University of Sunderland

BA (Honours) Accounting and Financial Management

APC310

Business Modelling for

Decision Making
Version 1.0

Flexible Learning
Helping you bring Learning to Life
Published by
The University of Sunderland

The publisher endeavours to ensure that all its materials are free
from bias or discrimination on grounds of religious or political
belief, gender, race or physical ability. These course materials are
produced from paper derived from sustainable forests where the
replacement rate exceeds consumption.

The copying, storage in any retrieval system, transmission,

reproduction in any form or resale of the course materials or any
part thereof without the prior written permission of the University
of Sunderland is an infringement of copyright and will result in
legal proceedings.

© University of Sunderland 2008

Every effort has been made to trace all copyright owners of

material used in this module but if any have been inadvertently
overlooked, the University of Sunderland will be please to make
the necessary arrangement at the first opportunity.

These materials have been produced by the University of

Sunderland Business School in conjunction with Resource
Development International.
Business Modelling for Decision Making

Contents

How to use this workbook

Unit 1
Decision Tree Analysis
Introduction 1
Basic probability 1
Decision trees 5
Summary 24
Appendix 25

Unit 2
Discrete Simulation
Introduction 29
The Simulation Process 29
Summary 63
Appendix – Pivot Tables 64

Unit 3
Linear Programming
Introduction 65
Worked Example 72
Summary 86

Unit 4
Optimisation
Introduction 89
Linear Equations 89
Non Linear Equations 91
Summary 130

University of
Sunderland
Unit 5
Basic Forecasting Techniques
Introduction 131
Summary 155

Unit 6
Inventory Control
Introduction 157
Types of Stock 157
Just-in-time (JIT) 181

Unit 7
Sampling Theory and Statistical Inference
Introduction 193
Hypothesis Testing 193
Two Tailed Tests 201
Hypothesis Tests of the Population Proportion (p) 207
Confidence Intervals 214
Chi Squared Test 225
Summary 232

Unit 8
Statistical Process Control and Financial Mathematics
Introduction 233
Deming’s 14 Points 233
Process Capability 241
Financial Maths 246
Summary 270

University of
Sunderland
How to use this workbook
This workbook has been designed to provide you with the course
material necessary to complete Business Modelling for Decision
Making by distance learning. At various stages throughout the module
you will encounter icons as outlined below which indicate what you are
required to do to help you learn.

This Activity icon refers to an activity where you are required to undertake a
specific task. These could include reading, questioning, writing, research,
analysing, evaluating, etc.

This Activity Feedback icon is used to provide you with the information
required to confirm and reinforce the learning outcomes of the activity.

This icon shows where the Virtual Campus could be useful as a medium for
discussion on the relevant topic.

It is important that you utilise these icons as together they will provide
you with the underpinning knowledge required to understand
concepts and theories and apply them to the business and management
environment. Try to use your own background knowledge when
completing the activities and draw the best ideas and solutions you can
from your work experience. If possible, discuss your ideas with other
students or your colleagues; this will make learning much more
stimulating. Remember, if in doubt, or you need answers to any
questions about this workbook or how to study, ask your tutor.

University of
Sunderland i
Unit 1

Decision Tree Analysis

Introduction
Managing any business involves making decisions on a wide range of
matters on a daily basis. Some of these decisions may be trivial or they
may require little more than simple common sense to make. The
solutions may be obvious. However, others will involve large sums of
money and have a considerable risk element attached to them.

There are, however, techniques which can be used to help reduce

complex problems to simpler, more manageable issues and these are
introduced within this unit.

We as individuals, make decisions based upon the current information

that is available to us. It is, therefore, essential to base decision-making
on current and valid information. Certain assumptions must be made
about the likelihood of events occurring, and these are based on
available information.

Random events occur due to the uncertainty or randomness of the

environmental factors which may affect our decisions. The task of the
decision-maker is, therefore, to try to make predictions of these factors
and assess their effects, usually in terms of cost to the organisation.

Basic probability
The first step we need to make so that we can develop our
decision-making techniques is to understand the concept of probability.
This is a relatively simple concept which involves only straightforward
arithmetic. It is, therefore, within the scope of any student, and not be a
cause for concern!

We can look at any event or occurrence and make an estimate about

how likely it is to happen. This might be whether a coin will land
showing heads, whether a dice will show a six, or whether a chosen
playing card is the ace of hearts. These are all events which have a
definite, objective likelihood of occurring. Conversely, there are events
such as rain falling tomorrow, someone passing their driving test or
Chelsea Football Club beating Manchester United, which can only be
predicted in a subjective way. We need to use our own judgment based

University of
Sunderland 1
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

upon any supporting data. We might, for example, say with confidence
that, as the last three meetings between the football teams have resulted
in a win for them, then Chelsea are favourites to win.

The probability of an event occurring could simply be stated as unlikely,

possible, probable or highly likely. However, this does not allow much
flexibility and is inadequate. A better method would be to allocate a
number from a percentage scale from 0 to 100, 0 indicating that
something will never happen and 100 that it will always happen. Thus,
the probability of the author growing wings and flying is 0%, and the
probability of the sun setting at night is 100%. These are extremes of
little interest to us, but it does indicate how few things have 0% or 100%
probability. The vast majority of events will have probabilities between
these two limits.

ACTIVITY
Draw up a list of about ten events in your daily life. Determine the probability
of each one occurring;

a) using words

b) using a percentage

ACTIVITY FEEDBACK
Your chosen events will have probabilities that are determined by your
lifestyle. For some, the probability of watching the news on TV may be high
(say, 80%), for others it may very low (perhaps, 15%).

However, this exercise will have enabled you to allocate some form of
numerical values to the words used.

University of
2 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

ACTIVITY
Looking back at our responses, now try to evaluate numerically each of your
chosen descriptive words. Using the following table, insert numerical values
for each of the terms provided.

Descriptive term Probability percentage

Highly unlikely

Unlikely

Possible

Probable

Highly likely

Almost certain

KEY POINT
In mathematics, we prefer to use a decimal value between 0 and 1 instead of a
percentage. Hence, a probability of 85% would be written as 0.85, 50% would
be 0.50 and 7% would be 0.07. We will use this convention from now on.

So far we have considered subjective probabilities. Now we need to look

at events which have definite probabilities that we are able to calculate.
These include events where there are a defined number of outcomes.

Consider a deck of 52 cards. There are four suits; i.e. clubs, hearts,
spades and diamonds, (2 suits are red and two suits are black). The
cards per suit are numbered from 1 to 10, plus Jack, Queen and King.

Selecting a specified card at random from a deck of cards can be

calculated from the basic probability formula:

Probability (p) = number of favourable outcomes/total number of

outcomes

University of
Sunderland 3
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

In our playing card example, if we want to determine the probability of

selecting any one of the four aces, the formula would become (as there
are four ace cards, and there are 52 cards in a pack)

p = 4/52

This gives a probability of approximately 0.08 – perhaps, fairly unlikely.

Similarly, selecting a heart would give us a probability of

p = 13/52 = 0.25

ACTIVITY
Using the basic probability formula, calculate the following probabilities:

a) Selecting a red card.

b) Selecting an even numbered diamond.

c) Throwing a dice and scoring a six.

d) Throwing a dice and scoring an even number.

e) Tossing a coin to get a head.

ACTIVITY FEEDBACK
a) 26/52 or 0.50

b) 5/52 or 0.10

c) 1/6 or 0.17

d) 3/6 or 0.50

e) 1/2 or 0.50

University of
4 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

In business situations, we are likely to be faced with probabilities that

are somewhere in between the extremes of the subjective-objective
spectrum. If information is available to support a probability value then
it is more likely to be accurate. Therefore, much effort will go into
obtaining as much information about an event so that any calculations
carried out can be as reliable as possible. After all, decisions affecting the
future of an organisation and involving large sums of money may be
made based upon calculations involving these figures.

We might consider that the probability of a new contract being

successful is 0.85, or that entering a new market might result in high
sales has a probability of 0.44. Appropriate research would be able to
provide these figures.

We now need to turn to combined probabilities. This might be useful

when we want to assess the probabilities of one event followed by
another. We calculate this simply by multiplying together the relevant
probabilities. For example, if we want to calculate the combined
probabilities of throwing an even number with a dice (0.5) and tossing a
head with a coin (0.5), we would carry out the following calculation:

p = 0.50 x 0.50 = 0.25

Similarly, we can calculate the probability of our organisation having a

successful new contract and achieving high sales (from our earlier
figures) in the following way:

p = 0.85 x 0.44 = 0.37

This is an important idea upon which much of the rest of this unit is
based. The method that we will use for analysing decisions in the next
sections involves the combination of multiple probabilities.

Decision trees
A manager contemplating a problem often sees a range of possible
decisions spread over a long period of time and not just one immediate
decision. Such a situation is particularly suitable for structuring in terms
of a diagrammatic approach to its solution. Such a diagram is known as
a decision tree. For example, although the initial decision facing an oil
executive may involve the location of a new refinery, later decisions
might involve whether or not to increase its capacity, or perhaps build
another one. By generalising our notions about the nature of decisions
problems we can cope with this situation, and the use of decision trees
becomes meaningful and relevant.

A decision tree enables us to break down a big decision problem into a

series of smaller problems which can be solved separately and then
combined to provide a solution to the larger problem. The device is

University of
Sunderland 5
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

especially useful where a more complex situation can be broken down

into a sequence of simpler problems which follow one another in some
natural order. Many decision problems are sequential in character. For
example, an engineer designing and building a plant would probably
not have to make one overall decision and then adhere to it completely.
He could perhaps make a few initial decisions, leaving others until later
in the design and construction process when he has more information
available. The entire plant construction operation would then consist of
a sequence of interconnected decision problems.

With decision tree analysis, the models revolve around decision nodes,
random nodes and terminal nodes. Decision nodes occur when we
make a choice; random nodes occur when there exists and element of
risk between two or more alternatives. Terminal nodes occur when we
exhaust all of the branches or pathways resulting from the decision
nodes and random nodes.

In such a tree the branches split into other branches: these points where
a split occurs are known as nodes. At the left-hand side of the diagram
the initial node gives rise to a set of branches representing the action
options or decisions that are the decision-maker’s choice. Consequently,
this type of node is known as the decision node and commonly
represented by a rectangular or, more traditionally, as a square box.
Following each of these branches leads to a second node from which
further branches grow. The decision-maker may have little or no control
over which further branch is followed, and so the situation may now be
different. Hence, this node is known as a random node, represented by a
circle. A typical decision tree, therefore, consists of a series of branches
stemming from nodes of two types. Each branch is labelled
appropriately with either an action or an outcome probability.

A decision node is represented by a rectangular or, more traditionally,

as a square box. These nodes represent a choice between expected
values (EVs).

A random node is represented by a circle. These nodes represent an

(EV)

A terminal node is represented by a vertical line and are inserted at the

end of the path of branches of the tree. Sometimes we may omit drawing
these! Terminal nodes traditionally represent points where there are
payoffs.

University of
6 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

These combine to form the structural elements of a decision tree. A

simple version of which is shown in Figure 1.1.

Alternative course of action Outcomes

Film good

Go to the
Film average
cinema

Film poor

Watch
TV

Programme average

A decision node Random nodes

Figure 1.1 Elements of a decision tree.

I have a choice of going to the cinema or staying indoors and watch

television. If I go to the cinema, the film might be good, it might be poor
or it might be average. Similarly, if I stay indoors, the TV programme
might be good, it might be poor or it might be average (I have only
included the average branch, but assume the ‘good’ and ‘poor’ branches
are there also!)

Payoff
For every action chosen within a decision tree, there will be a result or
consequence. This is called the payoff and it is this that we are
predominately interested in. The purpose of the process of creating and
analysing a decision tree is to produce an optimum payoff. Payoffs can
be positive values (e.g. revenues coming in to a company) or negative
values (e.g. costs to the company).

University of
Sunderland 7
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

KEY POINT
The payoff is the consequence of an outcome (or series of outcomes)
measured in terms of the objective.

When analysing a decision tree, we want to measure the payoff of each

possible outcome in terms of the decision-maker’s objectives. In most
cases this will be in terms of money, but sometimes we will be more
interested in measures such as time, investment return, market share,
etc.

Let us return to my choice of my evening’s entertainment. We will now

look at the outcomes of the original choice in terms of my enjoyment.

We can now add some more information to the diagram in the form of
the cost of the various options. This is shown in Figure 1.2.

Alternative course of action Outcomes Cost

Film good £5.50

Go to the
Film average £5.50
cinema

Film poor £5.50

Watch
TV

Programme average

Figure 1.2. Cost of options.

If I enjoyed the film, then I am likely to feel happy and that the cost of the
evening has been justified. If I thought the film was poor then I will
probably feel bored and discontented. These feelings are the payoffs of
the decisions made. This can be included in the diagram and is shown in
Figure 1.3.

University of
8 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

Alternative course of action Outcomes Cost Payoff

Film good £5.50 Happy

Go to the
Film average £5.50 Indifferent
cinema

Film poor £5.50 Bored

Watch
TV

Programme average Indifferent

Figure 1.3. Decision payoffs.

Expected Values
Once we have identified all the possible outcomes for a situation, it
would be helpful if we could in some way place a numerical value on
them. We can do this by using the concept of Expected Value (EV). This
can best be understood through the use of a simple example.

Consider an unbiased die. The probability of throwing any value is 1/6.

On a large number of throws, we would expect an equal number of 1s,
2s, 3s, 4s, 5s and 6s

However, what would be the Expected Value?

The expected value for an unbiased die is determined as follows:

Value(V) Probability(P) VxP

1 1/6 1*1/6 = 1/6

2 1/6 2*1/6 = 2/6

3 1/6 3*1/6 = 3/6

4 1/6 4*1/6 = 4/6

5 1/6 5*1/6 = 5/6

6 1/6 6*1/6 = 6/6

The third column is the product of the value and the probability, and is
therefore a ‘worth’ or value associated with each particular option. If we
then find the total of all these ‘worths’, we will get a weighted average of
the possible outcomes.

University of
Sunderland 9
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

Thus, we get a total VxP of 21/6 or 3.5 as the Expected Value. Note that
an expected value may or may not be an actual value!

KEY POINT
In most business applications of decision tree analysis, we would be dealing
with financial aspects of the decision. Therefore, we refer to Expected
Monetary Value (EMV).

Example

Consider a game where a coin is tossed. If the outcome is heads, then

you win £3, but if it is tails you lose £1. We will refer to a ‘loss’ value
with a negative number.

On a large number of throws, we would expect an equal number of

heads and tails (assuming, of course, that the coin is not biased). We can
show this as a simple decision tree.

0.5
heads WIN £3

0.5 LOSE £1
tails

The Expected Monetary Value is calculated by finding out the ‘worth’ of

each branch of the decision tree and then adding them together.

Hence, the EMV = (0.5 x 3) + (0.5 x -1) = 1.5 - 0.5 = £1

We would, therefore, expect to win £1 if we played this game. However,

remember that probability indicates what should happen, not what will
happen. Over a large number of plays, we might expect only to average
£1 per play.

University of
10 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

Example

Assume we have an opportunity to invest into a risky project, where the

return payoff of £3m is 75% certain but, we have a 25% chance of losing
£1m.

(0.75) GAIN £3

(0.25) LOSE £1

The Expected Monetary Value (EMV) will be

EMV = (3m x 0.75) + (-1m x 0.25) = £2m.

Forward and Backward passes

When the decision has been constructed, you are ready to evaluate it.
This is where you can work out which option has the greatest worth to
you. This is known as the forward and backward pass. For the forward
pass, we work from left to right in the model and we commence by
assigning a cash value or score as a payoff to the terminal node of each
possible outcome. Estimate how much you think it would be worth to
you if that outcome came about.

Next look at each circle (representing a random node) and estimate the
probability of each outcome. If you have data on past events you may be
able to make rigorous estimates of the probabilities. Otherwise write
down your best guess.

It is important to remember that the probabilities you assign to each of

the branches of a random node, should sum up to unity!

During the forward pass, if a benefit (e.g. a government subsidy) or a

cost (e.g. a capital investment cost) is required along a path throughout
the branches, then include this benefit or cost in an ‘end value’ at the
terminal node.

For example, suppose an expected payoff of £3M is attributed to a

terminal node, but, to achieve that payoff, it costs the company £1M of
investment. Then the end value at that node becomes the payoff minus
the cost of investment i.e. £3M- £1M = £2M. We will demonstrate this
later!

University of
Sunderland 11
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

The ‘end value’ instead of the payoff, then becomes the value to use in
the backward pass.

Once you have worked out the value of the outcomes, and have
assessed the probability of the outcomes of uncertainty, it is time to start
calculating the values that will help you make your decision.

For the backward pass, start on the right hand side of the decision tree,
and work back towards the left. As you complete a set of calculations on
a node (decision square or random node), all you need to do is to record
the result. You can ignore all the calculations that lead to that result
from then on.

Where you are calculating the value of uncertain outcomes (circles on

the diagram), do this by multiplying the value of the outcomes by their
probability. The total expected value for that node of the tree is the sum
total of these values.

When you are evaluating a decision node, simply select the branch with
the ‘best’ payoff or expected value. The ‘best’ payoff in terms of a benefit
to the company (e.g. a profit), is the biggest of the available values
relating to the decision node. In terms of a cost to the company, the ‘best’
payoff is then the smallest of the available values relating to the decision
node.

This can be more easily understood by reference to an example. Assume

we have a company and we are considering the possibility of
developing a new product (or not developing it!). If we decide to
develop, the development will be successful or unsuccessful (a random
node). If we are successful we might decide to manufacture at a high
level or at a low level. See Figure 1.4.

Manufacture
high level

Successful

Develop
Manufacture
low level

Not
successful

Don’t
develop

Figure 1.4. New development problem.

University of
12 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

However, at either high or low level, we know there will be high or low
demand (random nodes), so our diagram now looks like Figure 1.5.

High
demand

Manufacture
high level
Low
demand
Successful

High
demand
Develop
Manufacture
low level
Not
successful Low
demand
Don’t
develop

Figure 1.5. Completed branches for development problem.

We commence the forward pass by inserting the probabilities in the

branches of the random nodes. This is shown in Figure 1.6.

High demand
0.4

Manufacture
high level
Low
demand 0.6
Successful
0.7
High demand
0.4
Develop
Manufacture
low level
Not
successful Low
0.3 demand 0.6
Don’t
develop

Figure 1.6. Probabilities for development problem.

University of
Sunderland 13
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

Then we include any known payoffs and costs. These are included in
Figure 1.7.

High demand
0.4 3.5M

Manufacture
high level
Low
0.5M
demand 0.6
Successful
0.7
High demand
0.4 0.75M
Develop
-1M Manufacture
low level
Not
successful Low 0.75M
0.3 demand 0.6
Don’t
develop

Figure 1.7. Decision tree including payoffs and costs.

We are now able to evaluate the costs and payoffs into the end values.
This is done simply by finding the sum of the monetary figures along
each complete route. Hence, the uppermost end value of £2.5m is
obtained from filtering the £1M cost of development along the
respective path throughout the branches and subtracting it from the
payoff i.e. 3.5m + (-1m). The complete set of end values is shown in
Figure 1.8.

End values
High demand
0.4 3.5M 2.5M

Manufacture
high level
Low
0.5M -0.5M
demand 0.6
Successful
0.7
High demand
0.4 0.75M -0.25M
Develop
-1M Manufacture
low level
Not
successful Low 0.75M -0.25M
0.3 demand 0.6
Don’t
develop -1M

Figure 1.8. Completed end values.

University of
14 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

We now consider the backward pass.

Starting at the right-hand side, for each random node we calculate a

value as by combining the products of end value and probability for all
the branches. (This is sometimes known as the ‘sumproduct’) Thus, the
value of 0.7M, in figure 1.9, is obtained from

(2.5m x 0.4) + ( -0.5m x 0.6) = 0.7m.

End values
High demand
0.7M 0.4 3.5M 2.5M

Manufacture
high level
Low
0.5M -0.5M
demand 0.6
Successful
0.7
High demand
0.4 0.75M -0.25M
Develop
-1M Manufacture
low level
Not
successful Low 0.75M -0.25M
0.3 demand 0.6
Don’t
develop -1M

Figure 1.9. Start of backward pass.

This process is then repeated across the whole diagram. The completed
decision tree is shown in Figure 1.10.

(Note, at each node, where two or more branches originate, the highest
calculated value is retained. After all, the purpose of the business is to
make financial gain!)

University of
Sunderland 15
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

(2.5M x 0.4) + (-0.5M x 0.6) End values

High demand
0.7M 0.4 3.5M 2.5M

choice of 0.7M or -0.25M

Manufacture
0.7M high level
(0.7M x 0.7) + (-1M x 0.3) Low
0.5M -0.5M
demand 0.6
0.19M Successful
0.7
High demand
0.4 0.75M -0.25M
Develop
0.19M Manufacture
-1M
low level
Not
-0.25M
successful Low 0.75M -0.25M
0.3 demand 0.6
Don’t
develop -1M

0
(-0.25M x 0.4) + (-0.25M x 0.6)

Figure 1.10. Final decision tree.

By inspection, it can easily be seen that the optimum solution is to

develop at a high level of manufacture and supply the predicted high
demand. The expected advantage to the organisation would be
approximately £190,000.

Case study 1

Mandy’s Boutique
Mandy Townsend is the owner of a retail outlet called Mandy’s
Boutique and is currently considering a move to a new out of town
shopping mall. The current shop has been in the town centre for 20
years and Mandy has built up quite a sound business. If she moves to
the out of town mall, Mandy thinks that there is about a 20% chance that
the business might decline by £0.2M, a 30% chance that it will remain
stable and a 50% chance that it will increase by £0.35M due to the quality
of the sales promotion undertaken by the mall management. However,
the local authority is considering a revitalisation of the town centre area.
Mandy has heard that there is about a 70% chance that this will be
approved by the council and if it is, she would expect business to
increase by about £0.4M. If it is not built, Mandy thinks that business
will decline by about £0.1M due to the customer movement to the out of
town mall. Time is running out to make the decision on moving to the
new mall.

University of
16 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

Set up a decision tree and calculate the optimal decision for Mandy.

(The solution can be found in the Appendix to this unit.)

Case Study 2

The Metropole Royal Hotel

Hotel rooms are an extremely perishable product and letting them is
always an uncertain business. Mike Robinson, then manager of the
Metropole Royal, a large London hotel situated in the West End was
worried. This year had been exceptionally bad and the coming twelve
month period looked like offering no improvement. The drop in
tourism to London was not something that was likely to last for more
than the next couple of years, but his problem was, how was he to run a
hotel where 25 percent of the bedrooms were forecast to be empty over a
two year period?

By far, the most reliable part of Mike’s business, was the ‘crew contact’.
A crew contact is a contract with an airline for a fixed number of rooms
every night, for which it pays a fixed sum per year. Presently, Mike had
a contract with one of the largest Middle Eastern airlines, Arabia
Airlines, for ten rooms a night. This contract was entering its final year
of a five year period. Currently, the worth to the hotel was about
$100,000 per year.

Most airlines were well aware of the over capacity in the hotel market in
London and, taking advantage of this, Americal Airlines had recently
approached Mike with a proposal for a crew contact of the similar size
to his existing contract with Arabia. However, it insisted that it would
only place the contract with him if he could give Americal a substantial
discount on the price he was currently charging Arabia. Americal was
adamant that the most it would pay was $60 000 for a one year only
contract. Although this price was well below what Mike would
normally offer, he knew that, given the current state of the market, he
would be well able to offer the extra ten rooms to Americal with only a
marginal increase in his costs.

The major danger lay in the chance of offending his longstanding

Middle Eastern customer. Of course, the deal would be theoretically
confidential, and therefore Arabia would not hear of it. However, Mike
knew as well as anybody that crews do talk in the bar at night, and there
was a chance that eventually, Arabia would find out. Mike also
guessed that, if Arabia had no cause for complaint at the end of the year,
they would almost certainly renew the contract for further five years at
the existing price. However, if they realised that substantially better
terms had been offered to one of their competitors, then there was a
small chance that they would cancel the contract straight away (or,
more likely, not renew) unless they got the same terms as Americal had

University of
Sunderland 17
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

obtained when the next five year contract was negotiated. Mike
summed up his dilemma thus:

It all depends upon whether they find out about our deal with Americal
Airlines. You can never tell what will happen. They might be so
offended that they immediately cancelled the contract. Of course, they
would have to pay us a half rate under our agreement, which means
that we would get $50 000 for the next year, but they would have no
difficulty in getting accommodation elsewhere, and we would certainly
not get the contract immediately, then the ball is back in our court.

We can either keep our prices as they are, in which case our chance of
getting the next five year contract would be substantially reduced.
Alternatively, we could offer them the same terms as we offered
Americal Airlines, in which case, given that they haven’t already
cancelled, we could stand a reasonably good chance of getting the
contract, but obviously with reduced profits. My problem is that there
are quite a lot of unknowns in this. For example, I don’t know for sure
that the airline would find out about the Americal deal. I guess on the
whole it’s more likely that they don’t, but there is still a good chance that
they might, say 6 to 4 against them finding out. Similarly, we don’t
know whether they would immediately be so offended that they would
cancel the contract straight away. It’s fairly unlikely, but there is a finite
possibility – I wouldn’t give it more than one in ten chance of happening
though. If they don’t immediately cancel, at least it means that they are
not greatly offended. However, they are almost certainly likely to take
a closer look at the terms we offer them than they would otherwise have
done. If we offer them the same terms as we offered to Americal, it
would represent a 40% reduction in price and therefore I feel that they
would almost certainly be inclined to accept our offer – say a one in ten
chance of not accepting it. However, if we keep our original price, our
chances of getting the contract will be substantially reduced – I would
guess no better than 50:50.

Construct a decision tree for this scenario.

(The solution can be found in the Appendix to this unit.)

Sensitivity analysis
The objective of decision tree evaluation is to investigate and explore the
model in order to determine the underlying principles supporting its
strengths and weaknesses.

In our previous work, we consider the solution to our problems by

considering the logic of the problem and attempting to identify the
decision nodes and the random nodes. We then draw the model,
commence our forward pass by inputting the values of payoffs/costs
and, of course, the values of probability attributed to the branches

University of
18 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

following the random nodes. We end the forward pass by calculating

the end values attributed to the terminal nodes.

Evaluation of the decision tree is undertaken by completing the

backward pass, working from right to left, calculating the expected
value at each random node. At decision nodes, we simply select the
route of the ‘best’ payoff indicated by the expected values.

The work we will be considering in this section expands the previous

analysis by investigating the results of changes in those variables which
determine and thus affect the model.

First let us recap the steps we need to undertake when we attempt to set
up our decision tree model. Consider the Metropole Royal example.

The investigation into the case study continues into some depth. You
need to find the strong and weak points in the system. For example,
where is the point of weakness when a small change in one of the
variables can create maximum problems in the model?

Consider the Metropole Royal example where the ‘offer high rate’
option is considered into a little more depth. The expected value of 410k
is determined by the sumproduct of the probabilities and their end
values.

The alternative node (‘offer low rate’) was initially selected since it had a
better expected value (430k) as compared with the ‘offer high rate’
expected value of 410k.

Suppose now the expected value of the ‘offer high rate’ branch became
430k, i.e. the same as the expected value of the ‘offer low rate’ branch!
We now have an indecisive situation, i.e. do we choose the ‘offer high
rate’ or ‘offer low rate’ since both expected values are the same?

In reality, we would need to identify changes in the variables to help us

to determine one pathway or another. However, identifying the
changes to the variables required to provide us with an indecisive
situation can help us to determine the strengths and weaknesses in the
model.

So how do we do it? To help us to achieve the point of indecision (i.e.

two or more similar expected values in the model) we use some basic
algebra to assist or we make use of Excel’s ‘GoalSeek’ function which
does most of the necessary work for us.

Consider the situation where we have to worry about the offer of a high
rate or the offer of a low rate. The resultant expected value of 410 was
found by the sumproduct of the respective probabilities and end values
along the branch of the ‘offer of the high rate’.

Using algebra, i.e. 410 = (0.5*660)+(0.5*160), the point of indecision will

occur when the expected value of the choice between the offer of the

University of
Sunderland 19
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

high rate is identical to the expected value resulting from the offer of the
low rate, which happens to be 430.

Let us adjust the value of the probability of getting a new contract such
that our EV becomes 430 instead of 410. Should we have decided to
offer the high rate?

Let x represent the value of the new (adjusted) probability.

Hence, 430 = (x*660)+((1-x)*160)

So 430 = 660x + 160 - 160x and subtracting 160 from

both sides of the equation,

430 – 160 = 660x – 160x and simplifying,

270 = 500x

x = 270/500 or x = 0.54

So the probability of an expected value of 430 of ‘getting’ the new

contract if we decide to offer the high rate is 0.54.

As an alternative approach...

The ‘GoalSeek’ function works along the following format: Identify the
cell you want to change (in our case the cell containing the formula to
calculate the expected value of 410).

We then program Excel to change the value of 410 to 430 (to match the
expected value of the other branch of the decision node). Finally we
program Excel to change one of the variables until Excel determines our
present value.

Finally, we undertake a little arithmetic to determine the strengths and

weaknesses. For example, the original probability of the ‘offer high rate’
for the ‘get contract’ branch was 0.5. This value changes to 0.54 to
change the expected value from 410 to 430.

We determine the strength or the weakness in the model by

investigating the change in the value from 0.5 to 0.54 ., i.e. (0.54 –
0.5)/0.5 * 100 = 8%. And we interpret this as an 8% change in the
probability value for ‘getting the new contract’ is sufficient to give us an
indecisive situation.

We can also consider changes in the payoffs to create points of

indecision. However care should be taken since the payoffs between
the various terminal nodes may be interlinked. For example, in the
Metropole Royal example, whether or not the manager offers the high
or the low rate to the Arabians, there will always be the 60k offer from
the Americals and, changing this value at any one of the terminal nodes,
will create problems at the other terminal nodes where the Americal

University of
20 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

offer is available. In the review section that follows, you may relax this
assumption!

The work we have considered in this section has attempted to expand

the analysis and knowledge we studied earlier. We have investigated
the results of changes in those variables which determined and thus
affect the model.

Case Study 3

Vitler’s Plating
Charles Doman had recently taken over the well established but rather
old fashioned company of Vitler’s Plating from his father. The
company had been mainly concerned with silver plating cutlery and
decorative table ware, and chrome plating industrial parts, on a
sub-contract basis. When Charles took over the company, he forecast
(correctly) that both these markets were declining. After consideration,
he decided to re-equip the whole plant and move into an expanding
market, which would provide a higher margin. His programme of
re-equipment included closing and partially stripping his old automatic
silver plating line and buying a newer, technically more advanced, gold
plating line to get into the lucrative gold bathroom fittings market. The
company’s bank manager had been most helpful in providing loans for
the company’s capital investment programme.

Unfortunately, twelve months into the programme, the whole economy

had turned down and business had become very tight indeed. The new
gold plating line had hardly been used, and even when it was working,
it was giving trouble. Charles was most disillusioned:

It’s a technical problem more than anything else. The line just won’t
operate at the rate we had hoped. If we run the line at half of the speed
it is OK, but those rates of working just aren’t economic. At full speed,
the line often requires considerable maintenance and is sometimes
down for several days. Since we bought the line, business has been so
bad that we haven’t been under pressure, but this new order changes
things.

The new order to which Charles referred was a short lead time export
order for the gold plating of a large consignment of bathroom fittings.
The fittings were to be delivered to Vitlers in two weeks’ time, and the
finished goods were due for shipment two weeks after that. At full
speed, the new line could cope with order, but if the line went down for
more than two days, the delivery would be late, and penalties would be
extracted from the company by the customer.

One way of ensuring delivery on time was available. That was to

resurrect the old plating line, to take over if the new line went down.

University of
Sunderland 21
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

However, to make the old line operational again would be expensive.

The dilemma was summed up by Charles:

The major decision is whether to get the old line operational before the
consignment arrives. If we do, delivery can be guaranteed on time,
even though its operating costs are higher than the new line. If we don’t
activate the old line, and the line stops, we could still subcontract if we
arranged it as soon as the line goes down; otherwise we could take the
chance that the line would be repaired in under two days.

Charles’ engineer advised him that he thought that he had the new line
under better control than in the past, but that the chance of it going
down during the two weeks would be about 20%. Should the line go
down, he reckoned on a 50% chance of repairing it within two days.

Activating the old line straight away could result in profits of either
£28000 should the new line stop, or £32000 if the new line was OK. If
Charles did nothing, and the new line proved OK, then he would get the
maximum profit of £40000. If the new line stopped, he could either
subcontract immediately, in which case his profit would be reduced to
£20000, or take a chance and do nothing. If he did nothing, and the line
was down for more than two days, the penalty clauses would mean that
his final profit was unlikely to be more than £4000. If, however, he was
lucky, and the line was repaired in less than two days, his final profit
would be the full £40000.

Set up a decision tree to analyse the situation and report your findings.

(The solution can be found in the Appendix to this unit.)

Case Study 4

Harry’s Portfolio Dilemma

Harry has £25000 to invest. He is aware that he can invest it reasonably
safely into an interest bearing account such as a Building Society, where
it will grow at a steady but small rate. He is also aware that he can set
out his own portfolio and can split his £25000 into three risk areas (viz.
high, medium and low risk) and invest it accordingly and hopefully it
may return a fairly satisfactory return.

Additionally, Harry is aware that he can hire the services of a financial

advisor who will set up a similar portfolio for him. Harry’s mate
Mitchell, is a financial advisor and has offered to do the portfolio for
him at no charge but, has forewarned Harry that he is unable to
guarantee a better return than that Harry could find himself.

Harry decides to assess his options. If he decides to invest his £25000

into a Building Society, then the rate of return on his investment will be

University of
22 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

4%. If he decides to select his own portfolio, Harry, based upon

information received from the ‘Financial Times’, has forecasted the rate
of return on high, medium and low risk investments at 25%, 13% and
5% respectively and due to these rates, Harry has assessed his risk of
success of this portfolio investment at 10% (high), 25% (medium) and
65% (low). For those reasons, Harry has decided that if he were to select
his own investment, he is only prepared to invest 10% of his £25000 into
high risk, 30% into medium risk and the remainder into low risk.

If Harry decides to accept Mitchell’s offer, then based upon Mitchell’s

forewarning, Harry has assessed Mitchell’s offer at being 70:30 in
favour of success. Harry realises that Mitchell’s portfolio would be
similar to his own portfolio but should return better rates since Mitchell
is knowledgeable and qualified in this area and thus should be able to
find a decent deal.

If Mitchell’s offer is successful or not, then Harry would expect a return

of 6% if Mitchell invested the full £25000 into a Building Society (or
equivalent). Harry would expect Mitchell to make a decision between
the Building Society (or equivalent) and the risk portfolio, based upon
the best payoff.

Additionally, if Mitchell’s offer is successful or not, Harry has assessed

that he would expect a rate of return of at least 30%, 17% and 7% for the
respective high, medium and low risk investments. Based upon those
rates, Harry is prepared to invest 25% of his £25000 into high risk, 35%
into medium risk and the remainder into low risk. Harry is aware that a
successful financial advisor portfolio could work out at 0.3 probability
for high risk and 0.35 (each) for medium and low risk. However, Harry
is also aware that an unsuccessful financial advisor portfolio could
work out at 0.2 probability for high risk and 0.3 for medium risk and 0.5
for low risk.

What should Harry do?

(The solution can be found in the Appendix to this unit.)

REVIEW ACTIVITY
Tasks – Commence from the initial state of the Metropole Royal example and
by changing each of the variables in turn, find the points of strength and
weakness in the model. When you identify the weakest point, use the value of
the variable that determined the weakest point and, from there, find the next
weakest point in the model.

Continue throughout, until you have identified all potential weak points in the
system. Then try the same for the strong points!

University of
Sunderland 23
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

REVIEW ACTIVITY FEEDBACK

You might want to consider the possibility of working from left to right in the
model, at the point of the first random node. Consider the expected value of
555.2k and change this to 600k by changing the probabilities and end values in
turn to find the weakest point. For example, if the expected value of the first
random node is changed from 555.2k to 600k by changing the probabilities,
you need to set up the equation as follows:

600 = (x * 398) + ((1 – x) * 660) where x is the unknown probability

You should find that the probability needs to change from 0.4 to 0.229 and the
percentage change is found by (0.4 – 0.229)/0.4 * 100 = 42.75%

Using the expected value at the next random node to change 555.2k to 600k,
the equation is set up as follows:

600 = (0.4 * x) + (0.6 * 660) where x is the unknown expected value

You should find that the expected value needs to change from 398k to 510k
and the percentage change is found by (510 – 398)/398 * 100 = 28.14% Filter
this value of 510k throughout the model from left to right.

Summary
In this unit we have looked at the corner stone of decision making, the
concept of decision tree analysis. Building on an introduction to simple
probability, we have developed the theme to deal with multiple events.
This led to the comparison of different events and their consequences to
an organisation, and to the construction of decision trees.

The analysis stage then looked at payoffs and expected values, using the
forward and backward pass techniques. This is then able to provide us
with an indication of the likely financial implications of a chosen course
of action. By making appropriate changes to the values within the
completed decision tree, we are able to evaluate their effects. This is
known as sensitivity analysis.

University of
24 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

Appendix

Case Study 1 Solution

University of
Sunderland 25
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

Case Study 2 Solution

University of
26 Sunderland
Business Modelling for Decision Making Unit 1 – Decision Tree Analysis

Case Study 3 Solution

University of
Sunderland 27
Unit 1 – Decision Tree Analysis Business Modelling for Decision Making

Case Study 4 Solution

University of
28 Sunderland
Unit 2

Discrete Simulation

Introduction
Simulation can be defined as the process of designing a model of a real
system and conducting experiments with the model in order to fully
understand the mechanisms involved. It is used to assist in making
decisions in areas such as inventory control, factory or shop layouts,
sales forecasting and customer/ patient scheduling. It is currently one
of the most widely used analytical and planning tools.

Simulation techniques are in effect, the mathematically modelling of a

situation and are appropriate when analytical solutions are not possible
or are not straightforward. They can also give more insight and
understanding than would be obtained simply by calculating the ‘right’
answer.

The Simulation Process

Introduction

KEY POINT
Simulation can be defined as the process of designing a model of a real system
and conducting experiments with the model in order to:

! Understand the behaviour of the system.

! Evaluate various strategies for the operation of the system.

Physical simulation models of for example aircraft manufacturers are

well known and understood. However, business, manufacturing and
economic simulation models are now common place, and are used to
assist in making decisions in areas as diverse as inventory control,

University of
Sunderland 29
Unit 2 – Discrete Simulation Business Modelling for Decision Making

factory or shop layouts, sales forecasting and customer/patient

scheduling.

A common question is why use simulation when analytical tools such as

EOQ and linear programming techniques are available? The answer is,
in practice, if analytical techniques are to be applied then this often
means making over simplified assumptions.

Simulation techniques - in effect mathematically modelling of a

particular situation - are therefore appropriate when analytical
solutions are not possible or are not straightforward. E.g. when an
analytical approach leads to a very complex mathematical
representation, it is surprising how often this occurs even with well
known types of problems. Simulations can also give more insight and
understanding than would be obtained simply by calculating the right
answer.

Define the
problem

Formulate
the model

Collect data
and test
model

Run
simulation

Analyse the
results

Make any
changes and
re-run
Select the
best course of
action

Figure 2.1. The Simulation Process.

Simulations can be deterministic or stochastic.

Deterministic simulations are often large-scale systems and can be

used to answer ‘what if’ type of questions. For instance, if such a model
of a company is constructed it can be used to ascertain the financial
effects of, say, a ten percent drop in demand for a particular product.
The basic assumption of deterministic models is that the distribution of
any random variable can be represented by one value only, i.e., no
probability distributions are involved.

University of
30 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

Most models of the national economy are deterministic simulation

models. They allow such questions as ‘what will be the effect of
devaluing sterling by 20 per cent and maintaining other economic
policies?’ to be answered. Certain companies have large-scale models of
their operation, which enable questions such as ‘what is the effect of a 5
per cent drop in sales, or a 10 per cent rise in the price of energy?’ to be
answered.

Stochastic simulations, which incorporate random phenomena, are

sometimes known as Monte Carlo simulations. These models include
random variables which are represented by probability distributions.

Simulation models that do not include risk produce a deterministic (or

predetermined) result given the input values. Stochastic models are
deterministic simulation models that include variables which are not
known with certainty but have a known probability distribution. A
stochastic model is simulated a large number of times using randomly
selected values for the risky variables to estimate the probable outcomes
for key output variables (KOVs). The simulated sample of values for
each KOV constitutes an estimate of the variable’s probability
distribution which can be used to make decisions in a risky
environment.

Stochastic simulation involves simulating uncertain economic systems

that are a function of risky variables, for the express purpose of making
better decisions. In stochastic models future risk is assumed to mimic
historical risk, so past variability is used to estimate parameters for the
probability distributions of risky variables in a model. Probability
distributions are simulated a large number of times to formulate
probabilistic projections for the risky variables. The interaction of the
risky variables with other variables in the system allows the modeler to
project how risky a decision would likely perform under alternative
management strategies. In this way, models can provide decision
makers useful information about the likely outcomes of alternative
management decisions under risk.

A stochastic simulation model is a deterministic model with one or

more variables that have been made stochastic. Development of a
stochastic model usually starts with developing a deterministic model Ngoai sinh
and then converting it to be stochastic by making some of the exogenous
variables stochastic. Even when a model is stochastic it can be used to
generate deterministic results by making the risk components zero or
setting the simulation engine to its “Expected Value” mode. Discrete
simulation modeling techniques are used in ..

! Computer systems development.

! Manufacturing.
! Commerce and finance.
! Government/local authorities.

University of
Sunderland 31
Unit 2 – Discrete Simulation Business Modelling for Decision Making

! Ecological and environmental analysis.

! Social and behavioural analysis.
! Biosciences.

Advantages and Disadvantages

Advantages:

! Simulation may be the only method available.

! Can be used to analyse large and extremely complex
situations.

! Simulation does not interfere with the real world

situation.

! Allows ‘what if’ questions to be posed and answered.

Text
! Time compression is possible.
! Designs/methods can be tested prior to committing.
! Insight can be gained into the importance of specific
variables.

Disadvantages:

! Can be expensive to develop the model.

! Does
Textnot generate optimum solutions.
! Managers still need to generate all the solution
approaches they want to test.

Random Variables in a Simulation Model

Any variable the manager or decision maker cannot control can be
made stochastic. A random variable may be partly determined by
internal forces but still have a random component.

Another type of random variable is a forecasted variable which has risk

because the forecast is less than perfect. Variables affected by weather
are also stochastic, e.g., yields for crops or livestock response to feed.
Production can be stochastic due to risk caused by inconsistent quality
of inputs and response to the inputs. Market share is often a stochastic
variable because managers do not know how competitors will react to
their promotional campaigns or pricing strategies. The quantity sold

University of
32 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

each period is a stochastic variable depending on many factors outside

the manager’s control.

In all cases, stochastic variables are: crucial to the success of a business

decision, out of the control of the decision maker, and can be specified
by a probability distribution. If the variable is so risky that the decision
maker has no idea what the probability distributions is, then that is
uncertainty.

Simulating Uncertainty
Risk and uncertainty are out of the decision makers control and most
often they are the forces that affect the success of a business decision.
Risk is the portion that can easily be modeled using stochastic Phan
simulation. Risky variables have probability distributions that define
the nature of their risk. For example, sales per period may be normally
distributed with a mean of 10,000 and a standard deviation of 300.

Uncertainty is risk which can not be defined by a probability

distribution. An uncertain variable is one which does not have a known
distribution. One way to think about uncertainty is that the mean and
standard deviation for the variable’s probability distribution are risky.

Because uncertain variables do not have known distributions and

parameters they can not be simulated directly. Uncertainty can be
incorporated into simulation models two ways. The first is to use an
example from catastrophe theory and assume the worst outcome
happens at random with a probability of P. For example, a plant
manager may know that the main power supply can fail but has no idea
when it will occur. Each period the power failure variable is simulated,
if it equals 1 then the uncertain event occurred.

Another way to try to stress test a business for uncertainty is to test the
probability distributions for risky variables by making their means
stochastic using a sensitivity analysis. A range of means for each
distribution can be simulated to determine which is most critical to the
business decision. This approach can be extended to the variability
parameter for each risky variable.

ACTIVITY
Carry out some additional reading about Discrete Event Simulation. Look, if
possible, for a suitable example of its use in a specific case study.

University of
Sunderland 33
Unit 2 – Discrete Simulation Business Modelling for Decision Making

Monte Carlo Simulation

The basis of Monte Carlo simulation is effectively to gather sample data
to find out what is going on in a particular situation, and then use the
sample data to construct a probability distribution which is used as the
framework for simulating the required event(s).

KEY POINT
There are five simple steps:

1. Collect data and set up a probability distribution for important

variables.

2. Build a cumulative probability distribution for each variable.

3. Establish an interval of random numbers for each variable

4. Generate random numbers

5. Simulate a series of trials

The principles are best illustrated by considering a number of simple

examples. Students are recommended to study each one carefully and
decide the important information for themselves. In some cases, there
may be a number of possible routes to an appropriate solution.

A specialist eye surgery unit at a local hospital is considering its staffing phau thuat
arrangements following a series of complaints from patients about chuyen khoa
delays. Prior to making any changes the hospital management decides mat
to conduct a simulation study. As a first step historic data is used to
compile information about the number of patients that arrive at the unit
each day and the number of patients that are treated each day.

This data is shown in Figure 2.2

University of
34 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

Delayed from Random Total to be Random Number

Day Arrivals
Yesterday Number Treated Number Treated
1 0 52 3 3 37 3
2 0 6 0 0 63 0
3 0 50 3 3 28 3
4 0 88 4 4 2 1
5 3 53 3 6 74 4
6 2 30 2 4 35 3
7 1 10 0 1 25 1
8 0 47 3 3 3 1
9 2 99 5 7 29 3
10 4 37 2 6 60 3
11 3 66 3 6 74 4
12 2 91 5 7 85 4
13 3 35 2 5 90 5
14 0 32 2 2 73 2
15 0 99 5 5 59 3
Totals 20 42 40

Figure 2.2. Simulation table.

Simulation Results

! The average number of patients delayed to the next day =

20/15 = 1.33 patients per day.

! The average number of arrivals = 42/15 = 2.8 patients per

day.

! The average number of patients treated per day = 40/15

= 2.67.

One way of designing, in fact modelling, is to sketch out the problem in

a diagram. There several formal methods that can be adopted here, e.g.
spider-diagram, block-diagram, flowchart or causal-relationship
diagrams. All have the same thing in common, that is they attempt to
display the key variables of a problem and how they interact with each
other. In this module we shall not adopt any single method as the
correct one to use, the onus is on you to produce diagrams that you
understand and are thus able to describe your understanding of a
problem to others. This form of communication is crucial to good model
development. Once a relevant diagrammatic representation is available
it is then often easier to develop the simulation whether it is in tabular
form, such as a spreadsheet, or computational form using a computer
package or a tailor-made program.

University of
Sunderland 35
Unit 2 – Discrete Simulation Business Modelling for Decision Making

ACTIVITY
Now you try and draw your own diagram that represents the Eye Unit
scenario.

Monte Carlo
Simulation - Example

Arrivals at the Eye Unit :

Number of Probability Cumulative Random
of Arrivals Probability Number
0 13% 13% 1 - 13
1 17% 30% 14 - 30
2 15% 45% 31 - 45
3 25% 70% 46 - 70
4 20% 90% 71 - 90
5 10% 100% 91 - 00

Monte Carlo
Simulation - Example

Number of Patients Treated per day :

Treated Probability Cumulative Random
Patients Probability Number
1 5% 5% 1- 5
2 15% 20% 6 - 20
3 50% 70% 21 - 70
4 20% 90% 71 - 90
5 10% 100% 91 - 00

University of
36 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

Monte Carlo Simulation - Example Results

Day Delayed RN Arrivals Total to RN Number

from Yest. be treated Treated
1 - 52 3 3 37 3
2 0 06 0 0 63 0
3 0 50 3 3 28 3
4 0 88 4 4 02 1
5 3 53 3 6 74 4
6 2 30 1 3 35 3
7 0 10 0 0 24 0
8 0 47 3 3 03 1
9 2 99 5 7 29 3
10 4 37 2 6 60 3
11 3 66 3 6 74 4
12 2 91 5 7 85 4
13 3 35 2 5 90 4
14 1 32 2 3 73 3
15 0 00 5 5 59 3
Totals 20 41 39

Example

A small medical practice is thinking of expanding as the number of

patients being dealt with has increased. At present there are two doctors
working at the surgery (mornings only). Prior to making any changes
the medical surgery management decide to conduct a simulation study,
with a view to expansion and employing more staff. The first stage is to
conduct a simulation of the existing structure as improvements may be
made without expanding. When both doctors are available, patients
prefer to see Doctor No. 2.

Arrivals at the medical surgery

Probability Cumulative Random numbers Inter arrival

Probability

20% 20% 1 to 20 00:05

35% 55% 21 to 55 00:10

20% 75% 56 to 75 00:15

15% 90% 76 to 90 00:20

10% 100% 91 to 100 00:25

University of
Sunderland 37
Unit 2 – Discrete Simulation Business Modelling for Decision Making

Length of consultation

Probability Cumulative Random numbers Consultation

probability time

25% 25% 1 to 25 00:25

20% 45% 26 to 45 00:30

40% 85% 46 to 85 00:35

15% 100% 86 to 100 00:40

Simulation Results

The random numbers are allocated in proportion to the probability that

we want to simulate. For the arrival table above, thus there is a 20%
chance that patients will arrive after a five-minute interval. We
therefore generate random numbers between 1 and 100 inclusive and
allocate the first 20 of them to this event (numbers 1 to 20). Other
numbers are allocated in a similar fashion. In the model we generate
random numbers and for each one check which range it lies in. A
random number occurring between 1 and 20 inclusive gives an arrival
interval of 5 minute, a random number occurring between 21 and 55
inclusive gives an arrival interval of 10 minutes and so on.

A simulation of the first 20 patients can be seen in Figure 2.3. If you

follow this through you can see how the simulation builds up and that,
in this case, some patients are having to wait up to 40 minutes to see a
doctor. Conversely the doctor is not often idle, i.e. waiting for a patient.
Summary results for simulations in this case are also given.

University of
38 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

Figure 2.3. Simulation of twenty patients.

ACTIVITY
What do you think are the key values to be measured? (Hint: think as a patient
and as an employer)

Collect the appropriate values and give your view on the existing system by
making inferences from the values that you decided to collect. Draw a diagram
to represent the surgery scenario.

ACTIVITY
A consultant holds a clinic at the hospital every Wednesday afternoon for a
maximum of ten patients. They are expected to attend the clinic in two equal
groups at 2.00pm and 2.30pm. He is concerned that in recent weeks he has
wasted rather a large amount of time waiting for patients to arrive and,

University of
Sunderland 39
Unit 2 – Discrete Simulation Business Modelling for Decision Making

conversely, some patients have complained of excessive waiting times

themselves. He has decided to seek advice on ways of improving his clinic time.

You have been asked to analyse the situation by using a simulation and
determine the amount of waiting time, if any, for the consultant and his
patients. The consultant has kept a record of his last 100 patients and the
lengths of their appointments. This is shown in the table. Use the ten random
numbers given to simulate a sample afternoon.

Appointment length (mins) Number of patients

4 1

5 3

6 6

7 7

8 11

9 12

10 13

11 11

12 9

13 6

14 5

15 4

16 3

17 4

18 3

19 2

Random numbers; 40 07 46 12 96, 06 05 47 38 86

University of
40 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

ACTIVITY FEEDBACK
Assuming all the patients arrive at their allotted times, the data is represented
in the following table;

Appointment Number of Probability Cumulative Random

length (mins) patients probability number

4 1 0.01 0.01 1

5 3 0.03 0.04 02 to 04

6 6 0.06 0.10 05 to 10

7 7 0.07 0.17 11 to 17

8 11 0.11 0.28 18 to 28

9 12 0.12 0.40 29 to 40

10 13 0.13 0.53 41 to 53

11 11 0.11 0.64 54 to 64

12 9 0.09 0.73 65 to 73

13 6 0.06 0.79 74 to 79

14 5 0.05 0.84 80 to 84

15 4 0.04 0.88 85 to 88

16 3 0.03 0.91 89 to 91

17 4 0.04 0.95 92 to 95

18 3 0.03 0.98 96 to 98

19 2 0.02 1.00 99 to 100

This can then be converted into a schedule of events.

University of
Sunderland 41
Unit 2 – Discrete Simulation Business Modelling for Decision Making

Time Event Consultant’s Patients’ Patient

idle time waiting
time

2.00 Patients 1 to 5 arrive

2.00 P1 starts appointment 0 0 P1

(R/N 40 = 9 mins appointment)

2.09 P1 ends/P2 starts 0 9 P2

(R/N 07 = 6 mins appointment)

2.15 P2 ends/P3 starts 0 15 P3

(R/N 46 = 10 mins appointment)

2.25 P3 ends/P4 starts 0 25 P4

2.30 Patients 6 to 10 arrive

(R/N 12 = 7 mins appointment)

2.32 P4 ends/P5 starts 0 32 P5

(R/N 96 = 18 mins appointment)

2.50 P5 ends/P6 starts 0 20 P6

(R/N 06 = 6 mins appointment)

2.56 P6/ends/P7 starts 0 26 P7

(R/N 05 = 6 mins appointment)

3.02 P7 ends/P8 starts 0 32 P8

(R/N 47 = 10 mins appointment)

3.12 P8 ends/P9 starts 0 42 P9

(R/N 38 = 9 mins appointment)

3.21 P9 ends/P10 starts 0 51 P10

(R/N 86 = 15 mins appointment)

3.36 P10 ends

Overall, this gives

Total Consultant Idle Time (CIT) = 0

Total Patient Waiting Time (PWT) = 267 mins

Average PWT = 267 / 10 = 26.7 mins.

Now repeat the activity for a different set of random numbers. You must
decide on an appropriate way to generate these numbers.

Your results will give you another simulation of the afternoon’s clinic session.
How does this compare with your first simulation? Further simulations will
provide other information about waiting times and should be developed to
gain a better understanding of the current situations failures.

University of
42 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

In the previous activity, it was assumed that all patients arrived at their
designated appointment times. This is unrealistic. Our simulation can
be improved by considering a variation in arrival times. This is achieved
by using the data in the following table, which has been found by
comparing patient arrival times with their appointment times.

ACTIVITY
Patient reliability;

Time Probability Cumulative Random number

Probability

4 minutes early 0.10 0.10 1-10

2 minutes early 0.25 0.35 11-35

On time 0.35 0.70 36-70

2 minutes late 0.15 0.85 71-85

4 minutes late 0.10 0.95 86-95

Fail to appear 0.05 1.00 96-100

Using the set of random numbers given, generate a set of arrival times for the
ten patients. In this simulation, three patients have been asked to attend at
2.00pm, four at 2.30pm and three at 3.00pm.

Random numbers; 32 04 16, 44 45 93 22, 94 74 20

University of
Sunderland 43
Unit 2 – Discrete Simulation Business Modelling for Decision Making

ACTIVITY FEEDBACK
Patients’ time of arrival;

P1 1.58
P2 1.56
P3 1.58

P4 2.30
P5 2.30
P6 2.34
P7 2.28

P8 3.04
P9 3.02
P10 2.58

ACTIVITY
Using the patient arrival times from the previous activity, compile a schedule of
events for this simulation.

University of
44 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

ACTIVITY FEEDBACK
Assuming patients are seen in numerical order, the schedule of events
becomes;

Time Event Consultant’s Patients’ Patient

idle time waiting
time

1.56 Patient 2 arrives.

1.58 Patients 1 and 3 arrive

2.00 P1 starts appointment 0 2 P1

(R/N 32 = 9 mins appointment)

2.09 P1ends/P2 starts 0 13 P2

(R/N 04 = 5 mins appointment)

2.14 P2 ends/P3 starts 0 16 P3

(R/N 16 = 7 mins appointment)

2.21 P3 ends

2.28 Patient 7 arrives

2.30 Patients 4 and 5 arrive

2.30 P 4 starts 9 0 P4
(R/N 44 = 10 mins appointment)

2.34 Patient 6 arrives

2.40 P4 ends/P5 starts 0 10 P5

(R/N 45 = 10 mins appointment)

2.50 P5 ends/P6 starts 0 16 P6

(R/N 93 = 17 mins appointment)

2.58 Patient 10 arrives

3.07 P6 ends/P7 starts 0 39 P7

(R/N 22 = 8 mins appointment)

3.02 Patient 9 arrives

3.04 Patient 8 arrives

3.15 P7 ends/P8 starts 0 11 P8

(R/N 94 = 17 mins appointment)

3.32 P8 ends/P9 starts 0 30 P9

(R/N 74 = 13 mins appointment)

3.45 P9 ends/P10 starts 0 47 P10

(R/N 20 = 8 mins appointment)

3.53 P10 ends

University of
Sunderland 45
Unit 2 – Discrete Simulation Business Modelling for Decision Making

Overall,

Total Consultant Idle Time (CIT) = 9

Total Patient Waiting Time (PWT) = 184 mins

Average PWT = 184 / 10 = 18.4 mins.

Average PWT (of those patients who actually waited) = 184 / 9 = 20.44 mins.

As before, it is now necessary to repeat the simulation using different sets of

random numbers. It will also help if different groupings of arrival times are
selected, i.e, appointments made at 10 minute intervals, groups of two patients
every 15 minutes, etc. it is important to experiment with a wide range of
options.

This will provide a set of simulated clinic sessions from which it will be possible
to gain a better understanding of the existing system.

Other Simulation Models

There are a number of useful simulation models that can be used, but
perhaps the most common ones are:

! Operational Gaming
A business game where students/executives are given the
opportunity to test decision making skills.

! Systems Simulation
A business gaming system that permits users to model
the dynamics of large systems. Typical examples are
national economic models and trade models, such as
those used by the Treasury.

ACTIVITY
Carry out some research to find other simulation models. Briefly describe
each one, including details of how each works and in what situations it would
be useful.

University of
46 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

Case Study
The objective of the simulation case study evaluation is to investigate
and explore the simulation model, to determine the underlying
principles supporting its strengths and weaknesses. With discrete
simulation, the models revolve around queuing theory; for example,
when a customer waits in a line of people for service such as at a bank or
post office.

This will involve the generation of random numbers and, by using the
basic theory of probability, to determine the length of time spent in a
queue. This information is then used to simulate the problem and
permit us to determine the length of time spent queuing.

In this section, we will develop an analysis by investigating the result of

changes in those variables which determine and thus affect the model.

We begin by reviewing the steps needed when we attempt to set up our

simulation model:

1. Collect data and set up a probability distribution for important

variables.
2. Build a cumulative probability distribution for each variable.
3. Establish an interval of random numbers for each variable
4. Generate random numbers
5. Simulate a series of trials

Let us look now at an example of a surgery which has two doctors, and
patients arrive for scheduled appointments. There are two sets of
variables in this model:

! The distribution of the inter-arrival times.

! The distribution of the length of the consultation times.

When we study this system, our first observation is that we should

undertake more than one simulation. We would be able to summarise
the waiting times for doctors and patients following, perhaps, thirty
simulations.

Note, before each of the thirty simulations, the model was set up as the
original. In all case studies analysis, you should consider commencing
from the initial or original state of the model before you decide to
undertake any changes.

The results from thirty simulations are shown in Figure 2.4.

University of
Sunderland 47
Unit 2 – Discrete Simulation Business Modelling for Decision Making

Figure 2.4. Data from thirty simulations.

We can now use this information to calculate the error. This is found
from the following formula:

Error = ((Average - Original)/ Average) x 100

Hence, ((9 - 15) / 9) x 100 = -67% (or 67% error)

The results are shown in Figure 2.5.

Doctor 1 Doctor 2 Number of Patient

Idle time Idle time patients waiting time
(total) (total) waiting (AV)

Averages 9 15 5 9

Original 15 15 9 16

Error (%) 67 0 80 78

Figure 2.5. Calculation of error.

We continue the analysis on the simulation by adjusting the values of

the variables in turn. For example, we can arbitrarily adjust the values of
the patient inter-arrival times, such that we can experience some

University of
48 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

non-attendees and, the latest inter-arrival time is 20 minutes. (Assume

the probabilities remain constant!)

We can then undertake a series of simulations with this non-attendee

(NA) constraint and calculate the errors in a similar way.

University of
Sunderland 49
Unit 2 – Discrete Simulation Business Modelling for Decision Making

Now look at a further example. Consider the initial state of the scenario.
Let us now assume that Doctor 1 is called away between 10.20am and
returns at 11.15am. We need to try and simulate that situation. (Once
again, we can assume that all the probabilities remain constant.)

University of
50 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

We can then undertake a series of simulations with this (Call) constraint

and calculate the errors in the same way we did before.

The investigation into the case study continues in some depth. You need
to find strong and weak points in the system. For example, where is the
point of weakness where a small change in one of the variables can
create maximum problems in the model?

REVIEW ACTIVITY
Medical Consultation - Monte Carlo Method

The information presented below should be read in conjunction with the DR 1

and DR 3 templates provided.

Patients arrive at a clinic for a consultation. The pattern of arrivals is shown on

the template and is expressed in terms of ‘Arrival interval’ i.e. this shows the
time between patients arriving. For example if Patient 3 arrives at 09:30 and
the arrival interval for Patient 4 is 15 minutes, then Patient 4’s arrival time
would be shown as 09:45. The clinic opens at 08:00 and there are two doctors
on duty. Both doctors are available to see patients from 08:00 onwards but it
is known that patients generally prefer to see Doctor 1. The time required for
a consultation varies between 20 minutes and 35 minutes and the distribution
of consulting times is shown on the attached template.

As patients arrive they will see the first doctor that is free. If both doctors are
available they will see Doctor 1. The first patient arrives at 08:00 hours.

Requirements:

1. Consider the DR 1 template. Random number data is already inserted

into the template for you. You need to simulate the arrival and
consultation of the first 15 patients. Calculate the average waiting time
for patients and total idle time for each doctor. What conclusions can
you draw, and how could this type of information be used to establish
service standards and to plan staff usage?

University of
Sunderland 51
Unit 2 – Discrete Simulation Business Modelling for Decision Making

2. Consider the DR 3 template. All consultation times have been

reduced by 5 minutes. You should consider this as an application of
sensitivity analysis. As with the Dr 1 exercise, all random number data
is inserted for you. You may assume that the patient inter-arrival times
are constant (i.e. the same as for the Dr 1 exercise). Your task is to
simulate the arrival and consultation of the first 15 patients (similar to
the Dr 1 exercise). Calculate the average waiting time for patients
and total idle time for each doctor. Considering the Dr 1 exercise,
what conclusions can you draw from the Dr 3 exercise? What ‘new’
information (if any) do you have?

University of
52 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

University of
Sunderland 53
Unit 2 – Discrete Simulation Business Modelling for Decision Making

REVIEW ACTIVITY FEEDBACK

Patient Arrivals (Minutes)

Inter

Arrivals Percent Culm RN1

00:05 15% 15% 1-15 1 00:05

00:10 35% 50% 16-50 16 00:10

00:15 25% 75% 51-75 51 00:15

00:20 15% 90% 76-90 76 00:20

00:25 10% 100% 91-00 91 00:25

(where 00 is 100)

University of
54 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

Consultation (Minutes)

Times Percent Culm RN2

00:20 25% 25% 1-25 1 00:20

00:25 20% 45% 26-45 26 00:25

00:30 40% 85% 46-85 46 00:30

00:35 15% 100% 86-00 86 00:35

(where 00 is 100)

University of
Sunderland 55
Unit 2 – Discrete Simulation Business Modelling for Decision Making

Patient Arrivals (Minutes)

Inter

Arrivals Percent Culm RN1

00:05 15% 15% 1-15 1 00:05

00:10 35% 50% 16-50 16 00:10

00:15 25% 75% 51-75 51 00:15

00:20 15% 90% 76-90 76 00:20

00:25 10% 100% 91-00 91 00:25

(where 00 is 100)

Consultation (Minutes)

Times Percent Culm RN2

00:15 25% 25% 1-25 1 00:15

00:20 20% 45% 26-45 26 00:20

00:25 40% 85% 46-85 46 00:25

00:30 15% 100% 86-00 86 00:30

(where 00 is 100)

REVIEW ACTIVITY
Checkouts - Monte Carlo Method

Consider the Checkout template. You are instructed to complete this grid
similar to the two Doctor grids above. The difference here is that you need to
identify two completely different sets of random numbers and insert them into
the respective columns (i.e. columns RN1 and RN2) on the ‘checkout’
template. The seminar leaders are supplied with a solution to this problem
using columns B and G of random numbers. You should also use columns B
and G of the supplied random number (RNs) grid (available towards the rear of
this chapter) if you need your solutions verified.

(Instructions for the use of columns B & G - Use the random numbers
from column B of ‘RNs’ and commence with the first number (i.e. the value at
cell B3) into RN1 of the ‘checkout’ template at customer No. 2 (i.e. assume
customer No. 1 punctually arrives at 09:00 hours. The first value of the

University of
56 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

random numbers from column G data (i.e. the value at cell G3 from ‘RNs’),
should be inserted into RN2 of the ‘Checkout’ template, at customer No. 1
(i.e. we do not know how long the first customer will spend at the checkout!)
Continue inserting successive and respective random numbers throughout).

Your task is to simulate the customer arrival to the checkouts and the service
time of the checkouts for the first 15 customers. Calculate the average
waiting time for customers and total idle time for checkout operator. If you
were responsible for staff resourcing, what would be your conclusions?

University of
Sunderland 57
Unit 2 – Discrete Simulation Business Modelling for Decision Making

REVIEW ACTIVITY FEEDBACK

Customer Arrivals (Minutes)

Inter

Arrivals Percent Culm RN1

00:02 57% 57% 1-57 1 00:02

00:04 5% 62% 58-62 58 00:04

00:06 18% 80% 63-80 63 00:06

00:08 5% 85% 81-85 81 00:08

00:10 15% 100% 86-100 86 00:10

University of
58 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

Checkout Service (Minutes)

Times Percent Culm RN2

00:04 5% 5% 1-5 1 00:04

00:05 10% 15% 6-15 6 00:05

00:06 60% 75% 16-75 16 00:06

00:07 25% 100% 76-100 76 00:07

REVIEW ACTIVITY
Aberdeen Airways - Monte Carlo Method

The existing services of Airways Plc provides scheduled air services on a

number of routes. One of the routes that the company serves is between
Aberdeen and Amsterdam. Aberdeen Airways operates 3 daily return flights
between the two cities. These flights operate five days a week (Monday to
Friday) all year round and a significant part of the trade is accounted for by
business traffic. The daily service schedule currently being operated is as
follows:

ABERDEEN to AMSTERDAM AMSTERDAM to ABERDEEN

1.1.1 Departure times 1.1.2 Departure times

07.00 09.15

11.30 13.45

16.00 18.30

All departure times are based on GMT i.e. you can ignore any time differences
between the UK and the Netherlands.

The flying time between Aberdeen and Amsterdam, and vice versa, is
scheduled to be 1 hour 30 minutes.

The data shown contains departure times, flight times, turn around times and
the times that the aircraft is ready to return from Amsterdam for all the 07:00
departures from Aberdeen for the last 18 months.

University of
Sunderland 59
Unit 2 – Discrete Simulation Business Modelling for Decision Making

In the data file:

! Actual departure times are recorded to the nearest 5 minutes.

These do of course vary from the scheduled departure time
for a variety of reasons.

! Actual flight times are also recorded to the nearest 5 minutes

and again these do vary from scheduled flight times depending
on climatic and other conditions.

New Aircraft

In a months time Aberdeen Airways will be changing the aircraft it uses for the
Aberdeen - Amsterdam service. The new aircraft will be more economical and
marginally quicker than the existing aircraft, and the scheduled flying time will
be reduced by 5 minutes [to 1hour 25 minutes]. This reduction in scheduled
flying time is expected to be reflected in an exact equivalent reduction of 5
minutes in the actual flight times i.e. the impact of climatic and other conditions
on actual flight times is expected to be precisely as before.

Although the scheduled flight time will be reduced by five minutes the current
departure schedule will be maintained. The company does however, expect to
benefit from the improved aircraft economy and hopefully improved service
reliability.

Use a pivot table (see Appendix) to analyse the data provided in the file. Based
on the results obtained, produce a simple Monte-Carlo simulation model to
simulate twenty days of 07:00 departures from Aberdeen to Amsterdam based
upon using the new aircraft. Your simulation model output should show the
simulated values relating to the:

! Departure time from Aberdeen to Amsterdam

! Flight time

! Turnaround times

! Times that Amsterdam to Aberdeen flight is ready to depart

You need to be able to briefly describe your model; produce a table

summarising the results of the simulation exercise (i.e. set up your own
template similar to the templates below); comment on the results obtained.

(Hint, to help with your analysis, complete two simulations i.e. one for the
current situation and one for the proposed situation).

University of
60 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

REVIEW ACTIVITY FEEDBACK

University of
Sunderland 61
Unit 2 – Discrete Simulation Business Modelling for Decision Making

University of
62 Sunderland
Business Modelling for Decision Making Unit 2 – Discrete Simulation

Summary
In this unit, we have investigated the use of simulation methods to help
understand the processes in question. The concept of mathematical
modelling has been demonstrated through a number of different
examples.

To gain a more complete understanding of the mechanism being

simulated, it is often necessary to repeat the simulation a number of
times and observe the patterns that emerge from them. This provides a
valuable overview for analysis.

The student should not be put off by the apparent complexity of the
mathematics involved. There a many different ways of generating
random numbers, for example, and in the unit we have looked at some
ideas. If these appear complicated, there are other routes that can be
found which may be easier to accept.

University of
Sunderland 63
Unit 2 – Discrete Simulation Business Modelling for Decision Making

In reality, most simulations require only the repeated use of relatively

straightforward manipulation of data.

Appendix – Pivot Tables

What is a Pivot table?

A Pivot table is to a way to extract data from a long list of information,
and present it in a readable form. Remember the data we had from the
student scores spreadsheet? You could turn that into a pivot table, and
then view only the Maths scores for each pupil. Or view just Paul’s
scores, and nobody else’s.

To get a clearer idea of just what a Pivot Table is, examine the one below.

In this school, there is a test every month (it’s a tough school!). The Pivot
Table above shows the marks that Elisa got in January, February, and
March. There were tests for only 6 subjects. Notice the black
down-pointing arrows in the Pivot Table. On Row 1 we have Student
Elisa. If the black arrow were clicked, a drop-down box would appear
showing a list of the other students. We could click on a student and
view the marks he or she achieved. Or we could select which subjects to
view, or choose only one month.

But Excel does most of the work for you, and puts in those drop-down
boxes as part of the wizard. Explore this for yourself!

In order to fully understand how useful Pivot Tables can be, the student
is advised to obtain a suitable text book or visit the following websites to
see how they are constructed and used:

http://www.accountingweb.co.uk/cgi-bin/item.cgi?id=18517

http://www.ccsr.ac.uk/popgroup/docs/POPGROUPPivotTablesFeb
05.pdf

University of
64 Sunderland
Unit 3

Linear Programming

Introduction
Linear Programming is concerned with the management of scarce
resources. Suppose a company wants to make several different
products but each product makes different demands on the limited
available resources. The question is, how many of each one does the
company make to maximise profits?

Alternatively, how many of each one does the company make to

minimise costs? Linear Programming helps us to solve questions of this
type. Note that with linear programming, we are dealing with
inequalities rather than equalities!

As an introduction into Linear Programming, let us consider an

example. ‘Sunderland Optimum Shirts’ (SOS) is a company making
high quality shirts and are planning to make two distinct types
(‘Regular’ and ‘Deluxe’). The contribution to profits of the ‘Regular’
(‘Deluxe’) shirt is £5 (£8).

To manufacture the shirts, SOS have available to themselves 600m2

worth of cotton. Regular (Deluxe) shirts require 5m2 (6m2) worth of
cotton. Additionally, Regular (Deluxe) shirts take 1 (2) hour(s) to
manufacture. SOS have 20 machinists who work 8 hours/day.

To maximise profits, how many shirts of each should be made?

The first step is to identify the variables.

The two variables, of course, are the different amounts of the Regular
and the Deluxe shirts to produce.

Let R (D) represent the number of Regular (Deluxe) shirts to produce to

maximise profits.

The next step is to set up an equation with the variables. We know that
we want to maximise profits and we know that the Regular (Deluxe)
shirt contributes £5 (£8) towards profit. Hence, maximise £5R + £8D

The next step is to look for the constraints (if any!).

University of
Sunderland 65
Unit 3 – Linear Programming Business Modelling for Decision Making

The first constraint is the daily amount of available cotton (600m2) and, a
Regular (Deluxe) shirt, takes 5m2 (6m2) of material. So we can make 5R
and 6D shirts out of 600m2 cotton. Hence we can equate : 5R + 6D " 600

The second constraint is the daily length of time taken to produce one
shirt. For the Regular (Deluxe) shirt, it takes 1 (2) hour(s). There are 20
machinists who work 8 hours/day. Hence there are 160 available hours
to make 1R and 2D shirts.

Hence we can equate : R + 2D " 160

Let us summarise what we want to do, we want to maximise 5R + 8D

subject to:

5R + 6D " 600 and

R + 2D " 160

and, of course, neither R nor D can be a negative value, so R, D " 0

There are many values for R and D which will solve this problem
without maximising the profits. What we want to do, is to maximise
profits. Since there are only two variables (R and D), we will solve this
model by the Graphical method.

Let us look at the constraints first.

5R + 6D " 600 (cotton)

Let us assume 5R + 6D = 600

then when R = 0, 6D = 600, hence D = 100
and when D = 0, 5R = 600, hence R = 120

Similarly, R + 2D " 160 (labour)

Let us assume R + 2D = 160

then when R = 0, 2D = 160, hence D = 80
and when D = 0, R = 160

Recap : R D

cotton 0 100
120 0

labour 0 80
160 0

and we can plot these on a graph: (Cotton first!)

University of
66 Sunderland
Business Modelling for Decision Making Unit 3 – Linear Programming

Linear Programming
The Graphical Method
D
Cotton
100

0 120 R

Linear Programming
The Graphical Method
D
100 Cotton
80
x

Labour

0 120 R
160

Linear Programming
The Graphical Method
D
Cotton The region (F) is called
100
the feasible region and
80 is ‘bounded’ by 0, 80,
x x, 120, 0
F
Labour

0 120 R
160

Notice how the constraints have provided a feasible region. The feasible
region (F) represents the area of the graph which will satisfy all of the (R
and D) inequalities.

University of
Sunderland 67
Unit 3 – Linear Programming Business Modelling for Decision Making

Next, we need to add the profit equation to the graph. Recall that we
want to maximise 5R + 8D

To do this, let R = 40 and D = 20 (these are arbitrary values and could be

any value!)

then 5(40) + 8(20) = 360 (we then equate 5R + 8D to 360)

and we then solve for R and for D in turn,

i.e. when R = 0, 8D = 360, hence D = 45 and when D = 0, 5R = 360, hence R

= 72

and these values simply represent yet another linear line! - plot it!

Linear Programming
The Graphical Method
D
Cotton The region (F) is called
100
the feasible region and
80 is ‘bounded’ by 0, 80,
x x, 120, 0
45
F
Labour

0 R
72 120 160

What we do next, is simply to physically move the profit line parallel

from its current position until we touch the very edge of the feasible
region. Use a straight edge to help you to do this!

Linear Programming
The Graphical Method
D
Cotton The region (F) is called
100
the feasible region and
80 is ‘bounded’ by 0, 80,
x x, 120, 0
45
F
Labour

0 120 160 R
72

University of
68 Sunderland
Business Modelling for Decision Making Unit 3 – Linear Programming

Linear Programming
The Graphical Method
D
Cotton The region (F) is called
100
the feasible region and
80 is ‘bounded’ by 0, 80,
x x, 120, 0
45
F Labour
R
72 120 160

Linear Programming
The Graphical Method
D
100 Cotton The region (F) is called
the feasible region and
80 is ‘bounded’ by 0, 80,
x x, 120, 0
45
F Labour
R
72 120 160

Linear Programming
The Graphical Method
D
Cotton This is the point of
100
maximum profit
80
x
45
F Labour
R
72 120 160

On the last figure, and at the point indicated by ‘x’, we draw a vertical
line from ‘x’ to the horizontal axis. Similarly, we draw a horizontal line
from ‘x’ to the vertical axis. In both cases, we read off the respective
values at the points of intersection on the axes.

University of
Sunderland 69
Unit 3 – Linear Programming Business Modelling for Decision Making

This occurs when R = 60 and D = 50. Since we need to maximise £5R +

£8D, then we substitute the values for R = 60 and D = 50 into our profit
equation, so, 5(60) + 8(50) = 300 + 400 = £700 given the constraints (i.e.
the point of maximum profit).

Try the above example before you attempt the exercises.

Note the objective function was to maximise 5R + 8D subject to the

constraints of:

5R + 6D " 600
R + 2D " 160

And, of course, neither R nor D can be a negative value, so R, D # 0

Minimisation problems are usually set up as follows:

minimise ax1 + bx2 subject to :

cx1 + dx2 " e

fx1 + gx2 " h

(a to h inclusive are constants, x1, x2 are variables)

Notice how the inequality signs are reversed from those for the
maximisation problems.

And similarly, with minimisation problems, the feasible region changes

…

Linear Programming
The Graphical Method
D
Cotton The region (F) is called
100
the feasible region
80

F
From here, Labour
R
120 160

University of
70 Sunderland
Business Modelling for Decision Making Unit 3 – Linear Programming

Linear Programming
The Graphical Method
D
100
Cotton The region (F) is called
the feasible region
80
To here,

F
Labour
R
120 160

Linear Programming
The Graphical Method
D
100
Cotton The region (F) is called
the feasible region
80
F

Labour
R
120 160

And we attempt to minimise to the boundaries of the feasible region …

Linear Programming
The Graphical Method
Y

In this
direction!

One final point; the linear programming techniques above have related
to two constraints, i.e. two linear equations plus the profit (or cost)

University of
Sunderland 71
Unit 3 – Linear Programming Business Modelling for Decision Making

equation. If there are three or more constraints, the analysis is simply

extended; i.e. for three constraints, in addition to the profit or cost
equation, there are three linear equations to determine and, of course,
three lines to draw!

Worked Example
Remember, to solve Linear Programming examples by the graphical
method, all we need to do is to identify the (linear) constraints, calculate
their equations, plot them and use the equation of the ‘objective
function’ to identify the optimal solution!

Of course, we need to use the optimal solution to calculate the profit or

loss, then we need to explain and interpret our model.

Consider the following example:

A furniture factory makes two products: chairs and tables. The

products pass through three manufacturing stages i.e. woodworking,
assembly and finishing.

! The woodworking shop can make 24 chairs per hour or

12 tables per hour.

! The assembly shop can assemble 16 chairs per hour or 16

tables per hour.

! The finishing shop can finish 18 chairs per hour or 14

tables per hour.

! Workshops operate for 8 hours per day.

If the contribution to profit from each chair is £5 and from each table is
£4, determine by the graphical method, the number of chairs and tables
that should be produced to maximise profits.

Solution method:

Let chairs = x and tables = y.

Determine the objective function. In this case it is the maximisation of

profit given that the contribution to profit from each chair is £5 and from
each table is £4. (i.e. maximise 5x + 4y subject to ... [the constraints!])

Determine the number of constraints (in this case there are three -
woodworking, assembly and finishing), then calculate the equations for
each of the constraints.

University of
72 Sunderland
Business Modelling for Decision Making Unit 3 – Linear Programming

The woodworking constraint: 24 chairs (12 tables) per hour means that
it takes 1/24 (1/12) hours to make one chair (table) and there are 8 hours
available. Chairs (tables) are denoted by ‘x’ (‘y’). Hence:

x/24 + y/12 " 8

and we need to re-arrange the inequality so as to remove the fractions,

so we multiply both sides of the inequality by 24, we end up with

x + 2y " 192

This is the first of our three constraints!

We do likewise for the remaining two constraints:

The assembly constraint: 16 chairs (16 tables) per hour means that it
takes 1/16 (1/16) hours to assemble one chair (table) and there are 8
hours available. Chairs (tables) are denoted by ‘x’ (‘y’). Hence, we get

x/16 + y/16 " 8 (and multiplying both sides of the inequality by 16 (to
eliminate the fractions) ..

x + y " 128

This is the second of our three constraints!

The finishing constraint; 18 chairs (14 tables) per hour means that it
takes 1/18 (1/14) hours to finish one chair (table) and there are 8 hours
available. Chairs (tables) are denoted by ‘x’ (‘y’). Hence, we get

x/18 + y/14 " 8 (and multiplying both sides by (18 * 14) (to eliminate
the fractions) gives

14x + 18y " 2016 (and dividing both sides of the inequality by 2)

7x + 9y " 1008

This is the third of our three constraints!

After calculating the (three) constraints, the next step is to identify the
respective co-ordinates so we can plot the linear lines on a graph.

The woodworking constraint is x + 2y " 192.

Let x + 2y = 192 and when x = 0, (0 + 2y = 192) or 2y = 192 and y = 96.

When y = 0, (x + 2*0 = 192) so x = 192.

Plot these (x,y) co-ordinates (0,96) and (192,0) on the graph and connect
them with a linear line.

University of
Sunderland 73
Unit 3 – Linear Programming Business Modelling for Decision Making

y
Not to scale!

96 when x = 0, y = 96.

when y = 0,
x= 192
x
0 192

The assembly constraint is x + y " 128.

Let x + y = 128 and when x = 0, (1*0 + y = 128) y = 128. When y = 0, (x +

1*0 = 128) so x = 128.

Plot these (x,y) co-ordinates (0,128) and (128,0) on the graph and
connect them with a linear line.

y
Not to scale!

128 when x = 0, y = 128,

96

when y = 0,
X = 128

x
0 128 192

The finishing constraint is 7x + 9y " 1008.

Let 7x + 9y = 1008 and when x = 0, (7*0 + 9y = 1008) 9y = 1008 and y =

112.

When y = 0, (7x + 9*0 = 1008) so 7x = 1008 and x = 144.

Plot these (x,y) co-ordinates (0,112) and (144,0) on the graph and
connect them with a linear line.

University of
74 Sunderland
Business Modelling for Decision Making Unit 3 – Linear Programming

y
Not to scale!

128

112
when x = 0, y = 112
96

When y = 0,
x = 144

x
0 128 144 192

All constraints are now drawn, so we can consider the objective

function!

Our objective function is maximise profits subject to 5x + 4y.

Maximisation means we are working in the boundary of the axes and

the constraints!

y
Not to scale!

128 Remember..
112

96
This area (to the right of the
constraints) is for
MINIMISATION Problems!

x
0 128 144 192

Our objective function is maximise profits subject to 5x + 4y.

We need to set up an equation similar to the constraints where we can

equate the objective function to a specific number. There are no hard
and fast rules here. The selection of the specific number is arbitrary and
by chance. What we need to do, is to find co-ordinates on our objective
function such that the linear line we draw is not too small nor too large.

(Let x = 40 and let y = 50, (arbitrary values) then 5(40) + 4(50) = 200 + 200
= 400)

University of
Sunderland 75
Unit 3 – Linear Programming Business Modelling for Decision Making

So, let 5x + 4y = 400 (my arbitrary value), then when x = 0, y = 400/4, or y

= 100 and, when y = 0, x = 400/5 or x = 80. Hence my (x,y) co-ordinates
are (0,100) and (80,0). Plot these on the same graph and join them up
with a ‘broken’ linear line.

y
Not to scale!

128

112 when x = 0, y = 100.

100
96 when y = 0, x = 80.

‘Broken’ line of
objective function!
x
0 80 128 144 192

The next step is to physically ‘squeeze’ out all of the feasible region
bounded within the axes and the constraints (similar to what we did
earlier with the shirts example). We do this by ‘moving’ the linear line of
the objective function, parallel to itself within the feasible region,
‘squeezing out’ the feasible region, until we cannot squeeze out any
more. This point (where we are unable to squeeze out any more of the
feasible region) is known as the optimal solution.

y
Not to scale!

128

112
100
96

x
0 80 128 144 192

University of
76 Sunderland
Business Modelling for Decision Making Unit 3 – Linear Programming

y
Not to scale!

128

112
100
96

x
0 80 128 144 192

y
Not to scale!

128

112
It appears we are ‘squeezing’
100
96 out the feasible region
towards the point when y = 0
and x = 128!

x
0 80 128 144 192

y
Not to scale!

128

112
So the optimal solution occurs at
100
96 the point when y = 0 and x = 128!

x
0 80 128 144 192

University of
Sunderland 77
Unit 3 – Linear Programming Business Modelling for Decision Making

The optimal solution therefore, occurs when x = 128 and y = 0. Since ‘x’
(‘y’) represents chairs (tables) then given the constraints, the maximum
profit occurs when there are 128 chairs and no tables made! The
contribution to profit for chairs (tables) is £5 (£4) so our profit is:

(£5 * 128) + (£4 * 0) = £640!

ACTIVITY
What are the values of x and y which would maximise:

Profit = 5x + 3y

subject to

y " 400

x " 280

5x + 4y " 2000

2x + y " 640, and

x, y " 0 ?

Shade in the feasible region on your graph.

ACTIVITY FEEDBACK
Maximise 5x + 3y subject to:

1 y " 400

2 x " 280

3 5x + 4y " 2000

4 2x + y " 640

x, y " 0

University of
78 Sunderland
Business Modelling for Decision Making Unit 3 – Linear Programming

1 Let y = 400 (constant linear line at y = 400)

2 Let x = 280 (constant linear line at x = 280)

3 Let 5x + 4y = 2000, then

when x = 0, 4y = 2000 so y = 500

when y = 0, 5x = 2000 so x = 400

4 Let 2x + y = 640, then

when x = 0, y =640

when y = 0, 2x = 640 so x = 320

Objective

Let 5x + 3y = 600

then when x = 0, y = 200 and when y = 0, x = 120

Note that those constraints that limit the solution to the optimum point are
known as the ‘binding’ or ‘limiting’ constraints.

University of
Sunderland 79
Unit 3 – Linear Programming Business Modelling for Decision Making

ACTIVITY
Solve the following linear equation by the graphical method (minimisation)

What are the values of x and y which would minimise Cost = 12x + 42y subject
to

x + 2y " 3, x + 4y " 4, 3x + y " 3 and x, y " 0 ?

Shade in the feasible region on your graph and determine the value of the
minimum cost. If the cost function changed from 12x + 42y to 12x + 12y,
would the minimum cost remain at the same position on your graph as you
have initially determined? If not, calculate the new cost and explain the reasons
why the cost function has changed position.

ACTIVITY FEEDBACK

University of
80 Sunderland
Business Modelling for Decision Making Unit 3 – Linear Programming

If the objective function changes from 12x + 42y to 12x + 12y, the optimal
solution will move from x = 2.1, y = 0.5 to x = 0.6 and y = 1,2 (approx.) (see
dotted line)

Minimum cost is then 12(0.6) + 12(1.2) or 7.2 + 14.4 = 21.6

(let 12x + 12y = 24, then when x = 0, 12y = 24 so y =2 and when y = 0,

12x = 24 so x = 2)

x + 4y = 4

1 x + 2y = 3

2y = 1 so y = 0.5

when y = 0.5, x + 2y = 3

so x + 1 = 3

so x = 2

1 x + 2y = 3

(* 3)

3x + 6y = 9

3 3x + y = 3

so 5y = 6 and y = 1.2

when y = 1.2, x + 2(1.2) = 3

so x + 2.4 = 3

so x = 0.6

ACTIVITY
To maintain good health, an individual should consume a balanced diet. For the
purposes of this exercise, assume only three types of nutrients are required to
be consumed to attain a balanced diet. These are calcium, protein and vitamin
A. We also assume that the individual’s diet consists of two food items, i.e.
food 1 and food 2, whose price, content and minimum daily consumption levels
are depicted in the following table :

University of
Sunderland 81
Unit 3 – Linear Programming Business Modelling for Decision Making

Food 1 /lb Food 2 /lb Minimum

Requirements

Price £0.60 £1.00

Calcium/unit 10 4 20

Protein/unit 5 5 20

Vitamin A/unit 2 6 12

What is the combination of the two food items required which will satisfy the
daily requirements and minimise the cost of purchase ?

Hint : Let x = food 1 and y = food 2 and minimise Cost = 0.6x + y subject to :

10x + 4y " 20 (calcium constraint,

5x + 5y " 20 (protein constraint),

2x + 6y " 12 (vitamin A constraint) and

x, y " 0

(Shade in the feasible region on your graph).

Determine the shadow price of Protein and Vitamin A at the point of minimum
cost.

University of
82 Sunderland
Business Modelling for Decision Making Unit 3 – Linear Programming

ACTIVITY FEEDBACK
1. 10x + 4y = 20 When x = 0, y = 5 y

When y = 0, x = 2 5

2. 5x +5y = 20 When x = 0, y = 4 4

When y = 0, x = 4 3
Feasible
area

3. 2x + 6y = 12 When x = 0, y = 2 2
Optimal solution when
(approx) x = 3 and y = 1.
When y = 0, x = 6 1
Thus, minimum cost is
0.6(3) + 1(1) = £2.80

Let 0.6x + y = 3When x = 0, y = 3

0
1 2 3 4 5 6 x
When y = 0, x = 5

2 5x + 5y = 20

(* 2)

3 2x + 6y = 12

(* 5)

3 10x + 30y = 60

2 10x + 10y = 40

so 20y = 20 and y = 1

when y = 1, 2x + 6 = 12

or 2x = 6 so x = 3

University of
Sunderland 83
Unit 3 – Linear Programming Business Modelling for Decision Making

The following section looks at sensitivity analysis and how it is applied

to the linear programming method for solving business problems. The
first example considers the ‘shirt’ example as explained earlier.

REVIEW ACTIVITY
Use the ‘Sunderland Optimum Shirts’ example to calculate the following:

1. The shadow prices of Labour and Material (cotton),

2. Find the changes required in the slope of the objective function to

move the optimum point.

‘Sunderland Optimum Shirts’ (SOS)

‘Sunderland Optimum Shirts’ (SOS) is a company making high quality shirts and
are planning to make two distinct types (‘Regular’ and ‘Deluxe’). The
contribution to profits of the ‘Regular’ (‘Deluxe’) shirt is £5 (£8).

To manufacture the shirts, SOS have available to themselves 600m2 worth of

cotton. Regular (Deluxe) shirts require 5m2 (6m2) worth of cotton.
Additionally, Regular (Deluxe) shirts take 1 (2) hour(s) to manufacture. SOS
have 20 machinists who work 8 hours/day.

To maximise profits, how many shirts of each should be made? Also, what is
the maximum profit?

Solution equations:

maximise 5R + 8D subject to: 5R + 6D " 600 and R + 2D " 160 and of course,
neither R nor D can be a negative value, so R, D " 0

REVIEW ACTIVITY FEEDBACK

Solving the shadow price for Labour:

Assume that we have an increase in the labour constraint by one hour, i.e. total
hours of labour = 161 hours and hence, R + 2D = 161

R + 2D = 161 => *3 => 3R + 6D = 483

5R + 6D = 600 => => 5R + 6D = 600

and subtract. Hence, -2R = -117

University of
84 Sunderland
Business Modelling for Decision Making Unit 3 – Linear Programming

giving R = 58.5 and D = 51.25 (ignore decimal points for now)

We substitute the values of R = 58.5 and D = 51.25 into the objective function
of

5R + 8D i.e. 5(58.5) + 8(51.25) = £702.50 which gives us an increase in profit

of £2.50

(i.e. £2.50 profit for one additional hour!)

The £2.50 profit for one additional hour of labour is known as the Shadow Price
of labour, i.e. the change in the value of the objective function when the RHS of
the labour constraint is changed by one unit (is £2.50).

Solving the shadow price for cotton:

We now assume that we have an increase in the cotton material by one m2

i.e. cotton = 601m2 and hence, 5R + 6D = 601

We keep the labour constraint of R + 2D <= 160 constant and attempt to

solve both equations simultaneously:

R + 2D = 160 => *3 => 3R + 6D = 480

5R + 6D = 601 => => 5R + 6D = 601

and subtract. Hence, -2R = -121 giving R = 60.5 and D = 49.75

(ignore decimal points for now)

We substitute the values of R = 60.5 and D = 49.75 into the objective function
of

5R + 8D i.e. 5(60.5) + 8(49.75) = £700.50 which gives us an increase in profit

of £0.50

(i.e. £0.50 profit for one additional m2 of cotton!)

The £0.50 profit for one additional m2 of cotton is known as the Shadow Price
of cotton. i.e. the change in the value of the objective function when the RHS of
the cotton constraint is changed by one unit (is £0.50).

Changes in the slope of the objective function – movement occurs only from
the point at the optimal solution. In this case, the optimal solution occurs at R
= 50 and D = 60.

The slope of the objective function must equate in turn, to the slopes of the
binding constraints. The Labour constraint (as an equality) is R + 2D = 160
and we express this in terms of its equation of a linear line (which is 2D = 160 –
R). Dividing both sides of the equation by 2, gives D = 80 – 1/2R. Hence, the
‘slope’ of the labour constraint is -1/2. Similarly, the ‘slope’ of the objective

University of
Sunderland 85
Unit 3 – Linear Programming Business Modelling for Decision Making

function (the profit function) is -5/8 (from 5R + 8D = Profit and 8D = Profit –

5R (and dividing each side of the equation by 8 gives D = Profit/8 – 5/8R)).

y
Not to scale!

128

112
So the optimal solution occurs at
100
96 the point when y = 0 and x = 128!

x
0 80 128 144 192

To move from point B to point A, the slope of the objective function must
equate to the slope of the labour constraint (–1/2). The slope of the objective
function is –5/8 so to equate to –1/2, the profit on the Regular shirt must
decrease from £5 to £4 (i.e. 4/8 = 1/2) (Deluxe shirt profit remains constant)
OR, the profit on the Deluxe shirt must increase from £8 to £10 (i.e. 5/10 =
1/2) (Regular shirt profit remains constant).

Necessary percentage change in the profit of the Regular shirt is (5 – 4)/5*100

= 20%. Necessary percentage change in the profit of the Deluxe shirt is (10 –
8)/8*100 = 25%

To move from point B to point C, the slope of the objective function must
equate to the slope of the cotton constraint (which is –5/6). The slope of the
objective function is –5/8 so to equate to –5/6, the profit on the Regular shirt
must increase from £5 to £6.67 (Deluxe shirt profit remains constant) OR, the
profit on the Deluxe shirt must decrease from £8 to £6 (Regular shirt profit
remains constant).

Necessary percentage change in the profit of the Regular shirt is (5 –

6.67)/5*100 = 33.33%. Necessary percentage change in the profit of the
Deluxe shirt is (6 – 8)/8*100 = 25% (ignore negative values on the
percentages).

Summary
Linear programming is an important tool for maximising profit in a
variety of situations. This unit has introduced both a graphical and an

University of
86 Sunderland
Business Modelling for Decision Making Unit 3 – Linear Programming

algebraic method for determining solutions. These are based on the

initial problem constraints and the objective function.

The solution analysis allows us to understand the way in which the

technique models a real-life application in the search for an optimum
solution.

Sensitivity analysis helps us to understand how the problem will react

to small variations in the variables, and hence the outcome of the
objective function, i.e. contribution to profit or cost.

University of
Sunderland 87
Unit 4

Optimisation

Introduction
Our earlier analysis into the use of numerical methods to assist us to
make decisions, concerned the equation of the linear line

y = f(x) = a + bx

Now we are going to consider planning our decision making techniques

by the use of Non Linear Equations

In this unit we are going to look at:

! A recap of linear equations

! Non linear equations - extremum
! Calculating graphs of non linear equations
! A note on the constant term of NL equations
! Examples of non linear equations
! The demand function

Linear Equations
In our earlier years of study, we should have encountered the linear
regression model, i.e. the dependent variable (y) is assigned to be a
function of the independent variable (x) and we expressed the
relationship as y = f(x) and in general (for linear equations),

y = a + bx (the equation of the linear line of best fit)

where a and b are constants.

The constant ‘a’ represents the point where the line of best fit crosses the
vertical axis (the ‘y’ axis) and, the constant ‘b’ represents the gradient
(slope) of the line of best fit..

University of
Sunderland 89
Unit 4 – Optimisation Business Modelling for Decision Making

If the value of ‘b’ is > 0 (positive), the line of best fit slopes upwards. If
the value of ‘b’ is < 0 (negative), the line of best fit slopes downwards
and, if the value of ‘b’ is = 0, the line of best fit is horizontal and parallel
to the horizontal axis (the ‘x’ axis).

Optimisation - Extremum

y y = a + bx

Value of b > 0, hence

a line slopes upwards!

Optimisation - Extremum

y y = a - bx
a

Value of b < 0, hence

line slopes downwards!

Optimisation - Extremum

y y = a + 0x = a

a
Value of b = 0, hence
line is horizontal!

University of
90 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

So that was the linear line of best fit and, we use that line to ‘forecast’
changes in the dependent variable (‘y’), following changes in the
independent variable (‘x’).

Now, we are going to expand our knowledge of the linear system into a
non linear system.

Non Linear Equations

We will now use and apply our knowledge of the linear line to help us to
resolve problems relating to non linear systems. These are systems with
lines of best fit, usually depicted with ‘turning points’ - or maximum
and minimum points. These turning points are called - ‘Extremum’

Optimisation - Extremum

y y = f(x)
Maximum extrema

Minimum extrema
x

Examples of maximum extrema could be ‘Profit Maximisation’,

‘Revenue Maximisation’, ‘Sales Maximisation’, etc. The ‘best’ example
of minimum extrema is ‘Cost Minimisation’!

Optimisation - Extremum

y y = f(x)
Point of Maximum Profit

x
Output of ‘x’ to Maximise Profit

University of
Sunderland 91
Unit 4 – Optimisation Business Modelling for Decision Making

Optimisation - Extremum

y y = f(x)

Point of Minimum Cost

x
Output of ‘x’ to Minimise Cost

What we are attempting to do is to find values of our independent

variable (the ‘x’ variable) at either a maximum or at a minimum turning
point! (extremum)

In simple terms, suppose our variable ‘x’ represents the number of

commodities sold. Then if we can identify these extremum, we can plan
to restrict the supply of our commodities sold, to either maximise (e.g.
profits) or to minimise (e.g. costs).

Optimisation - Extremum

y y = f(x)

Maximum

Optimisation - Extremum

y y = f(x)

Minimum

University of
92 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

Non linear graphs of the form y = f(x) i.e. follow an equation of the form
2
y = cx + bx + a

This type of equation is also known as a ‘quadratic equation’. Notice

that our linear line of y = f(x) = a + bx is represented in the equation,
2
i.e. y = f(x) = cx + bx + a.

Hence, for non linear analysis, we need only consider extending our
previous knowledge by cx2 i.e. y = f(x) = cx2 + bx + a.

Textbooks typically follow the format of y = f(x) = ax2 + bx + c

and since the coefficients a, b and c are constant values, then we will
revise our formula of y = cx2 + bx + a, to follow the textbook format.

When we deal with a MINIMUM extrema, the general equation of the

non linear line will take the form of y = ax2 + bx + c, as we have already
stated!

Optimisation - Extremum

y y = f(x)

y = ax2 + bx + c

Minimum

When we deal with a MAXIMUM extrema, the general equation of the

non linear line will take the form of y = -ax2 + bx + c, (i.e. notice the
minus sign preceding the ax2 term!)

University of
Sunderland 93
Unit 4 – Optimisation Business Modelling for Decision Making

Optimisation - Extremum

y y = f(x)

y = -ax2 + bx + c

Maximum

Remember, when we deal with a MINIMUM extrema, the general

equation of the non linear line will take the form of y = ax2 + bx + c, and,
when we deal with a MAXIMUM extrema, the general equation of the
non linear line will take the form of y = –ax2 + bx + c (i.e. notice the minus
sign preceding the ax2 term!)

Calculating Graphs of Non Linear Equations

Suppose we let y = f(x) = 3x2 + 2x + 1

(i.e. let y = f(x) = ax2 + bx + c = 3x2 + 2x + 1)

Notice the value for ‘a’ in y = f(x) = ax2 + bx + c = 3x2 + 2x + 1 is 3 (or +3).

The positive value for ‘3’ in our equation y = f(x) = 3x2 + 2x + 1 indicates
our equation represents a graph which will display the properties of a
MINIMUM extrema, i.e. we are looking for a MINIMUM point!

Let us set up a simple model to calculate the co-ordinates to help us to

plot our non linear line. We let the value for ‘x’ take on arbitrary values
(e.g. from -3 to +3 inclusive, incrementing by unity) and in turn, we
substitute these values for ‘x’ in our equation y = f(x) = 3x2 + 2x + 1

x: -3 -2 -1 0 1 2 3

y: 22 9 2 1 6 17 34

and, for example,

when y = f(x) = 3x2 + 2x + 1 then when x = -3, y = 3(-3)2 + 2(-3) + 1

giving y = 3(9) - 6 + 1

= 27 - 6 + 1 = 22

University of
94 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

and we do similar with the other values of ‘x’, then we can use these
co-ordinates to plot the graph of our non linear line.

Optimisation - Extremum
Graph of y = 3x 2 + 2x + 1
40
35
30 Notice the minimum point occurs
somewhere between x = 0 and x = -1
25
y values

20
15
10
5
0
-3 -2 -1 0 1 2 3
x values

Notice the minimum point occurs somewhere between x = 0 and x = -1

(actually occurs when x = -1/3 - but will be explained later.)

Clearly, the output from a company cannot be negative (<0) but, we can
use this concept to identify extremum when conditions exist where our
extrema dictate values of x " 0.

Before we look at a business example, let us consider the role of the

constant ‘c’ in y = ax2 + bx + c

The constant term of non linear equations

The constant ‘c’ is simply a ‘shifting’ term i.e. its role is to ‘shift’ the non
linear line either up or down the vertical axis. Consider our equation:
y = f(x) = ax2 + bx + c and we used y = f(x) = 3x2 + 2x + 1

With this equation, we calculated the required co-ordinates to help us to

plot our non linear line i.e.

x: -3 -2 -1 0 1 2 3

y: 22 9 2 1 6 17 34

and, for example,

when y = f(x) = 3x2 + 2x + 1 then when x = -3, y = 3(-3)2 + 2(-3) + 1

giving y = 3(9) - 6 + 1

= 27 - 6 + 1 = 22

University of
Sunderland 95
Unit 4 – Optimisation Business Modelling for Decision Making

and we did similar with the other values of ‘x’, then we used these
co-ordinates to plot the graph of our non linear line.

Optimisation - Extremum
Graph of y = 3x 2 + 2x + 1
40
35
30
y values 25
20
15
10
5
0
-3 -2 -1 0 1 2 3
x values

Consider our equation: y = f(x) = ax2 + bx + c and let the value for ‘c’
change from 1 to 5. Hence, y = f(x) = 3x2 + 2x + 5

We calculate the ‘new’ co-ordinates to help us to plot our ‘new’ non

linear line i.e.

x: -3 -2 -1 0 1 2 3

y: 26 13 6 5 10 21 38

and, for example, when y = f(x) = 3x2 + 2x + 5 then when x = -3, y = 3(-3)2
+ 2(-3) + 5

giving y = 3(9) - 6 + 5

= 27 - 6 + 5 = 26

and by doing similar with the other values of ‘x’, we use these ‘new’
co-ordinates to plot our ‘new’ graph of our ‘new’ non linear line.

University of
96 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

Optimisation - Extremum
Graph of y = 3x 2 + 2x + 5
40
35
30
25
y values

20
15
10
5
0
-3 -2 -1 0 1 2 3
x values

and if we introduce the original non linear line (when y = f(x) = 3x2 + 2x
+1) then notice how our ‘new’ non linear line has ‘shifted’ up the vertical
axis, i.e. the ‘new’ non linear line is identical in shape to the original non
linear line

Optimisation - Extremum
Graph of y = 3x 2 + 2x + 5
40
35
30
25
y values

20
15
10
5
0
-3 -2 -1 0 1 2 3
x values

Note also, the value for ‘x’ at the minimum point has not changed. So
altering the value of the constant term ‘c’, simply moves the line up or
down the vertical axis.

Examples of non linear equations

MAXIMUM points are found very much in the same way. If we
consider our equation as
2
y = f(x) = -3x + 2x + 1

Notice the minus sign before the 3x2 term. The minus sign indicates our
equation represents a graph which will display the properties of a
MAXIMUM extrema, i.e. we are looking for a MAXIMUM point!

University of
Sunderland 97
Unit 4 – Optimisation Business Modelling for Decision Making

As before, we set up a simple table:

x: -3 -2 -1 0 1 2 3

y: -32 -15 -4 1 0 -7 -20

and when y = f(x) = -3x2 + 2x + 1 then when x = -3, y = -3(-3)2 + 2(-3) + 1

giving y = -3(9) - 6 + 1

= -27 - 6 + 1 = -32

and we do similar with the other values of ‘x’, then we can use these
co-ordinates to plot the graph of our non linear line.

Optimisation - Extremum
Graph of y = -3x 2 + 2x + 1
5
0
-5 -3 -2 -1 0 1 2 3
-10
y values

-15 Notice the maximum point occurs

-20 somewhere between x = 0 and x = +1
-25
-30
-35

x values

(Note the actual value for x occurs when x = +1/3 - but will be explained
later!)

ACTIVITY
Let us look at a business example:

A medium sized company produces a quantity of 9 units (q) per week. The
company’s total fixed cost (FC) = £10,000 and its total variable cost (VC) =
(-1,000q + 200q2).

The demand for these units are given by the equation (9,000 - 800q).

(Notice how the demand equation represents a linear line of the form

y = f(x) = a + bx)

University of
98 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

Given this information, find the company profit. We assume the units of output
represent a quantity of goods produced, e.g. one unit might represent 20,000
chocolate bars.

ACTIVITY FEEDBACK
To find the company’s profit equation, we need to find its total revenue and
total cost equations and difference them!

The company’s total fixed cost (FC) = £10,000 and its total variable cost (VC)
= (-1,000q + 200q2). To find the company’s total cost (TC), we equate the
Fixed Cost and the Variable Cost

TC = FC + VC

TC = (10,000) + (-1,000q + 200q2)

i.e. TC = 10,000 - 1,000q + 200q2

(Notice how the TC equation is a quadratic.)

We now need to find the company’s revenue function. Revenue is calculated

by the product of the quantity of commodities sold (quantity demanded) and
commodity price. i.e. p x q

e.g. If I can sell 10 chocolate bars at 50p each, then my revenue is 50p x 10 or
£5.00

The demand for the company’s units are given by the equation (9,000 - 800q).

Since our demand equation represents the number of units that can be sold at
various prices, then our demand equation represents PRICE.

To find our revenue function, we simply need to multiply the demand equation
by output (q).

Hence, Revenue = p x q or

Total Revenue (TR) = (9,000 - 800q) x q

and TR = (9,000q - 800q2)

We can now find our profit equation!

Profit = $= TR - TC

University of
Sunderland 99
Unit 4 – Optimisation Business Modelling for Decision Making

$ = (9,000q - 800q2) - (10,000 - 1,000q + 200q2)

and removing the brackets ..

$ = 9,000q - 800q2 - 10,000 + 1,000q - 100q2

(Notice how the signs of the TC equation changed when we removed the
brackets.)

i.e. the minus sign outside the brackets of the TC effectively changed the sign of
every term within the brackets when the brackets were removed!

$ = (9,000q - 800q2) - (10,000 - 1,000q + 200q2)

and removing the brackets ..

$ = 9,000q - 800q2 - 10,000 + 1,000q - 200q2

We now tidy up the terms ..

$ = 9,000q - 800q2 - 10,000 + 1,000q - 200q2

$ = 10,000q - 800q2 - 10,000 - 200q2

$ = 10,000q - 1000q2 - 10,000

and $ = 10,000q - 1000q2 - 10,000 is our profit equation! Notice how it

represents a quadratic equation with a negative sign in front of the ‘squared’
term.

Such an indication tells us that we are looking at a MAXIMUM point.

Let us plot a graph of our profit equation!

Optimisation - Extremum
Graph of Profit
15000

10000
Notice the maximum point occurs
5000 somewhere between q = 4 and q = 6

0
0 1 2 3 4 5 6 7 8 9
-5000

-10000
quant it y

(actually occurs when q = 5)

University of
100 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

The Demand Function

Consider the company’s demand function:

Demand = 9,000 - 800q

If the price set for the commodity is zero (q = 0) then demand is

maximised at 9,000.

Similarly, if we minimise demand to zero, then q = 11.25

i.e. 9,000 = 800q and q = 9,000/800

giving q = 11.25

So when q = 0, demand = 9,000 and when q = 11.25, demand = 0

i.e. two sets of co-ordinates which represent a linear line (remember the
linear programming exercises?)

Our demand equation (D = 9,000 - 800q) represents the line of best fit of
a series of data observations.

Optimisation - Extremum
Graph of Demand = 9,000 - 800q
9000
8000
7000
6000
Demand

5000
4000
3000
2000
1000
0
-1000 0 1 2 3 4 5 6 7 8 9 10 11 12

No. of Units (q)

Remember, our line represents our demand equation D = 9,000 - 800q

and D = 9,000 - 800q represents a linear equation of the form y = a + bx

This graph assumes all data observations lie on a linear line! So what
happens if the data do not form a linear line?

Well, if a given set of data observations do not lie on a linear line then,
we can use the data via our knowledge of regression analysis, to
determine the equation of the line of best fit for the demand. i.e. when y
= f(x) = a + bx.

University of
Sunderland 101
Unit 4 – Optimisation Business Modelling for Decision Making

Consider the following data ..

Optimisation - Extremum

q Dem and
(x) (y)
0 8600
1 8100
2 6900
3 7600
4 6400
5 5100
6 4000
7 3000
8 2500

Optimisation - Extremum
Scatte r plot of De m and for q
Demand

Clearly, the data observations

do not lie on a linear line !

No. of Units (q)

Hence, we need to calculate the regression line of best fit via our
knowledge of regression analysis. The equation determined by
regression analysis can be used to simulate our demand.

The required formulae ..

University of
102 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

Optimisation - Extremum

%XY ( ( X)( %Y)

&' ) ' Y ( &X
( %X 2 ) ( ( X)( %X)
Slope, Intercept

Given this formulae, we can see that we need to undertake some

manipulation of the data to help us to determine the regression
equation. Specifically, we need to identify the following:
2
%XY, %Y, %X and %X

q Dem and
(x) (y) xy x2
0 8600 0 0
1 8100 8100 1
2 6900 13800 4
3 7600 22800 9
4 6400 25600 16
5 5100 25500 25
6 4000 24000 36
7 3000 21000 49
8 2500 20000 64
9 1800 16200 81
Sum s 45 54000 177000 285

Using the respective sums in the formula for the slope term (b) ..

and b = (-66,000)/82.5 = -800

Using the respective sums in the formula for the intercept term (a) ..

and a = 5,400 + 3,600 = 9,000

Hence, demand = 9,000 - 800q

Notice our demand equation of demand = 9,000-800q is the same

demand equation as we have previously used. In this example, this was
designed to be so. We can have a perfectly linear demand equation or

University of
Sunderland 103
Unit 4 – Optimisation Business Modelling for Decision Making

we can determine the regression equation for demand given our data
observations.

Optimisation - The Derivative

In this section, all you will need to remember is:

1. Why the derivative is assigned to zero, dv/dx = 0

2. How to differentiate and calculate values for the independent
variable.

Earlier in this unit, we used and applied our knowledge of the linear line
to help us to resolve problems relating to non linear systems. If you
remember, non linear systems are lines of best fit, usually depicted with
‘turning points’ - or maximum and minimum points. These turning
points, we called - ‘Extremum’

We also said that examples of maximum extrema could be ‘Profit

Maximisation’, ‘Revenue Maximisation’, ‘Sales Maximisation’, etc. And
we also said that the ‘best’ example of minimum extrema is ‘Cost
Minimisation’.

What we were attempting to do with our non linear lines, was to find
values of our independent variable (the ‘x’ variable) at either a
maximum or at a minimum turning point (extremum).

If our variable ‘x’ represented the number of commodities sold. Then by

identifying the extremum, we could plan to restrict the supply of our
commodities sold, to either maximise (e.g. profits) or to minimise (e.g.
costs).

We learnt that non linear graphs of the form y = f(x)

i.e.

Optimisation - The Derivative

y y = f(x)
Maximum
were represented by an
equation of the form
y = -ax2 + bx + c

x
Remember the ‘-’ sign for a ‘Maximum’

University of
104 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

We also learnt that non linear graphs of the form y = f(x)

i.e.

Optimisation - The Derivative

y y = f(x) Minimum
were represented
by an equation of
the form
y = ax2 + bx + c

x
No ‘-’ sign for a ‘Minimum’!

Suppose we let y = f(x) = 3x2 + 2x + 1 (i.e. let y = f(x) = ax2 + bx + c = 3x2 +

2x + 1)

Notice the value of the ‘sign’ for ‘a’ in y = f(x) = ax2 + bx + c = 3x2 + 2x + 1

is 3 (or +3).

And if you remember, the positive value for the ‘3x2’ (i.e. the x2) in our
equation of y = f(x) = 3x2 + 2x + 1, indicated our equation represented a
graph which displayed the properties of a MINIMUM extrema, i.e. we
were looking for a MINIMUM point.

Optimisation - The Derivative

y y = f(x)
y = 3x2 + 2x + 1

Minimum

We then set up a simple model to calculate the co-ordinates to help us to

plot our non linear line, i.e. we let the value for ‘x’ take on arbitrary
values (e.g. from -3 to +3 inclusive, incrementing by unity) and in turn,
we substituted these values for ‘x’ in our equation y = f(x) = 3x2 + 2x + 1

University of
Sunderland 105
Unit 4 – Optimisation Business Modelling for Decision Making

x: -3 -2 -1 0 1 2 3

y: 22 9 2 1 6 17 34

and when y = f(x) = 3x2 + 2x + 1

so when x = -3, y = 3(-3)2 + 2(-3) + 1

giving y = 3(9) - 6 + 1 = 27 - 6 + 1 = 22

We use these co-ordinates to plot the graph of our non linear line.

We noticed the minimum point occurred somewhere between x = 0 and

x = -1, and, we said this value was actually when x = -1/3

Optimisation - The Derivative

Graph of y = 3x 2 + 2x + 1
40
35
30 We noticed the minimum point
occurred somewhere between x = 0
25
y values

and x = -1, and, we said this value

20 was actually when x = -1/3
15
10
5
0
-3 -2 -1 0 1 2 3
x values

The Derivative
Before we prove x = -1/3 in the previous equation, we need to forge an
understanding of the concepts of the derivative. Note that the
derivative is based upon the Differential Calculus method, invented by
Sir Isaac Newton and the German mathematician Wilhelm Leibniz circa
1675 AD.

The derivative is based upon our knowledge of both a linear line and a
non linear line.

We use the concepts of touching a non linear line with a linear line. This
is known as a ‘tangent’.

University of
106 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

Optimisation - The Derivative

y y = f(x)

Optimisation - The Derivative

y y = f(x)

The point of contact is

known as a TANGENT!

Notice that the tangent can form part of a right angled triangle and we
interpret this triangle in terms of its ‘y’ and ‘x’ values.

Optimisation - The Derivative

y y = f(x)

Notice that the tangent

can form part of a right
angled triangle!
x

University of
Sunderland 107
Unit 4 – Optimisation Business Modelling for Decision Making

Optimisation - The Derivative

y y = f(x)

x1 x2 x

Movement from x1 to x2 is known as the ‘change in x’ (dx)

Optimisation - The Derivative

y y = f(x)

y2

y1 *x
x1 x2 x

Movement from y1 to y2 is known as the ‘change in y’ (dy)

KEY POINT
Note that the slope of our tangent is determined by the ratio of dy/dx

Dy
' the slope of our tangent
Dx

Note also that if we reduce the size of the changes in our ‘y’ and ‘x’ variables
(equally) the slope of the tangent remains constant (i.e. the same!) and a
continual reduction in the changes in ‘y’ and ‘x’ eventually .. .. means our
triangle becomes small enough to be a ‘point’ on the curve.

University of
108 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

Optimisation - The Derivative

y y = f(x)

.. the slope of the

*y tangent remains
*x constant (i.e. the same!)

This tells us that we can find a point on a non linear curve, simply by
considering the relationship between the change in our ‘y’ and our ‘x’
variables. A small note here (to remember for later); when the change in
our ‘y’ and ‘x’ variables become small enough to become a point on our
non linear curve,

Dy
' the slope of our tangent
Dx
= the Derivative

Optimisation - The Derivative

Consider the role of the relationship between the

change in the ‘y’ variable and the change in
the ‘x’ variable!

*y
= slope of our tangent
*y *x

*x

University of
Sunderland 109
Unit 4 – Optimisation Business Modelling for Decision Making

Optimisation - The Derivative

If the change in our ‘y’ variable is identical to the

change in our ‘x’ variable, then the slope of
our tangent will lie at 45 degrees (45o) ..

*y
= slope of our tangent
*y *x

*x

Optimisation - The Derivative

If the change in our ‘y’ variable is identical to the

change in our ‘x’ variable, then the slope of
our tangent will lie at 45 degrees (45o) ..

*y 1 (unit)
+,o *y = 45o = =1
*x 1 (unit)
*x

The point of contact of the tangent to the non linear curve is not at a
maximum point!

Hence, the ratio of dy/dx = 1 is not of any use.

University of
110 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

Optimisation - The Derivative

Consider now, the position the tangent touches

the non linear curve when ..
*y 0.5
= = 0.5
*x 1
*y = 0.5
*x = 1

The point of contact of the tangent to the non linear curve is not at a
maximum point.

Hence, the ratio of dy/dx = 0.5 is not of any use.

Optimisation - The Derivative

The point of contact of the tangent to the non

linear curve is not at a maximum point!
Hence, the ratio of *y/*x = 0.5 is not of any use!

Consider now, the position the tangent touches the non linear curve
when

dy/dx = 0.25/1 = 0.25

University of
Sunderland 111
Unit 4 – Optimisation Business Modelling for Decision Making

Optimisation - The Derivative

Consider now, the position the tangent touches

the non linear curve when ..
*y 0.25
= = 0.25
*x 1
*y = 0.25
*x = 1

The point of contact of the tangent to the non linear curve is not at a
maximum point.

Hence, the ratio of dy/dx = 0.25 is not of any use.

Consider now, the position the tangent touches the non linear curve
when

dy/dx = 0.1/1 = 0.1

Optimisation - The Derivative

Consider now, the position the tangent touches

the non linear curve when ..
*y 0.1
= = 0.1
*x 1
*y = 0.1
*x = 1

The point of contact of the tangent to the non linear curve is not at a
maximum point.

Hence, the ratio of dy/dx = 0.1 is not of any use.

Finally, if we let the value for *y = 0, the position the tangent touches the
non linear curve occurs when

dy/dx = 0.0/1 = 0.0

University of
112 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

Perfect! - the point of contact of the tangent to the non linear curve is at
the maximum point.

Hence, the ratio of dy/dx = 0 is very useful.

We have identified the maximum point of our non linear curve simply
by permitting the ratio of the change in the ‘y’ variable to the change in
the ‘x’ variable to equal zero

dy/dx = 0.0/1 = 0.0

We can use the same argument to find MINIMUM Points.

Not at a minimum point so dy/dx = 1 is no good.

dy/dx = 1/1 = 1

Not at a minimum point so dy/dx = 0.5 is no good.

dy/dx = 0.5/1 = 0.5

Not at a minimum point so dy/dx = 0.25 is no good.

dy/dx = 0.25/1 = 0.25

At a minimum point so dy/dx = 0 is perfect.

If you remember, when the change in our ‘y’ and ‘x’ variables become
small enough to become a point on our non linear curve then, slope of
tangent = the derivative.

So, we are interested in a maximum point .. dy/dx = 0.0/1 = 0.0

or, we are interested in a minimum point .. dy/dx = 0.0/1 = 0.0

Hence, when dy/dx = 0, we have a turning point on our non linear line.

OK so we know that the slope of the tangent to our non linear line is zero
(0) at either a maximum or a minimum. So how does that help us to
identify the value for ‘x’ when we have a maximum or a minimum?

University of
Sunderland 113
Unit 4 – Optimisation Business Modelling for Decision Making

Optimisation - The Derivative

Graph of y = -3x 2 + 2x + 1
5
0
-5 -3 -2 -1 0 1 2 3
-10

y values
-15
-20
-25
-30
-35

x values

Easy! - we simply differentiate the equation with respect to ‘x’ (the

independent variable).

Differentiation
Differentiation consists of a number of rules of which some are fairly
complex. We are only interested in the simplest of all of the rules.

Consider our equation of y = f(x) = -3x2 + 2x + 1

we differentiate this equation and assign it to the value of the derivative

when the derivative is equal to zero (i.e. to identify the value for ‘x’ at
either a maximum point or at a minimum point).

When y = f(x) = -3x2 + 2x + 1, then dy/dx = 0, and therefore = -6x + 2 or

x = 1/3.

Optimisation - The Derivative

Graph of y = -3x 2 + 2x + 1
5
0
-5 -3 -2 -1 0 1 2 3
-10
y values

-15 At the maximum point, the value for ‘x’

-20 occurs when x = 1/3
-25
-30
-35

x values

University of
114 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

Basic Rules of Differentiation

Optimisation - The Derivative

If y = f(x) = axn then ..

dy
= naxn-1
dx
e.g. if y = f(x) = 3x2 then
dy
= 2*3x2-1 = 6x1 = 6x
dx

Another example ..

Optimisation - The Derivative

If y = f(x) = axn then ..

dy
= naxn-1
dx
e.g. if y = f(x) = 2x (= 2x1) then
dy
= 1*2x1-1 = 2x0 = 2*1
dx
(since x0 = 1 by definition!)

Another example ..

Optimisation - The Derivative

If y = f(x) = -axn then ..

dy
= n(-a)xn-1
dx
e.g. if y = f(x) = -3x2 then
dy
= 2*(-3)x2-1 = -6x1 = -6x
dx

University of
Sunderland 115
Unit 4 – Optimisation Business Modelling for Decision Making

Final example ..

Optimisation - The Derivative

If y = f(x) = axn then ..

dy
= naxn-1
dx
e.g. if y = f(x) = 5 (a constant) then
dy
= 0 (since there is no ‘x’ term!)
dx

Optimisation - The Derivative

Basic rules of differentiation ..

dy
If y = f(x) = axn then .. = naxn-1
dx
dy
If y = f(x) = -axn then .. = -naxn-1
dx
dy
If y = f(x) = a constant, then .. = 0
dx

Optimisation - The Derivative

A helpful rule for differentiation ..
dy
If y = f(x) = axn then .. = naxn-1
dx
e.g. suppose y = f(x) = 3x2 then ..
‘a’ = 3, ‘n’ = 2 and ‘n-1’ = 2-1 = 1
(write these values down before you attempt to
differentiate!, then substitute i.e.)
dy
= 2*3x2-1
dx

University of
116 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

Another helpful rule for differentiation ..

n
if y = f(x) = ±ax ± bx ± c

then since the axn term is separated from the bx term and the constant
(‘c’) term by either an addition sign or a minus sign, then we can
differentiate each term in turn - independent of the others, i.e.
differentiate the axn term first, then differentiate the bx term and finally
differentiate the constant (‘c’) term.

So, returning to our example;

Optimisation - The Derivative

Since when y = f(x) = -3x2 + 2x + 1, then ..
dy
= 0 (i.e. the value of the slope of
dx
the tangent is zero!
and differentiating the function ..
-6x + 2 = 0 (value of the slope of the tangent)
(Add 6x to both sides of the equality), hence 6x = 2
and 6x = 2, so x = 2/6 or 1/3

Finally, our example at the beginning of this session - i.e. when y = f(x) =
3x2 + 2x + 1, we said that the actual value of ‘x’ at the turning point was x
= -1/3 and would be explained later. So, we find our MINIMUM point
(no minus sign in front of the ‘x2’ term!), by differentiating our equation
and equating the result to the value of the slope of the tangent at a
minimum point (i.e. zero).

University of
Sunderland 117
Unit 4 – Optimisation Business Modelling for Decision Making

Optimisation - The Derivative

Since when y = f(x) = 3x2 + 2x + 1, then ..
dy
= 0 (i.e. the value of the slope of
dx
the tangent is zero!
and differentiating the function ..
6x + 2 = 0 (value of the slope of the tangent)
(Subtract 2 from both sides of the equality), hence
6x = -2 and x = -2/6 or x = -1/3

So there you are, all you will need to remember is:

1. Why the derivative is assigned to zero dy/dx = 0

2. How to differentiate and calculate values for the independent
variable.

Now try the following questions:

University of
118 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

ACTIVITY
Question 1

Consider the following two quadratic equations:

Y1 = f(x) = -3x2 – 2x – 2 (a); Y2 = f(x) = 4x2 + 3x + 1 (b);

1. Identify (with reasons) the extremum condition that each equation is

portraying (i.e. do we have a maximum or a minimum condition?).

2. Draw a graph of each of these two functions across the range of x

values from x = -3 to x = +3, incrementing each step by unity. (i.e. -3,
-2, -1 …)

From your graphs, identify the value of x at their respective extremum

positions.

3. Create a third equation (Y3) by adding equation (a) to equation (b).

Identify the extremum condition.

4. Create a fourth equation (Y4) by subtracting equation (a) from

equation (b). Identify (with reasons) the extremum condition.

Consider the following two equations:

-6x – 2 = 0 (c); 8x + 3 = 0 (d)

5. Compare equation (a) with equation (c). Then compare equation (b)
with equation (d). There is a similar type of pattern with both sets of
respective equations (i.e. equations (a) & (c) and equations (b) & (d)).
Can you (with reasons), identify the pattern?

6. Solve equation (c) to find the value for x. Solve equation (d) to find
the value for x. Compare each of these two values with the respective
values you determined from your respective graphs in section (2)
above. Are the respective values identical? If not, do you think the
respective values should be identical? – Explain.

University of
Sunderland 119
Unit 4 – Optimisation Business Modelling for Decision Making

ACTIVITY FEEDBACK

University of
120 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

ACTIVITY
Question 2

Consider the following equations:

Y1 = 4x2 + 3x + 2; Y2 = 4x2 - 3x + 2;

Y3 = -4x2 + 3x + 2; Y4 = -4x2 - 3x + 2;

1. Identify (with reasons) the extremum condition that the above

equations are portraying (i.e. do we have maximum or minimum
conditions?).

2. Sketch a graph of any one of the above four equations and from your
graph, determine the value for ‘x’ at the respective turning point.

3. Differentiate all four the equations with respect to ‘x’. You may
assume dy/dx = 0 and solve for x in all four differentiated equations.

4. Identify the equation you sketched in your graph as in section 2 above

and, compare the value you determined for ‘x’ at the turning point,
with the respective differentiated equation you calculated in section 3.
How do they compare?

University of
Sunderland 121
Unit 4 – Optimisation Business Modelling for Decision Making

ACTIVITY FEEDBACK

2
Graph of y1 = 4x + 3x + 2
50

45

40

35
y1 values

30

25

20

15

10

0
-4 -3 -2 -1 0 1 2 3 4
x values

2
Graph of y2 = 4x - 3x + 2
y2 values

x values

University of
122 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

2
Graph of y3 = -4x + 3x + 2

y3 values

x values

2
Graph of y4 = -4x - 3x + 2
y4 values

x values

ACTIVITY
Question 3

A small company produces a quantity (q) of 7 units per week. The company’s
total fixed cost (FC) = £50 and its total variable cost (VC) is given by the
equation (-10q + 10q2). The demand for these units is given by the equation
(240 - 15q). Find the equation for the company profit. Hint, Profit is Total
Revenue – Total Cost. Total Revenue is Demand times ‘q’.

Draw a graph of profit, across the range of ‘q’ from q = 0, 1, 2 .. 7. (Hint you
will need to set up a simple table similar to question (1.2), ranging from q = 0 to
q = 7. Then you will need to substitute the successive values for ‘q’ into your
profit equation and calculate the profit for each value of ‘q’. These values are
then co-ordinates – simply plot them on your graph). From your graph, find
the value of ‘q’ which maximises the profit.

University of
Sunderland 123
Unit 4 – Optimisation Business Modelling for Decision Making

On the same graph as your profit, draw a graph of demand and, of cost.
Explain what is happening with your graphs!

Consider your profit equation. Set up an equation for ‘q’ following the same
pattern as you identified in exercise (1) part 5 and solve for q. Explain (with
reasons) your findings.

ACTIVITY FEEDBACK

University of
124 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

ACTIVITY
Question 4

The following data relates to a small to medium sized company;

q Demand

(x) (y)

0.5 5500

1 4000

1.5 4000

2 4000

2.5 2500

3 2500

3.5 1000

4 2500

4.5 1000

5 500

University of
Sunderland 125
Unit 4 – Optimisation Business Modelling for Decision Making

Use regression analysis to calculate the company’s demand equation, i.e.

Demand = ) + &q.

The company produces a quantity (q) of 7 units per week and its total fixed
cost (FC) is 500 and its total variable cost (VC) is given by the equation (-500q
+ 100q2). If you have not already calculated this, then find the equation for the
company profit and ensure you are able to draw a graph of profit across the
range of ‘q’ from q = 0 to q = 5..

With your total cost equation, use the concepts of differentiation to solve for
q when cost is minimised.

Use the concepts of differentiation to solve for ‘q’ when profit is maximised.

Use the following to assist you with your analysis ..

q Demand

(x) (y) xy x2

1 0.5 5500

2 1 4000

3 1.5 4000

4 2 4000

5 2.5 2500

6 3 2500

7 3.5 1000

8 4 2500

9 4.5 1000

10 5 500

Sums 27.5 27500 0 0

University of
126 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

ACTIVITY FEEDBACK

q Demand

(x) (y) xy x2

1 0.5 5500 2750 0.25

2 1 4000 4000 1

3 1.5 4000 6000 2.25

4 2 4000 8000 4

5 2.5 2500 6250 6.25

6 3 2500 7500 9

7 3.5 1000 3500 12.25

8 4 2500 10000 16

9 4.5 1000 4500 20.25

10 5 500 2500 25

Sums 27.5 27500 55000 96.25

Revenue

Cost

Profit

q Demand Cost Profit

0 5500 500 -500

1 4500 100 4400

2 3500 -100 7100

3 2500 -100 7600

4 1500 100 5900

5 500 500 2000

6 -500 1100 -4100

7 -1500 1900 -12400

8 -2500 2900 -22900

University of
Sunderland 127
Unit 4 – Optimisation Business Modelling for Decision Making

y = -1000x + 5500
6000

5000
4000

3000

2000
1000

0
0 2 4 6

Graph of Profit
10000
5000
0
-5000 0 1 2 3 4 5 6 7 8
Profit

-10000
-15000
-20000
-25000
X values

Intercept = 5500

Hence, 5500 - 1000q

Demand
=

Const q q2

5500 -1000

500 -500 100

-500 6000 -1100

Differentiation

Profit Maximisation occurs

when q = 2.73

University of
128 Sunderland
Business Modelling for Decision Making Unit 4 – Optimisation

REVIEW ACTIVITY
(TR) = 9,000q - 800q2, (FC) = £10,000 and (VC) = (-1,000q + 200q2).

Tasks - Use the above revenue and cost equations and calculate:

1. The changes required in the coefficient of the ‘q’ variable of the cost
function to maximise profits when q = 4. (plus % changes).

2. The changes required in the coefficient of the ‘q2’ variable of the cost
function to maximise profits when q = 4. (plus % changes).

In all cases, commence from the initial state of the example.

ACTIVITY FEEDBACK
(TR) = 9,000q - 800q2, (FC) = £10,000 and (VC) = (-1,000q + 200q2).

Cost Equation

for the coefficient of the ‘q’ term to change to maximise profit when q = 4

$ = (9,000q - 800q2) – (10,000 - 1,000q + 200q2). Let 1,000q = xq

Hence, $ = 9,000q - 800q2 – 10,000 + xq - 200q2 or $ = 9,000q + xq – 1,000q2

– 10,000

And for $ max, = 0 = 9,000 + x – 2,000q

and we know $ max is to occur when q = 4, so, 0 = 9,000 + x – (2,000*4)

and 0 = 9,000 + x – 8,000 or 0 = 1,000 + x Hence x = -1,000

Check: $ = (9,000q - 800q2) – (10,000 + 1,000q + 200q2).

$ = 9,000q - 800q2 – 10,000 – 1,000q - 200q2 or $ = 8,000q – 1,000q2 – 10,000

d$
And for $ max, = 0 = 8,000 – 2,000q, so q = 8,000/2,000 = 4
dq

Note – this is a hypothetical situation!

University of
Sunderland 129
Unit 4 – Optimisation Business Modelling for Decision Making

Cost Equation

for the coefficient of the ‘q2’ term to change to maximise profit when q = 4

$ = (9,000q - 800q2) – (10,000 -1,000q + 200q2). Let 200q2 = xq

Hence, $ = 9,000q - 800q2 – 10,000 +1,000q - xq2 or $ = 10,000q – 800q2 –

xq2 – 10,000

d$
And for $ max, = 0 = 10,000 – 1,600q – 2xq
dq

and we know $ max is to occur when q = 4, so, 0 = 10,000 – (1,600*4) – 8x

and 0 = 10,000 – 6,400 - 8x or 0 = 3,600 - 8x Hence x = 3,600/8 = 450

Check: $ = (9,000q - 800q2) – (10,000 - 1,000q + 450q2).

$ = 9,000q - 800q2 – 10,000 + 1,000q - 450q2 or $ = 10,000q – 1,250q2 –

10,000

d$
And for $ max, = 0 = 10,000 – 2,500q, so q = 10,000/2,500 = 4
dq

Summary
This unit has used some complex maths to demonstrate how an analysis
of profit can be carried out by using a company’s revenue and cost
functions. The purpose of the demand function has also been described.

The key mathematical concept of differentiation has been explained,

both in theory and through examples of its application. Following on
from this it has been possible to identify the processes of profit
maximising and cost minimising.

University of
130 Sunderland
Unit 5

Basic Forecasting Techniques

Introduction
This unit provides an introduction to some of the basic forecasting
techniques employed in business modelling. The twin concepts of
regression and correlation are both explained and the derivation of the
necessary mathematics included. These are advanced from the idea of
scatter graphs through to their solution using both manual and
computer (via Microsoft Excel) methods.

The interpretation of the results provides a basis from which business

decisions can be made.

The unit includes some detailed examples in the form of activities for
the student to attempt.

The Slope of a Straight Line

A linear system can be represented as y = f(x) = a + bx. a and b are called
parameters and y and x are called variables.

Assume a and b are > 0 (positive values). We have already looked at

graphs (see the chapter on linear programming) and as a recap, the
dependent (independent) variable is usually depicted on the vertical
(horizontal) axis.

Rules for plotting graphs of the type y = f(x) :

1. For y = a + bx; let the value of x = 0, then y = a, implying the

graph will intercept the vertical axis at the value of y when y = a.
2. For y = a + bx; let the value of y = 0, then a + bx = 0, implying
bx = -a and x = -a/b. This implies the graph will intercept the
horizontal axis at the value of x when x = -a/b.
3. Since y = a + bx is a linear system (i.e. a polynomial of order
unity), then we can draw the graph of y = a + bx simply by
joining up the points y = a and x = -a/b .

University of
Sunderland 131
Unit 5 – Basic Forecasting Techniques Business Modelling for Decision Making

We do not need to limit our line to the points y = a and x = –a/b. We can
continue the line, but only because it is a linear system. If we were
dealing with a non-linear system, we would need to plot more points.

From this simple analysis of a linear system, we can develop the

concepts of differential calculus. We must assume that our linear line is
a curve, but with linear properties.

KEY POINT
The derivative of a function, finds the slope of a tangent to the curve at a
particular point, whether the curve is linear or non-linear. Let us examine the
extended graph we have recently plotted :

University of
132 Sunderland
Business Modelling for Decision Making Unit 5 – Basic Forecasting Techniques

We know that when x = 0, y = a and that when y = 0, x = –a/b. These

can be written in the following fashion :

(-a/b, 0), (0, a)

where the first (second) term in the brackets refers to the value of x (y) at
any particular point. These points are known as co-ordinates.

We can read from the graph, a value of y for any given value of x. Let us
examine the points y1 and y2, which have occurred due to plotting points
x1 and x2 respectively.

The distance from x1 (y1) to x2 (y2) is known as the change in the value of
x (y).

KEY POINT
Changes in value of y and/or x are normally written as :

*y and/or *x

If we divide the change in y by the change in x, we arrive at a ratio. If we

allow the changes between x1 and x2 (and hence y1 and y2) to reduce
towards zero, but not quite equal to zero, we arrive at the derivative.

University of
Sunderland 133
Unit 5 – Basic Forecasting Techniques Business Modelling for Decision Making

*y dy
'
*x dx

limit x –> 0

dy
Where is the derivative of the equation. More on the derivative later.
dx

Suppose y = 4 + 2x, then the ratio between y and x would be 2. i.e. when
x changes by one unit, y will change by 2 * one unit or by 2. Because y =
4 + 2x, then changes in x will produce changes in y of a similar direction.
(an increase in the value of x will produce an increase in the value of y
and vice-versa).

Suppose y = 4 - 2x, then the ratio between y and x would be -2. i.e. when
x changes by one unit, y will change by -2 * one unit or by -2. Because y
= 4 - 2x, then changes in x will produce changes in y of an opposite
direction. (an increase in the value of x will produce a decrease in the
value of y and vice-versa).

KEY POINT
As a recap, the ratio of the change in y by the change in x, gives us information
concerning

1. The direction of the change in y in response to a unit change in x.

(Whether or not both variables change in the same direction.)

2. The magnitude of the change in y in response to a unit change in x.

(The value of the change in y with respect to the change in x.)

ACTVITY
An Economic example would be : Qd = f(P) = 20 - 4P.

(i.e. Quantity demanded (Qd) is a function of Price (P) and is equal to the value
of a constant (20) minus Price (P) times 4)

Find the derivative of the function.

University of
134 Sunderland
Business Modelling for Decision Making Unit 5 – Basic Forecasting Techniques

ACTIVITY FEEDBACK
*Qd
' (4
*P

and is interpreted as an increase (decrease) in price by one unit will imply a

decrease (increase) in the quantity demanded by 4 units.

Market Demand and Supply

KEY POINT
Market Demand: The aggregate quantity demanded by all consumers
of a good/time period at various prices.

Market Supply: The aggregate quantity supplied by all sellers or

producers of a good/time period at various prices.

The general assumption is that a competitive market exists, i.e. no one

buyer or seller can influence the market mechanism.

Referring to our previous notes :

Qd1 = Qd1(P1, P2, Y)

i.e. quantity demanded of good 1 (Qd1) is a function of the price of good

1 (P1), the price of good 2 (P2) and income (Y).

A much better picture of the economy for one good, might be as follows:

Qd1 = Qd1(P1, P2, P3, ... , Pn, Y, L, D)

i.e. quantity demanded of good 1 (Qd1) is a function of the price of good

1 (P1), the price of good 2 (P2), the price of good 3 (P3), ... , the price of
good n (Pn), income (Y), Labour (L) and income distribution (D)).

Let us assume a simple model of Qd = a - bP (a, b > 0). We can sketch

this model as follows :

University of
Sunderland 135
Unit 5 – Basic Forecasting Techniques Business Modelling for Decision Making

Similarly the model for market supply might be as follows:

Qs1 = Qs1(P1, P2, P3, ... , Pn, F1, .. , Fn, T)

i.e. quantity supplied of good 1 (Qs1) is a function of the price of good 1

(P1), the price of good 2 (P2), the price of good 3 (P3), ... , the price of
good n (Pn), Factors of production (F1, .. , Fn) and Technology (T)).

Again, we can assume a simple model :

Qs = -c + dP (c, d > 0)

We can also construct the graphs with Price being the dependent
variable and Quantity Demanded/Supplied as the independent
variable.

University of
136 Sunderland
Business Modelling for Decision Making Unit 5 – Basic Forecasting Techniques

A Linear Partial Equilibrium Market Model

KEY POINT
Assumptions for partial analysis:

1. The good is bought and sold in an isolated market (i.e. there are no
interactions between this market and the rest of the economy).

2. Changes in marketing conditions have negligible effects on other parts

of the economy and vice-versa.

Consumer/producer behaviour patterns suggests a quantity

demanded and quantity supplied equilibrium will exist where Qd = Qs
= Qe = Pe (equilibrium) :

Let us analyse this scenario :

At the equilibrium point, Qd = Qs

but Qd = a - bP and Qs = -c + dP (a, b, c, d > 0)

Therefore, a - bP = -c + dP

We can determine the equilibrium solution value for P by simple

algebra :

University of
Sunderland 137
Unit 5 – Basic Forecasting Techniques Business Modelling for Decision Making

We add +c to both sides of the equation :

a- bP + c = dP

We add +bP to both sides of the equation :

a + c = dP + bP

We can factorise the right hand side of the equation :

a + c = (d + b)P

and we can divide both sides of the equation by (d + b) :

a -c
'P
(d - b)

The unique value for P is called the equilibrium value or the solution
value, and we normally depict equilibrium/solution values with a bar
above the character as such :

Having determined the equilibrium value for P, we can substitute for

this value of P into either Qd or Qs. Assume we use Qd :

Qd = a - bP hence Qd ' a ( b
.a - c /
. d - b/

ACTIVITY
An example should help clarify this analysis :

Let Qd = Qs ; Qd = 20 - 7P ; Qs = -4 + 5P ; Find P and Q

ACTIVITY FEEDBACK
Equating Qd to Qs : 20 - 7P = -4 + 5P adding +4 to both sides

20 - 7P + 4 = 5P adding 7P to both sides

20 + 4 = 5P + 7P and tidying the terms

University of
138 Sunderland
Business Modelling for Decision Making Unit 5 – Basic Forecasting Techniques

24 = 12P

Therefore P = 24/12 = 2

When P = 2 we can substitute for P = 2 into Qd or Qs. Let us substitute

into Qd :

Qd = 20 - 7P = Qd = 20 - 7(2) = Qd = 20 - 14 = 6 = Qd = Q

ACTIVITY
Another example :

Let Qd = Qs ; Qd = 27 - 4P ; Qs = -3 + 2P ; Find P and Q

ACTIVITY FEEDBACK
Equating Qd to Qs : 27 - 4P = -3 + 2P adding +3 to both sides

27 - 4P + 3 = 2P adding 4P to both sides

27 + 3 = 2P + 4P and tidying the terms

30 = 6P

Therefore P = 30/6 = 5

When P = 5 we can substitute for P = 5 into Qd or Qs. Let us substitute

into Qs :

Qs = -3 + 2P = Qs = -3 + 2(5) = Qs = -3 + 10 = 7 = Qs = Q

University of
Sunderland 139
Unit 5 – Basic Forecasting Techniques Business Modelling for Decision Making

Techniques for Estimating and Forecasting

Demand and Costs

Sampling via Surveys

These are used to derive values of the quantity demanded at different
values of the independent variables, e.g. where quantity is dependent
upon price:

Hence, a survey will be taken at random, for example to derive values of

quantity demanded at various prices, and keeping other variables
constant (e.g. prices of substitutes, etc). Surveys would be repeated
continuously, e.g. Tyne & Wear PTE. The objective of market surveys, is
to determine ELASTICITIES of demand.

Repeated sampling may not provide much information concerning

elasticities, since all variables associated with the quantity demanded
are subjected to changes. Hence, sampling techniques provide ARC
elasticities of demand rather than POINT elasticities.

KEY POINT
Surveys do not supply the following information :

(1) The values for the coefficients in the demand function, i.e. a and b in Y
= a + bX.

(2) The method for predicting the quantity demanded for combinations of
values of the explanatory variables, other than those given (i.e. given
Q, find P).

University of
140 Sunderland
Business Modelling for Decision Making Unit 5 – Basic Forecasting Techniques

To overcome these difficulties, we apply Regression Analysis.

Introduction to Regression Analysis

We need to CORRELATE the dependent and independent variables.

We achieve this by inserting a linear line, which best fits the data points.
This line, will provide information with respect to the values of the
coefficients of a and b:

The line of best fit can be inserted at any position between the data
points. In order to present the most efficient line of best fit, we adopt
what is known as the Ordinary Least Squares (OLS) technique.

The OLS technique, provides a `line of best fit’, via MINIMISING the
square of the vertical distance between each data point (data
observation) and the OLS line.

University of
Sunderland 141
Unit 5 – Basic Forecasting Techniques Business Modelling for Decision Making

The OLS regression model (or equation presented via the OLS `line of
best fit’) can be presented as :

Qt = a + bPt + Ut, where :

Qt = (in this case) Quantity demanded per time period,

Pt = the price per quantity demanded per time period,

Ut = DISTURBANCE or ERROR term (captures other factors,

not accounted for via Q = a + bP).

a, b are the coefficients and are expected to be greater than or equal to

zero.

The error term is stochastic, i.e. it occurs by chance.

Note that factors captured in the error term may include consumers
tastes for example. Techniques to measure tastes, are not readily
available.

Suppose we `fit’ a linear line to the sample data. It is unlikely that all the
data will lie on the line. Suppose the population is correlated and a true
line is fitted to the population data. See the following diagram:

The distance between the population line and each data point (x), is
known as the disturbance term or error term at that point (eg P1Q1
implies U1, P2Q2 implies U2 etc.) The true model is represented as :

Qt = a + bPt + Ut

The sample model is represented as Q t ' a - bPt - e t . Note .. the `hats’ (a

and b) represent ESTIMATED coefficients.

University of
142 Sunderland
Business Modelling for Decision Making Unit 5 – Basic Forecasting Techniques

The distance between the sample line and each data format (x), is
known as the sample error, termed RESIDUAL (e). The residuals are
assumed to affect the true error term (Ut).

When using regression analysis, the following procedure is adopted:

1. Choice of Variables:

Economic theory suggests the selection of those variables that are

considered to make economic sense, e.g. consumption depends upon
Income (fundamental relationship). Other variables would also have to
be considered, such as weather for example. Good weather can increase
the demand for soft drinks and ice cream, etc.

2. Form of the Model:

The decision to select a particular model, i.e.

(a) Linear or (b) Non Linear

If (a) is selected, then ordinary Least Squares (OLS) can be applied. If a

non linear model is selected (b), then it can be transformed (in some
instances), to a linear model, and OLS can be applied.

3. Obtaining the Data:

For many variables, the data can be found in governmental agencies,

e.g. The Central Statistical Office (CSO), etc. Other variables may have
to be researched and the data collected accordingly, e.g. via surveys or
experiments.

4. Estimating the Parameters of the Model:

Suppose that our model appears as : Qt = a + bPt + Ut

i.e. Quantity at time ‘t’ is a linear function of Price at time ‘t’.

The estimates of a and b (i.e. a and b) can be calculated via OLS.

a and b are estimates which minimise the sum of squared residuals (i.e.
the distances between an actual observation and a desired observation,
which would appear on the line of best fit).

Referring to the previous diagram, if Qt = a + bPt + Ut then our estimated

equation is given by :

Qt = a + bPt + et

This can be re-arranged and expressed in terms of the residuals (et), i.e.

et = Qt - (a + bPt) or et = Qt - a - bPt

University of
Sunderland 143
Unit 5 – Basic Forecasting Techniques Business Modelling for Decision Making

and we can find a solution for the sum of the squared residuals from :

. /
n 2
0e t '1
2
t ' Q t ( a ( bPt

We need to minimise the sum of the squared residuals to calculate our

OLS line of regression. We do this by differentiating (i) with respect to a
and b, to determine a MINIMUM turning point ..

d0 e 2 t
. /. (1 / ' 0
n
' 0 e 2 t ' Q t ( a ( bPt
da t '1

d0 e 2 t
. /. (P / ' 0
n
' 0 e 2 t ' Q t ( a ( bPt t
db t '1

The above two equations can be rearranged to permit the solutions for a
and b to occur :

n. 0 Pt Q t / ( . 0 Pt /. 0 Q t /
b'
.0 P / ( .0 P /
n t
2
t
2

a'
.0 Q / ( b .0 P /
t t

n n

In a, we would substitute for b, to solve.

We can simplify this equation to terms of Y and X, i.e. Y = a + bX

n. 0 XY / ( . 0 X/. 0 Y /
b'
n .0 X / ( . 0 X/
2 2

a'
. 0 Y / ( b . 0 X/
n n

where n = number of observations

The above formulae expressed in terms of X and Y denotes the general

formulae to estimate the coefficients a and b.

5. Interpretation:

This section allows us to interpret our results. Does the regression

equation permit us to make future forecasts? Does the regression
equation fit the data well?

University of
144 Sunderland
Business Modelling for Decision Making Unit 5 – Basic Forecasting Techniques

ACTIVITY
Let us look at an example. Given the following data:

C1 1.0 1.1 0.9 1.2 2.0 2.1 3.0 3.4 2.9 1.8

C2 1.2 1.0 1.1 1.3 1.8 2.2 3.4 3.1 2.8 1.9

C3 1.3 1.2 0.8 1.4 1.9 2.3 3.1 3.7 2.7 1.6

Use the general formulae as listed above to solve:

(1) C1 = a + bC2

(2) C1 = a + bC3

(3) C2 = a + bC3

(4) C2 = a + bC1

(5) C3 = a + bC1

(6) C3 = a + bC2

ACTIVITY FEEDBACK
For columns C1 and C2, let C1 = Y and C2 = X then

0 Y ' 19.4 ; 0 X ' 19.8 ; 0 (XY) ' 45.35 ;

.0 X/
2
0 (X 2
) ' 46.04 ; ' 392.04 ; n = 10 ;

n. 0 XY/ ( . 0 X/. 0 Y/
b'
.0 X / ( .0 X/
2 2
n

10. 45.35/ ( .19.8/.19.4/

b'
10. 46.01/ ( 392 .04

. ( 38412
4535 . 69.83
b' and b ' ' 101492
.
460.4 ( 392 .04 68.36

University of
Sunderland 145
Unit 5 – Basic Forecasting Techniques Business Modelling for Decision Making

a'
.0 Y/ ( b .0 X/
n n

a'
.19.4/ ( (1.01492) .19.8/
10 10

or 1.94 - 2.0095 implying a = - 0.0695

Since Y = C1 and X = C2 then C1 = -0.0695 + 1.01492C2

(The remaining solutions can be determined in a similar way.)

Correlation
The simple linear correlation coefficient (abbreviated to r) is a unit free
measure of the direction and strength of the linear association (i.e. a
linear line of best fit) between two (hence ‘simple’) random variables.

The crucial limiting property of the correlation coefficient, is that it

measures only the direction and strength of linear association between
the two random variables, i.e. the closer the observations lie on a linear
line, the higher will be the value of the correlation coefficient. The
correlation coefficient lies between and including 1 1, i.e.

-1 " r " +1

When r = 1 (= -1), the two random variables are perfectly positively

(negatively) linear associated. When r = 0, there is no linear association
between the two random variables.

When r is a positive (negative) value i.e. r > 0 (r < 0), the line is increasing
(decreasing).

The correlation coefficient is also known as the Pearson Product

Moment or Pearson Correlation Coefficient.

Examination of the following diagrams should help clarify the linear

association properties.

Also note, we can provide another measurement R2 and this value is

simply the square of the correlation coefficient. R2 allows us to measure
the proportion of the variation in the second sample that is best
‘explained’ by the first sample. R2 is known as the Coefficient of
Determination. It is usually listed in the analysis of variance (ANOVA)
within statistical packages such as Minitab or SPSS.

The correlation coefficient can be calculated as follows :

University of
146 Sunderland
Business Modelling for Decision Making Unit 5 – Basic Forecasting Techniques

2 . 0 X/. 0 Y / 5
4. 0 XY / ( 7
43 n 76
rXY '
2 . 0 X/ 5 2 .0 Y / 5
2 2

4.0 X /
43
2
(
n
784 .0 Y /
76 43
2
(
n
7
76

Examine the following diagrams. Clearly, the middle diagram depicts

a strong non linear relationship between the two variables.

We cannot effectively use Pearson’s correlation coefficient (r) in the

middle graph above, since (r) refers to a linear relationship only. So, we
need to define a correlation coefficient which will refer to a non linear
relationship. The correlation coefficient which measures the direction
and strength of non linear association between two variables is known
as:

University of
Sunderland 147
Unit 5 – Basic Forecasting Techniques Business Modelling for Decision Making

Spearman’s Correlation Coefficient

Spearman’s correlation coefficient is short-handed to 9 (rho) and uses
the ranks of the variables to determine the non linear association. Each
variable in turn, is ranked relative to the other values. Similar to
Pearson’s correlation coefficient :

(1 " 9 " 1

Spearman’s correlation coefficient is defined by the following formulae:

60 D 2
9 '1 (
.
n n2 ( 1 /
Where n = number of observations and D = difference between the
observations.

ACTIVITY
Example: Two art critics were asked to grade 12 paintings. Their results were
as follows :

Show Critic No 1 Critic No 2 D D2

A 8 3 5 25
B 9 7 2 4
C 3 10 -7 49
D 10 2 8 64
E 11 11 0 0
F 7 12 -5 25
G 6 5 1 1
H 4 1 3 9
I 2 8 -6 36
J 1 4 -3 9
K 5 6 -1 1
L 12 9 3 9

232

Determine whether there is a strong non linear association.

University of
148 Sunderland
Business Modelling for Decision Making Unit 5 – Basic Forecasting Techniques

ACTIVITY FEEDBACK
Using the formulae to solve for 9:

60 D2
9 ' 1(
.
n n 2 (1/
6.232 /
9 ' 1(
12 .144 -1/

and 9 = 0.19 and since 0.19 is close to zero, we can state that there is not a
strong non linear association between the two art critics, i.e. each art critic
appears to have reached his decision independent of the other.

The Basic Assumptions of the Simple Linear

Regression Model
1. The dependent variable (Y), is a stochastic linear function of the
independent variable (X), of the form :
Y ' ) - &X i - U i
i.e. a linear relationship is expected to exist.
2. The disturbance term (U) is a normally distributed random
variable, i.e. for every value of the independent variable (X), the
disturbance term is expected to be normally distributed.
For example, we can assume that the fundamental relationship
between income and expenditure, is affected by other factors
such as the weather, illnesses. tastes etc.
We can further assume that there may be as many positive
factors which affect expenditure (causing an increase in
expenditure), as there are negative factors (causing a decrease in
expenditure). If we assume that all these deviations are of a
similar size, and that the occurrence of negative deviations are
just as probable as the occurrence of positive deviations, then we
can further assume that the distribution of these deviations is
normal.
3. The mean of the disturbance term is expected to be zero i.e.
E(Ui) = 0

University of
Sunderland 149
Unit 5 – Basic Forecasting Techniques Business Modelling for Decision Making

If the assumptions in (2) above are true, then it can be assumed

that the net effect on the disturbance term is zero, and that the
disturbance term is expected to have a zero mean.
4. The variance of the disturbance term is a constant usually
denoted as :2 i.e. ...
var(Ui) = :2
It means that irrespective of the values of the independent
variable, the dispersion of the values of the disturbance term is
the same, implying homoskedasticity.
If we use the example of households consumption expenditure,
then this assumption implies that the dispersion of all
expenditure values by households, at all levels of income, is the
same.
In reality, this is not true of course, since households with large
incomes, would have more income left over for discretionary
consumption. Households with low incomes, would have little
(if any) left over after consumption on rent or mortgage, and
other fixed outgoings.
In practical applications, the variance of the disturbance term is
more than likely not constant. If this is the case, then the
disturbance term is said to be heteroskedastic.
We can re-write this assumption such that the expected value of
the variance of the disturbance term is :2 i.e. :
E(Ui2) = :2
If we group assumptions 2, 3 and 4 together, then we can assume
that the disturbance term follows a normal distribution with a
zero mean, and a constant variance :
Ui ~ N(0, :2)
5. The disturbance terms are independent of each other, i.e. the
covariance (cov) between any two values of the disturbance term,
is expected to be zero i.e. :
cov(UiUj) = 0 = E(Ui2Uj2)
Returning to the example on household expenditure. If any one
household spends above average on consumption, it is expected
that this household’s spending decisions, will not affect other
household’ spending decisions, i.e. other household’s will not
spend more or less on consumption in consequence.
If this assumption is not satisfied, then the disturbance terms are
said to be autocorrelated.
6. The independent variable (X) is a non-stochastic variable, fixed in
repeated sampling.

University of
150 Sunderland
Business Modelling for Decision Making Unit 5 – Basic Forecasting Techniques

Non-stochastic suggests that the values of the independent

variable do not occur by chance or probability. We are free to
select whatever values of the independent variable that we
require.
Fixed in repeated sampling means that we can repeat the
sampling process via permutating the samples of the
independent variable (from the true population), and calculating
the new sample values for the dependent variable.
In practice, the probable values of the independent variable will
occur via chance (i.e. stochastic). One final point, it is assumed
that the values of the independent variable are not all equal to
each other.

Given these six assumptions, we are able to represent our model as the
Classical Normal Linear Regression Model (CNLR). The CNLR model
has three unknown parameters; i.e. ), & and the variance (:2) of the
disturbance term (U).

We have already used Ordinary Least Squares (OLS) to solve for ) and
&. In this module, we are not interested in the variance of the disturbance
term.

REVIEW ACTIVITY
Question No. 1 (Manual work – use calculators)

The manager of the ‘Sizzling Gastropod’ bistro, provides a delivery service for
customers who telephone in an order. The manager would like to give callers
an idea of the time it will take to deliver their order. This depends upon the
distance of the customer from the bistro. The records of the last 12 deliveries
are shown in the following table:

! Calculate the coefficient of correlation. Does the result obtained

indicate that customer distance is a good predictor of delivery time?

! Determine the regression equation which would enable the

manager to estimate the time of delivery to the customer.

! Use your regression equation to estimate the time it would take to

deliver an order a distance of 5 miles.

University of
Sunderland 151
Unit 5 – Basic Forecasting Techniques Business Modelling for Decision Making

Distance Delivery Time

(miles) (minutes)

2.3 5

6.7 13

7.5 10

3.1 5

4.6 9

3.9 8

8.7 15

9.8 20

10.1 18

6.5 13

7.3 12

5.2 9

REVIEW ACTIVITY FEEDBACK

The data is summarised as follows:

0 X ' 75.7, 0 Y ' 137, 0 X 2

' 548.73, 0Y 2
' 1807,

0 XY ' 988.9, N = 12
X ' 6.308, Y ' 11.416, 0X 2
' 5730.49, 0Y 2
' 18769

Delivery time is our dependent (Y) variable.

The value of the correlation coefficient for this data is r = 0.948, which is very
close to unity. This indicates that distance would be a good predictor of time.

The regression equation (based upon the above observations) is Y = 0.3702 +

1.7511X

(Y represents time and X represents distance).

Thus, to cover a distance of 5 miles would take 9.1 minutes.

University of
152 Sunderland
Business Modelling for Decision Making Unit 5 – Basic Forecasting Techniques

REVIEW ACTIVITY
Question No. 2 (Computer work - use Excel)

A market analyst working for a large company wishes to establish a method for
predicting the number of sales a salesperson will achieve for a given number of
sales calls on customers. To this end, the analyst collects a sample of 20 sales
people and for each such person the analyst records :

a. The average weekly sales achieved by the sales person,

b. The average number of calls made in a week, and all results are
displayed as follows:

Sales Person No. of Calls No. of Sales Sales Person No. of Calls No. of Sales

1 26 11 11 23 11

2 13 7 12 25 7

3 21 8 13 38 18

4 37 20 14 33 14

5 17 9 15 12 2

6 20 12 16 30 15

7 17 4 17 30 10

8 28 16 18 10 4

9 28 11 19 18 7

10 6 2 20 21 10

( Draw a scatter plot for the data. Does it support the

contention that the relationship between these two variables
is linear?

( Calculate (and comment upon) the coefficient of correlation

between the two variables.

( Calculate a least squares regression equation which will make

the kind of predictions the analyst wishes.

( If a salesperson makes upon average, 37 calls per week, use

your regression equation to calculate his/her average weekly
sales. Comment upon the accuracy of your prediction.

University of
Sunderland 153
Unit 5 – Basic Forecasting Techniques Business Modelling for Decision Making

( Consider the regression coefficients (a & b OR ? & ?).

What do they tell us about the data? Do they supply
any useful information?

( Comment upon the general predictive accuracy of the

regression model. Make any calculations necessary to
support your comment.

REVIEW ACTIVITY FEEDBACK

! A study of the scatterplot indicates that linearity is justified.

! The remainder of the solution uses the following summary:

0 X ' 453, 0 Y ' 198, 0 X 2

' 11733, 0Y 2
' 2440,

0 XY ' 5238, N = 20 and r = 0.8962

! If the number of sales is our dependent

(Y) variable, then the slope = 0.5116 and the intercept is -1.6869,

i.e. Y = -1.6869 + 0.5116X

(or Y = 0.5116X - 1.6869)

! When X = 37, Y = -1.6869 + (0.5116 * 37). Hence, Y =

17.2423. The data contains a person who actually made 37
calls and achieved 20 sales. Therefore, we have a forecast

University of
154 Sunderland
Business Modelling for Decision Making Unit 5 – Basic Forecasting Techniques

error of about 2.76 which is 13.8% (i.e. (20 - 17.2423)/20 * 100).

Despite this poor estimation, the regression equation is reasonably
accurate (see answer to last part).

! The slope coefficient provides the rate at which sales are made for a
given change in calls. In our case, for every one unit change in the
number of visits, the number of sales will change by 0.5116 units.
Generally, the intercept coefficient does not provide us with any
useful information. In our case, when the number of visits is equal to
zero, the number of sales is -1.6869 (which does not make sense).

! We can use the coefficient of variation (r2).

r2 = 0.8032 (80.32%) which means that 80.32% of the number of

sales, can be explained by the number of visits. The other 20% or so
of sales can be explained by other variables e.g. advertising,
customers calling to collect themselves etc.

Summary
Regression analysis using two variables and the calculation of linear
correlation coefficients have been introduced in this unit and enable the
student to carry out an effective analysis of data. Methods for carrying
out the necessary calculations, either by hand or using appropriate
software, have also been described.

Although the derivation of some of the material in this unit involves

quite involved maths, the underlying principles are important to all
managers. Students are, therefore, recommended to study the activities
with care and refer to a suitable text book for additional worked
examples.

University of
Sunderland 155
Unit 6

Inventory Control

Introduction
Stock control, otherwise known as inventory control, is used to show
how much stock you have at any one time, and how to keep track of it.

It applies to every item you use to produce a product, from raw

materials to finished goods, or a service. It covers stock at every stage of
the production process, from purchase and delivery to using and
reordering the stock.

Efficient stock control allows you to have the right amount of stock in
the right place at the right time. It ensures that capital is not tied up
unnecessarily, and protects production if problems arise with the
supply chain.

Types of Stock
Everything you use to make your products, provide your services and
to run your business is part of your stock. The type of stock can
influence how much you should keep.

There are four main types of stock:

! Raw materials and components ready to use in

production.

! Work in progress - stocks of unfinished goods in

production.

! Finished goods ready for sale.

! Consumables - fuel and stationery.

Stock Value and Levels

You can categorise stock further, according to its value. For example,
you could put items into low, medium and high value categories. If

University of
Sunderland 157
Unit 6 – Inventory Control Business Modelling for Decision Making

your stock levels are limited by capital, this will help you to plan
expenditure on new and replacement stock.

You may choose to concentrate resources on the areas of greatest value.

However, low-cost items can be crucial to your production process and
should not be overlooked.

Stock levels depending on type of stock

The four main types of stock are:

Raw materials and components

Ask yourself some key questions to help decide how much stock you
should keep:

! How reliable is the supply and are alternative sources

available?

! Are the components produced or delivered in batches?

! Can you predict demand?
! Is the price steady?
! Are there discounts if you buy in bulk?

Work in progress - stocks of unfinished goods

Keeping stocks of unfinished goods can be a useful way to protect

production if there are problems down the line with other supplies.

Finished goods ready for sale

You might keep stocks of finished goods when:

! Demand is certain.
! Goods are produced in batches.
! You are completing a large order.

Consumables

For example, fuel and stationery. How much stock you keep will
depend on factors such as:

! Reliability of supply.
! Expectations of price rises.
! How steady demand is.

University of
158 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

! Discounts for buying in bulk.

How much stock should you keep?

The question then arises: how much stock should we have? It is this
simple question that inventory control theory attempts to answer.

There are two extreme answers to this question:

Keeping a lot:

! This ensures that we never run out.

! Is an easy way of managing stock.
! Is expensive in stock costs, cheap in management costs.

Advantages Disadvantages

Easy to manage. Higher stock, storage and insurance costs.

Low management costs. Certain goods might perish.

You never run out. Stock may become obsolete before it is used.

Your capital is tied up.

This might suit your business if sales are difficult to predict (and it is
hard to pin down how much stock you need and when), you can store
plenty of stock cheaply, the components or materials you buy are
unlikely to go through rapid developments or they take a long time to
re-order.

Keeping none or very little:

! This is known as Just-in-Time (JIT).

! Is a difficult way of managing stock.
! Is cheap in stock costs, expensive in management costs.

Advantages Disadvantages

Efficient and flexible. You only have what you need, Meeting stock needs can become complicated and
when you need it. expensive.

Lower stock and storage costs. You might run out of stock if there’s a hitch in the
system.

You can keep up-to-date and develop new products You are dependent on the efficiency of your suppliers.
without wasting stock.

University of
Sunderland 159
Unit 6 – Inventory Control Business Modelling for Decision Making

This might suit your business if it’s in a fast-moving environment where

products develop rapidly, the stock is expensive to buy and store, the
items are perishable or replenishing stock is quick and easy.

We shall consider the problem of ordering raw material stock but the
same basic theory can be applied to the problems of:

! Deciding the finished goods stock.

! Deciding the size of a batch in a batch production
process.

The costs that we need to consider so that we can decide the amount of
stock to have can be divided into stock holding costs and stock ordering
(and receiving) costs as below.

Note here that, conventionally, management costs are ignored here.

Holding costs - associated with keeping stock over time

! storage costs
! rent/depreciation
! labour
! overheads (e.g. heating, lighting, security)
! money tied up (loss of interest, opportunity cost)
! obsolescence costs (if left with stock at end of product
life)

! stock deterioration (lose money if product deteriorates

whilst held)

! theft/insurance

Ordering costs - associated with ordering and receiving an order

! clerical/labour costs of processing orders

! inspection and return of poor quality products
! transport costs
! handling costs

Note that a stockout occurs when we have insufficient stock to supply

customers. Usually stockouts occur in the order lead time, the time
between placing an order and the arrival of that order.

University of
160 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

Given a stockout the order may be lost completely or the customer may
choose to backorder, i.e. to be prepared to wait until we have sufficient
stock to supply their order.

Note here that whilst conceptually we can see that these cost elements
are relevant it can often be difficult to arrive at an appropriate numeric
figure (e.g. if the stock is stored in a building used for many other
purposes, how then shall we decide an appropriate allocation of
heating/lighting/security costs).

Stock control methods

To see how we can decide what stock level to adopt, consider the very
simple model below.

The basic function of stock (inventory) is to insulate the production

process from changes in the environment as shown below.

Raw materials stock Finished goods stock

Manufacturing

Work in progress
Cost

Revenue
Raw materials arrivals (restocking)
Sales of finished goods (destocking)

Stocks may be held for a variety of reasons. They may be stocks of raw
materials ready for production, they may be work-in-progress
(production part way through the production process) or they may be
stocks of finished goods. Whichever they are it is vital for the firm to
control the level of stocks very carefully. Too little and they may run
into production problems, but too much and they have tied up money
unnecessarily.

University of
Sunderland 161
Unit 6 – Inventory Control Business Modelling for Decision Making

The main theories about stocks then are to do with stock control. There
are various different ways to approach stock control:-

! Fixed re-order stock level

! Fixed time re-ordering
! Economic order quantity
! Just-in-time production

Fixed re-order stock level

This method of stock control is where a business decides the minimum

level of stocks it can tolerate, and then re-orders before the stocks reach
this level. The exact timing will depend how long the stocks take to
arrive. This can be illustrated as follows:-

Stock level Maximum stock

Reorder level

Minimum stock

Time

The distance between the re-order line and the minimum stocks level
depends how long it may take for the order to arrive - this time is known
as the lead time.

Fixed time re-ordering

This method is exactly as its title suggests. The firm re-orders stocks at a
fixed time each month or week. It can offer a good solution as it
represents a routine for the firm and ensures that stocks are regularly
supplemented. However, it may well mean the level of stocks
fluctuating quite a bit depending on the rate they are used up. It is a little
inflexible as a system as well unless used very carefully.

Economic order quantity

For any company there is an optimum level of stocks. The precise level
of this will vary in different firms and industries. They have to balance
the costs of holding stocks (the space taken, the money tied up, etc.)
with the costs of ordering stock. The more firms order at once, the better
the deal they will usually get.

University of
162 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

The level of stocks that strikes the balance between these two things is
known as the Economic Order Quantity. If this is taken to be the
optimum level of stocks it should help to minimise the firm’s costs - an
important pre-requisite to maximising profit.

Just-in-time production

Because stocks cost so much to keep, another method of stock-control

was developed in Japan and has now become much more common in
the UK. The just-in-time method involves keeping stocks to an absolute
minimum, and the raw materials are ordered only when they are
needed; in other words, just-in-time. This time period in some cases has
been reduced to minutes or hours, and the raw materials arrive on site
moments before they are needed.

This can be wonderful for helping to reduce the need for working
capital, but requires a very high level of organisational skill and a very
close relationship with suppliers.

These methods can be used alongside other processes to refine the stock
control system. For example:

Re-order lead time - allows for the time between placing an order and
receiving it.

Economic Order Quantity (EOQ) - a standard formula used to arrive at

a balance between holding too much or too little stock. It’s quite a
complex calculation, so you may find it easier to use stock control
software.

The formulae for the Economic Order Quantity is given as follows:

2DC
EOQ '
Ch

Batch control - managing the production of goods in batches. You need

to make sure that you have the right number of components to cover
your needs until the next batch.

If your needs are predictable, you may order a fixed quantity of stock
every time you place an order, or order at a fixed interval - say every
week or month. In effect, you’re placing a standing order, so you need to
keep the quantities and prices under review.

First in, first out - a system to ensure that perishable stock is used
efficiently so that it doesn’t deteriorate. Stock is identified by date
received and moves on through each stage of production in strict order.

University of
Sunderland 163
Unit 6 – Inventory Control Business Modelling for Decision Making

Inventory Costs
An inventory can best be described as the entire stock of a company.
This will include all completed items, which are awaiting dispatch,
semi-completed stock, raw materials and all supplies necessary for the
company to run in an efficient manner. Indeed, it will include every
item relevant to the operation of that business.

The reasons that an inventory is so important include the need to

predict future demand so that there is always sufficient stock to meet
demand. It is also necessary to have sufficient stock levels to cope with
times of extreme demand resulting from industrial action, etc.
Production of each item may take a significant time, and, therefore, it is
necessary to allow for the time delay between initiation and completion
of production.

Other factors that must be considered include discounts that may be

offered to bulk purchases. It is often sound business practice to ensure
that such offers are accepted and that stock levels are planned to cope
with this. However, restraints must also be applied to stock levels to
ensure that stock does not become obsolete or, where there is a
considerable fluctuation of prices within the market, become out-priced
in comparison to rival producers or retailers.

KEY POINT
The main aim of inventory control is to reduce all costs to a minimum, whilst at
the same time, maintaining an optimum amount of stock. The frequency of
re-ordering materials must also be calculated to allow the effective use of
inventory control.

Before investigating inventory control further, it would be useful to

define some important terms.

Lead time is the time elapsed between the initiation of an order and the
completed production of the item. This may be wholly within the
company or through to point of sale.

Minimum stock refers to the threshold level of stock below which it

would be unlikely for the company to be in a position to fulfill incoming
orders. Sometimes referred to as the safety stock.

Maximum stock is the level which a company considers it has no need

to exceed. During normal operation, such a level would ensure that all

University of
164 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

orders can be processed without any unnecessary demand placed on

the production process.

Reorder level is the level of stock at which a further order needs to be

placed in order to maintain adequate stock levels.

Reorder quantity is the size of order necessary to ensure adequate stock

levels.

Economic ordering quantity (EOQ) is the optimum value for external

order quantity such that inventory costs are kept at a minimum.

Economic batch quantity (EBQ) is an adaptation of the EOQ, dealing

with keeping inventory costs at a minimum through internal
production.

KEY POINT
Inventory costs fall into two main categories;

! Ordering costs – costs incurred each time an order is placed. These

can include administrative work, telephone calls, postage,
transportation, etc.

! Holding costs – costs of keeping an item in stock. These can include

capital costs, handling, storage, insurance, taxes, depreciation, etc.
Overheads, such as heating, lighting and rent, should also be
considered, as well as breakages and theft.

A smaller, but not insignificant, cost should not be neglected when stock
runs out. This may lead to penalty payments, loss of goodwill and staff,
and machinery idle. This cost can be almost totally avoided by adequate
inventory control.

University of
Sunderland 165
Unit 6 – Inventory Control Business Modelling for Decision Making

The Inventory Graph

Stock
level

b
d

a
e
c

A graphical description of the inventory control can be obtained by

plotting stock levels against time. This will show a comparison between
the supply of the stock and the demand for it, whilst at the same time
displaying the rates at which the stock can be replenished.

Consider the diagram above which shows a basic inventory graph. It

can be readily seen that the graph is made up of a number of discrete
sections. These show supply of stock and demand levels – in both cases
at differing rates. Section a indicates a gradual supply, or
replenishment, of stock levels. Section b indicates a period of no
demand and section c shows a steady demand.

Section e indicates differing rates of demand and section d shows

instantaneous supply. This latter stage is shown as a vertical line on the
graph. In reality, such instantaneous supply and demand would be
unlikely. However, for the purpose of modelling, it is useful to consider
this option.

There is also a period of no change shown, reflecting a period of zero

demand when stock levels are at an acceptable level. Once a working
inventory control is in place, the graph will tend to show a periodic
nature as supply is controlled and demand becomes more accurately
predicted. The inventory graph will then follow the sort of pattern
shown below. Note that in this idealised case there is instantaneous
supply, following a stock out point, returning stocks to their
pre-determined maximum level.

University of
166 Sunderland
 Business Modelling for Decision Making Unit 6 – Inventory Control

Q
Stock level

0
Time

Note: allow stock to fall to zero as no time needed to replenish stock

This periodic nature of the graph is referred to as the inventory cycle.

The cycle includes an ordering component and a supply component.
The length of the cycle is simply the time taken for the cycle to return to
the starting point of the next cycle.

An inventory graph provides the means to calculate the average

inventory level. This is obtained by calculating the total area under the
graph and dividing this by the total time elapsed. For example, in
Illustration 3.2, for each triangle formed on the graph, the area can be
calculated and then divided by the time, giving an average inventory
level of for each triangle,

Assuming the four triangles imply one cycle, and we let the maximum
stock level (Q) = 800 then

area = 0.5 x 800 x 4 = 1600

so for 3-cycle period = 1600 x 3 = 4800.

Thus average inventory level per time period = 4800 / 12 = 400.

KEY POINT
There are two standard methods of inventory control;

! Re-order level system – whereby for each stock item, a fixed

quantity of stock is ordered each time the stock level drops to, or

University of
Sunderland 167
Unit 6 – Inventory Control Business Modelling for Decision Making

below, a pre-determined level. The quantity of stock ordered

on each occasion is normally the Economic Order Quantity
(EOQ), previously defined.

! Periodic review system – whereby a review period is

established and at the end of each period the existing stock
level is brought back up to a pre-agreed level.

Reorder Level System

The re-order level system is the most commonly used system for
inventory control. Its main advantages are that it operates from a lower
level of stock, thus reducing the amount of money tied up in waiting
stock, and allows orders to remain economic in size. It also is able to
react to changes in demand and adjust stock levels appropriately.

To work effectively, this system defines three values. These act as

triggers or warnings for the company and demand response. The
re-order level must be established as a point at which a new order is
placed. It is, therefore, an action demand. The size of the new order will
generally be the EOQ. For a selected period of time – day, week, month,
etc – it is given by

Lro = maximum usage per period (L) x maximum lead time (ro)

The minimum level must be defined so that, when stock levels reach
this point, a warning is given. Such a point should only be reached as a
result of high, unpredicted demand or problems with supply.

Lmin = re-order level – (normal usage x lead time)

The maximum level must also be defined as a safeguard against

excessively high stock levels, in the case of low demand, perhaps.

Lmax = re-order level + EOQ – (minimum usage x maximum lead

time)

For example, consider the case of a particular inventory where the

following levels have been established;

Monthly normal usage 700 units

Monthly minimum usage 500 units
Monthly maximum usage 900 units
Lead time varies from 1 to 3 months, and EOQ is 750 units.

University of
168 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

Therefore, from the above definitions, we have;

Re-order level = 900 x 3

= 2700
Minimum stock level = 2700 – (700 x 2)
= 1300
Maximum stock level = 2700 + 750 – (500 x 1)
= 2950

Periodic Review System

The system, although less common, is also an effective means of
inventory control. Stock levels are regularly reviewed and a decision is
made to bring levels back in line with a pre-agreed optimum level. In
reality, the pre-agreed level may be rarely reached as the lead time may
mean that actual stock levels are lower when new orders arrive than
were previously calculated. Thus, the ‘top-up’ order would fail to reach
the optimum level.

However, this is not necessarily an important operating problem. One

advantage of this system is that orders may be placed in such a way as to
ensure maximum gain from large order discounts, etc.

In a particular company, stock levels might be reviewed weekly and

orders placed, based on those figures to return the levels to the agreed
optimum level. The amount of each order will always be the difference
between the current stock level and the optimum level.

ACTIVITY
A retailer expects to sell about 200 units of a product per year. The storage
space taken up in his premises by one unit of this product is costed out at £20
per year. If the cost associated with ordering is £35 per order, what is the
economic order quantity given that interest rates are expected to remain close
to 10% per year and the total cost of one unit is £100.

University of
Sunderland 169
Unit 6 – Inventory Control Business Modelling for Decision Making

ACTIVITY FEEDBACK
We use the EOQ formula,

2DC
EOQ '
Ch

Here (D) = 200, Co = 35 and the holding cost Ch is given by

Ch = £20 (direct storage cost per unit per year) + £100 x 0.10 (this term the
money interest lost if one unit sits in stock for one year)

i.e. Ch = £30 per unit per year

2DC
Hence EOQ '
Ch

; 35 >
' = 2 8 200 8 @ ' 21602
.
< 30 ?

But as we must order a whole number of units we have that:

EOQ = 22

We can illustrate this calculation by reference to the diagram below which

shows order cost, holding cost and total cost for this example.

University of
170 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

With this EOQ we can calculate our total annual cost from the equation

Total annual cost = Ch(Q/2) + Co( (D)/Q) (where ‘Q’ = EOQ value)

Hence for this example we have that

Total annual cost = (30 x 22/2) + (35 x 200/22) = 330 + 318.2 = £648.2

Note: If we had used the exact Q value given by the EOQ formula (i.e.
Q=21.602) we would have had that the two terms relating to annual holding
cost and annual order cost would have been exactly equal to each other

i.e. holding cost = order cost at EOQ point (or, referring to the diagram above,
the EOQ quantity is at the point associated with the Holding Cost curve and
the Order Cost curve intersecting).

i.e. (ChQ/2) = (CoR/Q) so that Q = (2RC o / C h )

In other words, as in fact might seem natural from the shape of the Holding
Cost and Order Cost curves, the optimal order quantity coincides with the
order quantity that exactly balances Holding Cost and Ordering Cost.

Note however that this result only applies to certain simple situations. It is not
true (in general) that the best order quantity corresponds to the quantity
where holding cost and ordering cost are in balance.

ACTIVITY
The demand for an item in production in an organisation is found to be
constant at 20 items per month. The unit cost is £50, the cost of processing an
order is £60 and the holding cost is £18 per unit per year.

What are the economic order quantity and cycle length?

University of
Sunderland 171
Unit 6 – Inventory Control Business Modelling for Decision Making

ACTIVITY FEEDBACK
Listing the values we know,

D = 20 x 12 = 240 units per year

Uc = £50 per unit

Co = £60 per order

Ch = £18 per unit per year

Hence, we calculate the following:

2DC o 2 8 60 8 40
EOQ ' Q o ' ' ' 40 units
Ch 18

The amount of stock delivered during a cycle is Q, and the amount leaving is
DT, where T is the length cycle. These must be equal,

So Q = DT, giving optimal values of Qo = DTo

which is 40 = 240 x To

Hence, the cycle length, To = 1/6 years or 2 months.

Therefore, the optimal policy is to order 40 units every 2 months.

Using software to find a solution

We can also solve this problem using a suitable software package or
Excel, the input and output being shown below. Note here that the
software can deal with more complicated factors than we have
considered in the simple example given above.

A sample software screenshot is reproduced below.

University of
172 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

Note the appearance here of the figure of 20,000 relating to material cost.
This is calculated from using 200 units a year at a unit cost of £100 each.
Strictly, this cost term should have been added to the total annual cost
equation (Ch(Q/2) + Co(R/Q)) we gave above. We neglected it above as
it was a constant term for this example and hence did not affect the
calculation of the optimal value of Q. However, we will need to
remember to include this term below when we come to consider
quantity discounts.

ACTIVITY
In our previous activity, suppose for administrative convenience we ordered
20 and not 22 at each order – what would be our cost penalty for deviating
from the EOQ value?

ACTIVITY FEEDBACK
With a Q of 20 we look at the total annual cost

= (ChQ/2) + (CoD/Q)

= (30 x 20)/2 + (35 x 200/20) = 300 + 350 = £650

Hence the cost penalty for deviating from the EOQ derived value of 22 and
ordering 20 at each order is £650 - £648.2 = £1.8

Note that this is, relatively, a very small penalty for deviating from the EOQ
value. This is usually the case in inventory problems i.e. the total annual cost
curve is flat near the EOQ so there is only a small cost penalty associated with
slight deviations from the EOQ value (see the diagram above).

University of
Sunderland 173
Unit 6 – Inventory Control Business Modelling for Decision Making

This is an important point. Essentially we should view the EOQ as a

ballpark figure. That is it gives us a rough idea as to how many we
should be ordering each time. After all our cost figures (such as cost of
an order) are likely to be inaccurate. Also it is highly unlikely that we
will use items at a constant rate (as the EOQ formula assumes).
However, that said, the EOQ model provides a systematic and
quantitative way of getting an idea as to how much we should order
each time. If we deviate far from this ballpark figure then we will most
likely be paying a large cost penalty.

The above cost calculation can also be done using suitable software - see
below.

(Note: it can be seen that the values in the right-hand column for Order
Quantity, Total Ordering Cost and Total Holding Cost come directly
from our earlier calculations. The Total Material Cost and the Grand
Total Cost are found by simple addition.)

Inventory Models
Two mathematical models can be applied to inventory control. These
are known as the basic model and the adapted basic model. Both the
models assume a cyclic nature and work most effectively over a yearly
period.

The basic model also assumes the following;

The annual demand, D, is constant and all demand can be

fulfilled.
The ordering cost per cycle, Co , is constant.
The cost of each item, P, does not vary with order size.
The holding cost, Ch, is stated per item.

University of
174 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

The quantity ordered per cycle, q, is supplied instantaneously

whenever the stock level falls to zero. In other words, there is no
lead time.

These factors can then be combined in the cost equation;

Total inventory cost = total ordering costs + total holding costs

= number of orders per year x order cost + mean inventory level x

holding cost per item

;D> ;q>
= C o == @@ - C h = @
<q? <2?

It can also be shown that that the minimum inventory cost is achieved
when q takes the value of the Economic Ordering Quantity (EOQ).

2DC o
EOQ '
Ch

Note also that

The number of orders per year = demand per year / EOQ

Length of cycle, in days, = 365 / number of orders per year
Mean inventory level = EOQ / 2

ACTIVITY
A component has a demand of 1800 per year. Each order placed costs £25.
Holding costs are £ 0.22 per item per year.

1. Calculate the EOQ.

2. Calculate the number of orders placed per year.

3. Evaluate the length of the inventory cycle.

University of
Sunderland 175
Unit 6 – Inventory Control Business Modelling for Decision Making

ACTIVITY FEEDBACK
Using the EOQ formula,

2DC o
EOQ =
Ch

2 (1800)(25)
=
0.22

= 639.6 Effectively, 640

Number of orders per year = demand per year / EOQ

= 1800 / 639.6

= 2.8 Effectively, 3

Length of inventory cycle = 365 / number of orders per year

= 365 / 2.8

= 130.4 days Effectively, 131 days

The adapted basic model allows for the gradual replacement of stock over a
period of time. This allows for the case where stock is manufactured on an
internal production line and, hence, becomes continually available.

The adapted basic model, or production run model, assumes the same
conditions except for:

A production run is started every time the stock level falls to zero and stops
when q items are available. The run takes time t, and is called the runtime.

The quantity ordered each cycle is called the run size, items being supplied at a
rate of r per year.

The adapted basic model gives a formula for the Economic Batch Quantity
(EBQ)

2DC o
EBQ =
; D>
C h =1 ( @
< r?

Note also

The number of runs per year = Demand per year / EBQ

University of
176 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

The length of cycle in days = 365 / number of runs per year

The run time = (EBQ x 365) / production rate

Peak inventory level = effective replenishment rate x run time

Mean inventory level = peak inventory level / 2

ACTIVITY
A manufactured item requires a continuous supply of 2800 items per year,
which are supplied from a factory production run at a constant rate of 6000
per year. The cost of each production run is £ 32 and each item has an
associated holding cost of £ 0.15.

Calculate the following:

The EBQ.

Number of runs each year.

Length of cycle.

Run time.

Peak inventory level.

Mean inventory level.

ACTIVITY FEEDBACK
Using the equation,

2DC o
EBQ =
; D>
C h =1 ( @
< r?

We get a value of EBQ of 1497.

University of
Sunderland 177
Unit 6 – Inventory Control Business Modelling for Decision Making

Therefore, we can now calculate the following values.

Number of runs per year = 2800 / 1497 = 1.9

Length of cycle, in days = 365 / 1.9 = 192.1

Run time = (1497 x 365) / 6000 = 91.1

Peak inventory level = (6000 – 2800) / 91.1 = 35.1

Mean inventory level = 35.1 / 2 = 17.6

ACTIVITY
A company has found that demand for an item is constant at 20 units per week,
the reorder cost is £125 an order, and the holding cost is £2 per unit per week.
If suppliers guarantee delivery within two weeks, what is the best ordering
policy?

ACTIVITY FEEDBACK
The information provided gives us:

D = 20 units per week

Rc = £125 per order

Hc = £2 per unit per week

L = 2 weeks

We can substitute these values to get:

2DC o 2 8125 8 20
EOQ ' Q o ' ' ' 50 units
Ch 2

Reorder level ROL = LD = 2 x 20 = 40 units

University of
178 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

The best policy, therefore, is to place an order for 50 units whenever stock
declines to a minimum level of 40 units. We can find the cycle length, “To”,
from

Qo = DTo

Thus, To = 50/20 = 2.5 weeks

The variable cost is given by

VC o ' .2R H D/ ' .2 8125 8 2 8 20/ ' £100 per week

c c

Stock Control Systems - Keeping Track Manually

Stocktaking involves making an inventory, or list, of stock, and noting
its location and value. It’s often an annual exercise - a kind of audit to
work out the value of the stock as part of the accounting process.

Codes, including barcodes, can make the whole process much easier
but it can still be quite time-consuming. Checking stock more frequently
- a rolling stocktake - avoids a massive annual exercise, but demands
constant attention throughout the year. Radio Frequency Identification
(RFID) tagging using hand-held readers can offer a simple and efficient
way to maintain a continuous check on inventory.

Any stock control system must enable you to:

! Track stock levels.

! Make orders.
! Issue stock.

The simplest manual system is the stock book, which suits small
businesses with few stock items. It enables you to keep a log of stock
received and stock issued.

It can be used alongside a simple re-order system. For example, the

two-bin system works by having two containers of stock items. When
one is empty, it’s time to start using the second bin and order more stock
to fill up the empty one.

Stock cards are used for more complex systems. Each type of stock has
an associated card, with information such as:

! Description

University of
Sunderland 179
Unit 6 – Inventory Control Business Modelling for Decision Making

! Value
! Location
! Re-order levels, quantities and lead times (if this method
is used)

! Supplier details
! Information about past stock history

More sophisticated manual systems incorporate coding to classify

items. Codes might indicate the value of the stock, its location and
which batch it is from, which is useful for quality control.

Stock Control Systems - Keeping Track using

Computer Software
Computerised stock control systems run on similar principles to
manual ones, but are more flexible and information is easier to retrieve.
You can quickly get a stock valuation or find out how well a particular
item of stock is moving.

A computerised system is a good option for businesses dealing with

many different types of stock. Other useful features include:

! Stock and pricing data integrating with accounting and

invoicing systems. All the systems draw on the same set
of data, so you only have to input the data once. Sales
Order Processing and Purchase Order Processing can be
integrated in the system so that stock balances and
statistics are automatically updated as orders are
processed.

! Automatic stock monitoring, triggering orders when the

re-order level is reached.

! Automatic batch control if you produce goods in batches.

! Identifying the cheapest and fastest suppliers.
! Bar coding systems which speed up processing and
recording. The software will print and read bar codes
from your computer.

! Radio Frequency Identification (RFID) which enables

individual products or components to be tracked
throughout the supply chain.

The system will only be as good as the data put into it. Run a thorough
stocktake before it goes “live” to ensure accurate figures. It’s a good
idea to run the previous system alongside the new one for a while,

University of
180 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

giving you a back-up and enabling you to check the new system and
sort out any problems.

Just-in-time (JIT)
Just-in-time (JIT) is easy to grasp conceptually, everything happens
just-in-time. For example consider my journey to work this morning, I
could have left my house, just-in-time to catch a bus to the train station,
just-in-time to catch the train, just-in-time to arrive at my office,
just-in-time to pick up my lecture notes, just-in-time to walk into a
lecture theatre to start the lecture. Conceptually there is no problem
about this, however achieving it in practice is likely to be difficult!

So too in a manufacturing operation component parts could

conceptually arrive just-in-time to be picked up by a worker and used.
So we would at a stroke eliminate any inventory of parts, they would
simply arrive just-in-time! Similarly we could produce finished goods
just-in-time to be handed to a customer who wants them. So, at a
conceptual extreme, JIT has no need for inventory or stock, either of raw
materials or work in progress or finished goods.

Obviously any sensible person will appreciate that achieving the

conceptual extreme outlined above might well be difficult, or
impossible, or extremely expensive, in real-life. However that extreme
does illustrate that, perhaps, we could move an existing system towards
a system with more of a JIT element than it currently contains. For
example, consider a manufacturing process - whilst we might not be
able to have a JIT process in terms of handing finished goods to
customers, so we would still need some inventory of finished goods,
perhaps it might be possible to arrange raw material deliveries so that,
for example, materials needed for one day’s production arrive at the
start of the day and are consumed during the day - effectively
reducing/eliminating raw material inventory.

Adopting a JIT system is also sometimes referred to as adopting a lean

production system.

JIT originated in Japan. Its introduction as a recognised

technique/philosophy/way of working is generally associated with the
Toyota motor company, JIT being initially known as the “Toyota
Production System”. Note the emphasis here - JIT is very much a
mindset/way of looking at a production system that is distinctly
different from what (traditionally) had been done previous to its
conception.

Within Toyota Taiichi Ohno is most commonly credited as the

father/originator of this way of working. The beginnings of this
production system are rooted in the historical situation that Toyota
faced. After the Second World War the president of Toyota said “Catch

University of
Sunderland 181
Unit 6 – Inventory Control Business Modelling for Decision Making

up with America in three years, otherwise the automobile industry of

Japan will not survive”. At that time one American car worker
produced approximately nine times as much as a Japanese car worker.
Taiichi Ohno examined the American industry and found that
American manufacturers made great use of economic order quantities -
the traditional idea that it is best to make a “lot” or “batch” of an item
(such as a particular model of car or a particular component) before
switching to a new item. They also made use of economic order
quantities in terms of ordering and stocking the many parts needed to
assemble a car.

Ohno felt that such methods would not work in Japan - total domestic
demand was low and the domestic marketplace demanded production
of small quantities of many different models. Accordingly Ohno
devised a new system of production based on the elimination of waste.
In his system waste was eliminated by:

! Just-in-time - items only move through the production

system as and when they are needed.

! Autonomation - (spelt correctly in case you have never

met the word before) - automating the production system
so as to include inspection - human attention only being
needed when a defect is automatically detected
whereupon the system will stop and not proceed until
the problem has been solved.

In this system, inventory (stock) is regarded as an unnecessary waste, as

is having to deal with defects.

Ohno regarded waste as a general term including time and resources as

well as materials. He identified a number of sources of waste that he felt
should be eliminated:

! Overproduction - waste from producing more than is

needed.

! Time spent waiting - waste such as that associated with a

worker being idle whilst waiting for another worker to
pass him an item he needs (e.g. such as may occur in a
sequential line production process).

! Transportation/movement - waste such as that

associated with transporting/moving items around a
factory.

! Processing time - waste such as that associated with

spending more time than is necessary processing an item
on a machine.

! Inventory - waste associated with keeping stocks.

! Defects - waste associated with defective items.

University of
182 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

Case Study – General Motors

An example of the use of JIT in General Motors is given below.

General Motors (GM) in the USA has (approximately) 1700 suppliers

who ship to 31 assembly plants scattered throughout the continental
USA. These shipments total about 30 million metric tons per day and
GM spends about 1,000 million dollars a year in transport costs on these
shipments (1990 figures).

JIT implies frequent, small, shipments. When GM moved to JIT there

were simply too many (lightly loaded) trucks attempting to deliver to
each assembly plant. GM’s solution to this problem was to introduce
consolidation centres at which full truckloads were consolidated from
supplier deliveries.

This obviously involved deciding how many consolidation centres to

have, where they should be, their size (capacity) and which suppliers
should ship to which consolidation centres (suppliers can also still ship
direct to assembly plants).

As of 1990 some 20% by weight of shipments go through consolidation

centres and about 98% of suppliers ship at least one item through a
consolidation centre.

All this has been achieved without sacrificing the benefits of JIT.

KEY POINT
JIT outline points

Originated in Japan.

Often said Japanese industry works - just-in-time, Western industry works -

just-in-case.

JIT is also known as stockless production or lean production.

JIT is a suitable production system when:

! Have steady production of clearly defined standard products.

! A reasonable number of units made.

! A high value product.

! Have flexible working practices and a disciplined workforce.

University of
Sunderland 183
Unit 6 – Inventory Control Business Modelling for Decision Making

! Short setup times on machines.

! Quality can be assured, e.g. zero defects either though good

working practices or though a cost penalty.

It is often said that:

! Materials Requirements Planning (MRP) = a ‘Push’

system

! JIT = a ‘Pull’ system

Some experts consider this is an incorrect analysis - MRP is a system

based on fulfilling predicted usage in a set time period.

JIT is a system based on actual usage - parts of the production system

are “linked” together via kanbans as the system runs.

It is this linkage that is the distinguishing difference between MRP and

JIT - JIT is a dynamic linked system, MRP is not

KEY POINT
JIT philosophy

! Elimination of waste in its many forms.

! Belief that ordering/holding costs can be reduced.

! Continuous improvement, always striving to improve.

Elements of JIT

! Regular meetings of the workforce (e.g. daily/weekly).

! Discuss work practices, confront and solve problems.

! An emphasis on consultation and cooperation (i.e. involving

the workforce) rather than confrontation.

! Modify machinery, e.g. to reduce setup time.

! Reduce buffer stock.

University of
184 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

! Expose problems, rather than have them covered up.

! Reveal bad practices.

! Take away the “security blanket” of stock.

JIT need not be applied to all stages of the process. For example we
could keep large stocks of raw material but operate our production
process internally in a JIT fashion (hence eliminating work-in-progress
stocks).

Classic JIT diagram

The classic JIT diagram is as below. There the company (the boat) floats
on a sea of inventory, lurking beneath the sea are the rocks, the
problems that are hidden by the sea of inventory.

The rocks are the problems hidden by the sea of inventory

If we reduce the inventory level then the rocks become exposed, as

below.

University of
Sunderland 185
Unit 6 – Inventory Control Business Modelling for Decision Making

Now the company can see the rocks (problems) and hopefully solve
them before it runs aground!

One plan to expose the problems is simply to:

1. Make a large amount of finished goods stock to keep the

customers supplied.
2. Try running the production system with less inventory to expose
problems.
3. Revert to the original levels of inventory until you have had time
to fix the problems you exposed.
4. Repeat the above - hence continuous improvement.

Benefits of JIT
The benefits of JIT are:

! Better quality products.

! Quality the responsibility of every worker, not just
quality control inspectors.

! Reduced scrap and rework.

! Reduced cycle times.
! Lower setup times.
! Smoother production flow.
! Less inventory, of raw materials, work-in-progress and
finished goods.

! Cost savings.
! Higher productivity.
! Higher worker participation.

University of
186 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

! More skilled workforce, able and wiling to switch roles.

! Reduced space requirements.
! Improved relationships with suppliers.

However, you should be absolutely clear that implementing a JIT

system is a task that cannot be undertaken lightly. It will be expensive in
terms of management time and effort, both in terms of the initial
implementation and in terms of the continuing effort required to run the
system over time.

Suppliers
Suppliers can be crucial to JIT success.

Supplier gets:

! Long-term, guaranteed, contract.

! A good price.
! Steady demand.
! Minimal paperwork (e.g. use electronic means to order -
such as email or Web or electronic data interchange, EDI).

In return the supplier agrees to

! Quality components (e.g. zero defects).

! Guaranteed delivery times.
! A “partnership” with its customer.
! Contingency plans to cope with disruptions, common
disruptions might be:

( the effect of bad weather

( a truck drivers strike blocking roads/ports
( a flu outbreak reducing the supplier’s workforce

Supplier selection criteria:

! Close to production plant (else potential transportation

delays).

! Good industrial relations (“involvement”, “value”,

“dignity”, “ownership”), no strike deals.

University of
Sunderland 187
Unit 6 – Inventory Control Business Modelling for Decision Making

! You believe that the suppliers can meet their promises

with respect to the list of factors that they are agreeing to.

With suppliers satisfying these criteria you can reduce the total number
of suppliers, indeed it seems logical so to do. If you had five suppliers
meeting all these criteria, why do you need five? Obviously you might
decide to have more than one supplier for safety reasons. Even the best
run supplier can suffer a factory fire or an earthquake, but probably no
more than two or three suppliers.

As an illustration of this, in 1997 Toyota was affected by a fire at a

supplier of brake parts that cost the company an estimated $195 million
and 70,000 units of production. The fire was at a plant that was the sole
supplier of brake parts for all but two Toyota models and forced the
company to shut its 18 assembly plants in Japan for a number of days.
As a result Toyota embarked on a review of components that were
sourced from a single supplier.

Having a single supplier may be attractive in cost terms, but one does
need to balance the risk (albeit a low probability risk - perhaps a fire
every 100-250 years, say) against the cost savings.

Case Study
Andrea Jones is managing director of Liversedge-based Systems
(Telecoms) Limited, a business specialising in the next-day delivery of
refurbished telecommunications equipment.

This is her description of the situation faced by her company:

“We bought our computer system with its accounting software, Sage
Line 100, when we were turning over £500,000 a year. It cost a lot of
money but I got it on a five-year lease and I only paid £50 a week. I
couldn’t have got anybody to do the stock work for £50 a week.

“I chose this system because I wanted something that integrated all my

accounting functions - my stock control, my buying … basically, to have
everything under one roof, as it were. And, importantly, I wanted
barcoding. I did a lot of phoning around software companies before
making my choice.

“As an item comes in it gets barcoded and then it’s logged on to the
system under a purchase order with the serial number, stock code and
details of the product. The product then goes down to our test room for
refurbishing.

“The system can tell you whether any item has been tested or not and
exactly where it’s located in the warehouse.

University of
188 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

“For the refurbished products side of my business there’s no such thing

as having too much stock. We can’t ask people to wait a week if they
want something so I buy anything that comes up for sale if it’s cheap
enough and I know it will move. I wouldn’t buy stuff if there was no
demand for it but that doesn’t happen. I know my market very well.

“We sell new installation products too and we re-stock that on a

demand-led basis. We set minimum and maximum stock levels on the
Sage system and when the stock hits the minimum level the screen
lights up telling me to reorder. But I prefer to print a report on stock
levels every Friday. That tells me what we’re low on and I always top up
to the maximum level.

“We use lots of paper, lever arch files, pens and ink cartridges.
Everything is kept on bookshelves and I can see instantly what’s in and
what we’re running low on. I do a stationery order every couple of
weeks or so - though I try to time it with when there are special offers in
the stationery catalogue I use.

“When I get home I can click on two buttons and I’m basically sitting in
my office. This means I can keep an eye on my stock at all times and I do
a lot of bidding for products from home, for example on eBay. I can also
complete purchase orders from home.”

REVIEW ACTIVITY
News vendor problem

Consider a news vendor who stands on the street and sells an evening paper.
How many copies should he stock?

He sells the paper to his customers for 35 pence a copy. He pays his supplier
20 pence a copy, but any unsold copies can be returned to the supplier and he
gets 10 pence back. This is known as a salvage value.

Assume that his demand for copies on any day is a Normal distribution of mean
100 and standard deviation 7.

REVIEW ACTIVITY FEEDBACK

Before we can compute the amount he should order we need to work out his
shortage cost per unit - how much does he lose if a customer wants a copy and
he does not have a copy available?

University of
Sunderland 189
Unit 6 – Inventory Control Business Modelling for Decision Making

As a first analysis he loses his profit (= revenue - cost = 35 - 20 =15) so we can

estimate his shortage cost (opportunity cost) as 15 (this ignores any loss of
goodwill and any loss of future custom that might result from a shortage).

Using a software package, we might get:

This tells us he should stock 104.7 (say 105) copies of the paper. This service
level of 75% means that, on average, he will be able to completely supply his
customers on 75 days in every 100, i.e. 3 days out of 4. The remainder of the
time (1 day out of 4) he will experience shortages, some customers will not be
able to buy a copy from him as he will have run out. (Note that in this example
the software has calculated an optimal order quantity based on the statistical
information provided and the arbitrarily chosen service level of 75%.)

More sophisticated variants of this simple model can be used, for example, to
decide how many copies of a magazine to have on a shelf in a newsagent (such
as W.H.Smith).

Note that there is an important conceptual difference between this news

vendor problem and the EOQ/discount problems considered above. In those
EOQ/discount problems we had a decision problem (how much to order)
even though the situation was one of certainty - we knew precisely the rate at
which we used items. In the news vendor problem, if we knew for certain how
many customers will want a paper each day then the decision problem
becomes trivial (order exactly that many).

In other words:

! for the EOQ problem we had a decision problem even though

there was no uncertainty.

! for the news vendor problem it was only the uncertainty that
created the decision problem.

University of
190 Sunderland
Business Modelling for Decision Making Unit 6 – Inventory Control

Summary
Effective inventory control depends on the nature of the business and
the market in which it operates. This unit has described a range of
inventory control methods, concentrating on optimal stock levels and
the Just-in-time model.

Throughout the unit, the use of manual methods has been

demonstrated. However, reference has been made to the use of
appropriate software packages that are specifically designed for
inventory control. These obviously make life considerable easier for
managers and it is recommended that students try to gain access to at
least one of them (or a suitably designed Excel spreadsheet) in order to
appreciate their value.

University of
Sunderland 191
Unit 7

Sampling Theory and

Statistical Inference

Introduction
Setting up and testing hypotheses is an essential part of statistical
inference. In order to formulate such a test, usually some theory has
been put forward, either because it is believed to be true or because it is
to be used as a basis for argument, but has not been proved, for example,
claiming that a new drug is better than the current drug for treatment of
the same symptoms.

This unit provides a basic introduction to this advanced section of

statistics, relating it to business applications where appropriate.

As with all statistical concepts, the student is strongly advised to carry

out their own reading from suitable textbooks in order to gain a more
thorough understanding of what can be a complicated subject area.

In this chapter, we will be making use of the ‘Normal Distribution’

tables (the ‘Z’ tables), the ‘Students ‘t’ distribution’ tables and the
‘chi-squared (A 2) distribution’ tables.

Hypothesis Testing
If we have some idea of a parameter’s value, (prior), we can use the
techniques of hypothesis testing to examine whether or not, our prior is
substantiated by the sample data.

e.g. let x (mean life) of a longlife lamp = 2000 hours (prior),

as stated by a company who manufactures them.

Suppose a consumer periodical decides to test a sample of 1000 lamps

and concludes that the sample mean = 1500 hours.

i.e. xs = 1500 hours

The consumer periodical might reject the company’s claim of x = 2000

hours, based upon the results of their tests. On the other hand, the

University of
Sunderland 193
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

consumer periodical may test another sample of 1000 lamps, and

conclude that the sample mean = 2100 hours.

i.e. xs = 2100 hours

In this case, the consumer periodical might accept the company’s claim
of x = 2000 hours, based upon the results of this second test.

To simplify our work within hypothesis testing, we refer to the prior as

the hypothesis and to the decision, as the decision rule.

e.g. if x = 2000 hours, then if xs # 1950 we accept the hypothesis.

This simply means that if the population mean is equal to 2000 hours
and, using a sample, we find the sample mean to be greater or equal to
1950 hours, then we accept the hypothesis, i.e. the manufacturers claim
to a population mean of 2000 hours is true.

Suppose xs = 1925 hours (where by chance, the sample contains an

unusually high number of short life lamps), then the decision to reject
the hypothesis is technically incorrect.

Similarly, suppose xs = 2100 hours (where by chance, the sample

contains an unusually high number of long life lamps), then the
decision to accept the hypothesis is technically incorrect.

Hence, two types of error exist. They are known as type I and type II
errors. We can represent these errors in the figure below.

Decision Hypothesis Correct Hypothesis Incorrect

Accept Hypothesis Correct Type II Error

Reject Hypothesis Type I Error Correct

We can relate this principle to an example. Suppose a defendant is on

trial for a very serious crime. Consider the following figure.

Decision Defendant Innocent Defendant Guilty

Proven Innocent Correct Type II Error (set free)

Proven Guilty Type I Error (execution) Correct

The probability of committing a type I error is denoted as ) and the

probability of committing a type II error is denoted as &.

The value of ) and/or &, is/are known as the significance level of the
test. Significance levels can be set to any value but, the most common
used levels are 0.1, 0.05 or 0.01

Hence, when ) = 0.05, then the probability of obtaining a type I error is

0.05.

University of
194 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

The higher the significance level, the larger is the probability of

committing a type I error.

Hypothesis

Usually written as Ho, e.g. Ho : x = 2000 hours. It is called the null

hypothesis. The decision rule is usually written as the alternative
hypothesis (H1).

e.g. H1 : x < 2000 hours (i.e. the mean is less than 2000 hours).

We usually present the hypotheses as follows :

e.g. Ho : x = 2000 hours; H1 : x < 2000 hours

The null hypothesis should always enclose the equality sign. E.g. Ho : x
= 2000 or Ho : x # 2000 or Ho : x " 2000

Errors

Given, for example, Ho : x # 2000 hours; H1 : x > 2000 hours

dist 0 implies the distribution of x when Ho : x " 2000 should be

accepted.

dist 1 implies the distribution of x when H1 : x > 2000 should be

accepted, or when Ho is incorrect and should be rejected.

This shown in the following figure.

University of
Sunderland 195
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

Suppose the critical value is 2050 hours (implying that values greater
than 2050 hours should cause Ho to be rejected).

In this case, then values of x under area A (which could rightfully have
originated under distribution 0), will be high enough to cause a rejection
of Ho, when in fact it is correct. This would be a type I error.

Similarly, values of x under area B (which could rightfully have

originated under distribution (1), will be sufficiently low enough to
cause an acceptance of Ho, when in fact it is incorrect. This would be a
type II error.

If we attempt to reduce (increase) the critical value, then area B (A)

increases. The only method available to us to reduce both, is to increase
the sample size.

One Tailed Hypothesis Tests of the Population

Mean (Right Tail Tests)
Consider the following hypothesis : H o : x " xo

H 1 : x > xo

(let x = B and xo = Bo and X is the sample mean)

Hence .. Ho : B " Bo and H1 : B > Bo

The Standardised Test Statistic:

.X ( B / ~ t
0
(n ( 1), )
s2
n

University of
196 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

t(n-1),) represents the value of t which leaves an area ‘)’ in the ‘t’
distribution.

If
.X ( B / C t
0
(n ( 1), )
s2
n

then reject Ho at the significance level ‘)’

Rejecting Ho when it is correct, will give a type I error.

Accepting Ho when it is incorrect, will give a type II error.

We can summarise the procedure for the hypothesis test (right tailed) :

1. Set the value of Bo in the null hypothesis (Ho)

2. Set up the hypothesis e.g. H o : B " Bo and H1 : B > Bo
3. Select a value for ‘)’ (the significance level) e.g. 0.01, etc.
4. Find t(n-1),) from the ‘t’ tables, where n = sample size.
5.
Find
.X ( B / 0

s2
n
where X = sample mean, s 2 = sample variance, n = sample size,
Bo = hypothesised value.
6. Compare t(n-1),) with the solution found in (5) above.
7. Given the hypotheses as stated as an example in (2) above, then if
the solution to (5) above, is less or equal to the value found in the
‘t’ tables, as in (4) above, we can accept H o at the significance
level. Otherwise, we would reject H o.

ACTIVITY
The government health department release the following statement: ‘The
average wait for an operation does not exceed 20 weeks’.

The National Patients Care Society, are sceptical and decide to test this
statement. Their findings are as follows :

sample average waiting time = 21.6 weeks, sample variance = 22.5,

and number of patients in the sample = 25

Given this data, is the government correct? (Use ) = 0.1 and ) = 0.05)

University of
Sunderland 197
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

ACTIVITY FEEDBACK
Ho : B0 " 20 and H1 : Bo > 20

Given two different significant values and constant information, we assume the
solution of the test statistic will be constant for both significance levels.
Therefore, we only need to calculate the test statistic once and compare its
value to the values obtained from the tables at ) = 0.1 and ) = 0.05.

n = 25; X = 21.6; s2 = 22.5. Hence, the test statistic ..

.X ( B / 0
=
.216. ( 20/ = 1.6865
s2 ; 22 .5 >
= @
n < 25 ?

at ) = 0.1 ...

t(n-1),) = t(25-1),0.1 = 1.3178 from ‘t’ tables.

and 1.6865 > 1.3178 ==> reject Ho @ ) = 0.1

at ) = 0.05 ...

t(n-1),) = t(25-1),0.05 = 1.7109 from ‘t’ tables.

and 1.6865 < 1.7109 ==> accept Ho @ ) = 0.05

At ) = 0.1 (0.05) the government are incorrect (correct) with their statement.

p - Values

KEY POINT
If Ho is true, the p - value is the probability of obtaining a sample result i.e. a
standardised test statistic value, which is more extreme than the one obtained.

This means the p - value is the smallest level of significance at which the given
sample observations would lead us to reject Ho.

University of
198 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

Procedure:

1. Set the value of Bo in the null hypothesis (Ho),

2. Set up the hypothesis e.g. H o : B " Bo and H1 : B > Bo
3. Select a value for ) (the significance level) e.g. 0.01 etc.
4.
Find
.X ( B / 0

2
s
n
where X = sample mean, s 2 = sample variance, n = sample size,
Bo = hypothesised value.
5. Determine the p - value associated with this test statistic (see
below).
6. Accept Ho if p - value " ), otherwise reject Ho.

Calculating p - values

To find the area or probability required, we have to interpolate.

1,6865 ( 1. 3178
' 0. 9379
1,7109 ( 1. 3178

Hence p - value = 0.1 - 0.9379(0.1 - 0.05) ==> 0.1 - 0.0469 giving the p -
value = 0.0531 and means the probability of obtaining a sample mean
which gives a standardised statistic greater than 1.6865 when B " 20, is
0.0531.

University of
Sunderland 199
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

Hence if p - value > ), then we accept Ho i.e. 0.0531 > 0.05 ==> accept Ho
@ ) = 0.05

What this means, is that we could have accepted Ho at any value up to )

= 0.0531.

Left Tailed Tests

These are similar in many ways to the right tailed tests. The usual
exception is that the hypotheses are defined as :

Ho : B # Bo and H1 : B < Bo

Consider the following activity on left tailed tests.

ACTIVITY
The average shelf life of a consumable is 60 days. A sample of n = 20 is taken
and tested. The results show that the sample mean life is 58 days, with a
variance of 13. Is the average life of 60 days justified ?

ACTIVITY FEEDBACK
n = 20; X = 58 and s2 = 13

Ho : B # 60 days; H1 : B < 60 days

let ) = 0.01, then t(n-1),) = t19,0.01 = -2.5395 (from t tables)

The minus sign indicates that we are in the left hand side of the distribution.

Using
.X ( B / 0
=
.58 ( 60/ = -2.4807
s2 ; 13 >
= @
n < 20 ?

and -2.4807 > -2.5395. ==> accept Ho @ ) = 0.01 i.e. B >= 60 days

University of
200 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

Two Tailed Tests

In this case we want to test:

Ho : B = Bo and H1 : B D Bo

i.e. in H1, B is greater or less than the value for Bo implying that Ho will
exist in between two confidence limits.

Note, it is quite easy to commit a type I error in this case.

KEY POINT
A small point on hypothesis testing :

When attempting to decide the type of test (i.e. right tailed, left tailed or two
tailed), use the alternative hypothesis (H1) as a guide. In most cases ..

H1 indicating the greater than (>) sign, means a right tailed test,

H1 indicating the less than (<) sign, means a left tailed test,

H1 indicating the not equal (D) sign, means a two tailed test.

The direction of the sign (e.g. > implies a right tailed test), indicates the
type of test.

Note: the equality sign (=) should always appear within the null
hypothesis (Ho).

i.e. Ho : x = value or Ho : x # value or Ho : x " value.

Procedure for the solution of two tailed tests:

1. Set the value of Bo in the null hypothesis (Ho).

2. Set up the hypothesis e.g. H o : B = Bo and H1 : B D Bo
3. Select a value for ) (the significance level) e.g. 0.01, etc.
4.
Find
.X ( B / 0

s2
n

University of
Sunderland 201
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

where X = sample mean, s 2 = sample variance, n = sample size,

Bo = hypothesised value.
5. Determine the p - value associated with this test statistic,
6. The decision rule is therefore:

Accept Ho if -t(n-1),) "

.X ( B / " t
0
(n-1),) otherwise reject Ho
s2
n

ACTIVITY
Assume a salary mean of £ 15000 pa, n = 28 employees,

(sample mean) X = £14650; s2 = 950 000 (sample variance)

We have reason to believe that salaries will be greater or less than £15000 (i.e.
not equal to £15000 pa). Is this true?

ACTIVITY FEEDBACK
Ho : B = £15000; H1 : B D £15000 (two tailed test)

using ) = 0.1 and on a two tailed test, we need to halve the significance level
and apply each half to each tail. Therefore, )/2 = 0.05

-t(n-1),)/2 = -t(27),0.05 = -1.7033 (‘t’ tables)

t(n-1),)/2 = t(27),0.05 = 1.7033 (‘t’ tables)

Using
.X ( B / 0
=
.14650 (15000/ = -1.9001
s2 ; 950000 >
= @
n < 28 ?

and -1.9001 is outside the limits of 1 1.7033 => reject Ho @ ) = 0.1 concluding
that we have evidence to suggest that the salaries are not equal to £15000 pa.

University of
202 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

The Probability of a Type II Error

Previous decision rules on hypothesis testing have provided us with
solutions based upon the significance of ), implying the probability of
making a type I error. However, we may need to know what the
probability of a type II error (&) may be.

ACTIVITY
Maximum mortgage from bank = 2.5 * Gross annual earnings of house buyer.

Recent competition has put pressure on this upper limit.

A sample of 40 recent loans provides:-

Earnings ratio : X = 2.55 (Sample mean) Sample Variance : s2 = 0.196

Question: Is the 2.5 earnings ratio rule being breached?

ACTIVITY FEEDBACK
Ho : Bo " 2.5; and H1 : Bo > 2.5 (one tailed RHS test)

let ) (significance level = 0.05)

test statistic:
.X ( B / ~ t
0
(i)
(n (1),)
s2
n

and t39,0.05 = 1.6849 (from ‘t’ tables)

and
.2 .55 ( 2 .5/ = 0.7143
; 0196
. >
= @
< 40 ?

and 0.7143 < 1.6849 implying we accept Ho @ ) = 0.05 and this

suggests that the mean lending to earnings ratio is NOT greater than
2.5

NOTE: if Ho is correct then Bo = 2.5 and we can rearrange (i) above to give :

University of
Sunderland 203
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

s2 ; 0196
. >
X = tn-1,) x + Bo or X = 1.6849 x = @ + 2.5 = 2.618
n < 40 ?

2.618 is the CRITICAL VALUE for acceptance of Ho @ ) = 0.05

Note. X " 2.618 implies we accept Ho @ ) = 0.05

X > 2.618 implies we reject Ho @ ) = 0.05

A type II error occurs when X " 2.618 and Ho : Bo = 2.5 is accepted, when in
fact the earnings ratio is actually > 2.5 (2.55 in this case). A type II error can
only be made if Ho is incorrect. Hence to calculate the probability of a type II
error :

(a) The correct value for Bo must be specified,

(b) In a right tail test, we need only consider correct values for Bo > 2.5.

Example ..

Suppose Bo = 2.57 then the probability of X > 2.618 implies the probability of
a type II error.

University of
204 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

Areas to the left hand side of the vertical line at 2.618, indicate the probability
of making a type II error, when Ho : Bo " 2.5 and when actual mean values are
(in fact) :

a. Bo = 2.57,

b. Bo = 2.64,

c. Bo = 2.68.

Bo (the mean value) occurs at the central point in all three respective
distributions.

Solution using (a) as an example:-

The area to the left of X (sample mean) = 2.618, which depicts the sampling
distribution of X, when Bo = 2.57, represents the probability of making a type II
error.

Area of LHS, i.e from the start of the left hand side of the distribution to Bo =
2.57 plus from Bo = 2.57 to the sample mean of 2.618

Clearly, the LHS area = 0.5 (since the distribution is symmetrical, hence half
the distribution is 50 % or 0.5),

The RHS value is calculated via :-

.X ( B / 0
=
.2 .618 (12 ,57/ = 0.686
s2 ; 0196
. >
= @
n < 40 ?

Using tables (Area of standardised normal tables) 0.686 (in tables), = 0.2483
(i.e. the area from 2.618 to the right hand side end of the distribution) and
implies:

University of
Sunderland 205
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

and the RHS area (from Bo = 2.57 to the sample mean of 2.618) = 0.5 - 0.2483
= 0.2517

Therefore, required area (from the start of the left hand side of the
distribution to the sample mean of 2.618) = 0.5 + 0.2517 or 0.7517

Hence, the probability of committing a type II error when Bo = 2.57 and the
sample mean is 2.618, is 0.7517 = &

Power of a Test

KEY POINT
The power of a test is defined as the probability of rejecting (correctly), the
null hypothesis when it is “false” and occurs at 1 - & where & = probability of a
type II error.

Determining Sample Size

Suppose we set values for ) and &. What would be the required sample
size (n) to achieve these values?

We would use the following formulae :

. /
2
: Z) - Z&
n'
d

Where : : = population standard deviation,

d = (Bo - B) i.e. the difference between the null

hypothesis value of B and the alternative possible value,

Z) = the value of Z which leaves ) in the tail of the

distribution,

Z& = the value of Z which leaves & in the tail of the

distribution.

University of
206 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

ACTIVITY
Suppose the lending to earnings ratio of borrowers has a mean of 2.5. Suppose
now, that it is decided to increase the mean to 2.6. If P(type I error) = 0.05, and
is kept constant, and P(type II error) " 0.1, then what should be the sample size?

ACTIVITY FEEDBACK
) = 0.05, &. = 0.1 (max); for ), Z0.05 = 1.645; for &., Z0.1 = 1.282;
d = 2.5 - 2.6 = - 0.1

: = UNKNOWN! let us use the ‘best’ likely value, e.g. historical evidence
(empirical evidence). Assume :2 = 0.196, therefore, : = 0196
. = 0.4427

and substituting into the formulae gives:

2
0.4427.1645
. . /
-1282
n' ' (167.9
(1

and we can ignore the negative value in favour of an actual value (absolute
value) which means a sample size of n = 168.

Using this value, then the new critical value is tn-1,) or t167,0.05 = 1.645 and we
would reject Ho if our ANSWER > 1.645

Hypothesis Tests of the Population

Proportion (p)
Bp = E(p) = $ ; :2p = $(1 - $)/n

Bp = mean of sample proportion (p),

E(p) = expected value of sample proportion,

:2p = variance of sample proportion,

n = number of observations/experiments, etc.

University of
Sunderland 207
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

There are three sets of hypothesis needed to consider:

Ho: $ # $o ; H1: $ < $o Left tailed test

Ho: $ " $o ; H1: $ > $o Right tailed test
Ho: $ = $o ; H1: $ D $o Two tailed test

A similar procedure is followed, to that of the population mean. ($o is

some arbitrary value that we may require to test)

Procedure for solving Hypothesis tests of Population Proportion :

(1) Choose a value for Bo for Ho.

(2) Choose ) (the significance level), accordingly.
(3) Set up and calculate the standardised test statistic:

Z'
.p ( $ 0 / ' p ($0
:p $ 0 .1 ( $ 0 /
n
(4) If using p-value approach, find p-value associated with the value
of the standardised test statistic.
(5) If using classical approach, find Z) or Z)/2 depending upon
whether a one or two tailed test is employed.
(6) Compare : either the p-value with ), or the critical value (Z) with
the standardised test statistic.
(7) Use the appropriate decision rule to either accept or reject H o.

Example

Right tailed test - Use of Classical Approach:

Suppose a survey suggests that only 75% of the households in the

sample recognise a company name. After a three month publicity
campaign, the result of a survey suggests that 80% of households in the
(new) sample now recognise the company name. Was the publicity
campaign successful? (Let n = 100)

Solution : Ho: $ " 0.75; H1: $ > 0.75; Let ) = 0.05

Therefore, Z) = Z0.05 = 1.645 (tables)

Standardised test statistic is

.p ( $ 0 / or
0.8 ( 0.75
' 1.1547
:p 0.75.1 ( 0.75 /
100

University of
208 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

and 1.1547 < 1.645, implying accept Ho @ ) = 0.05 i.e. there is not enough
evidence to suggest that the publicity campaign was successful.

Example

Right tail test - Use of p-value approach

In this case, Ho: $ " $o ; H1: $ > $o

Assume the same problem as above, i.e. that Ho: $ " 0.75 and H1: $ >
0.75

The Classical approach for the right tail test, gave us a critical value of
1.1547. To use the p-value approach, we need to find a value greater
than the critical value of 1.1547.

From Z tables, 1.1547 = 0.3764. Hence, we need to know the area under
the curve, in the right hand tail. See the figure below.

When Z = 1.1547, the shaded area represents 37.64 % of the total shaded
area under the curve.

We find the area to the right of Z = 1.1547 via 0.5 - 0.3764 = 0.1236,

that is, the probability of obtaining a sample proportion which gives a

value for the standardised test statistic greater than 1.1547, when B "
0.75 is 0.1236 (p-value)

Decision rule: accept HO if p-value # ), and since the actual p-value =

0.1236 ( > ) = 0.05), then accept Ho @ ) = 0.05.

We can continue to accept Ho up to ) = 0.1236 by re-arranging

University of
Sunderland 209
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

.p ( $ 0 /
:p

$ 0 .1 ( $ 0 /
to (p-value) * -$0
n

; ; 0.75.1 ( 0.75 / > >

i.e. = 0.1236= @ @ + 0.75 = 0.7554
= = 100 @@
< < ??

and we can accept Ho up to and including p = 0.7554

Left tailed tests of $

Similar to right tailed tests except the hypothesis takes the form of
Ho: $ # $o and H1: $ < $o (Similar analysis.)

Two tailed tests of $

Ho: $ = $o ; H1: $ D $o

Decision rule for:

(1) CLASSICAL APPROACH :

Accept Ho if - Z)/2 " (p - $o)/:p " Z)/2
(2) P-VALUE APPROACH :
Accept Ho if p-value # )

ACTIVITY
Suppose that prior to a large scale programme of privatisation, the
government believe that the proportion of males between 30-54 years of age,
who owned shares, was 0.1. A journalist takes a random sample of 50 males in
that age group, after the privatisation programme, and finds the sample
proportion of share holders to be 0.15. Has there been a significant change in
the proportion of share-owners?

University of
210 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

ACTIVITY FEEDBACK
Ho: $ = 0.1 ; H1: $ D 0.1 ; Let ) = 0.05 therefore )/2 = 0.025

and 0.025 from tables = 1.96.

The standardised test statistic is:

.p ( $ / 0
or
0.15 ( 0.1
' 1.178
:p 0.1.1 ( 0.1/
50

The decision rule is:

Accept Ho if - Z)/2 "

.p ( $ / 0
" + Z)/2
:p

and - Z)/2 = - 1.96 ; Z)/2 = + 1.96 and (clearly), -1.96 < 1.178 < 1.96

Therefore, we accept Ho @ ) = 0.05 (or )/2 = 0.025), i.e. there is not enough
evidence to suggest ownership of shares has changed from 0.1 of the population.

Hypothesis Tests of the Population Variance (:2)

This involves a similar procedure as in previous analysis:

(1) Choose a value for Ho: :2.

(2) Select ) (the significance level) = (whatever is appropriate).
(3) Set up and calculate the test statistic :

. n ( 1 /s 2
:2
(4) For CLASSICAL APPROACH, use CHI-SQUARE tables to
determine the critical value of A 2n-1,) (right tail test) or A 2n-1,1-)
(left tail test), or A 2n-1,1-)/2 and A 2n-1,)/2 (two tailed test).
(5) For p-value approach, determine the p-value associated with the
value of the standardised test statistic.
(6) Compare either :
(a) the critical A 2 value to the test statistic value OR

University of
Sunderland 211
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

(b) the p-value with ),

(7) Use the appropriate decision rule to accept or reject the H o value
of :2.

ACTIVITY
Right Tailed Test of :2

The manager of a supermarket expects customers to wait no longer than 10

minutes at the check-outs.

A sample of 30 customers reveals:

Xs = 8 mins; s2 = 16 (and sample standard error (SE) = 4)

The manager has previously set s2 = 9 (implying sd or STD error [SE] = 3 mins)
as the maximum error acceptable.

Does s2 = 16 imply a significant increase in the variance of waiting time?

ACTIVITY FEEDBACK
Ho: :2 " 9 ; H1: :2 > 9 ; Let ) = 0.01

Using the CLASSICAL method:

.n (1/s 2

=
.30 (1/16 = 51.56
:2 9

from A 2 tables, A 2n-1,) = A 229,0.01 = 42.56

for RHS tail test, if

.n (1/s 2

" A 2n-1,) then accept Ho

:2

Clearly, 51.56 > 42.56, therefore we REJECT Ho @ ) = 0.01, i.e. evidence

exists to support the theory that a significant increase in the variance of the
waiting time exists.

University of
212 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

Left Tailed Tests of the Population Variance :2

These are similar to the right tailed tests, except that the test statistic is
compared to the A 2n-1,1-) value,

i.e. Ho: :2 # :o2 ; H1: :2 < :o2 and we accept Ho if

. n ( 1 /s 2 # A 2n-1,1-)
:2

University of
Sunderland 213
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

Two Tailed Tests of the Population Variance :2

Similar to other tests except that the acceptance of Ho depends upon

critical values on the LEFT and RIGHT side of the probability
distribution ..

Ho : : 2 = : o 2 ; H1 : : 2 D : o 2

Accept Ho if A 2n-1,1-)/2 "

. n ( 1 /s 2 " A 2n-1,)/2
: 2

Confidence Intervals

Principles of a Confidence Interval

Generally, populations on the whole are too large to study effectively.
We usually select samples, and apply some statistical technique(s) to
them, and draw inferences about the population.

KEY POINT
Inferences imply estimates and we can identify two common estimators:

1. A point estimate, i.e. a discrete value (identified at a point),

University of
214 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

2. An interval estimate, i.e. a specified range containing the unknown

parameter(s).

The interval estimator uses the point estimator in the initial

estimation.

This part of the unit considers the interval estimate.

The interval estimate identifies two limits (upper and lower) and these
limits contain the unknown parameter(s). The accuracy of the interval
containing the unknown parameter is sometimes associated with some
measure of confidence. Hence, confidence intervals and they are
defined as an interval estimate, existing between an upper confidence
limit (clu) and a lower confidence limit (cll).

Probability is used to measure the confidence. For example, if a

confidence interval exists with a level of confidence of 99%, then we can
be 99% certain that our estimate will exist within the clu and cll. Or, in
other words, the probability of our estimate occurring within the clu
and cll, is 0.99.

We calculate the upper and lower limits by identifying the point

estimator, then we add (subtract) a specified value to determine the
upper (lower) limit. We make use of the central limit theorem (see notes
eea15) to determine our confidence intervals. If we use the standard
normal tables, then for a 95% confidence limit, we can determine that
our estimate (B) will lie between X 1 1.96 standard errors (:), i.e.

X - 1.96: " B " X + 1.96:

Hence, 95% of repeated sampling, will supply sample means within an

interval of X 1 1.96: which will contain B. Later on in these notes, we
will investigate the determinance of 1 1.96:.

Similarly, for a 99% confidence limit :

X - 2.575: " B " X + 2.575:

Interpretation of a Confidence Interval

Assume repeated sampling of students’ grades produces a mean mark
within 58 to 63 percent. Also assume that the confidence interval
significance value is set to 95%. Interpretation can take on two methods:

1. We are 95% certain that the unknown parameter (the population

mean) lies between 58 and 63 percent.

University of
Sunderland 215
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

Since the laws of probability assume either 0 (fail) or 1 (success),

then the unknown parameter will either exist or not within the
interval. The 95%, tells us that we are 95% certain (confident)
that it will appear within the interval.
2. If we construct an unlimited number of confidence intervals, then
95% of them will contain the unknown parameter.
Repeated sampling, more than likely, will produce different
means. Hence, based on these means, the construction of
confidence intervals in repeated sampling, will produce slightly
varying results.

Confidence Interval for the population mean (large

samples)
1. A Known Population Variance (:2) (or standard deviation (:)) :

For purposes of clarification, we will assume the mean is unknown.

Suppose a manufacturer has some idea of the variance of his output, but
has no idea of the mean. Also, let us assume that the manufacturer
decides to calculate a confidence interval for the mean. Previous
analysis suggests that we add (subtract) a fixed amount from the mean
to determine clu (cll). Since we do not know the value of the mean, the
problem is what is the value of the fixed amount we need to add
(subtract). The value of fixed amount will depend upon the
manufacturer’s precision. i.e. let us assume that he wants to be 95%
certain that the interval contains the actual population mean.

Analysis from other parts of this course advises us that the sampling
distributions of sample means will be normally distributed, providing
either one of the following two conditions are satisfied :

1. The population is known to be normally distributed,

2. The sample size is greater than 30.

Let us assume that the population is normally distributed. Hence, we

can use the standardised Z distribution tables :

X (B
z'
:2

where X is the mean value of the sample

B is the mean value of the population, and

:2 is the variance of the population.

University of
216 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

The above formula can be re-written as follows :

2 2
z: = X 1 B and B = X 1 z: or B = X 1 z(:/ n)

(where n is the number in the sample).

The 1 sign tells us that we need to add (subtract) to find clu (cll)

Hence, the confidence interval for B = X 1 z(:/ n). Let us assume the
following information:

z (at 95%) = 1.96 (since 95% indicates two halves of 47.5%, leaving 2.5%
in each tail of the distribution. 2.5% can be interpreted as 0.025 and 0.025
in the z tables = 1.96)

n = 150; X = 98 units; : = 45;

then the 95% confidence interval for the population mean, will occur
within :

; 45 > ; 45 > ' 98 1 1. 96 3.674 ' 98 1 7. 2015

98 1 1. 96= @ ' 98 1 1. 96= @ . /
< 150 ? < 12. 247 ?

==> the confidence interval for B = 90.7985 to 105.2015 or 90.7985 " B "
105.2015

Two inferences are available to the manufacturer :

1. S/he can be 95% certain that the true population mean will exist
within the confidence interval of 90.7985 " B " 105.2015, or

2. 95% of all similar formed confidence intervals, will include the

true value for the population mean (B).

Suppose we need a greater certainty (e.g. 99%), then we apply the same
formula as above, except that instead of the 1.96 obtained from the z
tables, we need to identify a different value.

z (at 99%) = 2.575 (since 99% indicates two halves of 49.5%, leaving 0.5%
in each tail of the distribution. 0.5% can be interpreted as 0.005 and 0.005
in the z tables E 2.575).

Inputting the new value for z into the above formula:

; 45 > ; 45 > ' 98 1 2. 575 3.674 ' 98 1 9. 4612

98 1 2. 575= @ ' 98 1 2. 575= @ . /
< 150 ? < 12. 247 ?

University of
Sunderland 217
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

==> the confidence interval for B = 88.5388 to 107.4612 or 88.5388 " B "
107.4612

Notice how the confidence interval has increased in value as the level of
confidence (z value) increases in value. The confidence interval will
decrease in value as the level of confidence decreases in value. This is
shown in the figure below.

Remember, a confidence interval of any value other than 100%,

indicates an error can exist. For example, if our confidence interval is
95%, then there is a 5% possibility that the interval does not contain B.

2. An Unknown Population Variance (:2) (or standard deviation (:)) :

In the event of the population variance being unknown, we can solve

our confidence intervals by using the sample variance (s2) (or sample
standard deviation (s)).

So, the confidence interval for B with :2 unknown is : B = X 1 z(s/ n).

Note, this method is valid, only if (n) is large or the population from
which the sample is taken is known to follow a normal distribution.
Either of these conditions will ensure normality in the sample
distribution.

Calculation of the confidence interval is similar to the example in the

above section, except that we substitute the sample standard deviation
(s) instead of the population standard deviation (:).

3. Sampling Populations that do not follow a Normal Distribution:

Section 1 and 2 above, assumed the populations were normally

distributed. This section relaxes this assumption.

Provided n > 30, then the central limit theorem will ensure normality in
the sampling distribution, and this implies that it will make no
difference whether or not the population follows a normal distribution.

University of
218 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

Solution of confidence intervals therefore, is identical to that described

above.

Confidence Interval for the population mean (small

samples)
This section deals with the situation when n " 30. When n " 30, we are
unable to use the z distribution. Instead, we use the Students ‘t’
distribution. To use the ‘t’ distribution, three conditions must be met :

1. The population is still assumed to follow a normal distribution,

2. n " 30, (i.e. the number of observations in the sample is less than
or equal to 30),
3. The population standard deviation (:) is unknown.

We calculate the ‘t’ statistic, similar to the ‘z’ statistic i.e. :

X (B
t'
s2

where X is the mean value of the sample

B is the mean value of the population, and

s2 is the variance of the sample;

n is the number in the sample.

The above formula can be re-written as follows :

2 2
ts = X 1 B and B = X 1 ts or B = X 1 t(s/ n)

The 1 sign tells us that we need to add (subtract) to find clu (cll).
Analysis is similar to that for the ‘z’ distribution, but we need to
consider the degrees of freedom (dof) associated with the sample size.
For example, suppose n = 25, then n-1 degrees of freedom = 24. If we
use the ‘t’ tables, then for a 95% confidence limit with 24 dof, we can
determine that our estimate (B) will lie between X 1 2.064 standard errors
(:).

i.e. X - 2.064: " B " X + 2.064:

Hence, 95% of repeated sampling, will supply sample means within an

interval of X - 2.064: which will contain B. Later on in these notes, we
will investigate the determinance of 1 2.064:.

University of
Sunderland 219
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

Similarly, for a 99% confidence limit (with 24 dof) :

X - 2.797: " B " X + 2.797:

Hence, the confidence interval for B = X 1 t(s/ n). Let us assume the
following information:

n = 15; X = 98 units; s = 45;

then the 95% confidence interval for the population mean, will occur
within :

t (at 95%, n-1 = 14) = 2.145 (in the ‘t’ tables, look under the 2) =
0.05 column for n = 14)
98 1 2.145(45/ 15) = 98 1 2.145(45/3.873) = 98 1 2.145(11.619) =
98 1 24.923
==> the confidence interval for B = 73.077 to 122.923 or 73.077 "
B " 122.923

Inferences about the population are the same as previous analysis.

Selecting the Proper Test Statistic:

Generally, we can use an algorithm as shown in the figure below.

University of
220 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

Estimating the Difference between Two Normal

Population Means
Suppose we want to compare the properties of two separate
populations. We can use either of the following two sampling methods:

1. Dependent Sampling (otherwise known as paired sampling).

2. Independent Sampling.

Dependent Sampling:

Both sets of observations are identical (or as similar as possible) but,

differ in one respect only. For example, consider a group of 100 hospital
patients, with similar illnesses. Let us assume that these 100 patients
are of a similar age, weight, sex, blood group etc. etc. We then split
these 100 patients into two groups and administer drug A, to the first
group and drug B to the second group. It is the two different drugs that
satisfy the above criteria.

Hence, we are able to compare the diagnosis between the two different
groups and decide if drug A is better (or worse) than drug B.

To assist with our understanding of this subject, consider the situation

of 10 students and their scores in the same general knowledge test,
before and after instruction :

Student Score before Score after di di2

Instruction Instruction
1 39 65 -26 676
2 88 92 -4 16
3 64 89 -25 625
4 37 27 10 100
5 78 87 -9 81
6 60 49 11 121
7 77 73 4 16
8 41 25 16 256
9 51 65 -14 196
10 50 58 -8 64
585 630 -45 2151

We need the mean and the variance of the differences. The mean is
simply (-45/10) = -4.5 i.e. di = -4.5. The variance is given by the following
formula :

0d ( n.d i / 2151 ( 10. (4. 5 /

2 2 2
2 i 1948. 5
sd ' ' ' ' 216. 5
n (1 9 9

University of
Sunderland 221
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

The standard deviation of the sample (s) is found by taking the square
root of 216.5 (= 14.714). Since n < 30 and the population standard
deviation is (:) unknown, then we make use of the ‘t’ distribution. (We
also assume that the d-values follow a normal distribution !). Hence, for
a 95% confidence level, with n-1 dof (= 9 dof) then the confidence
interval for the population mean mark (B) will exist between:

d 1 t(s/ n)

and substituting values :

Bd = -4.5 1 2.262(14.714/ 10) = -4.5 1 10.525

implying -15.025 " Bd " 6.025

We interpret this confidence interval by examining the means of the test

scores. The before (after) test score mean = 58.5 (63). Hence, since the
before test score mean is lower than the after test score mean, then the
confidence interval tells us that we can be 95% confident that the mean
of the after test scores will exist within the range of:

1. 6.025 marks above the before test mean and

2. 15.025 marks below the before test mean.

Or, in other words, we do not have enough evidence to suggest that

instruction will increase the mean test score, since the interval contains
zero.

Independent Sampling:

Occurs when two (or more) independent samples are selected from
different populations. Independent samples do not need to be of the
same size.

1. Known Variances :

Insurance companies have been known to charge a higher premium to

male motor policy holders than that of female motor policy holders.
This philosophy is based upon the scenario that male motorists in
general, are a greater risk than females. Risk levels are identified for
each sex and the point difference between the mean risk levels, can
serve as an estimate for the population means of both sexes, i.e. let x1
and x2 exist as an estimate for B1 and B2

Repeated sampling to determine point differences between the mean

risk levels, will more than likely produce slightly different values, since
no two samples will be exactly identical. This error is known as the
standard error of the sampling distribution of the differences between
two sample means (SESD). This error is important.

University of
222 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

If the population variances (:2) are known, the SESD can be calculated
via :

2 2
:1 :
: x1 ( x2 ' - 2
n1 n2

where : :12 and :22 are the variances of the two populations,
n1 and n2 are the two sample sizes (which do not need
to be equal)

Hence, the confidence interval for the difference between the

population means (B1 and B2) is given by :

B1 - B2 = (X1 - X2) 1 Z:x1-x2

Let us look at an example.

The North/South divide - does it exist with mortgages ?

A survey of 100 financial institutions in the north and 123 financial

institutions in the south, produced the following results :
2 2
X1 = 11.75%; X2 = 10.53%; :1 = 2.5%; :2 = 3.0%

You have the chance to live in either the north or the south. Which part
of the country has the higher mortgage levels? Construct a 95%
confidence interval.

0.5 - 0.475 = 0.025 and 0.025 in the Z tables gives 1.96. Hence, the
confidence interval for the difference between the population means (B1
and B2) is given by :

B1 - B2 = (X1 - X2) 1 Z:x1-x2

2. 5 3.0
' .11.75 ( 10. 53 / 1 1. 96 -
100 123

= 1.22 1 1.96(0.222) = 0.7849 " B1 - B2 " 1.656

University of
Sunderland 223
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

The interpretation is that you can be 95% confident that the mortgage
interest rates in the south, exceed that of the north by an amount
between 0.7849% and 1.656%. Hence, you should live in the north!

2. Unknown Variances (but assumed to be equal):

The assumption of unknown variances but equal, can best be explained

by a simple example. Suppose a canned soft drink manufacturer has a
production line, where cans are filled to a pre-determined level. It is
assumed that the cut off point will take on a certain value in fluid
ounces. A random sample of n observations will produce a sample
mean and a variance. Repeated random sampling should produce a
similar mean and a variance. Suppose now, that the cut off point is
adjusted. Random sampling will produce a different sample mean, but
it is expected (or assumed), that the variance will remain unaltered.
Hence, the same population, with different means, but similar (equal)
variances.

If population variances are unknown, we substitute the sample

variances. We must remember to consider the problems associated with
small samples, especially when n " 30. If n " 30, we use the ‘t’ statistic
instead of the ‘z’ statistic, and we must also assume that the populations
follow a normal distribution.

We use the following formula to calculate the pooled average of the

sample variances :

2
sp '
s1
2
. n1 ( 1 / - s 2 2 . n 2 ( 1 /
n1 - n 2 ( 2

where : sp2 is the pooled average of the sample variances,

s12 is the sample variance of sample No 1
s22 is the sample variance of sample No 2
n1 is the number of observations in sample No 1
and n2 is the number of observations in sample No 2

The confidence interval given for the two sample means is therefore
given by :

. /
B 1 ( B 2 ' X1 ( X 2 1 . t / s p . / 1
-
1
n1 n 2

3. Independent Samples (:12 D :22):

In some instances, the population variances are unequal, or there is no

reason to assume an equality. For example, suppose a firm sold two
similar types of shoes. Type 1 shoes were found to last slightly longer
than type 2 shoes, but there is no evidence to suggest that the variance of

University of
224 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

the wear on type 1 shoes, is equal to the variance of the wear on type 2
shoes, i.e. type 2 shoe wearers might be heavy on their feet.

Unfortunately, no-one has developed an accurate distribution which

would describe this sampling process. We have to use an
approximation, with the dof slightly altered.
2
; s1 2 s 2 2 >
= - @
=n n @
dof ' < 1 2 ?
2 2
; s1 2 > ;s 2 >
= @ = 2 @
=n @ = @
< 1 ? - < n2 ?
. n1 ( 1 / . n 2 ( 1 /
(Definitions are similar to those in the previous section). Since the dof
has been slightly altered, then we symbolise the ‘t’ statistic with the tilde
(~) :

2 2

.
Confidence interval for B 1 ( B 2 ' X1 ( X 2 1 t ~/ s1 s
- 2
n1 n 2

Chi Squared Test

The chi square test is used to test a distribution observed in the field
against another distribution determined by a null hypothesis.

Being a statistical test, chi square can be expressed as a formula. When

written in mathematical notation the formula looks like this :

. O ( E/
2

A2 '0
E

When using the chi square test, the researcher needs a clear idea of what
is being investigated. It is customary to define the object of the research
by writing a hypothesis. Chi square is then used to either prove or
disprove the hypothesis.

The chi-square test may be used both as a test of goodness of fit

(comparing frequencies of one attribute variable to theoretical
expectations) and as a test of independence (comparing frequencies of
one attribute variable for different values of a second attribute variable).
The underlying arithmetic of the test is the same; the only difference is
the way the expected values are calculated. However, goodness-of-fit
tests and tests of independence are used for quite different experimental
designs and test different null hypotheses, so we will consider the
chi-square test of goodness-of-fit and the chi-square test of
independence to be two distinct statistical tests.

University of
Sunderland 225
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

When to use it

The chi-squared test of independence is used when you have two

attribute variables, each with two or more possible values. A data set
like this is often called an “R x C table,” where R is the number of rows
and C is the number of columns. For example, if you surveyed the
frequencies of three flower phenotypes (red, pink, white) in four
geographic locations, you would have a 3 X 4 table. You could also
consider it a 4 X 3 table; it doesn’t matter which variable is the columns
and which is the rows.

It is also possible to do a chi-squared test of independence with more

than two attribute variables, but that experimental design doesn’t occur
very often and is rather complicated to analyse and interpret, so we
won’t cover it.

Null hypothesis

The null hypothesis is that the relative proportions of one variable are
independent of the second variable; in other words, the proportions at
one variable are the same for different values of the second variable. In
the flower example, the null hypothesis is that the proportions of red,
pink and white flowers are the same at the four geographic locations.

Ho: the variables are independent of each other, H1: Ho is untrue;

For some experiments, you can express the null hypothesis in two
different ways, and either would make sense. For example, when an
individual clasps their hands, there is one comfortable position; either
the right thumb is on top, or the left thumb is on top. Downey (1926)
collected data on the frequency of right-thumb vs. left-thumb clasping
in right-handed and left-handed individuals. You could say that the
null hypothesis is that the proportion of right-thumb-clasping is the
same for right-handed and left-handed individuals, or you could say
that the proportion of right-handedness is the same for
right-thumb-clasping and left-thumb-clasping individuals.

How the test works

For the test of independence, only the observed frequencies are used to
calculate the expected. For the hand-clasping example, Downey (1926)
found 190 right-thumb and 149 left-thumb-claspers among
right-handed women, and 42 right-thumb and 49 left-thumb-claspers
among left-handed women. To calculate the estimated frequency of
right-thumb-claspers among right-handed women, you would first
calculate the overall proportion of right-thumb-claspers:
(190+42)/(190+42+149+49) = 0.5395. Then you would multiply this
overall proportion times the total number of right-handed women,
0.5395*(190+149)=182.9. This is the expected number of right-handed
right-thumb-claspers under the null hypothesis; the observed number
is 190. Similar calculations would be done for each of the cells in this 2x2
table of numbers.

University of
226 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

The degrees of freedom in a test of independence are equal to the

number of rows - 1 times the number of columns -1. Thus for a 2x2 table,
there are (2-1)*(2-1)=1 degree of freedom; for a 4x3 table, there are
(4-1)*(3-1)=6 degrees of freedom.

Example

Suppose that the ratio of male to female students in the Science Faculty
is exactly 1:1, but in the Pharmacology Honours class over the past ten
years there have been 80 females and 40 males. Is this a significant
departure from expectation? We proceed as follows:

Set out a table as shown below, with the “observed” numbers and the
“expected” numbers (i.e. our null hypothesis).

Then subtract each “expected” value from the corresponding

“observed” value (O-E)

Square the “O-E” values, and divide each by the relevant “expected”
value to give (O-E)2/E

Add all the (O-E)2/E values and call the total “X2"

Notes:
*1
This total must always be the same as the observed total
*2
This total must always be zero
*3
The null hypothesis was obvious here: we are told that there are equal
numbers of males and females in the Science Faculty, so we might
expect that there will be equal numbers of males and females in
Pharmacology. So we divide our total number of Pharmacology
students (120) in a 1:1 ratio to get our ‘expected’ values.

Now we must compare our X2 value with a A 2 (chi squared) value in a

table of A 2 with n-1 degrees of freedom (where n is the number of
categories, i.e. 2 in our case - males and females). We have only one
degree of freedom (n-1). From the A 2 table, we find a “critical value” of
3.84 for p = 0.05 with one degree of freedom.

University of
Sunderland 227
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

If our calculated value of X2 exceeds the critical value of A 2 then we have

a significant difference from the expectation. In fact, our calculated X2
(13.34) exceeds even the tabulated A 2 value (10.83) for p = 0.001. This
shows an extreme departure from expectation. It is still possible that we
could have got this result by chance - a probability of less than 1 in 1000.
But we could be 99.9% confident that some factor leads to a “bias”
towards females entering Pharmacology Honours. (Of course, the data
don’t tell us why this is so - it could be self-selection or any other
reason.)

Now repeat this analysis, but knowing that 33.5% of all students in the
Science Faculty are males

Note *1: We know that the expected total must be 120 (the same as the
observed total), so we can calculate the expected numbers as 66.5% and
33.5% of this total.

Note *2: This total must always be zero.

Note *3: Although the observed values must be whole numbers, the
expected values can be (and often need to be) decimals. For ‘males’,
33.5% of 120 students is 40.2 males.

Now, from a table of A 2 we see that our data do not depart from
expectation (the null hypothesis). They agree remarkably well with it
and might lead us to suspect that there was some design behind this! In
most cases, though, we might get intermediate X2 values, which neither
agree strongly nor disagree with expectation. Then we conclude that
there is no reason to reject the null hypothesis.

REVIEW ACTIVITY
Two firms are about to horizontally merge and their respective union officials
are attempting to identify whether or not a significant difference exists
between the mean wage levels. As an accountant/economist, you are called in
to supply a solution. You are supplied with the following information:

University of
228 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

Firm No No of Sample Mean Sample Variance

observations Value

1 n1 = 28 X1 = £19.00/hour s12 = 101

2 n2 = 21 X2 = £16.98/hour s22 = 79

Calculate a 95% confidence interval for the difference between the mean levels
and interpret the result.

Hint, use the following formulae:

s1 .n1 (1/ - s 2 .n 2 (1/

2 2
2
sp '
n1 - n 2 ( 2

where : sp2 is the pooled average of the sample variances,

s12 (s22) is the sample variance of sample No 1 (No 2)

n1 (n2) is the number of observations in sample No 1 (No 2)

The confidence interval given for the two sample means is therefore given
by :

. /
B1 ( B 2 ' X1 ( X 2 1 . t/. s p /
1 1
-
n1 n 2

REVIEW ACTIVITY FEEDBACK

Firm No No of Sample Mean Sample Variance
observations Value

1 n1 = 28 X1 = £19.00/hour s12 = 101

2 n2 = 21 X2 = £16.98/hour s22 = 79

Substituting the above in to the formula for the pooled sample variance :

s1 .n1 (1/ - s 2 .n 2 (1/

2 2
2
sp '
n1 - n 2 ( 2

2 101.27/ - 79.20/ 2727 -1580 4307

sp ' ' ' ' 91.638 and s p ' 9.573
28 - 21 ( 2 47 47

University of
Sunderland 229
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

The confidence interval given for the two sample means is therefore given by :

. /
B1 ( B 2 ' X1 ( X 2 1 . t/. s p /
1 1
-
n1 n 2

A 95% confidence interval implies ) = 0.05 ()/2 = 0.025), with dof = 28 + 21 -

2 (i.e. 47), giving t)/2,47 E 2.000 to 2.021. (Use 2.000 since it is the lower value).

; 1 > ;1>
Hence, C.I. for B1 - B2 = (19 - 16.98) 1 (2.000)(9.573)( = @ - = @ ),
< 28 ? < 21?

giving: (2.02) 1 (19.146)( .0.08333/),

giving: (2.02) 1 (19.146)(0.2887) or (2.02) 1 (5.527)

and the confidence interval for B1 - B2 is -3.507 to 7.547 or -3.507 " B1 - B2 "
7.547

The interpretation is that we can be 95% confident that the firm with higher
hourly wage rate, lies on average, between -£3.51 to £7.55 of the other firm’s
average hourly wage.

Or in other words, we do not have enough evidence to suggest that there is

any difference between the hourly wage rates, since the interval contains zero.

REVIEW ACTIVITY
Consider the general knowledge test given in the notes. i.e. students were
given the same test twice before and after instruction. Let us modify the
scenario and assume now that the statistics are as follows :

Number in Group Sample Mean Sample Variance

18 X1 = 83% s12 = 109

23 X2 = 87.7% s22 = 128.7

Calculate a 95% confidence interval for the difference in the mean scores of all
the students and interpret the result.

University of
230 Sunderland
Business Modelling for Decision Making Unit 7 – Sampling Theory and Statistical Inference

Hint, use the following formulae :

2
; s1 2 s 2 2 >
= - @
= n1 n 2 @
dof ' < ?
2 2
; s1 2 > ; 2>
= @ = s2 @
= n1 @ = @
< ? - < n2 ?
.n1 (1/ .n 2 (1/
(Definitions are similar to those in the previous section). Remember, the dof
has been slightly altered, so we need to symbolise the ‘t’ statistic with the tilde
(~):

2 2

.
Confidence interval for B1 ( B 2 ' X1 ( X 2 1 t ~ / s1 s
- 2
n1 n 2

REVIEW ACTIVITY FEEDBACK

Number in Group Sample Mean Sample Variance

18 X1 = 83% s12 = 109

23 X2 = 87.7% s22 = 128.7

Substituting the above in to the formula for the degrees of freedom (dof):
2
; s1 2 s 2 2 >
= - @
= n1 n 2 @
dof ' < ?
2 2
; s1 2 > ; 2>
= @ = s2 @
= n1 @ = @
< ? - < n2 ?
.n1 (1/ .n 2 (1/
; ; 109 > ; 128.7 > >
== = @ -= @ @@
dof ' < < 18 ? < 23 ? ? '
6.056 - 5.596 11.651
' ' 3254
.
2; ; ; 109 > 2 > ; ; 128.7 > 2 > > 5 2157
. -1423
. 358
.
4= = = @ @ == @ @ @7
4 <= = 18 @
? - < = 23 ? @ @7
4 = = @ = @ @7
= 17 22
4 = @ = @ @7
43=< =< @ =
? <
@ @7
? ?6

University of
Sunderland 231
Unit 7 – Sampling Theory and Statistical Inference Business Modelling for Decision Making

We round down the dof to the nearest integer, i.e. 3.254 rounded = 3 = dof.
Substituting for dof = 3 into the confidence interval formula ..

2 2

. /
Confidence interval for B1 ( B 2 ' X1 ( X 2 1 t ~
s1 s
- 2
n1 n 2

A 95% confidence interval implies ) = 0.05 ()/2 = 0.025), with dof = 3,

giving t)/2,3 =3.182

; 109 > ; 128.7 >

Hence, C.I. for B1 - B2 = (83 - 87.7) 1 (3.182)( = @ -= @ ),
< 18 ? < 23 ?

giving: -4.7 1 (3.182)( 6.0556 - 5.596 )

= -4.7 1 (3.182)( 11651

. )

= -4.7 1 (3.182)(3.413)

and -4.7 1 10.861.

The confidence interval for B1 - B2 is -15.56 to 6.16 or -15.56 " B1 - B2 "

6.16.

The interpretation is that since the confidence interval contains zero, we

cannot be certain that there is any difference in the mean scores of all the
students.

Summary
This unit has provided an introduction to the principles of hypothesis
testing. After an introduction to the concept, we have studied one tailed
tests (both left and right hand). Then we looked at two tailed tests.

Confidence testing has been described, including the usefulness of

confidence intervals. To conclude the unit, we introduced the Chi
Square test for dependence.

University of
232 Sunderland
Unit 8

Statistical Process Control and

Financial Mathematics

Introduction
In this unit, we shall look at two important areas of statistical usage.
First of all we will consider the use of control charts for both variable
and attributes within an organisational setting. We will include a
discussion of Deming’s 14 points for business process quality
improvement, and also look at the concept of process capability.

In the second half of the unit the value of modelling financial

mathematics is explained through the use of Microsoft Excel. This
includes a study of its use in basic indicators of statistical spread. The
concept of interest, both simple and compound, is then used to explore
compound growth and real value of money. These discussions allow us
to gain an understanding of annuities, sinking funds, discounting
techniques and debt repayment calculations.

Deming’s 14 Points
Dr. W. Edwards Deming is known as the father of the Japanese post-war
industrial revival and was regarded by many as the leading quality
guru in the United States. He died in 1993.

Trained as a statistician, his expertise was used during World War II to

assist the United States in its effort to improve the quality of war
materials

Deming’s business philosophy is summarized in his famous “14

Points,” listed below. These points have inspired significant changes
among a number of leading US companies striving to compete in the
world’s increasingly competitive environment.

But the 14 Points pose a challenge for many firms to figure out how to
apply them in a meaningful way that will result in continual
improvement. Leadership Institute has developed powerful processes
for coaching executive teams, and eventually their entire organizations,
to begin accomplishing what Deming referred to as “the
transformation.”

University of
Sunderland 233
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

The 14 points are a basis for transformation of industry. Adoption and

action on the 14 points are a signal that management intend to stay in
business and aim to protect investors and jobs. Such a system formed
the basis for lessons for top management in Japan in 1950 and in
subsequent years.

The 14 points apply anywhere, to small organisations as well as to large

ones, to the service industry as well as to manufacturing. They apply to
a division within a company.

1. Create constancy of purpose for improvement and service.

Management must change from a preoccupation with the short
run to building for the long run. This requires dedication to
innovation in all areas, to best meet the needs of customers.
2. Adopt the new philosophy. Shoddy materials, poor
workmanship, defective products, and lax service become
unacceptable.
3. Cease dependence on mass inspection. Inspection is equivalent
to planning for defects; it comes too late and is ineffective and
costly. Instead, processes must be improved.
4. End the practice of awarding business on price tag alone. Price
has no meaning without a measure of the quality being
purchased. Therefore, the job of purchasing will change only
after management establishes new guidelines. Companies must
develop long-term relationships and work with fewer suppliers.
Purchasing must be given statistical tools to judge the quality of
vendors and purchased parts. Both purchasing and vendors must
understand specifications; but they must also know how the
material is to be used in production and by the final customer.
5. Constantly and forever improve the system of production and
service. Waste must be reduced and quality improved in every
activity; procurement, transportation, engineering, methods,
maintenance, sales, distribution, accounting, payroll, customer
service, and manufacturing. Improvement, however, does not
come from studying the defects produced by a process that is in
control, but from studying the process itself. Most of the
responsibility for process improvement rests with management.
6. Institute modern methods of training on the job. Training must
be restructured and centred on clearly defined concepts of
acceptable work. Statistical methods must be used for deciding
when training has been completed successfully.
7. Institute modern methods of supervising. Foremen must be
empowered to inform upper management about conditions that
need correction; once informed, management must take action.
Barriers that prevent hourly workers from doing their jobs with
pride must be removed.

University of
234 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

8. Drive out fear. Because of the tremendous economic losses

caused by fear on the job, people must not be afraid to ask
questions, to report problems or to express ideas.
9. Break down barriers between departments. Members of the
research, design, procurement, sales and receiving departments
must learn about problems with raw materials and specification
in production and assembly. Each discipline must stop
optimising its own work, and work together as a team for the
company as a whole. Multi-disciplinary quality control circles
can help improve design, service, quality and costs.
10. Eliminate numerical goals for the work force. Targets, slogans,
pictures, and posters urging people to increase productivity must
be eliminated. Most of the necessary changes are out of workers’
control so such exhortations merely cause resentment. Although
workers should not be given numerical goals, the company itself
must have a goal: never ending improvement.
11. Eliminate work standards and numerical quotas. Quotas focus
on quantity, not quality. Therefore, work standards practically
guarantee poor quality and high costs. Work standards that state
percentage defective or scrap goals normally reach those targets,
but never exceed them. Piecework is even worse, for it pays
people for building defective units. But if someone’s pay is
docked for defective units, that is unfair, for the worker did not
create the defects.
12. Remove barriers that hinder the hourly workers. Any barrier
that hinders pride in work must be removed, including not
knowing what good work is, supervisors motivated by quotas,
off-gauge parts and material, and no response to reports of
out-of-control machines.
13. Institute a vigorous programme of education and training.
Because quality and productivity improvements change the
number of people needed in some areas and the jobs required,
people must be continually trained and retrained. All training
must include basic statistical techniques.
14. Create a structure in top management that will push every day
on the above 13 points.

ACTIVITY
Visit the website of the Deming institute to find out more.

http://www.deming.org/theman/teachings02.html

University of
Sunderland 235
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

ACTIVITY
Visit a further website to see some more links to thought-provoking
information about each of Deming’s 14 points, including Pat Oliphant’s
amusing cartoon interpretation of the 14 points:

http://curiouscat.com/management/demings14pts.cfm

Control Charts
Control charts are a method of Statistical Process Control, SPC. (Control
system for production processes). They enable the control of
distribution of variation rather than attempting to control each
individual variation. Upper and lower control and tolerance limits are
calculated for a process and sampled measures are regularly plotted
about a central line between the two sets of limits. The plotted line
corresponds to the stability/trend of the process. Action can be taken
based on trend rather than on individual variation. This prevents
over-correction/compensation for random variation, which would lead
to many rejects.

Different types of control charts can be used, depending upon the type
of data. The two broadest groupings are for variable data and attribute
data.

! Variable data are measured on a continuous scale. For

example: time, weight, distance or temperature can be
measured in fractions or decimals. The possibility of
measuring to greater precision defines variable data.

! Attribute data are counted and cannot have fractions or

decimals. Attribute data arise when you are determining
only the presence or absence of something: success or
failure, accept or reject, correct or not correct. For
example, a report can have four errors or five errors, but
it cannot have four and a half errors.

Variables charts include

! x and R chart (also called averages and range chart).

! Chart of individuals (also called X chart, X-R chart,
IX-MR chart, Xm R chart, moving range chart).

! Moving average–moving range chart (also called

MA–MR chart).

University of
236 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

! Target charts (also called difference charts, deviation

charts and nominal charts).

! CUSUM (also called cumulative sum chart.

! EWMA (also called exponentially weighted moving
average chart).

! Multivariate chart (also called Hotelling T2).

Attributes charts include

! p chart (also called proportion chart).

! np chart.
! c chart (also called count chart).
! u chart.

Charts for either kind of data

! Short run charts (also called stabilized charts or Z charts).

! Group charts (also called multiple characteristic charts).

The control chart is a graph used to study how a process changes over
time. Data are plotted in time order. A control chart always has a central
line for the average, an upper line for the upper control limit and a lower
line for the lower control limit. These lines are determined from
historical data. By comparing current data to these lines, you can draw
conclusions about whether the process variation is consistent (in
control) or is unpredictable (out of control, affected by special causes of
variation).

Control charts for variable data are used in pairs. The top chart monitors
the average, or the centering of the distribution of data from the process.
The bottom chart monitors the range, or the width of the distribution. If
your data were shots in target practice, the average is where the shots
are clustering, and the range is how tightly they are clustered. Control
charts for attribute data are used singly.

When to use a control chart:

! When controlling ongoing processes by finding and

correcting problems as they occur.

! When predicting the expected range of outcomes from a

process.

! When determining whether a process is stable (in

statistical control).

University of
Sunderland 237
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

! When analysing patterns of process variation from

special causes (non-routine events) or common causes
(built into the process).

! When determining whether your quality improvement

project should aim to prevent specific problems or to
make fundamental changes to the process.

Basic control chart procedure:

1. Choose the appropriate control chart for your data.

2. Determine the appropriate time period for collecting and plotting
data.
3. Collect data, construct your chart and analyse the data.
4. Look for “out-of-control signals” on the control chart. When one
is identified, mark it on the chart and investigate the cause.
Document how you investigated, what you learned, the cause
and how it was corrected.
Out-of-control signals

! A single point outside the control limits. In Figure 8.1,

point sixteen is above the UCL (upper control limit).

! Two out of three successive points are on the same side of

the centerline and farther than 2 : from it. In Figure 8.1,
point 4 sends that signal.

! Four out of five successive points are on the same side of

the centerline and further than 1 : from it. In Figure 8.1,
point 11 sends that signal.

! A run of eight in a row are on the same side of the

centerline. Or 10 out of 11, 12 out of 14 or 16 out of 20. In
Figure 8.1, point 21 is eighth in a row above the
centerline.

! Obvious consistent or persistent patterns that suggest

something unusual about your data and your process.

University of
238 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

Figure 8.1. Control Chart: Out-of-Control Signals.

5. Continue to plot data as they are generated. As each new data

point is plotted, check for new out-of-control signals.
6. When you start a new control chart, the process may be out of
control. If so, the control limits calculated from the first 20 points
are conditional limits. When you have at least 20 sequential
points from a period when the process is operating in control,
recalculate control limits.

Case Study
‘Managing stock to meet customer needs’ from McDonalds.

Stock control charts

A stock control chart shows the balance of orders for new stocks against
sales. The system is dependent on figures for expected sales. For
example, if sales of burgers are going out of the system, then stocks of
beef patties need to be coming into the system.

Manugistics uses two years’ worth of product mix history to produce

forecasts for each restaurant. This uses time series analysis. The planner
will apply a causal factor (the blue blocks in the example) to the time
series for the start and end date of this promotion. Using complex
calculations, the graph then produces a forecast - seen below.

University of
Sunderland 239
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

Figure 8.2. McDonald’s control chart.

Entering data

Any system is only as good as the data that is provided. Therefore,

McDonald’s restaurant managers need to ensure that the data they
enter into the system is as accurate as possible. For example, each day,
restaurant managers record opening and closing stocks of key food
items. They record all other items weekly. The store computer system
identifies any stock count deviations from the last stock count so
managers can investigate. For example, the manager may have missed
off a box of organic milk whilst counting them earlier on in the shift.

Buffer stock

Restaurants hold a small buffer stock. This is an extra quantity of stock

held to meet unexpected higher demand. It is also the point at which
more goods are ordered – the re-order level.

Ordering

Figure 8.3. Order Proposals.

McDonald’s store managers use a simple web-based communication

tool called ‘WebLog’ to view and amend store Order Proposals.

University of
240 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

Every day WebLog creates a proposed order for the manager to analyse
and amend if necessary. WebLog enables managers and central
planners to see what quantities have been ordered, what the current
stock levels are and exactly how much stock is due to be delivered at a
particular time.

In the past, managers would have had to check their delivery for any
shortages and input every item they had received. The system now
automatically generates a delivery note that gives the exact quantities
and descriptions of the delivery. All managers need to do is simply click
‘confirm’ on WebLog. This saves valuable time and makes the process
more cost-effective.

ACTIVITY
Find out the views of other organisations to the use of control charts.

Process Capability
Process capability compares the output of an in-control process to the
specification limits by using capability indices. The comparison is made
by forming the ratio of the spread between the process specifications
(the specification “width”) to the spread of the process values, as
measured by six process standard deviation units (the process
“width”).

When to use it

! Use it when setting up a process, to ensure it can meet its

specification limits.

! Use it when setting specification limits, to ensure they are

neither too wide nor too narrow.

! Use it when investigating a process that is not meeting its

specification limits.

! Use it only when the process is stable and has a Normal

distribution.

University of
Sunderland 241
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

Figure 8.4. Using Process Capability in problem solving.

How to understand it

The outputs of any process will vary and it is common for specification
limits to be defined such that if the measured output of the process
exceeds the specified limits, the process is deemed to have failed. The
term ‘specification limits’ is most commonly used for the dimensions of
a manufactured item, but can be used in any process. Thus, for example,
the specification limits for the time a telesales operator may take to
answer a customer call may be between zero and five seconds.

The results of most processes will vary around a central value, and the
‘capability’ of the process is defined as the spread of results around this
value, with high capability occurring when process results group
closely around it. Thus a process that can produce parts to within
0.001mm of a target value is more capable than one which can only
produce them to within 0.015mm.

The most common measure of this spread is standard deviation, and

‘Process Capability’ may be defined as the range between three
standard deviations either side of the average.

Figure 8.5. Standard deviation.

University of
242 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

Specifications are often defined separately from the process that is being
measured and without a great deal of consideration of how easily the
process can meet them. This can result in either many failures and
rejects or effectively redundant specifications, as the variation in the
process fits badly or well within the specified limits.

Specification limits and process capability thus need to be considered

together. The limits still cannot be too tight as if calibration is done
under ideal conditions, process distribution may subsequently drift or
spread, for example as a result of wear in a machine tool.

Figure 8.6. Natural distribution fit within defined specification limits.

A common Process Capability measure, Cp (often called a Process

Capability Index), indicates how well the process distribution fits
within its specification limits, and is simply the ratio of the specification
width to the variation width. Thus, in the figure above , processes (a)
and (b) have Cp greater than one, (c) is equal to one and (d) is less than
one.

The problem with Cp is that it does not take account of how well the
process distribution is centered within its limits, which can result in a
process with both a low Cp and many rejects. The solution to this is a
second measure, Cpk, which measures a similar ratio, but considers
only the variation half that is closest to the specification limits, as in the
figure below. Thus Cp and Cpk, taken together, give a measure of both
the potential and centering of the process distribution within the
specification limits.

Process Capability measures are only as good as the data used, and
there is plenty of opportunity for misinterpretation. In particular,
Process Capability measurement is based on three important
assumptions which are thus preconditions for valid calculations:

University of
Sunderland 243
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

1. The process is in a state of statistical control, and there are no

special causes of variation. The implication of this is that before
Cp and Cpk can be measured, special causes must be found and
eliminated. This may be done using the Control Chart over a
period of time long enough to give confidence that this has been
successfully completed.
2. The process distribution is bell-shaped or ‘Normal’, which allows
the width of the distribution to be calculated as six times the
standard deviation. In practice, there are many situations where
the distribution is not normal, and in Process Capability
measurement the Central Limit Theorem does not act to
normalise this, as it does when using a Control Chart.

Figure 8.7. Capability measurement.

3. The measured data is representative of the process. This means

that it should be a randomly selected and large sample, taken
over a long period. Samples taken over a short period can suffer
from a limited range of changes in either external seasonal effects
or internal process variables, such as humidity or tool wear.

When interpreting values of Cpk, there are three significant regions

which may be considered, and a general rule is given in the table below .
The value of 3 as a ‘total confidence’ limit may be lowered if
measurements are taken as the average of sample batches. This
commonly happens when Cpk is measured using the same data that is
used to plot the Control Chart (e.g. the confidence limit reduces to 2 for
the common sample size of 4).

In the broader sense, studying Process Capability is more than just

measuring Cp and Cpk; it involves understanding the statistical
performance and operational working of the process. Most importantly,

University of
244 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

it means understanding what causes variation within the process,

under what conditions, and how these variables interact. The purpose
of doing this is to enable confident process improvement that steadily
reduces variation.

Figure 8.8. Cpk values and capability.

Example

A company producing kitchen worktops, specified the length of one

range at 120 cm ± 0.25 cm. As a normal part of production monitoring,
Control Charts were kept for such specified measures and all output
was inspected against specification limits. A request from the sales team
for tighter limits prompted the question of whether the limits could be
reduced to ± 0.1cm, as this could result in significant orders from a new
customer who preferred not to shave worktops to fit.

The production supervisor used the Control Chart data to draw a

Histogram and check for Normal distribution, then calculate Cpk,
across samples from a week’s work, as in the figure below . This showed
that the process was currently so capable that inspection could be
dropped and the new limits still met. This turned out to be as a result of
old specification limits coupled with several recent process
improvements.

As a result, the new orders were achieved and savings were also made
on inspection costs.

University of
Sunderland 245
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

Figure 8.9. Process Capability example.

Financial Maths
(Note: The following unit assumes you have access to Excel functions
PMT, CUMIPMT and CUMPRINC)

Mean, Median, Mode and Variance

The basic built-in functions of Excel include AVERAGE, MEAN,
MODE, COUNT, MAX and MIN. We will use the familiar example of a
class’s grades to illustrate the use of some of the more basic Excel
functions, like AVERAGE( ), MODE( ) AND MAX( ). Assume a class’s
grade distribution is as follows: 3, 0, 4, 4, 4, 2, 4, 1, 4, 0, 3, 3, 1, 1, 3. These
grades are based on a four-point scale with 4=A and 0=F and are entered
into an Excel worksheet shown below. Using the AVERAGE( )
function, we find the class’s average (or arithmetic mean) grade is a
disappointing 2.47, or a mid-C. The syntax for this common function is
=AVERAGE(number1, number2, ...) and is displayed in the screen shot
below. For more information using the AVERAGE function, see the
online help available within Excel.

University of
246 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

However we don’t get a clear picture of the class’s performance by

simply looking at its average. We can further analyse the data using the
MEDIAN( ) function. The median gives the middle number in a set of
numbers and its syntax is =MEDIAN(number1, number2,...). We see
from the screen shot below that the median grade is 3.0, meaning that
half of the grades are higher than 3.0, and half are lower. Therefore,
despite the low class average, more students scored 3’s and 4’s than 2’s,
1’s and 0’s.

Additionally, we can also analyse the grade distribution by using the

MODE( ) function. The mode gives the most frequently occurring value
(or values) of a set of numbers and its syntax is =MODE(number1,
number2,...). From the screen shot, we see that the mode grade is 4,
meaning that a score of 4 was the most common grade. Again, the
instructor of the class can take heart that, despite the low class average,
more students made A’s than any other grade. Be careful with the use
of the ‘MODE’ function in Excel as Excel tends to find one mode only.

Without going into too much detail, we can also use some of Excel’s
built-in functions to determine the number of grades entered, and the
maximum and minimum grades of the distribution. The syntax for
these functions are shown below in the bulleted list and also in the
screen shot.

! The COUNT( ) function gives the number of cells that

contain numbers. Its syntax is =COUNT(value1, value2,
...).

! The MAX( ) function returns the largest value in a set of

numbers. Its syntax is =MAX(number1, number2, ...).

! The MIN( ) function returns the smallest value in a set of

numbers. Its syntax is =MIN(number1, number2, ...).

University of
Sunderland 247
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

Figure 8.10. Basic Excel functions.

Financial functions and formulae

Excel contains many financial functions and this page includes
examples of PMT, PPMT, FV and IPMT. These calculate compound
interest, loan repayments and outstanding principal amounts. Many
other functions are only available if you load special add-ins after
installing the Analysis Toolpak. Also included on this page are some
formulae with which values can be calculated without using built in
functions.

The following example shows the results of various financial

calculations based upon both Annual and Monthly values. Note that the
cells C4:D8 have all been given names ending with ‘_A’ (annual) or ‘_M’
(monthly). The names have been used in the formulae rather than the
cell references.

University of
248 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

Figure 8.11. Sample financial calculations and explanations.

Simple and Compound Interest

Many people during their lives will decide to save some money for the
future. This will usually be placed in the hands of a bank in an account
that will give interest each year. This is interest will be given to the
owner of the capital some at the end of the year. For example, if
someone invests the sum of £500 in a bank account that yields 4% pa
they will receive the sum of £20 at the end of the year. Using the idea of
simple interest, each year the saver will receive the sum of £20 as long

University of
Sunderland 249
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

as the capital sum invested stays the same and the interest rate offered
by the bank is unchanged.

A formula to calculating the change in value as a result of simple

interest over a period of time would be

Future principal value, P = (number of years x capital x interest rate)

(where the interest rate is written as a decimal, i.e. 3% would be written

as 0.03)

Thus, £700 invested at a rate of 5% for three years would earn

3 x 700 x 5/100 = £105

(This shows the total interest across three years.)

However, in reality very few savers would be interested in such an

account for long-term investment. They are more likely to want the
interest to be added to the capital at the end of each year and so allow
the invested sum to grow. A bank would, therefore, collect the interest
accrued at the end of the first year and then add that to the original
capital. At the end of year 2, interest will then be paid on the combined
sum of the capital invested and the interest earned during the previous
year. In this way the amount of money earned in the form of interest will
increase each year. This is known as compound interest.

If our saver wishes to invest the sum of £500 at 4%, they will still receive
£20 at the end of the first year. However, this will be added to the £500 at
the start of year 2. Therefore, at the end of year 2 the capital sum will be
£520. Interest for that year will now rise to £20.80. The process is then
repeated for each year that the investment remains in the bank. The
balances at the end of each year will be as follows:

End of year Interest paid Balance

£ £

1 20.00 520.00

2 20.80 540.80

3 21.63 562.43

4 22.50 584.93

5 23.40 608.33

etc.

University of
250 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

ACTIVITY
Use the simple interest calculator at

http://www.maths.surrey.ac.uk/explore/matthewspages/simpleinterest.html

Insert a range of values to gain an understanding of the concept.

The mathematics now becomes a little more complicated, and the

formula for compound interest becomes:

Principal = Pn = Po( 1 + i )n

where: Po = the value at the beginning

Pn = the value after n years

i = the rate of interest, expressed as a decimal;

i.e. 6% would be 0.06.

n = the number of years.

ACTIVITY
I am going to invest the sum of £375 in a bank account offering me 7.5% simple
interest. How much interest will I have earned at the end of:

1. Year 1

2. Year 5

3. Year 12?

University of
Sunderland 251
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

ACTIVITY FEEDBACK
Your answers should be:

1 x 375 x 0.075 = £ 28.13

5 x 375 x 0.075 = £140.63

12 x 375 x 0.075 = £337.50

ACTIVITY
I now decide to invest the same amount of money in the same account but this
time earning compound interest.

How much capital will I have after the same time periods?

ACTIVITY FEEDBACK
Using the following definitions:

Po = £375

R = 0.075

n = the number of years for each part of the question

we get these answers:

375 x 1.0751 = £403.13 (i.e. £375 capital plus £ 28.13 interest)

375 x 1.0755 = £538.36

375 x 1.07512 = £893.17

University of
252 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

ACTIVITY
Visit the following website to use the compound interest calculator:

http://www.maths.surrey.ac.uk/explore/matthewspages/compoundinterest.html

Discounting Techniques
The concepts of compound interest and present value have been
introduced and these can be used to establish discounted values for a
flow of cash, where money flows in and out at different times during the
period under consideration.

Present value is defined as the value of a sum of money at some future

date, described in today’s terms. Let us begin by considering the
equation for present value:
n
PV = FV / ( 1 + i )

This could be written as

PV = FV x Discount Factor

where the Discount Factor (DF) is 1 / ( 1 + i )n

This discount factor provides us with an important tool for investment

appraisal. We can calculate discount factors for different values of
interest rate and investment duration. For example, if we invest for
three years at an interest rate of 5%, we get

DF = 1 / ( 1.05 )3
= 0.8638

More usefully, we could calculate DF values for a range of years, thus:

Discount Factor values for a rate of interest of 5%:

Year 1 0.9524

Year 2 0.9070

Year 3 0.8638

Year 4 0.8227

University of
Sunderland 253
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

Year 5 0.7835

Year 6 0.7462

Year 7 0.7107

Year 8 0.6768

Year 9 0.6446

Year 10 0.6139

Using discount factor values, calculate the present value of £2,500 in six
years time if interest rates are constant at 3%.

From tables we find that the discount factor for 3% over six years is
0.8375.

Therefore, the present value is 2500 x 0.8375 = £2,093.75

Calculate the discount factors for the first five years at interest rates of
6%, 10%, 15% and 20%.

This is rather a tedious process, so instead of recalculating each value

we can use a table of discount values. Such a table can be found at

http://greenbook.treasury.gov.uk/annex06.htm

It is important to note at this stage that there are different effects on

investments that should be understood. In most cases, we consider that
an investment at the bank will increase in value over a period of time.
However, factors such as depreciation will reduce the value of a fixed
asset. The same calculations are carried out, but you must be careful to
ensure that the answer you obtain is appropriate.

For example, money invested at 6% for two years will (from your table)
grow, but an asset depreciating by 6% per annum over two years will
fall in value. The respective calculations become:

The principal invested divided by 0.8900 but the value of the asset
multiplied by 0.8900.

(This shows that it is important to have an understanding of the

mathematical processes involved, not just use the formulae blindly!)

Hence, an investment of £600 in these circumstances will be worth

£674.16, but a machine costing £600 will have dropped in value to
£534.00.

University of
254 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

Fractional Years and Compounding Period

Until now, we have dealt only with time in convenient periods of whole
years. This is an unrealistic view of most situations. The subsequent
over-simplification is likely to introduce unacceptable inaccuracies into
our calculations.

Therefore, we need to be able to carry out our calculations with greater

accuracy. Fortunately, with Excel, this is very simple. If we are dealing
with incomplete years, it is simply a case of entering the period of time
in either fractions or decimals of a year. Thus, for a period of two years
and nine months, we would insert instead 2¾ or 2.75 years. This
remains the case even when we are dealing with numbers raised to a
power number. This is usually written as a decimal, i.e. x2.75 .

This simple adjustment, enables us to use Excel to determine exact

solutions in compounding problems involving difficult periods of time.

ACTIVITY
Calculate the following periods of time as decimals of a year:

1. 2 years 3 months

2. 3 years 4 months

3. 1 year 7 months

4. 8 months

ACTIVITY FEEDBACK
1. 2.25 years

2. 3.33 years

3. 1.58 year

4. 0.67 year

University of
Sunderland 255
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

Continuous Compounding
In economics, especially macroeconomics, ideas of growth and change
over time are very important. Moreover, economic analysis is not just
concerned with comparative statics but also with the dynamic analysis
of how an economy or a market moves from one static equilibrium
position to another. In the following, we’ll look at some the basic
quantitative techniques commonly employed in economic and public
policy analysis to deal with issues of time, growth and dynamics. First,
we’ll consider the questions of compound growth and growth rates.

ACTIVITY
Visit the following website to read more about continuous compounding:

http://www.mathscentre.ac.uk/resources/leaflets/mathcentre/business/
continuous_compounding.pdf

Discrete Compound Growth

Suppose a particular economic variable (Y) has an initial value today

(i.e. period 0) of Y0, and grows at a constant rate per period of r %. At the
beginning of the next period (period 1) it will have grown by r %. Hence
the value of Y at the beginning of period 1 will be:

Y 1 'Y 0 - rY 0 'Y 0 (1 - r )

By the beginning of period 2 Y will have grown by another r % from

what it was at the beginning of period 1. Hence:

Y 2 'Y 1 (1 - r ) 'Y 0 (1 - r )(1 - r ) 'Y 0 (1 - r ) 2

It’s clear from this that the value of Y in any period t will be given by the
formula:

Y 1 'Y 0 (1 - r ) 2

where Y0 is the initial value of Y (in period 0), r is the constant growth
rate per period, and t is the number of periods in the future. This is the
formula for discrete compound growth. (You may already know it as
the formula for calculating the effects of compound interest)

University of
256 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

Example

Suppose the value of Y in period 0, Y0 = 50. It grows at constant rate per

period of 5% (i.e. r = 0.05). What is the value of Y in 10 periods time?
Using the formula we get:

Y 10 ' 50(1 - 0.05 )10

' 50(1.05 )10

' 50 8 1.629
' 81. 45

Continuous Compound Growth

In the case of discrete compound growth discussed above the value of

the variable increases by r % at the end of each period - it does not grow
continuously over time. To see what we mean by continuous compound
growth and to derive the associated formula, let’s first consider a simple
numerical example.

Assume the initial value of a variable (Y) is 100. The growth rate per year
is 10%. However, this growth occurs twice per year; i.e. 5% over the first
half of the year, and 5% over the second. So by the end of the first 6
months Y has increased to 105, and by the end of the year it has risen
another 5% from 105 to reach 110.25 (= 105 x 1.05).

Notice carefully what has happened here. The annual growth rate has
been divided by 2 and this growth rate has been compounded twice a
year. Using the formula for discrete compound growth we could
therefore calculate the value of Y at the end of the year as follows:
2

y 1 ' 100;= 1 -
0.1 >
@ ' 100.1.05 / ' 110. 25
2

< 2 ?

Given any initial value of the variable, Y0 , and any annual growth rate,
r, we can write this formula in general terms as follows:
2t

Y t 'Y 0 ;= 1 - >@
r
< 2?

where t is the number of years in the future. This formula tells us the
future value of Y where the annual growth rate is compounded twice a
year. If the growth rate was compounded, say, four times a year the
formula would clearly become:
4t

Y t 'Y 0 ;= 1 - >@
r
< 4?

University of
Sunderland 257
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

Finally, generalising the formula to compounding n times per year the

formula would be:
nt

Y t 'Y 0 ;= 1 - >@
r
< n?

We still haven’t derived the formula for continuous growth but we’re
getting close! In the case of continuous compound growth the number
of times compounding occurs, n, tends towards infinity. To see the
effect of this we proceed as follows.

First, rewrite the above formula as:

n
2; r > 5
%

Y t 'Y 0 4= 1 - @ 7
3< n? 6

Now let m = n/r and substitute into the above equation to get:
n
2; r > 5
m

Y t 'Y 0 4= 1 - @ 7
3< m? 6

From the definition of m it’s clear that, for a given value of r, as n

approaches infinity so does m. So, for the continuous growth case the
formula can be written as:
n
lim 2 %
5
Y 0 4;= 1 - >@ 7
r
Yt '
m F G 3< n? 6

But it can be shown that the constant e (E 2.71828) is defined as follows:

m
lim ; 1 >
e' =1 - @
m F G< m?

Hence,
n
lim 2 m
5
Y 0 4;= 1 - >@ 7 'Y 0e n
1
Yt '
m F G 3< m? 6

Thus the formula for continuous compound growth is

Y t 'Y 0e n

where Y0 is the initial value, r is the growth rate per period, and t is the
number of periods in the future. It is, of course, the exponential function
that we first looked earlier.

University of
258 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

Example

Assume Y0 = 50 and r = 5% (i.e. 0.05) per year, then the formula becomes

Y t ' 50e 0 .05t

The value of Y in 3 years time with continuous compound growth is

Yt ' 50e 0 .05 ( 3 )

' 50e 0 .15

' 58.09

If we calculate the value of Y for a range of values of t from 0 to, say, 20

and plot these values of Y against t we get the following graph.

Figure 8.12.Exponential function.

This is the typical shape of the exponential function showing

continuous compounded growth of a variable.

Economic Example:

Growth of the Labour Force

In economic growth theory it is often assumed that the labour force

grows continuously at a given exponential rate. Using the formula
derived above, we can write the labour force growth equation as
follows:

L t ' L 0e gt

University of
Sunderland 259
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

where Lt is the labour force at time t, L0 is the current size of the labour
force (at time t = 0) and g is the (exogenous) growth rate.

Equivalent Annual Rate

A notional rate that is quoted on the interest paid on savings and
investments.

It is supposed to demonstrate what your return would be if the interest

was compounded and paid annually instead of, say, monthly.

Suppose an account pays interest more than once a year. The AER is
calculated by adding each interest payment to the original deposit and
then calculating the next interest payment, compounding the interest.

On accounts where interest is paid quarterly, the AER will be a bit

higher than the quoted gross rate because it allows for the interest paid
on the interest.

All advertisements for interest-bearing savings accounts quote AER, so

you can compare returns. AER normally excludes any bonus interest
that may be payable.

Calculation of the Annual Equivalent Rate (AER)

Interest that is calculated under the assumption that any interest paid is
combined with the original balance and the next interest payment will
be based on the slightly higher account balance. Overall, this means that
interest can be compounded several times in a year depending on the
number of times that interest payments are made.

In the United Kingdom, the amount of interest received from savings

accounts is listed in AER form.

Calculated as:
n

AER ' ;= 1 - >@ ( 1

r
< n?

Where: n = number of times a year that interest is paid

r = gross interest rate

The general formula set out at (a) is complicated because it attempts to

cater for any sort of deposit. The following notes aim to clarify its
assumptions and application:

It envisages a depositor making a series of payments (some of which

may be zero) into the account at annual intervals and the product

University of
260 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

provider paying or crediting interest annually which may be at a

different rate (possibly zero) each year.

Deposits are assumed to be made at the beginning of a year and interest

paid at the end of the year.

If deposits or interest payments are not made on anniversaries of the

start date, the formula can be operated using shorter periods and
fractions of the annual rate(s), and the answer compounded up to an
annual rate.

The interest payable in each year is the amount actually payable or to be

added to principal, not an accrual, so that, if paying 7% a year (simple)
after 2 years, the interest is 0% in year 1 and 14% in year 2.

The formula treats all interest as compounded because interest paid

(say) annually is worth more to the depositor than interest paid only
after two years. Interest paid is compounded at the contract rate to
avoid the introduction of an arbitrary reinvestment rate.

Growth Rate calculations

Consider a variable Y which is a function of time t, i.e.

Y ' f (t )

We know that the derivative dY/dt shows the actual rate of change of Y
with respect to changes in t. If we divide the derivative by Y we will get
the proportional (or percentage) rate of change of Y with respect to
changes in t, or, in other words, the instantaneous rate of growth of Y.
Thus

dY / dt 1 dY
'
Y Y dt

is the (instantaneous) proportional growth rate of Y.

The growth rate of a variable can also be expressed in terms of the

natural logarithm of the variable. To see this let’s consider the derivative
of ln Y with respect to time t. To determine this we need to use the
function of a function rule as follows:

d (lnY ) d (lnY ) dY 1 dY
' '
dt dY dt Y dt

From this we see that

1 dY d (lnY )
growth rate of Y ' '
Y dt dt

University of
Sunderland 261
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

So the growth rate of a variable is equal to the derivative of the natural

logarithm of the variable with respect to time. This is a very useful
property which often makes finding the growth rate much easier as the
examples below will demonstrate.

Example: Exponential Function

Consider the exponential function showing continuous compound

growth we discussed above:

Y t 'Y 0e n

To confirm that the growth rate is equal to r let’s take the natural logs of
both sides of the equation to get

lnY t ' ln(Y 0e n )

' lnY 0 - ln(e n )

' lnY 0 - rt ln e

(Note how we have used the various rules of logs here.) But remember,
by the definition of logs,

lne ' 1

Thus, in log-linear form, the continuous growth formula is given by:

lnY t ' lnY 0 - rt

If we now differentiate ln Y with respect to t we will get the

(instantaneous) proportional growth rate of Y. Thus:

d (lnY t ) d (lnY 0 - rt )
' 'r
dt dt

This confirms that the growth rate is equal to r.

Economic Example:

Growth in Unit Labour Costs

Unit labour costs (for the whole economy) are defined as the labour cost
per unit of output; i.e.

WN
ULC '
Y

where W = the average money wage rate, N is the total labour force
employed, and Y is real output (GDP). Hence, WN is the total labour
cost.

University of
262 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

This equation can be easily rearranged to give

W
ULC '
Y /N

which states that unit labour costs are equal to the money wage rate
divided by labour productivity (= Y/N i.e. output per worker). Taking
logs of both sides of this equation gives

lnULC ' lnW ( ln;=

Y >
@
<N ?

If we now differentiate with respect to time (t) we will get the

relationship between the growth rates. Thus:

d .lnULC / d .lnW / d .lnY / N /

' (
dt dt dt

This equation tells us that the rate of growth of unit labour costs equals
the rate of growth of money wages less the rate of growth of labour
productivity. Thus, for example, unit labour costs will remain constant
if money wages and productivity grow at the same rate.

Annuities and Sinking Funds

The term annuity is used in finance theory to refer to any terminating
stream of fixed payments over a specified period of time. This usage is
most commonly seen in academic discussions of finance, usually in
connection with the valuation of the stream of payments, taking into
account time value of money concepts.

An ordinary annuity (also referred as annuity-immediate) is an

annuity whose payments are made at the end of each period (e.g. a
month, a year). The values of an ordinary annuity can be calculated
through the following:

Let:

r = the annual interest rate.

t = the number of years.
m = the number of periods per year.
i = the interest rate per period.
n = the number of periods.

University of
Sunderland 263
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

Note:

r
i'
m
n ' tm

Also let:

P = the principal (or present value).

S = the future value of an annuity.
R = the periodic payment in an annuity (the amortized payment).
2 .1 - i / n ( 1 5
S 'R4 7 ' R HS n|r
43 i 76

Also:

2 1 ( (1 -1i ) 5
P 'R4 7 ' R Ha n|i
43 i 76

Clearly, in the limit as n increases,

lim R
P '
nF G i

Thus, even an infinite series of finite payments with a non-zero discount

rate has a finite Present Value.

Other types of annuities

! Fixed annuities - These are annuities with fixed

payments. They are primarily used for low risk
investments like government securities or corporate
bonds. Fixed annuities offer a fixed rate up to ten years
but are not regulated by the Securities and Exchange
Commission (SEC).

! Variable annuities - Unlike fixed annuities, these are

regulated by the SEC. They allow you to invest in
portions of money markets.

! Equity-indexed annuities - Lump sum payments are

made to an insurance company.

University of
264 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

Calculating an annuity

Before we use the annuity formula, let’s solve a short 3-year example the
“long way”.

First we need the compound interest formula which is:

Total = Principal x ( 1 + Rate )years

Now let’s say the amount that we invest annually is £2,000 per year and
the interest rate is 8%.

The £2,000 invested 3 years ago has become

£2,000 * (1.08)3 = £2,000 * 1.259712 = £2,519.424

The £2,000 invested 2 years ago has become

£2,000 * (1.08)2 = £2,000 * 1.1664 = £2,332.80

The £2,000 invested 1 year ago becomes

£2,000 * (1.08)1 = £2,000 * 1.08 = £2,160.00

Adding up all 3 yearly amounts, we obtain £7,012.22

As you can see, the mathematics of this can be a little cumbersome

especially when the time involved gets larger.

To make these calculations a little easier, there is a formula:

2 .1 ( r / .n -1/ ( 1 5
Total ' Amount H 4 7 ( Amount
43 r 76

where the AMOUNT is the annual amount invested each year,

‘n’ is the number of years and

‘r’ is the annual rate of the investment.

So, we have:

£2,000 * { [(1 + .08) (3 + 1) -1] I .08 } – £2,000

£2,000 * { [1.36048896 -1] I .08 } – £2,000
£2,000 * { .36048896 I .08 } – £2,000
£2,000 * { 4.506112 } – £2,000
£9,012.224 – £2,000
£7,012.22

University of
Sunderland 265
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

This is the same as the answer we obtained using the “long” method.

ACTIVITY
Luckily, you have a third and easier method - the annuity calculator. An
example can be found at:

http://www.local.odpm.gov.uk/pfi/annucalc.xls

Use this to help you understand how annuities work.

What is a sinking fund?

A sinking fund is a sum of money set up to build up an amount to pay
for major work, usually to a block of flats such as replacing a roof or
door-entry system. This means that when major work is carried out, the
bill will not be as large because of the money collected over a period of
time from different leaseholders. Your lease will say if you have to pay
towards this fund. If you do, the amount you have to pay will be shown
in your yearly service charge statement.

When new homes are built, we work out how much certain items would
cost to replace, and how long they will last. This means that we know
how much to collect each year, so we should have enough to cover some
or all of the costs in the future. We aim to review all the sinking funds
around every five years to make sure leaseholders are not paying too
much or too little.

If you are selling your home, you should tell your solicitor that you
contribute to a sinking fund so the buyers know that there is money
built up to cover some or all of the cost of major one-off work. This may
make your home more attractive to buyers. However, all contributions
you pay into the fund will stay in the fund if you decide to sell.

How does the sinking fund work?

When you buy the property the life cycle of all of the long-term elements
was calculated together with a replacement cost. The sinking fund was
then set up on the basis of building up enough money to meet these
long-term costs. This is reviewed annually to ensure the fund is not
falling behind at the level predicted for the future. If we did not have a
sinking fund then leaseholders may be faced with having to pay large
amounts in any one year if major repairs are required. The lack of a
sinking fund may put off future purchasers from buying your home.

University of
266 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

Debt repayment
Banks generally set repayments on loans and mortgages in equal
payments over a fixed period of time. Within these regular payments
the mix of interest and principal changes as time passes. A single
formula can easily calculate the repayments on a loan of £x,000 at y%
over z years. The real world is often more complicated and interest rates
change at irregular intervals, often part of the way through a particular
repayment period.

1. Basic data about loan

The initial stage of this example is simple. £35,000 is borrowed and to be

repaid over 1 year in equal monthly payments at an interest rate of 6%.

The Excel PMT() function is used in cell C7 to calculate the monthly

repayment. It takes the form: PMT (InterestRate, NumberOfPeriods,
Principal, FutureValue, PaymentsDue). The interest rate should relate
to a single repayment period and not the annual rate. In this example it
is 1/12th of 6%.

[C7] = - PMT(C6, B4, B3, 0, 0)

Note that in this example a negative operator (-) is placed in front of the
function in order to return a positive value. By default Excel will display
repayments as negative amounts.

For loans which are to be repaid over a long period of time it is possible
that the interest rate will change and the monthly repayments will be
revised to reflect the new rate. Calculating an interest rate change part
way through a period can be done in different ways and I don’t know
whether there is a standard procedure adopted by all banks. The result
can vary depending upon factors such as whether they treat all months
as equal 12ths of a year in terms of days. The Excel CUMPRINC function
can only deal with whole months (or periods) and treats any period
value as an integer.

University of
Sunderland 267
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

We can make the example more complicated and apply a new rate of
interest (5.0%) that comes into effect after 12 February. February is in
period 10 of the loan. It is necessary to perform a separate calculation for
February (when the rate changes part way through the month) and then
for the subsequent months. For the purposes of calculations, the unpaid
balance of the loan can then be treated as a new loan to be repaid in
equal amounts over the remaining number of periods.

A weighted average of the two interest rates determines the combined

rate for February (0.452% - See D16 in the diagram below). This is based
upon the number of days at which each rate is charged.

[C13] =B13/B15
[C14] =C15-C13
[D16] =SUMPRODUCT(C13:C14, D13:D14)

The new rate of 5% (=0.417%) will then apply to the whole months of
March and April.

Two new functions can be used to calculate the interest and the
principal amounts for any period (or range of periods) of the loan. The
sum of the interest and the principal equals the result of the PMT()
function.

2. Composite rate for month of change

The CUMPRINC function calculates the amount of principal repaid in

between a specified range of periods. It requires the following series of
arguments:

CUMPRINC(InterestRate, NumberOfPeriods, Principal, Start_period,

End_period, PaymentsDue). Similarly the CUMIPMT function returns
the interest paid in between a specified range of periods.
CUMIPMT(InterestRate, NumberOfPeriods, Principal, Start_period,
End_period, PaymentsDue).

University of
268 Sunderland
Business Modelling for Decision Making Unit 8 – Statistical Process Control and Financial Mathematics

In February the number of periods will be 3 (Feb - Apr) and the start and
end period will both be 1 because the interest rate applies to that period
only. The principal will be the remaining unpaid amount (£8,947) and
not the original £35,000.

Note that in common with many other Excel financial functions, these
both have a final parameter (PaymentsDue) which refers to whether the
payment is made at the beginning or end of the period. In this example
the value of ‘0’ indicates the end of the period.

3. Calculation of payments

[F19] =CUMIPMT(D19, C21-B19+1, E19, 1, C19-B19+1, 0)

[G19] =CUMPRINC(D19, C21-B19+1, E19, 1, C19-B19+1, 0)
[H19] =G19+F19
[I19] =H19/(C19-B19+1)
[E20] =E19+G19

The formula in cell [F21] might be easier to understand if shown as

=CUMIPMT(0.417%, 2, 5978.4, 1, 2, 0)

The sum of the three Principal values (in column G) exactly equals the
original value of the loan (£35,000).

This methodology is a sound accounting basis for deriving interest

charges and repayment terms. They may not however precisely match
the calculations of any particular lender - there are a lot of complicating
factors and variables. Of key importance in any such arrangements is
the frequency or periods at which the interest is calculated. Completely
different values will be obtained for a loan over one year if the
repayment periods are either annual, monthly or daily.

University of
Sunderland 269
Unit 8 – Statistical Process Control and Financial Mathematics Business Modelling for Decision Making

REVIEW ACTIVITY
Using your knowledge of the concepts discussed within this unit, prepare a
discussion paper for your senior management to demonstrate the value of
Microsoft Excel within your organisation. This should provide sufficient
information to justify a proposal to invest significant sums of money in
extensive management training in the use of Excel as a business modelling
application.

Include in your paper at least three detailed examples of its use in practical
situations.

Summary
This concluding unit considers a variety of topics linked to the use of
statistics in the processes of process control and financial modelling.

The initial discussion of Deming’s ideas leads onto the introduction of

control charts and process control.

The second section examines the use of Microsoft Excel in calculating

statistical spread and solving interest problems. This latter area is
particularly important in an economic climate where the value of
money changes so rapidly. Calculations about compounding,
equivalent annual rates and annuities all rely on the effective use of
interest percentages. The necessary calculations performed within a
spreadsheet enable a manager to model the changes within a specific
situation so that an optimum strategy can be formulated.

University of
270 Sunderland