Description
A series of laboratory double auction experiments is conducted to examine whether the order of information
releases affects market prices. Behavioral research on belief revision has shown that individuals are
influenced by the order in which a series of information items is presented. The experiments are designed
to provide a controlled investigation of whether order effects as displayed by individuals also can influence
prices in a market setting where outcomes are not a simple aggregation of individual behavior.
Signiticant evidence is found of a recency effect in the experimental asset markets.
Pergamon
Accounting, 0rganiration.s and Soctety, Vol. 22, No. 1, pp. 89-103, 1997
Copyright c 1996 Hsevier Science Ltd
Printed in Great Britain. All rights reserved
0361.3682/96 $17.00+0 00
PII: SO361-3682(96)00026-S
AN EXAMINATION OF MARKET EFFICIENCY: INFORMATION ORDER
EFFECTS IN A LABORATORY MARKET
BRAD TUTTLE, MARIBETH COLLER
Uni versi ty of South Carol i na
and
F. GREG BURTON
Uni versi ty of Nebraska
Abstract
A series of laboratory double auction experiments is conducted to examine whether the order of infor-
mation releases affects market prices. Behavioral research on belief revision has shown that individuals are
influenced by the order in which a series of information items is presented. The experiments are designed
to provide a controlled investigation of whether order effects as displayed by individuals also can influ-
ence prices in a market setting where outcomes are not a simple aggregation of individual behavior.
Signiticant evidence is found of a recency effect in the experimental asset markets. Copyright (Q 1996
Elsevier Science Ltd
The efficient market hypothesis states that mar-
kets impound new information quickly and
without bias into security prices. In this paper,
we report the results of a series of laboratory
double-auction experiments conducted to
examine whether the order of information
releases can affect market prices. In essence,
the study examines the descriptive validity of
the efficient market hypothesis against the
alternative hypothesis that systematic individual
biases can survive in a market setting. Prior
research has examined this implication of the
efficient market hypothesis using archival
exchange data. This stream of research, how-
ever, has provided mixed results. For example,
DeBondt & Thaler (1985) claim that stock pri-
ces overreact to new information and support
this hypothesis with exchange data. Although
DeBondt & Thaler (1987) examine the issue
further and find no evidence that their earlier
result is a manifestation of a size effect, Zarowin
(1989); Zarowin (1990) find that size may be
responsible for the DeBondt & Thaler (1985)
result.
Several explanations may be advanced for the
mixed results in exchange-based studies inves-
tigating market efficiency with respect to parti-
cular information items. The first relates to the
difficulty of gaining adequate control over
extraneous variables when using exchange-
based data. A second explanation is that the
pricing models used to determine the expected
efficient market price may themselves be mis-
specified. A third explanation relates to the
logical difficulty in assuming an information
item is valuation relevant and then testing for
efficient use of that item. If the item is not
relevant to valuation, then even the most effi-
cient market will not impound it into security
prices.
Authors are listed in reverse alphabetical order. We wish to thank John Dickhaut, Daniel Friedman, Adrian Harrell, Glenn
Harrison, Julia H&s, John Reisch, Charles S&nit&in, Earl Spiller, and participants at the 1995 Economic Science Association
meeting for comments on prior versions of this paper.
89
90 B. TUlTLE et al .
These difficulties inherent in exchange-based
research suggest that a potentially fruitful
direction is the use of laboratory methods.
Laboratory markets provide for direct control in
manipulating certain variables of interest while
eliminating other variables (or holding them at
a constant level).’ In the laboratory, it is also
possible to conduct a market where estimation
of an efficient price is not necessary for the
analysis and where the information is valuation
relevant by design. We accomplish this by
examining two markets that receive exactly the
same (valuation relevant) information, but in a
context where inefficiencies in information
processing could cause the different markets to
arrive at different prices.2 The literature on
individual decision making suggests that this
may be possible, for example, by manipulating
the order in which the information is received.
Prior behavioral research shows that when
individuals are presented with a series of infor-
mation items, their decisions are systematically
influenced by the order in which the items are
presented (Hogarth & Einhorn, 1992). These
effects have proven robust and have been
observed in various settings including social
judgments and attributions for behavior (e.g.
Feldman & Bernstein, 1978; Lichtenstein &
Srull, 1987; Luchins & Luchins, 1984) mock
legal trials (Put-&am, 1986; Tetlock, 1983;
Walker et al , 1972) judgments of probability
(Shanteau, 1970, 1972) and auditing (Asare &
Messier, 1991; Ashton & Ashton, 1988; Tubbs
et al , 1990). Despite the fact that investors in
organized markets regularly receive streams of
information that they use to revise their valua-
tion estimates, no prior research has examined
order effects in a market context3 This raises
the possibility that the order effects observed in
individual decision making may also occur in
asset pricing within a market. The objective of
the current study is to examine this possibility
in a laboratory market.
We conduct a series of experiments in which
participants trade shares of stock in fictitious
firms over five trading periods where relevant
information about the firm is publicly
announced between each of the trading peri-
ods. In one version of the experiment, the
information consists of two “good” information
items followed by two “bad” information items.
In the other version, the exact same informa-
tion is released, but in the opposite order, i.e.
bad followed by good. The subjects use this
information to update their beliefs regarding
the final dividend and, hence, to revise their
estimates of the security price. If these markets
are efficient, then the final valuation of security
prices should not differ between two markets
that receive different orders of the same infor-
mation. If, on the other hand, the market
mechanism does not remove individual biases
of this type, the final price should differ pre-
dictably depending on the order of information
arrival.
Research of this type is important for several
reasons. First, it is important to determine the
conditions under which economic theories,
such as market efficiency, are more or less
likely to hold. This is a necessary step towards
isolating the precise market mechanisms that
produce market efficiency. Second, this
research contributes to the literature on indivi-
dual belief revision by extending its findings to
a market setting. The competitive process
inherent in the market often is thought to
‘Similar arguments have been advanced in other laboratory studies investigating issues that have traditionally been exam-
ined with exchange based data (see, for example, Bloomfield, in press, who examines the effects of information asymmetry
on the bid-ask spread, and Coller, 1996, who examines pricing effects of noisy information releases).
aBloomfield and Libby, in press, use an experimental market to address another issue of market efficiency: whether prices
fully aggregate information held by differently informed investors. They compare prices between markets where the infor-
mation available to the market as a whole is the same, but where the availability of each information item is manipulated.
They also construct each of their three information items to be valuation relevant. They find that the distribution of infor-
mation among investors does affect prices.
3A developing stream of experimental research is beginning to examine certain other cognitive biases in market contexts.
For example, Ganguly et al . (1994) examine the persistence of the base-rate fallacy in an oral-double auction, and
Kachelmeier (1995) examines the effects of sunk costs on trading behavior in a computerized double auction.
influence how individuals formulate and revise the similarity of the present evidence to it. This
their beliefs, mitigating or even eliminating representativeness heuristic has been used to
cognitive biases. Whether or not the market explain departures from Bayes’ rule in various
process “trains” successful traders to overcome types of individual decision making.* Hogarth &
their individual biases is still an open question. Einhom (1992) suggest that another decision
Third, companies that are considering the tim- irrelevant variable, the order in which informa-
ing of their information and investors who are tion is received, systematically influences indi-
the recipients of the information wish to know vidual judgment. Although substantial evidence
what type of effect, if any, information order indicates that individual decision making often
has on the valuation of their securities. is inefficient, markets generally are assumed to
be (semi-strong) efficient even if individual par-
ticipants within the market are biased.
BACKGROUND Camerer (1987) summarizes several argu-
ments in defense of market theories despite
Jensen (1978, p. 96) defines an efficient market evidence indicating the irrationality of indivi-
as follows: “A market is efficient with respect to duals: (1) market participants have enough
information set 8, if it is impossible to make financial incentive and experience to avoid
economic profits by trading on the basis of mistakes; (2) random mistakes of individuals
information set &” The widely accepted semi- cancel out; (3) only a small number of rational
strong form of the efficient market hypothesis individuals are needed to drive the market to
further holds that markets are efficient with rationality; (4) less rational individuals learn or
respect to the information set that includes all follow the actions of rational individuals; and
publicly available information. When informa- (5) agents who are less rational are driven from
tion is widely known to the participants in the the market. Most of these arguments, however,
market, competition will drive prices so that may be countered. For example, some investi-
participants are unable to consistently earn gations have observed biased individual deci-
economic profits by trading on that informa- sion making with highly experienced subjects
tion. Economic profits are assumed to be net of and financial incentives (Grether, 1980). Addi-
all costs. If storage, transaction, and information tionally, most psychological biases are systema-
costs are zero in a market that is efficient, the tic, not random, and therefore may not cancel
expected price of an asset is equal to the sum of each other out in the aggregate. Furthermore,
its discounted future cash flows. Thus, widely institutional constraints on trading may prevent
available information that predicts future cash sufficient capital from being exerted by the
flows should be completely assimilated into the limited number of rational agents to make a
price of an asset. Conversely, such information difference. Finally, new traders, who pre-
received in a market that is not efficient may sumably are less likely to be rational, are con-
not result in a market price that properly stantly entering the market.
reflects the asset’s anticipated future cash Some evidence suggests that individual biases
flows. can in fact survive in a market (Anderson &
Although markets generally are assumed to Sunder, 1995; Camerer, 1987). Arrow (1982,
be efficient information processors, individuals p. 5) asserts that the representativeness bias ((...
are not. Research primarily from psychology typifies very precisely the excessive reaction to
suggests that individuals can be biased by irre- current information which seems to character-
levant variables. For example, Tversky & Kah- ize all the securities and futures markets.”
neman (1974) suggest that people base their DeBondt & Thaler (1985, p. 794), in reviewing
judgments of the likelihood of a future event by the work of Shiller (1981) and others, conclude
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET 91
‘Plott & Wide (1982) find evidence of the representativeness heuristic in a market setting.
92 B. TUlTLE et al.
that “... investors seem to attach dispropor-
tionate importance to short-run economic
developments.” DeBondt & Thaler (1985) also
cite evidence that the expectations of profes-
sional security analysts and economic fore-
casters have an overreaction bias. DeBondt &
Thaler (1985) explain that if stock prices sys-
tematically overreact to new information, then
their reversal should be predictable from past
return data alone: extreme price movements
should be followed by price movements in the
other direction, and the more extreme the
initial movement, the greater the subsequent
adjustment. This assertion is supported empiri-
cally, and DeBondt & Thaler (1985) conclude
that their data are consistent with the predic-
tions of the overreaction hypothesis. DeBondt
& Thaler (1987) provide further evidence that
their earlier finding is not simply a manifesta-
tion of a size effect and/ or differences due to
risk. Later research (Zarowin, 1989; Zarowin,
1990), however, finds that size may indeed be
responsible for the DeBondt & Thaler (1985)
finding.
The conflicting findings of studies using
exchange-based data may be attributable in part
to three difficulties inherent in such research:
(1) the lack of adequate control over potentially
confounding exogenous factors; (2) mis-
specifications in the assumed asset pricing
models; and (3) the inability to discriminate
between the market’s use of an item and the
importance of the item itself. Bernard & Schipper
(1994) discuss these general problems and note
that, “In principle, many of the difficulties faced
in interpreting research based on archival data
can be overcome in an experimental setting.”
Control over exogenous variables is achieved
in our study by employing a laboratory market.
This research method provides for the explicit
manipulation of the nature and timing of infor-
mation releases where the information is by
design valuation relevant. The experiments are
designed to allow a test of the market efficiency
assumption without requiring the exact func-
tional form of the link between information and
cash flows to be known. Specifically, we con-
duct an asset market that receives four pieces
of valuation relevant information. The tirst two
pieces of information suggest an increase in
future cash flows (good news). These are fol-
lowed by two items of information that suggest
a decrease in future cash flows (bad news). In
this situation, one would expect the market
price for the asset to first rise and then to fall,
the exact amount of change in price being
determined by the weight that the market pla-
ces on each piece of information. We then
conduct a second market for the same asset as
in the first market. This second market receives
the same information as does the first, but in
the opposite order. In this case, one would
expect the market price to first fall and then to
rise. The arrival of both markets at similar pri-
ces after all information is received and pro-
cessed would be consistent with efficient
information use. If, however, the final price
differs between markets, there is evidence of a
bias in the market’s processing of information.
Hence, by examining a situation in which order
might influence judgments, it should be possi-
ble to detect related market inefficiencies with-
out the need to use models of expected future
stock prices.
The decision-making literature has shown
that the order of information influences the
judgments of individuals in a predictable
manner. Hogarth & Einhom (1992) present a
model of sequential belief revision in which
predictions of order effects are based on a tax-
onomy of tasks along three dimensions: task
complexity, series length, and response mode.
Simple tasks typically are those in which the
information consists of a simple trait adjective
or single number and where the individual is
familiar with its meaning. Complex tasks
involve more or unfamiliar stimuli. Short series
involve between 2 and 12 pieces of informa-
tion. Long series have generally involved over
20 items. One type of response mode consists
of making a judgment after receiving each
information item and is termed “step-by-step”
processing. Another response mode, known as
“end-of-sequence” processing, consists of
processing alI pieces of information prior to
making a single, final judgment. Complex
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET 93
information received in a relatively short series
that is used to adjust beliefs in a step-by-step
mode characterizes many real-world capital
market conditions and is the task used in the
present study. Under these conditions, the
Hogarth & Einhorn (1992) model predicts a
particular order effect, the recency effect. That
is, recently received information receives
more weight than previously received infor-
mation.
The Hogarth and Einhorn model has impor-
tant implications for decisions involving
streams of mixed information where some
items are positive and some are negative. When
negative information is followed by positive
information, i.e. information order -+5, the
positive information will have the greater
impact because it is most recent. Conversely,
negative information will have a greater impact
when it follows positive information, i.e. infor-
mation order +--. Recency is implied because
the final position for the +- order is lower than
the final position for the -+ order. The combi-
nation of the two effects produces what has
become known as the “fishtail” result (Fig. 1). If
this model of individual sequential belief revi-
sion also describes market behavior, then a
market that receives good then bad news will
arrive at a lower asset price than a market that
receives bad then good news, when the spe-
cific items of information are identical.
The experiments described below seek to
determine whether this model of individual
belief revision also can describe price behavior
in an asset market. The extent to which propo
sitions based on individual behavior generalize
to aggregate market setting remains an unre-
solved issue (Camerer, 1987). Anderson 8t
Sunder (1995, p. 186) point out that market
behavior is not a simple aggregation of indivi-
dual behavior and state further that, “... it does
not follow that biased individual behavior
implies biased aggregate market outcomes.”
Although the step-by-step processing in asset
markets is parallel to the Hogarth and Einhorn
model, it is an open question whether any
resulting individual biases will affect the market
outcome.
The above arguments suggest the following
alternative hypothesis to the efficient market
hypothesis:”
Ha: Market price changes will be influenced by the
order of information arrival such that positive infor-
mation followed by negative information results in- a
lower price than the same information but received in
the order of negative information followed by positive
information.
THE EXPERIMENT
Procedure
The laboratory asset markets were conducted
using software developed at the University of
Arizona’s Economic Sciences Laboratory for
double-auction experiments.’ Each run of the
experiment consisted of four independent
rounds of trade where six traders were able to
make both bids and offers for shares of stock.
During each round, trade was conducted in a
separate company: round 1 was “Company A,”
round 2 was “Company B,” round 3 was
“Company C,” and round 4 was “Company D.”
At the beginning of trade in each company,
each subject was endowed with $30 (experi-
mental dollars’) and two shares of stock.
sWe follow the convention of labeling information as “+” for good news and “ -” for bad news.
60ne diiculty in analyzing order-effect data when order treatments are between subjects is that initial differences can
mask effects (Hogarth & Einhorn, 1992). For this reason, research using the sequential belief revision paradigm focuses on
the difference between the initial belief and the ending belief. Typically, this is accomplished by subtracting the initial belief
from subsequent responses, and comparisons between groups consist of changes from the initial belief. Accordingly,
references to market price in the present paper refer to price changes (i.e. the difference between first period price and
subsequent prices). The analytical results from Hogarth & Einhom (1992) suggest that this is appropriate because the initial
position (i.e. high or low) should have no effect on the size of the recency effect (Ashton & Ashtoa, 1988 p. 629).
‘Complete instructions for the asset markets are available from the authors upon request.
@Ihe exchange rate used in calculating cash payments was set at one experimental dollar=O.l IJS. dollar. Of the $30
endowment, $20 was an interest free loan that was repaid at the end of the trading round.
94 B. TUTTLE el al.
-3’
Period 1 Period 2
I
Period 3
I
Period 4
I
Period 5
Fig 1. Hypothesized order effect.
Information regarding the fmal liquidating divi-
dend to be received at the end of trade in each
company also was presented. This information
was identical for each company and was pre-
sented as a table containing possible dividends
and the associated probability of each. The
dividend distribution as shown to the subjects
is presented in Table 1. There are nine possible
dividends, ranging from $4 to $20, with the
expected value being $12.
TABLE 1. Dividend distribution used for each company.
Company A-Period 1: dividend expectations
Dividend amount Probability
$4
$6
$8
$10
$12
$14
$16
$18
$20
5%
5%
10%
15%
30%
15%
10%
5%
5%
Within each of the four trading rounds, there
were five three-minute trading periods. After
the first trading period in each round, trade
stopped and information regarding the com-
pany was publicly announced (both verbally
and shown via an overhead projector). Follow-
ing the release of information, the traders were
allowed time to consider its impact. After
everyone indicated they were ready to con-
tinue, they hit the ENTER key to begin another
three-minute trading period. At the conclusion
of this period, the market stopped, and the next
piece of information was announced. This pro-
cess continued until five periods of trade were
completed (with four information items
announced between periods). At the conclu-
sion of the fifth period of trade, the liquidating
dividend was publicly announced9, profits were
calculated, the market was re-initialized, and
trade began for the next company. The specific
information item released constitutes the order
manipulation as described below. Eight different
information items were used in the experiment.
These are listed in Table 2 and will be referred
to by the numbers presented in the table.
?lre actual dividend realized was equal to the expected value and was determined by “shifting” the expected value toward
the higher dividends after good news and toward the lower dividends after bad news by equal amounts ($3). For companies
with two items of good news (and two items of bad news), this results in the conditional expected value (after aIf infor-
mation is received) being identical to the unconditional expected value. Hence, the dividends paid to subjects were con-
sistent with the instructions provided to them.
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET 95
TABLE 2. Information items
Good news items
+1.
+2.
+.j.
+4.
Bad new items:
-5.
-6.
-8
An easing of fuel prices is expected to result in much lower shipping costs than expected during the
upcoming busy season. Because freight is a large portion of the company’s operating costs, a signiticant
savings will be realized.
Several of the company’s main competitors have manufacturing facilities located in the Southeast.
Recently, tropical storms severely damaged this area causing their facilities to shut down. As the com-
pany’s manufacturing is located elsewhere, it is expected to capture the competitors’ lost sales.
The company’s suppliers have unexpectedly reduced the cost of a primary component in a major
product line. This is expected to lower product costs substantially.
The company just introduced an important new product a full three months ahead of schedule. The
company’s competitors are not expected to have their own version of the product ready to release
until sometime next year. Consequently, the company expects to obtain much of the market.
A major customer has filed bankruptcy proceedings. This single customer represents about 15% of the
company’s total accounts receivable.
A competitor has introduced a revolutionary new product. The company is not expected to have its
own version of the product ready for some time. Consequently, the company is expected to lose sub
stantiaf market share.
The company has just been forced to recall nearly all of its latest model products to repair a mechanical
problem. Not only will this be very expensive, but the competition is taking full advantage of the
situation by aggressively pursuing the company’s customers.
The only source of customized parts for an important product will be unable to meet demand for the
next two quarters. This condition will reduce the company s sales by a considerable amount as no
other viable sources exist in the short term
The purpose of the experiments was to
examine price changes in a market that is
allowed to interpret information as it sees fit.
Therefore, no specific adjustment mechanism
was provided, nor was any direction offered on
how to weight any particular piece of informa-
tion. Subjects were told only that good (bad)
news would indicate a greater likelihood of a
higher (lower) dividend being realized. Many
laboratory markets specify a particular adjust-
ment procedure to be used in price formation.
The current study differs from prior ones in that
the absence of a specific adjustment procedure
is integral to our research question. Further-
more, this situation is more like that existing in
organized exchanges. In those settings, infor-
mation is often released that is not quantita-
tively specific, and the impact on market prices
must be “subjectively” determined by the market.
The first trading round, Company A, was
conducted “for practice”. Subjects were told
that they would not receive cash for their prof-
its from this round, although profits would be
calculated just as in the later rounds. This prac-
tice round was conducted to provide an
opportunity for the subjects to familiarize
themselves with the market and to practice
entering and accepting bids and offers without
concern for how their cash profits might be
affected. Subjects were paid all of the profits
earned during the remaining three rounds.
Average profits were $10 U.S. for the twohour
experiment plus a $3 U.S. participation fee.
Order of information manipulation
The experimental design was a repeated mea-
sures design (four rounds of trade) with order
of the information items manipulated between
subjects (two versions of the experiment). The
order manipulation was not imposed until the
third round as described below. The distin-
guishing feature of each trading round is the
specific information presented for each com-
pany. The sequence of events and the informa-
tion items released in each of the two versions
of the experiment are presented in Table 3. As
B. TUTTLB el al.
TABLE 3. Experimental design
Company/Round Period Information items
Version 1 Version 2
A (Practice round) 1 Dividend distribution Dividend distribution
2 +3 +3
3 -5 -5
4 +1 +1
5 +4 +4
End Dividend declared Dividend declared
Dividend distribution Dividend distribution
-7 -7
+2 +2
-6 -6
-8 -8
Dividend declared Dividend declared
D 1
2
3
4
5
End
Dividend distribution Dividend distribution
+1 -5
+2 -6
-5 +1
-6 +2
Dividend declared Dividend declared
Dividend distribution Dividend distribution
-7 +3
-8 +4
+3 -7
+4 -8
Dividend declared Dividend declared
Note: Information item numbers are preceeded by a sign designating whether it conveys positive or negative information to
the market. Information item numbers are presented here strictly as an expositional aid; subjects were not presented with
item numbers.
shown in Table 3, both versions received the
same information, in the same order, for Com-
panies A and B. For Company A, information
was provided in the order of +-++ (informa-
tion items +3,-5,+1,+4). For Company B,
information was provided in the order of
-+-- (information items -7,+2,-6-8). The
data generated in the first two rounds were not
used in hypothesis testing, but were used to
check whether the information manipulation
was successful. These early rounds also were
conducted to provide the subjects with an
opportunity to become familiar with the market
and to gain experience trading. Prior research
(Williams, 1980) ,demonstrates that traders
may require a training period to become familiar
with the task in computerized asset markets.
Note that all eight information items were
presented during the first two rounds so that
the subjects had experience with each piece of
information prior to the rounds used for data
analysis.
The two versions of the experiment differed
only in the order of the information presented
for Companies C and D. These final two rounds
of trade were designed to provide the strictly
controlled test of information processing effi-
ciency in the markets. For Company C of Ver-
sion 1, the information items were presented in
the order of ++-- (items +1,+2,-5,-6). In
Version 2, these items were presented in the
order of --++ (items -5,-6,+1,+2). By com-
paring the market price changes for Company
C in Version 1 versus those in Version 2, it is
possible to determine under strictly controlled
conditions whether the order of information
arrival per se affects the market price. Similarly, information arrival in the context of thinking
for Company D of Version 1, the information about appropriate stock prices.
items were presented in the order of --++ The pretest participants also were asked to
(items -7,~8,+3,+4); in Version 2, the items complete a manipulation check after providing
were presented in the order of ++-- (items their market prices. Here, they rated each of the
+3,+4,-7,-8). Each experiment version was eight information items as being “good news”,
run three times (a total of six experiments was “bad news”, or “can’t tell”. Ninety-six percent
conducted).” of the responses correctly identified the infor-
Pretests were used to select information mation as being good news or bad news. Of the
items that could be easily interpreted as good remaining 3.5%, all were in the “can’t tell”
or bad and that would be perceived to have category; no responses were in the direction
similar importance in judging company perfor- opposite the intended manipulation. The
mance. Prior to conducting the laboratory asset results from this check provided assurance that
markets, pretests also were conducted to the asset market subjects would perceive the
determine whether individual decision makers information items as intended.
exhibit order effects in this particular task.
Forty-four students completed an individual Subjects
survey asking for their estimate of market pri- The laboratory markets were conducted at the
ces for the securities of Companies A through D University of South Carolina and at the Uni-
under the same design as described above. versity of Nebraska. The experiments were
Pretest subjects first were presented with the designed for six traders, although one of the six
dividend distribution (as used in the asset mar- runs was conducted with only five (Run 2 of
kets) for the security and asked to state a mar- Version 1). Thus, a total of 35 subjects partici-
ket price. They next were provided with the pated. Participants were recruited from under-
same information in the same order as used in graduate business classes. Approximately haIf
the asset markets (one version of the pretest of the subjects were female. Just under half
corresponded to market Version 1, and the were accounting majors, with the remaining
other corresponded to market Version 2). After business disciplines evenly represented. The
receiving each new information item, pretest average participant had taken three or more
subjects answered the open question, “In light courses related to accounting, finance, compu-
of this new information, what do you believe ters and business policy.
should be the market price for a share of stock
in Company X?” The mean change between the
initial price provided in period 1 and the ending RESULTS
price provided in period 5 is significantly dif-
ferent between Version 1 and Version 2 for Preliminary analysis
both Company C (ti6.230, RO.001) and Com- Initial tests were conducted to assess the
pany D (t=2.819, eO.037). The mean percen- homogeneity of subjects across treatment con-
tage changes are in the predicted direction ditions (Version 1 vs. Version 2). Subjects do
such that prices indicated under the informa- not differ significantly (Chi-square tests:
tion order of ++-- changed in a negative -0.10) between treatment conditions as to
direction overall, and prices indicated under their major field of study, class standing (gradu-
the information order of --++ changed in a ate vs. undergraduate), or business courses
positive direction overall. These pretest results taken with the exception of Introduction to
suggest that individual decision makers are Finance. Neither do they differ significantly
systematically inlhtenced by the order of (t=1.474, -0.30) in age. Further support for
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET 97
“*The versions were administered in random order within the constraint that three runs of each version were conducted.
98 B. TU’lTLE el al.
TABLE 4. .Information items and manipulation check results, subject frequencies by response
Note: the items appeared in randomized order in the manipulation
Information represents:
check. No “+” or “-” designation appeared.
Good news Can’t tell Bad news
+ 1. An easing of fuel prices is expected to result in much lower 34 1 0
shipping costs than expected during the upcoming busy season.
Because freight is a large portion of the company’s operating
costs, a significant savings will be realized.
+2. Several of the company’s main competitors have manufactur- 31 1 3
Ing facilities located in the Southeast. Recently, tropical storms
severely damaged this area causing their facilities to shut down. As
the company’s manufacturing is located elsewhere, it is expected
to capture the competitors’ lost sales.
+3. The company’s suppliers have unexpectedly reduced the cost
of a primary component in a major product line. This is expected
to lower product costs substantially.
+4. The company just introduced an important new product a full
three months ahead of schedule. The company’s competitors are
not expected to have their own version of the product ready to
release until sometime next year. Consequently, the company
expects to obtain much of the market.
-5. A major customer has filed bankruptcy proceedings. This sir-
gle customer represents about 15% of the company’s total
accounts receivable.
29
33
-6. A competitor has introduced a revolutionary new product.
The company is not expected to have its own version of the pre
duct ready for some time. Consequently, the company is expected
to lose substantial market share.
-7. The company has just been forced to recall nearly all of its
latest model products to repair a mechanical problem. Not only
wilI this be very expensive, but the competition is taking full
advantage of the situation by aggressively pursuing the company’s
customers.
-8. The only source of customized parts for an important product,
will be unable to meet demand for the next two quarters. This
condition wiII reduce the company’s sales by a considerable
amount as no other viable sources exist in the short term.
5
2
0
2
1
0
27
34
35
33
the similarity of subjects assigned to Versions 1
and 2 is obtained by showing that their
responses on Companies A and B are insig-
nificantly different. Mean contract price
changes do not differ significantly between
groups for Company A @=0.893, P=O.45) or
Company B (tiO.419, P=O.70). These data sug-
gest that the subjects are homogenous across
treatments.
The good versus bad information manipula-
tion was tested using data collected at the con-
clusion of the experiment with the same
questionnaire as in the pretest. Table 4 shows
“bad news”, or “can’t tell” for each piece of
information. Of the total 280 responses (35
subjectsx8 information items), 256 are in the
correct cell. In only 4 cases did a subject
respond in the opposite direction. “Can’t tell”
responses are distributed evenly among the
market sessions. “Good news” responses were
coded as a 1, “can’t tell” responses were coded
as 0, and “bad news” responses were coded as
- 1. Using this data, a t-test shows that the mean
response on the four positive information items
is significantly higher (k40.505, RO.001) than
the mean response on the four negative infor-
the frequency of responses under “good news”, mation items.
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET
3-
-l- “‘.6x’
-2 - I I I I
Period 1 Period 2 Period 3 Period 4 Period 5
99
--o- B: -+--
Fig. 2. Mean price changes for Companies A and B.
As a further manipulation check, data from
the first two rounds (Companies A and B) were
initially examined to determine if the market
revised their beliefs in the appropriate direction
following the information releases. Recall that
all eight information items were presented dur-
ing these two rounds. Figure 2 presents the
mean price change (across all six experimental
runs) for each period of each of these rounds.
Mean contract prices follow the information
items quite well. For example, for Company A,
the market first received one item of positive
information followed by one item of negative
information. This is depicted in the figure as a
rise in the solid line followed by a decline. The
solid line also reflects the two subsequent posi-
tive pieces of information received prior to
trading in periods 4 and 5. Figure 2 provides
additional assurance that the subjects under-
stood the information and knew how to use it
in the market. Together, the data suggest that
the subjects understood the information items
and that the manipulation of information
valence was achieved.
Mai n resul ts
The market efficiency hypothesis suggests that
changes in market price between the end of
period 1 and the end of period 5 will not differ
between experiment versions that differ only in
the order of information arrival. If, however,
the market does not eliminate individual biases,
then an order effect should occur. This alter-
native hypothesis suggests that the price
change in the rounds with ++- - information
order will be substantially lower (or negative)
than the price change in the rounds with - - ++
information order. Figure 3 depicts the price
changes in the third round (Company C) and
Fig. 4 shows the changes in the fourth round
(Company D). Immediately obvious in both
rounds are the “fishtails” associated with a
recency effect.
The differences in price changes due to order
(Fig. 2 and Fig. 3) were tested for statistical sig-
nificance using a 2 x2 ANOVA. The data were
first tested for an overall information order
effect using the mean price changes, from per-
iod 1 to period 5, in Rounds C and D as the
dependent variable (A’= 12; 2 rounds from each
of 6 runs). Information order (++ - - vs. - - ++>
and round (C or D) served as the independent
variables. The overall model is significant
(F=10.35, P=O.OOS). Information order is sig-
nificant (F=16.64, P=O.O03) suggesting that the
order of Information influenced the movement
of market prices. The mean price changes
exhibit a recency pattern: in the ++-- order,
mean price change is -1.125; and in the --++
order, mean price change is +2.288. Neither
B. ‘I’VITLE et al.
‘r
P
3-
,,’
,’
2-
:’
.’
,’
..A.’
_
.._
. ..&.’
13
- 1
I
P-ziod 1 Period 2 Period 3 Period 4 Period 5
Isl:rt-_n2;--.tl
Fig. 3. Mean price changes for Company C.
_3L___‘-__‘-
J
Period 1 Period 2 Psrlod 3 Period 4 Period 6
Fig. 4. Mean price changes for Company D.
round nor the interaction term are significant,
signifying that the effect is unlikely to be
dependent on the particular set of information
items.
Additional tests for differences in changes in
the mean market prices (between period 1 and
period 5) between information orders within
individual rounds were conducted (N=6 for
each test). In Round C, the mean price change
in the ++--- order (-0.319) is significantly
lower (I=-3.351, one-tailed P=O.O14) than the
mean price change in the --++ order
(+3.16&I). In Round D, the difference in mean
price change in the ++-- order (-1.931) is
significantly lower (r=-2.323, one-tailed
EO.040) than the mean price change in the
- -++ order (+1.407). In both rounds, the
data reveal significant effects for order of
information.
To provide evidence that the statistical sig
nificance observed in the previous tests does
not result from averaging out the variance in
the individual contracts within each market,
changes from period 1 mean prices and prices
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET 101
for individual contracts in period 5 were ana-
lyzed (N=257). The mean price change for
individual contracts in the ++-- order (0.087)
is significantly lower (t=-2.984, P=O.O02) than
the mean change in the --++ order (0.678).
Together, these data suggest that individual
biases due to information order survived this
market.
CONCLUSIONS
Before discussing the implication of the results,
it is appropriate to mention the limitations
and strengths of the study. A limitation of the
study is that the data were collected using a
laboratory market methodology, hence, care
should be taken in generalizing the results to
field asset exchanges or other organized mar-
kets. On the other hand, the use of laboratory
markets permitted a relatively strong design,
thus, ensuring a high degree of internal validity.
A particular strength of the design is that the
subjects participated in ten periods of trade
prior to beginning the rounds used for hypoth-
esis testing. During these early rounds, the sub
jects were exposed to all of the information
items. As a result, during the treatment rounds,
the subjects were familiar with the operation of
the computer trading program as well as with
the information items and their possible effects
on liquidating dividends.
Prior behavioral research shows that when
individuals are presented with a series of infor-
mation items, their decisions are systematically
influenced by the order in which the items are
presented (Hogarth & Einhom, 1992). Eco-
nomic theory would suggest that in a market
context, information order is an irrelevant vari-
able to the price decision. This study extends
the prior findings to a market setting where
individuals contract to buy and sell assets in a
market setting.
A key economic theory is that markets use
information efficiently. This study was designed
to directly test the efficient market hypothesis
against the alternative hypothesis that systema-
tic individual biases can survive in a market.
In three replications over two companies,
separate markets traded on identical informa-
tion in different orders and arrived at different
positions. Had the markets used the informa-
tion without bias, one would expect the mar-
kets to arrive at identical positions. This
certainly did not happen; the observed order
effect was highly significant. The implication is
that markets may not remove every type of indi-
vidual decision bias. These results are consistent
with findings from other recent studies that
examine other types of cognitive biases in a
market context (Ganguly et al ., 1994 and
Kachelmeier, 1995).
These findings represent an important con-
tribution to the research on efficient markets
(DeBondt & Thaler, 1985; DeBondt & Thaler,
1987; Zarowin, 1989; Zarowin, 1990) due to
the method in which they were obtained. Prior
studies using exchange-based data may be sub
ject to difiiculties in interpretation based upon
misspecilied asset pricing models or due to
omitted variables. In the present study a pricing
model was not required. Rather, the markets
simply received identical information, but in
different orders. It also was possible to control
for unwanted variables using the laboratory
market method in a manner that cannot be
achieved using exchange-based data.
The findings suggest a possible boundary
condition for market efficiency to occur. The
data were obtained under conditions where the
individuals in the market likely were biased in
the same direction at the same time. In the
++-- order, individuals likely are biased
downwards, ceteri s pari bus, whereas in the
--++ order, individuals likely are biased
upwards. Because markets are composed of
individuals acting together, it is reasonable to
assume that if all (or a sufficiently large num-
ber) of the individuals are biased in the same
direction at the same time, then their collective
decisions will reflect this bias. The opposite
situation suggests a condition under which
market efficiency is most likely to occur:
when the information received by the market
does not systematically bias individual judg-
ment.
102 B. TUTTLE et al .
REFERENCES
Anderson, M. J. & Sunder, S., Professional Traders as Intuitive Bayesians, Organfzatfonaf Behavi or and
Human De&i on Processes (1995) pp. 155-202.
Arrow, K. J., Risk Perception in Psychology and Economics, Economi c hqufry (1952) pp. l-9.
Asare, S. K. & Messier, W. F., A Review of Audit Research Using the Belief-Adjustment Model , i n Ponemon,
L. and Gebhart, D. R. (eds), Audi ti ng: Advances i n Behavi oral Research (New York: Springer-VerIag.
1991)pp. 111-135.
Ashton, A. H. & Ashton, R. H., Sequential Belief Revision in Auditing, The Accounti ng Revfew (1955) pp.
623-641.
Bernard, V. L. & Schipper, K., Recognition and Disclosure in Financial Reporting, Unpublished Working
Paper, University of Chicago (1994).
Bloomfield, R., Quotes, Prices and Estimates in a Laboratory Market, J ouwzal ofFfnance, i n press.
Bloomfield, R. & Libby, R., Market Reactions to PifferentiaBy Available Information in the Laboratory,
J ournal of Accounti ng Research, i n press.
Camerer, C. F., Do Biases in Probability Judgment Matter in Markets? Experimental Evidence, Ameri can
Economfc Revi ew (1957) pp. 981-997.
Coi I er, M., Information, Noise, and Asset Prices: An Experimental Study, Revfew of Accountfng Studi es
(1996) pp. 31-46.
DeBondt, W. & Thaler, R., Does the Stock Market Overreact?, J ournal of Rnance (1955) pp. 793-505.
DeBondt, W. & Thaler, R., Further Evidence on Investor Overreaction and Stock Market Seasonal@,
J ournal of Fi nance (1957) pp. 557-551.
Feldman, R. S. & Bernstein, A. G., Primacy Effects in Self-attribution of Ability, J ournal of Personal @
(1975) pp. 732-742.
Furnham, A., The Robustness of the Recency Effect: Studies Using Iegal Evidence, Z?re J ournal of General
Psychol ogy (1986) pp. 351-357.
Ganguly, A. R., Kagel, J. H. & Moser, D. V., The Effects of Biases in Probability Judgments on Market Prices,
Accounti ng, Organi zati ons and Soci ety (1994) pp. 675-700.
Grether, D. M., Bayes Rules as a Descriptive Model: The Representativeness Heuristic, Q~atier? J OUWUZ~
of Economics (1950) pp. 537-555.
Hogarth, R. & Einhom, H., Order Effects in Belief Updating: The Belief-adjustment Model, CognWve Psy-
cbol ogy (1992) pp. l-55.
Jensen, M. C., Some Anomalous Evidence Regarding Market Efficiency, J ournal of Fi nanci al Economi cs
(1975) pp. 95-101.
Kachelmeier, S., The Influence of Sunk Costs in Accounting: Evidence from the Labomtory, Unpublished
Working paper, University of Texas at Austin (1995).
Lichtenstein, M. & Srull, T. K., Processing Objectives as a Determinant of the Relationship Between Recall
and Judgment, J ournal of E.+ri menkal Socfal Psychol ogy (1957) pp. 93-118.
Luchins, A. S. & Luchins, E. H., Conceptions of Personality and Order Effects in Forming Impressions, The
J ournal of General Psychol ogy (1954) pp. 165-l%.
Plott, C. R. & Wilde, L., Professional Diagnosis vs. Self-Diagnosis: An Experimental Examination of Some
Special Features of Markets with Uncertainty, In Smith, V. L., (ed.) Research i n Experi menkal Econom-
i cs (Greenwich, CN: JAI Press, 1952).
Shanteau, J. C., An Additive Model for Sequential Decision Making, J ournal of Experi mental P~cbol ogy
(1970) pp. 181-191.
Shanteau, J. C., Descriptive versus Nonnative Models of Sequential Inference Judgement. J ournal of
Experfmental Psychol ogy (1972) pp. 63-68.
Shiller, R. J., Do Stock Prices Move Too Much to be Justified by Subsequent Changes in Dividends?
Ameri can Economi c Revi ew (1951) pp. 421-436.
Tetiock, P. E., Accountability and the Perseverance of First Impressions, Soci al Psychol ogy Quarterl y
(1953) pp. 255-292.
Tubbs, R. M., Messier, W. F. & Knechel, W. R., Recency Effects in the Auditor’s Belief-Revision Process,
The Accounti ng Revi ew (1990) pp. 452-460.
Tversky, A. & Kahneman, D., Judgement under Uncertainty: Heuristics and Biases, Science (1974) pp.
1124-1131.
Walker, 1.. Thibaut, J. & Andreoli, V., Order of Presentation at Trial, Yule Law J ournal (1972) pp. 216-
226.
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET 103
Williams, A. W., Computerized Double-Auction Markets: Some Initial Experimental Results, J oumnl OJ
Business (1980) pp. 235-258.
Zarowin, P., Does the Stock Market Overreact to Corporate Earnings Information?, J ournal of Finance
(1989) pp. 1385-1400.
Zarowin, P., Size, Seasonality, And Stock Market Overreaction, J ournal of Financial and Quantitative
Analysis (1990) pp. 113-126.
doc_639488099.pdf
A series of laboratory double auction experiments is conducted to examine whether the order of information
releases affects market prices. Behavioral research on belief revision has shown that individuals are
influenced by the order in which a series of information items is presented. The experiments are designed
to provide a controlled investigation of whether order effects as displayed by individuals also can influence
prices in a market setting where outcomes are not a simple aggregation of individual behavior.
Signiticant evidence is found of a recency effect in the experimental asset markets.
Pergamon
Accounting, 0rganiration.s and Soctety, Vol. 22, No. 1, pp. 89-103, 1997
Copyright c 1996 Hsevier Science Ltd
Printed in Great Britain. All rights reserved
0361.3682/96 $17.00+0 00
PII: SO361-3682(96)00026-S
AN EXAMINATION OF MARKET EFFICIENCY: INFORMATION ORDER
EFFECTS IN A LABORATORY MARKET
BRAD TUTTLE, MARIBETH COLLER
Uni versi ty of South Carol i na
and
F. GREG BURTON
Uni versi ty of Nebraska
Abstract
A series of laboratory double auction experiments is conducted to examine whether the order of infor-
mation releases affects market prices. Behavioral research on belief revision has shown that individuals are
influenced by the order in which a series of information items is presented. The experiments are designed
to provide a controlled investigation of whether order effects as displayed by individuals also can influ-
ence prices in a market setting where outcomes are not a simple aggregation of individual behavior.
Signiticant evidence is found of a recency effect in the experimental asset markets. Copyright (Q 1996
Elsevier Science Ltd
The efficient market hypothesis states that mar-
kets impound new information quickly and
without bias into security prices. In this paper,
we report the results of a series of laboratory
double-auction experiments conducted to
examine whether the order of information
releases can affect market prices. In essence,
the study examines the descriptive validity of
the efficient market hypothesis against the
alternative hypothesis that systematic individual
biases can survive in a market setting. Prior
research has examined this implication of the
efficient market hypothesis using archival
exchange data. This stream of research, how-
ever, has provided mixed results. For example,
DeBondt & Thaler (1985) claim that stock pri-
ces overreact to new information and support
this hypothesis with exchange data. Although
DeBondt & Thaler (1987) examine the issue
further and find no evidence that their earlier
result is a manifestation of a size effect, Zarowin
(1989); Zarowin (1990) find that size may be
responsible for the DeBondt & Thaler (1985)
result.
Several explanations may be advanced for the
mixed results in exchange-based studies inves-
tigating market efficiency with respect to parti-
cular information items. The first relates to the
difficulty of gaining adequate control over
extraneous variables when using exchange-
based data. A second explanation is that the
pricing models used to determine the expected
efficient market price may themselves be mis-
specified. A third explanation relates to the
logical difficulty in assuming an information
item is valuation relevant and then testing for
efficient use of that item. If the item is not
relevant to valuation, then even the most effi-
cient market will not impound it into security
prices.
Authors are listed in reverse alphabetical order. We wish to thank John Dickhaut, Daniel Friedman, Adrian Harrell, Glenn
Harrison, Julia H&s, John Reisch, Charles S&nit&in, Earl Spiller, and participants at the 1995 Economic Science Association
meeting for comments on prior versions of this paper.
89
90 B. TUlTLE et al .
These difficulties inherent in exchange-based
research suggest that a potentially fruitful
direction is the use of laboratory methods.
Laboratory markets provide for direct control in
manipulating certain variables of interest while
eliminating other variables (or holding them at
a constant level).’ In the laboratory, it is also
possible to conduct a market where estimation
of an efficient price is not necessary for the
analysis and where the information is valuation
relevant by design. We accomplish this by
examining two markets that receive exactly the
same (valuation relevant) information, but in a
context where inefficiencies in information
processing could cause the different markets to
arrive at different prices.2 The literature on
individual decision making suggests that this
may be possible, for example, by manipulating
the order in which the information is received.
Prior behavioral research shows that when
individuals are presented with a series of infor-
mation items, their decisions are systematically
influenced by the order in which the items are
presented (Hogarth & Einhorn, 1992). These
effects have proven robust and have been
observed in various settings including social
judgments and attributions for behavior (e.g.
Feldman & Bernstein, 1978; Lichtenstein &
Srull, 1987; Luchins & Luchins, 1984) mock
legal trials (Put-&am, 1986; Tetlock, 1983;
Walker et al , 1972) judgments of probability
(Shanteau, 1970, 1972) and auditing (Asare &
Messier, 1991; Ashton & Ashton, 1988; Tubbs
et al , 1990). Despite the fact that investors in
organized markets regularly receive streams of
information that they use to revise their valua-
tion estimates, no prior research has examined
order effects in a market context3 This raises
the possibility that the order effects observed in
individual decision making may also occur in
asset pricing within a market. The objective of
the current study is to examine this possibility
in a laboratory market.
We conduct a series of experiments in which
participants trade shares of stock in fictitious
firms over five trading periods where relevant
information about the firm is publicly
announced between each of the trading peri-
ods. In one version of the experiment, the
information consists of two “good” information
items followed by two “bad” information items.
In the other version, the exact same informa-
tion is released, but in the opposite order, i.e.
bad followed by good. The subjects use this
information to update their beliefs regarding
the final dividend and, hence, to revise their
estimates of the security price. If these markets
are efficient, then the final valuation of security
prices should not differ between two markets
that receive different orders of the same infor-
mation. If, on the other hand, the market
mechanism does not remove individual biases
of this type, the final price should differ pre-
dictably depending on the order of information
arrival.
Research of this type is important for several
reasons. First, it is important to determine the
conditions under which economic theories,
such as market efficiency, are more or less
likely to hold. This is a necessary step towards
isolating the precise market mechanisms that
produce market efficiency. Second, this
research contributes to the literature on indivi-
dual belief revision by extending its findings to
a market setting. The competitive process
inherent in the market often is thought to
‘Similar arguments have been advanced in other laboratory studies investigating issues that have traditionally been exam-
ined with exchange based data (see, for example, Bloomfield, in press, who examines the effects of information asymmetry
on the bid-ask spread, and Coller, 1996, who examines pricing effects of noisy information releases).
aBloomfield and Libby, in press, use an experimental market to address another issue of market efficiency: whether prices
fully aggregate information held by differently informed investors. They compare prices between markets where the infor-
mation available to the market as a whole is the same, but where the availability of each information item is manipulated.
They also construct each of their three information items to be valuation relevant. They find that the distribution of infor-
mation among investors does affect prices.
3A developing stream of experimental research is beginning to examine certain other cognitive biases in market contexts.
For example, Ganguly et al . (1994) examine the persistence of the base-rate fallacy in an oral-double auction, and
Kachelmeier (1995) examines the effects of sunk costs on trading behavior in a computerized double auction.
influence how individuals formulate and revise the similarity of the present evidence to it. This
their beliefs, mitigating or even eliminating representativeness heuristic has been used to
cognitive biases. Whether or not the market explain departures from Bayes’ rule in various
process “trains” successful traders to overcome types of individual decision making.* Hogarth &
their individual biases is still an open question. Einhom (1992) suggest that another decision
Third, companies that are considering the tim- irrelevant variable, the order in which informa-
ing of their information and investors who are tion is received, systematically influences indi-
the recipients of the information wish to know vidual judgment. Although substantial evidence
what type of effect, if any, information order indicates that individual decision making often
has on the valuation of their securities. is inefficient, markets generally are assumed to
be (semi-strong) efficient even if individual par-
ticipants within the market are biased.
BACKGROUND Camerer (1987) summarizes several argu-
ments in defense of market theories despite
Jensen (1978, p. 96) defines an efficient market evidence indicating the irrationality of indivi-
as follows: “A market is efficient with respect to duals: (1) market participants have enough
information set 8, if it is impossible to make financial incentive and experience to avoid
economic profits by trading on the basis of mistakes; (2) random mistakes of individuals
information set &” The widely accepted semi- cancel out; (3) only a small number of rational
strong form of the efficient market hypothesis individuals are needed to drive the market to
further holds that markets are efficient with rationality; (4) less rational individuals learn or
respect to the information set that includes all follow the actions of rational individuals; and
publicly available information. When informa- (5) agents who are less rational are driven from
tion is widely known to the participants in the the market. Most of these arguments, however,
market, competition will drive prices so that may be countered. For example, some investi-
participants are unable to consistently earn gations have observed biased individual deci-
economic profits by trading on that informa- sion making with highly experienced subjects
tion. Economic profits are assumed to be net of and financial incentives (Grether, 1980). Addi-
all costs. If storage, transaction, and information tionally, most psychological biases are systema-
costs are zero in a market that is efficient, the tic, not random, and therefore may not cancel
expected price of an asset is equal to the sum of each other out in the aggregate. Furthermore,
its discounted future cash flows. Thus, widely institutional constraints on trading may prevent
available information that predicts future cash sufficient capital from being exerted by the
flows should be completely assimilated into the limited number of rational agents to make a
price of an asset. Conversely, such information difference. Finally, new traders, who pre-
received in a market that is not efficient may sumably are less likely to be rational, are con-
not result in a market price that properly stantly entering the market.
reflects the asset’s anticipated future cash Some evidence suggests that individual biases
flows. can in fact survive in a market (Anderson &
Although markets generally are assumed to Sunder, 1995; Camerer, 1987). Arrow (1982,
be efficient information processors, individuals p. 5) asserts that the representativeness bias ((...
are not. Research primarily from psychology typifies very precisely the excessive reaction to
suggests that individuals can be biased by irre- current information which seems to character-
levant variables. For example, Tversky & Kah- ize all the securities and futures markets.”
neman (1974) suggest that people base their DeBondt & Thaler (1985, p. 794), in reviewing
judgments of the likelihood of a future event by the work of Shiller (1981) and others, conclude
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET 91
‘Plott & Wide (1982) find evidence of the representativeness heuristic in a market setting.
92 B. TUlTLE et al.
that “... investors seem to attach dispropor-
tionate importance to short-run economic
developments.” DeBondt & Thaler (1985) also
cite evidence that the expectations of profes-
sional security analysts and economic fore-
casters have an overreaction bias. DeBondt &
Thaler (1985) explain that if stock prices sys-
tematically overreact to new information, then
their reversal should be predictable from past
return data alone: extreme price movements
should be followed by price movements in the
other direction, and the more extreme the
initial movement, the greater the subsequent
adjustment. This assertion is supported empiri-
cally, and DeBondt & Thaler (1985) conclude
that their data are consistent with the predic-
tions of the overreaction hypothesis. DeBondt
& Thaler (1987) provide further evidence that
their earlier finding is not simply a manifesta-
tion of a size effect and/ or differences due to
risk. Later research (Zarowin, 1989; Zarowin,
1990), however, finds that size may indeed be
responsible for the DeBondt & Thaler (1985)
finding.
The conflicting findings of studies using
exchange-based data may be attributable in part
to three difficulties inherent in such research:
(1) the lack of adequate control over potentially
confounding exogenous factors; (2) mis-
specifications in the assumed asset pricing
models; and (3) the inability to discriminate
between the market’s use of an item and the
importance of the item itself. Bernard & Schipper
(1994) discuss these general problems and note
that, “In principle, many of the difficulties faced
in interpreting research based on archival data
can be overcome in an experimental setting.”
Control over exogenous variables is achieved
in our study by employing a laboratory market.
This research method provides for the explicit
manipulation of the nature and timing of infor-
mation releases where the information is by
design valuation relevant. The experiments are
designed to allow a test of the market efficiency
assumption without requiring the exact func-
tional form of the link between information and
cash flows to be known. Specifically, we con-
duct an asset market that receives four pieces
of valuation relevant information. The tirst two
pieces of information suggest an increase in
future cash flows (good news). These are fol-
lowed by two items of information that suggest
a decrease in future cash flows (bad news). In
this situation, one would expect the market
price for the asset to first rise and then to fall,
the exact amount of change in price being
determined by the weight that the market pla-
ces on each piece of information. We then
conduct a second market for the same asset as
in the first market. This second market receives
the same information as does the first, but in
the opposite order. In this case, one would
expect the market price to first fall and then to
rise. The arrival of both markets at similar pri-
ces after all information is received and pro-
cessed would be consistent with efficient
information use. If, however, the final price
differs between markets, there is evidence of a
bias in the market’s processing of information.
Hence, by examining a situation in which order
might influence judgments, it should be possi-
ble to detect related market inefficiencies with-
out the need to use models of expected future
stock prices.
The decision-making literature has shown
that the order of information influences the
judgments of individuals in a predictable
manner. Hogarth & Einhom (1992) present a
model of sequential belief revision in which
predictions of order effects are based on a tax-
onomy of tasks along three dimensions: task
complexity, series length, and response mode.
Simple tasks typically are those in which the
information consists of a simple trait adjective
or single number and where the individual is
familiar with its meaning. Complex tasks
involve more or unfamiliar stimuli. Short series
involve between 2 and 12 pieces of informa-
tion. Long series have generally involved over
20 items. One type of response mode consists
of making a judgment after receiving each
information item and is termed “step-by-step”
processing. Another response mode, known as
“end-of-sequence” processing, consists of
processing alI pieces of information prior to
making a single, final judgment. Complex
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET 93
information received in a relatively short series
that is used to adjust beliefs in a step-by-step
mode characterizes many real-world capital
market conditions and is the task used in the
present study. Under these conditions, the
Hogarth & Einhorn (1992) model predicts a
particular order effect, the recency effect. That
is, recently received information receives
more weight than previously received infor-
mation.
The Hogarth and Einhorn model has impor-
tant implications for decisions involving
streams of mixed information where some
items are positive and some are negative. When
negative information is followed by positive
information, i.e. information order -+5, the
positive information will have the greater
impact because it is most recent. Conversely,
negative information will have a greater impact
when it follows positive information, i.e. infor-
mation order +--. Recency is implied because
the final position for the +- order is lower than
the final position for the -+ order. The combi-
nation of the two effects produces what has
become known as the “fishtail” result (Fig. 1). If
this model of individual sequential belief revi-
sion also describes market behavior, then a
market that receives good then bad news will
arrive at a lower asset price than a market that
receives bad then good news, when the spe-
cific items of information are identical.
The experiments described below seek to
determine whether this model of individual
belief revision also can describe price behavior
in an asset market. The extent to which propo
sitions based on individual behavior generalize
to aggregate market setting remains an unre-
solved issue (Camerer, 1987). Anderson 8t
Sunder (1995, p. 186) point out that market
behavior is not a simple aggregation of indivi-
dual behavior and state further that, “... it does
not follow that biased individual behavior
implies biased aggregate market outcomes.”
Although the step-by-step processing in asset
markets is parallel to the Hogarth and Einhorn
model, it is an open question whether any
resulting individual biases will affect the market
outcome.
The above arguments suggest the following
alternative hypothesis to the efficient market
hypothesis:”
Ha: Market price changes will be influenced by the
order of information arrival such that positive infor-
mation followed by negative information results in- a
lower price than the same information but received in
the order of negative information followed by positive
information.
THE EXPERIMENT
Procedure
The laboratory asset markets were conducted
using software developed at the University of
Arizona’s Economic Sciences Laboratory for
double-auction experiments.’ Each run of the
experiment consisted of four independent
rounds of trade where six traders were able to
make both bids and offers for shares of stock.
During each round, trade was conducted in a
separate company: round 1 was “Company A,”
round 2 was “Company B,” round 3 was
“Company C,” and round 4 was “Company D.”
At the beginning of trade in each company,
each subject was endowed with $30 (experi-
mental dollars’) and two shares of stock.
sWe follow the convention of labeling information as “+” for good news and “ -” for bad news.
60ne diiculty in analyzing order-effect data when order treatments are between subjects is that initial differences can
mask effects (Hogarth & Einhorn, 1992). For this reason, research using the sequential belief revision paradigm focuses on
the difference between the initial belief and the ending belief. Typically, this is accomplished by subtracting the initial belief
from subsequent responses, and comparisons between groups consist of changes from the initial belief. Accordingly,
references to market price in the present paper refer to price changes (i.e. the difference between first period price and
subsequent prices). The analytical results from Hogarth & Einhom (1992) suggest that this is appropriate because the initial
position (i.e. high or low) should have no effect on the size of the recency effect (Ashton & Ashtoa, 1988 p. 629).
‘Complete instructions for the asset markets are available from the authors upon request.
@Ihe exchange rate used in calculating cash payments was set at one experimental dollar=O.l IJS. dollar. Of the $30
endowment, $20 was an interest free loan that was repaid at the end of the trading round.
94 B. TUTTLE el al.
-3’
Period 1 Period 2
I
Period 3
I
Period 4
I
Period 5
Fig 1. Hypothesized order effect.
Information regarding the fmal liquidating divi-
dend to be received at the end of trade in each
company also was presented. This information
was identical for each company and was pre-
sented as a table containing possible dividends
and the associated probability of each. The
dividend distribution as shown to the subjects
is presented in Table 1. There are nine possible
dividends, ranging from $4 to $20, with the
expected value being $12.
TABLE 1. Dividend distribution used for each company.
Company A-Period 1: dividend expectations
Dividend amount Probability
$4
$6
$8
$10
$12
$14
$16
$18
$20
5%
5%
10%
15%
30%
15%
10%
5%
5%
Within each of the four trading rounds, there
were five three-minute trading periods. After
the first trading period in each round, trade
stopped and information regarding the com-
pany was publicly announced (both verbally
and shown via an overhead projector). Follow-
ing the release of information, the traders were
allowed time to consider its impact. After
everyone indicated they were ready to con-
tinue, they hit the ENTER key to begin another
three-minute trading period. At the conclusion
of this period, the market stopped, and the next
piece of information was announced. This pro-
cess continued until five periods of trade were
completed (with four information items
announced between periods). At the conclu-
sion of the fifth period of trade, the liquidating
dividend was publicly announced9, profits were
calculated, the market was re-initialized, and
trade began for the next company. The specific
information item released constitutes the order
manipulation as described below. Eight different
information items were used in the experiment.
These are listed in Table 2 and will be referred
to by the numbers presented in the table.
?lre actual dividend realized was equal to the expected value and was determined by “shifting” the expected value toward
the higher dividends after good news and toward the lower dividends after bad news by equal amounts ($3). For companies
with two items of good news (and two items of bad news), this results in the conditional expected value (after aIf infor-
mation is received) being identical to the unconditional expected value. Hence, the dividends paid to subjects were con-
sistent with the instructions provided to them.
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET 95
TABLE 2. Information items
Good news items
+1.
+2.
+.j.
+4.
Bad new items:
-5.
-6.
-8
An easing of fuel prices is expected to result in much lower shipping costs than expected during the
upcoming busy season. Because freight is a large portion of the company’s operating costs, a signiticant
savings will be realized.
Several of the company’s main competitors have manufacturing facilities located in the Southeast.
Recently, tropical storms severely damaged this area causing their facilities to shut down. As the com-
pany’s manufacturing is located elsewhere, it is expected to capture the competitors’ lost sales.
The company’s suppliers have unexpectedly reduced the cost of a primary component in a major
product line. This is expected to lower product costs substantially.
The company just introduced an important new product a full three months ahead of schedule. The
company’s competitors are not expected to have their own version of the product ready to release
until sometime next year. Consequently, the company expects to obtain much of the market.
A major customer has filed bankruptcy proceedings. This single customer represents about 15% of the
company’s total accounts receivable.
A competitor has introduced a revolutionary new product. The company is not expected to have its
own version of the product ready for some time. Consequently, the company is expected to lose sub
stantiaf market share.
The company has just been forced to recall nearly all of its latest model products to repair a mechanical
problem. Not only will this be very expensive, but the competition is taking full advantage of the
situation by aggressively pursuing the company’s customers.
The only source of customized parts for an important product will be unable to meet demand for the
next two quarters. This condition will reduce the company s sales by a considerable amount as no
other viable sources exist in the short term
The purpose of the experiments was to
examine price changes in a market that is
allowed to interpret information as it sees fit.
Therefore, no specific adjustment mechanism
was provided, nor was any direction offered on
how to weight any particular piece of informa-
tion. Subjects were told only that good (bad)
news would indicate a greater likelihood of a
higher (lower) dividend being realized. Many
laboratory markets specify a particular adjust-
ment procedure to be used in price formation.
The current study differs from prior ones in that
the absence of a specific adjustment procedure
is integral to our research question. Further-
more, this situation is more like that existing in
organized exchanges. In those settings, infor-
mation is often released that is not quantita-
tively specific, and the impact on market prices
must be “subjectively” determined by the market.
The first trading round, Company A, was
conducted “for practice”. Subjects were told
that they would not receive cash for their prof-
its from this round, although profits would be
calculated just as in the later rounds. This prac-
tice round was conducted to provide an
opportunity for the subjects to familiarize
themselves with the market and to practice
entering and accepting bids and offers without
concern for how their cash profits might be
affected. Subjects were paid all of the profits
earned during the remaining three rounds.
Average profits were $10 U.S. for the twohour
experiment plus a $3 U.S. participation fee.
Order of information manipulation
The experimental design was a repeated mea-
sures design (four rounds of trade) with order
of the information items manipulated between
subjects (two versions of the experiment). The
order manipulation was not imposed until the
third round as described below. The distin-
guishing feature of each trading round is the
specific information presented for each com-
pany. The sequence of events and the informa-
tion items released in each of the two versions
of the experiment are presented in Table 3. As
B. TUTTLB el al.
TABLE 3. Experimental design
Company/Round Period Information items
Version 1 Version 2
A (Practice round) 1 Dividend distribution Dividend distribution
2 +3 +3
3 -5 -5
4 +1 +1
5 +4 +4
End Dividend declared Dividend declared
Dividend distribution Dividend distribution
-7 -7
+2 +2
-6 -6
-8 -8
Dividend declared Dividend declared
D 1
2
3
4
5
End
Dividend distribution Dividend distribution
+1 -5
+2 -6
-5 +1
-6 +2
Dividend declared Dividend declared
Dividend distribution Dividend distribution
-7 +3
-8 +4
+3 -7
+4 -8
Dividend declared Dividend declared
Note: Information item numbers are preceeded by a sign designating whether it conveys positive or negative information to
the market. Information item numbers are presented here strictly as an expositional aid; subjects were not presented with
item numbers.
shown in Table 3, both versions received the
same information, in the same order, for Com-
panies A and B. For Company A, information
was provided in the order of +-++ (informa-
tion items +3,-5,+1,+4). For Company B,
information was provided in the order of
-+-- (information items -7,+2,-6-8). The
data generated in the first two rounds were not
used in hypothesis testing, but were used to
check whether the information manipulation
was successful. These early rounds also were
conducted to provide the subjects with an
opportunity to become familiar with the market
and to gain experience trading. Prior research
(Williams, 1980) ,demonstrates that traders
may require a training period to become familiar
with the task in computerized asset markets.
Note that all eight information items were
presented during the first two rounds so that
the subjects had experience with each piece of
information prior to the rounds used for data
analysis.
The two versions of the experiment differed
only in the order of the information presented
for Companies C and D. These final two rounds
of trade were designed to provide the strictly
controlled test of information processing effi-
ciency in the markets. For Company C of Ver-
sion 1, the information items were presented in
the order of ++-- (items +1,+2,-5,-6). In
Version 2, these items were presented in the
order of --++ (items -5,-6,+1,+2). By com-
paring the market price changes for Company
C in Version 1 versus those in Version 2, it is
possible to determine under strictly controlled
conditions whether the order of information
arrival per se affects the market price. Similarly, information arrival in the context of thinking
for Company D of Version 1, the information about appropriate stock prices.
items were presented in the order of --++ The pretest participants also were asked to
(items -7,~8,+3,+4); in Version 2, the items complete a manipulation check after providing
were presented in the order of ++-- (items their market prices. Here, they rated each of the
+3,+4,-7,-8). Each experiment version was eight information items as being “good news”,
run three times (a total of six experiments was “bad news”, or “can’t tell”. Ninety-six percent
conducted).” of the responses correctly identified the infor-
Pretests were used to select information mation as being good news or bad news. Of the
items that could be easily interpreted as good remaining 3.5%, all were in the “can’t tell”
or bad and that would be perceived to have category; no responses were in the direction
similar importance in judging company perfor- opposite the intended manipulation. The
mance. Prior to conducting the laboratory asset results from this check provided assurance that
markets, pretests also were conducted to the asset market subjects would perceive the
determine whether individual decision makers information items as intended.
exhibit order effects in this particular task.
Forty-four students completed an individual Subjects
survey asking for their estimate of market pri- The laboratory markets were conducted at the
ces for the securities of Companies A through D University of South Carolina and at the Uni-
under the same design as described above. versity of Nebraska. The experiments were
Pretest subjects first were presented with the designed for six traders, although one of the six
dividend distribution (as used in the asset mar- runs was conducted with only five (Run 2 of
kets) for the security and asked to state a mar- Version 1). Thus, a total of 35 subjects partici-
ket price. They next were provided with the pated. Participants were recruited from under-
same information in the same order as used in graduate business classes. Approximately haIf
the asset markets (one version of the pretest of the subjects were female. Just under half
corresponded to market Version 1, and the were accounting majors, with the remaining
other corresponded to market Version 2). After business disciplines evenly represented. The
receiving each new information item, pretest average participant had taken three or more
subjects answered the open question, “In light courses related to accounting, finance, compu-
of this new information, what do you believe ters and business policy.
should be the market price for a share of stock
in Company X?” The mean change between the
initial price provided in period 1 and the ending RESULTS
price provided in period 5 is significantly dif-
ferent between Version 1 and Version 2 for Preliminary analysis
both Company C (ti6.230, RO.001) and Com- Initial tests were conducted to assess the
pany D (t=2.819, eO.037). The mean percen- homogeneity of subjects across treatment con-
tage changes are in the predicted direction ditions (Version 1 vs. Version 2). Subjects do
such that prices indicated under the informa- not differ significantly (Chi-square tests:
tion order of ++-- changed in a negative -0.10) between treatment conditions as to
direction overall, and prices indicated under their major field of study, class standing (gradu-
the information order of --++ changed in a ate vs. undergraduate), or business courses
positive direction overall. These pretest results taken with the exception of Introduction to
suggest that individual decision makers are Finance. Neither do they differ significantly
systematically inlhtenced by the order of (t=1.474, -0.30) in age. Further support for
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET 97
“*The versions were administered in random order within the constraint that three runs of each version were conducted.
98 B. TU’lTLE el al.
TABLE 4. .Information items and manipulation check results, subject frequencies by response
Note: the items appeared in randomized order in the manipulation
Information represents:
check. No “+” or “-” designation appeared.
Good news Can’t tell Bad news
+ 1. An easing of fuel prices is expected to result in much lower 34 1 0
shipping costs than expected during the upcoming busy season.
Because freight is a large portion of the company’s operating
costs, a significant savings will be realized.
+2. Several of the company’s main competitors have manufactur- 31 1 3
Ing facilities located in the Southeast. Recently, tropical storms
severely damaged this area causing their facilities to shut down. As
the company’s manufacturing is located elsewhere, it is expected
to capture the competitors’ lost sales.
+3. The company’s suppliers have unexpectedly reduced the cost
of a primary component in a major product line. This is expected
to lower product costs substantially.
+4. The company just introduced an important new product a full
three months ahead of schedule. The company’s competitors are
not expected to have their own version of the product ready to
release until sometime next year. Consequently, the company
expects to obtain much of the market.
-5. A major customer has filed bankruptcy proceedings. This sir-
gle customer represents about 15% of the company’s total
accounts receivable.
29
33
-6. A competitor has introduced a revolutionary new product.
The company is not expected to have its own version of the pre
duct ready for some time. Consequently, the company is expected
to lose substantial market share.
-7. The company has just been forced to recall nearly all of its
latest model products to repair a mechanical problem. Not only
wilI this be very expensive, but the competition is taking full
advantage of the situation by aggressively pursuing the company’s
customers.
-8. The only source of customized parts for an important product,
will be unable to meet demand for the next two quarters. This
condition wiII reduce the company’s sales by a considerable
amount as no other viable sources exist in the short term.
5
2
0
2
1
0
27
34
35
33
the similarity of subjects assigned to Versions 1
and 2 is obtained by showing that their
responses on Companies A and B are insig-
nificantly different. Mean contract price
changes do not differ significantly between
groups for Company A @=0.893, P=O.45) or
Company B (tiO.419, P=O.70). These data sug-
gest that the subjects are homogenous across
treatments.
The good versus bad information manipula-
tion was tested using data collected at the con-
clusion of the experiment with the same
questionnaire as in the pretest. Table 4 shows
“bad news”, or “can’t tell” for each piece of
information. Of the total 280 responses (35
subjectsx8 information items), 256 are in the
correct cell. In only 4 cases did a subject
respond in the opposite direction. “Can’t tell”
responses are distributed evenly among the
market sessions. “Good news” responses were
coded as a 1, “can’t tell” responses were coded
as 0, and “bad news” responses were coded as
- 1. Using this data, a t-test shows that the mean
response on the four positive information items
is significantly higher (k40.505, RO.001) than
the mean response on the four negative infor-
the frequency of responses under “good news”, mation items.
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET
3-
-l- “‘.6x’
-2 - I I I I
Period 1 Period 2 Period 3 Period 4 Period 5
99
--o- B: -+--
Fig. 2. Mean price changes for Companies A and B.
As a further manipulation check, data from
the first two rounds (Companies A and B) were
initially examined to determine if the market
revised their beliefs in the appropriate direction
following the information releases. Recall that
all eight information items were presented dur-
ing these two rounds. Figure 2 presents the
mean price change (across all six experimental
runs) for each period of each of these rounds.
Mean contract prices follow the information
items quite well. For example, for Company A,
the market first received one item of positive
information followed by one item of negative
information. This is depicted in the figure as a
rise in the solid line followed by a decline. The
solid line also reflects the two subsequent posi-
tive pieces of information received prior to
trading in periods 4 and 5. Figure 2 provides
additional assurance that the subjects under-
stood the information and knew how to use it
in the market. Together, the data suggest that
the subjects understood the information items
and that the manipulation of information
valence was achieved.
Mai n resul ts
The market efficiency hypothesis suggests that
changes in market price between the end of
period 1 and the end of period 5 will not differ
between experiment versions that differ only in
the order of information arrival. If, however,
the market does not eliminate individual biases,
then an order effect should occur. This alter-
native hypothesis suggests that the price
change in the rounds with ++- - information
order will be substantially lower (or negative)
than the price change in the rounds with - - ++
information order. Figure 3 depicts the price
changes in the third round (Company C) and
Fig. 4 shows the changes in the fourth round
(Company D). Immediately obvious in both
rounds are the “fishtails” associated with a
recency effect.
The differences in price changes due to order
(Fig. 2 and Fig. 3) were tested for statistical sig-
nificance using a 2 x2 ANOVA. The data were
first tested for an overall information order
effect using the mean price changes, from per-
iod 1 to period 5, in Rounds C and D as the
dependent variable (A’= 12; 2 rounds from each
of 6 runs). Information order (++ - - vs. - - ++>
and round (C or D) served as the independent
variables. The overall model is significant
(F=10.35, P=O.OOS). Information order is sig-
nificant (F=16.64, P=O.O03) suggesting that the
order of Information influenced the movement
of market prices. The mean price changes
exhibit a recency pattern: in the ++-- order,
mean price change is -1.125; and in the --++
order, mean price change is +2.288. Neither
B. ‘I’VITLE et al.
‘r
P
3-
,,’
,’
2-
:’
.’
,’
..A.’
_
.._
. ..&.’
13
- 1
I
P-ziod 1 Period 2 Period 3 Period 4 Period 5
Isl:rt-_n2;--.tl
Fig. 3. Mean price changes for Company C.
_3L___‘-__‘-
J
Period 1 Period 2 Psrlod 3 Period 4 Period 6
Fig. 4. Mean price changes for Company D.
round nor the interaction term are significant,
signifying that the effect is unlikely to be
dependent on the particular set of information
items.
Additional tests for differences in changes in
the mean market prices (between period 1 and
period 5) between information orders within
individual rounds were conducted (N=6 for
each test). In Round C, the mean price change
in the ++--- order (-0.319) is significantly
lower (I=-3.351, one-tailed P=O.O14) than the
mean price change in the --++ order
(+3.16&I). In Round D, the difference in mean
price change in the ++-- order (-1.931) is
significantly lower (r=-2.323, one-tailed
EO.040) than the mean price change in the
- -++ order (+1.407). In both rounds, the
data reveal significant effects for order of
information.
To provide evidence that the statistical sig
nificance observed in the previous tests does
not result from averaging out the variance in
the individual contracts within each market,
changes from period 1 mean prices and prices
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET 101
for individual contracts in period 5 were ana-
lyzed (N=257). The mean price change for
individual contracts in the ++-- order (0.087)
is significantly lower (t=-2.984, P=O.O02) than
the mean change in the --++ order (0.678).
Together, these data suggest that individual
biases due to information order survived this
market.
CONCLUSIONS
Before discussing the implication of the results,
it is appropriate to mention the limitations
and strengths of the study. A limitation of the
study is that the data were collected using a
laboratory market methodology, hence, care
should be taken in generalizing the results to
field asset exchanges or other organized mar-
kets. On the other hand, the use of laboratory
markets permitted a relatively strong design,
thus, ensuring a high degree of internal validity.
A particular strength of the design is that the
subjects participated in ten periods of trade
prior to beginning the rounds used for hypoth-
esis testing. During these early rounds, the sub
jects were exposed to all of the information
items. As a result, during the treatment rounds,
the subjects were familiar with the operation of
the computer trading program as well as with
the information items and their possible effects
on liquidating dividends.
Prior behavioral research shows that when
individuals are presented with a series of infor-
mation items, their decisions are systematically
influenced by the order in which the items are
presented (Hogarth & Einhom, 1992). Eco-
nomic theory would suggest that in a market
context, information order is an irrelevant vari-
able to the price decision. This study extends
the prior findings to a market setting where
individuals contract to buy and sell assets in a
market setting.
A key economic theory is that markets use
information efficiently. This study was designed
to directly test the efficient market hypothesis
against the alternative hypothesis that systema-
tic individual biases can survive in a market.
In three replications over two companies,
separate markets traded on identical informa-
tion in different orders and arrived at different
positions. Had the markets used the informa-
tion without bias, one would expect the mar-
kets to arrive at identical positions. This
certainly did not happen; the observed order
effect was highly significant. The implication is
that markets may not remove every type of indi-
vidual decision bias. These results are consistent
with findings from other recent studies that
examine other types of cognitive biases in a
market context (Ganguly et al ., 1994 and
Kachelmeier, 1995).
These findings represent an important con-
tribution to the research on efficient markets
(DeBondt & Thaler, 1985; DeBondt & Thaler,
1987; Zarowin, 1989; Zarowin, 1990) due to
the method in which they were obtained. Prior
studies using exchange-based data may be sub
ject to difiiculties in interpretation based upon
misspecilied asset pricing models or due to
omitted variables. In the present study a pricing
model was not required. Rather, the markets
simply received identical information, but in
different orders. It also was possible to control
for unwanted variables using the laboratory
market method in a manner that cannot be
achieved using exchange-based data.
The findings suggest a possible boundary
condition for market efficiency to occur. The
data were obtained under conditions where the
individuals in the market likely were biased in
the same direction at the same time. In the
++-- order, individuals likely are biased
downwards, ceteri s pari bus, whereas in the
--++ order, individuals likely are biased
upwards. Because markets are composed of
individuals acting together, it is reasonable to
assume that if all (or a sufficiently large num-
ber) of the individuals are biased in the same
direction at the same time, then their collective
decisions will reflect this bias. The opposite
situation suggests a condition under which
market efficiency is most likely to occur:
when the information received by the market
does not systematically bias individual judg-
ment.
102 B. TUTTLE et al .
REFERENCES
Anderson, M. J. & Sunder, S., Professional Traders as Intuitive Bayesians, Organfzatfonaf Behavi or and
Human De&i on Processes (1995) pp. 155-202.
Arrow, K. J., Risk Perception in Psychology and Economics, Economi c hqufry (1952) pp. l-9.
Asare, S. K. & Messier, W. F., A Review of Audit Research Using the Belief-Adjustment Model , i n Ponemon,
L. and Gebhart, D. R. (eds), Audi ti ng: Advances i n Behavi oral Research (New York: Springer-VerIag.
1991)pp. 111-135.
Ashton, A. H. & Ashton, R. H., Sequential Belief Revision in Auditing, The Accounti ng Revfew (1955) pp.
623-641.
Bernard, V. L. & Schipper, K., Recognition and Disclosure in Financial Reporting, Unpublished Working
Paper, University of Chicago (1994).
Bloomfield, R., Quotes, Prices and Estimates in a Laboratory Market, J ouwzal ofFfnance, i n press.
Bloomfield, R. & Libby, R., Market Reactions to PifferentiaBy Available Information in the Laboratory,
J ournal of Accounti ng Research, i n press.
Camerer, C. F., Do Biases in Probability Judgment Matter in Markets? Experimental Evidence, Ameri can
Economfc Revi ew (1957) pp. 981-997.
Coi I er, M., Information, Noise, and Asset Prices: An Experimental Study, Revfew of Accountfng Studi es
(1996) pp. 31-46.
DeBondt, W. & Thaler, R., Does the Stock Market Overreact?, J ournal of Rnance (1955) pp. 793-505.
DeBondt, W. & Thaler, R., Further Evidence on Investor Overreaction and Stock Market Seasonal@,
J ournal of Fi nance (1957) pp. 557-551.
Feldman, R. S. & Bernstein, A. G., Primacy Effects in Self-attribution of Ability, J ournal of Personal @
(1975) pp. 732-742.
Furnham, A., The Robustness of the Recency Effect: Studies Using Iegal Evidence, Z?re J ournal of General
Psychol ogy (1986) pp. 351-357.
Ganguly, A. R., Kagel, J. H. & Moser, D. V., The Effects of Biases in Probability Judgments on Market Prices,
Accounti ng, Organi zati ons and Soci ety (1994) pp. 675-700.
Grether, D. M., Bayes Rules as a Descriptive Model: The Representativeness Heuristic, Q~atier? J OUWUZ~
of Economics (1950) pp. 537-555.
Hogarth, R. & Einhom, H., Order Effects in Belief Updating: The Belief-adjustment Model, CognWve Psy-
cbol ogy (1992) pp. l-55.
Jensen, M. C., Some Anomalous Evidence Regarding Market Efficiency, J ournal of Fi nanci al Economi cs
(1975) pp. 95-101.
Kachelmeier, S., The Influence of Sunk Costs in Accounting: Evidence from the Labomtory, Unpublished
Working paper, University of Texas at Austin (1995).
Lichtenstein, M. & Srull, T. K., Processing Objectives as a Determinant of the Relationship Between Recall
and Judgment, J ournal of E.+ri menkal Socfal Psychol ogy (1957) pp. 93-118.
Luchins, A. S. & Luchins, E. H., Conceptions of Personality and Order Effects in Forming Impressions, The
J ournal of General Psychol ogy (1954) pp. 165-l%.
Plott, C. R. & Wilde, L., Professional Diagnosis vs. Self-Diagnosis: An Experimental Examination of Some
Special Features of Markets with Uncertainty, In Smith, V. L., (ed.) Research i n Experi menkal Econom-
i cs (Greenwich, CN: JAI Press, 1952).
Shanteau, J. C., An Additive Model for Sequential Decision Making, J ournal of Experi mental P~cbol ogy
(1970) pp. 181-191.
Shanteau, J. C., Descriptive versus Nonnative Models of Sequential Inference Judgement. J ournal of
Experfmental Psychol ogy (1972) pp. 63-68.
Shiller, R. J., Do Stock Prices Move Too Much to be Justified by Subsequent Changes in Dividends?
Ameri can Economi c Revi ew (1951) pp. 421-436.
Tetiock, P. E., Accountability and the Perseverance of First Impressions, Soci al Psychol ogy Quarterl y
(1953) pp. 255-292.
Tubbs, R. M., Messier, W. F. & Knechel, W. R., Recency Effects in the Auditor’s Belief-Revision Process,
The Accounti ng Revi ew (1990) pp. 452-460.
Tversky, A. & Kahneman, D., Judgement under Uncertainty: Heuristics and Biases, Science (1974) pp.
1124-1131.
Walker, 1.. Thibaut, J. & Andreoli, V., Order of Presentation at Trial, Yule Law J ournal (1972) pp. 216-
226.
INFORMATION ORDER EFFECTS IN A LABORATORY MARKET 103
Williams, A. W., Computerized Double-Auction Markets: Some Initial Experimental Results, J oumnl OJ
Business (1980) pp. 235-258.
Zarowin, P., Does the Stock Market Overreact to Corporate Earnings Information?, J ournal of Finance
(1989) pp. 1385-1400.
Zarowin, P., Size, Seasonality, And Stock Market Overreaction, J ournal of Financial and Quantitative
Analysis (1990) pp. 113-126.
doc_639488099.pdf