BCC Irrationality Core Aspect of Institutional & Stakeholder Conduct Discussion
Description
apply the analytic concepts developed in the course to a situation of their choosing, explaining the apparent irrationalities in stakeholders behaviors and suggesting ways that one or more stakeholders could reframe the situation to their advantage.
12 attachmentsSlide 1 of 12attachment_1attachment_1attachment_2attachment_2attachment_3attachment_3attachment_4attachment_4attachment_5attachment_5attachment_6attachment_6attachment_7attachment_7attachment_8attachment_8attachment_9attachment_9attachment_10attachment_10attachment_11attachment_11attachment_12attachment_12.slider-slide > img { width: 100%; display: block; }
.slider-slide > img:focus { margin: auto; }
Unformatted Attachment Preview
Economics E-1035
Behavioral Economics and Decision Making
Prospect Theory
September 30, 2021
© Wilson Alan LLC | +1.978.402.8844 | Jon.Fay@WilsonAlan.com
How long should I expect to live?
100
90
80
70
Percentage of
Population
Still Alive
60
All US males
50
40
30
20
10
0
0
10
20
30
40
50
60
70
80
90
100
Age at Death
Source: https://www.ssa.gov/oact/STATS/table4c6.html
© DSMcIntosh LLC
Page
1
How long should I expect to live?
100
90
80
62 year old
US males
70
Percentage of
Population
Still Alive
60
All US males
50
40
30
20
10
0
0
10
20
30
40
50
60
70
80
90
100
Age at Death
Source: https://www.ssa.gov/oact/STATS/table4c6.html
© DSMcIntosh LLC
Page
2
Positive reinforcement and refinement
Challenges the
Assumptions of
Neoclassical Economics
A Model of Rationality
Behavioral Economics
A Model of Reality
Rational but Unrealistic
Realistic but Irrational
Rational
Bounded
Selfish
Social
Willful
Weak
But also:
Fast thinking
Intuitive
Provides the
Standard for
© DSMcIntosh LLC
Page
3
Session 1: The spectrum of rationality
S2
Degree of
Rationality
Decision
Framework
© DSMcIntosh LLC
Perfect
Rationality
S1
Genuine
Irrationality
Bounded
Rationality
Neoclassical
economics
Managerial
Economics
Heuristics
Behavioral
economics
Economic
ideal
Analytical
approximations
Rules of
thumb
Predictable
biases
Conceptual gold
standard
Calculable
solutions
The good gut
The bogus blink
Page
4
Session 2: Optimization
Economic agents adjust what they do until the rate at which they want to trade off one thing for
another thing exactly equals the rate at which they can trade off the two of them
POSSIBLE
When
tradeoffs
1 Define Feasible
tradeoffs
2
Set & Tradeoffs
Neoclassical
Economics
DESIRABLE
equal
y
OPTIMAL
we have an
Assess Value
Tradeoffs
solution
3 Equate Tradeoffs
on the Margin
y
y
Marginal calculus
Finds optimal point
x
x
x
Managerial
Economics
Discrete evaluation
Gets you close
Calculate tradeoffs on the margin, equate marginal tradeoffs across all choices, understand diminishing returns
© DSMcIntosh LLC
Page
5
Session 3: Heuristics
ESTIMATION HEURISTICS
Reliable
Mathematical Shortcuts
Decision Strategies
60 mph: mile a minute
Actual math
Birthday rule
and statistics
37% rule
Rule of 72
Evolutions Legacy
Depth perception
Non-verbal cues
A sweet tooth
Useful when
environmentally
appropriate
Biases
Availability
Anchoring
Confirmation
Unreliable
© DSMcIntosh LLC
Endowment
PROCESS HEURISTICS
Optimizing
MaxiMin
Lexicographic
Tallying
Defaults
Habits and routines
Mental accounts
Expert instincts
Group
Possibly useful
as data, but not
definitive
Explicit
choices
Consensus
Social imitation
Follow the leader
HiPPO
Bounded
rationality
and blinks
Multiple
stakeholders,
multiple
goals
Page
6
Session 4: Risk and Uncertainty
The EV of Utility may not be the
same as the Utility of the EV,
leading to a Risk Premium
Decision trees are a method for structuring the
problem and sequencing the Payoffs, Probabilties
and Choices for an economical calculation of a
good enough solution
E [ U ( L ) ] = S pi * U ( A i )
U [ E( L ) ] = U [ S pi * E (Ai ) ]
Breakeven analysis and overage/underage
calculations provide good enough answers for
practical applications
PO =
© DSMcIntosh LLC
CU
CU
(CU + CO)
CO
=
PO
PU
Page
7
Three approaches to decision making
Perfect
Rationality
Analytical
Rationality
Practical
Rationality
Theoretical Approach
Decision Trees
Process Charts
Uses all available evidence
Finite number of possible
options and outcomes
Flow from start to stop points
Requires knowledge of all
Options
Consequences
Benefits
Costs
Probabilities
Compares every possible set
of choices or decisions
© DSMcIntosh LLC
Requires knowledge of all
Values for each outcome
… either EV
or U(EV)
Probabilities for event
nodes
Sequence of events and
choices
Useful for breakeven
calculations
Event nodes usually Yes/No
Require real-time input
Examples:
Flow charts
Fast-and-frugal trees
Natural frequency trees
Useful for speeding up
decision making
Page
8
Types of Process Charts
Fast and Frugal Tree
Is lamp
plugged
in?
yes
Is bulb
burned
out?
no
Is it worth
repairing
lamp?
no
Trash
the lamp
© DSMcIntosh LLC
yes
Repair
the lamp
yes
Buy a
new light
bulb
Flow Chart
Start
no
Plug in
lamp
Alarm
rings
Ready
to get
up?
Delay
No
Hit snooze
button
Yes
Climb out of
bed
End
Page
9
A fast-and-frugal tree
EKG
worrisome?
no
Chest pain?
yes
Any other key
indicator?
no
Regular
Nursing
Bed
© DSMcIntosh LLC
yes
Coronary
Care
Unit
yes
Coronary
Care
Unit
no
Regular
Nursing
Bed
Fast-and-frugal decision tree:
An exit at every level
A specific ordering of the
decision criteria (cues)
Evaluated based on # of
errors (misses and false
alarms), given baseline test
probabilities
Page 10
Medical probabilities
Test for cancer came back positive (i.e., bad news)
Test has 80% accuracy rate (i.e., sensitivity)
False positives are rare, and only 10% of negatives register as positive
(i.e., 90% specificity)
Average incidence of this cancer in this demographic is 1%
POLL: What is the likelihood that the patient has cancer?
9 in 10
8 in 10
1 in 10
1 in 100
© DSMcIntosh LLC
Page 11
Medical probabilities
Test is
Positive
80%
Test is
Negative
20%
Reality is
Negative
© DSMcIntosh LLC
Reality is
Positive
Page 12
Medical probabilities
© DSMcIntosh LLC
Test is
Positive
10%
%
Test is
Negative
90%
20
Reality is
Negative
Reality is
Positive
Page 13
Medical probabilities
Test is
Positive
%
Test is
Negative
20
Reality is
Negative
Incidence
© DSMcIntosh LLC
99%
Reality is
Positive
1%
Page 14
Medical probabilities
Test is
Positive
9,900
800
Test is
Negative
89,100
200
Reality is
Negative
Reality is
Positive
What is the likelihood that the patient has cancer?
© DSMcIntosh LLC
Page 15
Medical probabilities
Test is
Positive
92.5%
7.5%
Test is
Negative
89,100
200
Reality is
Negative
Reality is
Positive
Base Rate Bias
© DSMcIntosh LLC
Page 16
Conditional probability
Bayes Law
(standard version)
P(A|B)=
P(A|B)=
Test is
B Positive
9,900
800
=
Test is
Negative
89,100
200
Reality is
Negative
Reality is
Positive
=
P(B|A)
x
P(A)
P(B)
( 800 / 1000 ) ( 1000 / 100,000 )
( 10,700 / 100,000 )
( .8 ) ( .01 )
( .107 )
7.47 %
A
© DSMcIntosh LLC
Page 17
Conditional probability
P(A|B)
Bayes Law
(odds version)
P ( ~A | B )
=
P(B|A)
P ( B | ~A )
Conditional
PProbability
(A|B)
~A | B )
1P- (Conditional
B
Test is
Positive
9,900
800
Probability
P(A|B)
P ( ~A | B )
Test is
Negative
89,100
200
Reality is
Negative
Reality is
Positive
A
© DSMcIntosh LLC
=
P(A|B)
P ( ~A | B )
P
B | ~A )
1 ( Specificity
.80
=
=
P(A|B) =
P(A|B)
PSensitivity
(B|A)
.10
x
x
x
P(A)
P ( ~A )
PIncidence
(A)
P ( Incidence
~A )
1
.01
.99
.08
.99
.08
=
.99 + .08
.08
.107
= 7.47 %
Page 18
Conditional probability
Standard Version
P(A|B) =
Odds Version
P(B|A) x P(A)
P(A|B)
P(B)
P ( ~A | B )
Use this when you are given:
Incidence (probability of A)
Sensitivity (true positives)
Probability of B (all positive tests)
P(B)
P (B|A)
=
P(B|A)
P ( B | ~A )
x
P(A)
P ( ~A )
Use this when you are given:
Incidence (probability of A)
Sensitivity (true positives)
Specificity (true negatives)
P (B|~A)
Sensitivity
P (B|A)
Sensitivity
Specificity
P(A)
Incidence
© DSMcIntosh LLC
P(A)
Incidence
Page 19
Conditional probability and decision trees
To calculate natural frequencies…
800
80%
1,000
Sensitivity %
Test is
Positive
10%
20%
80%
1%
200
99%
9,900
100,000
Test is
Negative
90%
20%
Specificity %
Reality is
Negative
99%
© DSMcIntosh LLC
10%
100,000 tests
Reality is
Positive
1%
Incidence %
10.7% of tests are positives
10,700 total positives
800 true positives
99,000
90%
89,100
7.47% of positive tests are
true positives
Page 20
Conditional probability and decision trees
Or, to calculate probabilities…
1
80%
.80
Sensitivity %
Test is
Positive
10%
20%
80%
90%
Reality is
Negative
99%
© DSMcIntosh LLC
Incidence %
10%
100,000 tests
Reality is
Positive
1%
Incidence %
1
99%
20%
Specificity %
0
1%
.107
Test is
Negative
Sensitivity %
10.7% of tests are positives
10,700 total positives
800 true positives
.10
Specificity %
90%
0
7.47% of positive tests are
true positives
Page 21
Decision trees
Hit: $ 24 mil
30%
$ 31 mil
Yes
Flop:
70%
$ 80 mil
$ 7 mil
$ 10 mil
Open Big?
$ 62 mil
No
$ 32 mil
$ 31 mil
Hit:
50%
Flop:
50%
Roll out
big?
Hit:
60%
Yes
Flop:
40%
No
$ 5 mil
$ 60 mil
$ 100 mil
$ 2 mil
$ 5 mil
$ 5 mil
$ 2 mil
$ 1 mil
© DSMcIntosh LLC
Page 22
We were given four sets of probabilities
p (A) = Probability of A occurring
p (A | B) = Prob of A given B
TEST:
Well
p(W) = 50%
Badly
p(B) = 50%
Probability of Flop
if test goes Well
Probability of Success
if test goes Well
p (F | W) = 40%
p (S | W) = 60%
Probability of Flop
if test goes Badly
Probability of Success
if test goes Badly
p (F | B) = 100%
p (S | B) = 0%
REALITY:
© DSMcIntosh LLC
Flop
p(F) = 70%
Success
p(S) = 30%
Page 23
We were given four sets of probabilities
p (A) = Probability of A occurring
p (A | B) = Prob of A given B
Natural frequencies
TEST:
Well
p(W) = 50%
Badly
p(B) = 50%
Probability of Flop
if test goes Well
Probability of Success
if test goes Well
p (F | W) = 40%
p (S | W) = 60%
20
30
Probability of Flop
if test goes Badly
Probability of Success
if test goes Badly
p (F | B) = 100%
p (S | B) = 0%
REALITY:
© DSMcIntosh LLC
50
0
Flop
p(F) = 70%
Success
p(S) = 30%
Page 24
We were given four sets of probabilities
p (A) = Probability of A occurring
p (A | B) = Prob of A given B
p(A ? B) = Prob of A and B occurring
TEST:
Well
p(W) = 50%
Badly
p(B) = 50%
Probability of Flop
if test goes Well
Probability of Success
if test goes Well
p (F | W) = 40%
p (S | W) = 60%
p (F ? W) = 20%
p (S ? W) = 30%
Probability of Flop
if test goes Badly
Probability of Success
if test goes Badly
p (F | B) = 100%
p (S | B) = 0%
p (F ? B) = 50%
REALITY:
© DSMcIntosh LLC
Flop
p(F) = 70%
p (S ? B) = 0%
Success
p(S) = 30%
Page 25
We were given four sets of probabilities
p (A) = Probability of A occurring
p (A | B) = Prob of A given B
p(A ? B) = Prob of A and B occurring
TEST:
Well
p(W) = 50%
Badly
p(B) = 50%
Probability of Flop
if test goes Well
Probability of Success
if test goes Well
p (F | W) = 40%
p (S | W) = 60%
p (F ? W) = 20%
p (S ? W) = 30%
Probability of Flop
if test goes Badly
Probability of Success
if test goes Badly
p (F | B) = 100%
p (S | B) = 0%
p (F ? B) = 50%
REALITY:
p (S ? B) = 0%
Flop
p(F) = 70%
Sensitivity
(true positives)
= 30% / 50%
= 60%
Specificity
(true negatives)
= 50% / 50%
= 100%
Success
p(S) = 30%
Incidence = 30%
© DSMcIntosh LLC
Page 26
Breakout exercise
Max Bialystock is producing his next Broadway show. He knows
that, of shows that open on Broadway, only 20% on average
become hits.
Assume that every show that opens on Broadway has a few
weeks of previews in Boston, and that every show that has
Boston previews goes on to Broadway, whether or not it is a
success in Boston.
The Boston critics are known for their discernment. Of shows that
eventually become hits in Broadway, the Boston critics have
historically liked 75% of them. And of shows that go on to flop,
they disliked 90% of them.
QUESTION:
If the Boston critics like a show, what is the probability that it will
be a success on Broadway?
© DSMcIntosh LLC
Page 27
We were given three sets of numbers
p (A) = Probability of A occurring
p (A | B) = Prob of A given B
Natural Frequencies
TEST:
Well
p(W) = 23
?
Badly
p(B) = 77
?
Probability of Flop
if test goes Well
Probability of Success
if test goes Well
p (F | W) = 10%
p (S | W) = 75%
8
15
Probability of Flop
if test goes Badly
Probability of Success
if test goes Badly
p (F | B) = 90%
p (S | B) = 25%
REALITY:
Bialystock
© DSMcIntosh LLC
72
5
Flop
p(F) = 80%
Success
p(S) = 20%
Sensitivity
(true positives)
= 15% / 23%
= 65%
Specificity
(true negatives)
= 72% / 77%
= 94%
Incidence = 20%
Page 28
The importance of priors
Test
Positive
10%
80%
Test
Positive
9,900
800
Test
Negative
90%
20%
Test
Negative
89,100
200
Reality Negative
Reality Positive
Reality Negative
Reality Positive
99%
1%
99,000
Prior
10%
80%
Test
Positive
9,252
5,981
Test
Negative
90%
20%
Test
Negative
83,271
1,496
Reality Negative
Reality Positive
Reality Negative
Reality Positive
© DSMcIntosh LLC
7.5%
New Prior
92,523
10,700
= 7.5%
1,000
Test
Positive
92.5%
800
5,981
= 39.3%
15,233
7,477
Page 29
More problems of estimation
Sample Size Fallacy
Conjunction Fallacy
Base Rate Neglect / Representative Heuristic
© DSMcIntosh LLC
Page 30
Sample size fallacy
In a small town nearby, there are two hospitals. Hospital A has an
average of 15 births per day; Hospital B is larger, and has an average
of 45 births per day. As we all know, overall the number of males born
is 50%. Each hospital recorded the number of days in which, on that
day, at least 60% of the babies born were male.
POLL: Which hospital recorded more days when
at least 60% of the babies born were male?
(a) Hospital A
(b) Hospital B
(c) Both equal
For small samples, bias is to:
Underestimate randomness
Overestimate informativeness
For large samples, bias is to:
Overestimate randomness
Underestimate informativeness
A
B
© DSMcIntosh LLC
Page 31
Conjunction fallacy
How likely is each?
(a)
My next-door neighbor drives a minivan
(b) My next-door neighbor, a father of three, drives a minivan
Minivan Drivers
Fathers of 3
An obvious error of logic and/or bad math
© DSMcIntosh LLC
Page 32
Base rate neglect
A cab was involved in a hit and run accident at night. A witness identified the cab
as Blue.
85% of the cabs in the city are Green, and 15% are Blue.
It has been reasonably established that the witness can correctly identify each
one of the two colors of taxis 80% of the time.
What is the probability that the cab involved in the accident was Blue,
rather than Green, knowing that this witness identified it as Blue?
Witness says
Green
30
680
Witness Says
Blue
120
170
Taxi was
Blue
Taxi was
Green
© DSMcIntosh LLC
P (TB | WB) = 120 / 290
= 41%
Page 33
Base rate neglect and confirmation bias
P(A|B)
P ( ~A | B )
=
P(B|A)
=
P ( ~A )
P ( B | ~A )
Conditional
Likelihood
=
Odds
Ratio
Posterior
x
Signal
P(A)
x
x
Unconditional
Odds
Prior
Base Rate Neglect /
Representative Heuristic
Underweight the prior
Overweight the new
information (i.e., the signal)
More likely when the new
information is salient
Confirmation Bias
Overweight the prior
Underweight the new
information (i.e., the signal)
More likely when the prior is
a deeply held belief
© DSMcIntosh LLC
Page 34
Expected monetary value (EV) vs. Expected utility
The EV of Utility may not be the same as the Utility of the EV
The Utility of the EV of the gamble is greater than the EV of the Utility of the gamble
So locking in the EV of the gamble is preferable to taking a chance: Risk Aversion
Utility
U($x2)
UTILITY of EV:
EV of utility:
U(0.5*$x1+0.5*$x2)
Risk Premium
Expected Utility
0.5U($x1)+0.5U($x2)
Observations:
U($x1)
Downward curving utility curve
means a diminishing marginal
utility of money
Using Expected Value rather
than Expected Utility means
assuming a linear curve (i.e.,
no marginal utility of money)
© DSMcIntosh LLC
$x1
(0.5*$x1)+(0.5*$x2)
Expected Value
$x2
Page 35
Losses hurt more than gains of equal amount
Utility
U($x2)
U($x0)
$?1
?2(U)
$?2
?1(U)
U($x1)
$x1
© DSMcIntosh LLC
$x0
$x2
$EV
Page 36
Prospect theory: loss aversion and framing
Utility
LOSS AVERSION
Impact of losses is much greater
than impact of comparable gains
$ EV
of Loss
$ EV
of Gain
FRAMING
Defining the base case
determines whether change is
seen as a loss or a gain
© DSMcIntosh LLC
Page 37
The fourfold pattern
Potential LOSSES
Potential GAINS
95% chance to LOSE $10,000
95% chance to WIN $10,000
5% chance to LOSE $10,000
5% chance to WIN $10,000
HIGH
Probability
LOW
Probability
© DSMcIntosh LLC
Page 38
The fourfold pattern
Potential LOSSES
0.95
HIGH
Probability
Bet
Lose $10,000
EV = $9,500
0.05
LOW
Probability
Bet
Bet
Maximum EV
Maximum EV
Lose $0
Sure Thing $501
People Choose
Win $10,000
EV = $9,500
0.05
0.05
Bet
Sure Thing $501
People Choose
Win $10,000
EV = $500
0.95
Maximum EV
Lose $0
Sure Thing $9,499
Lose $10,000
EV = $500
0.95
People Choose
0.95
Lose $0
Sure Thing $9,499
0.05
Potential GAINS
People Choose
Lose $0
Maximum EV
What are examples of each of the four types?
© DSMcIntosh LLC
Page 39
Four irrational actions
Utility
Lock in Gain
Lottery Ticket
$ EV
Insurance
Hail Mary
Pass
© DSMcIntosh LLC
Page 40
Probability weighting function (1948)
.95
Psychological
Probability
.50
.05
.05
.50
Mathematical Probability
.95
Source First experimental study of risk: Preston and Baratta (1948)
© DSMcIntosh LLC
Page 41
Probability weighting function Kahneman & Tversky
Kahneman & Tversky also observed that the subjective probabilities people use
(also known as psychological probabilities or decision weights) vary from objective
probabilities along an inverted s-curve
Decision Weight
.95
Steepness at 1
Certainty effect
Flatness at 0.5
Meh effect
.50
Steepness at 0
Possibility effect
.05
.05
© DSMcIntosh LLC
.50
Probability
.95
Page 42
The fourfold pattern
HIGH
Probability
LOW
Probability
© DSMcIntosh LLC
Potential LOSSES
Potential GAINS
95% chance to LOSE $10,000
95% chance to WIN $10,000
Hope to escape a likely loss
Fear of losing a likely gain
Risk SEEKING
Risk AVERSE
Take the Crazy Gamble
Lock in the Gain
5% chance to LOSE $10,000
5% chance to WIN $10,000
Fear of a possible loss
Hope to get a possible gain
Risk AVERSE
Risk SEEKING
Get the Insurance Policy
Buy the Lottery Ticket
Page 43
Prospect theory in a nutshell
SUBJECTIVE PROBABILITIES
Utility
LOSS AVERSION
Impact of losses is much greater
than impact of comparable gains
$ EV
of Loss
Events with probabilities close
to 0 or 1 are seen as having
disproportionate likelihood
and impact
$ EV
of Gain
FRAMING
Defining the base case
determines whether change is
seen as a loss or a gain
© DSMcIntosh LLC
Page 44
Prospect Theory, Loss Aversion, and the Fourfold Pattern
Utility
Steepness at 1
Certainty effect
Decision Weight
.95
Flatness at 0.5
Meh effect
.50
Steepness at 0
Possibility effect
.05
.05
.50
Probability
Lock in Gain
Lottery Ticket
$ EV
of Loss
Insurance
$ EV
of Gain
Hail Mary
Pass
.95
Utility
Neoclassical Risk Aversion
Probability
of Loss
SMALL chance to LOSE
Fear of a possible loss
Accept lower $EV deal
Risk Averse
HIGH chance to WIN
Fear losing a likely gain
Accept lower $EV deal
Risk Averse
SMALL chance to WIN
Hope to get a possible gain
Reject higher $EV deal
Risk Seeking
Probability
of Gain
HIGH chance to LOSE
Hope to escape a likely loss
Reject higher $EV deal
Risk Seeking
© DSMcIntosh LLC
Page 45
Allaiss paradox and the certainty effect
11%
$110,000
89%
10%
$500,000
90%
© DSMcIntosh LLC
$1,000,000
$0
100%
$1,000,000
$5,000,000
$0
$1,390,000
$1,000,000
0%
$0
89%
$1,000,000
0%
10%
$0
$5,000,000
Page 46
The behavioral model of risk and uncertainty
Simple EV model says
EV =
?
ps * $xs , but
s
People are really bad at
estimating Probabilities
Base Rate Neglect
92.5%
7.5%
89,10
0
200
Certainty & Possibility Effects
Utilities of uncertain outcomes are not
linear functions of monetary payoffs
Prospect Theory / Loss Aversion
Estimation Biases
Sample Size Fallacy
Conjunction Fallacy
Representative Heuristic
Leading to paradoxical patterns of behavior when making decisions under uncertainty
Allais Paradox
© DSMcIntosh LLC
Fourfold Pattern
Page 47
Homework #5
We all think our parents could do better. Take something that your parents have
done (or are thinking of doing) that would be considered irrational by neoclassical
economics and would illustrate one of the behaviors in the Fourfold Pattern.
In one page (either Word or PowerPoint), explain to them why they might be making
a mistake and what they might want to do differently. Keep in mind that these are
your parents youre talking to, so try to be clear, explanatory, helpful, and respectful.
The assignment is due on Tuesday,
October 5, at 8:00 pm (EDT).
© DSMcIntosh LLC
Potential LOSSES
Potential GAINS
HIGH
Probability
Take the
Crazy
Gamble
Lock in
the Gain
LOW
Probability
Get the
Insurance
Policy
Buy the
Lottery
Ticket
Page 48
Economics E-1035
Behavioral Economics and Decision Making
Uncertainty
September 24, 2021
© Wilson Alan LLC | +1.978.402.8844 | Jon.Fay@WilsonAlan.com
Positive reinforcement and refinement
Challenges the
Assumptions of
Neoclassical Economics
A Model of Rationality
Behavioral Economics
A Model of Reality
Rational but Unrealistic
Realistic but Irrational
Rational
Bounded
Selfish
Social
Willful
Weak
But also:
Fast thinking
Intuitive
Provides the
Standard for
© DSMcIntosh LLC
Page
1
Session 1: The spectrum of rationality
S2
Degree of
Rationality
Decision
Framework
© DSMcIntosh LLC
Perfect
Rationality
S1
Genuine
Irrationality
Bounded
Rationality
Neoclassical
economics
Managerial
Economics
Heuristics
Behavioral
economics
Economic
ideal
Analytical
approximations
Rules of
thumb
Predictable
biases
Conceptual gold
standard
Calculable
solutions
The good gut
The bogus blink
Page
2
Session 2: Optimization
Economic agents adjust what they do until the rate at which they want to trade off one thing for
another thing exactly equals the rate at which they can trade off the two of them
POSSIBLE
When
tradeoffs
1 Define Feasible
tradeoffs
2
Set & Tradeoffs
Neoclassical
Economics
DESIRABLE
equal
y
OPTIMAL
we have an
Assess Value
Tradeoffs
solution
3 Equate Tradeoffs
on the Margin
y
y
Marginal calculus
Finds optimal point
x
x
x
Managerial
Economics
Discrete evaluation
Gets you close
Calculate tradeoffs on the margin, equate marginal tradeoffs across all choices, understand diminishing returns
© DSMcIntosh LLC
Page
3
Session 3: Heuristics
ESTIMATION HEURISTICS
Reliable
Mathematical Shortcuts
Decision Strategies
60 mph: mile a minute
Actual math
Birthday rule
and statistics
37% rule
Rule of 72
Evolutions Legacy
Depth perception
Non-verbal cues
A sweet tooth
Useful when
environmentally
appropriate
Biases
Availability
Anchoring
Confirmation
Unreliable
© DSMcIntosh LLC
Endowment
PROCESS HEURISTICS
Optimizing
MaxiMin
Lexicographic
Tallying
Defaults
Habits and routines
Mental accounts
Expert instincts
Group
Possibly useful
as data, but not
definitive
Explicit
choices
Consensus
Social imitation
Follow the leader
HiPPO
Bounded
rationality
and blinks
Multiple
stakeholders,
multiple
goals
Page
4
Session 4: Uncertainty
(a) 11% chance of getting $1 million
89% chance of getting nothing
11% chance of
$1 million
© DSMcIntosh LLC
(b) 10% chance of getting $5 million
90% chance of getting nothing
10% chance of
$5 million
Page
5
Allaiss options
(a) 11% chance of getting $1 million
89% chance of getting nothing
11% chance of
$1 million
11%
Expected Value = ?
89%
(b) 10% chance of getting $5 million
90% chance of getting nothing
10% chance of
$5 million
$1,000,000
$0
$110,000
$0
Expected Value = PROBABILITY x PAYOFF
Expected Value =
? ( PROBABILITY
i
x PAYOFFi )
$110,000
© DSMcIntosh LLC
Page
6
Allaiss options
(a) 11% chance of getting $1 million
89% chance of getting nothing
11% chance of
$1 million
10% chance of
$5 million
11%
$1,000,000
10%
$5,000,000
89%
$0
90%
$0
$110,000
$500,000
$0
$0
$110,000
© DSMcIntosh LLC
(b) 10% chance of getting $5 million
90% chance of getting nothing
Expected Value =
$500,000
? ( PROBABILITY
i
x PAYOFFi )
Page
7
Allaiss options
(b) 10% chance of getting $5 million
90% chance of getting nothing
(a) 11% chance of getting $1 million
89% chance of getting nothing
11% chance of
$1 million
10% chance of
$5 million
11%
$1,000,000
89%
$0
10%
$5,000,000
90%
$0
$110,000
$500,000
$500,000
© DSMcIntosh LLC
Expected Value =
? ( PROBABILITY
i
x PAYOFFi )
Page
8
Allaiss options
(b) 10% chance of getting $5 million
90% chance of getting nothing
(a) 11% chance of getting $1 million
89% chance of getting nothing
11% chance of
$1 million
10% chance of
$5 million
11%
$1,000,000
89%
$0
10%
$5,000,000
90%
$0
$110,000
$500,000
$500,000
© DSMcIntosh LLC
Expected Value =
? ( PROBABILITY
i
x PAYOFFi )
Page
9
Allaiss options: Round 2
(b) 89% chance of getting $1 million
10% chance of getting $5 million
1% chance of getting nothing
(a) 100% chance of getting $1 million
0% chance of getting nothing
100%
$1,000,000
$1,000,000
$0
0%
$1,390,000
89%
$1,000,000
10%
$5,000,000
$1,390,000
© DSMcIntosh LLC
Expected Value =
? ( PROBABILITY
i
x PAYOFFi )
Page 10
Allaiss options: Round 1 vs. Round 2
(a) 11% chance of getting $1 million
89% chance of getting nothing
(b) 10% chance of getting $5 million
90% chance of getting nothing
11%
$1,000,000
89%
$0
10%
$5,000,000
90%
$0
$110,000
$500,000
$500,000
100%
$1,000,000
(a) 100% chance of getting $1 million
(b) 89% chance of getting $1 million
10% chance of getting $5 million
1% chance of getting nothing
0%
Expected Value =
$0
$1,390,0000
89%
$1,000,000
10%
$5,000,000
$1,390,000
© DSMcIntosh LLC
$1,000,000
? ( PROBABILITY
i
x PAYOFFi )
Page 11
Allaiss options: Round 1 vs. Round 2
(a) 11% chance of getting $1 million
89% chance of getting nothing
(b) 10% chance of getting $5 million
90% chance of getting nothing
11%
$1,000,000
89%
$0
10%
$5,000,000
90%
$0
$110,000
$500,000
$500,000
100%
$1,000,000
(a) 100% chance of getting $1 million
(b) 89% chance of getting $1 million
10% chance of getting $5 million
1% chance of getting nothing
0%
Expected Value =
$0
$1,000,0000
89%
$1,000,000
10%
$5,000,000
$1,390,000
© DSMcIntosh LLC
$1,000,000
? ( PROBABILITY
i
x PAYOFFi )
Page 12
How Economists Think About Uncertainty
Economists treat decisions with uncertainty as lotteries and then make four key
assumptions about preferences and lotteries:
Completeness
If I offer you a choice between a Powerball ticket and a scratch ticket, you know which
one you prefer (or you know youre indifferent)
Transitivity
If you prefer lottery L to lottery M to lottery N, then youll prefer lottery L to lottery N
Continuity
If you prefer L to M, and M to N, I can create a lottery of lotteries thats a
combination of L and N such that youre indifferent between it and M
Independence
Your relative preference between two options shouldnt depend on the presence or
absence of a third inferior option
Example:
A lottery L where you have a 5% of utility of 100
and a 95% chance of utility of 10
L: [ (.05,100) , (.95,10) ]
The expected value of utility (expected utility)
for this lottery is
E [ U(L) ] = .05*100 + .95*10
More generally, the expected utility for a lottery
with mutually exclusive outcomes is represented as
E [ U(L) ] =
© DSMcIntosh LLC
S pi * U(Ai )
Page 13
Expected monetary value (EV) vs. Expected utility
The EV of Utility may not be the same as the Utility of the EV
The Utility of the EV of the gamble is greater than the EV of the Utility of the gamble
So locking in the EV of the gamble is preferable to taking a chance: Risk Aversion
Utility
U($x2)
UTILITY of EV:
EV of utility:
U(0.5*$x1+0.5*$x2)
Risk Premium
Expected Utility
0.5U($x1)+0.5U($x2)
Observations:
U($x1)
Downward curving utility curve
means a diminishing marginal
utility of money
Using Expected Value rather
than Expected Utility means
assuming a linear curve (i.e.,
no diminishing marginal utility
of money)
© DSMcIntosh LLC
$x1
(0.5*$x1)+(0.5*$x2)
Expected Value
$x2
Page 14
Case study: A wide release for Miss Behaving?
Its 2025, the pandemic has receded, and movie theaters are open again. Crimson
Studios is about to release a new feature film, Miss Behaving. The studio needs
to decide whether to gamble on a wide release or test the waters with a limited
release.
OPTION 1:
WIDE RELEASE
OPTION 2:
LIMITED RELEASE w/ ROLLOUT
OPTION 3:
LIMITED RELEASE ONLY
1000 screens in USA
50 screens in USA
50 screens in USA
30% chance of being a hit,
grossing $80 million
50% chance of being a hit at
first, in which case it is then
rolled out nationwide, with
50% chance of being a hit,
grossing $5 million
70% chance it will be a flop,
grossing only $10 million
60% chance of being a hit,
grossing $100 million
40% chance of being a
flop, grossing $5 million
50% chance it will be a flop,
grossing only $2 million
NO ROLLOUT, EITHER WAY
50% chance it will be a flop in
limited release, grossing only
$2 million
© DSMcIntosh LLC
Page 15
Decision Trees
Yes
Hit:
30%
$ 80 mil
Flop:
70%
$ 10 mil
Hit:
60%
Open Big?
No
Hit:
50%
Flop:
50%
© DSMcIntosh LLC
Roll out
big?
$ 100 mil
Yes
Flop:
40%
No
$ 5 mil
$ 5 mil
$ 2 mil
Page 16
Decision Trees
Hit: $ 24 mil
30%
$ 31 mil
Yes
Flop:
70%
$ 7 mil
$ 10 mil
Open Big?
$ 62 mil
No
Hit:
50%
Flop:
50%
© DSMcIntosh LLC
$ 80 mil
Roll out
big?
Hit:
60%
Yes
Flop:
40%
No
$ 5 mil
$ 60 mil
$ 100 mil
$ 2 mil
$ 5 mil
$ 5 mil
$ 2 mil
Page 17
Decision Trees
Hit: $ 24 mil
30%
$ 31 mil
Yes
Flop:
70%
$ 80 mil
$ 7 mil
$ 10 mil
Open Big?
$ 62 mil
No
$ 32 mil
$ 31 mil
Hit:
50%
Flop:
50%
Roll out
big?
Hit:
60%
Yes
Flop:
40%
No
$ 5 mil
$ 60 mil
$ 100 mil
$ 2 mil
$ 5 mil
$ 5 mil
$ 2 mil
$ 1 mil
© DSMcIntosh LLC
Page 18
Decision Trees
Hit: $ 24 mil
30%
$ 31 mil
Yes
Flop:
70%
$ 80 mil
$ 7 mil
$ 10 mil
Open Big?
$ 62 mil
No
$ 32 mil
$ 31 mil
Hit:
50%
Flop:
50%
Roll out
big?
Hit:
60%
Yes
Flop:
40%
No
$ 5 mil
$ 60 mil
$ 100 mil
$ 2 mil
$ 5 mil
$ 5 mil
$ 2 mil
$ 1 mil
© DSMcIntosh LLC
Page 19
Expected Value of Perfect Information (EVPI)
$ 80 mil
Open
Big? Yes
$ 80 mil
Hit:
30%
No
$32.0 million
EV with perfect information
about success of Open Big
$42.4 million
Expected Value of
Perfect Information
$10.4 million
$ 32 mil
$ 42.4 mil
Flop:
70%
EV with uncertainty:
Open
Big? Yes
$ 32 mil
No
$ 32 mil
$ 10 mil
$ 62 mil
$ 31 mil
Hit:
50%
Flop:
50%
Roll out
big?
Hit:
60%
Yes
Flop:
40%
No
$ 5 mil
$ 60 mil
$ 100 mil
$ 2 mil
$ 5 mil
$ 5 mil
$ 2 mil
$ 1 mil
© DSMcIntosh LLC
Page 20
Decision Trees: Key Ideas
Two type of nodes: Events (probabilities) and Choices
At choice nodes, choose the highest EV
At event nodes, calculate the total probabilistic value of all possible events
Arithmethic flows from right to left (broadest to narrowest)
Populating a decision tree requires:
Estimating Payoffs and Probabilities
Sequencing Events and Choices
Acquiring more accurate estimates of payoffs and probabilities:
Has value, up to the Expected Value of Perfect Information
But the cost of getting better information is often unknown
Decision trees often work best as breakeven analyses
© DSMcIntosh LLC
Page 21
Pascals Wager
God
Exists
God Does
Not Exist
Should I live
«Qu’avez-vous
my life as if
à perdre?»
God
exists?
Yes
Eternal
Paradise
No golf on
Sunday
Morning
No
God
Exists
God Does
Not Exist
© DSMcIntosh LLC
Eternal
Damnation
Golf on Sunday
Mornings
Page 22
Other types of process charts
Fast and Frugal Tree
Flow Chart
Is lamp
plugged
in?
yes
Is bulb
burned
out?
no
Is it worth
repairing
lamp?
no
Trash
the lamp
© DSMcIntosh LLC
yes
Repair
the lamp
Start
no
Plug in
lamp
yes
Buy a
new
light
bulb
Alarm
rings
Ready
to get
up?
Delay
No
Hit snooze
button
Yes
Climb out of
bed
End
Page 23
Monty Hall Paradox
A car is behind 1 of 3 doors;
a goat is behind each of the other 2
You choose 1 of 3 doors
The game show host, Monty Hall,
opens one door, revealing a goat
He then offers you the chance to
switch your choice
Should you switch when
given the chance?
Choose a
strategy of
SWITCH
Choose
a strategy of
STAY
Car is
behind
1 of 3
doors
You
choose
1 of 3
doors
Monty
Hall
opens 1
of the
doors
…
…
…
…
…
Zoom Poll
© DSMcIntosh LLC
Page 24
Monty Hall Paradox
KEY:
Where is
the car?
Monty Hall opens 1 of the doors
Probability of each is TBD
0
You choose 1 of 3 doors
Probability of each is 1/3
2/3
Probability of each is 1/3
2/3
2/3
© DSMcIntosh LLC
1
0
Choose a strategy
of SWITCH:
EV = 2/3
Choose a
strategy of STAY
1
1
Car is behind 1 of 3 doors
1
1
1
…
1
0
111
112
113
121
122
123
131
132
133
211
212
213
221
222
223
231
232
233
311
312
313
321
322
323
331
332
333
2
Which did I
choose?
-Bad: 0
Bad: 0
–Good: 1
-Good: 1
—Good: 1
Bad: 0
-Bad: 0
Good: 1
—Good: 1
-Good: 1
–Bad: 0
Bad: 0
—
3
Which door
does Monty
open?
TWO RULES:
1. Monty Hall
does not
open a door
with a car
behind it
2. Monty
