Unit 2 Forecasting

Structure:
4.1 Introduction
Objectives
4.2 What is Forecasting?
4.3 The Strategic Importance of
Forecasting Human resources
Capacity
Supply chain management
4.4 Why Forecasting is required?
Benefits from forecasts
Cost implications of forecasting
Decision making using forecasting
4.5 Classification of Forecasting Process
4.6 Methods of Forecasting
4.7 Case-let
4.8 Forecasting and Product Life Cycle
4.9 Selection of the Forecasting Method
4.10 Qualitative Methods of Forecasting
4.11 Quantitative Methods
What is time series?
Naïve method
Moving average method
Weighted moving average
Exponential smoothing method
4.12 Associative Models of Forecasting
4.13 Accuracy of Forecasting
Mean Absolute Deviation (MAD)
Standard Error (SE) of estimate
4.14 Summary
4.15 Glossary
4.16 Terminal Questions
4.17 Answers
4.18 Case Study
4.1 Introduction
In the previous unit, we have dealt with the concepts of operations strategy,
competitive capabilities and core competencies, operations strategy as a
competitive weapon, linkage between corporate, business, and operations
strategy, developing operations strategy, elements or components of
operations strategy, competitive priorities, manufacturing strategies, service
strategies, and global strategies and role of operations strategy. In this unit,
we will deal with the concepts of forecasting, the strategic importance of
forecasting, why forecasting is required, classification of forecasting
process, methods of forecasting, forecasting and product life cycle, selection
of the forecasting method, qualitative and quantitative methods of
forecasting, associative models of forecasting, and accuracy of forecasting.
Every business activity aims to satisfy some needs and wants of the society
and hence tries to gauge the demand. Only when the demand is properly
understood and predicted with sufficient accuracy, it becomes possible to
develop and utilise the resources to cater to such demands. Thus, for any
business activity to be started, the first step would be to predict the demand
and then to develop the plans towards meeting the demand either partially
or fully. Hence, it is correctly said that forecasting the demand is the first
step and demand forecasting drives all the other activities of production
systems which include human resource planning, aggregate planning,
capacity planning, and scheduling. Even if a company decides to position
itself in a certain way, it has to have done forecasting. Thus good forecasts
are of critical importance in all aspects of a business.
The forecast is the only estimate of demand until the actual demand
becomes known. However, forecasts are seldom perfect because the
demand for a certain product or service is a complex function influenced by
a multitude of variables. Many of these variables are not controllable and
even not properly evaluated in terms of magnitude and frequency. This
means that outside factors which are not known to us or properly predicted
or controlled impact the forecast tremendously. Hence, it is essential to
allow for this reality. In other words, expecting an accurate forecast is self-
defeating.
Most forecasting techniques assume that there is some underlying stability
in the system, which is not the case always. Hence, product family and
aggregated forecasts are more accurate than individual product forecasts.
Objectives:
After studying this unit, you should be able to:
 define forecasting
 explain the importance of forecasting
 explain when to use the qualitative models
 apply the different methods of forecasting and compare the results
 compute the measures of forecast accuracy
 identify special cases like causal and seasonal models
 use a tracking signal for checking the accuracy and efficiency of
forecasting

4.2 What Is Forecasting?

As stated in Heizer and Render (2010), forecasting is the art and science of
predicting the future events. Forecasting is an art because subjective
assessment coupled with historical and contemporary judgment is required
to improve the accuracy of forecasts. It is a science because a wide variety
of numerical methods are used to obtain a number or several numbers and
further analysed using mathematical models to ascertain the accuracy of
forecast.
In most of the cases, forecasting may involve taking historical data and
projecting them into the future with some sort of mathematical model. It may
be a subjective or an intuitive prediction. Many times, the forecasting may
even resemble a kind of a wild guess or may involve extensive data analysis
involving several parameters. Sometimes, it may involve a combination of a
mathematical model adjusted by a manager’s good judgment.
Forecasting is synonymous with estimating and prediction, though
forecasting is considered to be more scientific rather than a crude or vague
guesswork.

4.3 The Strategic Importance of Forecasting

Forecast is very much required for all types of industrial activity and also for
those industries which are purely in the service sector like healthcare and
education. Good forecasts are of critical importance in all aspects of a
business: The forecast is the only estimate of demand until the actual
demand becomes known. Therefore, forecast is said to drive decisions in
many business areas. Forecast influences three key activities. They are:
 Human resources
 Capacity
 Supply chain management
Let us now discuss these three activities in detail.
4.3.1 Human resources
Typically, the number of persons required is a function of the production
output which, in turn, depends on demand forecasting. Hence hiring,
training, and laying off workers, all depend on the anticipated demand.
When fresh workers are hired anticipating a rise in the demand, it is
expected that they can quickly get into the required job. However, training is
required apart from developing good relations with the existing workers.
Similarly, if workers are removed, it sets a demoralising atmosphere.
Further, the removed workers will spread the news, and the industry will
suffer due to bad reputation and poor image.
4.3.2 Capacity
Capacity refers to the ability to meet the demand in terms of resources and
the preparedness on the part of the company. When the demand pattern is
well recognised and indicates a rise, the capacity build up happens and
ensures that there are no lost sales for want of product. On the other hand,
if the demand is showing a decline, it signals a decrease in capacity. Thus,
unnecessary investments are not made. Both for capacity-lead and
capacity-lag decisions, demand forecasting is vital.
4.3.3 Supply chain management
Supply chain management refers to all the activities that enable the right
product at the right place at the right price. Hence, demand forecasting has
to be done with utmost care to help identifying the vendors, pricing choices,
and material options. When the demand is properly forecasted, it is easy to
plan for the suppliers, logistics, and other intermediaries to ensure the
delivery of the product at the right time.
Self Assessment Questions

1. Forecasting is both art and science of predicting the future events.

(True/ False)
2. Demand forecasting is vital both for capacity-lead and capacity-lag
decisions. (True / False)
3. Forecasts are not always perfect because the demand for a certain
product or service is a complex function influenced by a multitude of
variables. (True / False)

4.4 Why Forecasting is required?

Forecasting is required for:
 Production planning
 Financial planning
 Personnel planning
 Scheduling planning
 Facilities planning
 Process design and planning
4.4.1 Benefits from forecasts
Forecasting basically helps to overcome the uncertainty about the demand
and thus provides a workable solution. Without the forecast, no production
function can be taken up. Hence, it can be stated that forecasting helps to:
 Improve employee relations
 Improve materials management
 Get better use of capital and facilities
 Improve customer service
4.4.2 Cost implications of forecasting
Forecasting requires special efforts and involves inputs from experts which
cost a lot to the companies. Well-trained experts and associations
substantially invest in human resources and hence charge their clients for
the service rendered. Thus, forecasting done in-house or carried out
externally requires significant investments. Thus, it can be said that more
the efforts put for forecasting, more will be the cost of forecasting. Because
of improved accuracy and better judgment, the losses that would occur
because of poor forecasting would decrease as more efforts are put in for
forecasting. Hence, higher the efforts, lower will be the losses. Because
effort is a direct function of forecasting, this cost goes up with increase in the
forecasting efforts. Figure 4.1 depicts the forecasting – cost implications
graphically.

Fig. 4.1: Forecasting – Cost Implications

From figure 4.1, it is to be understood that to keep the total cost of

forecasting to a minimum, it is necessary that the forecasting effort has to be
raised up to a level at which certain uncertainty is acceptable and hence,
there is preparedness for some possible loss. On the other hand, it doesn’t
make sense to increase the effort to improve the accuracy of forecasting
because the forecasts are subject to market dynamics and many other
unpredictable parameters which will not be known or controllable. For
example, the government’s import policy drastically affects the capacity and
thus any industry hoping of increasing the demand and expanding the
capacity will face a major threat and possible loss.
4.4.3 Decision making using forecasting
Forecasts are always subject to uncertainty because of the changing
environment and hence, any attempt to improve the forecast accuracy only
increases the cost but not the accuracy. Keeping this in mind, the
managerial decision makers adopt the following rule:
Actual decision = Decision assuming forecasting is correct + Allowance for
forecast error.
Further to account for the uncertainty and provide allowance, it is necessary
that the forecast output contains two numbers as follows:
1) Best estimate of the demand + 2) Error
Again the question, how much of error can creep in the forecast? It is
difficult to answer. However, the error in forecast is easy to calculate once
the actual demand is known.
Forecast error = Actual demand - Forecast demand
Figure 4.2 depicts the process of forecasting and the associated factors.

Fig. 4.2: Forecast Generation and Revision

What is being forecasted is called as the variable of interest
In the forecasting jargon, traditionally, it is assumed that it is the demand
that is always connected with sales. However in a general way, forecast
may refer to several things including the outcome of any process, naturally
occurring or created for a specific purpose. For example, election results,
material requirement, population, economic growth, weather, and even
tourists visiting a certain place.
Further, while forecasting, it helps to ask the following to improve the
accuracy of forecast:
 Input elements involved
 Process generating the variable
 Availability of data
 Stability of the process
 Accuracy of data
 Adequacy of data
 Representativeness of the data

4.5 Classification of Forecasting Process

According to Heizer and Render (2008), the forecasting methods can be
classified based on the context or focus. The different forecasting methods
are discussed below.
Based on the type of database, the forecasting methods can be classified
into 2. They are:
 Quantitative (Statistical forecasting)
 Qualitative (Subjective estimation)
Based on the forecast time period, the forecasting methods can be
classified into 3. They are:
 Short range – up to 1 year
 Medium range – 1 to 3 years
 Long range – 5 years or more
Based on the methodology, the forecasting methods can be classified into 3.
They are:
 Time – series methods
 Causal methods
 Predictive methods (Qualitative methods)
Let us now discuss these methods in brief.
When forecasting for an existing product or similar to it, historical data will
be available and hence statistical methods can be applied. These include
simple time series models to very extensive methods like response surface
methodology. If no numerical data is available, then the opinion-based data
containing qualitative descriptions like good, bad, acceptable, etc will have
to be used. Also in some methods, the expert’s opinion or panel’s comments
will be used. Regarding the time horizon, it is obvious that some elements or
variables of interest will have to be predicted over long ranges like housing
development or global warming effects. In some cases, particularly for short
periods, some parameters like fuel prices, material availability, labour
availability, etc have to be predicted. Keeping these situations, several
methods of forecasting have been proposed and not all of them will be
relevant or useful under all situations. Thus, the decision maker has to make
a careful evaluation of all the choices before choosing the method.
Self Assessment Questions
4. Forecasting basically helps to overcome the about the
demand and thus provides a workable solution.
5. Forecasting whether done in-house or carried out externally requires
significant investments. (True / False)
6. Forecast error = - Forecast demand.

4.6 Methods of Forecasting

The different methods of forecasting can be classified as follows:
Qualitative methods
 Market surveys
 Nominal group testing
 Historical analysis
 Jury of executive opinion
 Life cycle analysis
 Delphi method
Quantitative methods
Table 4.1: Quantitative methods classification
Time series analysis Causal methods

 Moving averages  Regression analysis

 Exponential moving averages  Input – output model
 Box – Jenkins method  Leading indicators
 Trend projections  Simulations model
 Fourier series  Economic models

4.7 Case-let
Demand for Light Commercial Vehicles (LCV) in India
At present, Tata Motors dominates the market of LCV in India with a market
share of about 52% owing the success largely to their model, Tata Ace.
Mahindra group enjoys around 25% and the other players are Ashok
Leyland, Piaggio, and Eicher Motors.
The demand for LCVs is affected by rising interest rates, decreasing
industrial output, and a considerable increase in the vehicle prices. The
operating cost and environment too have changed significantly affecting the
sales.
In view of the decreasing demand, the manufacturers have to face the
challenge of reduced capacity utilisation. With freight rates almost stagnant,
the market seems to be dull in the short run. However, the long-term growth
holds promise as there will be economic changes. The LCV industry is
expected to grow by 17-18% in the financial year 2011-12. According to the
research conducted by J.D. Power Asia Pacific, India will be the third largest
LCV market by 2020. The report also says that given India’s poor road
infrastructure and concerns about fuel consumption, micro-van and mini
truck models in the LCV segment are likely to be popular.
(Source: Patel, J, LCV Industry – Begging to Differ, Business India, 4 th march
2012, page 30)
Discussion Questions:
a. What forecasting method might be suitable for the LCV segment?
b. How do you differentiate between poor growth in the short term and
promising growth in the long term?

4.8 Forecasting and Product Life Cycle

The demand for a product keeps changing as it passes through different
stages in its life cycle. The demand starts with zero value and keeps rising
as the product moves along the life cycle and gradually diminishes once the
product is outdated or obsolete. For example, iPad and iPhone are currently
in the growth stage and hence, the forecast can be trend based and on the
rise. But a product like a fixed telephone instrument is in the decline stage
and already being phased out. Hence, its forecast has to be carefully done
keeping in mind the decreasing sales. If methods are applied mechanically
without observing the current stage of the product, the forecast is bound to
be erroneous and leads to catastrophe. Figure 4.3 depicts the product life
cycle and volume of demand graphically.

Fig. 4.3: Product Life Cycle and Volume of Demand

Similarly, many of the web or internet-based transactions like online broking,
mail ordering, voice messaging, and online banking are witnessing a surge
in the demand and more people will start availing the service. In the banking
sector, the IT-enabled services are growing rapidly and hence have
decreased the demand for staff. Hence, when forecasting in such cases,
people need to understand the possible life left in the product and estimate
the demand. Further, there are also instances where the new product has
not gone through all the stages but became a failure after it was launched.
For example, pagers had a short life before giving way to mobile phones.

4.9 Selection of the Forecasting Method

Given the fact that several methods are available, the question is how to
select the right method for forecasting. Because cost, time, and skills are
involved, the choice of a forecasting method is based on several factors.
They are:
 Form of forecast required
 Forecast horizon, period, and interval
 Data availability
 Accuracy required
 Behaviour of process being forecasted (demand pattern)
 Cost of development, installation, and operation
 Case of operation
 Management comprehension and cooperation

4.10 Qualitative Methods of Forecasting

The different qualitative methods of forecasting are as follows:
 Market surveys
 Nominal group testing
 Historical analysis
 Jury of executive opinion
 Life cycle analysis
 Delphi method
Let us now discuss these qualitative methods in detail.
Market surveys
Conducting surveys among the prospective buyers or users is a very old
method of forecasting. Here, a questionnaire is prepared and circulated
among the people and their responses are obtained. The responses are
collated and analysed to reveal possible clues towards acceptance or
otherwise about a new product or service. Based on the overall decision, the
forecasting is done. This method is typically done for new products or at
new places where a product is to be launched. In this method, the number
of respondents and how responses are gathered like through oral
interviews, personal talks, internet based, postal ballots, etc, have to be
established before survey. The common limitations are the sample size and
the way of drawing the sample like random, convenient, or judgmental.
Sample bias is not completely ruled out.
Nominal group testing
In the nominal group testing method, the product or service may be given a
trial use to a specified group like students, employees, neighbors, etc and
their responses are collected and analysed.
Historical analysis
The historical analysis method is based on the fact that the past is an
indicator of the future. People try to associate the events that happened
earlier with the events that are likely to happen in the future.
Jury of executive opinion
In the jury of executive opinion method, the opinion of a group of experts is
collected and used as an estimate to obtain the forecast.
Life cycle analysis
In the life cycle analysis method, an assessment of the life cycle stage in
which the product lies is made first and an opinion is formed.
Delphi method
In the Delphi method, the experts give their opinions which are collected by
the coordinator and several rounds of discussion may be held before a
consensus is reached. This forms the basis for forecasting.
4.11 Quantitative Methods
Under the quantitative methods, adequate data is collected and different
statistical techniques are applied to reveal some patterns which will serve as
forecast. Because these methods heavily rely upon the data, it is essential
that the collected data is free from errors and bias so that proper
conclusions can be made. Another advantage is the availability of several
softwares including MS Excel which help to analyse the data extensively
which makes these methods easy to learn and apply.
The quantitative methods are divided into two groups. They are time series
analysis and causal methods.
Table 4.2: Classification of Quantitative methods

Time series analysis Causal methods

 Moving averages  Regression analysis

 Exponential moving averages  Input – output model
 Box – Jenkins method  Leading indicators
 Trend projections  Simulations model
 Fourier series  Economic models

In the time series methods, one set of data or several sets are analysed to
obtain the forecast. In the causal methods, the association between two
variables forms the basis for forecasting.
4.11.1 What is a time series?
A time series is defined as a set of values pertaining to a variable collected
at regular intervals (weekly, quarterly, or yearly). For example, the
temperature recorded every one hour is time series. Similarly, the annual
rainfall or agricultural output forms a time series. However, it is to be noted
that to draw a reasonable conclusion, at least observations should be
available. With very little number of values, say 7 or 8, the forecasts will not
be accurate.
A time series consists of four components namely:
1) Trend
2) Cyclic
3) Seasonality
4) Random
Let us now discuss these components in brief.
Trend refers to the gradual upward or downward movement of data over a
long period of time. Cycles are repetitions of data in a certain pattern at
regular intervals like several years. Business cycles are very commonly
used to understand the mood of the markets. Seasonal pattern is also a
repetitive pattern but observed at much lesser frequency. The season length
could be every hour in a day, a day, a week, or months. Variations are
noticed at each time period and patterns are observed. Random variations
are difficult to predict and are caused by chance factors or unusual events.
For example, tsunami wrecked the tourist inflow at several popular holiday
destinations all over the world. Figure 4.4 depicts the four components of a
time series.

Fig. 4.4: Components of a Time Series

4.11.2 Naïve method

In the naive method, a set of observations pertaining to a certain variable
like sales, production, or consumption is observed and the forecast is taken
as the same value as that of the most recent period.
For example, as stated in www.icra.in the number of small car offerings is
given in the table 4.3 below.

Table 4.3: Number of small car offerings

1 2 3 4 5 6 7
Year 2007 2008 2009 2010 2011 2012 2013

Number of small
9 14 17 23 26 31 35
car offerings

The forecast as given by ICRA is 35 for the year 2013.

Fig. 4.5: Graph showing number of small car offerings

Using the naïve method, the forecast for the year 2013 is 31.

4.11.3 Moving average method

A moving average is obtained by summing and averaging the values of a
time series over a given number of periods repetitively, each time deleting
the oldest value and adding a new value. Usually, the number of time
periods chosen will be an odd number like 3 or 5 or 7. Rarely, there will be a
need to go beyond 7 periods.
Consider the same example as given before.
The three-year moving average for the first three periods = (9+14+17)/3 =
13.33
Every time an old period data is dropped and the new period data is
included to get the average. Therefore, the next three-year moving average
= (14+17+23)/3 = 18
This calculation is continued.
Similarly, the five-year moving average for the first five periods =
(9+14+17+23+26)/5 = 17.8
The next five-year moving average = (14+17+23+26+31)/5 = 22.2
Table 4.4: Moving averages
1 2 3 4 5 6 7
Year 2007 2008 2009 2010 2011 2012 2013
Number of small
9 14 17 23 26 31 35
car offerings
Moving average
13.333 18 22 26.667 30.667
3 periods
Moving average
17.8 22.2 26.4
5 periods

To decide upon the number of time periods, the following guide line may be
used:
(AP = Averaging Period)
Table 4.5: Guide line
Noise Dampening
Impulse Response Accuracy
Ability
AP = 3 Low High Low
AP = 5 Medium Medium High
AP = 7 High Low Medium

Noise dampening refers to the ability to smooth out the variations. Impulse
response enables to detect immediate changes and accuracy implies
minimum forecast error.
Drawbacks of moving average methods:
 All the past periods in the averaging period are weighted equally
 No provision is made for seasonal patterns
 Several periods of historical data must be carried forward from period to
period for calculating the forecasts
4.11.4 Weighted moving average
In the simple moving average, all the past periods in the averaging period
are weighted equally and hence, the forecast is sometimes influenced by
bigger values. Secondly, older values are not relevant, particularly in the
changing environments. Hence, to reflect those changes, the simple moving
average method is modified by using different weights to different time
periods. A weighted average generally gives more weight to recent
observations than older ones. This has an advantage over simple moving
averages that, older values are given less importance than more recent
values of a series and that the number of values included in the average can
be large while still achieving responsiveness to changes through judicious
selection of weights. However, the choice of weights is somewhat arbitrary
and is often based on trial and error approach.
For example consider the following data:
Table 4.6: Data showing the sales for six months

Sales in lakhs of
Month
Rupees
Jan 90
Feb 70
Mar 80
Apr 85
May 82
June ?
Simple moving average (4 months average)
Forecast for the month of June = ¼ [70+80+85+82] = 79.25 lakh of Rupees
Here the weights are = 1/4 for all the values.
Weighted moving average (Using weights 0.1, 0.2, 0.3, 0.4)
Forecast for the month of June
= 0.1 X 70 + 0.2 X 80 + 0.3 X 85 + 0.4 X 82 = 81.3 lakh of rupees
This simple modification to the moving average method allows forecasters
to specify the relative importance of past periods of data.
4.11.5 Exponential smoothing method
This method is a modified weighted moving average method using weights
in an exponentially increasing way. The name exponential smoothing is
derived from the way the weights are assigned to historical data: The most
recent values receive most of the weight and weights decrease
exponentially as we go back in the periods.
Each new forecast is based on the previous forecast plus a percentage of
the difference between that forecast and the actual value of the series at
that point.
New forecast = Old forecast + α [Actual value - old forecast value]
Where α = a percentage and (actual – old forecast) = an error
Mathematically,
Ft =Ft-1 + α (At –1 –Ft-1)
Where Ft = Forecast for the period t
Ft-1 = forecast for the period (t-1)
α = Smoothing constant, varying from 0 to 1
At-1 = Actual value for the period (t-1)
Typically values of α range from 0.01 to 0.50. Higher values of α respond
more rapidly to any discrepancies, i.e., the changes in time series are more
closely tracked by higher values of α.
One important limitation of simple exponential smoothing is that it is ill–
suited for data that includes long-term upward or downward movements (i.e.
trend). Use of simple exponential smoothing in such instances would
produce forecasts that are too low for upward movements and too high for
downward movements. Therefore, when trend is present in time series data,
simple exponential smoothing should not be used, instead, exponential
smoothing adjusted for trend should be used.
Exponentially weighted moving averages is the term used for exponential
smoothing.
Sometimes α is calculated as 2/(AP +1), where AP = Averaging Periods
Exponential smoothing methods have been further extended to include
other possible variations and two such methods are:
1. Double Smoothing 2. Box – Jenkins methods
These methods are beyond the scope of the present topic.
 Self Assessment Questions
7. Forecasting is broadly classified as and .
8. Delphi method is a method of forecasting.
9. A is defined as a set of values pertaining to a variable
collected at regular intervals (weekly, quarterly, or yearly).
10. The method of forecasting is based on the fact that the past
is an indicator of the future

4.12 Associative Models of Forecasting

The methods earlier discussed analysed the data pertaining to a single
variable and applying statistical methods developed the forecast. But in
industries, it has been observed that there are many situations where the
data values of one variable have some association with the data values of
another variable. Though this is not exactly a cause and effect situation, it is
possible to find out the extent of association and hence, when one variable’s
value is known, the value of the other variable can be estimated using the
mathematical relationship. Correlation and regression analyses are
commonly used to establish such relationships and also help to obtain the
forecast.
Regression and correlation techniques are means of describing the
association between two or more such variables. Regression means
“dependence” and correlation measures the degree of dependence. Using
the regression equation, it is possible to estimate the value of a ‘dependent’
variable, Y, from an ‘independent‘ variable, X.
Two types of regressions are possible:
a) Simple regression involves only one independent variable which affects
the dependent variable. For example, the gold price is influenced by,
say, rising income.
b) Multiple regressions involve two or more independent variables
influencing the dependent variable. For example, the real estate price is
influenced by rising income, shortage of land, and urban migration.
Regression is also categorised as linear and non–linear regression based
on the severity of relationship and characteristics. The following table 4.7
shows the examples.
 Table 4.7: Linear and non–linear regression

Simple Multiple
Linear Y= a +b x Y=a+bx1+cx2+dx3
2 3
Non-linear Y= a+bx2 Y=A+BX1+CX2 +DX3
It is to be remembered that known variable = Independent variable and
unknown variable = Dependent variable.
The forecasting procedures using regression involves the following steps:
1. The variables are plotted along Cartesian or rectangular coordinates
2. A trend equation is developed
3. The equation is used for forecasting
4. The variables are not necessarily related on a time basis
The most popular form of the simple linear regression equation is Y = a+bX,
where
Y= Dependent variable
X= Independent variable
a = Y intercept
b = Slope
For a unit change in the value of X, a change in the value of Y occurs in a
straight line manner and hence, it is easy to predict the values.
To find the values of constants a and b, the following equations are used:
∑y = na + b∑x
∑xy = a ∑x +b∑x2
Solving the above two equations, the values of a and b are obtained and
then substituted into the regression equation along with the value of X and
the value of Y is determined.
Example:
A departmental store has collected data about sales figures and profits
during the last 12 months. Obtain a regression line for the data and predict
the profit when sale is 10 Rs. lakh.
 Table 4.8: Data on sales
Sales X
7 2 6 4 14 15 16 12 14 20 15 7
(Rs. Lakh):
Profit Y
(Rs. 15 10 13 15 25 27 24 20 27 44 24 17
thousands)

First plot the data and decide if a linear model is reasonable (i.e. do the
points seem to scatter around a straight line?) Next compute the quantities
∑x, ∑y, ∑xy and ∑x2.
Let linear regression equation is Y = a+bX, where
Y= Dependent variable, profit
X= Independent variable, sales
a = Y intercept
b = Slope
The following table is constructed:
Table 4.9: Table Construction for regression

X Y XY X
2
Y
2 Ŷ
7 15 105 49 225 16.15
2 10 20 4 100 8.186
6 13 78 36 169 14.558
4 15 60 16 225 11.372
14 25 350 196 625 27.302
15 27 405 225 729 28.895
16 24 384 256 576 30.488
12 20 240 144 400 24.116
14 27 378 196 729 27.302
20 44 880 400 1936 36.86
15 34 510 225 1156 28.895
7 17 119 49 289 16.15
Total ∑ 132 271 3529 1796 7159

Solving for a and b, we get a = 5.06 and b = 1.593

Therefore, the regression line is
Y = 5.06 + 1.593 x
When sales X=10,
Y = 5.06 +1.593 X = 20.99 Rs. Thousands
Self Practice Problem
Maxwell Motors has collected the following data on annual spare parts sales
and new car registrations:
Table 4.10: Data on annual spare parts
Annual spare parts sales 1.0 1.4 1.9 2.0 1.8 2.1 2.3
(in million $)
New car registrations (in 10 12 15 16 14 17 20
thousands)
Develop the linear regression equation and estimate the spare parts sales
when new car registrations are 22,000.
Answer:
Linear Regression equation:
Y = -0.16 + 0.13 X
Spare parts sales when new car registrations are 22,000 = 2.7 million $

4.13 Accuracy of Forecasting

Finally, we come to an important question. How accurate are the forecasts
obtained by the different methods. Any forecast method results in values
that may not exactly match the actual values. Hence, deviations are
expected. Higher the deviation, more will be the error. Several measures of
error in forecast have been developed to examine the issue of error in
forecast. Here, we look at two widely used and popular measure applicable
to a wide variety of methods. These two measures are: (1) Mean Absolute
Deviation (MAD), and (2) Standard Error of Estimate (SE).
4.13.1 Mean Absolute Deviation (MAD)
In short range forecasting, MAD is often used to measure how closely
forecast values are matching the actual data. MAD is computed as follows:
MAD = Sum of absolute deviations for n periods / number of periods.
Here deviation = Difference between the actual value and the forecast
value.
These deviations are added and divided by the number of time periods to
get the MAD.
4.13.2 Standard Error (SE) of estimate
The standard error of estimate measures the variability or scatter of the
observed values around the regression line.
The formula for calculating SE is given below:

where
y = values of the dependent variable
yest = Estimated values from the estimating equation that correspond to each
Y value.
n = number of data points used to fit the regression line.
It is important to note that the error values should be as low as possible
while making comparison and selection.

4.14 Summary
Let us now summarise the key learnings of this unit:
 For any business activity to be started, the first step would be to predict
the demand and then to develop the plans towards meeting the demand
either partially or fully. This process of estimating is called forecasting
 Forecasting basically helps to overcome the uncertainty about the
demand and thus provides a workable solution.
 Supply chain management refers to all the activities that enable the right
product at the right place at the right price. Hence, demand forecasting
has to be done with utmost care to help identifying the vendors, pricing
choices, and material options.
 Forecasting is broadly classified as quantitative and qualitative
 Market survey, Delphi method, historical analysis are some of the
qualitative methods of forecasting
 The quantitative methods are divided into two groups. They are time
series analysis and causal methods.
 Several measures of error in forecast have been developed to examine
the issue of error in forecast. There are two widely used and popular
measure applicable to a wide variety of methods. These two measures
are: (1) Mean Absolute Deviation and (2) Standard Error of Estimate.

4.15 Glossary
 Survey: A detailed study of a market or geographical area to gather
data on demand for a product or service, attitudes, opinions, satisfaction
level, etc
 Questionnaire: A form containing a set of questions submitted to
people to gain statistical information

4.16 Terminal Questions

1. What is meant by forecasting?
2. Discuss the role of forecasting in modern business context.
3. List the benefits of forecasting.
4. Distinguish between moving average method and weighted moving
average method. What are the advantages and disadvantages of these
methods?
5. How do you measure the accuracy of forecast? Describe one measure
of error used in forecasting.

4.17 Answers

Self Assessment Questions

1. True
2. True
3. True
4. Uncertainty
5. True
6. Actual demand
7. quantitative, qualitative
8. qualitative
9. time series
10. historical analysis