Lecture 3: Prospect Theory

IB9Y20 Behavioural Finance

1/41
Outline

In this lecture we will discuss the second pillar of rationality,

Expected Utility Theory (EUT)

We will rst illustrate the basics of EUT, and then look at

some famous violations

We will then discuss perhaps the most popular behavioral

model, Prospect Theory.

2/41
EUT and Ecient Markets

Last time we saw that the price of an asset depends on how

much expected return investors require to be compensated for
the risk of the stock, E(rE ), P ricet = E(CashF lowt+1 )
1+E(rE )

E(re ) is the risk premium for this company, and reects the
(systematic) risk of the company and investors preference
toward risk.

According to "rational" asset pricing theories E(rE ) comes

from an EUT-based model.

3/41
Decisions Under Risk

Some notation: G = (x1 , p1 ; x2 , p2 ; ..., xn , pn )

GamblePG yields monetary payos xi with probabilities pi ,

where ni=1 pi = 1.

4/41
Expected Utility Theory

Jon Von Neumann and Oscar Morgenstern in 1953 showed

that, if the preferences of a person satisfy a few logical axioms,
then it must be the case that the decision maker makes
choices according to some utility function, E[U (W )].
This basically means that the person applies their utility
function to rank lotteries with risky payos, and chooses the
lottery with the largest E[U (W )].
Some notation: If we have two lotteries P and Q, if P  Q
(P preferred to Q) this means that E[U (P )] ≥ E[U (Q)].

5/41
Expected Utility Theory

von Neumann-Morgenstern shows that if preferences satisfy

the following logical rules, then choices have an E[U(.)]
representation:
Completeness: For all lotteries P and Q, either P  Q or
P  Q.
Transitivity: For all lotteries P , Q, Z , if P
 Q and Q  Z
then P  Z
Continuity: For all lotteries P , Q, Z , if P  Q and Q  Z
then Q ∼ αP + (1 − α)Z for some α ∈ (0, 1)
Independence: For all lotteries P , Q, Z , if P  Q then
αP + (1 − α)Z  αQ + (1 − α)Z for some α ∈ (0, 1)

6/41
EUT Application

Assume our current level of wealth is W , and out utility

function is U (W )
We are presented with a gamble G = (x1 , p1 ; x2 , p2 ; x3 , p3 )
The decision makes calculates E[U(G)]:
p1 × U (W + x1 ) + p2 × U (W + x2 ) + p3 × U (W + x3 )

The decision maker compares G with other gambles on oer

and chooses the gamble with the higher utility value.
The price of the gamble depends on ow much people are
willing to pay for the right to play it.

7/41
Shape of utility function

Evidence from experiments strongly indicate that peoples' utility is increasing in

wealth but at a decreasing rate (indicating risk aversion)

8/41
EUT Hallmarks

According to EUT people integrate gambles with their overall

wealth, and choose the gamble that maximizes their total
expected utility
EUT is a "rational" model since its axioms are logical rules
that peoples preferences should satisfy
For this reason EUT is known as normative model of choice,
depicting how rational people should behave to maximize their
well-being.

9/41
Lets look at some decision problems

10/41
Expected Utility Theory: Violations
Choose between A1 and A2 :
A1 : (£1M, 1)
A2 : (£5M, 0.1; £1M, 0.89; 0, 0.01)

11/41
Expected Utility Theory: Violations
Choose between A1 and A2 : Now choose between A3 and A4 :
A1 : (£1M, 1) A3 : (£5M, 0.1; £0, 0.9)
A2 : (£5M, 0.1; £1M, 0.89; 0, 0.01) A4 : (£1M, 0.11; £0M, 0.89)

11/41
Expected Utility Theory: Violations
Choose between A1 and A2 : Now choose between A3 and A4 :
A1 : (£1M, 1) A3 : (£5M, 0.1; £0, 0.9)
A2 : (£5M, 0.1; £1M, 0.89; 0, 0.01) A4 : (£1M, 0.11; £0M, 0.89)
Most people in experimental studies choose A1 and then A3 , which
violates the independence axiom → the Allaix Paradox.
A1 A2 A4 A3
¿ p ¿ p ¿ p ¿ p

1M 0.89 1M 0.89 0M 0.89 0M 0.89

1M 0.11 0M 0.01 1M 0.11 0M 0.01
5M 0.10 5M 0.10
The stu in red are common consequences in both sets of lotteries, so
according to the independence axiom should not aect our preferences.
Either choose A1 and A4 or A2 and A3 .
11/41
Expected Utility Theory: Violations

Choose between B1 and B2 :

B1 : (£2K, 0.9; 0K, 0.1)
B2 : (£4K, 0.45; £0K, 0.55)

12/41
Expected Utility Theory: Violations

Choose between B1 and B2 : Now choose between B3 and B4 :

B1 : (£2K, 0.9; 0K, 0.1) B3 : (£2K, 0.002; £0, 0.998)
B2 : (£4K, 0.45; £0K, 0.55) B4 : (£4K, 0.001; £0M, 0.999)

12/41
Expected Utility Theory: Violations

Choose between B1 and B2 : Now choose between B3 and B4 :

B1 : (£2K, 0.9; 0K, 0.1) B3 : (£2K, 0.002; £0, 0.998)
B2 : (£4K, 0.45; £0K, 0.55) B4 : (£4K, 0.001; £0M, 0.999)
Most people choose B1 over B2 and then B4 over B3, which violates
EUT.
Choice B1 over B2 implies:
0.9U (2K) ≥ 0.45U (4K) ⇒ U (2K) ≥ 0.5U (4K)
Choice B4 over B3 implies:
0.001U (4K) ≥ 0.002U (2K) ⇒ U (2K) ≤ 0.5U (4K).
The ratio of probabilities attached to 4K and 2K in the two choices is the
same 0.5, so according to EUT common ratios should lead to identical
choices.
However it seems that people underweight the dierence between 0.002
and 0.001.The violation is known as the common ratio eect.
12/41
Expected Utility Theory: Violations

Choose between C1 and C2 :

C1 : (£4K, 0.8; 0K, 0.2)
C2 : (£3.2K, 1)

13/41
Expected Utility Theory: Violations

Choose between C1 and C2 : Now choose between C3 and C4 :

C1 : (£4K, 0.8; 0K, 0.2) C3 : (−£4K, 0.8; £0, 0.2)
C2 : (£3.2K, 1) C4 : (−£3.2K, 1; )

13/41
Expected Utility Theory: Violations

Choose between C1 and C2 : Now choose between C3 and C4 :

C1 : (£4K, 0.8; 0K, 0.2) C3 : (−£4K, 0.8; £0, 0.2)
C2 : (£3.2K, 1) C4 : (−£3.2K, 1; )
Most people choose C2 and then C3, which again violates EUT.
this is known as the the reection eect.
It implies risk aversion in the domain of gains and risk seeking in
the domain losses.
This means that people's risk attitude is not constant but
domain-dependent.
Incompatible with the notion that certainty is generally preferable.
Rather it seems that certainty is only preferred when choosing
among gains.
13/41
Expected Utility Theory: Violations

Imagine you are 20K richer and you have two options:
A1: get 5K for sure
A2: get 50% chance to win 10K and 50% chance to win 0.

14/41
Expected Utility Theory: Violations

Imagine you are 20K richer and you have two options:
A1: get 5K for sure
A2: get 50% chance to win 10K and 50% chance to win 0.

Imagine you are richer by 30K and you have two options:
B1: Lose 5K for sure
B2: 50% chance of losing 10K and 50% of losing 0.

14/41
Expected Utility Theory: Violations

Imagine you are 20K richer and you have two options:
A1: get 5K for sure
A2: get 50% chance to win 10K and 50% chance to win 0.

Imagine you are richer by 30K and you have two options:
B1: Lose 5K for sure
B2: 50% chance of losing 10K and 50% of losing 0.
People commonly feel that the two problems are dierent, one with
losses and the other gains,choosing A1 and then B2.
This relativity eect violates an important rule of rational decision
making- that the two problems are identical when formulated in
terms of nal states of wealth.

14/41
Expected Utility Theory: Violations

These violations of EUT are at the center of behavioral

economics
Many models have been developed that modify EUT in some
specic way, to explain these violations.
By far the most important model is Prospect Theory (PT)
proposed by psychologists Daniel Kahneman and Amos
Tversky in 1979, in a paper published in Econometrica.
This is perhaps the most cited paper in all the social sciences,
with 60K citations!

PT is a positive model of decision making, i.e., showing how

people actually behave
15/41
Prospect Theory

Suppose we have a gamble:

P = (x−m , p−m ; x−m+1 , p−m+1 ; ..; x0 , p0 ; xn−1 , pn−1 ; ..; xn , pn )

EUT says we value it as: ni=−m pi U (W + xi ).

P

where W is our current wealth position.

Prospect theory (PT) says we value it as: i=−m πi v(xi ).

Pn

Payos according to PT are evaluated in terms of scales v and π .

v is the value function assigns to each outcome x a number
v(x), which reect its subjective value to the decision maker.
This function does not obey all the axioms of EUT
π(p) is the probability weighting function which transforms the
probability of the payo p.
16/41
PT:The Value function

Kahneman and Tversky propose v(.) takes the form:

(
xα x≥0
v(x) =
−λ(−x)α x<0

where α < 1 and λ > 1

The argument of v(.) is xi , so people get utility from changes

in wealth relative to a reference point, rather than absolute
wealth levels
The slope of the function is steeper for losses, i.e., utility drops
much faster for decreases in wealth. This is captured by loss
aversion λ, which makes v(.) steeper in the loss domain
17/41
PT:The Reference Point

xi is coded as gain or loss relative to our reference point.

In many applications the reference point is the status quo,

i.e., if a bought something for ¿10, then 10 is the reference
point
In recent work is has been highlighted that the reference point
is adaptive, i.e., it depends on expectations, i.e., if i bought at
10 and expect to sell at 12, my reference point is 12.
This makes applications of PT in economics/nance
particularly challenging.

18/41
PT:The Probability Weighting Function

Kahneman and Tversky propose π(p) takes the form:

pδ
π(p) = (pδ +(1−p)δ )1/δ

When δ = 1 decision weights π(p) = p. But KT nd that

δ ≈ 0.65

This means small probabilities are over-weighted, and large

probabilities are under weighted, i.e., π(p) > p for small p, and
π(p) < p for large p.

19/41
PT

20/41
EUT vs. PT

Three key dierences between EUT and PT

1 In PT alternatives are evaluated as changes relative to a
reference point, not nal wealth positions.
2 In PT risk attitide is domain dependent; people are risk averse
in gains and risk seeking in losses
3 In PT probabilities are distorted into "decision weights"
according to the probability weighting function.

21/41
PT and stock market anomalies

22/41
The Equity Premium Puzzle

In EUT-based models asset returns are driven by:

1 Risk: covariance of returns with systematic risk factors.
2 Risk aversion: The concavity of the utility function.
Therefore, the equity risk premium, E[ri,t ] − rft depends
positively on (1) and (2).
Economists calculate that based on observed risk and
reasonable risk aversion the equity risk premium should be
around 1-2%.
Equity premium between 1926-2000: US 7.8%; U.K. 4.6%;
Germany 6.6%; France 6.3%.
Thus the risk premium seems to high, the equity premium
puzzle.

23/41
The Equity Premium Puzzle

Can this puzzle be resolved within an EUT framework?

Covariance between stocks and consumption is not very large,
so stocks are not that risky.
Maybe people are more risk averse? The equity risk premium
requires that one prefers:
A : (£51, 201; 1) to B : (£50K, 0.5; £100K, 0.5).
No one is that risk averse!
The equity premium puzzle is one of the most dicult
questions in macro-nance

24/41
Benartzi and Thaler 1995

Benartzi and Thaler (1995) (BT) resolve this puzzle using two
(very plausible) behavioral phenomena
1 Investors are loss averse, so very concerned about the
probability of loss (PT preferences).
2 Investors are myopic when calculating probability of losses, so
consider only recent history.

25/41
Pr(losses) with 1-year evaluation periods

26/41
Pr(losses) with 20-year evaluation periods

27/41
Benartzi and Thaler 1995

The probability of losses in stocks is higher for shorter

evaluation periods.
So myopia, leads investors to perceive a higher probability of
loss in equities.
Because they are also loss averse, the potential for losses is
particularly hurtful, so investors require a higher risk premium
to hold stocks.
Assuming a λ = 2.25, and assuming an evaluation period of 1
year, BT show that the implied equity premium is around 7%,
thus resolve the equity premium puzzle!

28/41
Implied equity premium with myopic PT investors

29/41
Implied equity premium with myopic PT investors

30/41
PT and risk taking

In PT people are risk seeking in the domain of losses (since

they hate losses so much they want to take risk in order to
have some chance of avoiding them)
This has implications for risk taking behavior: Investors will
take more risk after they have experienced nancial losses
Shefrin and Statman (1985) predicted that this behavior
should make investors hold onto losing position longer than
winning positions. This is known as the disposition eect.
Note that the disposition eect is (probably) harmful from a
tax point of view. By selling losers, your capital losses can
oset other gains, therefore reduce your tax bill.

31/41
PT and the Disposition Eect

32/41
Disposition Eect, Odean 1998

Are Investors Reluctant to Realize Their Losses? 1783

Table I
PGR and PLR for the Entire Data Set
This table compares the aggregate Proportion of Gains Realized ~PGR! to the aggregate Pro-
portion of Losses Realized ~PLR!, where PGR is the number of realized gains divided by the
number of realized gains plus the number of paper ~unrealized! gains, and PLR is the number
of realized losses divided by the number of realized losses plus the number of paper ~unrealized!
losses. Realized gains, paper gains, losses, and paper losses are aggregated over time ~1987–
1993! and across all accounts in the data set. PGR and PLR are reported for the entire year, for
December only, and for January through November. For the entire year there are 13,883 real-
ized gains, 79,658 paper gains, 11,930 realized losses, and 110,348 paper losses. For December
there are 866 realized gains, 7,131 paper gains, 1,555 realized losses, and 10,604 paper losses.
The t-statistics test the null hypotheses that the differences in proportions are equal to zero
assuming that all realized gains, paper gains, realized losses, and paper losses result from
independent decisions.

Entire Year December Jan.–Nov.

PLR 0.098 0.128 0.094
PGR 0.148 0.108 0.152
Difference in proportions 20.050 0.020 20.058
t-statistic 235 4.3 238

33/41
PT and Sophisticated Investors

Coval and Shumway (2005) examine whether loss aversion

inuences professional market makers trading Treasure Bond
futures contracts at CBOT

These people trade, on average, options worth $200M per day,

so are quite sophisticated
Coval and Shumway (2005) split the trading in two sections,
morning and afternoon, and ask the following question: Do
losses in the morning lead to more risk taking in the afternoon,
consistent with loss aversion?
[Coval, Joshua D., and Tyler Shumway. "Do behavioral biases aect
prices?." The Journal of Finance 60.1 (2005): 1-34]

34/41
PT and Sophisticated Investors

Figure 1: Morning Profit Percentile and Afternoon Risk-Taking

Traders
Thiswith
figure morning losses averages
plots the time-series are more of 236likely to take above
daily cross-sectional average risk in
semi-parametric
the afternoon
regressions of afternoon total dollar risk on morning profit percentile. The regressions are
35/41
Loss aversion and the cross section of stocks

Does loss aversion inuence the pricing of individual stocks?

Barberis, Abhiroop and Wang (2016): when thinking

about allocating money to a stock, investors mentally
represent the stock by the distribution of its past returns (5
years), and then evaluate this distribution in the way described
by prospect theory.

Loss making stocks will be under-invested (undervalued) and

therefore will outperform prot making stocks

36/41
Loss  aversion and the cross section of stocks

1.2

1.0

0.8

0.6

0.4

0.2

0.0
Low-1 2 3 4 5 6 7 8 9 High-10
-0.2

-0.4
EW-TK
 

Low
0.6 PT-value stocks out-perform high PT-value stocks by 1.4% per
month
37/41
Loss aversion and the cross section of stocks

Additional ndings by Barberis, Abhiroop and Wang (2016):

The results are stronger among small-capitalization stocks,
where less sophisticated investors are likely to have a bigger
impact on prices, and where limits to arbitrage are more
binding.

BAW test the hypothesis in 46 international stock markets,

and nd a similar pattern in the majority of them.

38/41
PT and the Pricing of Skewness

Barberis and Huang (2008) incorporate PT in asset pricing,

focusing on probability weighting.
Their model predicts that stocks with positive skewness will
be overpriced.
This is because, due to probability weighting,the small
probability of gain is inated, which results to an overpricing of
"lottery-like" assets

39/41
PT and the Pricing of Skewness

This logic can explain several anomalies:

"Lottery-type" stocks have lower average returns (IPO's,
low-priced and high volatility stocks). This particularly
challenging for EUT-based models because such stocks are
riskier but have lower average returns

Out-of-the-money options have lower average returns.

40/41
Readings

Essential Readings:
Barberis, N., Mukherjee, A., & Wang, B. (2016). Prospect theory
and stock returns: an empirical test. The Review of Financial
Studies, 29(11), 3068-3107.
Kahneman D. and Tversky A. (1979), Prospect Theory: An Analysis
of Decision under Risk, Econometrica, 47(2),pp. 263-291
Benartzi, Shlomo, and Richard H. Thaler, 1995, "Myopic loss
aversion and the equity premium puzzle." The Quarterly Journal of
Economics, 73-92.
Odean, T., (1998), Are investors reluctant to realize their losses?
Journal of Finance 53, 1775-1798
Additional Reading:

Barberis, N., 2012, "Thirty Years of Prospect Theory in Economics: A Review and Assessment",
Journal of Economic Perspectives.
Grinblatt, Mark, and Bing Han. "Prospect theory, mental accounting, and momentum." Journal
of Financial Economics 78.2 (2005): 311-339.
Frazzini, Andrea. "The disposition eect and underreaction to news." The Journal of Finance
61.4 (2006): 2017-2046.
41/41