Operations Management 18ME56

Module 2 : FORECASTING

Production and Inventory control activities:

A few years ago, a small test equipment manufacturer in Bombay received a corporation
directive to improve their business operations.

With the help of a consultant, they decided to discard their manual production control system and
undertake a five-phase program to gain better control of their costs. Here’s what happened.

• The material requirements, estimated costs and inventory records were

computerized within four months, they began to check actual inventories against
the computerized data base and analyse any variances.

• The general ledger and financial data were integrated into the system a month
later.

• Ten months after that, the payroll and labor distribution information was
transferred from their bank to the system;

• It was automatically interfaced to the job costing system. Finally, the order entry
information and invoicing was incorporated.

The manufacturing control system took about two years to implement and saved the firm
Rs. 152000/- in the first year of operations.

In the second chapter we had discussed on the operations strategy, which is embodied in the long
range operations/production plan.

While all elements of operations management are important, I view forecasting as one of the
key elements in the operations structure. In this chapter, helps us to recognize the models and
when to use for our needs.

Department of Mechanical Engineering, ATMECE 1

Operations Management 18ME56

Specifies positioning, strategy, product process and technology plans, strategic allocation of
resources and facility planning shown in fig 1.

Department of Mechanical Engineering, ATMECE 2

 Operations Management 18ME56

Industry

Market and competition

Organizational strategy

Profit or return
Source of funds
Product or service quality

Operational policy
Conversion characteristics: design
Product design flexibility
Delivery capability location of facilities
Processing technology
Control systems

Managing conversion operations

Quality
Efficiency
Schedule

Results

Fig:1 Strategic perspective

Forecast of a product is an estimate of its future demand. However, it is not a prediction. A
forecast is however based upon scientific analysis of past data, if available and by other techniques.

Once these are in place, the fundamental structure of the operation function is established.
Before, resources can be planned but, it is critical to estimate or forecast long-range and shortrange
demand for products and services.

These forecasts guide the strategic allocation of resources. Based on the expected levels
of demand, decisions are made concerning product, process and service designs, facility
capacity, location and layout, operations technologies and allocation of operations resources.

Other issues involving the strategic allocation of resources include managing quality,
planning service operations and managing projects.

Department of Mechanical Engineering, ATMECE 3

 Operations Management 18ME56

Forecasting meals on airline flights

Providing in-flight meals to the airline passengers is big business.
Few companies’ which have business are listed below
Northwest airlines and continental’s food budget per year: $ 300 million dollars
Delta serves about 135,000 meals per day
American airlines spends around $800 million each ear on food with each meals cost is $8.20
With this huge expense, the airlines are interested in accurately forecasting the number of meals
that will be needed in each flight.
Factors that make airline meal forecasting:
• Passengers purchasing tickets just before a flight
• Cancelled flights
• Passengers no-shows
• Complicate maters
• Some passengers decide not to have meals,
• Children can request a kid’ meals
• Some passengers request special-diet meals,
• First class passengers receive different meals than economy class passengers and
may have two or more choices of meals.
• Some flights may have 60% full and while others may be 100%
• If an airline orders too many meals for a flight, extra meals must be thrown away,
although some items such as boxes cereal might be given to charity.
• If it does not order enough meals, then hungry passengers may be upset and may not
fly on that airline in the future.

Shortages of meals statics:

Last year: 1% shortage

Continental had average meal shortage : 0.6 %
Excess meal : 3.5%
At Home base: 5%
To satisfies the customers of first class passengers the airline orders 125percent instead of 100
percent.
Accurate demand forecasting is critical to providing good customer service in a cost-efficient
manner.
Forecasting enables his company to respond more quickly and accurately to market changes.
How does forecasting relate to the management processes of planning, organizing and
controlling? These processes are not independent processes. They interrelate and overlap.
If operations have been properly planned and organized, control is easier and smoother. These
were forecasting comes in. cost can be reduced and accurately goods and services can be
estimated and this in turn improves operating efficiently increases. The figure below explain the
relationships of P O C and the forecasting plan.

Department of Mechanical Engineering, ATMECE 4

 Operations Management 18ME56

Fig: 2 operations and production management activities

Operations managers need long range forecasts to make strategic decisions about products,
processes and facilities. They also need short-range forecasts to assist them in making decisions
about operations issues that span for few days or weeks. The following table 1 shows
summarizes some of the reasons why operations managers must develop forecasts.
Table 1 Some Reason Why Forecasting Is Essential in Operations Management

1. New facility planning. It can take as long as five years to design and build a new
factory or design and implement a new production process. Such strategic activities in POM
require long –range forecasts of demand for existing and new products so that operation
managers can have the necessary lead time to build factories and install process to produce
the products and services when needed.
2. Production Planning. Demand for products and services vary from month to month.
Production and
Services rates must be scaled up or down to meet these demands. It can take several months
to change the capacities of production processes. Operation managers need medium-range
forecasts so that they can have the lead time necessary to provide the production capacity to
produce these variable monthly demands.
3. Workforce scheduling. Demands for products and services vary from week to week. The
workforce must be scaled up or down to meet these demands by using reassignment,
overtime, layoffs, or hiring. Operations managers need short-range forecasts so that they
can have the lead time necessary to provide workforce change to provide the weekly
demands.

Department of Mechanical Engineering, ATMECE 5

 Operations Management 18ME56

In the table 2 shows examples of things that are commonly forecasted.

Forecasting Horzion Time Span Example of Things Some typical

Long range Years That must Be Forecasted units of forecasts
New Products lines Dollars
Old Products lines Gallons, hours, pounds
Factory Capacities Units or customers per
Time periods

Dollars
Capital funds Space,Volume
Facility needs
Medium range Months Product groups Units
Department capacities Hours, strokes, pounds,
Gallons, Units or customer
per time period

Workforce Workers, hours

Purchased materials Units, pounds, gallons

Inventories Units,dollars

Short range Weeks Specific products Units

Labour skill classes Workers, hours
Machine capacities Units , hours, gallons
Pounds or customers per
time period

Cash Dollars

Inventories units, dollar

Forecasting is an integral part of business planning. The inputs are processed through forecasting
models or method to develop demand estimates.

Theses demand estimates are not the sales forecasts; rather, they are the starting point for
management teams to develop sales forecasts.

The sales forecasts become inputs to both business strategy and production resource forecasts.

Department of Mechanical Engineering, ATMECE 6

Operations Management 18ME56

Fig: 3 Forecasting as an Integral Part of Business Planning

WHAT IS A FORECAST?

• Forecasting is the basis of planning ahead. It involves estimating the future and the
expected demand of the company’s product.
• Forecasts of future demand is the company’s expectation with the outside
environment that permits planning functions to commence activities.
• While forecasting is not exactly planning it just puts planning action into motion.
• Forecasts are estimates of the occurrence, timing, or magnitude of future events.
• They give operations managers a rational basis for planning and scheduling
activities, even though actual demand is quite uncertain.

Department of Mechanical Engineering, ATMECE 7

 Operations Management 18ME56

WHY DO FIRMS FORECAST?

• Forecasting of independent demands, item by item, is required to maintain the

supply of materials for production in anticipation of future demand.
• Forecast is important I case of advance commitment to procure or to produce.
From forecast of demands optional plans are adapted.
• Accurate projections of future activity levels can minimize short term fluctuations
in production and help balance workloads.
• This lessens hiring, firing, and overtime activities and helps maintain good labor
relations.
• Good forecasts also help managers to have appropriate levels of materials available
when needed.
• Forecast enable managers to make better use of facilities and give improved
services to customers.
• Benefits from forecasts
1. Improved employee relations
2. Improved Materials management
3. Better use of capital and facilities
4. Improved customer service

COST OF FORECASTING

• As forecasting activities increases the data requirements also increases, hence

increasing the cost of data collection and analysis.
• The system for reporting and control must also be expanded resulting in
increased cost.
• On the other hand if forecasting is not done it might lead to reduced activities
and result in loses in terms of unplanned labor, material, capital costs, expediting
costs, and ultimately lead to lost revenues.

OUR APPROACH TO FORECASTING

To gain an appreciation of the value of forecasting and an understanding of some of the more
widely used techniques, we discuss the following methods of forecasting.

1. Judgmental
2. Time series
3. Exponential smoothing
4. Regression methods

Department of Mechanical Engineering, ATMECE 8

 Operations Management 18ME56

FORECASTING VARIABLES

Forecasting activities are a function of the following

1. Type of forecast
2. Time horizon being forecast
3. Database available
4. Methodology employed

We discuss one by one in detail

Type of forecast.

Most of the items produced in a firm do not need forecast in a formal way, because they are
components, subassemblies or required services that are part of a finished product.
Forecasts should be used for end items and services that have uncertain demand. Other types
of forecasts

Purpose:

The purpose of forecasting activities is to make the best use of the present information to
guide decisions toward the objectives of the organization. Managers should continually
make decisions about;
1. Purchasing new equipment
2. Setting employment levels
3. Carrying inventories
4. Scheduling production …etc.

Types of variables being forecast:

There are two types of variables being forecast;

1. Controllable
2. Uncontrollable

Example: Sales of a firm is a function of both controllable variables such as advertising efforts
and inventory levels where as the uncontrollable variables are competition in the market and
raw-material cost.

Forecasting methodology help by providing information about the uncontrollable variables.

Accuracy:

Department of Mechanical Engineering, ATMECE 9

 Operations Management 18ME56

• Forecasts tend to be more accurate when the uncontrollable variables of a

variable can be identified and isolated.
• In general the more the random effects can be isolated, the better the forecast
will be.
• Whereas individual forecasts are susceptible to error due to spontaneous random
effects (which cannot be anticipated), when several projects are aggregated
together, the error effect is dissipated throughout the group, and compensating
effects occur.
• One products demand may exceed the forecast, while another’s might fail to
meet it.
• But as a whole, the aggregate forecast generally tends to be more accurate than
individual product forecast.

Types of forecast:

• Example; the manager who must decide whether to invest in a computer system
this year (or wait till next year) faces a different problem from the one who must
decide how much inventory to place in stock. Here the former must grapple with
the pace of technology whereas the latter must project future demand.
• Manager must select or develop those types of forecasts that will be most useful
to them in their specific area of concern.
• Forecasts of demand are specifically important to operations managers because
they guide the firm’s scheduling and production control activities.
• Reliable forecasts enable managers to formulate material and capacity plan
directing how their system will respond.
• Technological forecasts are concerned with the pace of new developments in
technology, such as developments in storage devises that will increase the
capacity and decrease the cost of computers.
• Environmental forecasts are concerned with the social, political, and economic
state of the environment.
• Econometric forecasts provide forecasts of the gross national product, consumer
prices, unemployment, housing starts or other economic variables of particular
interest to the firm.

Forecasting and operations subsystems:

In the production units – number for televisions in a plant, the number of patients fed in a hospital,
the number of books circulated in a library, or the number of lots of common stock sold in a
brokerage house – the resource forecasts are used to plan and control operation subsystems, as
shown in figure 4.

Department of Mechanical Engineering, ATMECE 10

 Operations Management 18ME56

Information on most recent

demand and production

Demand forecast for

operations

Planning the system Scheduling the system Controlling the system

(designing) Aggregate production Production control
Product design planning Inventory control
Process design Operations scheduling Labor control
Equipment investment and Cost control
replacement
Capacity planning

Output of
goods and services

Fig4: Using demand forecasting and production/operations subsystems

Planning (designing) the system:

Managers need to forecast aggregate demands so they can design or redesign processes
necessary to meet demand.
The degree of automation for example: Depends a great deal upon future product demand.
• Automated
• Continuous flows facilitate high production volumes
• Manual or semi automated
• Intermittent flows (batching)

The demand forecast is critical to this design decision. Once process design, product design and
equipment investment decision have been made for an anticipated volume, mangers are locked
into a facility of specified capacity.

There may be wide variations between anticipated demand and actual demand can result in
excessive production and operating cost.
Department of Mechanical Engineering, ATMECE 11
 Operations Management 18ME56

Capacity planning that makes use of long-run forecasts is one of the areas in
production/operations that is both critical and not well understood or developed. In the steel,
power generation and other basic industries, Ex: jet aircraft, Mc Donnell Douglas and Airbus,
facilities becomes idle some time.

Scheduling the system:

When deciding how best to use the existing conversion system, accurate demand forecasts are very
important.
Managers need intermediate-run demand forecasts for three months, six months and a year into
the future.
Forecast must be established form the current and future work force levels and production
rates. Job scheduling in intermittent and continuous operations is more stable if demand
forecasts are accurate.

Controlling the system:

Managers need forecasts of demand to make decisions about controlling inventory, production,
labor and overall costs.
Accurate forecasts are needed for the immediate future – hours, days and weeks ahead.

The demand patterns are shown in the following figures

150
140
130

80
5 10 15 20
Fig: 5.1 Steady Demands

150
140
130

80
5 10 15 20
Fig: 5.2 Demand with increasing trend.

Department of Mechanical Engineering, ATMECE 12

 Operations Management 18ME56

150
140
130

80
5 10 15 20
Fig: 5.3 Seasonal Demands.

150
140
130

80
5 10 15 20
Fig: 5.4 Seasonal Demand with rising trend.

Noise in Demand:

® ®®
® ®
®® ®®
®® ®® ®®
®® ®®

® Low Noise
High Noise
Time

To describe the points clustered about a pattern, we use the term NOISE.
we have two type of NOISE
LOW NOISE: Means all or most of the points lie very close to the pattern.
HIGH NOISE: Means many of the points lie relatively far away from the pattern.
In decision making, we deal with devising future plans. The data describing the decision
situation must thus be representative of what occurs in the future. For ex:
• An inventory control, we base our decisions on the nature of demand for the
controlled item during a specified planning horizon.
• In financial planning, we need to predict the pattern of cash flow overtime.

We know forecasts are of two kinds

Department of Mechanical Engineering, ATMECE 13

 Operations Management 18ME56

• Long range
• short range

The long range is making forecast on capacity, location and layout. The short range makes
forecast on the individual items. The following figure3 shows different types of planning
decisions depend on different types of information, which in turn depend on what are called the
forecasting time horizons, of the future times to which the forecasting points.

Differentiate between Forecast and Prediction:

Forecast Prediction
A forecast is an estimate of a future event A predication is an estimate of a future event
achieved by systematically combining and achieved through subjective consideration other
casting forward in a predetermined way data than just past data; this subjective consideration
about the past. need not occur in any predetermined way.

Type of Representative
Decision Information
Needs

Short Specific
Planning Item
Decisions Demands

Aggregate
Demands

Long – run Strategies

Planning and
Decisions Facilities
Present Five Years
Hence
Fig. 3 Forecasting time Horizon

EX: for Forecasting

A TV manufacturer, for example can use past data to forecast the number of picture screens
required for next week’s TV assembly schedule.
A fast food restaurant can use past data to forecast the number of hamburger buns required for this
weekend’s operations.

Department of Mechanical Engineering, ATMECE 14

 Operations Management 18ME56

EX: for Prediction:

Suppose the manufacturer offers a new TV model or the restaurant decides to offer a new item.
Since, no past data exist to estimate first year sales of the new product, prediction, not
forecasting is required.
For predicting good subjective estimates can be based on the manger’s skill, experience and
judgment; but, forecasting requires statistical and management science techniques.

TIME HORIZON

• Forecasts are often classified according the time period.

• Short – range----up to 1 year (typically 0-3 months); these forecasts serve primarily
as guides for current operation.
• Medium – Range----1 to 3 years;
• Long – range----5 years or more; Medium and longer range forecasts are often of
more comprehensive or aggregated nature.
• A 3-5 year forecast may be necessary to support plant capacity decisions, whereas
product- line and plant location decisions may require longer forecasts.
• Product life and seasonal factors affect the length of forecasts.
• Products in their earlier stage of development will require longer forecasts than those
in the declining stage.
• The forecasts are needed for planning different employment and inventory levels as
the product phases through the various stages of growth and maturity.

DATABASE: QUANTITATIVE AND QUALITATIVE

• Most forecasting rely on quantitative data- it is the basis for scientific decision
making. It enhances the objective of the model and forces precision.
• Some variables cannot be quantified, or the quantification process itself is biased.
• In some cases, the models that a firm designs (or can afford) cannot accommodate
the variable that the firm might like to include. Some judgmental allowance must be
made for the models inadequacy.
• Even the most sophisticated models used need the balance of a good judgment.
• Testing the model on past data or simulated data can be an effect check of its
adequacy.
In the modern forecasting techniques. The techniques have been grouped into qualitative models,
time series models and causal models
The most frequently used techniques in operations management are the qualitative and time
series models
The casual models are often more costly to implement and do not offer the increased accuracy
for short-term forecasting typically needed by the production/operations manager.

Department of Mechanical Engineering, ATMECE 15

 Operations Management 18ME56

The table shows the representative forecasting techniques

Model type Description
Qualitative models
Delphi method Questions panel of experts for opinions
Historical data Makes analogies to the past in a judgmental
manner
Nominal group technique Group process allowing participation with
forced voting.
Time Series (Quantitative Models)
Simple average Averages past data to predict the future based
on that average
Exponential smoothing Weights old forecasts and most recent
demand
Causal Quantitative Models
Regression analysis Depicts a functional relationship among
variables
Economic modeling Provides an overall forecast for a variable
such as gross national product (GNP)

1. Delphi Technique: A qualitative forecasting technique in which a panel of

experts working separately and not meeting arrive at a consensus through the
summarizing of ideas by a skilled coordinator. The procedure works as
follows:
• A coordinator poses a question, in writing; to each expert on a panel. Each
expert writes a brief prediction.
• The coordinator brings the written predictions together, edits them and
summarizes them.
• On the basis of the summary, the coordinator writes a new set of questions
and gives them to the experts. Theses are answered in writing.
• Again, the coordinator edits and summarizes the answers, repeating the
process until the coordinator is satisfied with the overall prediction
synthesized from the experts.
The key to the Delphi technique lies in the coordinator and experts. The experts frequently
have diverse backgrounds; two physicists, a chemist, an electrical engineer, and an economist
might make up a panel.
The coordinator must be talented enough to synthesize diverse and wide-ranging statements and
arrive at both a structured set of questions and a forecast.
2. Historical data
It is based on the past data with informed judgment.

Department of Mechanical Engineering, ATMECE 16

 Operations Management 18ME56

3. Nominal group technique: a qualitative forecasting technique in which a panel

of experts working together in a meeting arrive at a consensus through
discussion and ranking of ideas.
The process works like this. Seven to ten experts are asked to sit around a table in full
view of one another, but they are asked not to speak to one another.
A group facilitator hands out copies of the question needing a forecast. Each expert is
asked to write down a list of ideas about the question.
After a few minutes, the group facilitator asks each expert in turn to share one idea from
his or her list.
A recorder writes each idea on a flip chart sp that everyone can see it.
The experts continue to give their ideas in a round-robin manner until all the ideas have
been written on the flip chart.
When all the discussion has ended, the experts are asked to rank the ideas, in writing,
according to priority.

Quantitative Models:
Many models use historical data to calculate an average of past demand.
There are several ways of calculating an average.
Simple average: a simple average (SA) is the average of the demands occurring in all previous
periods. The demands of all periods are equally weighted:
Sum of demands for all periods
SA=------------------------------------------
Number of periods
∑Di
=------ where, n = the number of periods
n Di=the demand in the ith period
Example: at weld supplies, demand for a new welding rod was 50 dozen in the first quarter, 60,
dozen in the second,, and 40 dozen in the third. The average demand has been:
D1 + D2 +D3
SA=----------------
3
50+60+40
=----------------
3
= 50
A forecast for all future quarters could be based on this simple average and would be 50 dozen
welding rods per quarter.

Department of Mechanical Engineering, ATMECE 17

 Operations Management 18ME56

Simple Moving Average:

A simple moving average(MA) combines the demand data from several of the most recent
periods, their average being the forecast for the next period. Once the number of past periods
to be used in the calculations has been selected, it is held constant. We may use a 3-period
moving period or 20 moving period moving average.
A simple moving average is calculated as follows:

Sum of demands for periods

MA=----------------------------------------
Chosen number of periods
∑ Di
MA =------------ where Di =the demand in the ith
period N n= chosen number of
periods

The average effectively smoothes out fluctuations while preserving the general data.
The adaptability of the moving average is the source of major disadvantages;
however, there is no equation for forecasting.
In place of equation we use the latest moving-average value as the forecast for the next
period.
In the process of averaging gives equal importance to the most recent demand.
It ignores any trend in the period over which the data is averaged. However, we can assign
weights to components of the moving average before averaging them
∑(wt) X
MAwt = ----------

∑ wt or
Forecast for the next period is given by
E= w1 D1 + w2 D2 +w3 D3 + ……..+ wk
Dk Where: Di = demand for I periods back
wi = weight to be assigned to demand Di
k = number of periods

FORECASTING METHODOLOGY
• The complexity of forecasting methodology sometimes tends to correspond to the
event to which future events are evaluated in an objective or professional manner.
• As the amount of uncertainties of future events increases, firms tend to rely more
upon inferences and correlations based upon the present.
• When these inferences in turn come from the analysis of the data, the methodology
becomes more objective but also more complexes. Complexity does not guarantee
accuracy.

Department of Mechanical Engineering, ATMECE 18

Operations Management 18ME56

• Some techniques are best suited to long-range or new-product forecasts, whereas

others are more appropriate for production and inventory control.
• Opinion methods although subjective, are widely used, especially by small firms.
• To Large extent they rely upon personal insights, imagination, or perhaps even
guesswork. The cost is low but the accuracy is too.
• Judgments are an improvement over pure opinion in that they call on past
experience, consensus with others, or perhaps knowledge of historically analogous
situations.
• Time series methods which capitalize upon the identification of trend and seasonal
effects are data-based and are likely to be more accurate than opinion methods.
• The basic assumption is that history follows a pattern that will continue.
• Exponential smoothing methods are of this same type, for they are trajectory, or
trendbased. They are however, readily adaptive to current levels of activity and
have become increasingly popular in production and inventory control
applications.
• Regression and correlation methods are associative in nature and depend upon the
casual relationship or interaction of two or more variables.
• Box-Jenkins is a combination time series-regression approach that incorporates
some advantages of both methods.

TIME SERIES METHODS

• A time series is a set of observations of some variable over time. The series is
usually tabulated or graphed in a manner that readily conveys the behavior of the
subject variable.
• Components of a series;
1. Trend (T)
2. Cyclical (C)
3. Seasonal (S)
4. Random (R) or irregular
• In the classical model of time series analysis, the forecast (Y) is a multiplicative
function of these components : Y=TCSR
• The trend represents a long-term secular movement, characteristic of many
economic series.
• Cyclical factors are long-term swings about the trend line and are usually
associated with business cycles.
• Seasonal effects are similar patterns occurring during corresponding months of
successive years.
• Random or irregular components are sporadic effects due to chance and usually
occurrences.

Department of Mechanical Engineering, ATMECE 19

 Operations Management 18ME56

FORECASTING PROCEDURE:
1. Plot historical data to confirm the type of relationship (for example linear,
quadratic..)
2. Develop a trend equation to describe the data
3. Develop a seasonal index
4. Project the trend into the future
5. Multiply the monthly trend values by the seasonal index
6. Modify the projected values by a knowledge of:
a) Cyclical business conditions(C)
b) Anticipated irregular effects( R)
Methods of estimating trend
Freehand:
• A freehand Curve drawn smoothly through the data points is often an easy and
perhaps adequate representation of the data but this method suffers from
subjectivity.
Moving Average:
• A moving average is obtained by summing and averaging the values from a
given number of periods repetitively, each time deleting the oldest value and
adding a new value.
MA = ∑X

Number of periods
Where one X value is exchanged each period.
Example;
Compute a 3-year moving average for the aluminium tube shipments.
Note that the moving average is recorded in the center position of the data it averages. The
3.7-ton figure in the above example would thus be centered on July 1, 1978.

Example: Find the forecast for the period 11 by the method of weighted moving average by
assigning 0.5, 0.3 and 0.2 to demand of periods 10, 9, and 8. The demand are 66, 67 and 70
for the month of 8, 9, 10.

Department of Mechanical Engineering, ATMECE 20

 Operations Management 18ME56

Solution: Expected demand for period 11= 66 x 0.2+67 x 0.3 +70 x 0.5 = 67.7.

Worked Examples:
1. the sales pattern of a manufacturing firm is given below.compute the 3yearly moving
trend and find out the sales forecast for the year 1993.
Year 1985 1986 1987 1988 1989 1990 1991 1992

sales 8 8.5 9 10 9.5 11 11.5 12

SOLUTION: the 3 yearly moving average trend is computed in the following table.
year Sales(Rs. 3 yearly moving total 3 yearly moving
Lakhs) average
1985 8
1986 8.5
1987 9 25.5 8.5
1988 10 27.5 9.2
1989 9.5 28.5 9.5
1990 11 30.5 10.2
1991 11.5 32.0 10.7
1992 12 34.5 11.5

The forecast for 1993 in 11.5 which is the average of last 3 years.

Department of Mechanical Engineering, ATMECE 21

 Operations Management 18ME56

2.The demand for a product during past 20 periods and the forecasted demand by method of
moving average are given in below table. For comparison results have been given for forecast
made by averaging past demands of four and two moving average.
period Demand Moving Average Moving Average
(4 periods) (2 periods)
1 121
2 125
3 124 123
4 118 125
5 134 122 121
6 127 125 126
7 124 126 130
8 141 126 126
9 133 131 133
10 135 131 137
11 141 133 134
12 139 138 138
13 144 137 140
14 152 140 141
15 142 144 148
16 149 144 147
17 145 146 145
18 140 147 147
19 132 144 143
20 130 141 136

Fig2: comparison of forecasts with moving averages of two and four periods.
3. A food processor uses a moving average to forecast next month’s demand. Past actual demand
(in units) is as shown in the accompanying table.

Department of Mechanical Engineering, ATMECE 22

 Operations Management 18ME56

a) Compute a simple 5-month moving average to forecast demand for month 52.
b) Compute a weighted 3-month moving average where the weights are highest for
the latest months and descend in order of 3,2,1.
Month Actual demand
43 105
44 106
45 110
46 110
47 114
48 121
49 130
50 128
51 137
52

Solutions:
∑X
a) MA= -------------------------
number of periods
114+121+130+128+137
=-------------------------------
5
= 126
units b)
∑(wt)(X)
MAwt=-------------
∑wt
Where wt X value = total
3 X 137 = 411
2 X 128
= 256 1X
130 = 130
--- ------
6 797
797
MAwt=------------- = 133 units
6

Department of Mechanical Engineering, ATMECE 23

Operations Management 18ME56

4. The ABC Floral shop sold the following number of geraniums during the last 2 weeks.
Day Demand Day Demand

1 200 8 150

2 134 9 182

3 157 10 197

4 165 11 136

5 177 12 163

6 125 13 157

7 146 14 169

i. Determine the forecast for the number of Geraniums demanded on the 15th day
using three period moving average as well as five period moving average.
ii. Depict graphically the difference between forecast and the actual demand

SOLUTION:
i. ‘3’ period moving average.

Day Demand 3 day moving 3 day moving Round off to

total average nearest figure

1 200

2 134

3 157 491 163.7 164

4 165 456 152.0 152

5 177 499 166.3 166

6 125 467 155.7 156

7 146 448 149.3 149

8 150 421 140.3 140

9 182 478 159.3 159

10 197 529 176.3 176

11 136 515 171.6 172

Department of Mechanical Engineering, ATMECE 24

 Operations Management 18ME56

12 163 496 165.3 165

13 157 456 152.0 152

14 169 489 163.0 163

Forecast for the 15th day is 163.0 ii.

‘5’ period moving average

Day Demand 5 day moving 5 day moving Round off to

total average nearest figure

1 200

2 134

3 157

4 165

5 177 833 166.6 167

6 125 758 151.6 152

7 146 770 154.0 154

8 150 763 153.0 153

9 182 780 156.0 156

10 197 800 160.0 160

11 136 811 162.0 162

12 163 828 166.0 166

13 157 835 167.0 167

14 169 822 165.0 165

ii. The forecast value is the average of past sales. The forecast values (trend
values) of 3 period and 5 period moving averages and the actual demand can
be shown on the graph by plotting the trend values and actual demand on y-
axis and No. of days on x-axis.

Department of Mechanical Engineering, ATMECE 25

 Operations Management 18ME56

Least Squares:
1. Least squares is one of the most widely used methods of fitting rends to data
because it yields what is mathematically described as a “line of best fit”
2. The Trend line has the following properties;
a) The summation of all vertical deviations about it is zero
b) The summation of all vertical deviations squared is a minimum
c) The line goes through the means X and Y
Linear Equation
∑Y = na + b∑X
∑XY = a∑X + b∑X²----(1)
Where the data can be coded so that ∑X =0, two terms in the above expression drop out, and we
have:

∑Y=na
∑XY = b∑X²---(2)

Coding is easily accomplished with time series data, we simply designate the center of the
time period as X=0 and have equal number of plus and minus periods on each side which sum
to zero.

Example;

Use the least square method to develop a linear trend equation for the data below. State the equation
complete with signature, and forecast a trend value for 1992

Department of Mechanical Engineering, ATMECE 26

 Operations Management 18ME56

year X Y XY X²

Year Shipment
coded (tons)
1977 -5 2 -10 25
1978 -4 3 -12 16
1979 -3 6 -18 9
1980 -2 10 -20 4
1981 -1 8 -8 1
1982 0 7 0 0
1983 1 12 12 1
1984 2 14 28 4
1985 3 14 42 9
1986 4 18 72 16
1987 5 19 95 25
total 0 113 181 110

Rearranging equation---2, we have

a=∑Y /n = 113/11 = 10.3
b= ∑XY/∑X² = 181/110 =1.6

Therefore the forecasting equation is of the form Y=a + bX

Y= 10.3=1.6X (1982=0, X= years, Y=tons)
Forecast for 1992: Because 1992 is 10 years distant from the origin,
Y=10.3+1.6(10)=26.3 tons.
• The above example assumes that a linear equation adequately describes the data.
• The appropriateness of a linear function should always be checked first; this can
be simply done by graphing the data and observing whether a straight line would
provide a satisfactory fit.

2.The sales of a product during the last five years is tabulated below
Year: 1973 1974 1975 1976 1977
Sales: 4 8 6 10 4
Using least square method forecaster,

Department of Mechanical Engineering, ATMECE 27

 Operations Management 18ME56

SOLUTION:
Year x y x² xy

1973 -2 4 4 -8

1974 -1 8 1 -8

1975 0 6 0 0

1976 +1 10 1 +10

1977 +2 4 4 +8

∑x=0 ∑y=32 ∑x²=10 ∑xy=2

The linear equation is y=a+bx

Equation of regression line is given by

y=6.4+0.2x

Regression and correlation methods

Regression and correlation techniques are means of describing the association between
two or more such variables. They make no claim to establishing cause and effect but, instead
merely quantify the statistical dependence or extent to which the two or more variable are related.
Regression means “dependence” and involves estimating the value of a dependent
variable, Y from an independent variable X.
In simple regression only one independent variable is used, whereas in multiple
regression two or more independent variables are involved.
Simple Linear regression model
Yc = a+bX where Y=dependent variable, X=independent variable.
Multiple linear regression equation
Yc=a+bX+cX2+dX3.
The forecasting procedure using regression is similar to that of time series in that data are
first obtained and plotted to be sure the correct form of a model is chosen. The method of
converting the data into a forecasting equation is the same in that the normal equations are used
as in the case least square. The equations are always solved for the values of the slope b and
intercept a, they are often rewritten in the more convenient form:
_ _
∑XY -nXY
b=-----_ -------
∑X2-nX
_ _
a=Y - bX

Department of Mechanical Engineering, ATMECE 28

 Operations Management 18ME56

_ _
Where X = ∑(x)/n and Y = ∑(Y)/n
1. The general manager of a building material production plant feels the demand for plaster
board shipments may be related to the number of construction permits issued in the country
during the previous quarter. The manager has collected the data shown in the accompanying
table.
Construction Plaster
permits (X) board
shipments
(Y)
15 6
9 4
40 `6
20 6
25 `3
25 9
15 10
35 16

Find
a) Graph the data to see whether they can be satisfactorily described by linear equation.
b) Use the normal equations to derive a regression forecasting equation.
c) Confirm the values of (b) and (a)
d) Determine a pint estimate for plaster board shipments when the number of construction
permits is
30.
Solutio
n: a)
Graph

Department of Mechanical Engineering, ATMECE 29

 Operations Management 18ME56

A scatter diagram shows that the data are nor perfectly linear but approach
linearity over this short range. b)
Construction Plaster board XY X2 Y2
permits (X) shipments
(Y)
15 6 90 225 36
9 4 36 81 16
40 `6 640 1600 256
20 6 120 400 36
25 `3 325 625 169
25 9 225 625 81
15 10 150 225 100
35 16 560 225 256
∑184 ∑80 ∑2146 ∑5006 ∑950

N=8 pairs of observations

∑Y=na+b∑X 80 = 8a+184b (1)
∑XY = a∑X + b∑X 2
2146 = 184a+5006b (2)
Multiplying (1) by (-23):* -1840 =-184a – 4232b (3)
Adding (2) and (3) 360= 774b (4)

Therefore b= 306/774 =0.395

Substituting in(1) 80 = 8a + 184 (0.395)

8a=80 – 72.7
A= 7.3 / 8 = 0.91

Department of Mechanical Engineering, ATMECE 30

 Operations Management 18ME56

Equation is Y=0.91+ 0.395X

Where X = permits and
Y=shipments c) Alternately,
_
X = ∑(x)/n = 184/8=23
_
Y = ∑(Y)/n = 80/8 =10
_ _
∑XY –nXY 2,146-
8(23)(10) b=-----_ ------- = ---------
-------------- = 0.395
∑X2-nX 5006 – 8(23)(23)
_ _
a=Y – bX = 10-0.395 (23) =0.91

d) letting X=30,
Y =0.91 +0.395 (30)
= 12.76 = 30 shipments.

Simple linear Regression analysis

X=independent variable values X= values of x that lie on the trend line

y=dependent variable values Y=values of y that lie on the trend line

n=number of observations Y=a+bX

a=vertical axis intercept r=coefficient of correction

b=slope of the regression line r²= coefficient of determination

y =mean value of the dependent variable

∑x²∑y- Y=a+bX
∑x∑y a= ----- n∑xy - ∑x∑y r2 = ----------
---------- ----------------------
n∑x²-(∑x)²
√[n∑x²-(∑x) ²][n∑y²-(∑y)²]
n∑xy -∑x∑y
b= --------------- n∑x² - (∑x) ²

Department of Mechanical Engineering, ATMECE 31

 Operations Management 18ME56

Example 1. Simple Linear Regression Analyses: A Time Series

1. Aroma Drip Coffee Inc. Produces commercial coffee machine that are over the world. The
company’s production facility has operated at near capacity for over a year now. Wayne conners.the
plant manager thinks that sales growth will continue, and he wants to develop longrange forecasts
to help plan facility requirements for the next 3 years. Sales records for the past 10 years have been
complied:
year Annual sales year Annual sales
(Thousands of units) (thousands of units)
1 1000 6 2000
2 1300 7 2200
3 1800 8 2600
4 2000 9 2900
5 2000 10 3200
We study the formula and variable definition in table and then we construct the following table
to establish the values to us in the formula (It is helpful to use a spreadsheet such as Microsoft
Excel to perform many of the calculation.)

Year Annual Times x² xy

Sales periods (x)
(Thousands of
units)
1 1000 1 1 1000
2 1300 2 4 2600
3 1800 3 9 5400
4 2000 4 16 8000
5 2000 5 25 10000
6 2000 6 36 12000
7 2200 7 49 15400
8 2200 8 64 20800
9 2600 9 81 26100
10 3200 10 100 32000
Totals ∑y=21000 ∑x=55 ∑x²=385 ∑xy=133,300

Solution:
1. Let us now solve for the a and b values:

= 8,085000 – 7,331,500 = 753,500 =913.333

Department of Mechanical Engineering, ATMECE 32

 Operations Management 18ME56

,850 – 3,025 825

b= n∑xy -∑x∑y = (10) (133,300) – (55) (21,000)

n∑x² - (∑x) ² 825
= 1,333,000-1,155,000 = 178000 = 215.758
825 825
2. Now that we know the values of a and b,the regression equation can be used to
forecast future years’ sales:

Y=a+bX =913.333+215.758X

3. If we wish to forecast sales in thousands of units for the next three years, we would
substitute 11, 12,and 13, the next three values for x,into the regression equation for
X:

Y11 = 913.333 + 215.758(11) = 3,286.7 or 3,290 thousand units

Y12 = 913.333 + 215.758(12) = 3,502.4 or 3,500 thousand units
Y13 =913.333 + 215.758(13) = 3,718.2 or 3,720 thousand units

The forecasts are rounded to one significant digit more than the original data. Notice than the sales
data contain only two significant digits; the forecasts are carried to three.

2. Jack Weis, the general manager of precision Engineering Corporation, thinks that his firm’s
engineering services supplied to highway construction firms are directly related to the amount
of highway construction contracts let in his geographic area. He wonders if this is really so
and if it is can this information help him plan his operation better? Jack asked Maria Cortez,
one of his engineers, to perform a simple linear regression analysis on historical data. Maria
Cortez, one of the following: a. Develop a regression equation for predicting the level of
precession’s services. b. use the regression equation to predict the level of demand for the next
four quarters’. C.
Determine how closely demand is related to the amount of construction contracts released.

Solution:
a. Develop a regression equation :
1. Maria goes back through local, state, and federal records’ to gather the dollars amount
of contracts released in the Geographic’s area for two years by quarters.
2. She examines the demand for her firm’s services over the same period.
3. The following data are prepared:

Department of Mechanical Engineering, ATMECE 33

Operations Management 18ME56

Year Quarter Sales of precision Total Amount of

engineering services contracts released
(Thousands of units) (thousand of Rs)
1 Q1 8 150
Q2 10 170
Q3 15 190
Q4 9 170
2 Q1 12 180
Q2 13 190
Q3 12 200
Q4 16 220

4. Maria now develops the totals required to perform the regression analysis. The
formulas and n=variable definitions are found in table 5. (It is helpful to use a
spreadsheet to perform many of the calculations.)

Time periods Sales x² xy y²

(y) Contracts (x)
1 8 150 22500 1200 64
2 10 170 28900 1700 100
3 15 190 36100 2850 225
4 9 170 28900 1530 81
5 12 180 32400 2160 144
6 13 190 36100 2470 169
7 12 200 40000 2400 144
8 16 220 48400 3520 256
9 9 81 26100
Totals ∑y=95 ∑x=1470 ∑x²=273,300 ∑xy=17,830 ∑y²=1,183

5. Use these values in the formula in table 3.45 to compute a and b:

= 25,963,500 – 26,210,100 = 753,500 = -9.671

Department of Mechanical Engineering, ATMECE 34

 Operations Management 18ME56

2,186,400 – 2,160,900 25,500

b= n∑xy -∑x∑y = (8) (17,830) – (1470) (95)

n∑x² - (∑x) ² 25,500

= 142,640-139,650 = 2,990 = 0.1173

25,500 25,500

6. The regression equation is therefore Y= -9.671 +0.1173X.

b. Forecast the level of demand for the next four quarters:

1. Maria calls representatives of the contracting agencies and prepares

estimates of the quarterly contracts for the next four quarters in thousands of
dollars. These were 260,290,300 and 270.

2. Next, Maria forecasts the demand for precision’s engineering services

(in thousands of dollars)for the next four quarters by using the regression
equation Y—9.671+0.1173X:

Y1 =-9.671 +0.1173(260) Y2=-9.671 +0.1173(290)

= -9.671 + 30.498 =-

9.671 + 34.017 = 20.827

= 24.346

Y3 =-9.671 +0.1173(300) Y4 =-9.671 +0.1173(290)

= -9.671 + 35.190 =-

9.671 + 31.671 = 25.519

= 22.000

The total forecast (in thousands of dollars) for the next years is the total of the four quarter
forecasts:
20.827 +24.346 +25.5819 + 22.000 =$92.7
Notice that the forecast is rounded to one significant digit more than the original data.

Department of Mechanical Engineering, ATMECE 35

 Operations Management 18ME56

c. Evaluate how closely demand is related to the amount of the construction contracts
released:

r= n∑xy - ∑x∑y = 2900

[n∑x²-(∑x) ²][n∑y²-(∑y)²] [25,550][8(1,183) – (95)²

= 2,900 = 2,900 = 2,900

[25,550][9,464 -9,025] (25,500)(439) 11,194,500

= 2990 =.894

3,345.8 r²

= 0.799

The amount of contracts released explains approximately 80 %( r² = 0.799) of the observed

variation in quarterly demand for precision’s services.

3. The table below gives a sales record of a firm. Determine the regression line for the firm and
find the forecast of sales in the month of Jan for next year.
Month Sales(in units)/(Demand)
Jan 90
Feb 111
Mar 99
April 89
May 87
June 84
July 104
Aug 102
Sept 95
Oct 114
Nov 103
Dec 113

SOLUTION:

Regression equation is,

y=a+ bx

Department of Mechanical Engineering, ATMECE 36

 Operations Management 18ME56

Month x y x² xy
Jan 0 90 0 0
Feb 1 111 1 111
Mar 2 99 4 198
April 3 89 9 267
May 4 87 16 348
June 5 84 25 420
July 6 104 36 624
Aug 7 102 49 714
Sept 8 95 64 760
Oct 9 114 81 1026
Nov 10 103 100 1030
Dec 11 113 121 1243
∑x=66 ∑y=1191 ∑x²=506 ∑xy=6741

Where;
X = month of the year
Y= sales in the respective month
n= number of observations

Using above values, the constants a and b are calculated as follows

The equation of the regression line is

given by y=a+ bx
∴y=91.92+1.332x
From the regression line the estimated sales for next year ‘Jan’ is
Put x=12( since it is calculated for next year ’jan’)
y=91.92+1.332*12
=107.904=108
∴Forecast for next ‘Jan’ is 108 units.

Department of Mechanical Engineering, ATMECE 37

 Operations Management 18ME56

2nd method: the above problem can be worked in a simplified manner by taking the deviation
from the middle year, such that ‘∑x’ will be equal to zero, and the values of ‘a’ and ‘b’ can be
given by Where;
X = month of the year
Y= sales in the respective month
n= number of observations

Month x y x² xy
Jan -6 90 36 -540
Feb -5 111 25 -555
Mar -4 99 16 -396
April -3 89 9 -267
May -2 87 4 -174
June -1 84 1 -84
July +1 104 1 +104
Aug +2 102 4 +204
Sept +3 95 9 +285
Oct +4 114 16 +456
Nov +5 103 25 +515
Dec +6 113 36 +678
∑x=0 ∑y=1191 ∑x²=182 ∑xy=226

The required equation is y=a +bx

Put x=+7(since we are taking the deviation from the middle year, such that ∑x=0)
To find forecast for ‘Jan’

∴ Forecast for next ‘Jan’ is 108 units.

4. The sales of a product during the last five years is tabulated below
Year: 1973 1974 1975 1976 1977

Sales: 4 8 6 10 4

Department of Mechanical Engineering, ATMECE 38

 Operations Management 18ME56

Using linear forecaster, calculate

i. Sales in the years 1978 and 1979 ii. Standard
error of estimate and give its significance.
SOLUTION:
Year x y x² xy
1973 -2 4 4 -8
1974 -1 8 1 -8
1975 0 6 0 0
1976 +1 10 1 +10
1977 +2 4 4 +8
∑x=0 ∑y=32 ∑x²=10 ∑xy=2

The linear equation is y=a+bx

Equation of regression line is given by

y=6.4+0.2x
i. Estimated sales for,
Year 1978
Put x=3 in the regression equation, y=6.4+0.2*3=7
Year 1979
Put x=4 in the regression equation, y=6.4+0.2*4=7.2
Forecast values y1978=7; y1979=7.2

ii. To determine standard error estimate standard deviation

Where y=Demand values (sales) y'=calculated

values from regression equation
Year y y' (y'-y) (y'-y)²
1973 4 6 +2 4
1974 8 6.2 +1.8 3.24
1975 6 6.4 +0.4 0.16
1976 10 6.6 +3.4 11.56
1977 4 6.8 +2.8 7.84
26.80

∑ (y'-y)²=26.80
=2.315

Its significance: the standard error estimate simply shows that 95% of the data are expected
to fall within ±2 limits of the regression line. 2 =2*2.315=4.63

Department of Mechanical Engineering, ATMECE 39

 Operations Management 18ME56

5. the sales of machine tools in the last 8 years is lakhs of Rs are

5.0,4.5,10,9.0,11.0,18.5,17.5 and
22.0.Find the linear regression line and calculate the sales for the 10th year.

Solution:
Regression line is given by y=a+bx

Year y x x² xy
1 5.0 -7 49 -35.0
2 4.5 -5 25 -22.5
3 10.0 -3 9 -30.0
4 9.0 -1 1 -9.0
5 11.0 +1 1 +11.0
6 18.5 +3 9 +55.5
7 17.5 +5 25 +87.5
8 22.0 +7 49 +154
∑y=97.5 ∑x=0 ∑x²=168 ∑xy=211.5

Put x=11 in the equation y10=12.19+1.3*11=26.49

∴y10=26.49 lakhs of rupees.

6. A manufacturer of children’s cycle believes that the demand for the cycles is
correlated to the birth of babies in the area during the previous year. The
following data shows the relationship.
Compute the probable sales in the ninth year, given the number of births in the previous year as
1, 66,000
Year No. of births in the previous year Cycles sold during the year

1 40000 3000

2 48000 3200

3 66000 3700

Department of Mechanical Engineering, ATMECE 40

 Operations Management 18ME56

4 78000 4000

5 92000 5200

6 1,05,000 7900

7 1,25,000 9000

8 1,40,000 10000

SOLUTION: y=a+bx

Year y*1000 x*1000 xy x²

1 3 40 120 1600

2 3.2 48 153.6 2304

3 3.7 66 244.2 4356

4 4 78 312 6084

5 5.2 92 478.4 8464

6 7.9 105 829.5 11025

7 9 125 1125.0 15625

8 10 140 1400.0 19600

∑y=46 ∑x=694 ∑xy=4662.7 ∑x²=69078

Forecast for 9th year is given x=1,66,000 y9= [-0.82+0.0757*1,66,000] =11744

cycles.

7. A manufacturer of tyres believes that a relationship exists between the

automobiles sold in the year and the sales of the tyres two years later. The data
for the past 10 years are given below.

Department of Mechanical Engineering, ATMECE 41

 Operations Management 18ME56

Sales of 4.0 4.5 4.2 5.5 5.8 5.5 6.2 7.2 6.7 7.9
automobiles(in
lakhs)

Sales of tyres 8.0 7.9 8.1 8.4 8.1 8.6 9.1 8.9 9.1 9.6
2 years later(in
lakhs)

Establish a linear regression fit to forecast the sales of tyres on the basis of the sales of automobiles
2 years earlier. Also find what will be the sales of tyres given that the sale of automobiles 2 years
earlier was 6.1 lakhs.
Solution: Required equation is, y=a+bx

Given sales of automobiles 2 years earlier=6.1 lakhs

Forecast for tyres
Put x=6.1 in the regression equation
y=16.23+116*61=9.154 =9.2
Forecast sales of tyres=9.2 lakhs

Department of Mechanical Engineering, ATMECE 42

 Operations Management 18ME56

Exponential Smoothing: (Time Series Analysis)

Exponential smoothing models are well known and often used in operations
management.
Exponential smoothing is a type of moving-average forecasting technique which
weights past data in an exponential manner so that the most recent data carry
more weights in the moving average.
Simple exponential smoothing makes no explicit adjustment for trend effect.
Whereas adjusted exponential smoothing does take trend effects into account.
Simple exponential smoothing:
The forecast is made up of the last-period forecast plus a portion of the difference between the
lastperiod actual demand and the last-period forecast.
Ft = Ft-1 + α (Dt-1 – Ft-1)
Where
Ft =current –period
forecast Ft-1=last
period forecast α =
smoothing constant
Dt-1 = last period
demand
If demand was above the last period forecast, the correction will be positive and if
demand was below, the correction will be negative.
The smoothing constant, α actually dictates how much correction will be made.
It is number between 0 and 1 used to compute the forecast Ft.
The value of α is often kept in the range of 0.005 to 0.30 in order to “Smooth” the
forecast.
The exact value depends upon the response to demand that is best for the individual firm.

Problem 1. A firm uses simple exponential smoothing with α =0.1 to forecast demand.
The forecast for the week of February 1 was 500 units, whereas actual demand turned out
to be 45o units.
a) Forecast the demand for the week of February 8
b) Assume that the actual demand during the week of February 8 turned out to be
505 units. Forecast the demand for the week of February 15.Continue on
forecasting through March 15, assuming that subsequent demands were actually
516, 488, 467, 554 and10 units

Solution;
(a) Ft = Ft-1 =α (Dt-1 – Ft-1)
= 500=0.1(450-500) = 495 units
Department of Mechanical Engineering, ATMECE 43
 Operations Management 18ME56

(b) Arranging the procedure in tabular form, we have;

Week Demand Old-forecast Forecast Correction α New forecast(Ft)

Dt-1 Ft-1 error Dt-1 – (Dt-1 – Ft-1) Ft = Ft-1 =α (Dt-1 – Ft-
Ft-1 1)

Feb.1 450 500 -50 -5 495

8 505 495 10 1 496

15 516 496 20 2 498
22 488 498 -10 -1 497
Mar.1 467 497 -30 -3 494
8 554 494 60 6 500
15 510 500 10 1 501

• In this example, an initial forecast value was available.

• If no previous value is known, the old forecast starting point may be
estimated or taken to be an average of the values of some preceding periods.
Smoothing coefficient selection:
Smoothing coefficient: A numerical parameter that determines the weighting of old
demands in exponential smoothing.
To begin forecasting, some reasonable estimate for an old beginning forecast is necessary.
Likewise, a smoothing coefficient, α, must be selected.
A high smoothing coefficient could be more appropriate for new products or items for
which the underlying demand is shifting about (dynamic or unstable).
A of 0.7, 0.8, 0.9 might be best for these conditions, i.e. if unstable conditions are known to
exist.
If demand is very stable and believed to be representative of the future, the forecaster
wants to select a low α value to smooth out any sudden noise that might have occurred.
Under the stable conditions, an appropriate value may be 0.1, 0.2, and 0.3. When
demand is slightly unstable smoothing coefficients of 0.4, 0.5 or 0.6 might provide the most
accurate forecasts.
Selecting forecasting parameters and comparing models:
The procedure for selecting forecasting parameters is given in the first four steps that follow; the
fifth step is used for comparing and selecting models.
1. Partition the avialble data into two subjects, one for fitting parameters (the test
set) and the other for forecasting.
2. Select an error measure to evaluate forecast accuracy of the parameters to be
tried. MAD and /or bias are useful error measures.
3. Select a range of α values. Using one of the α values apply the forecasting
model to the test set of data, recording the resulting ofrecast errors. Then,
selecting a new values in the selected range have been tested.

Department of Mechanical Engineering, ATMECE 44

 Operations Management 18ME56

4. Select the α value that resulted in the lowest forecast error when applied to the
test set. Your model is now fitted to the demand data.
5. Forecast using the balance of the data with the exponential (or moving average)
model that you have fitted to the test set. Use the results to compare alternative
models that have previously been fitted to representative demand data.
Forecast Error:
When we evaluate different forecasting methods, it is necessary to measure the
effectiveness. Forecast error is the numeric difference of forecasted demand and actual demand.
Mean Absolute Deviation (MAD):
A forecast error measure that is the average forecast error without regard to direction;
calculated as the sum of the absolute value of forecast error for all periods divided by the total
number of periods evaluated.
Sum of the absolute value of forecast error for all periods
MAD = -----------------------------------------------------------------------
number of periods

∑ |forecast error|
= -----------------------
n

∑|forecast demand – actual demand|

= --------------------------------------------
----
n

Where n is the number of periods

There is a relationship between mean absolute deviation and the classical measure of
dispersion for the forecast error, the standard deviation (σe). If the forecast is working properly,
forecast errors are normally distributed. When this is so, the smoothed mean absolute deviation
(SMAD) is used to estimate the standard deviation.

σe ≈ 1.25SMAD
Smoothed MAD as an average MAD over time.

Bias : a forecast error measure that is the average of forecast error with regard to direction and
shows any tendency consistently to over-or under forecast; calculated as the sum of the actual
forecast error for all periods divided by the total number of periods evaluated.

Sum of forecast error for all

periods Bias =----------------------------------
-------------
number of periods

Department of Mechanical Engineering, ATMECE 45

 Operations Management 18ME56

∑ forecasted error
= --------------------------
n

∑(forecasted demand - actual demand)

= -------------------------------------
----
n
the α value may be calculated by an approximate equivalent to an arithmetic moving average,
in terms of the degree of smoothing can be estimated by
2
α= -------
n+1

How to monitor and control a forecasting model.

Forecasts can be monitored and controlled by setting upper and lower limits on how much the
performance characteristics of a model can deteriorate before we change the parameters of the
model. One common way that we can track the performance of forecasting models is to use
what is called a tracking signal:
Algebraic sum of errors over n periods
Tracking signal =----------------------------------------------------
Mean absolute deviation over n periods

∑(Actual Demand – forecast demand)

=--------------------------------------------------
MAD
∑ (Actual demand – Forecast demand)

=-------------------------------------------------
∑|Actual demand – Forecast demand|
----------------------------------------------
--
n
The tracking signal measures the cumulative forecast error over n period in terms of MAD.

Problem 1. Forecast for 9th year using the following data by exponential technique.

Year 1 2 3 4 5 6 7 8

Demand(Rs 90 100 107 113 123 136 144 155

in lakhs)

Smoothing constant =0.5 and initial forecast F=85.

Department of Mechanical Engineering, ATMECE 46

 Operations Management 18ME56

SOLUTION:

Year Demand Forecast New forecast Ft= Ft-

1+ (Dt-1- Ft-1)
Dt-1 Ft-1

1 90 85 85.75

2 100 85.75 87.89

3 107 87.89 90.76

4 113 90.76 94.1

5 123 94.1 98.44

6 136 98.44 104.07

7 144 104.07 110.06

8 155 110.06 116.8

Where;

Ft = current period forecast

Ft-1 = last period forecast Dt-1 = last period

demand α = smoothing constant
The forecast value 116.8 made in the 8th year, is the forecast for the next year (i.e., 9th year)

∴F9=116.8 Rs. In lakhs.

Problem 2. monthly sales of a product in thousands of rupees for the past 2 years are shown
below:

Month Jan Feb Mar Apr May June July Aug Sep Oct Nov Dec

2
years 253 236 245 246 260 251 249 242 234 244 246 251
ago

1 year 250 252 248 241 247 244 244 249 511 238 249 252
ago

a) Fit a line to the data and determine a forecast of the next month.
b) Select an initial forecast from part ‘a’ and use =0.2 to determine the forecast for the
next
Jan by exponential smoothing.
c) Compare the forecast from (a) and (b), which one would you select.
Department of Mechanical Engineering, ATMECE 47
 Operations Management 18ME56

SOLUTION:
a) Required equation y=a+bx
Month Average x xy x² y' Error
E=(y'-y)
‘y’

Jan 251.5 -11 -2766.5 121 245.8 -5.7

Feb 244.0 -9 -2196.0 81 247.9 +3.9

Mar 246.5 -7 -1725.5 49 250.1 +3.6

Apr 243.5 -5 -1217.5 25 252.2 +8.7

May 253.5 -3 -760.5 9 254.5 +0.9

June 247.5 -1 -247.5 1 256.5 +9.0

July 246.5 +1 +246.5 1 258.7 +12.2

Aug 245.5 +3 +736.5 9 260.8 +15.3

Sep 372.5 +5 +1862.5 25 262.9 -109.6

Oct 241.0 +7 +1687.0 49 265.1 +24.1

Nov 247.5 +9 +2227.5 81 267.3 +19.8

dec 251.5 +11 +2766.5 121 269.4 +17.9

∑y=3091 ∑x=0 ∑xy=613 ∑x²=572 ∑|(y'-

y)|=230.7

∴y=257.6+1.072x
Forecast for the next month (i.e.,jan)
Put x=13 in the above equation
∴y=257.6+1.072*13=271.536
∴yjan=271.5
The forecast values for all the months are calculated using regression equation and tabulated.
The mean absolute deviation is calculated by taking the average value of absolute error.

Department of Mechanical Engineering, ATMECE 48

 Operations Management 18ME56

=19.23

MAD=19.23

b) Initial Forecast = 271.5

Smoothing constant = α = 0.2
Month Demand Dt-1 Forecast Ft-1 New Forecast Error

Ft= Ft-1+ (Dt-1- E = (Ft-1- Dt-1 )

Ft-1)
Jan 251.5 271.5 260.3 +56.0

Feb 244.0 260.3 257.04 +16.3

Mar 246.5 257.04 254.9 +10.54

Apr 243.5 254.9 252.6 +11.4

May 253.5 252.6 252.8 -0.9

June 247.5 252.8 251.7 +5.3

July 246.5 251.7 250.7 +5.2

Aug 245.5 250.7 249.7 +5.2

Sep 372.5 249.7 274.3 -122.8

Oct 241.0 274.3 267.6 +33.3

Nov 247.5 267.6 263.6 +20.1

Dec 251.5 263.6 261.2 +12.1

∑|(Ft-1- Dt-1 )| =
299.14

MAD = 24.92

c) The deviation (MAD = 19.23) in the I method is less. Hence the least square method is
preferred.

Department of Mechanical Engineering, ATMECE 49

 Operations Management 18ME56

Problem 3. The pesticide manufacture has experienced the following monthly demand for an
environmentally sound pesticide.
Month Actual demand
(in tones)
Feb 62
Mar 84
April 77
May 95
June 100
Using first order exponential smoothing technique forecast the demand for the month of July.
Choosing α = 0.3, compare graphically the forecast with the actual demand from the months
march through June, assuming that the forecast for the February was 60.
Solution :
Month Actual demand Initial forecast New forecast (Ft)

Feb 62 60 60.6

Mar 84 60.6 67.62

April 77 67.62 70.43

May 95 70.43 77.8

June 100 77.8 84.5

Specimen Calculation
F1= Ft-1+ (Dt-1- Ft-1)
=60+0.3(62-60)
=60+0.6
=60.6 = Forecast for the month of July = FJuly =84.5

Department of Mechanical Engineering, ATMECE 50

 Operations Management 18ME56

Problem 4: Number of daily calls for repair has been recorded as follows:
Day 1 2 3 4 5 6

Calls 132 170 95 110 120 135

Prepare exponentially smoothed forecasts for α =0.1 and F1 =130. Compute the error’s of
Bias and absolute Deviation. Forecast for 7th day.
Solution:
Given α =0.1, F1 =130
Day Calls Initial forecast New forecast Error = (Ft-1 - Dt-1)
Dt-1 Ft-1 Ft
1 132 130 130.2 -2
2 170 130.2 134.18 -39.8
3 95 134.18 130.26 39.18
4 110 130.26 128.23 20.26
5 120 128.23 127.41 8.23
6 135 127.41 128.17 -7.59
7 128.17
New forecast of 6 day is the forecast for 7th day.
th

The forecast value for 7th day =129 calls

(i) Mean Absolute Deviation
∑|Error| 117.06
=------------- = ------------ = 19.51
n 6

∑ Error 18.28
(ii) Bias = ---------- = ----------- = 3.05
n 6

Problem 5: sales of plywood in rupees of a particular size have been tabulated below.
Year 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989

Sales 15 16 12 22 16 21 30 12 31 40
in Rs.
x 105

(i) What is the expected sale in 1990 by method of least squares.

(ii) Select an initial forecast from (i) and use α = 0.1 to determine the forecast for 1990
by exponential smoothing.
(iii)Compare the forecast (i) and (ii) which one would you select.

Department of Mechanical Engineering, ATMECE 51

 Operations Management 18ME56

Solution:
(i) By method of least squares

Year Sales ‘y’ in ‘X’ XY X2 Forecast from regression Error (Y1-Y)

Rs.x 105 equations ‘y’

1980 15 -9 -135 81 11.69 -3.31

1981 16 -7 -112 49 13.87 -2.13

1982 12 -5 -60 25 16.05 4.05

1983 22 -3 -66 09 18.23 -3.77

1984 16 -1 -16 01 20.41 -4.41

1985 21 1 21 01 25.59 1.59

1986 30 3 90 09 24.77 -5.23

1987 12 5 60 25 26.95 14.95

1988 31 7 217 49 29.13 -1.87

1989 40 9 360 81 31.31 -8.69

∑ y=215 ∑ X=0 ∑ ∑ X2 ∑|Y1-Y|=50

XY=359 =330

Y=a+bx
a= ∑ y/n = 215 / 10 = 21.5
b = ∑ XY / ∑ X2 = 359/330 = 1.09

Y=a+bx
= 215+1.09X
Forecast for the year 1990 Put – X=10 Y =215 + 1.09 x10 =32.4
Y1990 = 32.4 rs. x105 y1990 =32.4 x 105

∑| Error| 50
MAD = ------------ = -------- = 5
n 10
(ii) Exponential smoothing
Technique α = 0.1
Initial forecast = 32.4

Department of Mechanical Engineering, ATMECE 52

 Operations Management 18ME56

Month Sales Initial Forecast Ft-1 New Forecast Error

Dt-1 Ft= Ft-1+ (Dt-1- Ft- E = (Ft-1- Dt-1 )
1)
1980 15 32.4 30.66 17.4

1981 16 30.66 29.2 14.66

1982 12 29.2 27.48 17.2

1983 22 27.48 26.93 5.48

1984 16 26.93 25.84 10.93

1985 21 25.84 25.36 4.87

1986 30 25.36 25.82 -4.64

1987 12 25.82 24.44 13.82

1988 31 24.44 25.09 -6.56

1989 40 25.09 26.58 -14.91

∑|Error|= 110.44

Forecast for the year 1980 is 26.58 x105

∑|Error | 110.44
MAD = ------------ = ------------- = 11.044
n 10
(iii) The Absolute deviation is
less in case of method of
least squares. Hence, it can
be preferred.

Department of Mechanical Engineering, ATMECE 53

 Operations Management 18ME56

Seasonal indexes:
A seasonal index (SI) is a ratio that relates a recurring seasonal variation to the corresponding
trend value at that given time.
Several methods of computing are available, but the most widely used is a ratio-to-
movingaverage method.
The procedure is to tabulate the data in monthly terms and compute 12-month moving-average
values over a period of several years.
Seasonalized forecast= seasonal index(trend forecast)
Ysz =(SI)Yc

Problem 1. The production manager of a natural gas pipeline company has projected trend
values for next august, September, and October of 2.1, 2.2 and 2.3 million cubic meters,
respectively. Seasonal indexes for the three months have been found to be, 0.80, 1.05, and
1.20, respectively. What actual seasonalized (adjusted) production should the manager plan
for?
Solution:
Ysz =SI (Yc)
For August: = (0.80) (2.1) = 1.68 million cubic meters
For September: = (1.05) (2.2) = 2.31 million cubic meters
For October: = (1.20) (2.3) = 2.76 million cubic meters
• After seasonal adjustments have been made, similar adjustments can be made for cyclical or
irregular effects if data are available.

2. Wayne Conners, the plant manager of Aroma drip coffee Inc., is trying to plan cash,
personnel and materials and supplies requirements for each quarter of next year. The quarterly
sales data for the past three years seem to reflect fairly the seasonal output pattern that should
be expected in the future. If Wayne could estimate quarterly sales for next year, the cash,
personnel, and materials and supplies needs could be determined.

Solution:
1. We compute the seasonal indexes.
Year Quarterly Sales (thousands of Units) Annual total
Q1 Q2 Q3 Q4
8 520 730 820 530 2600
9 590 810 900 600 2900
10 650 900 1000 650 3200
Totals 1760 2440 2720 1780 8700
Quarter 586 2/3 813 2/3 906 2/3 593 1/3 725*
average
Seasonal 0.809 1.122 1.251 0.818
**
index (S.I)
* Overall quarter average= 8700/12
=725

Department of Mechanical Engineering, ATMECE 54

 Operations Management 18ME56

** S.I = quarter average / overall quarter

average.
2. Next, we deseasonalize the data by dividing each quarterly value by its S.I
(seasonal index) for instance, 520 ÷0.809=642.8, 730÷1.122=650.6 and so on

Deseasonalized Adjusted quarterly Data

Year Q1 Q2 Q3 Q4

8 642.8 650.6 655.5 647.9

9 729.3 721.9 719.4 733.5
10 803.5 802.1 799.4 794.6

3. Now, we perform a regression analysis on the deseasonalized data (12 quarters)

and forecast for the next 4 quarters:
Time Period X Y Y2 X2 XY
Year 8, Q1 1 642.8 413,191.84 1 642.8
Year 8, Q2 2 650.6 423,280.36 4 1301.2
Year 8, Q3 3 655.5 429,680.25 9 1966.5
Year 8, Q4 4 647.9 419,774.41 16 2591.6
Year 9, Q1 5 729.3 531,878.49 25 3646.5
Year 9, Q2 6 721.9 521,139.61 36 4331.4
Year 9, Q3 7 719.4 517,536.36 49 5035.8
Year 9, Q4 8 733.5 538,022.25 64 5868.0
Year 10, Q1 9 803.5 645,612.25 81 7231.5
Year 10, Q2 10 802.1 643,364.41 100 8021.0
Year 10, Q3 11 799.4 639,040.36 121 8793.4
Year 10, Q4 12 794.6 631,389.16 144 9535.2
totals ∑x=78 ∑y=8700.5 ∑y =6,353,909.75 ∑x =650
2 2
∑xy=58,964.9
4. Now, to find the value of a, b and Y

= 650(8700.5) – 78(58,964.9) / 12(650) – (78)2

= 12(58,964.9) - 78(8700.5) / 12(650) – (78)2

Y = a + b x = 615.421 +16.865X
5. Now, we substitute the values 13,14, and 16 – the next four values for x – into
the regression equation. These are deseasonalized forecasts, in thousands of
units, for the next four quarters.
Y13 =615.421+16.865(13)=834.666 Y14 =615.421+16.865(14)=851.531
Y15 = 615.421+16.865(15) =868.531 Y16 =615.421+16.865(16)=885.261

Department of Mechanical Engineering, ATMECE 55

 Operations Management 18ME56

6. Now, we use the seasonal indexes (SI) to seasonalize the forecasts:

Quarter S.I. Deseasonalized Seasonalized forecasts (S.I.X
forecasts Deseasonalized forecasts)
(thousands of Units)
Q1 0.809 834.666 675
Q2 1.122 851.531 955
Q3 1.251 868.396 1086
Q4 0.818 885.261 724

Life cycle effects upon forecasting methodology:

Decline

Maturity

Growth

Introduction

Time
Introduction
Data No data available : rely on qualitative method
time Need long horizon
methods Judgment, Delphi and historical analogy were useful, market surveys important
Growth
Data Some data available for analysis
time Still need long horizon; trends and cause-effect relationships important
methods Market surveys and historical comparison still useful. Regression and computer
simulation models justified. Tracking product history now important
Maturity
Data Considerable data available on demand, inventory levels et.
time More uses of short-term forecasts; still need long-term projections’, but, trends
change only gradually.
methods Statistical and quantitative methods more useful. Time series help for trend,
seasonal. Regression and correlation use associations and leading indicators.
Exponential smoothing very useful. Econometric methods feasible.
Decline
Data Abundant data (but not necessarily on decline).
time Shorter horizon.
methods Continue use of maturity methods as applicable. Judgment, historical analogies,
and market surveys may signal changes.
How to select a forecasting method:
Several factors should be considered in the selection of a forecasting method

Department of Mechanical Engineering, ATMECE 56

 Operations Management 18ME56

Cost
Accuracy
Data available
Time span
Nature of product and services
Impulse response and noise dampening
Some reasons for ineffective forecasting:
1. Failure of the organizations to involve a broad cross section of people in
forecasting. Individual effort is important, but the need to involve everyone who
has pertinent information and who will need to implement the forecast is also
important.
2. Failure to recognize that forecasting is integral to business planning
3. Failure to recognize that forecasting will always be wrong. Estimates of future
demand are bound to be subject to error, and the magnitude of error tends to be
greater for forecasts that cover very long spans of time. When operations
mangers have unrealistic expectations of forecasts , the fact that the forecasts
were not on the nose is often used as an excuse for poor performance in
operations
4. Failure to forecast the right things. Organizations may forecast the demand for
raw materials that go into finished products. The demands for raw materials need
not be forecast because theses’ demands can be computed form the forecasts for
the finished products. Forecasting too many things can be overload the
forecasting system and cause it to be too expensive and time consuming.
5. Failure to select an appropriate forecasting method.
6. Failure to track the performance of the forecasting models so that the forecast
accuracy can be improved. The forecasting models can be modified as needed
to control the performance of the forecasts.
Sources of forecasting data
♣ Auto sales
♣ Consumer confidence index
♣ Consumer price index
♣ Durable goods
♣ Employment
♣ Factory orders
♣ Gross domestic product
♣ Housing starts
♣ Index of leading economic indicators
♣ Industrial production
♣ Merchandise trade
♣ Personal income and consumption
♣ Producer price index
♣ Purchasing price indies
♣ Retail sales

Department of Mechanical Engineering, ATMECE 57