Block 1: Decision-making under uncertainty and

modelling
There is an introduction video on the VLE, you can access it here:
https://emfssvideo.s3.amazonaws.com/MT%26ST/ST2187/ST2187_Block1_Intro.mp4
We begin the course by considering decision-making under uncertainty. Decisions are taken in the
present with uncertain future outcomes. In business, decision-making is problematic precisely because
the outcomes of decisions are unknown at the time the decision is made. To assist us, we construct
models to simplify the (complex) real world, hopefully retaining the most important aspects of reality
such that we do not depart from reality too much.
After completing this block, you should be able to:
 describe how decisions taken in the present have uncertain future outcomes
 discuss the features of a good model, in that it achieves maximum simplicity while retaining
maximum realism
 explain how models typically have simplifying assumptions and appreciate the importance of
the validity of assumption.

Readings
Albright, S and Winston, W.L, Business Analytics Data Analysis & Decision Making, (Cengage
Learning, 2017) 6th edition [ISBN 9781305947542] Chapter 1.

Decision-making under uncertainty and modelling

The Monty Hall problem

You are taking part in a gameshow. The host of the show, who is known as Monty, shows you three
outwardly identical doors. Behind one of them is a prize (a sports car), and behind the other two are
goats.
You are asked to select, but not open, one of the doors. After you have done so, Monty, who knows
where the prize is, opens one of the two remaining doors.
He always opens a door he knows will reveal a goat, and randomly chooses which door to open when
he has more than one option (which happens when your initial choice contains the prize).
After revealing a goat, Monty gives you the choice of either switching to the other unopened door or
sticking with your original choice. You then receive whatever is behind the door you choose. What
should you do, assuming you want to win the prize?
Rationale
The famous ‘Monty Hall’ problem is a classic example of decision making under uncertainty. We will
solve this problem formally, but for now appreciate that at each round of the game you, as the player,
do not know where the sports car is.
To begin with, the only certainty you have is that the sports car must be behind one of the three doors.
You may, or may not, initially chose the ‘correct’ door (assuming you want to win the prize!) but
there is no certainty in your choice.
Upon revealing a goat behind one of the doors you did not choose, you still face uncertainty - the only
certainty you have is that the sports car must be behind one of the two unopened doors.
The ‘controversy’ arose over the American game show ‘Let’s Make a Deal’, and the New York
Times (among others) devoted two pages to the problem, readers’ letters etc.
Bewildered game show players wrote to Marilyn vos Savant, an advice columnist for Parade
Magazine, and asked for her opinion in her ‘Ask Marilyn’ column. Savant - who is credited by the
Guinness Book of Records as having the highest IQ of any woman in the world - gave her decision.
She said, ‘you should change your choice’. There then followed a long argument in the
correspondence columns, some supporting Savant’s decision and others saying that it was nonsense.
Before reading on, what do you think, and why?

Bayesian updating
We now solve probabilistically the Monty Hall problem.
Suppose the three doors are labelled A, B and C. Let us define the following events.
 DA, DB, DC: the prize is behind Door A, B and C, respectively.
 MA, MB, MC: Monty opens Door A, B and C, respectively.
Suppose you choose Door A first, and then Monty opens Door B (the answer works the same way for
all combinations of these). So Doors A and C remain unopened.
What we want to know now are the conditional probabilities P(DA | MB) and P(DC | MB).
You should switch doors if P(DC | MB) > P(DA | MB), and stick with your original choice otherwise.
(You would be indifferent about switching if it was the case that P(DA | MB) = P(DC | MB).)
Suppose that you first choose Door A, and then Monty opens Door B. Bayes’ theorem tells us that:
𝑃(𝑀𝐵 |𝐷𝐶 ) 𝑃(𝐷𝐶 )
𝑃(𝐷𝐶 |𝑀𝐵 ) = .
𝑃(𝑀𝐵 |𝐷𝐴 ) 𝑃(𝐷𝐴 ) + 𝑃(𝑀𝐵 |𝐷𝐵 ) 𝑃(𝐷𝐵 ) + 𝑃(𝑀𝐵 |𝐷𝐶 ) 𝑃 (𝐷𝐶 )
We can assign values to each of these.
 The prize is initially equally likely to be behind any of the doors. Therefore, we have
P(DA)=P(DB)=P(DC) = 1/3.
 If the prize is behind Door A (which you choose), Monty chooses at random between the two
remaining doors, i.e. Doors B and C. Hence P (MB|DA) = 1/2.
 If the prize is behind one of the two doors you did not choose, Monty cannot open that door,
and must open the other one. Hence P (MB | DC) = 1 and P (MB | DB) = 0.
Putting these probabilities into the formula gives:
1
1 ×3
𝑃(𝐷𝐶 |𝑀𝐵 ) = = 2/3
1 1 1 1
×
2 3 + 0 × 3 + 1 × 3
and hence P(DA | MB) = 1 − P(DC | MB) = 1/3 (because also P(MB | DB) = 0 and so P(DB | MB) = 0).
The same calculation applies to every combination of your first choice and Monty’s choice.
Therefore, you will always double your probability of winning the prize if you switch from your
original choice to the door that Monty did not open.
The Monty Hall problem has been called a ‘cognitive illusion’, because something about it seems to
mislead most people’s intuition. In experiments, around 85% of people tend to get the answer wrong
at first. The most common incorrect response is that the probabilities of the remaining doors after
Monty’s choice are both 1/2, so that you should not (or rather need not) switch.
This is typically based on ‘no new information’ reasoning. Since we know in advance that
Monty will open one door with a goat behind it, the fact that he does so appears to tell us nothing new
and should not cause us to favour either of the two remaining doors - hence a probability of 1/2 for
each (people see only two possible doors after Monty’s action and implicitly apply ‘classical’
probability by assuming each door is equally likely to reveal the prize).
It is true that Monty’s choice tells you nothing new about the probability of your original choice,
which remains at 1/3. However, it tells us a lot about the other two doors. First, it tells us everything
about the door he chose, namely that it does not contain the prize. Second, all of the probability of that
door gets ‘inherited’ by the door neither you nor Monty chose, which now has the probability 2/3.
So, the moral of the story is to switch! Note here we are using updated probabilities to form
a strategy - it is sensible to ‘play to the probabilities’ and choose as your course of action that which
gives you the greatest chance of success (in this case you double your chance of winning by switching
door). Of course, just because you pursue a course of action with the most likely chance of
success does not guarantee you success!
If you play the Monty Hall problem (and let us assume you switch to the unopened door), you
can expect to win with a probability of 2/3, i.e. you would win 2/3 of the time on average. In any
single play of the game, you are either lucky or unlucky in winning the prize. So you may switch and
end up losing (and then think you applied the wrong strategy - hindsight is a wonderful thing!) but in
the long run you can expect to win twice as often as you lose, such that in the long run you are better
off by switching!
If you feel like playing the Monty Hall game again, I recommend visiting this site.
In particular, note how at the end of the game it shows the percentage of winners based on multiple
participants’ results. Taking the view that in the long run you should win approximately 2/3 of the
time from switching door, and approximately 1/3 of the time by not switching, observe how the
percentages of winners tend to 66.7% and 33.3%, respectively, based on a large sample size. Indeed,
in statistical inference as the sample size increases we tend to get a more representative (random)
sample of the population. Here, this equates to the sample proportions of wins converging to
their theoretical probabilities.
Note also the site has an alternative version of the game where Monty does not know where the sports
car is! Good luck!

Decision-making under uncertainty

To study, or not to study? To invest, or not to invest? To marry, or not to marry?
These, among others, are decisions many of us face during our lives. Of course, decisions have to be
taken in the present, with uncertain future outcomes.
In the workplace, for example, making decisions is the most important job of any executive.
However, it is also the toughest and riskiest job. Bad decisions can damage a business, a reputation
and a career, sometimes irreparably. Good decisions can result in promotion, a strong reputation and
making money!
Today we are living in the age of technology, with two important implications for everyone.
 Technology has made it possible to collect vast amounts of data - the era of ‘big data’.
  Technology has given many more people the power and responsibility to analyse
data and make decisions on the basis of quantitative analysis.
A large amount of data already exists, and it will only increase further in the future. Many companies,
rightly, are seeing data-driven decision-making as a source of competitive advantage. By using
quantitative methods to uncover and extract the information in the data and then acting on this
information - guided by quantitative analysis - they are able to gain advantages which their more
qualitatively-oriented competitors are not able to gain.
Today, demand for people with quantitative skills far exceeds supply, creating a ‘skills deficit’. With
demand set to increase further, and supply failing to keep pace with demand, then Economics 101 (or
rather EC1002 Introduction to economics!) will tell you that the price increases whenever demand
exceeds supply. Of course, the ‘price’ being referred to here is that of an employee, i.e. the salary
which quantitative staff can command (already high) is set to rise even further.
Decision-making is a process when one is faced with a problem or decision having more than one
possible outcome. The possible results from the decision are a function of both internal variables
(which we can control) and external variables (which we cannot control), each of which cannot be
expressed with certainty. Hence the outcome cannot be known in advance with certainty.
When evaluating all decision-making, we start with structuring the problem.
Example: ‘What price should we charge for our new product?’
Determine the set of possible alternatives, for example:
 £1.00
 £2.00
 £3.00
 £4.00.
Determine the possible criteria which could be used to evaluate the alternatives:
 qualitative analysis
 quantitative analysis.
An example of a qualitative analysis:
 ‘Well, last time we brought a new product to the market, we priced it at £2.00 and we sold out
on the first day. This time let’s price it higher.’
 This is all very well, but how much ‘higher’? How would we justify a specific increase of
£𝑥x?
An example of a quantitative analysis:
 ‘What do we know about current market demand?’
 ‘What do we know about competitive market factors?’
 ‘Where will we manufacture the new product and how much will it cost to bring it to the
market?’
 ‘How will we advertise and how much will the advertising cost?’
 ……
Note this is not an exhaustive list, but clearly market demand, competition, production costs and
advertising expenditure (among other factors) are likely to be relevant to the price-setting problem.
For all decisions, we need to determine the influencing factors which could either be internal or
external, such as:
 demand and competitive supply
 availability of labour and materials
 ……
which are then used to derive expected results or consequences. Of course, determining which are the
influencing factors, and their corresponding weights of influence, is not necessarily easy, but a
thoughtful consideration of these is important due to their cumulative effect on the outcome.
In a qualitative analysis, once we have determined a preliminary list of the factors which we think
will affect the possible outcomes of the decision:
 the management team ‘qualitatively’ evaluates how each factor could affect the decision
 this discussion leads to an assessment by the decision-maker
 the decision is made followed by implementation, if necessary.
For example, in a qualitative analysis we might describe the potential options in a decision tree in
which we can include the concept of probable outcomes. We could make this assessment using the
(qualitative) qualifiers of:
 ‘optimistic’
 ‘conservative’
 ‘pessimistic’.
However, a qualitative approach inevitably is susceptible to judgement and hence biases on the part of
the decision-makers. ‘Gut instinct’ can lead to good outcomes, but in the long run is far from optimal.
In a quantitative analysis, once we have determined a preliminary list of the factors which we think
will affect the possible outcomes of the decision, we need to ask the following questions.
 What do we know?
 What data can we ‘mine’ which will help us understand the factors and the effect each will
have on the possible outcomes?
In a quantitative analysis, the evaluation becomes a process of using mathematics and statistical
techniques. These are used to find predictive relationships between the factors, the potential outcomes
of the problem we are seeking to understand and the decision we are seeking to make.
Our objective becomes to define mathematically the relationships which might exist. Next, we
evaluate the significance of the predictive value of the relationships found. An assessment of the
relationships which our analysis defines leads us to be able to quantitatively express the expected
results or consequences of the decision we are making.

Uncertainty in the news

‘News’ is not ‘olds’. News reports new information about events taking place in the world - ignoring
fake news!
Especially in business news, you will find numerous reports discussing the many uncertainties being
faced by business. While uncertainty certainly(!) makes life exciting, it makes decision-making
particularly challenging. Should a firm increase production? Advertise? Cut back? Merge?
Decisions are made in the present, with uncertain future outcomes. Hence many media reports will
comment on the uncertainties being faced.
Of course, some eras are more uncertain than others. Indeed, 2016 was the year of the ‘black swan’ -
low-probability, high-impact events - with the Brexit referendum vote and the election of Donald
Trump to the White House the main geopolitical stories. Both outcomes were considered unlikely, yet
they both happened. Some prediction markets priced in a 25% probability of each of these outcomes,
but a simple probability calculation (as will be covered next week) would equate this to tossing a fair
coin twice and getting two heads - perhaps these were not such surprising results after all!
Taking Brexit as an example, immediately after the referendum result was known, uncertainty arose
about exactly what ‘Brexit’ meant. Exiting the single market? Exiting the customs union?
Financial markets, in particular, tend to be very sensitive to news. Even stories reporting comments
from influential people, such as politicians, can move markets - sometimes dramatically! For
example, read ‘Flash crash sees the pound gyrate in Asian trading’.
At one stage it fell as much as 6% to $1.1841 - the biggest move since the Brexit vote - before
recovering. It was recently trading 2% lower at $1.2388. It is not clear what triggered the
sudden sell-off. Analysts say it could have been automated trading systems reacting to a news
report. The Bank of England said it was ‘looking into’ the flash crash. The sharp drop came
after the Financial Times published a story online about French President François Hollande
demanding ‘tough Brexit negotiations’.
Increasingly, quantitative hedge funds and asset managers will trade algorithmically, with computers
designed to scan the internet for news stories and interpret whether news reports contain any useful
information which would allow a revision of probabilistic beliefs (we have already seen an example
of this with the Monty Hall problem) which can be viewed as an example of ‘Bayesian updating’.
Here, the demand for ‘tough Brexit negotiations’ by the then French President would be interpreted as
being bad for the UK, which would lead to a further depreciation in the pound sterling.
‘These days some algos trade on the back of news sites, and even what is trending on social
media sites such as Twitter, so a deluge of negative Brexit headlines could have led to an algo
taking that as a major sell signal for the pound,’ says Kathleen Brooks, research director at
City Index.
So, from now on, when you read (or listen to) the news, keep an eye out for (or ear open to!) the word
‘uncertainty’ and consider what kinds of decisions are being made in the face of the uncertainty.

Simple vs. complexity – the need for models

Is the real world:
a. nice, simple and easy?
b. big, horrible and complicated?
Answer (b)!
Although we care about the real world, seek to understand it and make decisions in reality, we have
an inherent dislike of complexity. Indeed, in the social sciences the real world is a highly complex
web of interdependencies between countless variables. For example, in economics, what determines
the economic performance of a country? From national income accounting in Economics 101 (or
rather, again, EC1002 Introduction to economics!) you might say consumption, investment,
government spending and net exports, but what affects, say, consumption? Consumer confidence?
Perhaps, but what drives consumer confidence? Consumers’ incomes? Consumers’ inflationary
expectations? Fears of job insecurity? Perceived level of economic competency of the government.
The weather? Etc.
So in order to make any sense of the real world, we will inevitably have to simplify reality. Our tool
for achieving this is a model.
A model is a deliberate simplification of reality. A good model retains the most important features of
reality and ignores less important details.
Immediately we see that we face a trade-off. The benefit of a model is that we simplify the complex
real world. The cost of a model is that the consequence of this simplification of reality is a departure
from reality. Broadly speaking, we would be happy if the benefit exceeded the cost, i.e. if the
simplicity made it easier for us to understand and analyse the real world while incurring only a
minimal departure from reality.
The London Underground map
The world-famous London Underground map is an excellent example of a model for getting from
point A to point B. The map contains the most important pieces of information for reaching your
intended destination:
 distinct names and colours for each line
 the order of stations on each line
 the interchange stations between lines
while less important details are ignored, such as:
 the depth of each tunnel
 the exact distance between stations
 the non-linear nature of the tunnels under the ground.
Of course, an engineer would likely need to know these ‘less important details’, but for a tourist
visiting London such information is superfluous and the map is very much fit-for-purpose. However,
we said above that a model is a departure from reality, hence some caution should always be
exercised when using a model. Blind belief in a model might be misleading. For example, the map
above fails to accurately represent the precise geographic location of stations.
If we look at the geographically-accurate map we see the first map can be very misleading in terms of
the true distance between stations - for example, the quickest route from Holborn to Temple (LSE's
nearest stations) is to walk!
Also, even line names can be a model - the Circle line (in yellow) is clearly not a true circle! Does it
matter? Well, the Circle line forms a loop and it is an easy name to remember, so arguably here the
simplification of the name outweighs the slight departure in reality from a true circle!
Our key takeaway is that models inevitably involve trade-offs. As we further simplify reality (a
benefit), we further depart from reality (a cost). In order to determine whether or not a model is
‘good’, we must decide whether the benefit justifies the cost. Resolving this benefit-cost trade-off is
subjective - further adding to life’s complexities.

Safe to assume? Beware, when model assumptions go wrong!

We have defined a model to be a deliberate simplification of reality. To assist with the process of
model building, we often make assumptions - usually simplifying assumptions.
Returning to the (geographically-accurate) London Underground map, the Circle line (in yellow) is
not a perfect geometric circle, but here it is reasonable to assume the line behaves like a circle as it
does go round in a loop. So adopting the name ‘Circle line’ assumes its path closely approximates a
circle. I do not think anyone would seriously suggest the name ‘Circle line’ is inappropriate!
Moving to statistical models, we often make distributional assumptions, i.e. we assume a particular
probability distribution for a particular variable.
Recall the normal distribution, the familiar bell-shaped curve:

The normal distribution is frequently-used in models. One example is that financial returns on assets
are often assumed to be normally distributed.
Under this assumption of normality, the probability of returns being within three standard deviations
of the mean is approximately 99.7%. This means that the probability of returns being more than three
standard deviations from the mean is approximately 0.3%. (In the graph above, the mean is 0 and the
standard deviation is 1, so ‘mean ±± 3 standard deviations’ equates to the interval [−3,3][−3,3]. This
means that 99.7% of the total area under the curve is between −−3 and 3.)
Assuming market returns follow a normal distribution is fundamental to many models in finance, for
example Markowitz’s modern portfolio theory and the Black-Scholes-Merton option pricing model.
However, this assumption does not typically reflect actual observed market returns and ‘tail events’,
i.e. black swan events (which recall are low-probability, high-impact events), tend to occur more
frequently than a normal distribution would predict!
For now, the moral of the story is to beware assumptions - if you make a wrong or invalid
assumption, then decisions you make in good faith may lead to outcomes far from what you expected.
As an example, the subprime mortgage market in the United States assumed house prices would only
ever increase…… but what goes up usually comes down at some point!
Computing requirements of the course

Microsoft Excel requirements of the course

Although in the final examination you will not have access to a computer, and hence will not
be using Microsoft Excel to answer questions, clearly most of the examples and problems seen in the
course use spreadsheet modelling which is best demonstrated with Excel files, denoted by the file
extension ‘.xlsx’.
In order to gain maximum (career!) benefit from the course, you are strongly encouraged to follow the
numerous Excel spreadsheets provided in each block (with the exceptions of the introductory Block 1
and the Tableau-focused Block 4).
Many of the examples in the accompanying course textbook make use of the powerful DecisionTools
Suite by Palisade Corporation, which is a suite of separate add-ins. For the assessment of this course,
it is not essential to use these add-ins, although if you wish to replicate the worked examples you will
surely benefit from them!
To download and install the DecisionTools Suite, use this link. Note you will need to use the index of
the back of the textbook for the requested password.
Of the add-ins, examples in this course will mainly use:
o StatTools - an add-in for statistical data analysis
o PrecisionTree - a graphical-based add-in for creating and analysing decision trees
o @RISK - an add-in for simulation.
Note for Mac users: these add-ins are supported in a Windows environment only.
Excel’s Analysis ToolPak add-in can perform many types of statistical analysis. For details, please
consult the document Using Excel’s Analysis ToolPak Add-In.
Depending on your level of previous experience using Excel, you may find the following tutorials
informative.
o For Windows users: Excel Tutorial for Windows.xlsx
o For Mac users: Excel Tutorial for Mac.xlsx

Tableau requirements of the course

Block 4 is a stand-alone block of the course focusing on data visualisation using the popular
software Tableau.
When you reach Block 4, the block explains how you may access a free licenced version of Tableau
as well as accessing the excellent free training videos produced by Tableau to help you orientate
yourself with the Tableau environment.
In this course you will be required to use Tableau to complete the 30%-weighted coursework
component of the assessment. Please see the Assessment block on the VLE for details.

Block 1: Test your understanding

This activity is a multiple-choice quiz, please complete it on the VLE using the link below:
https://emfss.elearning.london.ac.uk/mod/quiz/view.php?id=28445
Learning Outcomes Checklist
Use this to assess your own understanding of the chapter. You can always go back and amend the
checklist when it comes to revision!
 describe how decisions taken in the present have uncertain future outcomes
 discuss the features of a good model, in that it achieves maximum simplicity while retaining
maximum realism
 explain how models typically have simplifying assumptions and appreciate the importance of
the validity of assumption.