as new product development or the market for some specific product. A
structure of a questionnaire is employed in which the responses from a prior set of questions are
used to design a subsequent questionnaire. In the expert group, the members tend to respond to the
questions and provide their several opinions. After getting the opinions from the expert
members, find out the similarities between these opinions. If the variations among the opinions
are too much, the summary of opinions is sent to the same group of experts for another round
and the participants may choose to reconsider their previous opinions. These rounds continue
until consensus has arrived.

While the procedure for the Delphi method involves the following sequential
steps:
1. The first step of the Delphi method is to select a panel of experts (people). These
experts should have some applied knowledge with research and data processing. They
should also find a neutral person within the organization.
 2. Send to each of the experts in the set of questionnaire requesting a forecast. They should
independently answer the questions and send them back to the central coordinator.
3. The results of this questionnaires should be collectively summarized in a statistical
presentation. The experts who are distant from the average (mean) forecast values should be
asked to justify their deviation.
4. Again, collect and summarize reports on the newly forecast values and ask if
participants reconsider their prior forecast.
5. The above process is repeated until a group of experts' consensus on a specific
forecast remains constant. Then the final report is prepared.

Each of the above steps of the Delphi method comes with its challenges. The Delphi
method has some limitations, especially that
a. The Delphi process is quite time-consuming. At that time, the group of people considered
to be experts may change, the bias in the results, or disappointed with the process.
b. The experts' responses in the Delphi process may be less meaningful and may not be
truly independent.
c. Below standard questionnaires will result in confusing and unaccepted conclusions. d. It
is difficult to determine in a short time to establish a better consensus in the
Delphi procedure.
The above shortcomings should be judiciously considered before the Delphi
method is used.
7.4.2 Market Research

Market research is also another qualitative technique of forecasting. It is a systematic

approach to conducting market surveys of some companies regarding specific consumer purchases.
For example, designing a set of questionnaires that requests economic, demographic, and social
information from every consumer interviewed and their opinions on particular products or
services of interest to the firm. The surveys can be conducted to gather information through
telephone polling, mailings, personal interviews, or questionnaires from the customers. This
method requires extensive statistical analysis is to be applied to survey results to test
hypotheses regarding consumer behavior.

Market research is suitable to forecast demand for the short, medium, and long term. The correctness of
forecast demand is outstanding for the short term, respectable for the medium term, and reasonable for
the long term. Positively, you might not have avoided phone calls asking you about product
preferences, your income level, behavior, and so on. Market research is usually employed to forecast
existing and new product sales or sales of established products in new markets. The research may be
controlled to limit the cost and time to a sample from the potential customers. However, another
limitation is the possibility that the research results do not exactly reflect the opinions of the
market. Lastly, the survey might produce derivative rather than innovative ideas because the
purchaser's reference opinion is often inadequate.

7.4.3 Time Series Analysis,

Time series is a set of numerical measurements on a time-dependent variable that has been
arranged at periodic intervals. Time is the most important factor in the time series analysis
because the variable is related to time which may be a year, month, week, day, hour, minute, second,
etc. It is another quantitative method of forecasting that uses time-series data usually involves the
assumption that previous experience reflects probable future experience.

Time series analysis techniques employ historical data covering one or more trend components,
seasonal variations, cyclical variations, and random variations. Trends are the upward and downward
movements of a time series of data over a long period of time. Seasonal variations are repetitive
movements in business activity that occur systematic variations in time of period. Cyclical
variations are long-time activities that represent consistently periodic up and down inactivity. The
term cycle refers to recurrent variations of more than one year related to economic and
political conditions. Sometimes identifying and using these components to forecast is
difficult because of the presence of random and irregular variations in a time series.
Random variation has having unpredictable outcomes, and all results just as probable
consequences. An irregular variation is one that occurs predictably and generally during short
periods. It does not follow a particular model.

The following Figure 7.2 illustrates the different possible trends and variations that might
be seen in time-series data.
Figure 7.2 Some Possible Trends and Variations in Time-Series Data
Demand
Demand
Trend with random
 2

Year

Cyclical variations
Demand
Seasonal variations

3
0
1
7
3
5

Year
Demand
Random variations
around an average

2
3
4
5
0
1
2
3
4
5

Year
Year

Selecting a technique to use with time-series data requires some knowledge of

components a series of trends to exhibit. Time series analysis should be used as a foundation for
forecasting when data are available for a long-time period and tendencies exposed by the trend.
However, time series analysis plays a dynamic role in business decision making for planning and
evaluating present and future operations.

7.4.4 Simple Moving Average

A simple moving average (SMA) technique is used to computing the average of a

demand time series and useful in removing these random variations for forecasting. Applying
this method is to generate the forecast for the next time period by averaging the actual demand for the n
 immediate past time periods.
To compute the next time period, if the demand is known, the previous demand for
the previous average is replaced with the immediate past demand, and the average is calculated
again. A forecast using this method will tend to gap the trend if one happens.
The simple moving average (SMA) forecast can be calculated using the following formula:

where
SMA1
=
t-1

t=1

SMA, = Forecasted demand for period t

D1 = Actual demand in period t
n = Total number of periods in the average
D-1= Actual demand in the period immediately preceding
period t
Dt-(n-1)+Dt-(n-2)+ ...... +D12+ D1-1 + Dt
For example, a simple moving average would imply a three-period moving
average.

Example 7.1
Compute a three-period moving average forecast for the arrival of customers in a shop for
the last seven periods:
Period
Customer

(days)
Arrivals
1
30
2
3
4
5
6
 7
34
28

32
30
32
34

If the actual number of customer arrivals on day 8 is 35, what is the forecast for day 9?

Solution

This is demonstrated using a problem in Table 7.1

Table 7.1

Period
Three Days Simple Moving Averages

Customer Total Moving

(days) Arrivals Moving Average (MA)
Forecast

1
30

2
3
2
34
28
92
30.67
 4
32
94
31.33
30.67
5
30
90
30
31.33
6
32
94
31.33
30

7
34.
96
32
31.33

The values written under column (MA) are obtained as:

Total number of periods in the moving average, n = 3
The simple moving average at the end of days 3, 4, 5, 6, and 7 are:
30+34+28
MA2
=
=

30.67
3 34+28+32
MA1

MA1
=

3
28+32+30

3
31.33

30
 MA
=

32+30+32

3
30+32+34
=

31.33

MA1
=

3
32

The simple moving average forecast at the end of day 7 is:

SMA = MA,
30+32+34

3
32

The forecast for day 9 requires the actual arrivals from days 6-8, the three immediate
past days of data.

SMA,
32+34+35

3
= 33.67

Therefore, the forecast at the end of day 7 would have been 32 customers for day 8. The
forecast for day 9, made at the end of day 8, would have been 33.67 customers.
The moving average can incorporate as much data as used to estimate trends. It is
relatively simple to update the moving total then subtract the previous value from the
recent value and added that amount to the moving total, for each update.

Day Customer
Arrival
Estimate Trends

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

30 34 28 32 30 32 34 35 34 35 30 32 34 32 29 34 31 33 34 32 35 34 32 35 30
 Figure 7.3 Daily Customer Arrivals at a Shop
40

35

30

25
Customer arrivals
5

62 83
10
20

15

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

Day

Figure 7.3 shows a three-period simple moving average plotted against customer arrival over 25
periods as a line chart. The stability of the arrival series usually determines how many times (the
value of n) to include and smooth the forecasted values are compared to the actual values.

7.4.5 Weighted Moving Average

A weighted moving average (WMA) is very much related to a simple moving average and is
attained by mathematically weighing the values used to calculate the moving average. The
weighted moving average assigns more weight to the more recent values in a time period
1 UJU
13J

than to others. Such as in a three-period weighted moving average method with a weight
of 0.50 assigned to the most recent period, a weight of 0.30 assigned to the second
most recent period, and a weight of 0.20 assigned to the third most recent period. The sum of
the weights should add up to 1 (0.50 +0.30+ 0.20 = 100) and we cannot assign a negative weight. The
formula for computing weighted moving average is as follows:

where
WMA,= 0.50D, +0.30D.1 +0.20Dt-2
 WMA,= Weight for the time period t
D1 = Actual arrival in period t
Dt.1 = Actual arrival for last time period

Figure 7.4 Daily Weighted Moving Average Forecast Trends

40

35
Customer arrivals
30

25

20

15

10

0
5
10
15
20

Day

-Actual Customer Arrival

Four-Value WMA

Example 7.2

Use the data from example 7.1 to calculate:

a) a four-day weighted moving average for day 8, where the weight of 0.40 was assigned to the
most recent period, while the three previous days are assigned a weight of 0.30, 0.20,
and 0.10, respectively.
b) If the actual customer arrival for day 8 is 35, forecast the customer arrival for day
9.
Solution

a) The weighting moving average forecast value for day 8 is calculated using the formula
 above as follows:

WMA8 = (0.40×34) + (0.30×32) + (0.20×30) + (0.10×32)

=

13.60 + 9.60 + 6 + 3.20 = 32.40

b) Now given that customer arrival for day 8 is 35. The weighting moving average
forecast
value for day 9 is
WMA, = (0.40×35) + (0.30×34) + (0.20×32) + (0.10×30)
1=

14+ 10.20 +6.40+3=33.60

It is noted that in preparing the four-period moving average, four

weights are used.
From the above formula, we are using the weighted actual customer arrival during the last
four days to forecast the next day's arrival. Then, the process is repeated until we finally use
the last four days of arrival to forecast the seventh day's arrival.

Table 7.2 Forecast Values Using a Weighted Moving Averages Technique

Period

(days)
Actual
Arrivals
Forecast*

1
30

2
3
 4
34

28

32

5
30
31

6
32
32

7
34
30.5

* The forecast is based on a weighted moving average of four values with weights of 0.40,
0.30, 0.20, and 0.10, respectively.

7.4.6 Exponential Smoothing

Exponential smoothing is the most commonly used quantitative model in forecasting. It is

a sophisticated form of weighted moving average that uses a single time period to forecast
the future depending on the formulation. It is the most commonly used of all forecast methods
because of its ease. Exponential smoothing requires three types of data: the last period's
forecast, the demand for this period, and smoothing constant a.

The formula for the exponential smoothing technique is:

where
ES1 = ES-1+ a (Dt-1 - ESt-1)

ES1 = The smooth average forecast for the period t

ES = The previous period (t-1) forecast value
a = Smoothing constant or mathematical weight (0 ≤a≤1) Dt-1 =
Actual demand or sales for the previous period (t-1) value
 The difference between D., and ES, is the forecasting error in the last forecast period. The
smoothing constant, a (alpha), denotes a percentage of the earlier error and which has a value
between 0 and 1. If the earlier the value of a is to zero, the slower the forecast will
be to control to forecast error or the better the smoothing. Similarly, the closer the
value of a is to 1.0, the greater the sensitivity and the less the smoothing.

Example 7.3
Calculate the exponential forecast for the ten weeks for a firm with the following data. The
forecast for the first week was 250 units. Forecast the demand for up to ten weeks when the firm
uses exponential smoothing with a is 0.20.
Solution

The forecast value for the second week is calculated as follows:

ES ES-1+a (Dt-1 - ESt-1)

=

250 +0.2 (260-250)

= 250 + 2 = 252 units

The calculation of the exponential smoothed forecast for the first week to ten weeks is
given in Table 7.3.

Table 7.3

Actual
Smoothed Average Forecasts

Old Forecast Week

Demand, Forecast Error
(Dt-1 - ESt-1)
Correction
Forecast
a (Dt-1-ESt-1)| ESt=ESt-1+α (Dt-1-ESt-
1)
(Dt-1)
(ESt-1)

1
260
 250
10
2
252.0

2
262
252
10

3
244
254
-10

4
242
252
-10

2
2
2
254.0

252.0

250.0

5
255
250
5
1
251.0

6
 261
251
10
2
253.0

7
273
253

8
277
257
2220
4
257.0

4
261.0

9
281
261
20
4
265.0

10
264
265
-1
-0.2
264.8

Note that if you have no previous forecast value, the old forecast starting value may be
estimated or taken to be an average of the values of some preceding periods.

This method is applicable for forecasting data with no perfect trend. For instance,
the data in Figure 7.5 do not exhibition any clear trending behavior.
Figure 7.5 Exponential Smoothing Average Forecast
Trends
285

280

275

270
 265

260

255

250

245

240

0
2
4
6
8
10
12

-Actual Demand
Old Forecast
-Forecast

7.4.7 Regression Analysis

Regression analysis is a statistical equation that has been enriched to study and determine
the statistical relationship between two or more variables. The regression analysis has two
types of variables: the dependent variable is Y and the independent variable is X. In
simple regression, only one independent variable is used, while, in multiple regressions,
two or more independent variables are involved.

The simple regression can be expressed as follows:

where
ŷ=a+bx

ŷ = The average forecasted value of the dependent

variable
=

a The constant value (Y-axis intercept value)

b = The slope (trend) of the line

x = The independent variable

These are shown in the following Figure 7.6.

Figure 7.6 Simple Regression Line

YA
Demand
 a

ŷ = a + bx
b

Forecast variable
X

In the above figure, we have a scatter diagram of data that have been regressed to a single
line. At this time, 'a' is the intercept or constant value and 'b' is the slope of the
regression
line.

Making the values of 'a' and 'b' in the regression equation requires the following
formulas:

where

Σxy.
ExΣy
b
n

Σx2-
(Σ x)2
n

and a = y - bx
 Ex
X

ΣΥ
x=
and y
n
n

A multiple linear regression model provides a numeric measure of the relationship between the
dependent variable and a set of independent variables.

The model for multiple linear regression can be expressed as follows:

ŷ=a+b1x1+b2x2 +b3x3 +.........+b2x ̧

where

ŷ
=

The average forecasted value of the dependent variable

a = The constant value (Y-axis intercept value)

b = The slope (trend) of the line

x = The set of the independent variable

n = The number of independent variables

Example 7.4
The industry believes that its annual sale depends on its expenditure on an advertisement.
The sales (in lac taka) and advertisement (in lac taka) for the period 2010-2016 are given in
the following table. Determine the sale when the expenditure is 8 lacs.

Year
2010 2011 2012
2013 2014 2015
2016

Expenditure for
advertisement (x)
 Sales (y)
4
5
2
6
7
6
8

30
34
36
32
38
40
?
Solution

The calculation of simple regression is shown below:

Simple Regression Model Using Least-Squares Method
Table 7.4

Sale
Year
(y)
Expenditure for
advertisement (x)
xy
x2

2010
30
4
120
16

2011
34
5
170
25

2012
36
2
 72
4

2013
32
6
192
36
2014
38
7
266
49
2015
40
6
240
36

2y=210
Ex=30
Σxy= 1060 | Σx2=166
where

x=
X

Ex_30 -

n6
= 5; y=
ΣΥ 210
=35
n
6
X

Σxy
ΣxΣy
30×210
1060
b=
n
 6
1060-1050
=

Σx2-
(Ex)2
= 0.625
X
30×30
166-150
166-
6
n

a=y-bx =
n

Σy Σχ
b

Therefore the regression equation is, ŷ= 31.875+0.625x

The sale when expenditure is 8 lacs:
ŷ 31.875+0.625x
y = 31.875 +0.625×8=36.875 lacs.
X 210
30
—

0.625 ×
= 31.875
n

6
6

Example 7.5

The following are given the figures for sales (in metric tons) of a sugar factory. Compute the
sales forecast for 2020.
 Year

Sales (in m. tons) 80

2013 2014 2015 2016 2017 2018
90 92 83
2019

94
99
92

Solution

Here the number of years, n = 7(odd). Hence we want to shift the origin to the arithmetic
mean of the middle year.
Let the time indicator, x = (X-2016) and Sale = y
Table 7.5
Simple Regression Model Using Least-Squares Method
Deviation from
Sale (in metric
Year
tons), y
the middle year,
xy
x2
2

x=(X-2016)
2013
80
-3
-240
9

2014
90
-2
-180
4

2015
92
-1
-92
 1

2016
83
0
0
0

2017
94
1
94
1

2018
99
2
198
4

2019
92
3
276
9

Ly=630
Σx=0
Exy=56 Ex2=28

Let the equation of the straight line is ŷ = a +

bx
where

Σxy-
X

ΣxΣy
0
56.
b=
n
7
=

Σx2-
 (Ex)2
0
X
28
7
n
56
588
= 2 28

and a =
=

y-bx=
Σy Σx_630
b
n
n
7
—

2x
7
= 90
=

Hence the values for our least square regression equation is, ŷ=90+2x The
forecast for sales for the year 2020 is computed by substituting X
equation.
y=90+2 (X - 2020)
= 90+2 (2020-2016) = 90+2 x 4 = 94 metric tons

7.5 Choosing the Best Forecasting Technique

=

2020 in the above

At present, forecasting has become an essential part of our life. Successful forecasting
begins with careful thinking about the choice of a particular forecasting technique.
Choosing a forecasting technique depends on the purpose of the forecast, the desired
accuracy, the previous experience, the time required to develop a forecast, and the cost-
benefit of the forecast to the company. Many forecasting techniques have been developed in
recent years. Two general types of forecasting techniques are used for demand forecasting:
qualitative (e.g. timeliness of forecast) and quantitative (e.g. cost of the forecast). This
measure may be a total error of the forecast values from the actual demands. Determining
forecast error is an important part of forecasting because forecast errors can determine which
forecasting technique will be the best performing.