See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/229026287

Framing Effects, Selective Information and Market Behavior-An

Experimental Analysis

Article · March 2004

CITATIONS READS

17 732

10 authors, including:

Erich Kirchler
University of Vienna
372 PUBLICATIONS 13,255 CITATIONS

SEE PROFILE

All content following this page was uploaded by Erich Kirchler on 16 May 2014.

The user has requested enhancement of the downloaded file.

 Framing Effects, Selective Information and
Market Behavior
- An Experimental Analysis -∗
Erich Kirchlera, Boris Maciejovskyb, Martin Weberc

February 2004

Abstract
The results of an asset market experiment, in which 64 subjects trade two assets on
eight markets in a computerized continuous double auction, indicate that objectively
irrelevant information influences trading behavior. Moreover, positively and nega-
tively framed information leads to a particular trading pattern, but leaves trading
prices and trading volume unaffected. In addition, we provide support for the dis-
position effect. Participants who experience a gain sell their assets more rapidly than
participants who experience a loss, and positively framed subjects generally sell their
assets later than negatively framed subjects.

Keywords: financial markets, prospect theory, anchoring and adjustment,

experimental economics, disposition effect

JEL-Classification: C90; D44; D80; G12

∗ The authors acknowledge financial support by the Austrian National Bank (Jubiläumsfonds 8382). Valuable comments by Jordi
Brandts, Werner Güth, Hans Haumer, Christian Helmenstein, Manfred Königstein, Christian Schade, Erik Theissen, and Anthony
Ziegelmeyer are gratefully acknowledged. We also strongly benefited from comments by seminar participants at the Humboldt-
University of Berlin, the ENDEAR-workshop in Amsterdam, the ESA-meeting in Barcelona, the IAREP-conference in Bath, and the
public choice meeting in San Diego. Thanks are also due to Tarek El-Sehity, Eva Hofmann, and Herbert Schwarzenberger, who helped
to run the experiment at the University of Vienna.
a University of Vienna, Department of Psychology, Unit for Economic Psychology, Universitätsstrasse 7, A-1010 Vienna, Austria. Tel.:
+43/1/4277 47880, Fax: +43/1/4277 47889, E-mail: [email protected]
b Max Planck Institute for Research into Economic Systems, Strategic Interaction Group, Kahlaische Str. 10, D-07745 Jena, Germany.
Tel.: +49/3641/686 626, Fax: +49/3641/686 623, E-mail: [email protected]
c University of Mannheim, Chair of Business Administration and Finance, esp. Banking, L5-2, D-68131 Mannhein, Germany. Tel.:
+49/621 181 1532, Fax: +49/621 181 1534, E-mail: [email protected]
1 Introduction

The communication of asset return distribution is a central issue in finance. Investment

advisors are legally obliged to inform their clients about the potential risk of their invest-
ments, whereby financial risk is usually expressed by the variance or the standard deviation
of the underlying distribution of the future returns on the investment. Investors are thus,
at least, implicitly required to accurately perceive and interpret statistical information, ir-
respective of how the information is presented.
In recent years new possibilities of acquiring financial information have become avail-
able to investors. One of the main information sources is the internet, which provides
investors with a vast quantity of investment data. Yahoo!Finance, for instance, offers free
market information, business news, and personal finance plans, and BigCharts provides
investors with the option of creating personalized interactive charts. Market data provider
eSignal even assumes a positive relationship between information quantity and investment
success by promising “You’ll make more, because you know more.”
This evidence suggests that investors generally benefit from the provision of infor-
mation. Empirical studies, however, indicate that more information does not necessarily
imply an increase in actual knowledge. In the psychological literature, this tendency is
referred to as the illusion of knowledge and is empirically confirmed for a variety of de-
cision domains. Park (2001), for instance, shows that even when news media recipients
are socially involved with issues covered in the media, they are prone to the illusion of
knowledge. This tendency will increase the more recipients use the media. In the finance
domain, Barber and Odean (2001) investigate the performance of investors who switched
from phone-based trading to internet trading. While those traders who opted for internet
trading initially beat the market by about 3% prior to going online, their performance de-
creased after going online, resulting in a performance of 2% below the market. Similarly,
Choi et al. (2002) report evidence of underperformance in the market-timing of online

1
traders in 401(k) plans. Access to vast quantities of investment data on the internet does
therefore not necessarily imply better performance.
Information does not only play an important role in individual investment decisions,
but also in market environments. Market efficiency, for instance, requires that neither ob-
jectively irrelevant information nor selectively distributed information affects aggregate
market prices. If, for example, some traders receive a positive signal about the likely re-
turns on an asset, whereas an equal number of traders receive the opposite signal, this
information should be completely revealed, leaving aggregate market prices unaffected.
Thus, while individual investors may be prone to biases, such as the illusion of knowledge,
aggregate market prices are considered to be unbiased.
In this paper we focus on the communication and the quality of information in the
context of a competitive asset market. More precisely, we investigate the impact of objec-
tively irrelevant information on trading behavior by drawing upon a novel type of fram-
ing. Traders are confronted with randomly distributed selective information about perfor-
mance chances of potential investments. The presented information is either provided in
a positive or a negative frame and is essentially irrelevant to the decision problem at hand.
Since we symmetrically distribute the additional information among traders, aggregate
market behavior is expected to remain unaffected. In addition, we also investigate the ro-
bustness of the disposition effect in a competitive market environment with real time data
available.
Our results indicate that objectively irrelevant information influences trading behav-
ior. Moreover, positively and negatively framed information leads to a particular trading
pattern, but leaves trading prices and trading volume unaffected. In addition, our findings
support the disposition effect. Participants who experience a gain sell their assets more
rapidly than participants who experience a loss. Furthermore, this effect is mediated by

2
framing. Positively framed market participants generally sell their assets later than nega-
tively framed participants.
This section proceeds with a discussion of framing effects and the disposition effect.
In section 2, the experimental design and the procedure are introduced. Section 3 covers
the results, and finally in section 4, our main findings are discussed.

1.1 Framing effects

Expected utility theory assumes, among other things, descriptive invariance, implying that
different representations of the same choice problem should yield the same preference.
However, several empirical studies indicate that this axiom is frequently violated in indi-
vidual decision making. McNeil et al. (1982), for instance, showed that the same medical
statistics, framed either in terms of mortality rates or in terms of survival rates, lead to
different preferences. Framing effects were also observed in decisions involving risky lot-
teries and monetary payoffs (Kahneman and Tversky 1983, Tversky and Kahneman 1981).
More recently, Statman (1995) as well as Kahneman and Riepe (1998) applied the concept
of framing to financial decisions, such as dollar-cost averaging.
Weber et al. (2000) investigated the impact of endowment framing on market prices
in an experimental asset market. Participants were either put in long position by receiv-
ing some amount of cash plus a certain amount of positively valued risky assets (positive
framing) or were put in short position by receiving a larger amount of cash and certain
state-contingent liabilities (negative framing). The objective value of initial endowments
was identical in terms of final wealth. In line with the predictions of prospect theory, We-
ber et al. (2000) found that overpricing1 was observed more often for negatively framed
market participants than for positively framed participants.
In contrast to Weber et al. (2000) who altered participants’ initial actual endowments,
the present study investigates whether framing effects are also robust under weaker con-

3
ditions, for instance, when participants only obtain different and more importantly irrele-
vant information. Our experimental procedure differs from the way framing effects were
originally studied by Tversky and Kahneman (1981). In their experiments subjects were
presented with scenarios in which a hypothetical decision problem was either semantically
framed in terms of “gains” or “losses.” However, the concept of framing in studies that em-
phasize the role of language in the representation of the decision problem lack conclusive
empirical evidence. Kühberger (1995) found that a variation of missing items of informa-
tion in the decision problem produced markedly different framing effects. Moreover, with
fully described decision problems, no framing effects emerged at all. While the results of
a meta-analysis of 136 empirical studies indicate that generally framing effects are a reli-
able phenomenon (Kühberger 1998), the results of a further meta-analysis, particularly
focusing on Asian disease-like studies indicate that risk preference depends on the size of
payoffs, the probability levels, and the type of good at stake (Kühberger et al. 1999).
In our experimental procedure, we employ a novel type of framing, which is not based
on semantic variations of a decision problem. Instead, participants are informed that divi-
dends are randomly determined and drawn from a normal distribution with a commonly
known fixed µ and a fixed σ, whereby we assume that µ serves as the aspirational refer-
ence payoff. For a given probability p, p ∈ {0, ..., 0.5}, we let Xp and Xp denote the 100p
and 100(1 − p) percentiles, respectively. For a given p, subjects are told that dividends will
be less than Xp with probability p (negative framing), or that dividends will exceed Xp
with probability p (positive framing). We distinguish between two independent markets,
A and B, in which percentile information follows two different probabilities. The framed
information on market A deviates more extremely from the aspirational reference payoff
µ than the framed information on market B (pA < pB ).
Our experimental approach is not only related to framing effects, but also to the “an-
choring and adjustment”bias (Tversky and Kahneman 1974), which refers to a sequential

4
decision situation in which initial information serves as an anchor, from which adjust-
ments in the decision process are only made insufficiently.2 In our design, the positively
and negatively deviating dividend information represents the initial information, the an-
chor. If subjects respond to this additional percentile information, we expect that trad-
ing behavior is influenced in a systematic way. Positive information should increase the
traders’ dividend expectations, whereas negative information should lower them, leading
to a particular trading pattern. Positively framed buyers are expected to purchase assets
from negatively rather than from positively framed sellers, whereas in turn negatively
framed sellers are expected to sell their assets to positively rather than to negatively framed
buyers.
If the framed information has indeed a systematic impact on the traders’ dividend ex-
pectations we hypothesize a differential trading activity on the two markets, A and B. The
additional irrelevant information deviates more strongly from the aspirational reference
payoff µ on market A than on market B. Thus, the more extreme information on the for-
mer should create more diverging dividend expectations on part of the traders, which is
assumed to increase the likelihood that pairs of participants willing to trade will actually
meet on the market. The experimental design also allows us to investigate whether fram-
ing effects vanish if the decision problem is fully described, as suggested by Kühberger
(1995), based on individual decision tasks. Since in our experiment positively and nega-
tively framed information is symmetrically distributed among traders, the market posseses
complete information.
One might argue that the percentile information in our approach might serve two
different roles; an informational and a framing role. Although from a normative perspec-
tive the additional percentile information is logically redundant, given knowledge of the
mean and variance of the normal distribution, individuals may perceive it as useful in the
decision-making process; for instance, in order to learn about the shape of the distribution.

5
However, since the percentile information is symmetrically distributed among subjects in
the context of a competitive market environment, given that information dissemination
takes place and that subjects have statistical training, the informational role is experimen-
tally ruled out.3 Observed behavioral regularities in the experiment are therefore likely to
be due only to the framing role.

1.2 Disposition effect

The disposition effect is one implication of prospect theory (Kahneman and Tversky 1979,
Tversky and Kahneman 1992). In contrast to the utility function implied by expected util-
ity theory, the value function v postulated by prospect theory is defined in terms of gains
and losses relative to a reference point and not on the basis of absolute levels of final wealth.
Prospect theory assumes that the value function is concave for gains and convex for losses.
In a financial context, one can therefore expect that winner assets will be sold more read-
ily than loser assets in order to collect the gain and undo or “repair” the loss, respectively
(Shefrin and Statman 1985).
This hypothesis has been supported empirically, for instance, for field data (Heisler
1994, Odean 1998), and in experimental asset markets (Heilmann et al. 2000, Weber and
Camerer 1998). Odean (1998) analyzed trading records for 10,000 accounts at a large dis-
count brokerage house and found that investors held losing stocks for a median of 124
days, whereas winners were only held for 104 days. Heilmann et al. (2000) showed on an
experimental call market that in periods of rising trading prices with respect to the previ-
ous trading period the number of assets offered as well as the number of assets sold was
higher than in periods of falling trading prices.
In contrast to Heilmann et al. (2000) who used the price of the previous trading period
as the reference point, the present study focuses on individual behavior and defines the
reference point in the way Weber and Camerer (1998) did, namely as the subject’s purchase

6
price. But unlike the experimental procedure followed by Weber and Camerer (1998) who
determined the prices not on the basis of the trading actions of subjects but by a random
process, our market prices are determined solely by the market participants themselves on
a computerized experimental asset market. Our contribution to the existing literature is
to study the disposition effect in the context of a competitive market environment with
real time data available. We expect that a purchase price lower than the previous trading
price implies a gain that will lead to more rapid selling, whereas a purchase price higher
than the previous trading price implies a loss that will lead to less rapid selling.

2 The experiment

2.1 Participants

Overall, 64 participants, all students either at Vienna University or the Vienna University
of Economics and Business Administration, participated in eight sessions of an experi-
mental asset market. On average, participants earned ¤ 19.14 with a standard deviation of
¤ 14.94. Twenty-two females and 42 males, aged 19 to 31 (M = 22.52, SD = 2.90), par-
ticipated in the experiment. The time required for conducting the experiment was about
2 hours and 15 minutes. Forty-nine participants were students of economics, whereas the
remaining 15 participants were enrolled in other social science disciplines. All participants
had attended at least introductory courses in statistics during their studies.

2.2 Experimental design

The experiment was conducted in a 2 x 2 factorial design in order to study the inter-
action of differently framed participants within one market. The independent variables
were (i) the framing of dividend information (positively versus negatively framed infor-
mation) as a between-subjects factor and (ii) the probability of the framed information
(low versus high probability; pA = .05 and pB = .45) as a within-subjects factor. Par-

7
ticipants were randomly assigned to one of the two framing conditions. All participants
were informed that dividends would be randomly drawn from a normal distribution with
µ ∈ {95, 135, 105} and σ ∈ {20, 30, 40}. For the combination and the sequence of µ and
σ see Table 1. In order to keep subjects’ attention levels high we balanced µ and σ across
trading periods.

[Table 1 about here.]

Prior to the trading periods participants were provided with the actual µ and σ as
well as with additional irrelevant percentile information, Xp (negative framing) and Xp
(positive framing), respectively. Figure 1 exemplarily displays the available information to
subjects at the beginning of the trading periods.

[Figure 1 about here.]

2.3 Experimental procedure

The experiment consisted of four phases. In the first phase, subjective propensity toward
risk was measured experimentally by the methods of certainty equivalents and by binary
lottery decisions in order to control for possible differences in individual risk attitude. In
the second phase, the experimental asset market was opened and assets were traded. In the
third phase, participants were asked to complete a short questionnaire, and finally in the
fourth and last phase, the procedure to control for risk attitude was repeated. The exact
sequence of events in the experiment is shown in Figure 2.

[Figure 2 about here.]

Phase 1: After brief instructions, participants were asked (i) to reveal their certainty
equivalent for a lottery that offers a payoff of 100 Experimental Currency Units (ECU)4
with a probability of p = .50, and zero ECU otherwise; and (ii) to make seven decisions

8
among risky lotteries.5 The payoffs of the lotteries are listed in Table 2. As a control for
position effects, the lotteries were systematically varied with respect to a1 (highest possible
payoff) and a2 (lowest possible payoff) as well as to A (certain payoff), and to the sequence
of a1 /a2 (risky payoff).

[Table 2 about here.]

The certainty equivalent allows the experimenter to infer participants’ attitude toward
risk. More precisely, it allows to discriminate between risk aversion, risk neutrality and
risk seeking behavior. A certainty equivalent that is lower than the expected value of the
lottery, which is 50 ECU, indicates risk aversion, whereas a certainty equivalent equal to
the expected value indicates risk neutrality, and finally a certainty equivalent above the
expected value indicates risk seeking behavior. Also, the seven decisions among lotteries
can be used to infer risk attitude. However, since each lottery has the same expected value
in each of its two components, namely the certain payoff and the risky payoff, the design
only allows to distinguish between risk aversion (certain payoff) and risk neutrality (risky
payoff).
One of the seven decisions was randomly selected in order to determine the individual
payoff. The payoff from the lotteries was added to the total payoff from the market. Phase
1 took 15 to 20 minutes.
Phase 2: After receiving instructions about the experimental asset market6 and a short
questionnaire to check the understanding of the instructions, subjects participated in two
trial periods of six minutes each in order to become familiar with the selling and buying
procedures on the market. After the trial periods, the asset market was opened. Overall,
eight sessions were run with eight subjects each on a computerized asset market, using
the software z-Tree (Zurich Toolbox for Readymade Economic Experiments, Fischbacher
(1999)).

9
 The computer screen for the auction is displayed in Figure 3. Each market participant
was entitled (i) to submit bids and asks, (ii) to accept standing bids and asks, whereas
only better offers, i.e., higher bids and lower asks, were allowed, or (iii) to remain passive.
Bids and asks were automatically ranked, indicating the most favorable offer. Informa-
tion about the trading history, provided as a chronological list of contracts, was displayed
throughout the trading periods.

[Figure 3 about here.]

The market was performed as a continuous anonymous double auction. Participants

were endowed with 1,000 ECU (100 ECU equal ¤ 0.18) plus five risky assets A and five
risky assets B. The assets were traded separately on market A and market B. To ensure com-
parability between the sessions, the sequence of the two markets was chosen in advance
and applied to all eight sessions. Dividends were randomly determined and drawn from a
normal distribution (see Table 1). Participants were informed that the markets would be
open for at least eight and at most twelve periods. The probability that the markets would
end after the eighth, ninth, tenth, or eleventh period was 25 percent. Participants were in-
formed that at the end of the final market period the liquidation value of the asset would
be zero. Again to ensure comparability between the sessions, the last market period was
randomly chosen once for all eight sessions. According to this random selection, it was
determined that each market ended after the ninth period. Each trading period lasted for
180 seconds.
Before the market was opened, participants (i) were told which market (A or B) and
which trading period (1 to 9) were opened and received information about the last average
market price as well as the closing price of the asset traded, (ii) received either positively
framed dividend information or negatively framed information, and (iii) had to predict
the next average trading price of the assets. Phase 2 took about 80 to 90 minutes.

10
 Phase 3: Participants were asked to fill out a computerized postexperimental question-
naire with items designed to measure how well they had understood the experiment and
how much effort they had put into arriving at accurate decisions. Phase 3 took about 15
to 20 minutes.
Phase 4: Finally, participants again had to reveal their certainty equivalent for a lottery
offering a payoff of 100 ECU with a probability of p = .50 and zero ECU otherwise; and to
make seven decisions among lotteries (100 ECU ¤ 0.73). The payoffs were identical with
those used in phase 1 and displayed in Table 2. Phase 4 took about 15 to 20 minutes.

3 Experimental results

3.1 Descriptive data analysis

Over the eight sessions with two times nine trading periods each, participants submit-
ted 6, 983 offers out of which 3, 168 contracts were concluded. Thus, on average partic-
ipants concluded 22 contracts per period (SD = 9.19), ranging from a minimum of
four contracts to a maximum of 68 contracts. The average market price was 368.15 ECU
(SD = 390.71).
Figures 4 and 5 indicate that over the trading periods, the number of concluded con-
tracts decreased both in market A (χ2 (1) = 112.91, p < .001) and market B (χ2 (1) =
73.83, p < .001), while the number of offers not accepted increased in market A (χ2 (1) =
75.02, p < .001) and market B (χ2 (1) = 20.16, p < .05).

[Figure 4 about here.]

[Figure 5 about here.]

One explanation as to why the number of accepted offers decreased would be that
prices increased over trading periods. This conjecture was confirmed; average trading
prices were statistically significantly higher in the last period of both markets, A (MA,9 =

11
235.77, SDA,9 = 216.29) and B (MB,9 = 354.34, SDB,9 = 425.30), compared to the
first period (MA,1 = 150.06, SDA,1 = 83.73; F (1; 649) = 40.52, p < .001; MB,1 =
163.94, SDB,1 = 82.45; F (1; 625) = 50.74, p < .001). However, Figure 5 indicates that
average trading prices on both markets sharply declined in late trading periods when un-
certainty about market duration was important. Thus, uncertainty about market termina-
tion depressed average trading prices, although these were still higher in the last period
than in the first. Figure 6 also indicates that in times of high uncertainty, especially in late
trading periods, the variance of market prices increased.

[Figure 6 about here.]

To control for possible differences with respect to individual risk attitude, it was inves-
tigated whether individual risk attitude differs between sessions and between experimental
conditions with respect to the elicitation method of certainty equivalents and with respect
to the lottery decisions. The average certainty equivalent that was revealed by the subjects
was 44.23 (SD = 31.20), indicating a slight degree of risk aversion. Certainty equivalents
did not differ significantly between the eight sessions (F (7; 56) = 0.48, p = .84). An in-
dex for risk attitude ranging from 0=risk neutrality to 7=risk aversion was computed out
of the seven decisions among lotteries. Participants’ average risk attitude amounted to 3.66
(SD = 2.15). Again no statistically significant difference between the eight sessions was
observed (F (7; 56) = 0.95, p = .47). Nor was there any statistically significant difference
between positively and negatively framed subjects with respect to the certainty equivalent
(F (1; 62) = 0.09, p = .76) and the lottery decisions (F (1; 62) = 0.05, p = .82). Thus,
any differences in observed behavior between experimental conditions are not likely to be
caused by different underlying risk attitudes.
In addition, risk attitude does not differ between the first measurement before the
market was performed and the second measurement after the market, both for the cer-
tainty equivalents (MI = 44.23, SDI = 31.20; MII = 45.58, SDII = 31.91; F (1; 62) =

12
0.15, p = .70) and for the lottery decisions (MI = 3.34, SDI = 2.10; MII = 3.67, SDII =
2.44; F (1; 62) = 1.22, p = .27). The results thus indicate that the market behavior did
not have a recursive impact on individual risk attitude.
Questionnaire results reveal that instructions were clear and easy to understand (M =
7.16, SD = 2.00) and confirmed that participants carefully considered their buying or-
ders (M = 6.13, SD = 1.91) and selling orders (M = 6.09, SD = 2.08). Subjects also
emphasized that they had tried to maximize their earnings (M = 6.83, SD = 1.94). All
questions were formulated as statements that subjects could disagree or agree with (rang-
ing from 1=I do not agree to 9=I fully agree).
In the following, we present the results with respect to (i) framing effects, (ii) the
variation of the probability of the framed information, and (iii) the disposition effect.

3.2 Framing effects

The results confirm our conjecture that positively framed buyers purchase assets from
negatively rather than from positively framed sellers (χ2 (1) = 6.61, p < .01), whereas in
turn negatively framed sellers sell their assets to positively framed buyers rather than to
negatively framed buyers (χ2 (1) = 11.26, p < .001).Tables 3 and 4 show the observed
and the expected trading volume between positively and negatively framed subjects.

[Table 3 about here.]

[Table 4 about here.]

Second, we expected that varying the probabilities of the framed information would
shape individual price expectations. Since the framed dividend information on market A
was more extreme than on market B, it was expected that the trading volume on the former
would be higher than the trading volume of the latter, due to more diverging dividend
expectations.

13
 The results at least weakly support our conjecture. The total number of concluded
contracts was higher on market A than on market B (χ2 (1) = 3.41, p = .07). More pre-
cisely, a total of 1,636 contracts was concluded on market A compared to 1,532 contracts
concluded on market B. However, the total number of offers not accepted did not differ
between the two markets (χ2 (1) = 0.12, p = .73). On market A 1,918 offers were not ac-
cepted by other market participants, whereas on market B 1,897 offers were not accepted.
However, since Figure 6 indicates that participants seemed not to distinguish between
the two asset markets, the observed higher trading volume on market A may be due to the
unbalanced sequence of periods of the two markets. Figure 6 indicates that prices followed
an upward trend on both markets up to the sixth period of market B, and then sharply
decreased in late trading periods. Note that in the beginning of the experiment, when
participants were still highly inexperienced, market A was opened more often than market
B, whereas in later trading periods this pattern was reversed. Thus, the higher number of
concluded contracts may be a result of this particular sequence of trading periods.
We also investigated whether the observed matching of unequally informed subjects
leads to different trading prices. For this analysis we distinguished between the following
trading patterns: trades (i) with negatively framed sellers and buyers, (ii) with positively
framed sellers and buyers, and (iii) with mixed pairs of sellers and buyers. We ran a re-
peated ANOVA with the trading pattern as a between-subjects factor and the market, A
or B, as a within-subjects factor. The data were aggregated by replacing the nine periods
of the two markets by the overall mean of each market. The results indicate that trading
prices were not statistically significantly different between trading patterns (F (2; 21) =
0.07, p = .93), but prices differed across the two markets (F (2; 21) = 28.71, p < .001).
Our findings indicate that objectively irrelevant information influences trading be-
havior: Positively framed buyers purchase assets from negatively rather than from posi-
tively framed sellers, whereas in turn negatively framed sellers sell their assets to positively

14
framed buyers rather than to negatively framed buyers. The observed matching of un-
equally informed subjects, however, does not lead to different trading prices. We also find
that a probability variation of the framed information impacts trading volume. We believe,
however, that this result is primarily driven by the unbalanced sequence of trading periods
across markets.

3.3 Disposition effect

Based on the predictions of prospect theory we expect that a purchase price lower than the
previous market price implies a gain situation that leads to more rapid selling, whereas a
purchase price higher than the previous trading price implies a loss situation that leads to
less rapid selling.
To test this conjecture, two different scenarios, one describing a gain situation and
another describing a loss situation, were distinguished. The gain scenario was defined as
a situation in which the purchase price was below the previous market price, whereas the
loss scenario was defined as a situation in which the purchase price was higher than the
previous market price. The software z-Tree (Fischbacher 1999) used in the experiment en-
ables us to calculate the exact time in seconds between a subject’s buying and a subject’s
selling. It can be shown that market participants who experienced a gain sold their assets
significantly earlier (MG = 10.12, SDG = 23.30) than market participants who expe-
rienced a loss (ML = 13.66, SDL = 26.90; F (2; 1, 306) = 3.01, p < .05). This effect
was mediated by framing (see Figure 7). Positively framed market participants generally
sold their assets later than negatively framed market participants (MP = 13.90, SDP =
27.62; MN = 10.32, SDN = 23.72; F (1; 1, 307) = 6.34, p < .05). Thus, it can be as-
sumed that framing shaped individual expectations and thereby influenced market behav-
ior. Positively framed participants seemed to be more optimistic than negatively framed
participants about the likely performance and profit of the assets and were thus also more

15
patient, both in gain and loss situations.

[Figure 7 about here.]

4 Conclusions

In this paper we investigate the impact of objectively irrelevant information on trading

behavior. Moreover, we draw upon a novel type of framing that is not based on seman-
tic variations of a decision problem. Participants are given complete information about a
distribution, and receive additional percentile information, which either positively or neg-
atively deviates from an aspirational reference payoff. Normative decision theories, such
as the expected utility theory, require that this additional information be neglected in the
decision process so that it does not influence behavior in a market environment. From a
behavioral perspective, however, we expect that the additional information serves as an
anchor in the decision process and thereby systematically influences individual behavior,
even in market environments. We also investigate the impact of a probability variation
of the framed information and the robustness of the disposition effect in a competitive
market environment with real time data available.
Our results (i) indicate that objectively irrelevant information influences individual
trading behavior. Moreover, positively and negatively framed information leads to a par-
ticular trading pattern: Positively framed buyers purchased assets from negatively rather
than from positively framed sellers, whereas in turn negatively framed sellers sold their
assets to positively rather than to negatively framed buyers. The observed matching of un-
equally informed subjects, however, does not lead to different trading prices. We find (ii)
weak support for the conjecture that a probability variation of the framed information
impacts trading volume. However, we believe that this effect is not due to the available
information, but rather to the unbalanced sequence of trading periods of the two asset
markets. Our results (iii) confirm the disposition effect. Participants sold their assets more

16
readily in gain situations than in loss situations. Furthermore, this effect was mediated by
framing. Positively framed market participants generally sold their assets later than nega-
tively framed participants, indicating that the framing of dividend information influenced
individual expectations and thereby also market behavior.
Since objectively irrelevant information influenced market behavior, our findings vi-
olate expected utility theory and the invariance axiom. Moreover, the results are also not
consistent with refinements of the expected utility theory which do not account for fram-
ing effects, such as rank-dependent utility theories (Quiggin 1982), which are similar to
cumulative prospect theory (see e.g. Weber and Camerer (1987), for a more in-depth dis-
cussion). Our experimental approach starts out with how information is perceived, which
is neither captured by expected utility theory nor by rank-dependent utility theories. Both
theories do not capture the perception of information, but only start with the next step,
its processing.
The results of our study may have important implications for financial decision mak-
ing. On financial markets a huge amount of more or less reliable investment information
is available to an increasing number of potential investors all around the world. Standard
finance theory assumes that markets are able to filter out relevant information allowing
individuals to arrive at unbiased decisions. Specifically, standard theory assumes that even
if information does not add to a decision problem, it does not bring negative externalities
to bear on it. Our findings cast doubt on this prediction. Additional irrelevant informa-
tion does not leave the decision problem unchanged, but systematically influences trading
behavior; even in competitive market environments.

17
Notes

1
Overpricing refers to market prices that exceed the total value of the lotteries traded
(Rietz 1993).

2
Tversky and Kahneman (1974) asked subjects to estimate the percentage of African
countries in the United Nations after a number between 0 and 100 had been drawn by
spinning a wheel of fortune. Estimates were dependent on the initially determined num-
ber. In the case of a high number having been selected, the median percentage of African
countries exceeded the corresponding percentage in the case of a low number having been
determined.

3
There is indeed strong experimental evidence suggesting that asset markets are highly
informationally efficient (for a survey on the literature see Sunder 1995).

4
One hundred ECU equal ¤ 0.73.

5
Correspondence between the two measures of certainty equivalents and lottery choices
is investigated, e.g. by El-Sehity et al. (2002) and by Fellner and Maciejovsky (2002).

6
An English translation of the instructions is to be found in the appendix.

18
References

Barber, B. M. and Odean, T.: 2001, The internet and the investor, Journal of Economic
Perspectives 15, 41–54.

Choi, J. J., Laibson, D. and Metrick, A.: 2002, How does the internet affect trading? Evi-
dence from investor behavior in 401(k) plans, Journal of Financial Economics 64, 397–
421.

El-Sehity, T., Haumer, H., Helmenstein, C., Kirchler, E. and Maciejovsky, B.: 2002, Hind-
sight bias and individual risk attitude within the context of experimental asset markets,
Journal of Psychology and Financial Markets 3, 227–235.

Fellner, G. and Maciejovsky, B.: 2002, Risk attitude and market behavior: Evidence from
experimental asset markets, Papers on Strategic Interaction (34). Max Planck Institute
for Research into Economic Systems.

Fischbacher, U.: 1999, Z-tree: Zurich toolbox for readymade economic experiments: Ex-
perimenter’s manual, Working Paper 21. Institute for Empirical Research in Economics
at the University of Zurich.

Heilmann, K., Läger, V. and Oehler, A.: 2000, The disposition effect: Evidence about the
investors aversion to realize losses: A contribution to behavioral finance through the use
of experimental call markets, Mimeo . University of Bamberg.

Heisler, J.: 1994, Loss aversion in a futures market: An empirical test, Review of Futures
Markets 13, 793–822.

Kahneman, D. and Riepe, M. W.: 1998, Aspects of investor psychology, Journal of Portfolio
Management 24, 52–65.

19
Kahneman, D. and Tversky, A.: 1979, Prospect theory: An analysis of choice under risk,
Econometrica 47, 263–291.

Kahneman, D. and Tversky, A.: 1983, Choices, values, and frames, American Psychologist
39, 341–350.

Kühberger, A.: 1995, The framing of decisions: A new look at old problems, Organizational
Behavior and Human Decision Processes 62, 230–240.

Kühberger, A.: 1998, The influence of framing on risky decisions: A meta-analysis, Orga-
nizational Behavior and Human Decision Processes 75, 23–55.

Kühberger, A., Schulte-Mecklenbeck, M. and Perner, J.: 1999, The effects of framing, reflec-
tion, probability, and payoff on risk preference in choice tasks, Organizational Behavior
and Human Decision Processes 78, 204–231.

McNeil, B. J., Pauker, S. G., Sox, H. C. and Tversky, A.: 1982, On the elicitation of prefer-
ences for alternative therapies, New England Journal of Medicine 306, 1259–1262.

Odean, T.: 1998, Are investors reluctant to realize their losses?, Journal of Finance 53, 1775–
1798.

Park, C. Y.: 2001, News media exposure and self-perceived knowledge: The illusion of
knowing, International Journal of Public Opinion Research 13, 419–425.

Quiggin, J.: 1982, A theory of anticipated utility, Journal of Economic Behavior and Organi-
zation 3, 323–343.

Rietz, T. A.: 1993, Arbitrage asset prices and risk allocation in experimental markets, Work-
ing Paper 109. Department of Finance, Kellogg Graduate School of Management, North-
western University, Illinois.

20
Shefrin, H. and Statman, M.: 1985, The disposition to sell winners too early and ride losers
too long, Journal of Finance 40, 777–790.

Statman, M.: 1995, A behavioral framework for dollar-cost averaging, Journal of Portfolio
Management 22, 70–78.

Sunder, S.: 1995, Experimental asset markets, in J. H. Kagel and A. E. Roth (eds), Handbook
of Experimental Economics, Princeton University Press, Princeton, pp. 445–500.

Tversky, A. and Kahneman, D.: 1974, Judgment under uncertainty: Heuristics and biases,
Science 185, 1124–1131.

Tversky, A. and Kahneman, D.: 1981, The framing effect of decisions and the psychology
of choice, Science 211, 453–458.

Tversky, A. and Kahneman, D.: 1992, Advances in prospect theory: Cumulative represen-
tation of uncertainty, Journal of Risk and Uncertainty 5, 297–324.

Weber, M. and Camerer, C.: 1987, Recent developments in modelling preferences under
risk, OR Spektrum 9, 129–151.

Weber, M. and Camerer, C.: 1998, The disposition effect in securities trading: An experi-
mental analysis, Journal of Economic Behavior and Organization 33, 167–184.

Weber, M., Keppe, H.-J. and Meyer-Delius, G.: 2000, The impact of endowment framing
on market prices: An experimental analysis, Journal of Economic Behavior and Organi-
zation 41, 159–176.

21
Appendix: Instructions about the market

Thank you for participating in our experiment. The experiment will last for about 2 hours
and 15 minutes. You will trade assets on a market, whereby your payoff is contingent on
your decisions.
In the following, the trading mechanism is explained in detail. You will learn how
to place buying and selling offers, and how to accept offers by other market participants.
After reading the instructions, there will be time to ask questions. Afterwards there will be
a short test to check whether you have understood the trading rules. The experiment will
not begin until all participants have correctly answered all questions in the test. Then you
will participate on a trial market with two periods: You will have the opportunity to try
out the buying and selling procedures without affecting your payoffs. The two trial periods
will last for six minutes each. After the trial market the real asset market will be opened.
Let us now explain how the asset market works. Generally, there are two possibilities
to buy assets and two possibilities to sell assets.
Let us start with the buying of assets: You can buy assets in two ways: You can either (i)
submit a bid to the market, or you can (ii) accept a standing ask made by another market
participant.
(i) If you want to submit a bid, you have to type your maximal buying price in the
input box “your bid” and press the button “bid” (ii) If you want to accept a standing ask
made by another market participant, you have to press the button “buy.” Standing asks for
the assets are ranked according to prices and listed in columns. Of course, the best offer for
you, and all other potential buyers, is the lowest ask. The lowest ask is listed at the bottom
of the column.
Let us now explain the selling of assets: You can sell assets in two ways: You can either
(i) submit an ask to the market, or you can (ii) accept a standing bid made by the other
market participants.
(i) If you want to submit an ask, you have to type your minimal selling price in the
input box “your ask” and press the button “ask.” (ii) If you want to accept a standing bid
made by another market participant, you have to press the button “sell.” Standing bids for
the assets are ranked according to prices and listed in columns. Of course, the best offer
for you, and all other potential sellers, is the highest bid. The highest bid is listed at the
bottom of the column.
Note that you can engage simultaneously in buying and selling activities. However,
you cannot buy more assets than your cash holdings allow, and you cannot sell more assets

22
than you own. If you have submitted a bid to the market, then your available money for
further activities is reduced by this amount. On the other hand, if you have submitted an
ask to the market, then your available asset holdings are reduced by this one offer. That is,
we do not grant any credit, nor do we allow short selling.
Only improving offers, i. e., higher bids and lower asks, are allowed on the market.
During a trading period you can buy assets, sell assets, or remain passive. You can engage
in all three possibilities at all times. In fact, you can simultaneously submit buying offers
and selling offers, and accept standing offers made by other market participants.
You are also informed about the remaining trading time, the current period number,
and about the previous trades and their trading prices. All trades are chronologically listed
in the column “previous trades.”
You will now have the opportunity to try out the buying and selling procedures with-
out affecting your payoffs. The trial market consists of two periods, each lasting for six
minutes.
[Trial market]

Now the “real” markets will be opened. You will trade assets on two separate markets,
referred to as market A and market B. At the beginning of each trading period you will
be reminded whether market A or B will be opened. The sequence of the market was
determined randomly. Each trading period lasts for 180 seconds.
At the beginning of the markets you will be endowed with 1,000 Experimental Cur-
rency Units (100 ECU equal ¤0.18) plus five assets on each of the two markets, A and B.
Note that the monetary endowment is carried forward on both markets, whereas assets of
type A can only be traded on market A and assets of type B only on market B, respectively.
The minimum number of trading periods for each of the two markets is 8 and the
maximum number of trading periods is 12. The probability that the market ends after the
8th, 9th, 10th, and 11th period is 25 percent. This means there is a chance of 3 over 4 that
the market continues after period 8. Similarly, there is an equal chance that the market
continues after period 9, 10, and 11, once this period has been reached. Only when period
12 is reached, is it certain that this will be the final market period.
Dividends are randomly drawn from a normal distribution with a certain mean and
variance. At the beginning of each period you will be informed about the mean as well
as the variance of the distribution. At the end of the final market period, the liquidation
value of the asset is zero. This means that once the final market period is reached, the assets
carry no intrinsic value; they are worth zero ECU.

23
 In this period dividends are randomly drawn from a normal distribution with a mean of
95 ECU and a variance of 20 ECU.

With a probability of 5 percent the next dividend will be larger (smaller) or equal to 134
ECU (56 ECU). This means that on average in five out of 100 cases the observed
dividend will be larger (smaller) or equal to 134 ECU (56 ECU).

Note: The information provided to subjects in the negative framing condition is displayed
in parantheses.

Figure 1: Available information at the beginning of the first trading period of market A for
positively and negatively framed subjects

24
 Instructions
Instruction
lotteries
lotteries

Certainty
equivalent

7x
Lottery A Lottery a1, a2 Lottery A Lottery a 2, a 1
a 1, a 2 A a2, a1 A

Random
selection

Payoff
information

Instructions
Instruction
asset
asset market
market

T rial market

M arket
A/B

Positive N egative
framing framing
2x9

Prediction

Auction

Payoff
information

Q uestionnaire

Certainty
equivalent

7x
Lottery A Lottery a 1, a 2 Lottery A Lottery a 2, a 1
a 1, a 2 A a2, a1 A

Random
selection

T otal payoff

Figure 2: Sequence of events in the experiment

25
 Rem aining Tim e : 12

Y our Current D eviation

Purchasing Price M arket Price A sset M arket
G uilders 306

B
A sset B 5 40 75 + 35

Asks M ark et Prices B ids

90 50 65

85 40 66

75 75
YIhr
ou r
Y our Bid
Kau fangebot
Ask

70 9 6

Ask Buy Sell Kaufangebot

B id

Figure 3: Computer screen for the auction

26
 Percentage

80
70
60
50
Accepted offers
Reihe1
40
Reihe2
Not accepted offers
30
20
10
Period
0
1 2 3 4 5 6 7 8 9

Figure 4: Percentage of accepted and not accepted offers for market A

27
 Percentage

80
70
60
50
Accepted offers
Reihe1
40
Reihe2
Not accepted offers
30
20
10
Period
0
1 2 3 4 5 6 7 8 9

Figure 5: Percentage of accepted and not accepted offers for market B

28
 Average
trading prices
Standard
800 deviation

700

600
500

400

300

200

100 Period
0
1A1 B21 32
A 43
A 5A4 6B 2 A
75 83
B 94
B 10
A6 11
A7 12
B 5 13
B 6 14
B 7 15 168 17
B8 A B 9 18
A9

Figure 6: Average trading prices and standard deviations for market A and market B across
periods

29
 Time in seconds
Time in
16
14
12
10
Reihe1
Negatively framed participants
8
Reihe2
Positively framed participants
6
4
2
0
Scenario
Scenario
1
Gain scenario
2
Loss scenario

Figure 7: Average time difference between a subject’s buying action and the next selling
action for positively and negatively framed participants in gain and loss scenarios

30
Table 1: Positive and negative dividend information of markets A and B for all periods

Market and σ1 = 20 Market and σ2 = 30 Market and σ3 = 40

(Period) Xp − Xp (Period) Xp − Xp (Period) Xp − Xp
µ1 = 95 A(1) 56 - 134 A(2) 36 - 154 A(3) 17 - 174
B(9) 92 - 98 B(8) 91 - 99 B(7) 90 - 100
µ2 = 135 A(4) 96 - 174 A(5) 76 - 194 A(6) 57 - 214
B(6) 132 - 138 B(5) 131 - 139 B(4) 130 - 140
µ3 = 105 A(7) 66 - 144 A(8) 46 - 164 A(9) 27 - 184
B(3) 102 - 108 B(2) 101 - 109 B(1) 100 - 110

31
 Table 2: Lottery payoffs in Experimental Currency Units

Lottery Payoff p Expected value

1 a1 160 .20 88
a2 70 .80
A 88 1.00 88
2 a1 150 .32 99
a2 75 .68
A 99 1.00 99
3 a1 178 .28 106
a2 78 .72
A 106 1.00 106
4 a1 140 .35 101
a2 80 .65
A 101 1.00 101
5 a1 135 .40 105
a2 85 .60
A 105 1.00 105
6 a1 188 .25 98
a2 68 .75
A 98 1.00 98
7 a1 130 .30 102
a2 90 .70
A 102 1.00 102

Note: A denotes the certain payoff, whereas a1 and a2 denote the risky payoff of the lottery.

32
Table 3: Observed and expected trading volume between negatively framed sellers and pos-
itively and negatively framed buyers

Positively framed buyers

Observed trading volume Expected trading volume
Negatively framed sellers 870 820.4
Positively framed sellers 634 683.6
P
1,504

33
Table 4: Observed and expected trading volume between negatively framed sellers and pos-
itively and negatively framed buyers

Negatively framed sellers

Observed trading volume Expected trading volume
Positively framed buyers 905 839.5
Negatively framed buyers 634 699.5
P
1,539

34 View publica