

Perform Quantitative Risk Analysis

Lecture # 11
Perform Quantitative Risk Analysis

 Perform Quantitative Risk Analysis is the process of

numerically analyzing the combined effect of identified
individual project risks and other sources of uncertainty
on overall project objectives.

 The key benefit of this process is that it quantifies

overall project risk exposure, and it can also provide
additional quantitative risk information to support risk
response planning.

 This process is not required for every project, but where

it is used, it is performed throughout the project.
 Important Considerations

 High quality risk data

 Sound baselines

 Specialized risk software

 Expertise in risk models

 Additional cost and time

 Two-way facilitation to plan risk responses

 Only reliable method to assess overall project risk

 When to use?

 Specified in RMP

 Large, complex projects

 Strategically important projects

 Contractual requirement

 Stakeholder requirement

 High risks
 Quantitative Risk Assessment

When all identified risks have been scored using qualitative risk
analysis, high priority risks can be selected from a ranked list for
more detailed analysis i.e. quantitative risk analysis.

The task here is to assess, quantitatively in some way, the three

components of each risk – the probability of the occurrence of the
risk event; the magnitude or impact of the risk event if it should
occur; and the period during which the risk will be “live”, when the
risk taker will be “at risk”.

Assessing:
– Probability
– Impact
– Duration
 Assessing Probability

 The probability (chance) of an event occurring can

be established in any one of three ways:
 Subjectively
 a priori
 Objectively
 Subjective Probabilities

 Where objective or a priori probabilities are

inappropriate or impossible to assign, subjective
probabilities may be necessary.
 This is often the case with real world problems, where
risk events are rarely replicated exactly, but we have
some opinion about them.
 Subjective probabilities are reflections of peoples’
opinions about the likelihood of events occurring.
 Subjective probabilities are prone to some degree of
human error.
Subjective Probabilities: Delphi

 How do we establish subjective probabilities, and

what errors are they prone to?
 One method of establishing subjective probabilities
is the “Delphi” technique.
 With this technique, a group of experts is formed,
and each is asked to give an opinion about the
probability to be established.
 The experts are asked for their individual opinions,
and do not consult each other.
Subjective Probabilities: Delphi

 The opinions are transformed into ordinal values (e.g. 5-

point rating scales), or cardinal values (lowest / highest /
most likely value to be encountered)
 The Delphi panel data are collected and collated, and
descriptive statistics calculated (mean, range, standard
deviation).
 This information is given back to each expert, together
with his or her original opinion.
 Each expert is now invited to amend his or her opinion.

 The process is repeated until no further changes in the

statistics are detected, and the mean of the final values
is accepted as the established probability.
Subjective Probabilities: Delphi

 The Delphi process can be lengthy.

 It may also be necessary to devise leading questions

to focus the group.

 Highlighting inconsistencies by asking similar

questions can help to minimize errors.

 Delphi is also used in research to elicit expert

opinion about a range of issues, with the aim of
arriving at a consensus view.
Subjective Probabilities – Errors and Biases

 People tend to over-estimate low probabilities.

 People do not generally calibrate their subjective

assessments, i.e. they do not monitor their own
performance to check the consistency and accuracy of
their assessments.
 People tend to depend upon heuristics (rules of thumb),
e.g. They will use a Pareto 80/20 rule without
consideration of the specifics of the situation.
 Examples that employ heuristics include using trial and
error, a rule of thumb, an educated guess, an intuitive
judgment, a guesstimate, profiling, or common sense.
 Types of Subjective Biases

Base Rate Sample Availability

Fallacy Size Fallacy Bias
Subjective Probabilities – Errors and Biases

 Base rate fallacies occur when other diagnostic

information is preferred (without proper
justification) to the basic parameters.

 or when we are distracted by misleading

information.
Subjective Probabilities – Errors and Biases
Example

 Records over the past eight years show that tiling sub-
contractor completes 25% of all projects on time.
 You find that, out of his last five jobs, two were late but
the last three jobs were finished on time.
 You are wondering whether to award him another
contract.
 What is the probability that he will complete this new
contract on time?
 (a) 0.5? (b) 0.25? (c) 0.75? (d) 0.60?
Subjective Probabilities – Errors and Biases

 The 8 year base rate indicates that the probability

of completing on time is 0.25.

 Without knowing the particular circumstances, the

data for the last five projects should be treated as
an unrepresentative small sample of the whole of
Tiling projects over the past 8 years. We should
therefore stick with the base rate.
Subjective Probabilities – Errors and Biases

 The potential to commit a base rate fallacy should

force decision-makers to consider:

 Have the underlying conditions of the base rate

changed sufficiently to warrant a change in the base
rate itself?

 We should have asked if anything had happened

to Tiling organization recently (new contract
manager, incentive scheme introduced) which
would justify making a change in the base rate.
 Subjective Probabilities – Errors and Biases

In the sample size fallacy, people tend to ignore a basic property of

sample sizes.
 Assume that, on average, the re-work rate in construction in Pakistan is 25%
(i.e. 25% of all work has to be re-done).
 Two construction firms have similar quality control procedures. Company A
has 2000 employees, of whom 1500 are site operatives.
 Company B has 32 employees, of whom 24 are site operatives.
 Over a period of a year, which company is likely to record more days with re-
work exceeding 30%, and less than 20%?
 Company A? Company B?
 Both about the same number of days?
Subjective Probabilities – Errors and Biases

 The availability bias occurs when a probability is

over-estimated due to more immediate retrieval of
particular instances.

 i.e. a probability value is preferred because it is

more immediately available to us through the
influence of personal experience.
Subjective Probabilities – Errors and Biases

 You will be more likely to assign a higher probability

to a scaffolding accident occurring, if you have had
personal experience of one, or have recently heard
of one.

 For example, if the real probability of a scaffolding

accident is 1 in 100,000; personal experience of
such an accident happening might cause us to raise
our estimate of the probability to as high as 1 in
1,000.
 Assessing Probability

 The probability (chance) of an event occurring can

be established in any one of three ways:
 Subjectively
 A priori
 Objectively
 A Priori

 a priori probabilities for risk management are

established by assigning prior known probabilities
taken from other contexts.

 Using a six-sided dice, what is the chance of

throwing a six?

 1/6 (or 16.67% or 0.1667)

 A Priori – Joint Probability

 What is the probability of throwing a number

greater than 3 in a single throw?

 1/2 (1/6 + 1/6 + 1/6 or 50% or 0.5)

 This is a joint probability.

 If I bet on two horses in the six-horse race, then the

probability of my winning when each horse has a
equal chance of winning: 1/6 + 1/6 = 0.34
A Priori – Compound Probability

 Compound probability occurs where a number of

independent events can occur together.

 For example, If I bet on two separate races such

that If the first horse wins then the winnings are
placed on the horse in the 2nd race, then the P of
winning = 1/6 x 1/6 = 0.028

 The probability is lower because winning on the

second horse is dependent on winning on the 1st
horse.
 A Priori – An Example

 Assume that your company owns four items of equipment.

 Assume there is a 10% a priori chance of any piece of
equipment breaking down during the year.
 What is the chance of all four items breaking down at the
same time?
 Compound Probability: 0.1 x 0.1 x 0.1 x 0.1 = 0.0001 (one
chance in ten thousand)
 Note the result if poor maintenance policies increase the
single breakdown chance to 30%?
 (0.3 x 0.3 x 0.3 x 0.3) = 0.0081 (almost 1 in a hundred)
 Assessing Probability

 The probability (chance) of an event occurring can

be established in any one of three ways:
 Subjectively
 A priori
 Objectively
 Objective Probability
Objectively determined probabilities for risk management are
established by statistical analysis of previous similar risk-related
events.

For example:

 Assume you are responsible for a train scheduling project.

 You need to plan the peak hour frequency for a suburban train
line.

 Historical records are available to provide data on peak hour

passenger loads for the line.

 You have data on the approximate number of passengers arriving

at the terminal on each of 21 occasions.
Objective Probability - Example

 On one occasion there were between 851-900 passengers.

 Twice there were between 801 and 850 passengers.
 Three times there were 751-800 passengers.
 Seven times there were 701-750 passengers.
 Four times there were 651-700 passengers.
 Three times there were 601-650 passengers.
 On one occasion there were fewer than 600 passengers.
 Plot the data...
 Objective Probability - Example

%
Average # of Cumulative Cumulative
Passengers Frequency Frequency Frequency
Less than 600 1
601-650 3
651-700 4
701-750 7
751-800 3
801-850 2
851-900 1
 Objective Probability - Example
• Plot a data (bar chart)
• Plot shows that frequency distribution is close to Normal distribution

Frequency
8
7
Number of Occasions

6
5
4
3
2
1
0
Less than 601-650 651-700 701-750 751-800 801-850 851-900
600
Average number of Passengers per Train
 Objective Probability - Example
Calculate cumulative frequency and plot the data..

%
Average # of Cumulative Cumulative
Passengers Frequency Frequency Frequency
Less than 600 1 1
601-650 3 4
651-700 4 8
701-750 7 15
751-800 3 18
801-850 2 20
851-900 1 21
Objective Probability - Example

Cumulative Frequency
25
Number of Occasions

20

15

10

0
Less than 601-650 651-700 701-750 751-800 801-850 851-900
600
Average number of Passengers per Train
 Objective Probability - Example
Now convert the cumulative frequency distribution plot into a percentage cumulative
frequency distribution plot

Average # of Cumulative % Cumulative

Passengers Frequency Frequency Frequency
Less than 600 1 1 1/21x100 = 5%
601-650 3 4 4/21x100 = 19%
651-700 4 8 8/21x100 = 38%
701-750 7 15 15/21x100 = 71%
751-800 3 18 18/21x100 = 86%
801-850 2 20 20/21x100 = 95%
851-900 1 21 21/21x100 = 100%
 Objective Probability - Example
A curve can be fitted to the resulting plot to create a probability distribution function
curve
% Cumulative Frequency
120%
Number of Occasions

100%
80%
60%
40%
20%
0%
Less than 601-650 651-700 701-750 751-800 801-850 851-900
600
Average number of Passengers per Train
 Objective Probability - Example
This curve allows to measure the probability associated with any of the passenger
load rates chosen for the scheduling exercise.
Number of Occasions

Average number of Passengers per Train

 Suppose you decide that an appropriate rate would be 840 passengers per train

There is a 95% chance that the train loads will be less than 840 passengers and there is 5%
chance that the train loads will be more than 840 passengers.
Number of Occasions

Average number of Passengers per Train

 What is the chance that the actual load will be between 701-800 passengers per train?

There is a 93% chance that the train loads will be less than 800 passengers and there is
70% chance that the train loads will be more than 700 passengers.
Number of Occasions

Therefore, the chance that the

actual load will be between 701-
800 passengers will be equal to the
difference between 93% and 70%
i.e. 93% - 70% = 23%.

Average number of Passengers per Train

Objective Probability - Example
 This information allows us to make decisions about scheduling
the trains, with information about the risks of making wrong
decisions.

 The probabilities are statistically deduced from the (objective)

observations.

 Objective probabilities for risk management are established by

calculation from previous similar risk events.

 The insurance industry relies on this method for its policy

premium calculations

 The objective method requires large numbers of historical cases

(e.g. human death rates; thefts; car accidents) to establish
reliable probabilities.
 Probability Distribution
 Continuous probability
distributions, extensively
used in modeling and
simulations represent the
uncertainty in values such as
durations of scheduled
activities and costs of
project components.
 These distributions depict
shapes that are compatible
with the data typically
developed during
quantitative risk analysis.
 Probability Distributions

 Shape or distribution is defined by two parameters;

 Central Tendency (mean, median, mode, etc.)
 Dispersion (SD, Range, Variance, etc.)
Probability Distribution
 Uniform Distribution
Normal Distribution

Triangular Distribution

Poisson Distribution Beta Distribution

 Risk Analysis - Decomposition

Three methods are;

 Fault Tree Analysis (FTA)

 Event Tree Analysis (ETA)

 Decision Trees
 Fault Tree Analysis

 FTA is a diagrammatic approach.

 Uses a top-down deductive process by starting with a

risk event.
 Then traces a logical sequence of possible contributory
factors.
 FTA detail helps with probability and duration, but not
usually with impact assessment.
 It is necessary to assess the probabilities in the FTA
diagram.
 Event Tree Analysis

 This analytical approach is opposite to FTA.

 It differs from Fault Tree Analysis in that it focuses

upon impacts or consequences, instead of tracing
causes. It inductively traces a sequence of possible
outcomes flowing from a trigger event.

 Useful for exploring probabilities, impacts and

durations, but needs clear identification of the risk
event(s).
 Assessing Impact

The assessable outcome of the risk event for the risk taker.
May be incorporated into a decision-analysis technique.

 Decision Trees
 Expected Utility (EU)
 Expected Monetary Value (EMV)

 Monte Carlo Simulation

 Premium/Contingency Sums

 Sensitivity Analysis
 Decision Tree Analysis

 Used to support selection of the best of several alternative course of

action.
 Alternative paths through the project are shown in the decision tree
using branches representing different decisions or events, each of
which can have associated costs and related individual project risks
(including both threats and opportunities).
 The end-points of branches in the decision tree represent the
outcome from following that particular path, which can be negative
or positive.
 The decision tree is evaluated by calculating the expected monetary
value of each branch, allowing the optimal path to be selected.
 Expected Utility (EU) and Decision Tree

 EU=Expected utility

 Assumes values can be derived for probability of occurrence and

worth of outcomes.
 EU can be displayed as a Decision Tree analysis.

 Remember that utility shows satisfaction or happiness derived from a

good/service while value shows monetary value.
 Example: Project Consultant Mariam wants to decide between
alternative methods of travelling from Islamabad to Lahore.
 The proposed alternatives are:

-Fly -Drive -VF train

What will be the EU of each (and hence the decision) if due to bad weather the delay in
flight increases to 225 minutes and probability of delay increases from 0.6 to 0.9?

EU Fly:
=(0.6 x 60) + (0.4 x 100)
= 36 + 40
= 76

EU Drive:
=(0.2 x 25) + (0.8 x 30)
= 5 + 24
= 29

EU VFT:
=(0.2 x 40) + (0.8 x 50)
= 8 + 40
= 48
 Expected Monetary Value (EMV) and
Decision Tree
 EMV is a financial adaptation of EU

 It is a statistical concept that calculates the average outcome

when the future includes scenarios that may or may not
happen, i.e. analysis under uncertainty.
 It helps in calculating the amount required to manage all
identified risks.
 The EMV of opportunities will generally be expressed as
positive values, while those of threats will be negative.
 Risk exposure = probability of occurrence x magnitude of loss
Expected Monetary Value (EMV) and
Decision Tree

 Generally the EMV is reliable only over a large

number of similar risk events (e.g. car accident
insurance).

 The insurance industry relies on the large sample

size for the stability of its risk statistics, and is
therefore predominantly mathematical.

 Projects (unique in nature) do not have such a large

sample size, and risk management is more
dependent upon expert opinion.
Expected Monetary Value (EMV)
 EMV and Decision Trees
Note 1

 The decision tree shows how to make a decision between alternative capital strategies
(represented as “decision nodes”) when the environment contains uncertain elements
(represented as “chance nodes”).

Note 2

 Here, a decision is being made whether to invest $120M US to build a new plant or to
instead invest only $50M US to upgrade the existing plant. For each decision, the demand
(which is uncertain, and therefore represents a “chance node”) must be accounted for. For
example, strong demand leads to $200M revenue with the new plant but only $120M US
for the upgraded plant, perhaps due to capacity limitations of the upgraded plant. The end
of each branch shows the net effect of the payoffs minus costs. For each decision branch,
all effects are added (see shaded areas) to determine the overall Expected Monetary Value
(EMV) of the decision. Remember to account for the investment costs. From the
calculations in the shaded areas, the upgraded plant has a higher EMV of $46M – also the
EMV of the overall decision. (This choice also represents the lowest risk, avoiding the
worst case possible outcome of a loss of $30M).
 Modeling & Simulations
 Models
 Representation of real life
 Probabilistic and Deterministic
Models

 Simulation
 A technique that is used by models
to generate future scenarios
 Modeling and Simulations

 A project simulation uses a model that translates the specified detailed

uncertainties of the project into their potential impact on project objectives.

 Iterative simulations are typically performed using the Monte Carlo

technique.

 In a simulation, the project model is computed many times (iterated), with

the input values (e.g cost estimates or activity durations) chosen at random
for each iteration from the probability distributions of these variables.

 A probability distribution (e.g. total cost or completion date) is calculated

from iterations.

 For a cost risk analysis, a simulation uses cost estimates and for schedule risk
analysis, the schedule network diagram and duration estimates are used.
 Monte Carlo Simulation

 Monte Carlo simulation performs risk analysis by

building models of possible results by substituting a
range of values—a probability distribution—for any
factor that has inherent uncertainty. It then calculates
results over and over, each time using a different set of
random values from the probability functions.
 It is a technique used to understand the impact of risk
and uncertainty in prediction and forecasting models.
 It helps one visualize most or all of the potential
outcomes to have a better idea regarding the risk of a
decision.
 Probabilistic Analysis
 Estimates are made of potential
project schedule and cost outcomes
listing the possible completion
dates and costs with their
associated confidence level.

 These are often expressed as a

cumulative distribution, can be
used with stakeholder risk
tolerances to permit qualification of
cost and time contingency reserves.

 These contingency reserves are

needed to bring the risk of
overrunning the objectives to an
acceptable level.
 Monte Carlo Simulations

 Monte Carlo simulation performs risk analysis by

building models of possible results by substituting a
range of values—a probability distribution—for any
factor that has inherent uncertainty. It then calculates
results over and over, each time using a different set of
random values from the probability functions.
 It is a technique used to understand the impact of risk
and uncertainty in prediction and forecasting models.
 It helps one visualize most or all of the potential
outcomes to have a better idea regarding the risk of a
decision.
 Premiums/Contingency Sums

 The inclusion of additional monetary sums in estimates,

budgets, tenders, etc. to allow for the impacts of specific
or general risks.
 May be:
 an estimate of the financial impact of a risk occurrence;
 an arbitrary amount included to compensate the risk-
taker for taking the risk;
 or the cost of transferring the risk to another party (e.g.
insurance premium).

 Rarely a precisely-calculated amount.

 Sensitivity Analysis
 Sensitivity analysis involves the investigation of alternative risk
parameters by analysing the effect of changing the values of the
factor variables in a stepwise manner.
 E.g. the effect of different probabilities of occurrence might be
investigated by calculating the risk outcomes after increasing or
decreasing probability values in regular steps (e.g. 0.1 > 0.11 > 0.12
etc. or 0.1 > 0.09 > 0.08, etc.).
 Limited to changing ONE factor variable by ONE step at a time.

 Most risk situations involve multiple factors, each of which may be

variable in value.
 Some factors may be inter-dependent.
 Risk Duration (Live Risk)

 The period during which the risk taker is exposed to

a particular risk.

 Risk impact assessment should also consider the

time aspect associated with the risk (i.e. it may be
acceptable to retain a low probability/high impact
risk if the total period of risk exposure is short).
 Risk Duration (Live Risk)
 Example 1: Removing structural support of an existing foundation
(high impact but low probability and short period).
 Example 2: weather risk on a pipe-line project may be constant
(probability and impact).
 Example 3: weather risk on an office building should reduce (impact)
after roof is fixed.
 Example 4: accident risk on pipe-line project may be constant
(probability and impact).
 Example 5: accident risk on an office building should reduce
(probability) after scaffolding is removed, but impact would remain
constant. [The impact will be constant (a death is a death), but the
probability of an accident occurring should diminish over time for the
building project as working conditions inside become less dangerous.]
Thank you!