American Economic Association
Asset Markets: How They Are Affected by Tournament Incentives for Individuals
Author(s): Duncan James and R. Mark Isaac
Source: The American Economic Review, Vol. 90, No. 4 (Sep., 2000), pp. 995-1004
Published by: American Economic Association
Stable URL: http://www.jstor.org/stable/117320 .
Accessed: 17/04/2014 14:10
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
.
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The
American Economic Review.
http://www.jstor.org
This content downloaded from 199.79.169.81 on Thu, 17 Apr 2014 14:10:45 PM
All use subject to JSTOR Terms and Conditions
Asset Markets:How They Are Affected by Tournament
Incentives for Individuals
By DUNCAN JAMES AND R. MARK ISAAC*
Tournamentincentives have been extensively
analyzed by economists, experts in organizational behavior, and the business press. The
analysis of tournamentincentives has most often looked at the effect of tournamentcontracts
for individualson individualbehavior.This paper examines the effect of tournamentincentives on overall marketperformance.
In an asset marketsetting, a numberof questions about market performance assert themselves. The most importantof these questions
is: does the use of such tournamentcontractsfor
traders affect bubbleformation? This question
is especially topical: mutual funds are increasingly dominantin capital markets, and mutual
fund managersare generallyheld to be compensated in proportionto the degree to which they
"beat the market."
In particular,we seek to determineif tournament contracts are helpful, harmful, or irrelevant to asset double auction performance.To
this end, we present a theoreticaldiscussion of
the existence and natureof equilibriumin asset
marketswith tournamentincentives in place for
the decision makers. We present the results of
laboratoryexperiments designed to illuminate
the kind of behavior we might expect from
actual people facing tournamentincentives in
an actual institution(an asset double auction).
The paper is organized as follows. Section I
provides a discussion of the tournamentincentives literature.Section II presents an a priori
theoreticaldiscussion of the effect on asset market equilibrium of introducing tournamentin* James: Departmentof Economics, Fordham University, Bronx, NY 10458; Isaac: Departmentof Economics,
University of Arizona, Tucson, AZ 85721. We thank Bill
Fenwick, without whom this paper would not have been
written.We acknowledge with many thanks the computerized double auction trading program developed by
ArlingtonWilliams of IndianaUniversity and the financial
support of the National Science Foundation (Grant No.
SBR9809741). Two anonymousreferees made many helpful suggestions. Any errorsare our own.
centives for individuals. Section III details the
hypotheses at which we aim our experiments.
Section IV details our experimentaldesign. Section V reportsour empiricalresults. Section VI
concludes.
I. LiteratureReview
The literature on tournamentshas included
researchfocusing on such topics as employment
contracts (Barry J. Nalebuff and Joseph E.
Stiglitz, 1982), sportsevents (RonaldG. Ehrenburg and Michael L. Bognanno, 1990), and
work teams inside the firm (Clive Bull et al.,
1987). There are two regularities across this
literature.First,the focus of analysis is typically
the role of tournamentcontractsinside the firm.
The marketin which the firm using the tournament contract operates is seldom considered.
Second, the tournamentcontractis usually described as increasing the wealth of the firm's
owners, and as being an appropriateresponse to
an environmentcharacterizedby moral hazard
(an environmentin which employees shirk).
These regularitiesare,of course, not perfectly
descriptive of the literature.First, there is an
acknowledgmentthattournamentcontractsmay
have effects on aspects of behavior other than
effort. Examples here include Nalebuff and
Stiglitz, and Ehrenbergand Bognanno, both of
which see tournamentsas potentiallyalteringan
individual's adoption of risky strategies. Second, there is a relatively small subliterature
which looks at the effects of tournamentcontracts for individuals on individual behavior in
market settings. An example is the work of
Keith C. Brown et al. (1996). The authors argue that paying money managers to "beat the
market"'sets up a tournamentwithin the mutual
fund industry. From this point they argue that
managerswho trail the marketmidway through
1 That is, all the other fund managers.
995
This content downloaded from 199.79.169.81 on Thu, 17 Apr 2014 14:10:45 PM
All use subject to JSTOR Terms and Conditions
THEAMERICANECONOMICREVIEW
996
SEPTEMBER2000
the evaluationperiod will reallocate their holdings to relatively risky assets.2
These latter papers suggest that tournament
incentives can lead to risky, or even destructive,
individual behavior. How such individual behavior might impact marketprocesses and outcomes is an intriguing question as yet
unexplored.In the theory section of our paper,
we provide a specific example of effects of
tournamentcontracts on individual agent behavior, and further show the implications of
these effects for asset marketperformance.
Finally, we point out that tournamentcontracts are of more than academic interest. Examples of tournamentcontractsin the business
world abound-for instance, the following anecdote refers to a series of events at Fidelity
Investments.
equilibrium constructed by Tirole). The SSW
result suggests that, in laboratoryasset markets
based on the Tirole model, the relative size of
expected capital gain and expected dividend
payouts is endogenous, and over the course of
repeated experiments converges to expected
dividend payout dominating expected capital
gain. When this has come to pass, subjects
coordinateon intrinsicvalue as the unique asset
valuation as in the finite horizon Tirole model.
What happens to this model, and its predictions, when tradersare compensatedby means
of a "beat-the-market"
tournamentcontract instead of absolute earnings?Simply put, tournament contractsfor the individualtradersdestroy
simple notions of intrinsicvalue. There are two
components of Tirole's construction that are
relevant for our analysis:
Dismayed by the change of directionand
all of the risk-aversedicta-and lured by
higher salaries at companies whose funds
were performingbetter-the stars began
their exodus. "It was a lose/lose situation," explains one of the departingcrew.
"Everything,your compensation,all your
incentives, were tied to your performance
against other funds in your class. Obviously, in orderto departfrom the mean, to
do better than the average you had to do
something different. You had to take
risks." (Andrew Cohen, 1997)
1. In a rationalexpectationsequilibriumof a T
period market, prices converge in the final
period to intrinsic value, consisting of the
expected value of period T dividends.
2. The establishmentof equilibriumin the previous periods is based on backward induction from the period T result.
People do notice and react to the kind of incentives we investigatein a controlledenvironment.
II. Theory
Jean Tirole (1982) gives a model of asset
pricing, using a backwardinduction argument.
Tirole's model suggests intrinsic value as the
unique asset valuation(and that all agents know
this, and hence that no trade occurs).
Empirically, Vernon L. Smith et al. (1988;
hereafterSSW) found that repeatedplay in an
asset market using the same group of participants eventually leads toward the development
of common expectations for those particular
people at that particular time (and toward the
2 On this basis, Brown et al. perform an ANOVA analysis within which about52 percentof the data are consistent
with the authors'conjectureabout asset reallocation.
The introduction of tournament performance
contractsalters each of the two components of
the standardmodel.
First, one can no longer conclude that prices
in period T will converge to the expected value
of the final dividenddraw (intrinsicvalue), even
in the presence of common expectationsruling
out period T capital gains. With tournament
contracts,intrinsic value may not be a marketclearing equilibrium; there may be mutually
profitabletrades at prices higher or lower than
intrinsicvalue. Furthermore,a trader'swillingness to accept or proposea trademay dependon
the distributionof shares and cash to specific
potential tradingpartners.
Second, since convergence to intrinsic value
need not occur in the last period, a backward
induction equilibriumcannot be constructed.
In the absence of a backwardinductionequilibrium,what can transpirein an asset marketin
which agents have tournamentincentives?With
perfectly informed agents, we will observe
path-dependentequilibria,which depend on the
accrual of capital gains and dividend draws to
the various traders.Such equilibriaallow price
This content downloaded from 199.79.169.81 on Thu, 17 Apr 2014 14:10:45 PM
All use subject to JSTOR Terms and Conditions
VOL 90 NO. 4
JAMESAND ISAAC:TOURNAMENT
INCENTIVESIN ASSETMARKETS
paths other than those that follow expected
dividends.
The following example demonstratesthat the
price path need not follow intrinsic value (in
this example, price increases even as intrinsic
value decreases).
Example: Assume that we are in the second to
last periodof a T periodmarket.Assume thatno
other residual expectations of capital gains remain. Assume that there are three traders, all
risk neutral. Assume that there is exactly one
shareof stock in the market,which is set to pay
two more dividends; per period there is a 1percent chance of a $0.30 dividend, and a 99percent chance of a $0.00 dividend. Assume
thatTraderA has $1.00, TraderB has $0.90 and
a share,andTraderC has $0.80. Finally, assume
that agents are paid max{twice the difference
between their final earnings and the marketaverage final earnings, 0}.
If there is no trading, A expects to earn
$0.1999, B expects to earn $0.008, and C expects to earn $0.00. What happens if C offers,
say, $0.10 to TraderB for her share?TraderB
would accept, as it increasesher expected profit
to $0.1999. TraderC proposesthe tradebecause
her expected profit increases to $0.004. Now
assume that (the preceding trade having been
made) the dividend paid after period T-1 turns
out to be $0.30. Then at the startof period T all
threetradershave $1.00, and TraderC owns the
share. What happens if A or B offers C $0.11
for the share? C accepts, as it increases her
expectedpayoff from $0.004 to $0.2180. A or B
would consent to such a trade, because it increases the expected profit from $0.00 to
$0.0018 (for either of them).
We thus observe that price can increase as
expected dividends decrease.3
The preceding example uses tractablesituations and assumes agents who share common
expectationsand knowledge of all aspects of the
economy except future realized dividends. In
field situations, the computational complexity
of trading situations would be greater, and the
information on competitors' positions imper3 Similar examples can be constructed with different
patterns of deviation of trading from intrinsic value. For
example, one can show that mutually beneficial trades can
occur below intlinsic value.
997
fect, while the agents themselves would not
have infinite memories or infinitely quick computationalabilities.4What then might transpire
in an actual asset market with actual people
facing tournament incentives? The (rational)
pricing away from intrinsic value whose possibility was just shown is one possibility. But
given computational complexity and agents'
imperfectknowledge of the asset market,common expectations may not exist, and may not
develop. Hence price bubbles due to the lack of
common expectations may occur in additionto
or instead of the type of non-intrinsic-value
pricing shown in this example.5
Ourtheory of price formationin the presence
of tournamentincentives has the following two
implications for laboratorymarkets. First, we
can expect to see marketsthat do not converge
to the expected value of remaining dividends.
Second, thereis no reasonto believe thatlack of
convergence will mitigate with more traderexperience with tournamentcontracts.In fact, just
the opposite is possible: as traders gain more
experience with tournamentcontracts,they become more sophisticated at working out the
strategicpossibilities for "beatingthe market,"
resulting in tradingfurther from intrinsic value
with more experience.
Some might object that stocks in field situations do not have known terminal dates, and
hence the fact that tournamentincentives rule
out the intrinsicvalue backwardinductionequilibriumis irrelevantfor field asset pricing. Even
if such is the case (and it may not be, as there
are evaluation periods that punctuate time for
fund managers), then it is still possible that
tournament incentives induce pricing away
from intrinsic value by alternative means. In
4 This last point takes on particularsignificance when
one notes that tradingwould take place in real time.
S Suppose, for instance,thatone traderobserves the price
increasing in a way that is not justified by expected dividends. If she does not know exactly how her competitorsare
faring in the tournament,does she take this as evidence of
a rational commitment to betting on particular dividend
draws-a course of action that is forced on tradersby the
tournamentcontract? Or as evidence of another trader's
trying to bluff her into participatingin a more conventionally defined bubble? Beyond that, what do other traders
think she thinks?Whatdoes she thinkthey think she thinks?
Clearly, this is a situationwhere there need not be common
expectations, and hence one where bubbles in the sense of
Tirole may exist.
This content downloaded from 199.79.169.81 on Thu, 17 Apr 2014 14:10:45 PM
All use subject to JSTOR Terms and Conditions
998
THEAMERICANECONOMICREVIEW
particular,Tirole notes thatthe presence of riskloving agents can lead to the destructionof the
intrinsic value equilibrium. This is important
because tournament incentives may induce
agents who, in other situations,would behave in
a risk-neutral,or even risk-averse, manner to
behave as if they were risk loving. This has
been suggested elsewhere (e.g., Nalebuff and
Stiglitz, 1983) and can be seen in this instance
in the natureof the contractfaced by the money
manager;it is convex or "option-like"(Brown
et al., 1996 p. 88). These alternativechannelsby
which tournamentincentives might affect asset
market outcomes are not mutually exclusive.
One, some, or all of the following could also
induce pricing away from intrinsic value: induced changes in risk-takingbehavior, the destructionof intrinsic value as a focal point, or
the introductionof Knightianuncertainty.
III. Hypotheses
HYPOTHESIS 1: HO: Tournamentincentives
for all traders in an asset double auction increase divergence from the "intrinsic value
price" relative to similar marketswithouttournament incentives.
H]: -HO.
HYPOTHESIS 2: HO: Tournamentincentives
lead to greater asset reallocation in the latter
half of an experimentalsession (i.e., an evaluation period), as predicted by Brown et al.
H]: -HO.
HYPOTHESIS 3: HO: Tournamentincentives
lead to greater volume of trading than is observed in similar markets without tournament
incentives.6
H]: -HO.
6
This hypothesisis motivatedas follows. In a theoretical
Tirole asset market with common expectations, no trade
occurs. With repetition, experimental asset markets have
been shown to converge toward the Tirole prediction(e.g.,
SSW). Thus "in the limit," i.e., with exhaustive repetition,
we would expect no trade in experimental asset markets.
However, the theory section of our paperdemonstratesthat
with tournamentincentives in place, gains from tradecan be
present,even when there are common expectations.Hence,
we would expect volume undertournamentincentives to be
greater than or equal to volume under linear incentives,
given exhaustive repetition. The conjectured effect, how-
SEPTEMBER2000
IV. Experimental Design
We reportresults from a design consisting of
a 15-periodasset double auctionmarketwith no
asset reinitialization.We choose value and informationconditions comparableto those used
in previous experiments.In doing so, we can be
assured that, whether our results do or do not
supportthe hypotheses in the previous section,
the outcomes are not drivenby surprisesderived
from employing an unchartedpartof the parameter space.
The work of Smith et al. (SSW, 1988) approachesthe problemof laboratorymarketasset
designwith two principaldifferencesfromrelated
works, such as that of Robert Forsythe et al.
(1982). First, SSW did not createheterogeneous
dividend categories.Second, SSW providedfor
no reinitializationof assetsin most of theirexperiments. Thus, in a 15-periodexperimentassets
were carried forward through all periods, and
there was no infusion of new cash, except that
which came from the stochasticdividend.7
Three results from SSW are particularlyimportantfor our design. First, specifying different privatedividend values was not a necessary
condition for tradingin a laboratoryasset market. Second, the modal classification of market
outcomes at low levels of subject experience
was one of price bubbles.Third,marketstended
to converge toward intrinsic value pricing and
lower volume as groups of subjects became
more experienced. The critical design change
appears to be the lack of asset reinitialization
and the resulting 15-periodasset life. SSW conclude that "realpeople in any environmentusually do not come off the stops with common
expectations.... With experience, and its lessons in trial-and-errorlearning, expectations
tend strongly to converge and yield [a rational
expectations] equilibrium." This learning included both within-experimentlearning (bubbles tend to crashnearthe known end point) and
across-experiment learning (there was some
ever, may not be observed at less than exhaustive levels of
repetition.
7 We are conducting a similar set of experiments with
marketsin which the assets have a life of only two periods,
and are then reinitialized. We have found preliminaryresults in these marketsthat are at least qualitativelysimilarto
those reportedin this research.
This content downloaded from 199.79.169.81 on Thu, 17 Apr 2014 14:10:45 PM
All use subject to JSTOR Terms and Conditions
VOL.90 NO. 4
JAMESAND ISAAC:TOURNAMENT
INCENTIVESIN ASSETMARKETS
tendency for groups brought back multiple
times to avoid price bubbles).
These results from SSW and related work
provide the groundworkfor a criticaltest of our
hypothesesregardingtournamentcontracts.The
SSW results suggest that price bubbles in asset
markets are the result of a (natural) lack of
common expectations. After sharing multiple
experiences in the same laboratory environment, many groupsdevelop sharedexpectations
supportingrationalexpectations equilibria.But
our hypotheses suggest that tournamentcontracts can cause distortedmarketperformance,
even after the convergencetowardcommon expectations pointed out in SSW.
It would not be surprisingto observe price
bubbles in early trials of an SSW-style asset
market, because bubbles in such an environment are common, even in the absence of tournament contracts. In recognition of this, we
implementa deliberatelysequencedexperimental design BBTTBT, where B representsbaseline experimental payment contracts and T
representstournamentcontracts (the treatment
condition). The first switchover between the
two compensationcontractscomes at the point
that groups of tradersusing baseline contracts
($1 experimental = $1 U.S., as in SSW) have
historically demonstratedclear convergence to
intrinsicvalue pricing (between the second and
third market sessions). The second switchover
is intendedto control for the passage of time; it
gives us a baseline measurement late in the
experiment. The third switchover is likewise
intended to give us a treatmentmeasurement
late in the experiment.
The purpose of this sequencing is to see
whetherthe introductionof tournamentcontracts
will distortan asset market,while controllingfor
the evolutionof expectationsand the passage of
time. With our experimentaldesign, it shouldbe
possible to determinewhethertournamentcontractscan distortan asset market,even aftercommon expectations have been developed and
intrinsicvalue pricinghas been achieved.
The specific values of the experimentaldesign are set out in Table 1.
V. Results
The results for these experimentsare unambiguously supportiveof the theoretical predic-
TABLE 1-SPECIFIC
999
VALUES OF THE EXPERIMENTAL DESIGN
Number of traders:9
(same group of participantsin each marketsession)
Number of marketsessions: 6
(i.e., each market session took 2 hours; marketsessions
were held from 6:00 p.m. to 8:00 p.m. every second
work day over two weeks)
Life of assets in each marketsession: 15 periods
Total numberof periods per marketsession: 15
Sequencing of compensationcontractsacross the six
marketsessions: BBTTBT
(three switches between a baseline and a treatment
condition; i.e., the first, second, and fifth nights used
the baseline condition, $1 experimental = $1 U.S.)
Distributionof dividends: discrete uniform distribution
over ($0.00, $0.08, $0.28, $0.60)
Dividend draws: homogenous
(i.e., all tradersreceived the saine draw per share in a
given period)
Initial endowments:heterogeneous
Categories of initial endowments:3
3 shares; $3.60 (3 traders)
2 shares; $7.20 (3 traders)
1 share; $10.80 (3 traders)
Tournamentcompensationearnings into US$:
-5.00 + 2[(end of experimentcash)i - (average
startingcash + averagerealizeddividends)]if [.] > 0
= 5.00 if [-] ' 0
Evaluationperiod for tournament:each 15-periodmarket
session when treatmentwas in effect
Informationon evaluation criterionrevealed to
participants:cumulative average marketearnings
(i.e., average startingcash + average realized
dividends) announcedafter each period (underboth
baseline and treatmentconditions).
tion that tournament contracts lead to
divergence from pricing at intrinsic value.
The most importantstylized fact thatemerges
from the previous 15 years of research in experimentalasset marketsis thatrepeated,shared
trading experience under the baseline contract
promotes convergence toward intrinsic value
pricing. We are able to replicate this finding.
However, we also find that the imposition of
tournament contracts reverses this process!
Repeated,sharedtradingexperience when tournament incentives are in place promotes divergence from intrinsic value pricing.
The data are visually reportedin the following form. The price data are the mean trading
prices per period, measured as deviation from
intrinsic value. These are displayed in Figure
1 in sequentialorder.Volume by period is similarly displayed in Figure 2. Additionally,
This content downloaded from 199.79.169.81 on Thu, 17 Apr 2014 14:10:45 PM
All use subject to JSTOR Terms and Conditions
SEPTEMBER2000
THEAMERICANECONOMICREVIEW
1000
4
2
Session 1
Baseline
Session 2
Baseline
---
----
--
-
Session 4
Tournament
Session 3
Tournament
-
-
________.
--
_
Session 6
Tournament
Session 5
Baseline
_
--
--
_-
.20
Co
~
CL
-1
*
---0C)
0
-'1
____.-
-.
-
1
C~1
C' C'
C)
C)
~)
0
__
C)
~
D
0
-
___
0~c~
'
10
LO0
.._ .__
(0
co
... _
0
._
CD
f-
.-
__
0
f-
C1
1
w
Gaps indicate no trades.
-2
-
_
_
_
--
-
_
_
_
_-
-
_
_
-
-
-
-
-
-
-
-
_
_
-
-
-3Trading Period
FIGURE1. MEAN PRICEDEVIATIONBY PERIOD
empirical density functions for volume appear
in Figure 3.
The firsttwo marketsessions (baseline incentives) show a classic convergencefrom a bubble
towardintrinsicvalue, as has been observed by
other researchers.8The third session is difficult
to interpret.It is closer to intrinsic value than
'The second session was much closer to converging to
intrinsic value than can be apparentfrom the price graphs.
Trader4 was the buyerin 20 of the 30 tradesabove intrinsic
value. He had been the top earnerin the first session, which
bubbledseverely. He was clearly trying to starta bubble in
this session also, and drove his earnings down to $1.40
(comparedwith average earningsof $13.04) in the attempt.
We reasoned that it was appropriateto go forward with
the switch of treatmentsbecause all the other tradershad
demonstrateda willingness to sell into the buying pressure
that Trader4 was trying to generate,and furtherthat Trader
4 would be disciplined by this experience, and would develop similarexpectationsto the rest of the groupbefore the
startof the thirdsession. We believe that Trader4's behavior in the third session is consistent with our conjecture;in
an environment theoretically conducive to pricing away
from intrinsic value, he did not attempt to start a bubble
single-handedly, as before. Rather, he "scalped" on the
session 1, but there also appearto be multiperiod moves away from intrinsicvalue. As such,
were session 3 to be judged on a stand-alone
basis, it would be hard to draw clear conclusions. Fortunately,such is not the case; thereare
furthersessions of data that can make the case
conclusively. The fourth session (tournament
incentives) diverges furtherfrom intrinsicvalue
than session 3; this is odd in that convergence
has been found by other researchers to be
roughly monotonic in subject experience. At
this point one might claim that these are simply
subjects who do not understandwhat they are
doing; that they might never converge, regardless of the type of incentive treatment used.
Hence session 5 again employs baseline incentives; the subjects locked on to trading at (or
very close to) the intrinsic value price throughout the experiment. At this point one might
claim that the subjects converged to intrinsic
bid/offer fluctuations generated by the other, previously
converged traders.
This content downloaded from 199.79.169.81 on Thu, 17 Apr 2014 14:10:45 PM
All use subject to JSTOR Terms and Conditions
1001
INCENTIVESIN ASSETMARKETS
JAMESAND ISAAC:TOURNAMENT
VOL.90 NO. 4
20 -Session 1
Session 2
Session 3
Session 4
Session 5
Session 6
Baseline
Baseline
Tournament
Toumament
Baseline
Toumament
18-
16
14
12
=10
8
6-
a
4-
ro-
C
ct)
CD
N
Co4 0
C>l
cN
O
C>l
o
')LO
c
C(
O
C
C
LO U Co
r-
r-_
- r-
N
C
Co
o
CO
C
Co
TradingPeriod
FIGuRE
2.
VOLUME BY TRADING PERIOD
14E
12*
E
10
Baseline
Tournament
.
*0
---
0)8
0~
E
z
-]---
li
46-204
17
2
0
1 2
3
4
5
6
7
8
9 10 11 12 13 14 15 16 17 18 19 20
Volume Per Period
FIGURE
3.
DISTRIBUTION OF VOLUME BY TREATMENT
value in session 5 because they were fourfold
more experienced, but failed to converge in
session 4, not because they were faced with
tournamentincentives, but because they were
only threefoldmore experienced.Hence session
6 again employs tournamentincentives; trading
diverges wildly from intrinsic value. Note that
this occurs after the subjects had demonstrated
This content downloaded from 199.79.169.81 on Thu, 17 Apr 2014 14:10:45 PM
All use subject to JSTOR Terms and Conditions
1002
THEAMERICANECONOMICREVIEW
the ability to tradeat intrinsicvalue for an entire
experiment,given baseline incentives.9'10
We now turn to a time-series statisticalanalysis of this data. Time (as a proxy for the
convergence towardcommon expectations)has
been suggested by prior researchersto be the
foremost explanator of price deviations from
equilibrium. Hence our econometric analysis
comparesthe explanatorypower of a regression
employing a time trend, intercept treatment
dummies, and slope treatmentdummies as regressors to the explanatorypower of a regression employing only a time trendas a regressor.
The dependent variable in the regressions
9 Trader8 did not show up at the preset time for session
6. A protocol describing how this situation was handled is
available from the authorson request.
10Four of the last six tradesin the last period of session
6 suggest extreme subject frustrationwith the tournament
contract. In particular,two traderswho were not going to
make money under the tournamentcontract entered what
appearto be frustrationbids of (sequentially) $5.00, $5.00,
$9.99, and $1.22. These bids were of course snappedup by
other traders, who were thus pushed above the market
average,and so made money underthe tournamentcontract.
There are two explanations for these four trades: frustration or collusion. Under the tournament contract, the
maximum extractionof cash from the experimenteroccurs
if all nine tradersconspire to concentrateall the working
capital and sharesin the experimentin the hands of a single
trader(and then divide thatmoney among themselves later).
Hence there is a collusive optimum for the subjects that is
not of interest to the experimenter.This collusive outcome
would be likely if all nine subjectsknew each otherwell and
could coordinate their actions in the time between market
sessions. For example, with subjects drawn from, say, the
same floor of a dormitory,this would be a problem.Otherwise, the collusive outcome is extremely unlikely. Why?
Because for a "sweetheart"bid or offer to make it from one
conspiratorto another,it has to pass by seven other traders
in our design. Thus, ex ante, collusion appearsdifficult, and
potential conspirators (of number less than nine) would
realize this (ex ante). Ex post, analysis of tradessuggests not
only that collusion was never a problem, but also that
frustrationbids/offers were limited to the last 90 seconds of
the two-week series of marketsessions. For example, using
maximum holding value of a share as a way to diagnose
frustration(or collusion) trades,we find thatpriorto the last
90 seconds of the last periodof the last experimentonly two
tradesoccurredabove maximumholding value. Those were
both duringthe bubble (by the then inexperiencedsubjects)
in the first session, which employed baseline incentives
(with which collusion is not an issue).
Given that these four trades in session 6 incorporate
frustrationbids, we drop them from the analysis. Given that
these trades occurredfar above intrinsic value, this adjustment biases the results against our theoretical prediction
about pricing away from intrinsic value.
SEPTEMBER2000
TABLE2-REGRESSION
INCLUDING
TOURNAMENT
DUMMY
VARIABLES
Independentvariable
Coefficient
estimate
T statistic on
individual
coefficient
1.23
-0.036
4.80
-6.11
-4.26
-4.55
Intercept
Time trend
Experiments3 and 4
interceptdummy
Experiments3 and 4 slope
dummy
Experiment6 intercept
dummy
Experiment6 slope
dummy
0.095
-24.73
0.31
4.56
-5.64
5.94
Notes: Dependent variable: natural log of summed price
deviations (by period). R2 = 0.49; n = 75.
below deserves one quick, separatecomment. It
is the natural log of summed price deviations
per period. This is arguablya more meaningful
dependentvariable than mean or closing price,
as it incorporatesvolume. (After all, which is
more persuasiveevidence of bubbleactivity in a
given period: one contract 50 percent above
intrinsic value, or eight contracts 40 percent
above intrinsic value?'1)
The statisticalresults for the unrestrictedregression (employing dummies for experiments
using tournamentincentives) are given in Table
2. The line-fit for this regression is quite striking, and is reproducedin Figure4.12 The line-fit
in Figure 4 illustratesthat the effects of tournament incentives are directly counter to the previously documentedeffects of repeated,shared
tradingexperience.
" Using naturallogs meant that
nonpositive values had
to be addressed.These were 15 of the 90 observations.The
results reportedhere involve removing the nonpositive values. These results are robust with respect to a number of
alternativeways for dealing with the nonpositive values.
12 The reported regression contains separate slope and
intercept dummies for session 6 (the last T) but not for
session 5 (the last B). We conducted a similar regression
where sessions 5 and 6 are treatedsymmetrically,and thus
each design block ([BB], [TT], [B], and [fl) is accounted
for in the structureof the dummy variables.The results are
as follows: in the "symmetric"regression,all of the original
coefficients (on the dummy variables) keep the same sign
and virtually the same magnitude, and remain significant.
The time-trendcoefficient loses its significance. The coefficients for the two additionaldummy variablesfor session
5 are insignificant.The adjustedR2 increases to 0.54.
This content downloaded from 199.79.169.81 on Thu, 17 Apr 2014 14:10:45 PM
All use subject to JSTOR Terms and Conditions
INCENTIVESIN ASSETMARKETS
JAMESAND ISAAC:TOURNAMENT
VOL.90 NO. 4
43
TABLE3-REGRESSION WITHOUT
TOURNAMENT
DUMMYVARIABLES
A
0
2
X
8
2-
0
1003
A
AAii
A
[;
9
Pred'icted
Variable
Intercept
Time trend
Time(TradingPeriod)
Coefficient estimate
T statistic on
individual
coefficient
0.50
-0.015
1.71
-2.84
Notes: Dependent variable: natural log of summed price
deviations (by period). R2 = 0.09; n = 75.
LINE-FIT
FIGURE4. REGRESSION
Notes: Horizontal axis is "time" measured as consecutive
tradingperiods. Vertical axis is ln(sum of price deviations).
In contrast,the regression using only a time
trend as an explanatoryvariable yields the results in Table 3. An F-test for joint significance
of the variables excluded from this regression
(the tournamenttreatment dummies) strongly
rejects that they do not have collective explanatory power. Specifically, the calculated Fstatistic is 15.68, whereas the critical value for
F(4, 69) is 3.60 at the 99-percent level.
All of the above-mentionedways of examining the data point to one conclusion: that Hypothesis 1 is supportedempirically.What about
the other hypotheses?
Hypothesis 2 maps the Brown et al. (1996)
conjectureinto our asset marketby comparing
the relative proportions of "early" and "late"
trading in the baseline and tournamentconditions. (The computerizedmarketdoes not keep
a recordof subjects' working capital at the time
of each trade, and asking subjects to do so
would interfere greatly with the mechanics of
trading.Thereforewe cannottest the conjecture
using informationon whethersharebuyers trail
the market at the time of each trade, as do
Brown et al., 1996.) Hypothesis 2 is not supportedat the 95-percentlevel, but it is supported
at the 90-percentlevel. Specifically, employing
a differencein proportionstest thatassumesthat
the difference in sample proportionsbetween
two populations is asymptotically normal, the
calculated value is 1.61, whereas the critical
value is 1.65 at the 95 percent level or 1.29 at
the 90 percent level (Wiiliam C. Merrill and
Karl A. Fox, 1970). In other words, the tournament condition has more tradingin the second
half of each marketsession than does the base-
line condition (an approximationto the BrownHarlow-Starksconjecture).
Hypothesis 3 would be "given its best shot"
in an experiment with exhaustive repetition.
Unfortunately, other design considerations
ruled this out.13Given the design we used we
found a total of 153 tradesunderthe tournament
and 120 under the baseline.14 A number of
statistical tests are potentially useful here, and
all of them-a difference in means test pitting
voluine in sessions 1, 2, and 5 againstvolume in
sessions 3, 4 and 6; a differencein means test on
the analogous data from only sessions 5 and 6;
and a Kolmogorov-Smirnovtest on the histograms in Figure 3-have the right sign to be
consistent with the null, but the nonexistence of
this effect is not rejected statistically.
VI. Conclusion
Tournamentcontractscan have clear and destructiveeffects in asset markets.They generate
misleading prices. Both theory and experimental evidence suggest that this problemcan exist.
Furthermore,there are multiple paths by which
13
In determiningthe treatmentsequencing, we wished
for the overall length of the experimental series to be
exogenous, and known to the subjects to be so, and for the
motivationfor treatmentswitchovers to be unknown to the
subjects. Both of these considerationswere importantfor
the same reason:we did not want the subjectsadjustingtheir
behavior so as to manipulateour choice of treatments.For
example, if we had (endogenously) run the initial baseline
sequence until complete convergence to the Tirole equilibrium, and then switched treatments,the subjectsmight infer
that not trading would produce a treatment switch, and
might find it or imagine it to be in their self-interestto bring
such about.Furthermore,such a design, in additionto being
vulnerableto subject manipulation,would have potentially
been extremely expensive.
14 This count omits the "frustration"
trades omitted as
discussed in footnote 11.
This content downloaded from 199.79.169.81 on Thu, 17 Apr 2014 14:10:45 PM
All use subject to JSTOR Terms and Conditions
1004
THEAMERICANECONOMICREVIEW
tournamentincentives might affect marketoutcomes. Beyond the breakdownof backwardinduction,the results from the laboratorycould be
the result of a tournament contract induced
change in risk-takingbehavior, the presence of
Knightianuncertainty,the destructionof intrinsic value as a focal point, or all of the preceding.
As such, even if backwardinduction does not
matter in a field situation, the applicability of
our laboratoryresults to field situations cannot
yet be ruled out.
Tournamentcontracts impair rational price
formation in asset markets; if this regularity
carries over to field situations, it could lead to
the misallocationof capitalresourcesand attendant undesirablemacroeconomiceffects. Similar things have occurredin the recent past. The
FSLIC debacle is a particularlycostly example
of risk-inducingincentives for agents working
to the regret of the principals.Ironically,in the
case of tournamentincentives, contractswhose
inspiration was the elimination of shirking
within the firm may worsen the principal-agent
conflict between the firm and its clients (mutual
fund depositors).Moreover,one should not lose
sight of the fact thatsociety as a whole-including those without mutualfund deposits-stands
to lose if choice among the investments that
transformthe economy throughtime is warped
by the contractsfacing money managers.
Although there may be institutional details
present in field situations that mitigate the effects documented here, it is not obvious what
those details might be. This does not imply that
mitigating factors do not exist, but rathersuggests two courses for future research. First,
thereis a need for researchinto the existence (or
lack) of such mitigating factors.15Second, if
such factors cannot be found, there is a need to
investigate regulatory and incentive alterna15
For example, in new researchwe are investigatingthe
effects of having the tournamentcontract apply to only
some of the traders.
SEPTEMBER2000
tives. We need to know, "Would a regulatory
cure be any better than the disease?" We need
furtherstudy-by policy makers as well as by
theoreticians and experimentalists-to do justice to that policy question.
REFERENCES
Brown, Keith C.; Van Harlow, W. and Starks,
Laura T. "Of Tournaments and Temptations:
An Analysis of ManagerialIncentives in the
Mutual Fund Industry." Journal of Finance,
March 1996, 51(1), pp. 85-110.
Bull, Clive; Schotter, Andrew and Weigelt, Keith.
"Tournamentsand Piece Rates: An Experimental Study." Journal of Political Economy,
February1987, 95(1), pp. 1-33.
Cohen,Andrew."Abby's Road."Boston Magazine, April 1997, pp. 50-55, 91-94, as reposted on Boston Magazine Online, 1997.
Ehrenburg, Ronald G. and Bognanno, Michael L.
"Do TournamentsHave Incentive Effects?"
Journal of Political
Economy, December
1990, 98(6), pp. 1307-24.
Forsythe, Robert; Palfrey, Thomas R. and Plott,
CharlesR. "AssetValuationin an Experimental Market." Econometrica, May 1982, 50(3),
pp. 537-67.
Merrill, William C. and Fox, Karl A. Introduction
to economic statistics. New York: Wiley,
1970.
Nalebuff, Barry J. and Stiglitz, Joseph E. "Prizes
and Incentives:Towardsa GeneralTheory of
Compensationand Competition."Bell Journal of Economics, Spring 1983, 14(1), pp.
21-43.
Smith, Vernon L.; Suchanek, Gerry L. and Williams, Arlington W. "Bubbles, Crashes, and
Endogenous Expectations in Experimental
Spot Asset Markets."Econometrica,September 1988, 56(5), pp. 1119-53.
Tirole,Jean. "On the Possibility of Speculation
Under Rational Expectations." Econometrica, September 1982, 50(5), pp. 1163-82.
This content downloaded from 199.79.169.81 on Thu, 17 Apr 2014 14:10:45 PM
All use subject to JSTOR Terms and Conditions