Вы находитесь на странице: 1из 16

Journal of Behavioral Finance

ISSN: 1542-7560 (Print) 1542-7579 (Online) Journal homepage: http://www.tandfonline.com/loi/hbhf20

Value Functions for Prospect Theory Investors: An


Empirical Evaluation for U.S. Style Portfolios

Evanthia K. Zervoudi

To cite this article: Evanthia K. Zervoudi (2017): Value Functions for Prospect Theory
Investors: An Empirical Evaluation for U.S. Style Portfolios, Journal of Behavioral Finance, DOI:
10.1080/15427560.2018.1405005

To link to this article: https://doi.org/10.1080/15427560.2018.1405005

Published online: 13 Dec 2017.

Submit your article to this journal

Article views: 11

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=hbhf20

Download by: [University of New England] Date: 04 January 2018, At: 22:16
JOURNAL OF BEHAVIORAL FINANCE
https://doi.org/10.1080/15427560.2018.1405005

Value Functions for Prospect Theory Investors: An Empirical Evaluation for U.S.
Style Portfolios
Evanthia K. Zervoudi
Athens University of Economics & Business

ABSTRACT KEYWORDS
The main aim of this article is to provide a general behavioral analysis that proposes a series of Prospect theory; Probability
different value functions for prospect theory (PT) investors incorporated into behavioral reward-risk distortion; S-shaped value
models that are finally solved so as to provide some specific optimal solutions. To do this, general function; Loss aversion;
behavioral reward-risk models, which contain all the basic elements of the PT, are first set up. Two Behavioral portfolio
optimization; Upper partial
reward and risk measures, the upper partial moment and the lower partial moment, are
Downloaded by [University of New England] at 22:16 04 January 2018

moment; Lower partial


subsequently used to create the various value functions. The technical difficulties arising during the moment
behavioral maximization process are overpassed by adapting the Rubinstein [1982] algorithm. The
results show that agents differentiate their behavior according to their type of preferences (S-
shaped, reverse S-shaped, kinked convex, and kinked concave value function) but they seem to
always prefer small capitalization and high positively skewed value stock portfolios. Probability
distortion also affects the optimal solutions of the problem, independently of the employing
weighting functional form; when subjective probabilities are employed the optimal weights of the
most risky positively skewed assets seem to increase. Probability distortion has an additional
important effect on optimal perspective values of the problem driving to a significant increase.

Introduction
depending on the loss aversion of the agent. The most
For several years, expected utility theory (EUT) has been representative preference function for this type of invest-
one of the most popular methods in decision-making ors is an S-shaped value function (e.g., a function con-
analysis. EUT models represent rational and totally risk- cave for gains and convex for losses). The importance of
averse investors whose preferences are better described the aforementioned elements of PT in the economic
by an increasing and totally concave utility functions. decision making process has been discussed in several
Uncertainties are expressed in terms of probabilities studies among which Barberis [2013] and Bleichrodt,
(objective and common for all agents) and outcomes in Pinto, and Wakker [2001].
terms of utilities. Many studies, however, demonstrate The main aim of this article is to empirically evaluate
that several deviations from the EUT framework, such as and compare behavioral reward-risk models that contain
probability transformation, loss aversion, scale compati- all the basic elements of PT such as S-shaped value func-
bility, and the prominence effect, take place in practice. tion, loss aversion, and probability distortion. I employ 2
In 1979, Kahneman and Tversky observed that preferen- reward and risk measures, the upper partial moment
ces differ from investor to investor: some agents tend to (UPM) and the lower partial moment (LPM) to create
be risk averse, others risk seeking, and others are neither multiple value functions and the corresponding behavioral
totally risk averse nor totally risk seeking and not even reward-risk models. Their goal is to determine the optimal
consistent with their decision making, although they all asset allocations and the resulting optimal perspective val-
behave rationally in the sense that they prefer more than ues (the values resulting by the value function under con-
less. Based on this observation, they develop the most sideration if the optimal asset allocations are used as
popular alternative to EUT, prospect theory (PT), where weights) for each of these models. The optimization pro-
investors tend to be risk averse for returns considered as cess in a model with nonconcave (and nonconvex) value
gains and risk seeking for returns considered as losses. function, which also has a kinked point at the reference
Gains and losses are determined by a reference point point, differs from the classical optimization method with

CONTACT Evanthia K. Zervoudi zervoudiev@aueb.gr Department of International and European Economic Studies, Athens University of Economics &
Business, Patission 76, Athens, Attica 104 34, Greece.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/hbhf.
© 2017 The Institute of Behavioral Finance
2 E. K. ZERVOUDI

a strictly concave utility function. So, I first create an algo- model. I extend this line of research by incorporating
rithm so as to overpass the difficulties of maximizing a loss aversion and probability distortion so as to create a
nonconcave and nondifferentiable (at the reference point) general behavioral model that includes all the basic
function. A series of studies, such that of Levy and Levy components of the PT. At the same time I solve this
[2004], De Giorgi and Hens [2009], and Best and Grauer problem to provide specific optimal solutions, by fol-
[2016], also provide some specific optimal solutions lowing behavioral optimization methods and maximiz-
resulting from behavioral reward-risk models developed ing the whole value function under the appropriate
in a PT framework. The main aim of these articles is to constraints. This is quite important because by maxi-
make a comparison between the PT and the mean-vari- mizing the whole value function (without separating
ance analysis. Other studies that empirically evaluated PT the analysis into maximization of the UPM or minimi-
models are that of De Giorgi, Hens, and Mayer [2007] zation of the LPM (i.e., I do not differentiate their anal-
and Hens and Vlcek [2011], who examined if the disposi- ysis for outcomes below and above the reference point)
tion effect could be explained by the PT; Pirvu and I take under consideration mixed bets with both posi-
Schulze [2012] made a multistock portfolio selection anal- tive and negative returns which are quite common in
ysis under PT (for stocks that follow multivariate elliptical finance. However, the maximization of a function that
distributions). The purposes of the present article differ is non–quasi-concave and nonsmooth creates a series
Downloaded by [University of New England] at 22:16 04 January 2018

significantly by those papers’ goals. of problems in optimization process that are overpassed
More specifically, I present original evidence and employing the Rubinstein [1982] algorithm.
contributes to the relevant literature in several ways. Second, I set under examination 5 different value
First, I provide a general behavioral portfolio optimiza- functional forms among which 2 of the most widely
tion model for investors with arbitrary preferences and used functions. More specifically, I examine the S-
various risk attitudes taking into account higher shaped value function proposed by Kahneman and
moments, such as skewness and kurtosis, which are sig- Tversky [1979], 2 types of the reverse S-shaped value
nificant elements in real market analysis. The first main function proposed by Markowitz [1952b], the kinked
component of the model is the value function; to create concave, and the kinked convex value functions. Here it
the value function, I combine the upper partial moment is important to underline that the value function cap-
(as reward measure) and the lower partial moment (as tures the risk and the potential attitude of an investor
risk measure). The article that first introduced the LPM and shows how this may change according to the out-
as a risk measure in financial models is that of Bawa come of the problem. Employing these 5 value func-
and Lindenberger [1977]. Specifically, Bawa and Lin- tional forms, I examine how the optimal allocations are
denberger developed a capital asset pricing model differentiated according to the agents’ reward and risk
(CAPM) into a mean-LPM framework and they showed attitudes. As the value functions under consideration are
that when the probability and the portfolio returns dis- not always totally concave (or quasi-concave), the first-
tributions are normal, stable, or Student t then the order conditions are necessary but not sufficient which
mean-lower partial moment CAPM is reduced to the means that they can only describe local optima (i.e.,
traditional mean-scale CAPM. After this article, various such a problem may not have a global maximum but
studies employed these reward-risk measures to only local maxima). After that, in each of these value
develop UPM/LPM analyses (see Fishburn [1977], Sor- functions I have incorporated loss aversion that captures
tino, Van der Meer, and Plantinga [1999], Moreno, the psychological phenomenon according to which
Nawrocki, and Olmeda [2006]). Fishburn [1977] pro- investors are more sensitive to losses than to gains. The
posed the LPM model so as to explain the risk-seeking using loss aversion index is that of Kahneman and Tver-
behavior above and the risk-averse behavior below the sky [1979], which is equal to 2.25 (the pain from losses is
reference return; Sortino et al. [1999] proposed as per- more than double than the pleasure from gains). The
formance measure the UPM/LPM ratio to capture more incorporated loss aversion index makes the value func-
realistic investment behaviors while Cumova and tions steeper for losses than for gains, a fact that creates
Nawrocki [2014] developed a general UPM/LPM port- an additional difficulty in maximization process.
folio optimization model (see also Moreno, Nawrocki, Third, I take the analysis one step further by using 2
and Olmeda [2006]). Most of these articles follow clas- weighting schemes for the outcomes of the problem; I
sical optimization methods where the LPM is mini- use both linear and nonlinear weighting schemes to
mized under the constraint the UPM to remain above a investigate whether the introduction of probability dis-
target return level or the UPM is maximized under the tortion in the model may affect the empirical results.
constraint the LPM to remain below a target return During the years it has been observed that people do not
level, as Markowitz [1952a] did in his mean-variance weight outcomes linearly but they tend to underestimate
JOURNAL OF BEHAVIORAL FINANCE 3

large and moderate probabilities and overestimate small is introduced in the model especially for agents who fol-
probabilities. The probability distortion captures this low more aggressive strategies. The rest of the article is
underweighting of large and overweighting of small organized as follows: the second section discusses the
objective probabilities, through a nonlinear, continuous, methodology, the third section contains data and empiri-
and strictly increasing probability weighting function. In cal results, and the fourth section concludes the article.
the case of probability distortion each outcome is
weighted by a decision weight that is subjective for each
Methodology
agent and arises from her personal information. There
are various studies where probability distortion and S- UPM/LPM
shaped, concave, and convex value functions are com-
The classical reward-risk model is composed by a reward
bined into one model (see among others Tversky and
measure m and a risk measure r; the tradeoff between
Kahneman [1992], Wu and Gonzalez [1996, 1999], Pre-
reward and risk is given by the utility function, which is
lec [1998], Hens and Vlcek [2011], Best and Grauer
increasing in the reward measure and decreasing in the
[2016]). The role of the probability distortion in the pres-
risk measure. I employ as a reward measure the UPM
ent article is to examine if the optimal solutions of the
measure and as risk measure the LPM measure. The
maximization problem are differentiated using subjective
UPM measure gives the positive deviations of returns
Downloaded by [University of New England] at 22:16 04 January 2018

and objective probabilities. To ensure that their results


from a target return. In the model the reference point is
are not due to a specific probability weighting functional
the risk-free rate. This reference return represents a min-
form, I use 2 weighting functions: (1) Tversky and Kah-
imum target return that an investor requires so as to
neman [1992] and (2) Prelec [1998].
cover her liabilities. These positive deviations in the new
Their results indicate that the value functional form
model are not minimized anymore as in the mean-vari-
does influence the outcome. Each value function repre-
ance model. Mathematically the UPM is formulated as:
sents a specific investment strategy; investors who follow
a conservative strategy assign higher weights to less risky
assets and invest smaller percentages of their wealth to X
T

very risky assets. These investors have smaller optimal UPM D pt  ½maxð0; rt ¡ rpÞa (1)
tD1
perspective values relatively to investors who follow
aggressive strategies. The most conservative strategy is
that of agents whose preferences are described by a In Equation 1, rt are the assets’ returns, a is the degree
kinked concave value function while the most aggressive of the UPM and rp is the reference point of the model
and the most profitable strategy is that of agents with a (risk-free rate). The parameter a has to do with the
kinked convex value function. This result confirms the investors’ potential attitude. Specifically:
expectations, as agents with by a kinked concave value  a D 1 implies potential neutrality which means that
function are totally risk averse while agents with a kinked an investor is indifferent to an increase or a decrease
convex value function are characterized by (total) risk of the reward.
seeking. Agents with S-shaped and reverse S-shaped  a < 1 implies potential aversion which means that
value functions also differentiate the allocation of their the higher the returns above the reference return,
wealth according to their risk and potential attitudes the smaller the satisfaction of an investor.
with potential seekers to follow more aggressive strate-  a > 1 implies potential seeking, which means that
gies than potential averse agents. the higher the returns above the reference return,
Probability distortion also influences empirical results, the greater the satisfaction of an investor.
although the specific weighting functional form does not The LPM captures only downside deviations from a
play an important role. Optimal asset allocations are benchmark return. Mathematically, the LPM is formu-
affected by probability distortion as follows: positively lated as following:
skewed assets become more attractive to investors when
I assume probability distortion; their weights are X
T
increased remarkably relatively to the other assets, for all LPM D pt ½maxð0; rp ¡ rt Þb (2)
tD1
value functions. However, the allocations are mostly
related to the strategy that an investor follows (deter-
mined by her preferences) and less by the probabilities. In Equation 2, rt are the assets’ returns, b is the degree
The effect of probability distortion is much clearer on of the LPM, and rp is the reference point of the model.
optimal perspective values in the sense that perspective The parameter b has to do with investors’ risk attitude.
values increase significantly when probability distortion Specifically:
4 E. K. ZERVOUDI

 b D 1 implies risk neutrality which means that an Regarding the value functions, different combinations
investor is indifferent to an increase or a decrease of of the parameters may lead to several value functional
the risk. forms such as the reverse S-shaped value functions of
 b < 1 implies risk seeking (e.g., risk-loving invest- Friedman and Savage [1948] and Markowitz [1952b]
ors), which means that the more the returns fall and the S-shaped value function of PT proposed by Kah-
below the target return, the more investors prefer neman and Tversky [1979]. An important issue is
them. whether the functional form chosen for the analysis
 b > 1 implies risk aversion (e.g., risk-averse invest- affects the empirical results. To investigate this issue in
ors), which means that the more the returns fall more detail I employ a number of different value func-
below the target return, the more investors dislike tional forms and the corresponding strategies as follows:
them.  a; b > 1: High (above unity) parameter values
imply potential seeking and risk aversion and a
reverse S-shaped value function (concave for losses
Value functions
and convex for gains). I try both the cases where
The utility function of the classical reward-risk model a D b D 1.8 and the case where a D 1.8 and b D
can be represented as: 3.6 (the case where the risk parameter b is higher
Downloaded by [University of New England] at 22:16 04 January 2018

than the potential parameter a (i.e., when investors


X
T consider protection against losses more important
U ðxÞ D ½½maxð0; rt ¡ rpÞa C ½maxð0; rp ¡ rt Þb  (3) than their exposure to gains). Note that these
tD1 parameters are close the ones in Cumova and
Nawrocki [2014] (i.e., a D 2 and b D 4).
PT preferences can be represented as:  0 < a; b < 1: Parameter values between 0 and
unity suggest potential aversion and risk seeking
X
T   and an S-shaped value function (concave for gains
Vðrt Þ D w.pt /½½max 0; rt ¡ rp a and convex for losses). This value function is the
tD1
classical PT value function. I use the estimations of
  Kahneman and Tversky (i.e., a D b D 0.88).
¡ L½max 0; rp ¡ rt b  (4)  a > 1 and 0 < b < 1: These values imply both
potential and risk seeking and suggest a value function
In Equation 4, w.pt / are the decision weights that that is in general a kinked convex (because of the
correspond to returns rt 8 t D 1; 2; ::; T (I shall use 2 kinked point at the reference point) value function for
different probability weighting functions), L is the loss both losses and gains. I use as parameters a D 1.8 and
aversion index that takes the value of 2.25 (see Kahne- b D 0.88.
man and Tversky [1979], He and Zhou [2011], among  0 < a < 1 and b > 1: These values imply poten-
others), implying that losses are more than twice painful tial aversion and risk aversion and suggest a kinked
for investors than the pleasure of similar gains. I use the concave value function for both losses and gains
same loss aversion parameter (for all value functional resembling the classical concave function of the
forms) so as to make a direct comparison among the EUT. I use as parameters a D 0.8 and b D 1.88.
optimal allocations of agents who are “equally” loss To weight outcomes 3 approaches are employed: the
averse (have the same loss aversion degree) but they classical linear probability weighting (objective probabili-
have different preferences (i.e., see how the optimal solu- ties) and 2 approaches with probability distortion (Kah-
tions are differentiated according to the value functional neman and Tversky [1992] Prelec [1998]) in which
form under consideration). The evaluation of assets is decision weights (w(p)) subjective for each agent (arising
done by investors according to gains .rt > rp ) and losses from her personal information) are used. The 2 last
.rt < rp ). The first part of the value function represents approaches are employed to take into account the proba-
the positive deviations of assets returns from the refer- bility transformation that is a common violation of the
ence point (gains), that is, the reward measure m, EUT, but also ensure that the results are not due to a spe-
whereas the second part represents the negative devia- cific probability weighting functional form. The experi-
tions of assets’ returns from the reference point (losses), mental outcomes are considered as equally likely. As the
that is, the risk measure r. Having found the optimal analysis is developed into a PT framework, the transfor-
asset allocations, I replace these (optimal) allocations to mation is applied directly on the individual objective
the value function under consideration to find the opti- probabilities of the outcomes and not on the cumulative
mal perspective values. ones as in the CPT framework. Thus, the subjective
JOURNAL OF BEHAVIORAL FINANCE 5

probabilities that weight the experimental outcomes will purchase assets whose prices
PNare less or equal to his wealth
be equal and they will be calculated via the subsequent to maximize his welfare i D 1 λi D 1/, and the short-sell-
probability distortion functions by applying the transfor- ing constraint, that the weights of portfolios assets must be
mation directly on the individual objective probabilities. nonnegative (λi  0 8i). In this way, I avoid the case of
The transformation function w() is a nondecreasing infinite leverage and a great risk exposure.
function w : ½0; 1 ! ½0; 1 with wðpÞ > p for small proba- Hence, the asset allocation model is formulated as fol-
bilities and wðpÞ < p for large probabilities while wðpÞ D p lows:
for p D 0 and p D 1. More specifically, I first use the
"" ! !#a
reverse S-shaped probability distortion function proposed X
T X
N
by Tversky and Kahneman [1992], that is: max w.pt /  max 0; rit  λi ¡ rp
λ
tD1 iD1

pg " !!#b #
wðpÞ D (5) X
N
.pg C ð1 ¡ pÞg /16 g
¡ L max 0; rp ¡ rit  λi
iD1
In Equation 5, the parameter g  0.65 determines the
X
N
overweighting of small probabilities and the underweight- s:t: λi D 1
Downloaded by [University of New England] at 22:16 04 January 2018

ing of large probabilities. This functional form gives the iD1


subjective probabilities that agents give to each event and
it is in accordance with the observation that investors λi  0 8i (7)
tend to overweight small objective probabilities and P
underweight moderate and large objective probabilities. In Equation 7, NiD 1 rit  λi D rtT  λ is the portfolio
Second, I use the reverse S-shaped probability weight- return at each t D 1…T, rt is the vector of the assets’
ing function proposed by Prelec [1998], where: returns at each t D 1…T, and λ the vector of the assets’
weights. There are several difficulties in maximization pro-
wðpÞ D expð ¡ bð ¡ lnpÞÞ~a (6) cess of a behavioral reward-risk model. First, as the value
function is not quasi-concave (and of course not totally
Such a transformation deflates the large probabilities concave) the first-order condition may only describe local
and inflates the small probabilities. The parameter a ~ maxima. Another difficulty is that the loss aversion index,
determines the overweighting of small probabilities and which is incorporated in the value function, makes the
the underweighting of large probabilities. Prelec’s proba- objective function nonsmooth and results a nondifferenti-
bility weighting function is a well-behaved probability ability at the reference point. In the case of a piecewise lin-
distortion function, which is regressive, asymmetric, ear value function this problem can easily be dealt with by
reflective, and reverse S-shaped (concave for the initial using linear programming. Specifically, a nonlinear pro-
interval of small probabilities and convex for the proba- gramming problem with a piecewise linear concave func-
bility interval of large probabilities). As the risk aversion tion maximizing under linear constraints can be easily
parameter a ~ decreases the function becomes more reformulated to a linear programming problem using aux-
regressive, more subproportional and more (reverse) S- iliary variables instead of the deviations from the reference
shaped while for a ~ D 1 the weighting function is the point. In the general case of changing risk aversion, which
identity one w.p/ D p. Moreover, Prelec’s weighting is the case of the value functions that I examine in this arti-
function is a 1-parameter subproportional functional cle, several computational issues arise. For example, the
form with a fixed and an inflection point at p D 16 e D value function is nonsmooth but it has a kinked point at
0:37 (asymmetric function). The fact that the fixed point the reference point where the value function is nondiffer-
is below the probability 0.5 drives to a decrease of an entiable; here I use smoothing techniques to get around
uncertain outcome’s weight relatively to the certain out- the reference point. Also, the value function is non–quasi-
comes’ weights and thus makes investors more risk concave, which implies that there are upper sets that are
averse for gains and more risk seeking for losses, because nonconvex sets and consequently, the portfolio optimiza-
of the reflection. I set b D 1 and a ~ D 0.74, the aggregate tion problem may have several local maxima but not a
least-squares estimations by Wu and Gonzalez [1996]. global solution.
To formulate the asset allocation problem I consider as I follow a process similar to De Giorgi et al. [2006],
λ the vector of the assets’ weights. In this model I assume introducing probability distortion, specific risk and
N assets in the investor’s portfolio with λi i D 1 ,.., N the reward measures, and different value functional forms.
weight of each asset. The constraints that I impose in the First, the nonsmoothness of the value function is treated
model are the budget constraint, that an agent can using cubic spines. Specifically, to deal with the
6 E. K. ZERVOUDI

nondifferentiability problem, I replace the nonsmooth the problem. The choice of the n starting points is crucial
value function V(x) by the smooth V~ ðxÞ: for the quality of the solution. I repeat this process by
increasing n until the solution is unchanged (see De Giorgi
~ ðxÞ
V et al. [2007]).

a0 x3 C a1 x2 C a2 x C a3 ; if x 2 ½rp ¡ d; rp C d
D Data and empirical results
V ðx Þ ; elsewhere
(8) For the empirical analysis I employ data on U.S. style
portfolios (stocks listed in on the NYSE, AMEX, and
In Equation 8, d D 10¡6. The goal of this method is to NASDAQ). More specifically, the sample consists of 10
replace the nonsmooth part of the value function by a canonical decile portfolios formed on size (market equity
smooth and thus differentiable cubic polynomial. I take a [ME]) and 10 decile portfolios formed on value (book to
too small random number d to be as close as possible to market equity ratio [B/M]). All data are available from
the reference point and I approach the nonsmooth part Kenneth French’s online data library. I employ annual
of the value function using cubic splines to make it real returns for the period 1927–2013 (87 years). Table 1
smooth. So, around the reference point the value func- presents descriptive statistics of the sample portfolios.
Downloaded by [University of New England] at 22:16 04 January 2018

tion is replaced by a cubic polynomial, which is now dif- There seems to be an reverse relation between firm size
ferentiable. The rest of the value function remains the and stock returns, that is, the smaller the market capitali-
same, as away from the reference point the value func- zation of the firms included in a portfolio is, the larger
tion is smooth and differentiable and thus there is not the portfolios’ return is. Also, for portfolios formed on
problem in maximization process. B/M ratio, the higher the B/M ratio is, the higher the
Thus, the asset allocation model is now formulated as: average return is (i.e., value stocks seems to outperform
!! growth stocks). Small and value stocks seem to have
X
T X
N
higher variance/standard deviation than large and
max ~
w.pt /  V rit  λi
λ growth stocks, indicating that higher returns may be a
tD1 iD1
compensation for the higher variance (which in general
X
N is considered as a measure of risk for financial assets).
s:t: λi D 1 Note, however, that small and value stocks also have
iD1 positive skewness while large and growth stocks are nega-
λi  0 8i (9) tively skewed. This is a quite desirable feature of assets
especially in PT where investor have a clear preference for
The second problem is that the value function is non– positive-skewed assets. More specifically, if a stock has
quasi-concave, which implies that the portfolio optimiza- positive skewness, investors believe that they have a
tion problem may have several local maxima but not a chance (even a very small one) of becoming very wealthy.
global solution. To overpass this problem I consider N This belief is reinforced by the probability distortion in
starting points that have been randomly generated on the PT where investors tend to overweight too small chances.
P In other words, if positively skewed stocks offer them
unit simplex {λ j N i D 1 λ i D 1, λ i 0 8i}, according to the even a very small chance to have a great return from this
uniform distribution over the unit simplex. I employ the
investment this chance is subjectively distorted and makes
algorithm of Rubinstein [1982] to find the optimal alloca-
this investment to be considered as a great opportunity.
tion in their problem. Briefly, this algorithm starts with the
After that, the maximum return that small and value
generation of n pseudorandom numbers (xi 8 i D 1 … n)
stocks may offer is in excess of 100% and too high rela-
according to the uniform distribution on [0, 1], continues
tively to the maximum return of large and growth stocks.
with their transformation to yi D ¡log (xi) 8 i D 1 … n,
This fact may enhance the belief of PT investors that the
corresponding to the exponential distribution with parame-
positively skewed and more profitable small and value
ter 1, and terminates with their normalization as
stocks seems to be a better investment than the negatively
following Pyni corresponding to the uniform distribu-
y
iD1 i
skewed and less profitable large and growth stocks.
tion over the unit simplex. This process is repeated n times. Another article that also points out the preference of PT
Thus, starting from n starting points I get n locally optimal investors to positively skewed stocks is that of Barberis
allocations for each of the value functions which are used and Huang [2008]. Specifically, authors in their article
to find the corresponding optimal perspective values (at study asset pricing issues in a cumulative PT (CPT)
each of these optimal allocations). The allocation with the framework emphasizing the probability transformation
highest perspective value is considered as the solution of feature of CPT and the probability weighting function
JOURNAL OF BEHAVIORAL FINANCE 7

Table 1. Descriptive statistics.


Mean Variance Standard Deviation Kurtosis Skewness Sharpe Ratio Min Max

Panel A: Portfolios formed on Market Capitalization


Small 0.1919 0.1551 0.3939 1.1246 0.7736 0.4871 ¡0.5437 1.4272
Dec 2 0.1711 0.1286 0.3586 3.1056 0.8874 0.4772 ¡0.5553 1.7165
Dec 3 0.1679 0.0992 0.3150 2.5915 0.6410 0.5329 ¡0.5420 1.4776
Dec 4 0.1592 0.0868 0.2946 0.7666 0.3699 0.5405 ¡0.4911 1.1476
Dec 5 0.1539 0.0759 0.2755 1.0778 0.2161 0.5585 ¡0.4788 1.1539
Dec 6 0.1537 0.0679 0.2606 0.6752 0.1021 0.5899 ¡0.4765 1.0141
Dec 7 0.1463 0.0653 0.2554 1.4656 0.1824 0.5727 ¡0.4451 1.0829
Dec 8 0.1369 0.0549 0.2344 1.4449 0.0507 0.5842 ¡0.4716 0.9793
Dec 9 0.1298 0.0471 0.2171 0.6579 ¡0.3904 0.5980 ¡0.5053 0.7136
Large 0.1117 0.0372 0.1929 ¡0.0896 ¡0.4499 0.5788 ¡0.4172 0.4861
Panel B: Portfolios formed on B/M Ratio
Growth 0.1102 0.0507 0.2252 ¡0.4265 ¡0.1836 0.4894 ¡0.4108 0.5829
Dec 2 0.1211 0.0393 0.1983 ¡0.2360 ¡0.2394 0.6106 ¡0.3606 0.5751
Dec 3 0.1186 0.0391 0.1978 0.09036 ¡0.2386 0.5994 ¡0.3886 0.634
Dec 4 0.1215 0.0494 0.2223 2.1081 0.1204 0.5466 ¡0.5239 0.9575
Dec 5 0.1320 0.0507 0.2251 1.9613 ¡0.0407 0.5867 ¡0.6142 0.9126
Dec 6 0.1327 0.0491 0.2215 1.7590 ¡0.1960 0.5988 ¡0.6160 0.7671
Dec 7 0.1343 0.0576 0.2400 1.7269 ¡0.1073 0.5596 ¡0.6318 0.9626
Downloaded by [University of New England] at 22:16 04 January 2018

Dec 8 0.1561 0.0705 0.2655 2.6297 0.4950 0.5881 ¡0.5448 1.2063


Dec 9 0.1641 0.0721 0.2686 2.5199 0.4397 0.6112 ¡0.5256 1.2408
Value 0.1751 0.1127 0.3358 1.9364 0.3705 0.5216 ¡0.5799 1.4991

Note. Annual return descriptive statistics on U.S. style portfolios (i.e., 10 canonical decile portfolios formed on size [market equity (ME)]) and 10 decile portfolios
formed on value (book to market equity ratio (B/M)]). All data are available from Kenneth French’s online data library. Sample period: 1927–2013 (87 years).

Figure 1. Various value functions.


8 E. K. ZERVOUDI

effects; their main results that an asset’s own skewness Optimal asset allocations
may be (over)priced and that the probability distortion
has a significant role in the decision-making process of The following Table 2 presents results for the various
PT investors are in accordance to the results of the pres- value functions when I assume no probability distortion,
ent article, and especially to the ascertainment that PT for size portfolios (panel A) and B/M portfolios (panel
investors prefer positively skewed stocks such as small B). The optimal asset allocation for investors with an S-
and value stocks, a preference that is reinforced by the shaped function is to invest roughly 35% in the smallest
probability distortion (as the too small gain probabilities capitalization portfolio (small portfolio) and roughly
of these assets are overweighted). 53% in high value portfolios (16.42% in decile 8, 22.15%
in decile 9, and 15.51% in the value portfolio). For invest-
ors with a reverse S-shaped function the results indicate
an optimal asset allocation that weighs heavily toward
Value functions the smallest capitalization portfolio. For instance, when
Regarding value functions, Figure 1 presents the different the 2 parameters are equal (a D b D 1.8), agents give the
functions that I obtain with their dataset for various coeffi- greatest weight to the small stocks’ portfolio (85.84%)
cient assumptions. As can be seen from the figure, when and the second larger weight to size decile 2 (9.41%).
Downloaded by [University of New England] at 22:16 04 January 2018

both parameters are set to less than 1 (a D b D 0.88) the When the 2 parameters are not equal (a D 1.8, b D 3.6),
resulting value function is S-shaped (i.e., strictly concave for agents again give the greatest weight to the small stocks’
gains and strictly convex for losses). This type of functional portfolio (87.35%) and the second larger weight to size
form represents agents who tend to be risk averse for out- decile 2 (10.41%). B/M portfolios do not seem to be
comes considered as gains and risk seeking for outcomes important in this case (e.g., the weight of the value port-
considered as losses. When both parameters are set to be folio varies between 0.005% and 0.22%). For a kinked
greater than 1 (a D b D 1.8) the resulting value function is convex value function the 2 smallest size decile portfolios
reverse S-shaped (i.e., strictly convex for gains and strictly have a combined weight of roughly 54% (32.5% and
concave for losses). Such a function represents investors 22.08%), while the value portfolio has a weight of
whose behavior is characterized by risk aversion for losses 11.37%. For the kinked concave function the smallest
and potential seeking for gains. In this case, it makes size portfolio has a weight of 36.76% while the 2 high B/
sense to examine separately what happens when the risk M value portfolios have a combined weight of around
parameter b is higher than the potential parameter a 43% (36.46% for value decile 9 and 6.46% for decile 10).
(a D 1:8 < b D 3:6/ because investor behavior is char- So, in the case where objective probabilities are employed
acterized by risk aversion for losses (i.e., the protection in the model, investors choose to invest the highest pro-
against losses [expressed by the parameter b]) is more portion of their wealth to the riskiest and most profitable
important than the exposure to gains (expressed by the positively skewed small stock portfolio for all types of
parameter a). When the potential parameter a is set to be preferences (i.e., S-shaped, reverse S-shaped, kinked con-
greater than unity and the risk parameter b smaller than vex, and kinked concave value functions). Agents who
unity (a D 1.8 and b D 0.88) then the resulting value func- are potential averse (S-shaped and kinked concave value
tion is a kinked convex function for both losses and gains. function) give smaller weights to this too risky asset
Such a function represents investors whose behavior is char- while potential seekers (reverse S-shaped and kinked
acterized by risk seeking and potential seeking. This type of convex value function) who derive great satisfaction
investors is aggressive enough and is willing to undertake from very high potential, they give higher weights since
great risks by investing to most risky assets that are related this asset has the highest average return.
to high average returns, to derive the greatest possible satis- The following Table 3 presents results for the various
faction (potential seeking). When the potential parameter a value functions when I assume probability distortion and
is smaller than 1 and the risk parameter b is greater than 1 employ the Kahneman and Tversky [1979] weighting
(a D 0.88 and b D 1.8) then the resulting value function is a function, for size portfolios (panel A) and B/M portfolios
kinked concave function for losses and for gains, which (panel B). The optimal asset allocation for investors with
resembles the classical concave value function of the expected an S-shaped function and probability distortion is to
utility theory. Such a function represents a conservative invest 37.49% in the smallest capitalization portfolio and
investor whose behavior is characterized by risk and poten- roughly 47% in value portfolios (34.96% in decile 9 and
tial aversion. This investor is unwilling to undertake great 12.22% in the value portfolio). For investors with a
risks since the satisfaction that she will derive by the addi- reverse S-shaped function the results indicate, as previ-
tional potential won’t be adequate to compensate her great ously, that optimal asset allocation weighs heavily toward
risk exposure (potential aversion). the smallest capitalization portfolio. For instance, when
JOURNAL OF BEHAVIORAL FINANCE 9

Table 2. Optimal asset allocation: No probability distortion.


Value Function

S-shaped a D 0.88 Reverse S-shaped a D 1.8 Reverse S-shaped a D 1.8 Convex a D 1.8 Concave a D 0.88
b D 0.88 b D 1.8 b D 3.6 b D 0.88 b D 1.8

Panel A Style Portfolio: Market Capitalization


Small 0.3478 0.8584 0.8735 0.3250 0.3676
Dec 2 0.0078 0.0941 0.1041 0.2208 0.0003
Dec 3 0.0040 0.0170 0.0033 0.1871 0.0009
Dec 4 0.0172 0.0171 0.0020 0.0478 0.0037
Dec 5 0.0103 0.0030 0.0016 0.0064 0.0031
Dec 6 0.0098 0.0007 0.0012 0.0044 0.0033
Dec 7 0.0072 0.0009 0.0014 0.0006 0.0031
Dec 8 0.0035 0.0011 0.0011 0.0046 0.0032
Dec 9 0.0046 0.0008 0.0008 0.0020 0.0029
Large 0.0038 0.0004 0.0005 0.0018 0.0023
Panel B Style Portfolio: B/M Ratio
Growth 0.0023 0.0001 0.0005 0.0020 0.0016
Dec 2 0.0064 0.0005 0.0005 0.0027 0.0018
Dec 3 0.0067 0.0005 0.0007 0.0025 0.0020
Dec 4 0.0031 0.0006 0.0010 0.0035 0.0031
Downloaded by [University of New England] at 22:16 04 January 2018

Dec 5 0.0143 0.0007 0.0009 0.0102 0.0074


Dec 6 0.0088 0.0003 0.0006 0.0017 0.0080
Dec 7 0.0017 0.0003 0.0009 0.0019 0.0042
Dec 8 0.1642 0.0003 0.0015 0.0024 0.1523
Dec 9 0.2215 0.0029 0.0018 0.0588 0.3646
Value 0.1551 0.0005 0.0022 0.1137 0.0646
Optimal value 0.2252 0.1740 0.2040 0.2770 0.1590

Note. Optimal asset allocations for value functions without probability distortion and for various assumptions about the coefficients. The sample consists of U.S.
style portfolios (i.e., 10 canonical decile portfolios formed on size [market equity (ME)]) and 10 decile portfolios formed on value (book to market equity ratio [B/
M]). All data are available from Kenneth French’s online data library. Sample period: 1927–2013 (87 years).

the 2 parameters are equal (a D b D 1.8), agents give the are not equal. For a kinked convex value function the 2
greatest weight to the small stocks’ portfolio (88.25%) smallest size decile portfolios have a combined weight of
and the second larger weight to size decile 2 (10.48%); roughly 60% (39.85% and 21.56%), whereas the value
the same happens when I assume that the 2 parameters portfolio has a weight of 15.8%. For the kinked concave

Table 3. Optimal asset allocation: Kahneman and Tversky [1992] weighting function.
Value Function

S-shaped a D 0.88 Reverse S-shaped a D 1.8 Reverse S-shaped a D 1.8 Convex a D 1.8 Concave a D 0.88
b D 0.88 b D 1.8 b D 3.6 b D 0.88 b D 1.8

Panel A Style Portfolio: Market Capitalization


Small 0.3749 0.8825 0.8809 0.3985 0.3830
Dec 2 0.0012 0.1048 0.1058 0.2156 0.0000
Dec 3 0.0080 0.0004 0.0002 0.0018 0.0026
Dec 4 0.0069 0.0013 0.0021 0.1079 0.0015
Dec 5 0.0023 0.0010 0.0006 0.0036 0.0012
Dec 6 0.0013 0.0010 0.0006 0.0224 0.0016
Dec 7 0.0008 0.0009 0.0005 0.0035 0.0016
Dec 8 0.0104 0.0008 0.0004 0.0067 0.0020
Dec 9 0.0084 0.0004 0.0008 0.0059 0.0020
Large 0.0096 0.0004 0.0008 0.0028 0.0011
Panel B Style Portfolio: B/M Ratio
Growth 0.0086 0.0004 0.0004 0.0061 0.0007
Dec 2 0.0097 0.0004 0.0006 0.0054 0.0010
Dec 3 0.0075 0.0005 0.0007 0.0039 0.0019
Dec 4 0.0100 0.0006 0.0008 0.0030 0.0014
Dec 5 0.0174 0.0008 0.0006 0.0023 0.0062
Dec 6 0.0091 0.0001 0.0007 0.0013 0.0197
Dec 7 0.0031 0.0006 0.0006 0.0035 0.0022
Dec 8 0.0391 0.0008 0.0012 0.0174 0.1291
Dec 9 0.3496 0.0012 0.0012 0.0304 0.3919
Value 0.1222 0.0011 0.0006 0.1580 0.0493
Optimal Value 0.7023 0.5949 0.7100 0.8620 0.2233

Note. Optimal asset allocations for value functions with probability distortion (Kahneman and Tversky [1992]) and for various assumptions about the coefficients.
The sample consists of U.S. style portfolios (i.e., 10 canonical decile portfolios formed on size [market equity (ME)]) and 10 decile portfolios formed on value
(book to market equity ratio (B/M)]). All data are available from Kenneth French’s online data library. Sample period: 1927–2013 (87 years).
10 E. K. ZERVOUDI

function the smallest size portfolio has a weight of 38.3% kinked concave value functions) with potential averse
whereas the 2 high B/M value portfolios have a com- agents to give smaller weights to this too risky asset than
bined weight of around 44% (39.19% for value decile 9 potential seekers.
and 4.93% for decile 10). Overall, the results in this subsection indicate that dif-
The results for the case in which I assume probability ferent value functions result to different optimal asset
distortion and employ the Prelec [1998] weighting func- allocations: all value functions suggest a higher weighting
tion are presented in Table 4 and indicate that the choice for small capitalization firms compared with value firms;
of the weighting function does not matter much for the particularly the reverse S-shaped functions suggest a sig-
optimal asset allocation. In other words the results in nificantly higher weighting for small firms. Also, proba-
Table 4 are qualitatively similar to the results in Table 3. bility distortion seems to play a role in the sense that it
For instance, the optimal asset allocation for investors results an increase of extreme style portfolios’ weights
with an S-shaped function is to invest about 40% (small-cap and high B/M); however, the exact weighting
(37.49% in Table 3) in the smallest capitalization portfo- functional form does not play a substantial role. The
lio and roughly 45% (47% Table 3) in value portfolios; results indicate that PT investors believe that investing to
for investors with an reverse S-shaped function the a positively skewed asset has a chance (a very small one)
results indicate that optimal asset allocation weighs to produce very high returns with high risk; thus they
Downloaded by [University of New England] at 22:16 04 January 2018

heavily toward the smallest capitalization portfolio; for a assign the highest weight to the most profitable and posi-
kinked convex value function the 2 smallest size decile tively skewed asset (small stock portfolio). When proba-
portfolios have a combined weight of roughly 57% (from bility distortion is incorporated in the model, the very
60% in Table 3); for the kinked concave function the small probabilities are overweighed, making such a risky
smallest size portfolio has a weight of 41.84% (from asset to seem even more attractive. This suggests that
38.3% in Table 3) while the 2 high B/M value portfolios probability distortion affects decisions (without altering
have a combined weight of around 42% (from 44% in the clear preference to positively skewed assets) driving
Table 3). As in the case of nonprobability distortion, investors to further increase the percentage of their
agents choose to invest the highest proportion of their wealth invested to positively skewed assets and especially
wealth to the riskiest and most profitable positively to the small-stock portfolio that offers very high average
skewed small stock portfolio for all types of preferences returns. These results (Tables 2–4) are also summarized
(i.e., S-shaped, reverse S-shaped, kinked convex, and graphically in Figures 2–6.

Table 4. Optimal asset allocation: Prelec [1998] weighting function.


Value Function

S-shaped a D 0.88 Reverse S-shaped a D 1.8 Reverse S-shaped a D 1.8 Convex a D 1.8 Concave a D 0.88
b D 0.88 b D 1.8 b D 3.6 b D 0.88 b D 1.8

Panel A Style Portfolio: Market Capitalization


Small 0.3999 0.8518 0.8790 0.4003 0.4184
Dec 2 0.0038 0.1083 0.1061 0.1681 0.0009
Dec 3 0.0100 0.0042 0.0004 0.0819 0.0419
Dec 4 0.0045 0.0222 0.0047 0.0506 0.0022
Dec 5 0.0029 0.0039 0.0006 0.0021 0.0018
Dec 6 6.5462 0.0006 0.0005 0.0096 0.0030
Dec 7 0.0019 0.0006 0.0005 0.0119 0.0032
Dec 8 0.0012 0.0006 0.0005 0.0133 0.0075
Dec 9 0.0006 0.0005 0.0005 0.0149 0.0067
Large 0.0008 0.0006 0.0009 0.0060 0.0018
Panel B Style Portfolio: B/M Ratio
Growth 0.0008 0.0003 0.0007 0.0090 0.0038
Dec 2 0.0011 0.0003 0.0006 0.0131 0.0021
Dec 3 0.0008 0.0005 0.0007 0.0143 0.0046
Dec 4 0.0011 0.0003 0.0008 0.0130 0.0041
Dec 5 0.0023 0.0004 0.0005 0.0226 0.0051
Dec 6 0.0002 0.0004 0.0009 0.0172 0.0190
Dec 7 0.0003 0.0004 0.0005 0.0090 0.0025
Dec 8 0.1216 0.0006 0.0006 0.0003 0.0457
Dec 9 0.3828 0.0030 0.0971 0.0382 0.3336
Value 0.0633 0.0006 0.0008 0.1045 0.0921
Optimal Value 0.6067 0.5109 0.6477 0.8451 0.2225

Note. Optimal Asset Allocations for value functions with probability distortion (Prelec [1998]) and for various assumptions about the coefficients. The sample con-
sists of U.S. style portfolios (i.e., 10 canonical decile portfolios formed on size (market equity, ME) and 10 decile portfolios formed on value (Book to Market
Equity ratio, B/M). All data are available from Kenneth French’s online data library. Sample period: 1927–2013 (87 years).
JOURNAL OF BEHAVIORAL FINANCE 11

Figure 5. Optimal asset allocation for the kinked convex value


Figure 2. Optimal asset allocation for the S-shaped value function.
function.
Downloaded by [University of New England] at 22:16 04 January 2018

Figure 6. Optimal asset allocation for the kinked concave value


function.
Figure 3. Optimal asset allocation for the reverse S-shaped value
function (a).
reverse S-shaped, kinked convex, and kinked concave
value function). I obtain these optimal values as follows:
in the optimization process I employed 700 initial points
for each of the value functions; for every initial point I
obtained an optimal asset allocation and its correspond-
ing optimal perspective value through the asset allocation
model (9) and I reported the highest optimal perspective
value as the solution of the optimization problem (for
each of the value functions). Figure 8 presents a sum-
mary of these results. Their goal is to see how the optimal
perspective values are differentiated according to invest-
Figure 4. Optimal asset allocation for the reverse S-shaped value ors’ reward and risk attitudes represented by a specific
function (b). value function. Thus, for the S-shaped function (a D
0.88 and b D 0.88) the optimal value is around 20%
Overall, the findings in this section indicate that (1) agents without probability distortion, around 70% with the
give the maximum weight to the smallest stock portfolio; (2) Kahneman and Tversky [1979] weighting function, and
the second larger weight is given to high value stocks; (3) around 60% with the Prelec [1998] weighting function.
when probability distortion is incorporated in the model it For the reverse S-shaped functions the optimal value is
seems that the very small probabilities are overweighted, around 20% without probability distortion, around 60–
making risky positively skewed assets (and especially the 70% with the Kahneman and Tversky [1979] weighting
small-stock portfolio) to seem more attractive; and (4) the function, and around 50–60% with the Prelec [1998]
used weighting function is of second importance. weighting function. For the kinked convex function (a D
1.8 and b D 0.88) the optimal value is around 85% with
and 28% without probability distortion, whereas for the
Perspective value
kinked concave value function (a D 0.88 and b D 1.8)
Figure 7 presents the optimal perspective values obtained the optimal value is around 16% without and 22% with
for the different value functions of this article (S-shaped, probability distortion.
12 E. K. ZERVOUDI
Downloaded by [University of New England] at 22:16 04 January 2018

Figure 7. Optimal perspective values for different value functions.

As I see there are significant differentiations among assets, which are related to highest average returns, to
the optimal perspective values from one value func- derive the greatest possible satisfaction (potential
tion to the other. The lowest optimal perspective val- seeking).
ues are obtained for the kinked concave value The probability distortion is another important factor
function. The reason behind this is that this strategy that also affects the optimal perspective values. This is so
is the least risky (i.e., agents with this type of prefer-
ences (totally risk and potential averse investors) are
the most conservative and prefer to invest higher per-
centages of their wealth to less risky assets, which are
related to lower average returns, to reduce their risk
exposure (risk and potential aversion); in this way
they sacrifice a part of their potential returns lowering
their optimal perspective values, as well. As expected,
the highest values are obtained for the kinked convex
value function. The reason behind this is that this is
the most risky strategy, that is, agents with this type
of preferences (totally risk and potential seeking
investors) are the most aggressive and prefer to invest Figure 8. Comparison of optimal perspective values for different
higher percentages of their wealth to most risky value functions.
JOURNAL OF BEHAVIORAL FINANCE 13

because the objective probabilities that weight the out- value functions, with and without probability distortion.
comes of the problem are too small, lowering in this way The fact that the optimal perspective values are much
the resulting perspective values. When probability distor- higher than the 2 benchmarks in all cases implies that PT
tion is introduced, these very small probabilities are over- investors whose preferences are described by the value
weighed increasing directly the optimal perspective functions examined in this article (S-shaped function,
values. After that, the overweighting of very small objec- reverse S-shaped functions, kinked convex function, and
tive chances to become very wealthy by investing to posi- kinked concave function) choose strategies that are highly
tively skewed assets makes these assets more attractive to profitable and achieve returns considerably above the aver-
agents driving them to increase their weights in their port- age market return as it is expressed by the average risk-free
folios, but the positively skewed assets are also related to rate and the average excess return on the market portfolio.
high average returns which implies a significant increase Note here that such a rate is often the benchmark return
to the optimal perspective values, as well. (the pursuit) for classical risk-averse agents. This implies
The analysis is completed by comparing the optimal that PT investors differentiate their behavior, choosing to
perspective values resulting by all value functions to 2 follow more risky but also much more profitable strategies
benchmarks: (1) the average risk-free rate for the period to derive a great satisfaction in terms of utility. This finding
under examination 1927–2013 that is equal to 3.5% and may be amplified by the fact that PT agents have a clear
Downloaded by [University of New England] at 22:16 04 January 2018

(2) the average excess return on the market portfolio for preference to positively skewed stocks that are considered
the same period 1927–2013 that is equal to 4.8%. The as quite profitable assets and thus may drive to great poten-
choice of these 2 benchmarks is justified by the fact that tial returns and very high optimal perspective values.
PT investors quite often use these rates (the average risk-
free rate and the average excess return on the market port-
folio) as reference points to evaluate their portfolios. The
Conclusions
results show that the optimal perspective values are much
higher than the 2 benchmarks in all cases. This finding In this article, I built a general behavioral reward-risk model
becomes more conspicuous when probability distortion is that contains all the basic elements of the PT. I used 2 reward
introduced in the model. More specifically, the lowest and risk measures, the UPM and the LPM, to create different
optimal perspective values obtained for the kinked con- value functions for their problem that I solved by maximiz-
cave value function is around 16% without and 22% with ing the resulting value functions under the budget and the
probability distortion (i.e., more than 4 times higher than short-selling constraints. I find that when agents are potential
the average risk-free rate 3.5% without probability distor- seekers and their preferences are represented by a reverse S-
tion and more than 6 times higher than the average risk- shaped or by a kinked convex value function, they assign
free rate 3.5% with probability distortion. Similarly, the much higher weights (almost 86%) to the small size portfolio
optimal perspective values obtained for the least risky (the riskiest and most profitable positively skewed asset). In
strategy, representing by the kinked concave value func- addition, they assign the second larger weight to the decile 2
tion, are more than 3 times higher than the average excess portfolio formed on size, which is the second riskiest and
return on the market portfolio without probability distor- second most profitable positively skewed asset. In other
tion and more than 4 times higher than the average excess words, potential seekers follow very aggressive strategies and
return on the market portfolio with probability distortion. invest almost all of their wealth in only 2 very risky and very
The differentiations are impressive if we compare the opti- profitable assets to increase their potential and thus derive a
mal perspective values obtained for the most risky strat- great satisfaction. The strategy is also consistent with the
egy, representing by the kinked convex value function, to clear preference of PT investors to positively skewed assets,
the 2 benchmarks (i.e., more than 8 times higher than the which are quite risky but they offer the (small) chance of sig-
average risk-free rate without probability distortion and nificant returns. When preferences are represented by an S-
more than 24 times higher than the average risk-free rate shaped or a kinked concave value function, potential averse
with probability distortion. Similarly, the highest optimal investors follow more conservative strategies without altering
perspective values are more than 5 times higher than the their clear preference to the positively skewed assets. In these
average excess return on the market portfolio without cases, agents allocate the greatest part of their wealth among
probability distortion and more than 17 times higher than the small stocks’ portfolio (about 36%) and the high-value
the average excess return on the market portfolio with portfolios (about 38% in the 2 highest value deciles). This
probability distortion. strategy is less aggressive than the strategy of potential
The reason for this comparison is to give an economic seekers in the sense that potential averse investors choose to
meaning to the comparative analysis of the optimal alloca- invest to more and less risky (but always positively skewed)
tions and the resulting optimal perspective values for all assets to reduce their risk exposure.
14 E. K. ZERVOUDI

Probability distortion seems to affect the optimal asset general behavioral reward-risk models that are solved
allocations, but it does not alter at all the clear preference under the appropriate constraints. This analysis is devel-
of PT investors to positively skewed assets, independently oped into a PT framework similar to the studies of Best
of the employing weighting functional form. In fact, when and Grauer [2016] and Pirvu and Schulze [2012] that,
subjective probabilities are employed the optimal weights however, are in a different direction; these studies mostly
of the most risky positively skewed assets increase. This focused on the optimization process of a PT model under
takes place because the objective probabilities for an agent the appropriate constrains and subsequently to the com-
to become very wealthy by investing to a quite risky posi- parison of the resulting solutions with that of the classical
tively skewed asset are too small. But, when probability mean-variance problem. One interesting issue for future
distortion is incorporated in the model, the very small research would be the development of the present behav-
probabilities are overweighted making this asset more ioral analysis into a CPT framework. The main differenti-
attractive to investors and driving them to increase the ation between PT and CPT is that in the PT framework
percentages of their wealth that they invest to this asset. In the experimental outcomes are equally likely because the
general, probability distortion increases the attractiveness transformation is applied directly on the individual objec-
of the positively skewed assets, driving investors to tive probabilities of the outcomes whereas in the CPT
increase the weight that they give to these assets, indepen- framework equally likely outcomes may take different
Downloaded by [University of New England] at 22:16 04 January 2018

dently of the employing weighting functional form. probability weights because the transformation is applied
Finally, probability distortion has a significant effect on on the cumulative probabilities rather than on the individ-
optimal perspective values; without probability distortion ual ones (see Tversky and Kahneman [1992], Wakker and
the optimal perspective values are much lower than the Tversky [1993], among others). In this direction are the
optimal perspective values with probability distortion. working article “Theory Matters for Financial Advice!” of
This happens because without probability distortion the Hens and Mayer [2014b] and especially the working arti-
objective probabilities, which weight the outcomes are too cle “Cumulative Prospect Theory and Mean Variance
small while with probability distortion the small probabili- Analysis: A Rigorous Comparison” of Hens and Mayer
ties are overweighted, and thus the perspective values [2014a] in which algorithms for optimization into a CPT
increase significantly. The lowest optimal perspective framework are presented, the probability distortion is
value (around 20% with probability distortion) is observed applied on cumulative rather than on individual probabili-
in the case of the kinked concave value function, which ties while it is taken into account the additional non-
represents more conservative investors (large weights are smoothness that is introduced in the model because of the
assigned to the less risky assets) and the highest optimal sorting of scenarios.
perspective value (around 85% with probability distor- Two basic issues would make interesting the develop-
tion) is observed in the case of the kinked convex value ment of the present behavioral analysis into a CPT
function which represents the most aggressive investors framework. The first main issue is that models developed
(large weights are assigned to the most risky assets). High into a CPT framework may avoid the violation of the
optimal perspective values are also obtained in the case of first-order stochastic dominance rule, which practically
the reverse S-shaped value function, which represents means that PT investors may choose dominated strate-
risk-averse and potential seeking investors; the weight that gies. However, there are studies such that of Rieger and
these investors give to the very profitable small stocks Wang [2008], which support that the violation of the
portfolio is very high (around 86%), resulting quite large first order stochastic dominance rule is not necessarily a
returns. This is due to the potential-seeking attitude, problem because in descriptive decision-making theories
which makes agents to be willing to undertake greater such as the PT, investors often choose dominated strate-
risks to have higher returns and thus greater satisfaction. gies. Even if the most economic theories preassume
On the other hand, in the case of the S-shaped value func- rational investors, the theory of Kahneman and Tversky
tion, which represents risk seeking and potential averse [1979] does not adopt this traditional assumption of
investors, the optimal perspective values are smaller agents’ rationality because it is considered at most a
because the potential aversion attitude makes agents descriptive theory and not a normative one. The second
unwilling to undertake high risks, lowering in this way issue is that the PT is not continuous (i.e., small changes
their optimal perspective values. In general, the riskier the in lotteries can drive to great changes in PT value). The
strategy is, the higher the optimal perspective values are, problem arises from the fact that the PT is applied only
with and without probability distortion. on lotteries with small number of outcomes and on dis-
It would be useful here to point out that this article is a crete distributions. The CPT could offer a solution to
general behavioral analysis that proposes a series of differ- this problem since the CPT is applied on lotteries with
ent value functions for PT investors incorporated into any number of outcomes. However, as the number of the
JOURNAL OF BEHAVIORAL FINANCE 15

outcomes goes to infinity, the limit of the continuous dis- Gonzalez, R., and G. Wu. “On the Shape of the Probability Weight-
tributions becomes independent of specific weighting ing Function.” Cognitive Psychology, 38, (1999), pp. 129–166.
functional forms. In this way, an analysis developed into He, X. D., and X. Y. Zhou. “Portfolio Choice under Cumulative
Prospect Theory: An Analytical Treatment.” Management
a PT framework becomes computationally less costly Science, 57, (2011), pp. 315–331.
than an analysis developed into a CPT framework. This Hens, T., and J. Mayer. “Cumulative Prospect Theory and
simplification offers a significant advantage to the PT (or Mean Variance Analysis: A Rigorous Comparison.”
the smooth PT), especially in behavioral finance where (2014a). Swiss Finance Institute Research Paper No. 14–23,
the amounts of data (e.g., historical stock returns are forthcoming to Journal of Computational Finance.
Hens, T., and J. Mayer. “Theory Matters for Financial Advice!”
huge, comparing to the CPT that is numerically and ana-
(2014b). Swiss Finance Institute Research Paper No. 14–22,
lytically costly. From all these, it becomes clear that both RR for Journal of Computational Finance.
behavioral analyses developed into a PT framework and Hens, T., and M. Vlcek. “Does Prospect Theory Explain the
a CPT framework are quite important and interesting. Disposition Effect?” Journal of Behavioral Finance, 12,
(2011), pp. 141–157.
Kahneman, D., and A. Tversky. “Prospect Theory: An Analysis of
References Decision under Risk.” Econometrica, 47, (1979), pp. 263–292.
Levy, H., and M. Levy. “Prospect Theory and Mean-Variance
Barberis, N. “Thirty Years of Prospect Theory in Economics: A Analysis.” Review of Financial Studies, 17, (2004), pp.
Downloaded by [University of New England] at 22:16 04 January 2018

Review and Assessment.” Journal of Economic Perspectives, 1015–1041.


27, (2013), pp. 173–196. Markowitz, H. “Portfolio Selection.” Journal of Finance, 7,
Barberis, N., and M. Huang. “Stocks as Lotteries: The Implica- (1952a), pp. 77–91.
tions of Probability Weighting for Security Prices.” Ameri- Markowitz, H. “The Utility of Wealth.” Journal of Political
can Economic Review, 98, (2008), pp. 2066–2100. Economy, 60, (1952b), pp. 151–158.
Bawa, V., and E. Lindenberger. “Capital Market Equilibrium in Moreno, D., D. Nawrocki, and I. Olmeda. “A Genetic Algo-
a Lower Partial Moment Framework.” Journal of Financial rithm for UPM/LPM Portfolios.” Society for Computa-
Economics, 5, (1977), pp. 189–200. tional Economics, Computing in Economics and
Best, M., and R. Grauer. “Prospect Theory and Portfolio Selec- Finance (2006).
tion.” Journal of Behavioral and Experimental Finance, 11, Pirvu, T. A., and K. Schulze. “Multi-stock Portfolio Optimiza-
(2016), pp. 13–17. tion Under Prospect Theory.” Mathematics and Financial
Bleichrodt, H., J. Pinto, and P. Wakker. “Making Descriptive Use of Economics, 6, (2012), pp. 337–362.
Prospect Theory to Improve the Prescriptive Use of Expected Prelec, D. “The Probability Weighting Function.” Econometr-
Utility.” Management Science, 47, (2001), pp. 1498–1514. ica, 66, (1998), pp. 497–527.
Cumova, D., and D. Nawrocki. “Portfolio Optimization in an Rieger, M. O., and M. Wang. “Prospect Theory for Continuous
Upside Potential and Downside Risk Framework.” Journal Distributions.” Journal of Risk and Uncertainty, 36, (2008),
of Economics and Business, 71, (2014), pp. 68–89. pp. 83–102.
De Giorgi, E., and T. Hens. “Making Prospect Theory Fit for Rubinstein, R. Y. “Generating Random Vectors Uniformly Dis-
Finance.” Financial Markets and Portfolio Management, 20, tributed Inside and On the Surface of Different Regions.”
(2006), pp. 339–360. European Journal of Operational Research, 10, (1982), pp.
De Giorgi, E., and T. Hens. “Prospect Theory and Mean-vari- 205–209.
ance Analysis: Does it Make a Difference in Wealth Man- Sortino, F. A., R. Meer, and A. Plantinga. “The Dutch Trian-
agement?” Investment Management and Financial gle.” The Journal of Portfolio Management, 26, (1999), pp.
Innovations, 6, (2009), pp. 122–129. 50–57.
De Giorgi, E., T. Hens, and J. Mayer. “Computational Aspects Tversky, A., and D. Kahneman. “Advances in Prospect Theory:
of Prospect Theory and Asset Pricing Applications.” Cumulative Representation of Uncertainty.” Journal of Risk
Computational Economics, 29, (2007), pp. 267–281. and Uncertainty, 5, (1992), pp. 297–323.
Fishburn, P. “Mean-Risk Analysis with Risk Associated with Wakker, P., and A. Tversky. “An Axiomatization of Cumula-
Below-Target Returns.” American Economic Review, 67, tive Prospect Theory.” Journal of Risk and Uncertainty, 7,
(1977), pp. 116–126. (1993), pp. 147–176.
Friedman M., and L. J. Savage. “Utility Analysis of Choices Wu, G., and R. Gonzalez. “Curvature of the Probability
Involving Risk.” Journal of Political Economy, 56, (1948), Weighting Function.” Management Science, 42, (1996), pp.
pp. 279–304. 1676–1690.