Professional Documents
Culture Documents
Cultural Finance
Cultural Finance
Cultural Finance
Finance
Prof. Marc Oliver Rieger
University of Trier
www.banking-finance.uni-trier.de
Research areas:
Behavioral Finance, structured financial products, decision theory
About geography
any
September 2018).
Germ
Born 1974 in Berlin, Germany
2001 PhD in mathematics at the Max-Planck-Institute for
Mathematics in the Sciences, Leipzig, Germany
2001-2003 Carnegie Mellon University, Pittsburgh, USA
2003-2004 Scuola Normale Superiore, Pisa, Italy
2004-2010 University of Zurich, Switzerland
3
Research areas:
About Trier
any
September 2018).
Germ
Born 1974 in Berlin, Germany
2001 PhD in mathematics at the Max-Planck-Institute for
Mathematics in the Sciences, Leipzig, Germany
2001-2003 Carnegie Mellon University, Pittsburgh, USA
2003-2004 Scuola Normale Superiore, Pisa, Italy
Founded2004-2010
by the Romans, Northern
University capital
of Zurich, of the Roman empire, residence
Switzerland
of the archbishop of Trier, now a medium sized city (pop. 100000), on the
4
border toResearch
Luxembourg.
areas:
About University of Trier
any
September 2018).
Germ
Born 1974 in Berlin, Germany
2001 PhD in mathematics at the Max-Planck-Institute for
Mathematics in the Sciences, Leipzig, Germany
2001-2003 Carnegie Mellon University, Pittsburgh, USA
2003-2004 Scuola Normale Superiore, Pisa, Italy
2004-2010 University of Zurich, Switzerland
Founded in the middle age, but closed down by Napoleon.
5
Reopened in 1970,areas:
Research today approx. 12000 students, located outside town.
And where are you from?
(Please turn on your camera!)
6
What are Cultural
Economics and Finance?
Cultural
Economics and
Finance
7
Why do we study
Cultural Economics and Finance?
8
Main Research
Methodologies
‣ Surveys
‣ Experiments
9
Table of Content
10
Schedule
Week 1
‣ Lectures will be Mon/Tue/Wed/Thu/Fri, 4x45 min.,
including discussions on group projects
Week 2
‣ Mon/Tue/Wed/Thu: time for the work on the group
project
‣ Friday: group discussions on projects
Saturday, September 18
‣ presentations of group projects, concluding discussion
11
Information on
Group Project (1)
12
Information on
Group Project (2)
13
Information on
Group Project (3)
15
Further Information on
Group Project
‣ Most topics are from ongoing research projects.
‣ Sometimes I already know what should come out of the data
analysis, sometimes I do not know yet. (In any case, feel free
to surprise me with new observations!)
‣ You are quite free regarding your precise research
methodology. OLS Regressions are often a good choice.
Make sure that the method fits the problem and is not
unnecessarily complicated.
‣ You may use any software for the analysis (SPSS, Stata,
SAS, R, Mathlab), but not only Excel!
‣ Make sure to consider statistical significance of your findings!
‣ Written report on presentation (including bibliography) to
be submitted by the official deadline via email, please
16
use the subject: UZH Seminar
Chapter 1
Foundations:
Classical and Behavioral
Preferences
17
1.1 Expected Utility Theory
Important remark:
We first discuss only preferences regarding risk, not
regarding time.
18
St. Petersburg paradox
19
Expected value of the lottery
20
Expected utility (1)
Nutzen
Vermögen
21
Expected utility (2)
22
following way:
Definition 2.6 (Concavity and risk-averse behavior). We call a
function u : R ⌅ R concave on the interval (a, b) (which might be R) if
for allRisk x2 ⇧ (a,and
x1 , aversion b) and
the utility (0, 1) the following inequality holds:
⇧function
Howu(x can1) +
we(1describe
)u(x2 )risk
⇥ u( x1 + (1with)x
aversion 2 ). utility function?
the (2.1)
We call uTheorem:
strictly concave if the above inequality is always strict (for
x1 ⌃= x2 ).AWeconcave utility function
call a person risk-averseu describes a risk
if he prefers averse value
the expected
behavior.
of every lottery overAtheconvex
lotteryutility
itself.function
5 u describes risk seeking
2.2 Expected Utility Theory 17
behavior.
We see that the application of Expected Utility Theory leads to quite
realistic results. We also u(x0see
) that a crucial factor for the explanation
of the attractiveness
u(x1 ) + (1 of insurances
)u(x 2)
and the solution of the St. Peters-
burg Paradox is the concavity of the utility function. Roughly spoken,
concavity corresponds to risk-averse behavior. We formalize this in the
following way:
Definition 2.6 (Concavity and risk-averse behavior). We call a
function u : R ⌅ R concave
x1 on the interval
x0 (a,xb)2 (which might be R) if
0 =1) xthe
for all x1 , x2 ⇧ (a, b) and ⇧x(0, 1 + following
(1 )x2 inequality holds:
Fig. 2.6. A strictly concave function. 23
24
Further axioms
25
von Neumann-Morgenstern Theorem
26
1.2 Framing effect
27
Examples
28
Examples
29
Examples
Another example:
30
Examples
31
Examples
32
Examples
33
Examples
35
1.3 Prospect Theory
36
making under risk for a long period
Decision
Prospect under
Theory Risk: A choice between prospects or gambles.
n
A Prospect is a contract (x1 , p1 ; ...; xn , pn ) with i=1 pi =1
where
xi denotes the outcomes i = 1...n;
pi denotes the probabilities i = 1...n;
(x,p) denotes the prospect (x,p;0,1-p).
37
Certainty Effect
People always overweight certain outcomes in comparison with
outcomes merely probable.
Prospect Theory
38
Violation from EUT: Several Counter Phenomena
Certainty Effect
Prospect Theory
Certainty Effect
Prospect Theory
Problem 5:
A: 0.50 chance to win a B: A one-week tour of UK
three week tour of UK, with certainty.
France and Italy;
N = 72 [22] [78]*
Problem 6:
C: 0.05 chance to win a D: 0.10 chance to win a
three week tour of UK, one-week tour of UK with
France and Italy; certainty.
N = 72 [67]* [33]
QEM 2 Pengfei Sun Seminar “Behavioral Decision Theory and Finance”
40
Certainty Effect
Prospect Theory
Problem 7:
A: (6,000,0.45), B: (3,000,0.90),
N = 66 [14] [86]*
Problem 8:
C: (6,000,0.001), D: (3,000,0.002),
N = 95 [73]* [27]
41
rom EUT: Several Counter Phenomena
ection Effect
Prospect Theory
Reflection effect:
Probabilistic Insurance
Violation from EUT: Several Counter Phenomena
Probabilistic Insurance
Prospect Theory
Probabilistic Insurance
Probabilistic
Probabilistic insurance:
Insurance
People pay part of the premium and also have part of the chance
People pay part of the premium and also have part of the chance
to gettoallgetcover of the
all cover losses.
of the losses.
Problem 9: Now
Problem 9: Now youyou
cancanpay payhalf
halfof of the premium,inincase
the premium, case of of
damage, you you
damage, havehave0.50
0.50 chance
chancetotopay pay the other otherhalf halfofofthethe
premium
premium to covers
to covers all all
thethelosses;or
losses;or have
have 50 50 percent
percentchance chance to to
get back
get back insurance
insurance paymentand
payment andsuffer
suffer all
all the
thelosses.
losses.Would Would
you buy this insurance?
you buy this insurance?
N = 95 Yes[20] No[80]*
N = 95 Yes[20] No[80]*
If (w x, p; w , 1 p) ⇤ (w y ), then
If (w (w x,x, w ⇤y(w
p;(1w , 1r )p;p) , rp; w y ),rythen
, 1 p) ⇥ (w y ), 0 < r <
(w 1.x,⌅, (1 u isr )p; w y , rp; w ry , 1 p) ⇥ (w y ), 0 < r < 1
concave.
1. ⌅,This
u iscase
concave.
is inconsistent with concavity of u
This case is inconsistent
QEM 2
with concavity
Pengfei Sun
of u
Seminar “Behavioral Decision Theory and Finance”
Isolation Effect
Prospect Theory
Isolation Effect
PeopleIsolation
alwaysEffect
disregard the same components shared by the
Peopleand
prospects, always disregardwith
concerned the same components shared
the components by the them.
distinguish
prospects, and concerned with the components distinguish them.
44
Isolation Effect
Prospect Theory
Isolation Effect
Prospect Theory
Problem 11: If you have been give 1,000. Now choose between
A: (1,000,0.50), B: (500),
N = 70 [16] [84]*
Problem 12: If you have been give 2,000. Now choose between
C: (-1,000,0.50), D: (-500),
N = 68 [69]* [31]
46
Prospect Theory
47
Prospect Theory
48
Prospect Theory
Valuation Phase
Prospect Theory
Valuation phase:
49
Valuation Phase
Prospect Theory
Concave for gains and convex for losses; for x > 0, v ”(x) < 0;
for x < 0, v ”(x) > 0
Properties
Prospect Theory of Weighting Function
Graphics
Prospect Theory
Value function
Weighting function
53
QEM 2 Pengfei Sun Seminar “Behavioral Decision Theory and Finance”
Seminar “Behavioral Decision Theory and Finance”
Prospect Theory: An Analysis of Decision under Risk
Prospect Theory
Overweighting
Prospect Theory
Small Probabilities
Overweighting of small probabilities:
Problem 13:
A: (5,000,0.001), B: (5),
N = 72 [72]* [28]
Problem 13’:
C: (-5,000,0.001), D: (-5),
N = 68 [17] [83]*
Reference
What aboutpoint
if the Shifts of the reference happened?
56
Prospect theory
x x
p p
1-p 1-p
y 0
w(p) > p
Losses Gains
58
Prospect Theory –
Four-fold pattern of risk attidues
Losses Gains
59
Prospect theory – an exercise
60
Prospect theory
61
1.4 Cumulative Prospect
Theory (CPT)
Definition:
If for any given amount x, the lottery A has a higher (or at least
equally good) chance to yield x or more than B, then we say
“A stochastically dominates B”.
It’s quite rational to chose A in this case!
PT, however, does not always predict this, see the example in
the group discussion or this, even more extreme, one:
A: certain payoff of 100€
B: lottery with 1/100 probability each for 99€, 99.01€, 99.02€..
Here, the small probabilities will be overweighted, resulting in
a total PT-utility exceeding that of A.
Lottery A Lottery B
14 Euro 5% 90 Euro 5%
Another example:
One of four teams — the Cleveland Indians or Seattle
Mariners from the American League, or the Atlanta Braves
or Cincinnati Reds from the National League — will win the
1995 World Series.
The prospect below offers hypothetical amounts of money
depending on the outcome of the World Series.
Choice A: If an American League team wins the World
Series, You win $220. Otherwise You win $0.
Choice B: If the Cleveland Indians win the World Series,
You win $220. If the Seattle Mariners win the World Series,
You win $200. Otherwise You win $0.
Which one do you choose?
65
Cumulative prospect theory
66
Decision Theory
Prospect Theory
Cumulative
Cumulative Prospect
Prospect TheoryTheory
68
us first extend CPT violation of stochastic
to arbitrary dominance,
lotteries. Since we lack of continuity) and p
all the
sume state-independent modified theory still
preferences, a good
we can descriptive
describe model of behavior
lotteries
Let us first
bability measures, see Appendix A.4 extend CPT to arbitrary lotteries. Since
for details.
timetheory
Cumulative prospect assume state-independent preferences, we can descri
tion 2.35. Let p be by an probability
arbitrary probability
measures,measure,
see Appendixthen theA.4 for details.
ized form ofCPTCPTfor 17 continuous
reads as distributions:
Definition
! +∞ ⇤ " 2.35. Let ⇥ p#be an arbitrary
⇤ probability measur
generalizedd d form of CPT17 reads as ⇥
CP T (p) := v(x) w± (F (x))
w(F (t))|t=x
v(x)dx,dx = w± (F (x))v(x)p(x) dx
(2.9)
−∞ dxdt ! +∞ " #
d
‣Here we have redefined CP T (p) w, :=
since−∞ v(x) wew(F
otherwise (t))|
have to t=x dx,
dt
write two integrals:
! t one for gains and one for losses...
(Exercise:
Fwhere
(t) :=how dodp. we have to transform w?)
−∞ ! t
cognoscenti we‣F remark
is the cumulative
that the formula (2.6)
for
probability: F (t) :=
lotteries dp.
with
−∞
many outcomes is just a special case of (2.9) when choosing p
‣ The second
ite sum of Diracs. For theformulation
cognoscentiis we only valid ifthat
remark certain
the formula (2.6) for lo
nition 2.35 pathsmathematical
the way to
finitely conditions
applications
many holdof(smoothness,
outcomes isCPT
just in decay
financial
a special caseatof (2.9) when
infinity).
ics and other areas as aIt finite
where ismodels
derived using
sumrequire partial
more than
of Diracs. integration.
just a cou-
otential outcomes. Although it looks
Definition 2.35atpaths
first glance
the way muchto more
applications of CPT
69
d than its finite counterpart
economics(compare
and other Definition
areas where ??),models
a closerrequire more than
Advances in Prospect Theory - Cumulative Prospect Theory
Value Function and Cumulative Weighting Function
Qualitative Properties
Cumulative prospect theory
Value Function
Diminishing Sensitivity.x ⇤ 0, v ”(x) < 0; x ⇥ 0, v ”(x) > 0
Loss Aversion.v (x) < v ( x) for x > 0 (in the case 𝛼=𝛽)
Proposition 2.36.
2.39 CPT does not violate stochastic dominance, i.e. if
A is stochastic dominant over B then CP T (A) > CP T (B).
17
Here we consider again only the form defined in this book. In the original for-
mulation we would need to write down two integrals for negative and positive
outcomes and invert the direction of integration on the latter one. Compare the
remark after Def. 2.34
71
ulting function v has a “kink” at zero, a marginal loss is considere
robability events in than
lot more important the mid-range
a marginalofgain. outcomes play ameasured a
λ is usually
mewhere between 2 and 2.5.
Cunmulative prospect theory – classical form of value
her related theory,
and weighting functionsRank Dependent Utility (RDU),
73
1.5 CPT and the St.Petersburg
paradox
74
Petersburg Paradox”. He concluded that unless the utility function is bounded, it is
impossible to discriminate all possible probability distributions. One could argue,
however, it is not necessary to discriminate all possible probability distributions,
because no individuals or organizations can offer a lottery with unlimited EV.
Arrow result
Actually, Arrow [2] pointed out that if we only consider distributions with finite
EV, we can still guarantee finite expected utility even though the utility function is
Arrow (1974) gives the following precise result:
unbounded. More precisely he found the following result:
Proposition 1 Let p be a probability measure with finite EV E(p) < ∞ and let
! : R → R be a strictly increasing, concave utility function, then the utility U (p) :=
u
u dp is finite.
(We remark again that this statement is a generalization of the case of discrete
outcomesIn prospect
where the theory
integral there is a by
is replaced potential problem:
a sum. Hence the usual formulation of
the St.probability weighting
Petersburg problem gives
in terms of amore weight
discrete lotterytoisextreme
included.)events,
In thus
other the utility
words, the might still become
St. Petersburg paradoxinfinite.
can be resolved by allowing only
for “realistic” lotteries, i.e., under the assumption of a finite EV, a (not necessarily
strictly) concave utility function is sufficient to guarantee that the expected utility
In the following we will discuss this problem in details.
is finite.
Even though this fundamental statement is mathematically easy in the frame-
work of expected utility theory, it turns out to be false in the context of CPT. In fact,
we will show in the Section 2 that in cumulative prospect theory a gamble with a
finite EV can have an infinite subjective value – independent of the concavity75 of the
value function u.
The central result of this section is the following Theorem 1. It provides precise
conditions under which the St. Petersburg paradox in CPT occurs. We will later
derive several results from this theorem which are more accessible and which cover
General result
situations of particular interest, compare Theorems 3–5.
Theorem 1 (General result) Let U be a CPT subjective utility given by
! 0 ! +∞
d d
U (p) := u(x) (w− (F (x))) dx + u(x) (w+ (F (x))) dx,
−∞ dx 0 dx
where the value function u is continuous, monotone, convex for x < 0 and concave
for x > 0. Assume that there exist constants α, β ≥ 0 such that
u(x) |u(x)|
lim = u1 ∈ (0, +∞), lim = u2 ∈ (0, +∞), (3)
x→+∞ xα x→−∞ |x|β
and that the weighting functions w± are continuous, strictly increasing functions
from [0, 1] to [0, 1] such that w± (0) = 0 and w± (1) = 1. Moreover assume that
w± are continuously differentiable on (0, 1) and that there are constants δ, γ > 0
such that
′
w− (y) 1 − w+′
(y)
lim = w1 ∈ (0, +∞), lim = w2 ∈ (0, +∞). (4)
y→0 y δ−1 y→1 (1 − y)γ−1
Let p be a probability distribution with E(p) < ∞. Then U (p) is finite if α <γ
and β < δ. This condition is sharp as can be seen from Theorem 2 below.
76
In particular, the subjective utility may be infinite for distributions with 2006)
(Rieger, Wang, finite
Cumulative
Exampleprospect theorywith
for a lottery and the St.expected
finite Petersburgvalue,
paradox
but 671
infinite subjective utility
Theorem 2 (Finite EV, but infinite subjective utility) Let γ, α ∈ (0, 1), q > 2.
Let the probability density p of possible outcomes be given by
!
0, x≤1
p(x) :=
Cx−q , x > 1,
"∞
where C := 0 x−q dx.
Let the weighting function w+ : [0, 1] → [0, 1] be given by (2) and the
value function on u : R+ → R be given by (1), i.e. value- and weighting func-
tion are "chosen as in the original work by Tversky and Kahneman [15]. Let
x
F (x) := −∞ p(x) dx be the total probability for an outcome less than x. Then we
have E(p) < +∞ and u strictly concave, but for α >γ and q sufficiently close to
2 the subjective utility is infinite, i.e.
#
d
U (p) := u(x) (w(F (x))) dx = +∞.
dx
77
This result shows that it is not possible to resolve the St. Petersburg paradox in
U (p) = u(x) w(F (x)) dx ≤ u(b) w(F (x)) dx
directly.
We firstWeconsider
assume afor conditionsdx
simplicityon that w := w− =
the probability dx
w+a, use the
distribution of monotonicity
the outcomes.of Theu
and
St. Petersburg ∈=[0u(b)(w(F
that w(F ) paradox (b)) − w(F
, 1]: can obviously not(a)))
occur≤if u(b) < ∞.ourselves to a finite
we restrict '
⊓
set of possible outcomes, ! b but even a bounded set of !possibleb
outcomes suffices to
preventMany
Special interesting
infinite
cases utility: d distributions (e.g. normald Gauss distributions) do
probability
U (p) = u(x) w(F (x)) dx ≤ u(b) w(F (x)) dx
not have a boundedasupport.dxTherefore the following extension a dx is useful:
Theorem 3 (Finiteness = u(b)(w(F for (b))
special
− w(Fdistributions
(a))) ≤ u(b) I) Let
< U∞.be a CPT subjective
Theorem 4 (Finiteness foraspecial distributions II) with
Let Ubounded
be a CPT subjective '
⊓
utility functional and p be probability distribution support, i.e.
utilitypfunctional
supp :=interesting
{x ∈ R; andp(x) p be >a probability
[a, b],distribution
0} ⊂distributions
where(e.g. with
−∞exponential
a >normal and +∞.decay
Then
b < distributions) +∞,
at U (p)
Many probability Gauss do
there exist a, b, c > 0such that p(x) ≤ ae
i.e.finite.
is
−bx
for all x ≥ c. Moreover, assume
not have a bounded support. Therefore the following extension is useful:
that (4) holds. Then U (p) < +∞. (The corresponding condition for −∞ would
ensure This
Proof.
Theorem (p) > −∞.)
that4 Ufollows
(Finiteness fromfor thespecial
general result, Theorem
distributions 1, U
II) Let butbeit acan
CPTalso be seen
subjective
directly.
utility We assume
functional andfor p besimplicity
a probability w := w− =with
that distribution w+ ,exponential
use the monotonicity
decay at +∞,of u
Proof.
and that This
w(F result
) ∈ [0 is, 1]:
an immediate consequence −bxof Theorem 6. '
⊓
i.e. there exist a, b, c > 0such that p(x) ≤ ae for all x ≥ c. Moreover, assume
that If(4)one holds.
wantsThento!allowUb (p)for +∞. (The
<arbitrary corresponding
probability ! b condition
distributions, a for −∞
general would
finiteness
d d
ensure thatU (p)
U (p)
= >
result can be given for bounded −∞.)u(x) w(F (x))
value dx ≤
functions: u(b) w(F (x)) dx
a dx a dx
Proof.
Theorem This5 result= u(b)(w(F
(Bounded value(b))
is an immediate w(F (a)))
−consequence
functions) Let U≤of Theorem
beu(b)
a CPT 6.
∞.
< subjective utility func-'
⊓
'
⊓
tional with bounded value function |u(x)| ≤ C and let p be a probability distribu-
If one wants to allow for arbitrary probability distributions, a general finiteness
tion.Many
Theninteresting
U (p) is finite. probability distributions (e.g. normal Gauss distributions) do
result
not have canabe given for
bounded bounded
support. value functions:
Therefore the following extension is useful:
Proof. Again, this is a corollary of Theorem 1, but a direct proof, following the
Theorem 5 (Bounded value functions) Let U be a CPT subjective utility78 func-
Theorem
ideas of the 4 proof
(Finiteness
of Theorem for special distributions
3 is also easy. II) Let U be a CPT subjective '
⊓
tional with bounded value function |u(x)| ≤ C and let p be a probability distribu-
utility functional and p be a probability distribution with exponential decay at +∞,
674 M.O. Rieger and M. Wang
Parameter
T estimates
able 1. Experimental for prospect
values of α, β andtheory from
γ, δ from various
various studies, compare (3) and (4) for the precise
¯definition of α, β, γ, δ
stuides
Abdellaoui [1]
gains: 0.89 0.60 no
losses: 0.92 0.70 no
79
Possible solutions – summary
80
Possible solutions – summary
81
Possible solutions – summary
82
1.6 Continuity and
prospect theory
83
problem of continuity: a small change of a lottery (e.g. splitting a single
outcome into two similar outcomes) changes the PT-value often substantially.
This is the cause of several theoretical problems of PT that could only be
resolved using cumulative prospect theory. Already Kahneman and Tversky
(1979) suggest a so-called “editing phase” before the evaluation of the formula
for the
Approximation PT-value. In the editing phase in particular nearby outcomes are
idea
combined. We formalize this process and call the resulting modified theory
smooth prospect theory (SPT). We show that this theory is in fact continuous.
Mathematical idea:
Its key idea is to approximate
make the outcomes “fuzzy”,ai.e.probability distribution
replacing an outcome at, say, by
discrete probabilities
x by an (“histograms”):
outcome-distribution around x. Nearby outcomes are then handled
as something in between two separate outcomes and one combined outcome,
Fig. 1 Non-discrete outcome distributions are approximated by finer and finer histograms of
uniform width
84
and Tversky (1979) was only defined for lotteries with, at most, two non-zero
outcomes, it can be generalized to n outcomes. Generalizations have been
used by various authors, e.g., Fennema and Wakker (1997), Camerer and Ho
(1994), Wakker
Karmarkar (1989), Schneider and Lopes (1986). In this article, we study
formulation
the original formulation for n outcomes by Karmakar (1978). The value of a
We need to use a “normalized” form of PT, introduced by
lottery with outcomes xi , each of probability pi , is then given by
Karmarkar (1978):
!n
i=1 w( pi )v(xi )
PT = !n . (1)
i=1 w( pi )
The measures p̄n are still infinite# sums of Diracs, but since p is a probability
measure, it is easy to see that [z/n,(z+1)/n) dp → 0 for |z| →∞ , thus we can
neglect all, but finitely many intervals by making an arbitrarily small error.
We call the resulting measure pn . By a small lemma (McCann 1995), this
⋆
approximates in fact p; in mathematical language: pn ⇀ p.
Generally, we can use any decomposition of R into equally sized intervals
[xi , xi+1 ) where the size hn := |xi − xi+1 | → 0 and the union of these intervals
89
natural possible choice that we can make.2
We will demonstrate in the next section how incorporating the editing phase
of prospect theory into our consideration makes it possible to derive this limit
for arbitrary approximating
Convergence sequences.
theorem for prospect theory
and let pn be defined as above. Assume v ∈ C1 (R). Moreover, assume that for
the weighting function w : [0, 1]→[0, 1]there exists some α ∈ (0, 1]. Then the
"
PT utility of Schneider and Lopes (1986)
%
J Risk Uncertainty (2008) 36:83–102
" 91
PT( pn ) = w( pn,z)v(z/n)
z
converges to
"
∞, if α < 1,
lim !PT( pn ) = !
n→∞
3 The assumption can be weakened, e.g. to Cp(x)
· EU( p),
α dx < +∞ ififv isαbounded.
= 1, 91
1.7 SP/A
SP/A theory
‣introduced by Lopes and Oden (1999)
‣ based on studies on emotions
‣Security, potential and aspiration
92
Goal
93
SP/A theory
i
mathematically, where F = 0 and F = ∑i p j for
where F0 0= 0 and Fi =i ∑ j = 1j p=j 1for i = 1,...,N
i = 1,...,N and
and w w± mean
± means that
s mathematically,
e ofofSP/A
ase SP/Acan
can be
be functionww
function ++is isapplied
appliedforforall all positive
positive outcomes
outcomes and wand w
− for
negativeoutcomes.
negative outcomes. Here
Here v isvaisconvex–concave
a convex–concave shaped
shaped value valu
func
and
and ww±±areareprobability
probability weighting
weighting functions in the
functions in gain
the and
gain
2
2 CPT is usually defined by a decumulative weighting in gains and a cumul
CPT is usually defined by a decumulative weighting in gains and a
weighting in losses, but redefining w± can be used to arrive at our more com
weighting in losses, but redefining w± can be used to arrive at our m
more details. formulation. 96
ore details. formulation.
A) = SP + h(A). Since the probability weighting h is increasing, t
function is admissible for SP/A.
Let us denote the resulting “utility” by SPA, then we can rewrit
asReformulating SP/A Theory
with
%
0; x b α;
aðxÞ =
1; x ≥ α:
97
3. Conclus
We se
psycholog
think at fi
within on
gives som
should loo
different m
Acknowle
I am gr
related to
Fig. 1. The function v implied by SP/A is the sum of a concave function u and a jump at α Wang for
(which we set to zero for convenience). Centre of
98
with
%
0; x b α; 3.
aðxÞ =
x≥
Connecting1;SP/A andα:
CPT
Let us denote v(x) = u(x) + a(x) and w(D) = h(D), then we have
reached Eq. (1), just that value and weighting functions are defined
slightly differently: av
100
Uncertain aspiration levels
101
hor's personal copy
ally, v and
in outcome
o the ideas
w how it can
ven by f(SP,
easing, this Again we obtain a
concave–convex–
n rewrite it concave shape.
Fig. 2. The function v implied by SP/A when we use a “fuzzy” aspiration level resembles
for small payoffs the convex–concave shape of standard CPT.
102
1 Þ$
1.8 Quasi-hyperbolic time
discounting
Previous studies
‣ Discount rate reflects the marginal rate of substitution
between current and future consumption.
103
Homogeneous or not?
Frederick, Shane, George Loewenstein, Ted O’Donoghue: Time discounting and time
preference: a critical review. Journal of Economic Literature, 40:350–401, 2002.
104
Time consistent or not?
Frederick, Shane, George Loewenstein, Ted O’Donoghue: Time discounting and time
preference: a critical review. Journal of Economic Literature, 40:350–401, 2002.
105
Quasi-hyperbolic time discounting
People are more patient for the far future than for the near
future.
Discount rate between t=0 and t=1:
Discount rate between t and t+1 (when t>1):
106
Endogenous time preference
107
Chapter 2
How Culture Affects
Economic Behavior
2.1 What is culture?
什什么是⽂文化?
Culture is something stable over time that
120 distinguishes different groups.
Example: Hofstede culture dimensions
Power Distance Index (PDI):
80 inequality of power and wealth.
Individualism (IDV):
degree the society reinforces individual or
40
collective achievement.
Masculinity (MAS):
dominance of traditional “male” values
0
PDI IDV MAS UAI LTO over traditional “female” values.
Uncertainty Avoidance Index (UAI):
USA Germany China
intolerance for uncertainty and ambiguity
Long-Term Orientation (LTO):
(Hofstede, 1980/2003) long-term commitments vs. respect 109
for
traditions.
What is culture?
什什么是⽂文化?
110
What is culture?
什什么是⽂文化?
111
Example: Uncertainty avoidance index
112
Other concepts of cultural dimension
关于⽂文化维度的其它概念
114
Culture and risk attitudes
⽂文化与⻛风险态度
General findings:
Chinese are more risk tolerant than Americans
(or Westerners), but only to the financial domain.
In the social domain, the pattern is reversed.
(Weber & Hsee, 1998 ; Hsee & Weber, 1999; Fan & Xiao,
2006)
Possible Explanations:
1. Cushion hypothesis:
Financial risk of Chinese is mitigated by family and friends.
2. (Lack of) experience with financial products.
115
How to study risk attitudes?
怎样研究⻛风险态度?
Typical question:
After one year you will either get a return of 1500 CHF
or of 0 CHF, both with a 50-50 chance.
How risky (on a scale from 0 to 100) do you rate this investment?
How much would you be maximally willing to pay for participating on this
investment?
116
Perceived risk vs. risk-taking tendency
⻛风险认知 与 冒险倾向
250
Americans
Germans
Polish
0 0
Chinese
Perceived risk Willingness to pay
The results of the study by Hsee & Weber
can be summarized as follows:
Chinese perceive a slightly lower risk and are willing to pay much more.
117
This
supports the cushion hypothesis.
Risk attitudes may depends on how you frame it
When asked about such a decision (in the context of health insurance),
subjects self-reported different frames:
60
Americans
Chinese
0
Gain frame Final wealth frame Loss frame Status-quo frame
However, none of the Chinese in the loss frame bought an insurance, but 71% of
the Americans did!
Americans
Chinese
Possible Explanations:
1. Holistic vs. analytical cognitive style:
East Asians are more focused on the context, whereas
Westerners are not distracted by it.
121
2.3 INTRA: a closer look at
time discounting
International test of
risk attitudes
122
122
INTRA: How to organize a large-scale international survey?
Questionnaire
124
Tendency to wait
Part I. Decisions
I prefer ! A ! B
____________________________________________________________
Depending on his choice, a subject has
“Interest rate” per month would be 11.7%
generally either high or
ences. A fill
Please subject
in thewith lowfor
amount time preferences
which (patient
you consider participant)
the alternatives A aw
attractive: 125
126
Proportion of subjects willing to wait
(red=low, green=high)
Choropleth Map of the World Population
127
Waiting tendency
10 3 RESULTS
$!""#
!,"#
percentage choosing to wait
!+"#
!*"#
!)"#
!("#
!'"#
!&"#
!%"#
!$"#
!""#
-./012345 92:;75 <388;.#=1>?# 9>31# =1>?#=@/7A.# B1C2# B1C2#=@/7A.# 9D/341#
67/834# 90./3412# 90./341#
129
Some theory
130
umption in period t and t + 1, respectively, we can display the possible
Whatever the reason, we have to consider the possibility that people d
umption streams induced by a decision for obtaining A at time t or X
not have the possibility to arrange the money in a free way between the tw
me t + 1.Rational
Figuretime discounting
1 shows the following cases:
time periods, but indeed have to stick to what is offered.
Complete market
In thisIncase, without
the
a perfect restrictions:
choice
market,between
the answerboth
an optionsA
Xamount
follows are only
atthe
with time equally
t and X at tim
formula:
good
t + 1 ifis both lines coincide,
(applying i.e. ifutility
the classical X = model
RA, where R is the market
for intertemporal decisions) th
interest. On the other hand, without a financial market, we would
choice between
have tothe utilitiesthe expected utilities
compare
(a): perfect
market
(b): without
(a) ct (b) ct
market w+A w+A
(c): differing
interest rates ct+1 ct+1
(d): no loans
w+X w+X
0.60
Lithuania
Lithuania
Estonia
Greece
Russia
0.50
Malaysia
Moldova Azerbaijan
China
Nigeria
0.40
Italy
Colombia Croatia
Turkey
Slovenia Mexico
Israel
0.30 Germany Hong Kong
Austria Angola
Canada Ireland
Taiwan Japan
Australia Denmark Czech Rep.
Portugal US UK Sweden
Vietnam Hungary South Korea Chile
Switzerland Argentina
New Zealand
0.20 Norway
Lebanon
Spain
Thailand
0.10
0.00 2.50 5.00 7.50 10.00 12.50
implicit interest rate (1 year horizon) 135
Quasi-hyperbolic discounting parameters
136
Waiting tendency vs.
time pace
Linear Regressi
The waiting tendency is .90 Austria
Germany
•Walking speed
•Postal service speed
•Clock accuracy .50
Greece R Sq Linear = 0.
.40
-6.00 -4.00 -2.00 .00 2.00 4.00
137
time pace
Waiting tendency vs.
time pace
138
GDP per capita vs.
Waiting tendency
Linear Regression
1.00
Switzerland
Germany Norway
Austria
Frage 1 gender weighted native Denmark Canada
.80 Estonia Japan
Hungary Israel
South Korea Ireland
Lebanon Slovenia
USA
China Colombia Malaysia Taiwan
Portugal
Turkey
.60 Lithuania
Moldova Argentina Croatia Australia
Angola Romania Greece
Azerbaijan Italy
Chile Russia
.40
Bosnia.Herzegovina
.20
R Sq Linear = 0.452
Nigeria
.00
*N>40 .00 1.00 2.00 3.00 4.00
logGDPcapita
139
Growth rate vs. Present bias
Linear Regression
1.00 Australia
Lebanon
Denmark
.40
Croatia
Greece
Malaysia Azerbaijan
Chile
Russia Angola
.20
Romania
Lithuania
Estonia
R Sq Linear = 0.351 Bosnia.Herzegovina
Georgia
.00
-2.00 -1.00 .00 1.00 2.00 3.00
140
Ln(growth rate)
Application (1)
Innovation
Macroeconomic view:
‣Paul Romer: technological innovation as
endogenous variable in economic models.
‣ Small time discounting is an important factor.
(Thomas Edison: “Inventions are 1% inspiration
and 99% transpiration.”)
Do time preferences indeed play a role for
innovations?
We can test this empirically with our INTRA data...
141
Tendency to wait vs. Innovation factor
Linear Regression
6.00
USA
Japan Switzerland
Germany
Taiwan Denmark
Israel
Austria
South Korea
5.00 UK Canada
Norway
Ireland Hong Kong
New Zealand
Spain China Slovenia
Chile Italy
Portugal Estonia
4.00
Thailand
Lithuania
Nigeria Azerbaijan Croatia
Russia Turkey Hungary
Greece Mexico Colombia
Romania
Argentina
Georgia
3.00
Bosnia.Herzegovina Moldova
R Sq Linear = 0.442
2.00
.00 .20 .40 .60 .80 1.00 142
Percentage choosing to wait
Regression results
143
Application (2)
Environmental Protection
144
Environmental Sustainability
145
Discussion
146
UP Time
147
3.5
Fig 2. Bubble plots of 20 various variables and universal preferences for time (UP time, on the
3.0
x-axis). In most cases, we can see a clear dependence. The sizes of the bubbles correspond to the
0 2.5
weight
-1.5 of the
-1.0 country
-0.5 data.
0.0 Weighted
0.5 regression
1.0 1.5 lines
-1.5 in red.
-1.0 -0.5 0.0 0.5 1.0 1.5
UP Time results
Gasoline price per liter Body mass index
Average
2.5credit rating Innovation
30 factor
6.0
2.0
100 28
5.5
1.5
80 26 5.0
1.0
60 24 4.5
4.0
40
0.5 22
3.5
20
0.0 20
3.0
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
0 2.5
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
Resilience index
Average years in school
Gasoline14price per liter 110 mass index
Body
2.5 100 30
90
12
80 28
2.0
10 70
60
1.5 26
8 50
40
1.0 24
6 30
20 148
0.5 22
4 10
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
2.4 INTRA: risk preferences
and culture
149
6 3 METHODOLOGY
were translated into the local languages following the same procedure as the
INTRA: risk preferences and culture
questionnaires and were then read aloud by the local lecturer.
How
We to measure
measured risk risk preferences?
preferences by eliciting the participants’ willingness
to pay for hypothetical lotteries. Six of the lotteries were solely about gains
Standard method: hypothetical lotteries, e.g.
with di↵erent probabilities (where one of the lotteries had a very high gain),
two were solely about losses, and two mixed-outcome lotteries were included
to elicit loss aversion (not reported in this paper). The gain lotteries were
formulated similar to the following:2
Imagine you are o↵ered the lotteries below. Please indicate the maximum
amount you are willing to pay for the lottery:
The following lotteries involve losses. Imagine you have to play these lotteries,
unless you pay a certain amount of money beforehand. What is the maximum
amount you would be willing to pay to avoid playing the lottery? This corresponds
to buying insurance that saves you from su↵ering potential losses.
Table study
For the INTRA 1: Payo↵s
the for the lotteries
following in our
lotteries survey.
have been used:
Lottery Outcome A($) Prob(A) Outcome B($) Prob(B) Expected Value ($)
1 10 0.1 100 0.9 91
2 0 0.4 100 0.6 60
3 0 0.1 100 0.9 90
4 0 0.4 10,000 0.6 6000
5 0 0.9 100 0.1 10
6 0 0.4 400 0.6 240
7 0 0.4 80 0.6 48
8 0 0.4 100 0.6 60
153
INTRA: risk preferences and culture
154
INTRA: risk preferences and culture
155
Violations of internatlity and data quality
The data strongly suggests to estimate different RRP for gain and
loss lotteries – consistent with CPT (confirmed by factor analysis).
158
Are there country differences?
159
General findings: risk aversion in gains, risk seeking losses
160
Gender differences
Wealth effects
Table 3: Regression results for risk preferences in gain lotteries.
163
Cultural dimensions:
Table 3: IDV
Regression results for risk preferences in gain lotteries.
166
Cultural dimensions:
Table 3: UAI
Regression results for risk preferences in gain lotteries.
169
2.5 INTRA: CPT parameters
and probability weighting
170
ght to be probability
n-linear larger than one) that leads
weighting to a steeper
function w. value function in losses than in
. Fourth, small probabilities
this article, we present =have
(1 −overly
C P T results for w+CPTp))v(A)
(large + wclassical
decision
with the + ( p)v(B),
weight, as expressed by
specification of
on-linear probability weighting function w.
ky and Kahneman (1992). The model is given for lotteries in gains as follows:
dthis article,
forINTRA: CPTwe
lotteries in present
parameters
lossesandresults forweighting
probability
as: CPT with the classical specification of
sky and Kahneman P T(1992).
= (1 − The
C specification:w+model is given
( p))v(A) + wfor lotteries in gains as follows:
+ ( p)v(B),
Model
C P T = (1 − w− ( p))v(B)
C P T = (1 − w+ ( p))v(A)
+ w− ( p)v(A),
+ w+ ( p)v(B), (in gains)
r lotteries in losses as:
here A andinBlosses
or lotteries are the
as: payoffs of the lottery ( A < B), p is the probability to o
C P T = (1 − w− ( p))v(B) + w− ( p)v(A), (in losses)
e higher outcome B, w± are the probability weighting functions in gains/losse
C P T = (1 − w− ( p))v(B) + w− ( p)v(A), 3
sAthe value
and BValuefunction.
are the w ± and v are given by:
payoffs of the lottery (A < B), p is the probability toM.obtain
function: O. Rieger et a
gher outcome !
e A and B areB, ± are the
thewpayoffs ofprobability
the lotteryx α(weighting
,A3 < B), pfunctions
isx the 0 in gains/losses,
≥ probability and
to obtain
igher outcome
e value B, w
function. w±±are
andthe are :=
v probability
v(x) given by: weighting functions in gains/losses, and
where
he > 0 is called
valueλ function. w and −λ(−x)
the “loss-aversion”
v are given by:
β , x and
3 coefficient, < 0,α, β > 0 (usually α, β ≤ 1
± ! α
describe the risk attitudes for! gains
x , and losses.x ≥0
The standard v(x) :=function
weighting x α,
is β , xx ≥ 00, (1)
Probability v(x)
weighting
:= −λ(−x)
function: < (1)not focus
We tried different specifications of w−λ(−x)
± and v βand
, also
x < alternative
0, PT models, but we do
he current article, see Rieger and Bui (2011) for details.
pγ
w ( p) :=
ied different specifications of w± and± v and also alternative
γ + (1 − PTp) γ )1/γbut we do not focus on this
models,
(2
ried different
( p
urrent article, specifications
see Rieger and ofBui and v and
w± (2011) foralso alternative PT models, but we do not focus on this
details.
current article, see Rieger and Bui (2011) for details.
with the parameter γ = γ± ∈ (0, 1] describing
123
171
172
Reflection effect
*In Tanzania, only 57% showed a reflection effect. Sample size in Tanzania, however,
was small (N=47).
173
Probability weighting
174
Factor analysis
175
Finding CPT parameters
Compare Table 5.
178
Key findings of parameter estimates
179
Key findings of parameter estimates
180
Culture?
181
Estimating cumulative prospect theory parameters…
Table 7 Regression results on pairs of countries for differences between countries in CPT value versus
Culture?
cultural and economic differences
Model 1 Model 2
*** Significant on the 1% level, ** Significant on the 5% level, * significant on the 10% level
‣There
probability is a significant
weighting. relation
12 We obtain between
a correlation these
coefficient measures,
of 81.0% with p < 0.001.
i.e., that
This implies countries that are of
the measurement very different
λ, indeed, with the
captures respect toof
concept wealth
loss aversion,
and/or
and is not drivencultural factors,errors
by measurement tendsofto have
β and γ−also
. very different
CPT
By now, weparameters andthat
have established vicein versa.
most countries, our estimates are robust and
that country differences are significant. The question arises whether these differences
182
are related to economic or cultural differences between these countries. While this has
Estimating cumulative prospect theory parameters…
A closer look at probability weighting
Consider two binary lotteries A and B, both with outcomes 0 and X > 0. Lottery
A gives a payoff of X with probability pA and lottery B with probability pB . If a
decision maker can be described by CPT, but does not have any probability weighting,
his certainty equivalences for the two lotteries (C E A and C E B ) have to have a fixed
ratio, as the following computation shows, where v is the value function:
By definition of the certainty equivalence, we have
u (C E A ) u (C E B )
= .
pA pB
u (C E A ) pA
= .
u (C E B ) pB 183
(CEEAA)) u u(C(CEE
uu(C B )B )
== . .
ppAA ppBB
Normative rule without probability weighting
Rewritingthis,
Rewriting this, we
we find
(CEEAA) )= pp
uu(C AA
= ..
uu(C
(CEEB ) ) ppB
B B
Weiscan
a normative rule that
apply this general everybody
result to three ofwho does not
our lotteries show2, 3, and 5) with
(lottery
We probability
can apply this general result to three of our lotteries (lottery 2, 3, and 5) with
X = 100 and p2 =weighting should
0.6, p3 = 0.9, and follow.
p5 = 0.1. This leaves us with three conditions
X = 100‣If and p2find= 0.6, p3 = 0.9, and = 0.1. This
p5 means leaves us with three conditions
that need towe > instead
be satisfied of =,
if a decision that
maker doesthat a person
not have probability weighting.
that need overweights
Moreover, toone
becan
satisfied if aoverweighting
showsmall
that decision maker
probabilities. does not
of smaller have probability
probabilities occurs if,weighting.
in our
Moreover,
case, one one
of thecan show that
following overweighting
conditions is of smaller probabilities occurs if, in our
satisfied:
‣ In the case of the INTRA lotteries we obtain three
case, one of the following conditions is satisfied:
inequalities:
u (C E 5 ) 1
u (C E 5 )> ,
1 (5)
u (C E 2 ) > 6 , (5)
uu(C
(CEE2 )2 ) 26
> , (6)
u (C E
u (C E 3 )2 ) 3 2
> , (6)
uu(C
(CEE5 )3 ) 13
> . (7)
uu(C
(CEE3 )5 ) 91 184
> . (7)
Empirical results on probability weighting:
M. O. Rieger et al.
percentage of subjects
185
90%
Fig. 2 Proportion of subjects showing probability overweighting in gain lotteries in the respective three
tested probability ranges or in at least one of them. We see that most probability overweighting took place
in small probabilities, as CPT would predict
Correlations between different measurements of probability
Table 8 Correlations between different probability weighting measurements
weighting
Probability weighting
γ+ γ− (10%/60%) (10%/90%) (60%/90%) (any)
*** Significant on the 1% level, ** significant on the 5% level, * significant on the 10% level
186
Probability weighting, cultural and economic factors
Probability weighting,
Table 9 Ranked cultural
regression of γand economic factors
+ with cultural and economic factors (country level)
Ranked γ+
Model 1 Model 2 Model 3 Model 4
PDI 0.178
(1.29)
IDV −0.423***
(−3.42)
MAS 0.044
(0.31)
UAI 0.339**
(2.68)
log(GDP/capita) −0.379*** −0.336*** −0.425*** −0.364***
(−2.74) (−2.76) (−2.97) (−2.88)
N 48 48 48 48
Adj. R 2 in % 18.9 33.5 16.1 27.3
*** Significant on the 1% level, ** significant on the 5% level, * significant on the 10% level 188
*** Significant on the 1% level, ** significant on the 5% level, * significant on the 10% level
Table 10 Ranked regression of the proportion of subjects that showed probability overweighting with
Probability weighting, cultural and economic factors
cultural and economic factors (country level)
PDI −0.121
(−0.83)
IDV 0.244*
(1.77)
MAS −0.160
(−1.09)
UAI −0.312**
(−2.33)
log(GDP/capita) 0.307** 0.289** 0.288* 0.278**
(2.10) (2.09) (1.96) (2.07)
N 48 48 48 48
Adj. R 2 in % 9.7 14.2 10.7 18.0
*** Significant on the 1% level, ** significant on the 5% level, * significant on the 10% level 189
M. O. Rieger et al.
Table 11 Logit regression results on individual level for non-parametric probability weighting. (Wald
Probability
scoresweighting, cultural and economic factors
in parentheses.)
Table 12 Logit regression results on individual level for non-parametric probability weighting for different
Probability weighting,
probability ranges.cultural andineconomic
(Wald scores parentheses.)factors
192
2.6 INTRA: loss aversion
193
1 4 Journal of Behavioral Decision Making
2
3Eliciting lossthe lottery) was therefore zero by definition. A similar formu-
aversion Survey
4 lation was used in Tversky and Kahneman (1992) (p. 312). As men
5 Question used:formulations were as follows:
The precise sion in
6 In the following lotteries you have a 50% chance of of Risk
7 winning or losing money. The potential loss is given. Please of Zuri
8 state the minimum amount $X for which you would be was tran
9 willing to accept the lottery. or trans
10 the amo
11 50% chance loss of $25 purchas
12 50% chance win of $X come a
13 X should be at least $____ to make the lottery acceptable. Hofs
14 mogeno
15 50% chance loss of $100 control
16 50% chance win of $Y chose u
17 ses at r
18 Y should be at least $____ to make the lottery acceptable. first yea
19 The loss aversion parameter θ has been defined by194 ground
34 1 wealth and
35 for the smaller stake is significantly smaller than the one the individ
36 for the larger stake (2.81 vs. 2.96, p<.001). The magnitude level GDP
37Eliciting of this difference, however, is relatively small. Moreover,
loss aversion
from the W
38 Cronbach’s α of the two measures is .73, demonstrating good use countr
39 reliability.
Since Therefore,ofour
the estimation loss aversion
lambda parameter
is numerically θ is (as
difficult de- we World Fac
40 have as theand
finedseen) average of the not
therefore tworobust
subindices
enough to be used for
41 individual level data, we use instead theta:
! "
42 1 X Y
43 θ :¼ þ :
2 25 100
44 Data prep
45 Survey dat
The In resulting
later values we
sections, show significant,
report results and nottojust
related the random,
aggre-
46 differences betweenparameter
cultural clusters: this can b
47 gated loss aversion θ. When we repeat the analy- bounded) a
When
ses withcombining
θ1 and θsimilar
2, the countries
results show intovery
cultural clusters
similar (e.g.,
patterns
48 of loss ave
Anglo-American
(not reported here). countries and Eastern European countries,
49 when analy
based To on Chhokar,
measure the Brodbeck, and House,
Hofstede cultural 2008),19
dimensions, between-
ques-
50 to deal wi
cluster
tions variations
based on the are large,VSM94
Hofstede as an ANOVA test at(Hofstede
questionnaire the level of
51 cultural clusters reveals (F(8, 6683) =survey.
16.44, Demographic
p <.001). highest and
52 & McCrae, 2004) were included in the ranked reg
53 information (gender and age) and information about personal the impact
195
M
sio
tes
bo
As
ate
hy
an
* hy
M
of
wh
196
*There are large differences within some clusters, especially the Asian one! so
38 clusterTable
culture dummy variables on culture cluster dummy variables
2. Regression 103
)
39 104
Loss aversion (Ranked θ)
40
Regression analysis 105
Model412 Model 1on ranked theta
Regressions Model 3 Model 106
Model 12
42 107
!239.1***(3.81) !214.6***(3.55)
6***(3.55) Gender (1 = male) !212.5***(3.48) !239.1***(3.81)
43 Age !10.0 (1.00) 108
!4.6(.45)
(1.00)
!4.6(.45) !4.2 (.39) !10.0
44
2***(2.61) –.2***(2.69)
Risk avers. gain –.2***(2.59) 109
–.2***(2.61)
–.2***(2.69)
45 !33.1 (.44)
ln(GDP/cap) (.15) 25.0 !33.1110(.44)
31.346(.59) African !178.7 (.71) !131.3111(.59)
7**47(2.15) Anglo-American !298.1**(2.08) !311.7** (2.15)
!132.6
112(.53)
32.648(.53) Latin American !151.2 (.56)
92.2(.48) Latin European 110.8 (.57)
113
92.2(.48)
49
2***(2.83) Eastern European 447.0 ***(2.58) 114
460.2***(2.83)
72.550(1.47) Germanic/Nordic 186.1(1.58) 172.5
115(1.47)
07.051(.82) Middle Eastern !127.5 (.80) !107.0116(.82)
41.352(.23) Asian 21.7(.10) 41.3 (.23)
3828.0***(9.99) 117
3451.1***(12.10)
1***(12.10) Constant 3440.0*** (7.69) 3828.0***(9.99)
53 6177 118
6177
6177 N 6177 6177
1.9254 R2(%) .50 1.93 119
1.92
.50 197
55
2 67
3 Table 3. Regression analysis on cultural dimensions
68 and religions
4 Loss aversion (Ranked θ) 69
5 70
Regression
Model 3 analysis
6 Model 4 ModelModel
5 1 71 Model 2 Model
!182.5***(3.31)7 Gender (2.87)
!185.9*** !239.1*** (3.81) 72 !195.7***(3.03)
(1 = male) !170.4***(2.80) !182.5***(
!9.2(1.28)8 Age!3.4 (.35) !10.0(1.00) 73
!7.8(.96) !9.7(.92) !9.2(
–.2***(8.09)9 Risk avers. gain
–.2***(3.26) –.2***
–.1***(4.23) 74 –.1***(3.34) –.2***(
.8***(2.87)10 Risk avers. loss
.8***(2.91) .7***(2.69) 75
.8***(3.30) .8***(2.67) .8***(
39.7(.52) 11 log 43.5(.67)
GDP/cap 48.0(.74)
!33.1(.44) 32.6(.50) 39.7(
log 48.9(.35)
growth rate 76 3.8(.03) 37.9(
37.9(.24) 78.8(.58)
!15.4(.11)
10.8*(1.96)
12 IDV8.6*(1.75)
average 8.7*(1.79) 77 10.8*(
1.4**(2.32)13 IDV ind. di. 1.4**(2.26) 78 1.4**(
7.2(1.50)14 PDI average
6.6(1.26) 5.8(1.13) 79 7.2(
1.6**(2.41)15 PDI ind. di. 1.5**(2.34) 80 1.6**(
8.0***(3.43)16 MAS average
7.0***(2.00) 7.0***(3.01) 8.0***(
MAS ind. diff. 81 .2(
.2(.91) .2(.91)
1.5 (.30)
17 UAI7.0 average
(1.47) 7.9*(1.74) 82 1.5 (
.4(.99) 18 UAI ind. diff. .3(.95) 83 .4(
19 Protestant
476.6(1.24) 421.0*(1.80) 84 !194.7(.47)
20 Catholic
!80.5 (.31) !131.5 (.51) 85 !191.2 (.47)
21 Orthodox
533.7*(1.88) 490.9* (1.80) 738.1***(2.67)
Muslim 86 !618:0(1.63)
!372.6(1.06) !397.6(1.14)
22 Jewish (0.98)
!247.9 !273.4(1.16) 87 6.7(.02)
23 Buddhist
299.3(1.04) 255.9(0.94) 88 Religions65.3(0.17)
are country
1938.5**(2.66)24 Constant
1635.8**(2.42) 3827.9***(9.99) 89 average
1691.8**(2.57) 3640.5***(9.62)
data! 1938.5**(
5801 25 N 6146 58016177 90 6177 5801
2
2.60 26 R (%)
3.01 3.25 .50 1.80 198
2.60
91
Impact of religion: recent evidence
‣ Key results:
• Individualism, power distance, and masculinity
increase loss aversion.
• Religion also might have an influence.
• In comparison, the connection of loss aversion to
macroeconomic variables seems to be much smaller.
200
2.7 Framing and culture
201
Holistic vs. Analytic
Thinking Style
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 205
Example
Imagine that you are about to purchase a jacket for 125 Euro,
and a calculator for 15 Euro. The calculator salesman informs
you that the calculator you wish to buy is on sale for 10 Euro at
another branch of the store, located 20 minutes-drive-away.
Would you make the trip to the other store?
Imagine that you are about to purchase a jacket for 15 Euro, and
a calculator for 125 Euro. The calculator salesman informs you
that the calculator you wish to buy is on sale for 120 Euro at
another branch of the store, located 20 minutes drive away.
Would you make the trip to the other store?
‣ An analytic thinker might only focus on the most crucial
information and notice that it’s only about 5 EUR difference.
‣ A holistic thinker might integrate all information and notice that
it’s about spending 135 EUR or 130 EUR.
‣ Framing effect only occurs “in the middle” between these
extremes. That’s why thinking style does not influence the
framing effect much.
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 207
Illustration
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 208
Example
Recall the lost vs. found scenario of Levinson and Peng (2006).
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 210
Survey
‣ We conducted an online survey with N=296 undergraduates
(146 in Germany, 150 in Vietnam).
‣ The students were randomly assigned to groups A or B.
(Pretests had shown that within-person measures of framing
effect fail.)
‣ We asked four framing questions in different frames.
‣ Three gain/loss questions (A: gain/loss/gain frame, B: loss/
gain/loss frame) and one purchasing framing question (the
above mentioned calculator & jacket question by Tversky and
Kahneman, 1981).
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 211
Measuring
Holistic Thinking Style
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 212
Country Results
15% effect
Framing 22.6% 21.1% 27.6% 8.0%
Vietnamese sample (n=151)
Gain frame 58.7% 44.7% 50.7% 48.0%
7.5%
Loss frame 53.9% 37.3% 30.3% 34.2%
Framing effect 4.7% 7.4% 20.4% 13.8%
0%
Difference between Germans
17.9% 13.7% 7.2% -5.8%
andOliver
Marc Vietnamese
Rieger, Professor of Banking and Finance, University of Trier 213
Gain/loss 1 Gain/loss 2 Gain/loss 3 Purchasing
A Note on Significance
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 214
Holistic vs. Analytic
When sorting by high/low AHS or categorization, the
holistic thinking style groups showed less average
framing effect:
Categorization AHS
German sample (n=145)
analytic 20.7% 22.4%
holistic 17.5% 12.6%
Vietnamese sample (n=151)
analytic 16.7% 14.7%
holistic 9.6% 9.3%
Total sample (n=296)
analytic 18.7% 18.5%
holistic 13.5% 10.9%
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 215
Statistical Significance
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 216
Summary so far
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 217
Demographic Effects
on Framing
In both countries, males and business administration students
showed less framing than others. How long they were studying
did not matter; the effect of age differed between Vietnam and
Germany (small sample size of young students in Germany!).
gender age years studying major
N 93 55 80 71 87 64 112 39
N 70 75 11 134 91 54 109 36
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 218
Chapter 3
How Culture Affects
Game Theory
Behavioral game theory
220
V/76+-5$> 0LMOL2F D+1*%$E +F 4&+%+,)& <='+*>@
1 C 2 C 1 C 2 C 1 C 2 C
100
S S S S S S 100
1 0 98 97 99 98
1 3 98 100 99 101
222
Play a beauty contest game
223
Beauty contest result
Beauty Conest Game -Fall Semester 2008
5
Frequency
0
0 6 10 16 20 26 30 36 40 46 50 More
Average=19.82
Target number=13.21
Winning number =12
224
What do we learn?
225
Games as social allocations (1)
Ultimatum game
226
Games as social allocations (2)
227
Experiments on ultimatum
games
228
Ultimatum bargaining
across cultures (1)
229
Ultimatum bargaining
across cultures (2)
230
Impacts of market integration
and payoff cooperation
Market integration:
How much do people rely on market exchange in their daily lives?
Payoff to cooperation: How important and how large is a group’s payoff from cooperation in
economic production?
231
Chapter 4
A Closer Look at
East Asian Culture
4.1 East Asia: Background
233
East Asia: Geography
234
East Asia: History*
*Sorry for gross simplifications and omissions and for still being a bit lengthy. To put a few thousand years in a few slides is hard!
China:
‣ Area with many ancient cultures, ancient writing system
‣ First unification of the core area in 221 BC by Qin (“China”)
‣ Expansion, mostly to West and South until ca. 760
‣ Strong influence on neighboring countries and Japan
around that time, while itself being influenced by buddhism
‣ More focus on the Eastern and Southern part of nowadays
China from ca. 760 to the Mongol invasion
‣ 1276 the Mongols conquered all of China (until 1370)
‣ During their rule, Marco Polo visited China
‣ After the Mongols, the Great Wall was built
‣ Later, China was ruled again by a Northern tribe (the
Manchu) until 1911 235
East Asia: History
237
East Asia: History
Japan:
‣ Strong cultural influence from China, esp. around 400-700
‣ Writing, arts, science, religion imported from China
‣ Basically completely closed from around 1500-1850
‣ US forced an opening that led to political reforms in Japan
‣ Rapid adaption of Western technology and governmental
system from 1868 onwards.
‣ Conquest of various islands (including Taiwan) and Korea
and Mandchuria.
‣ Large scale nationalism, collaboration with Nazi Germany
and start of the war with China (1938) and the US (1941).
‣ After the nuclear bombing of Hiroshima and Nagasaki defeat
against the US, loss of all conquered territories.
‣ Since 1945 democracy with a solely representative emperor.
238
East Asia: History
Korea:
‣ On the intersection between China and Japan, often
depending on China or even conquered.
‣ Invention of a unique alphabet with 24 letters, replacing the
Chinese characters in the 15th century.
‣ Occupied by Japan in 1910, after the war, the North was
occupied by the Sowjet Union and became communist.
‣ In 1950, the North attacked the South. UN troops were
fighting back and conquering the North. China intervened
unofficially with “volunteer troops”. The war ended 1953 with
a draw.
‣ North Korea is one of the least free countries in the world.
‣ South Korea is a flourishing, wealthy democracy.
239
East Asia: History
Taiwan:
‣ Inhabited by pacific tribes (similar to the South sea)
‣ Colony of the Netherlands (1622-1662)
‣ Conquered by the fleeing Ming armies of China (1662),
afterwards Chinese settlers
‣ Conquered by the Japanese in 1895, part of Japanese
empire
‣ After World War II, Japan had to return Taiwan to China
‣ In 1949, Chinese fleeing from the communists flood Taiwan,
Taiwan becomes the sole land of the “Republic of China”.
‣ Until 1970s, representing China internationally, afterwards
international isolation.
‣ Democratisation in the 1990s, now fully working democracy
and relatively wealthy country, but still under threat of China.
East Asia: History
Vietnam:
‣ Strictly spoken not East Asia, but South East Asia.
‣ In its long history often partially under Chinese rule or influence.
‣ Other influences from the South (Thailand, Cambodia, India).
‣ French colony 1858-1954, 1945 briefly occupied by Japan.
‣ After independence war, separated into Northern (communist)
and Southern Vietnam.
‣ After invasions of Northern troops into the South, the USA got
more and more involved into the war (Vietnam war) with
atrocities on both sides.
‣ When the war was no longer acceptable for US citizens, the US
withdrew. Not long afterwards, Northern Vietnam occupied the
South and unified the country under communist rule.
‣ Wars with Laos, Cambodia and China, later economic reforms.
Economic development of
East Asian countries
242
4.2 Why do we Study East Asia?
Rich* East Asian culture
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 243
Some special points about
East Asian Cultures
245
4.3 The PANDA study
General motivation:
After the completion of our global INTRA
study (International Test on Risk Attitudes)
with nearly 7000 subjects in 53 countries, we
broaden the view to more diverse economic
decisions, but focus on a particularly
interesting region of the world: East Asia.
The surveys in this study are being
conducted in Taiwan, China, Hong Kong,
Japan and Vietnam, but also in Germany,
Estonia and India (for comparison).
246
PANDA study
‣ Partially classroom (non-incentivized), partially online
(semi-incentivized), partially lab (incentivized)
‣ Countries in East Asia (China, Hong Kong, Japan,
Taiwan, Vietnam), Europe (Estonia, Germany) etc.
‣ Topics include…
247
Innovation
Learning Creativity
Probabilities Trust
Economics
Culture
Updating beliefs Market economy
Social redistribution
Main research questions of
PANDA (1)
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 249
Main research questions of
PANDA (2)
Marc Oliver Rieger, Professor of Banking and Finance, University of Trier 250
Main research questions of
PANDA (3)
252
4.4 Social Difference
Acceptance
‣ Online survey (N=1621)
‣ Several items on fairness and social differences:
1. Only if differences in income and social standing
are large enough is there an incentive for individual
effort.
2. Differences in social standing between people are
acceptable because they basically reflect what
people made out of their opportunities they had.
3. All in all, I think social differences in my country are
justified.
4. Generally speaking, business profits are distributed
fairly in this country.
SDA index computed based on agreement to these items
(Cronbach’s Alpha 0.73). 253
Social Difference Acceptance
255
Two qestions
256
Discussion
257
Government productivity reform
Suppose the government wants to undertake a reform to improve the
productivity of the economy. As a result, everyone will be better off, but the
improvement in life will not affect people equally. A million people (people
who respond energetically to the incentives in the plan and people with
certain skills) will see their incomes triple while everyone else will see only a
tiny income increase, about 1 %. Would you support the plan?
Government reform / unequal distribution
90%
68%
45%
23%
0%
Estonia Taiwan China Vietnam Germany Japan
No Yes
258
Flower Price Increase by Country
On a holiday, when there is a great demand for flowers, their prices
usually go up. Is it fair for flower sellers to raise their prices like this?
75%
50%
25%
0%
Estonia Taiwan China Vietnam Germany Japan
No Yes
259
Fairness and the state
260
Flower Price Limits by Country
Should the government introduce limits on the increase in
prices of flowers, even if it might produce shortage of flowers?
Acceptance for Flower Price Limits
90%
68%
45%
23%
0%
Estonia Taiwan China Vietnam Germany Japan
No Yes
261
4.5 Perception of stocks
262
Online survey
263
Stock market investments - real
Have you ever invested in stocks or funds?
100%
75%
50%
25%
0%
Estonia Taiwan China Vietnam Germany Japan
No Yes
264
Stock market investments - potential
60%
40%
20%
0%
Estonia Taiwan China Vietnam Germany Japan 265
266
Differences in attitudes towards stocks
267
Social Inequality Aversion (SIA)
268
Skepticism About Fairness (SAF)
269
Regressions for potential investment
270
Regression results
272
Some preliminary results on cultural differences in fairness
perception…
273
4.6 Information processing,
critical thinking and fake news
275
Example
277
Results
China
USA
Elsewhere
the extent of conspiracy theories: subjects think that it is most likely that the virus
20%
10%
20% 20%
10%
5%
10% 10%
Additional discussion 0%
0%
0% 0% 2 3 4 5 6 7 8 9
And 24 a3 comparison
here 5 4 65 67 78 8 9 9 ≥10
between10 Germans4 5 and
6 Chinese
7 8 9 in≥10
Germans Chinese
Germany, demonstrating
Germans
Germans Chinese how politics influences
Germans these
Chinesebeliefs:
Germans Chinese
East Europe: another example of the
interaction between culture and politics
282
Overview: Communist and
Western Europe in 1989
283
Culture and Institutions
Culture: “those customary beliefs and values that ethnic, religious,
and social groups transmit fairly unchanged from generation to
generation.” (Guiso et al.,2006, p.23)
285
Communist Institutions and
Preferences
288
Descriptive Overview of Risk and
Time Preferences
communistic effects?
OLS-Regression –
Post-Communist Dummy
291
OLS-Regression – Covariates
(ln) Highest rate of unemployment 0.0152 0.00648 0.00764 0.00367 0.118 0.0759
between 1990 and 1999 (0.56) (0.23) (0.26) (0.11) (1.57) (0.94)
Controls + + + + + +
293
Results: Time Preferences
(OLS-Regression)
Beta (β) Delta (δ) Waiting Tendency
Model 1 Model 2 Model 1 Model 2 Model 1 Model 2
(ln) Highest rate of unemployment 0.106** 0.164*** 0.00433 0.0129 -0.0396 -0.0266
between 1990 and 1999 (2.61) (3.58) (0.34) (0.96) (-1.32) (-0.80)
Controls + + + + + +
294
Possible Causes for Differences
in Preferences
1) Did communist systems effectively shape the risk and time preferences of
individuals?
296
Convergence in Europe
297
Chapter 5
Cultural Finance
on the Market Level
5.1 Is culture relevant to
finance?
299
Overview
300
Example 1: Momentum (Griffi, Ji and Martin (2003))
(Griffi, 301
Ji and Martin, 2003)
Trend chasing around the world
1.8
1.4
0.9
Europe
Profit of
momentum 0.5
Asia
0.0
-0.5
0 50 100 IDV
Profitability of momentum strategies (trend chasing) depends on Hofstede index of individualism (Chui, Titman & Wei, 2005).
Control for institutional differences (La Porta et al.).
302
Example 2: Equity premium and ambiguity aversion
(Rieger & Wang, Finance Research Letters, 2012)
303
B. a payment of X $ in 10 years from now
X has to be at least _____ $, such that B is as attractive as A.
__________________________________________________________________________
Measure of Ambiguity Aversion
Question 4. In an urn, there are 100 balls with three colors (red, yellow, and blue).
30 balls are red, whereas the remaining 70 consist of yellow and blue balls.
30 balls 70 balls
red yellow blue
Imagine a ball is randomly drawn from the urn. You are offered the following two
Lotteries. Which lottery would you prefer?
A. If the color of this ball is red, you win 100 $; otherwise you win nothing.
B. If the color of this ball is yellow, you win 100 $; otherwise you win nothing.
I prefer _ A _B 304
Example 3: Stock market participation and trust
305
Discussion
306
5.2 Dividend payout policy
Key result:
Strong dependence of average dividend payout ratios of
companies in 46 countries with the INTRA survey values
of patience, loss aversion and ambiguity aversion.
(Breuer, Rieger, Soypak, Journal of Banking and
Finance, 2014)
307
Behavioral Decision Theory
308
Behavioral Decision Theory
309
Example: dividend payout policy
Behavioral model:
‣Investors like dividends, because they can spend them
without psychological disutility.
‣Firms pay dividends because investors demand them.
‣Myopic, loss averse and ambiguity averse investors
prefer more dividends, as can be shown with a
mathematical model.
310
ginal work, which focuses on the possible influence of wealth
transfer among accounts on household savings, we take a closer
look at the subjective perception of these simple wealth transfers
Example: dividend depending
by investors payout policy on their preferences.
Current Future
asset S2 income
S1
account account
x1
Current
d1 income
account
312
e idend policy d1 that maximizes the (representative) investor’s
l overall utility U which is computed as the sum of utility
d uðd1 !
ðRÞ
and uðS
d1 Þ payout 1 !
ðRÞ
in the first period and the subjectively
S1 Þ
Example: dividend policy
e R1 ðRÞ
discounted expected utility ð2 ! dÞ " 0 uðS2 ! S2 Þf ðr1 Þdr 1 þ
- The
R 0 expectedðRÞ
utility of an investor receiving a dividend d1
y d " can uðScomputed
!1 be 2
Þdr 1 of the second period:
! S2 Þf ðr1as:
e ðRÞ
Uðd1 Þ : ¼ uðd1 ! d1 Þ þ uðS1 ! S1 Þ
ðRÞ
- Z 1 Z 0
ðRÞ ðRÞ
- þ b " ð2 ! dÞ " uðS2 ! S2 Þf ðr 1 Þdr 1 þ b " d " uðS2 ! S2 Þf ðr 1 Þdr 1 : ð1Þ
0 !1
e
e In (1), the index ‘‘(R)’’ denotes reference values to distinguish
- Here the
between index
gains ‘‘(R)’’
and denotes
losses, reference
i.e. for x e {d, S}values to
we assume
- distinguish between
8 gains and losses.
ðRÞ a
þ
<
𝜆,
ðRÞ
g u is a ðRÞ ðx ! x
CPT-valuet function
t Þ x P x
with losst aversion
t ;
uðxt ! xt Þ ¼ ð2Þ
- : !k " ðxðRÞ ! x Þa! x < xðRÞ ;
𝛽 is the quasi-hyperbolic
t time
t discounting
t t parameter.
- 𝛿 (between
with k being a1loss
andaversion
2) is an ambiguity
parameteraversion
typicallyparameter.
greater than 1 and
e a+ as well as a! being variables that determine the curvature
313
of u
the future incom
e idend corporate
policy dividend
d1 that policymaximizes
in a more formalthe (representative)
manner. However, they investor’s
assume a value function that is not completely in line with pros-
l overallpectutility U which is computed as the sum of 2 utility
ðRÞ
S 2 ! S ¼ ðx1 ! d
theory, they refrain from taking ambiguity aversion into ac-
d uðd ! count
d
ðRÞ and they do not ðRÞ
1 dividendÞ and uðS 1 ! S distinguish between different mental
Þ in the first period and the The future inc
subjectively
Example: 1
accountspayout policy
for dividends 1 asset holdings which are at the core
and (1)). In addition, w
e R1 ðRÞ
discounted
of the generalexpected
ideas of Shefrin utility
and Thaler ð2 (1988).
! dÞIn"what0
uðS
follows,
2 ! S 2
ever,
Þf as
ðr 1 a genera
Þdr 1þ
- The
R 0 we expected
mainly attempt
ðRÞ
utility
to of an the
depict investor
approachreceiving
of Shefrin anda dividend
Thaler d
mally1 model am
y d " can uðScomputed
(1988)
!1 be 2
! aSmore
in 2 Þf ðr Þdr 1 offramework.
quantitative
as:
1 the second period:
As opposed to their ori- mainly interested
ginal work, which focuses on the possible influence of wealth one main conseq
e transfer
Uðd1 Þ : ¼ uðd1 ! among
d1 Þ accounts
ðRÞ
þ uðS1 !on S1household
ðRÞ
Þ savings, we take a closer evaluating a futu
- look at the subjective
Z 1 perception of these simple wealth Z 0 transfers ambiguity averse
by investors depending on their ðRÞ preferences. ðRÞ reducing the ove
- þ b " ð2 ! dÞ " uðS2 ! S2 Þf ðr 1 Þdr 1 þ b " d " uðS2 ! S2 Þfisðrdisturbing
1 Þdr 1 : ð1Þfor a
0 !1
e guity aversion by
and 2 where for d
e In (1), the index ‘‘(R)’’ denotes reference values to distinguish ity aversion is in
- Here the
between index
gains and ‘‘(R)’’ denotes
losses, i.e. for
Current
reference
x {d, S}values
e we assume to
Future subjectively disco
316
to per- b. As the investor should be willing to hold the asset until t =
endent for high enough probabilities of positive rates of returns, o
Example: dividend payout policy
me t = 1 dividends d1 should be decreasing in the investor’s patience
theless, We canDependence
also prove that on d:this term just
Again, is positive:
the last two summands of g d
ned by on d,
Z 1
values @g
d levels ¼ b % u 0
ððx1 $ d1 Þ % r 1 Þ % r1 % f ðr 1 Þdr 1
@d 0
otential Z 0
as long $b% u0 ððx1 $ d1 Þ % r1 Þ % r1 % f ðr1 Þdr1 :
at time $1
318
Idea of empirical test
This table presents the results of various robustness tests, including results with320industry dum
four large countries and excluding mandatory dividend countries. Sample period is 1992–201
Empirical results (2)
Robustness checks:
‣Various standard checks, controls, high versus low etc.
‣Different proxies (e.g. future orientation instead of
patience)
‣More globalized firms have fewer local investors, thus
their relation to the local behavioral parameters is
weaker.
Reference:
Breuer, Rieger, Soypak: The behavioral foundations of
corporate dividend policy a cross-country analysis,
Journal of Banking & Finance, 42 (2014), pp. 247–265.
321
Conclusions from this part
323
Cash holdings
325
Dividend Dummy Size Dummy Age Dummy
2.216*** 1.120 4.133*** -0.369 2.329** -0.509
t Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained
3.288 1.603 5.390 -0.646 2.013 -0.926
Table III. Financial constraints -1.173** and the valuation
0.878 of cash
-1.560** as a function
0.306 of ambiguity
-1.715** aversion 0.639
Ct×Ambiguity Aversion -0.351 -1.194*** -0.331 -0.962** -1.644*** -0.258
t -2.395 1.350 -2.448 0.614 -2.111 1.012
-0.997 -3.065 -0.933 -2.488 -3.214 -0.573
This
Cash table presents the
holdings results of 0.350***
0.357***
regressions on the excess 0.295***
0.427***
stock returns0.385***
distinguishing between two
0.268***
subsamples:
Et the -1.071**constrained
financially 1.145***
and 0.907*** unconstrained
financially 0.159 firms.1.052**
The main 0.109 of inter-
variable
D t 9.714 6.986 9.119 7.616 5.836 6.473
est is the interaction term, -2.050 4.978 . We use dividend
3.281 0.485 payouts to assets, the
2.551 value of to-
0.284
‣ Regression on stock returns: negative impact of ambiguity
tal assets t and company 0.074***
-0.010 0.098***
age as defining 0.117***
criteria for
-0.078** the level of
-0.048 0.064*** 0.099***
financial constraints
-0.056* -0.020 0.082*** (1) and
in Columns
-0.192***
NFt
(2), (3) andaversion × Cash holdings for constraint firms.
5.138
(4) and (5)-0.341 5.693
and (6), respectively.
-2.361 6.291
In Columns
-1.416 (1), (3)4.529 3.868
and (5), we-0.428
-1.873 4.341 for finan-
report the results
-5.557
cially constrained firms and in Columns
2.216***
0.696*** 1.120 (2), (4) 4.133***
0.652*** and (6) for unconstrained
0.768*** -0.369
0.539*** firms.
2.329**Robust0.454***
0.941*** standard
-0.509 errors
Ct t
are clustered at the firm level and we11.132use year and12.486
industry dummies.
3.288
12.401 1.603 5.390 9.573 All t-values
-0.646 2.013 are reported
10.380 7.320 under the
-0.926
coefficient estimates.-0.002*** Statistical
-0.351 significance
-1.194***
-0.001***
at -0.002***
the 1 %, 5 %-0.962**
-0.331 and 10 % level
-0.000*
are indicated
-1.644***
-0.001***
by ***, **
-0.258
-0.000
and *, respectively.
Ltt
-0.997 -3.065 -0.933 -2.488 -3.214 -0.573
-6.545 -4.674 -7.349 -1.720 -4.702 -0.945
Dividend Dummy
-1.071**
0.233*** 1.145***
0.182*** 0.907***
0.316*** Size Dummy 0.159
0.183*** 1.052**
0.286***Age Dummy 0.109
0.155***
CDt-1 t
-2.050
Constrained
9.751 4.978
Unconstrained
5.422 3.281
Constrained
9.480 0.485
Unconstrained
8.358 2.551
Constrained
6.739 0.284
Unconstrained
6.480
-0.010
-1.173**
-0.176** -0.078**
0.878
-0.301*** -0.048
-1.560**
-0.365*** -0.056*
0.306
-0.082 -0.020
-1.715**
-0.076 -0.192***
0.639
-0.117*
CNF
Ct×Ambiguity
t ×C t Aversion
t-1
-0.341
-2.395
-3.201 -2.361
1.350
-3.725 -1.416
-2.448
-4.291 -1.873
0.614
-1.554 -0.428
-2.111
-0.736 -5.557
1.012
-1.894
0.696***
-0.005**
0.357*** 0.652***
-0.004
0.350*** 0.768***
-0.005**
0.427*** 0.539***
-0.003
0.295*** 0.941***
-0.009**
0.385*** 0.454***
-0.001
0.268***
CtC
E×L
t t
12.401
-2.282
9.714 11.132
-1.285
6.986 12.486
-2.068
9.119 9.573
-1.389
7.616 10.380
-2.573
5.836 7.320
-0.405
6.473
-0.002***
-0.223***
0.074*** -0.001***
-0.059
0.098*** -0.002***
-0.259***
0.117*** -0.000*
-0.174***
0.064*** -0.001***
-0.250***
0.099*** -0.000
-0.107*
0.082***
Ambiguity
Lt Aversion
t
-3.085
-6.545
5.138 -1.534
-4.674
5.693 -5.082
-7.349
6.291 -3.039
-1.720
4.529 -3.521
-4.702
3.868 -1.954
-0.945
4.341
0.068***
0.233***
2.216*** 0.015*
0.182***
1.120 0.068***
0.316***
4.133*** 0.011
0.183***
-0.369 0.042***
0.286***
2.329** 0.004
0.155***
-0.509
Common C
Law
t-1 t Dummy
3.900
9.751
3.288 1.688
5.422
1.603 6.086
9.480
5.390 0.874
8.358
-0.646 3.002
6.739
2.013 0.307
6.480
-0.926
0.204
-0.176**
-0.351 1.038***
-0.301***
-1.194*** 1.618***
-0.365***
-0.331 -1.255***
-0.082
-0.962** 0.866**
-0.076
-1.644*** 0.453
-0.117*
-0.258
Inflation
Ct×C t t-1
0.486
-3.201
-0.997 3.715
-3.725
-3.065 4.343
-4.291
-0.933 -2.894
-1.554
-2.488 2.155
-0.736
-3.214 326 1.039
-1.894
-0.573
-0.008 -0.015*** 0.003 -0.022*** -0.012** -0.022***
This table presents the results of regressions of cash holdings divided by net assets (Cash/NetAssets) on
various
This tablefirm characteristics.
presents the resultsTheofmain variableonofthe
regressions interest
excessis our
stockpreference parameter, Ambiguity
returns distinguishing between Aver
two
sion. We usethe
subsamples: dividend payouts
financially to assets,
constrained andthe value of unconstrained
financially total assets andfirms.
company age as
The main defining
variable criteria
of inter-
for is
est thethelevel of financial
interaction term,constraints in Columns (1). and (2), dividend
We use (3) and (4) and (5)
payouts and (6),the
to assets, respectively.
value of to-In
Columns
tal
Cash assets
holdings(1), company
and (3) and (5),ageweasreport thecriteria
defining results for
for the
financially
level of constrained firms andininColumns
financial constraints Columns (1)(2),
and(4
and (3)
(2), (6) and
for (4)
unconstrained
and (5) and firms. Robust standard
(6), respectively. errors (1),
In Columns are (3)
clustered
and (5),atwe
thereport
firm level and we
the results for use yea
finan-
cially ‣ Regression
and industry dummies.
constrained firms Allont-values
and incash
Columns holdings:
are reported
(2), (4) and negative
under impactestimates.
the unconstrained
(6) for coefficient of ambiguity
firms. Statistical
Robust significance
standard errors
at the 1 %, 5% at and 10 %level
are clustered
aversionthe firm
for level arewe
and
constraint indicated
use yearbyand
firms.***, **, and
industry
(Again, *, respectively.
dummies.
many All t-values are reported under the
controls.)
coefficient estimates. Statistical " significance at the 1 %, 5 % and 10 % level are indicated by ***, **,
and *, respectively.
Dividend
Dividend Dummy
Dummy Size
SizeDummy
Dummy Age
Age Dummy
Dummy
Constrai-
Constrained Unconstrained
Unconstrained Constrained
Constrained Unconstrained
Unconstrained Constrained
Constrained Unconstrained
Unconstrained
ned
-0.309***
-1.173** 0.031
0.878 -0.500***
-1.560** 0.253***
0.306 -0.416***
-1.715** 0.081
0.639
Ambiguity Aversion
Ct×Ambiguity Aversion
-2.395
-3.902 1.350
0.640 -2.448
-6.790 0.614
4.703 -2.111
-3.592 1.012
1.587
0.357***
-0.003* 0.350***
0.001 0.427***
-0.006*** 0.295***
-0.008** 0.385***
0.004* 0.268***
-0.001
Et Size
Real
9.714
-1.807 6.986
0.758 9.119
-2.936 7.616
-2.304 5.836
1.646 6.473
-0.669
0.074***
0.109*** 0.098***
0.068*** 0.117***
0.089*** 0.064***
0.087*** 0.099***
0.102*** 0.082***
0.058***
‣ The
t
MB-Ratio effect5.138
disappears
13.134
5.693when only
8.540
6.291 considering
11.498
4.529
9.560
companies
3.868
10.177
4.341
6.214
with a large
2.216***
-0.632***
proportion 1.120 of foreign
-0.576***
4.133***ownership
-0.647***
-0.369 (>30%).
-0.526***
2.329**
-0.844***
-0.509
-0.377***
t
NWC/Assets
‣Controlling 3.288 1.603
for risk -14.676
-14.452 aversion 5.390
does -0.646
-15.075 not change
-12.236
2.013
the-13.485 -0.926
picture. -8.910
-0.351 -1.194*** -0.331 -0.962** -1.644*** -0.258
t -0.005*** -0.005*** -0.005*** -0.005*** -0.006*** -0.004***
Lt -0.997 -3.065 -0.933 -2.488 -3.214 -0.573
-15.324 -17.310 -14.767 -14.996 -14.077 -12.643
-1.071** 1.145*** 0.907*** 0.159 1.052** 0.109
Dt 0.054*** 0.009 0.061*** 0.008 0.042*** -0.026***
SG -2.050 4.978 3.281 0.485 2.551 3270.284
4.810
-0.010 1.034
-0.078** 5.401
-0.048 1.067
-0.056* 2.832
-0.020 -3.234
-0.192***
5.4 Value investment
Behavioral explanation:
Value stocks are only attractive if risk aversion is not too
high and patience is not too low.
Hens and Schindler (JEBO, 2020) derive this predictions
from a mathematical model and show that it is highest
where patience is low and risk aversion is high…
328
T-statistics, calculated with Newey and West (1994) heteroscedasticity and autocorrelation con-
sistent standard errors, are reported in brackets. Min and max sample size for each time period’s
cross-sectional regression are provided in the bottom section along with the start time period for
the datapremia:
Value sample through December 2011, with the exception of Panel C because our private credit
data are limited
Patience to aversion
and risk December as2010.
factors
Panel A: Panel B: Panel C: Panel D
Behavioral Macro. Dev. Financial Dev. Fin. Characteristics
330
5.5 Foreign investment
and culture: discussion
333
Investing on foreign markets
334
Predicting developments
335
Conclusions and future research
336
Discussions on group projects
337
Chapter 6
Group Projects
Master/PhD thesis
Remark:
339