-
Zurich Open Repository andArchiveUniversity of ZurichMain
LibraryStrickhofstrasse 39CH-8057 Zurichwww.zora.uzh.ch
Year: 2001
Does Money Illusion Matter?
Fehr, Ernst ; Tyran, Jean-Robert
Abstract: This paper shows that a small amount of
individual-level money illusion may cause considerableaggregate
nominal inertia after a negative nominal shock. In addition, our
results indicate that negativeand positive nominal shocks have
asymmetric effects because of money illusion. While nominal inertia
isquite substantial and long lasting after a negative shock, it is
rather small after a positive shock.
DOI: https://doi.org/10.1257/aer.91.5.1239
Posted at the Zurich Open Repository and Archive, University of
ZurichZORA URL: https://doi.org/10.5167/uzh-95345Journal
Article
Originally published at:Fehr, Ernst; Tyran, Jean-Robert (2001).
Does Money Illusion Matter? American Economic
Review,91(5):1239-1262.DOI:
https://doi.org/10.1257/aer.91.5.1239
-
Does Money Illusion Matter?
By ERNST FEHR AND JEAN-ROBERT TYRAN*
This paper shows that a small amount of individual-level money
illusion may causeconsiderable aggregate nominal inertia after a
negative nominal shock. In addition,our results indicate that
negative and positive nominal shocks have asymmetriceffects because
of money illusion. While nominal inertia is quite substantial
andlong lasting after a negative shock, it is rather small after a
positive shock. (JELC92, E32, E52)
Until recently, the notion of money illusionseemed to be
thoroughly discredited in moderneconomics. James Tobin (1972)
described thenegative attitude of most economic theoriststowards
money illusion as follows: “An eco-nomic theorist can, of course,
commit no greatercrime than to assume money illusion” (p. 3). Asa
consequence, money illusion has been anath-ema to the profession
for several decades. Theindex of the Handbook of Monetary
Economics(Benjamin M. Friedman and Frank M. Hahn,1990), for
example, does not even mention theterm “money illusion.” In
principle, money il-lusion could provide an explanation for the
in-ertia of nominal prices and wages and, thus, forthe
nonneutrality of money. The stickiness ofnominal prices and wages
seems to be an im-portant phenomenon (see, e.g., George A.Akerlof
et al., 1996; Ben S. Bernanke andKevin Carey, 1996; David Card and
DeanHyslop, 1997; Shulamit Kahn, 1997; Truman F.
Bewley, 1998; Alan S. Blinder et al., 1998). Ithas puzzled
economists for decades because it isquite difficult to explain in
an equilibriummodel with maximizing individuals. Instead ofmoney
illusion other factors like informationalfrictions (Robert E.
Lucas, Jr., 1972), staggeringof contracts (e.g., Stanley Fischer,
1977; JohnB. Taylor, 1979), costs of price adjustment (N.Gregory
Mankiw, 1985), and near-rationality(Akerlof and Janet L. Yellen,
1985) have beeninvoked to explain nominal inertia.
In this paper we do not contest the potentialrelevance of these
explanations. We do, how-ever, argue that money illusion has
prematurelybeen dismissed as a potential candidate for
theexplanation of sluggish nominal price adjust-ment. Our argument
is based on rigorous exper-imental evidence from a price-setting
game thatisolates money illusion from other potential de-terminants
of nominal inertia. In particular, weshow that after a fully
anticipated negative nom-inal shock, long-lasting nominal inertia
pre-vails, even if informational frictions, costs ofprice
adjustment and staggering are absent. Ourresults indicate that the
direct and indirect ef-fects of money illusion are the major
determi-nants of this long-lasting nominal inertia. Weshow, in
addition, that money illusion causesmuch less nominal inertia after
a fully antici-pated positive nominal shock. This result
isreminiscent of the Keynesian proposition thatdownward wage
rigidity causes asymmetric re-sponses to monetary shocks. Yet,
since we ob-tain our result in a price-setting game, theasymmetric
response cannot be directly relatedto downward wage rigidity. Our
results suggestthat the asymmetry is caused by a particularform of
money illusion arising from people
* Fehr: Institute for Empirical Research in Economics,University
of Zurich, CH-8006 Zurich, Switzerland; Tyran:Department of
Economics, University of St. Gallen, CH-9000 St. Gallen,
Switzerland. We are particularly gratefulfor two excellent referee
reports and for comments byGeorge Akerlof, Linda Babcock, Jim Cox,
Urs Fischbacher,Simon Gächter, Ed Glaeser, Lorenz Goette,
CharlesGoodhart, Reinhard Selten, Dick Thaler, Frans van Winden,and
Michael Waldman. In addition, we acknowledge help-ful comments by
the participants of many seminars aroundthe globe. Ernst Fehr
gratefully acknowledges the hospital-ity of the Center for Economic
Studies (CESIFO) in Mu-nich. Valuable research assistance has been
provided byMartin Brown, Beatrice Zanella, and Tobias Schneider.
Weare grateful for financial support by the Swiss NationalScience
Foundation under Project No. 1214-051000.97/1,and by the EU-TMR
Research Network (Project No.FMRX-CT98-0238).
1239
-
taking nominal payoffs as a proxy for real pay-offs. After a
negative money shock, nominalpayoffs decline because prices tend to
decline,while after a positive shock nominal payoffsincrease
because prices tend to rise. If thesechanges in nominal payoffs are
taken as a proxyfor changes in real payoffs there will be
morereluctance to adjust prices to the new equilib-rium after a
negative shock.
Our experiments also allow us to judge therelative importance of
the direct and indirecteffects of money illusion on nominal
inertia.The direct effects of money illusion are definedas those
effects that are the direct result ofindividual optimization
mistakes. The indirecteffects of money illusion are defined as
thoseeffects that arise because some agents expectthat others are
prone to money illusion and, as aconsequence, they behave
differently. The dis-tinction between the direct and the indirect
ef-fects of money illusion is important becausemany economists seem
to believe that moneyillusion is not a widespread phenomenon at
theindividual level, i.e., that the direct effects ofmoney illusion
are small. The textbook examplewhere all nominal prices and nominal
incomesare doubled nicely illustrates this view. It ishard to
believe that many people make an indi-vidual optimization mistake
by choosing a dif-ferent bundle of goods when prices and incomesare
doubled. Our results clearly show, however,that it would be
misleading to conclude thatmoney illusion is largely irrelevant
because thedirect effects of money illusion are small. In
ourexperiments the direct effects of money illusionon nominal
inertia after the negative shock arealso rather small but the total
effects neverthe-less are very large. The reason for this finding
isthat money illusion renders price expectationsvery sticky after
the negative shock, which—under conditions of strategic
complementar-ity—induces agents to choose sticky prices.This result
lends support to theories that stressthat small amounts of
individual-level irratio-nality can have large aggregate effects
(Akerlofand Yellen, 1985; John Haltiwanger andMichael Waldman,
1985, 1989; Thomas Russelland Richard Thaler, 1985). It also lends
supportto the view of George W. Evans and GareyRamey (1992, 1998)
that costly expectation for-mation causes expectations and prices
to adjustonly gradually to the rational expectations equi-
librium. Although there are no direct costs offorming
expectations in our experiments, it isquite likely that the task of
forming expectationsinvolves cognitive costs. Taken together,
theresults of our experiments suggest that moneyillusion matters,
i.e., money illusion should beconsidered as a serious candidate in
the expla-nation of nominal inertia.
The rest of the paper is organized as follows:In Section I we
discuss the notion of moneyillusion and its potential aggregate
implicationsin more detail. In Section II we argue thatexperimental
methods are appropriate for study-ing whether money illusion
matters and wepresent our experimental design. In Section IIIthe
experimental results of the design with thenegative nominal shock
are presented. SectionIV argues that the nature of money illusion
inour experiment suggests that after a positivenominal shock there
should be less nominalinertia. This conjecture is tested in a
design witha positive nominal shock. In the final section
wesummarize and interpret our main results.
I. Money Illusion at the Individual
and the Aggregate Level
A. Money Illusion at the Individual Level
Wassily Leontief (1936) defined money illu-sion as a violation
of the “homogeneity postu-late.” This postulate stipulates that
demand andsupply functions are homogeneous of degreezero in all
nominal prices which means that theydepend only on relative and not
on absoluteprices. Although other authors have usedslightly
different definitions, the intuition be-hind their definitions
seems to be rather similar.This intuition says that if the real
incentivestructure, that is, the objective situation, an
in-dividual faces remains unchanged, the real de-cisions of an
illusion-free individual do notchange either. Two crucial
assumptions underlythis intuition: First, the objective function of
theindividual does not depend on nominal but onlyon real
magnitudes. Second, people perceivethat purely nominal changes do
not affect theiropportunity set. For example, people have
tounderstand that an equi-proportionate change inall nominal
magnitudes leaves the real con-straints unaffected. Whether people
are, in fact,able to pierce the veil of money, i.e., whether
1240 THE AMERICAN ECONOMIC REVIEW DECEMBER 2001
-
they understand that purely nominal changesleave their objective
circumstances unchanged,is an empirical question. Irving Fisher
(1928),for example, was convinced that ordinary peo-ple are, in
general, prone to money illusion.
More recently Eldar Shafir et al. (1997) pro-vided interesting
questionnaire evidence indi-cating that frequently one or both
preconditionsfor the absence of money illusion are violated.Their
results suggest that the preferences ofmany people as well as their
perceptions of theconstraints are affected by nominal
values.Moreover, the answers of many people do notonly indicate
that they themselves are prone tomoney illusion but that they also
expect otherpeople’s behavior to be affected by moneyillusion.
Since the absence of money illusion meansthat an individual’s
preferences, perceptionsand, hence, choices of real magnitudes are
notaffected by purely nominal changes, it is naturalto view money
illusion as a framing or repre-sentation effect. From this
viewpoint, an indi-vidual exhibits money illusion if his or
herdecisions depend on whether the same environ-ment is represented
in nominal or real terms.There is a large body of experimental
researchthat shows that alternative representations of thesame
situation may well lead to systematicallydifferent responses
(Reinhard Selten and ClausC. Berg, 1970; Amos Tversky and
DanielKahneman, 1981). Representation effects seemto arise because
people tend to adopt the par-ticular frame that is presented and
evaluate theoptions within this frame. Because some op-tions loom
larger in one representation than inanother, alternative framings
of the same op-tions may provoke different choices.
It is important to note that the nominal rep-resentation of an
economic situation is probablythe natural representation for most
people. Thisis so because most economic transactions inpeople’s
lives involve the use of money and,hence, are framed in nominal
terms. Therefore,it is likely that people often perceive and
thinkabout economic problems in nominal termswhich may induce money
illusion. A ratherbasic form of money illusion occurs when peo-ple
take nominal values or changes in nominalvalues as a proxy for real
values or changes inreal values, respectively. Note that this rule
ofthumb makes perfect sense in an environment
with a given aggregate price level. However,this rule is
inappropriate in situations where theaggregate price level is
changing. Therefore, theapplication of this rule in an environment
withchanging aggregate prices constitutes a form ofmoney
illusion.
B. Money Illusion at the Aggregate Level
In the past, economists frequently invokedthe assumption of
money illusion to accountfor the short-run nonneutrality of money
(e.g.,Fisher, 1928). However, since the success ofthe rational
expectations revolution, econo-mists have been extremely reluctant
to invokemoney illusion to explain the short-run non-neutrality of
money. A common feature of themodels of New Classical and New
Keynesianmacroeconomists is that they exclusively fo-cus on the
equilibrium states of their econo-mies. In general, they remain
silent on howeconomic agents move from one equilibriumto the other.
In models that exclusively focuson equilibrium, the assumption of
the absenceof money illusion is very intuitive because itis
difficult to imagine that an illusion couldpersist in equilibrium.
However, there is astrong a priori argument that money illusionis
likely to affect the adjustment process of aneconomy after a fully
anticipated monetaryshock. This argument is based on the simplefact
that in an interactive situation the failureof some agents to fully
adjust to the nominalshock will, in general, provide incentives
forother agents to not fully adjust to the shock,either. Thus,
there may be a snowball effectthat causes less than full adjustment
for aprolonged period of time.
This can be illustrated in the context of amonopolistically
competitive economy as ana-lyzed in, for example, Akerlof and
Yellen(1985) or Olivier Jean Blanchard and NobuhiroKiyotaki (1987).
To keep the argument simplewe focus solely on the firms’ behavior.
Thereduced-form real profit function for firms inthese models can
be written as
(1) p i 5 p i ~P i /P# , M/P# !, i 5 1, ... , n
where pi is firm i’s real profit, Pi is the nominalprice set by
firm i, P# is the aggregate pricelevel, M denotes the supply of
money, and n the
1241VOL. 91 NO. 5 FEHR AND TYRAN: DOES MONEY ILLUSION
MATTER?
-
number of firms.1 In these models M/P# is pro-portional to real
aggregate demand. For sim-plicity, we assume identical firms, the
absenceof menu costs and informational frictions, and aunique and
symmetric equilibrium P*i 5 P*j , forall i, j. In this equilibrium
each firm maximizesreal profits by setting P*i 5 P# *. Since (1)
ishomogeneous of degree zero in Pi , P# , and M, achange in M to lM
(l Þ 1) leads to post-shockequilibrium values of lP*i and lP#
*.
Suppose now that there is one group of agentswho suffers from
money illusion and does,therefore, not fully adjust their nominal
pricesto lP*i. Suppose further that there is a secondgroup of
agents that anticipates the behavior ofthe first group. The second
group, therefore,anticipates a change in real aggregate demandM/P#
such that their members, in general, havean incentive to choose a
price that differs fromlP*i , too. Whether the interaction between
thesegroups causes aggregate nominal inertia de-pends in an
important way on the strategic en-vironment, that is, whether
agents’ actions arestrategic complements or strategic
substitutes.Haltiwanger and Waldman (1989) have shownthat in the
presence of strategic complementar-ity between agents’ decisions,
the existence of asmall group of nonrational subjects can havelarge
effects on the process of adjustment toequilibrium. In the
above-mentioned model ofmonopolistic competition, strategic
comple-mentarity means that firm i’s profit maximizingnominal price
P9i is positively related to theaggregate price level P# . This
means that firmswhich believe that, because of money illusion,the
prices of other agents are kept close to thepre-shock equilibrium
have a rational reason tochoose a nominal price that is also close
to thepre-shock equilibrium.
Thus, under strategic complementarity ratio-nal firms have an
incentive to partly imitate thebehavior of the nonrational firms
which givesthe latter a disproportionately large impact onthe
aggregate price level. In contrast, in thepresence of strategic
substitutability, i.e., if P9i isnegatively related to P# ,
rational firms have an
incentive to partly compensate the behavior ofthe nonrational
ones so that the latter have adisproportionately small impact on
the aggre-gate outcome. The results of Haltiwanger andWaldman
(1989) thus suggest that, given stra-tegic complementarity, the
existence of a smallgroup of subjects that suffer from money
illu-sion may generate substantial nominal inertia.However, while
this is a plausible theoreticalargument, there is, to our
knowledge, no empir-ical evidence for the claim that a small
amountof money illusion may generate substantialnominal
inertia.2
II. An Experimental Approach
to Money Illusion
One way to rigorously examine whethermoney illusion matters, is
to look for a naturalexperiment in which an exogenous and
fullyanticipated monetary shock occurs. In order tounambiguously
identify whether the shock isfully anticipated, the researcher
needs to knowindividual information sets before the shock. Tojudge
whether the anticipated shock causes adisequilibrium and nominal
inertia, the re-searcher has to know the equilibrium values
ofnominal prices before and after the shock.Moreover, to examine
whether money illusioncauses nominal inertia, the researcher
shouldidentify two similar natural experiments. In oneexperiment
the “world” should be framed innominal terms while in the other
experiment itshould be framed in real terms. In our view, itseems
extremely difficult, if not impossible, tomeet the above
requirements with field data. Infact, the exogeneity of monetary
policy and thecausality between money and output is a matterof
considerable debate (e.g., Christina D.Romer and David H. Romer,
1989, 1994; KevinD. Hoover and Steven J. Perez, 1994; WilburJohn
Coleman, 1996). In addition, full knowl-edge of the pre- and
post-shock equilibrium
1 Equation (1) already incorporates (i) the maximizingbehavior
of all households, (ii) the cost-minimizing behav-ior of all firms
for given output and wages levels, (iii) theequilibrium real wage,
and (iv) the equilibrium relationbetween real aggregate demand and
real money balances.
2 Since strategic complementarity is important for ourargument
in favor of the aggregate relevance of (beliefsabout) money
illusion, one would like to know to whatextent it does prevail in
naturally occurring economies.Seonghwan Oh and Waldman (1990,
1994), Russell Cooperand Haltiwanger (1996), and Blinder et al.
(1998) provideevidence in favor of the relevance of strategic
complemen-tarity in real economies.
1242 THE AMERICAN ECONOMIC REVIEW DECEMBER 2001
-
values of nominal prices is clearly beyond theinformation
content of presently available fielddata. In reality, almost all
business transactionsare shrouded in nominal money, i.e., it is
verydifficult to find real-world examples of a realframe.
In an appropriate laboratory setting, however,the
above-mentioned data requirements can bemet. The techniques of
experimental economicsallow the implementation of exogenous
andfully anticipated nominal shocks and the exper-imenter can exert
full control over pre- andpost-shock equilibrium values of
nominalprices. In addition, the experimenter controlsthe framing of
the situation, e.g., whether sub-jects receive the payoff
information in nominalor in real terms. These enhanced control
oppor-tunities suggest that laboratory experimentsprovide valuable
information regarding the im-pact of money illusion on nominal
inertia,which may complement and help to interpret theresults of
studies based on field data (for fieldevidence see, e.g., Michael
Abbott and OrleyAshenfelter, 1976; Beth T. Niemi and CynthiaB.
Lloyd, 1981). The use of experimental meth-ods also distinguishes
our examination from thestudy of Shafir et al. (1997). While these
authorsasked subjects hypothetical questions, we di-rectly observe
the evolution of individual andaggregate behavior after a nominal
shock.
A. General Description of theExperimental Design
To study the impact of money illusion, wedesigned an n-player
pricing game with stra-tegic complementarity and a unique
equilib-rium. The pricing game was divided into apre-shock and a
post-shock phase. All nplayers simultaneously had to determine
theirnominal prices in each period of the game.
They were free to change their nominal pricesin each period at
no cost. The players interactedanonymously via computer terminals.
Eachtreatment condition had 2T periods. During thefirst T periods
of a session the money supplywas given by M0. Then we implemented a
fullyanticipated monetary shock by reducing themoney supply to M1.
This shock and the factthat the post-shock phase again lasted T
periodswas common knowledge.
Our major interest concerns subjects’ pricingbehavior in the
post-shock phase. The pre-shockphase serves the purpose to make
subjects ac-quainted with the computer terminal and thedecision
environment. In addition, and moreimportantly, the pre-shock phase
allows us tosee whether subjects reach equilibrium in thepre-shock
phase. After all, one can only arguethat money illusion is a
disequilibrating force ifequilibrium has in fact been reached
before theshock.
The real payoff of subject i, pi , is given by
(2) p i 5 p i ~P i , P# 2i , M! i 5 1, ... , n
where Pi denotes i’s nominal price, P# 2i repre-sents the
average price of the other n 2 1 groupmembers while M denotes a
nominal shockvariable (money supply). The nominal payoff ofsubject
i is given by P# 2ipi. In total, we havefour treatment conditions
and the payoff func-tions (2) are the same in all conditions. The
fourconditions differ along two dimensions (seeTable 1). The first
dimension concerns the fram-ing of the situation, i.e., whether
payoffs arerepresented in real or in nominal terms. In thereal
treatments, denoted by RC and RH, subjectsreceived the payoff
information in real termswhile in the nominal treatments, denoted
by NCand NH, payoffs were represented in nominalterms. Thus, to
compute their real payoffs in the
TABLE 1—TREATMENT CONDITIONS
Payoffs in Real Terms Payoffs in Nominal Terms
Computerized opponents Real treatment with computerized
opponents(RC): 22 groups with 1 human and n 21 computerized players
in each group
Nominal treatment with computerizedopponents (NC): 24 groups
with 1human and n 2 1 computerizedplayers in each group
Human opponents Real treatment with human opponents (RH):10
groups with n human players in eachgroup
Nominal treatment with human opponents(NH): 11 groups with n
human playersin each group
1243VOL. 91 NO. 5 FEHR AND TYRAN: DOES MONEY ILLUSION
MATTER?
-
nominal treatments subjects had to divide theirnominal payoffs
P# 2ipi by P# 2i.
The second dimension concerns the factwhether our experimental
subjects face n 2 1preprogrammed computerized players orwhether
they face n 2 1 other human subjects.The crucial point here is that
in the computer-ized condition where one human subject facesn 2 1
preprogrammed computers, the subject isinformed about the aggregate
response rule ofthe computers in advance. The response rule ofthe
computers is given by the best replies of thecomputers [based on
the computers’ payofffunctions (2)]. Therefore, there is no
strategicuncertainty and, hence, no need to form expec-tations
about the behavior of the other players inthis condition. Moreover,
since the computersplay best replies, their behavior rules out
anymoney illusion or any other form of irrational-ity. In contrast,
in the condition with humanopponents each subject faces the task of
form-ing expectations about the other players’ pricechoices. This
necessarily also involves a guessabout the extent to which other
players areaffected by money illusion.
The conditions with computerized players es-sentially boil down
to individual decision-making experiments in which human
subjectscan maximize their money earnings by playingoptimally
against the known aggregate best re-ply of the n 2 1 computerized
players. Notethat in the computerized conditions the
indirecteffects of money illusion, which operate via
theexpectations that other players are affected bymoney illusion,
can play no role because thecomputers play best reply. These
conditions,therefore, allow us to examine to what extentmoney
illusion has direct effects on nominalinertia, i.e., to what extent
it simply causesindividual optimization mistakes. In the
condi-tions with human opponents the indirect effectsof money
illusion can, in addition, also play arole.
The experimental design in Table 1 allows toisolate various
potentially important determi-nants of nominal inertia. In the RC,
moneyillusion is ruled out at the individual and, hence,also the
aggregate level. Therefore, if we ob-serve in the RC a slow
adjustment of the nom-inal price chosen by the human subject after
theshock, money illusion cannot be the source ofthis nominal
inertia. Thus, with the help of the
RC we can test the hypothesis whether there areindividual-level
irrationalities other than moneyillusion.
In the NC, in contrast, money illusion canaffect the behavior of
individuals because as apart of the individual optimization problem
hu-man subjects have to correctly deflate nominalpayoffs at the
various (Pi , P# 2i) combinations.Hence, by comparing the
post-shock prices ofhuman subjects in the RC and the NC we
canobserve whether there exists money illusion atthe individual
level.
In the RH, as in the RC, individual-levelirrationality other
than money illusion can playa role. However, in the RH the
adjustment to thenew post-shock equilibrium is not just an
indi-vidual optimization problem for the human sub-jects. In the
RH, adjustment to the newequilibrium also involves the solution of
a com-plex coordination problem.3 It cannot be takenfor granted
that subjects instantaneously suc-ceed to act according to the new
post-shockequilibrium. A plausible reason for this is thatthe
complexity of subjects’ task is greater in theRH compared to the
RC.
The RH and the RC are used to examineto what extent
individual-level irrationalities,other than money illusion,
together with thecoordination problem contribute to nominal
in-ertia. The difference in price adjustment be-tween the RH and
the RC measures the impactof the coordination problem plus the
impact ofthe interaction between the coordination prob-lem and the
individual irrationalities that are notrelated to money illusion.
Interaction effectsoccur when these individual irrationalities
causeslow adjustment by some subjects after theshock which—due to
strategic complementar-ity—induces the other subjects to adjust
slowly,too. A particularly interesting case arises if wefind no
nominal inertia in the RC while in theRH nominal inertia prevails.
In this case all ofthe nominal inertia in the RH can be
attributedto the coordination problem because individual
3 There is an important literature on coordination prob-lems in
macroeconomic models (see, e.g., Cooper, 1999)that is based on the
existence of multiple equilibria. We usethe term “coordination
problem” in a different way because,even in the case of a unique
equilibrium, subjects face acoordination problem: Nash-equilibrium
play presupposesthat subjects have coordinated (equilibrium)
expectations.
1244 THE AMERICAN ECONOMIC REVIEW DECEMBER 2001
-
irrationalities (other than money illusion) areabsent.
In the NH, subjects face the same coordina-tion problem as in
the RH. We are, however,particularly interested in the impact of
addingthe nominal frame to this coordination problem,i.e., in a
comparison of the NH and the RH. Thiscomparison allows us to
isolate the total effectsof money illusion in an environment where
sub-jects face a coordination problem. The totaleffects of money
illusion in this environmentconsist of the direct effects of
individual-levelmoney illusion as exhibited in the NC plus
theindirect “multiplier” effects of individual-levelillusion. These
“multiplier” effects may arisebecause in our setting with human
opponents,subjects with money illusion can also affect
theexpectations and thus the behavior of the sub-jects without
money illusion. If, for example,some subjects exhibit money
illusion by taking(variations in) the nominal payoff as a proxy
for(variations in) the real payoff, adjustment toequilibrium in the
NH may be slower than in theRH. The reason is that after a negative
shock,adjustment requires a decrease in nominal prices.By
definition, a decrease in nominal prices isassociated with a
decrease in nominal payoffnumbers in the NH. Therefore, subjects
whoexhibit the above form of money illusion mis-takenly believe
that real payoffs decrease withlower nominal prices. Thus, they
prefer to stayat higher nominal prices, which may have adirect
adjustment-reducing effect. Moreover, ifsome subjects believe that
others suffer fromthis form of money illusion, they have an
in-centive to slow down adjustment, too. In theRH, in contrast,
this effect cannot occur becausepayoffs are represented in real
terms. In the RH,it is, therefore, completely transparent that
gen-eral price reductions are not associated withlower real
payoffs.
If the deviation from the post-shock equilib-rium is larger and
lasts longer in the NH com-pared to the RH, we have support for
thehypothesis that money illusion contributes tonominal inertia.
If, in addition, the price expec-tations are more sticky in the NH
than in the RHwe have an indication for indirect effects be-cause
the indirect effects become effective viasticky expectations.
However, our design alsoenables us to isolate the indirect effects
ofmoney illusion at the behavioral level. The dif-
ference in the deviations of the post-shock av-erage prices from
equilibrium between the NCand the RC, DPNC 2 DPRC, measures
theaggregate impact of individual-level money il-lusion on nominal
inertia. Note that DPNC 2DPRC 5 PNC 2 PRC because the
equilibriumprice is identical across conditions. The differ-ence
between the NH and the RH, DPNH 2DPRH 5 PNH 2 PRH, measures the
total ef-fects of money illusion which consist of
theindividual-level effects plus the indirect effects.Thus, if
there are no indirect effects of moneyillusion the total effects
must be equal to theindividual-level effects: PNH 2 PRH 5 PNC 2PRC.
If there are, however, indirect effects weshould observe that PNH 2
PRH . PNC 2PRC.4
B. General Properties of the Payoff Functions
Before we proceed to the specific numericalparameters of our
experiment, it is useful toprovide a general description of the
payoff func-tions (2). They have the following properties:
(i) They are homogeneous of degree zero in Pi ,P# 2i , and
M.
(ii) The best reply is (weakly) increasing in P# 2i.
In addition, our functional specification5 ofequation (2)
implies that the equilibrium
(iii) is unique for every M,(iv) is the only Pareto-efficient
point in payoff
space, and(v) can be found by iterated elimination of
weakly dominated strategies.
4 Note also that if there are no individual irrationalitiesother
than money illusion, DPRC 5 0, and the conditionfor the existence
of indirect effects, DPNH 2 DPRH .DPNC 2 DPRC, can then be written
as DPNH . DPRH 1DPNC. This means that if the deviation from
equilibriumin the NH condition is larger than the summed
devia-tions in the RH and the NC condition, indirect
effectsprevail.
5 The functional form is presented in the Appendix.Readers who
are interested in the full set of instructionsshould consult the
Appendix in Fehr and Tyran (2000),which can be downloaded from
http://www.iew.unizh.ch/wp/iewwp045.pdf. The instructions are also
available fromthe authors upon request.
1245VOL. 91 NO. 5 FEHR AND TYRAN: DOES MONEY ILLUSION
MATTER?
-
Note that the real payoff pi does not depend onthe average price
P# of all group members but onP# 2i. This feature makes it
particularly easy toplay a best reply for a given expectation
aboutthe other players’ average price. If we made pidependent on P#
, so that Pi affects P# , it wouldhave been much more difficult for
i to computethe best reply (see also below). It is also worth-while
to point out that the nominal payoff foreach subject i is given by
P# 2ipi and not byP# pi. This makes the computation of the
realpayoffs from a given nominal payoff much eas-ier because the
deflator is independent of one’sown price choice.
Properties (i) and (iii) above were imple-mented because our
analysis focuses on theimpact of money illusion on the
adjustmentprocess of an economy with a unique money-neutral
equilibrium P*i , i 5 1, ... , n. To seethat properties (i) and
(iii) imply neutrality, notethat a change in M from M0 to lM0
leaves real pay-offs unaffected if prices change to lPi andlP# 2i.
Moreover, if P9i , i 5 1, ... , n, is a bestreply to P# 2i at M0,
lP9i also is a best replyto lP# 2i at lM0. Thus, lP*i for all i is
thepost-shock equilibrium.
Property (ii) captures strategic complementa-rity and was
implemented for the reasons givenin Section I, subsection B. In our
pilot experi-ments we initially implemented a price-settinggame
with monopolistic competition. However,it turned out that subjects
quickly realized thatunder monopolistic competition
cooperativegains can be achieved by out-of-equilibriumbehavior.
Therefore, both in the nominal as wellas in the real frame,
subjects systematicallytried to achieve real payoff gains through
out-of-equilibrium behavior. Only towards the endof each phase
these attempts vanished. Thus, inthe pre- as well as in the
post-shock phase of ourpilot experiments, adjustment towards
equilib-rium was strongly retarded by attempts to co-operate. To
remove this confound with the othersources of nominal inertia we
chose payofffunctions that ensured that the equilibrium wasthe
unique Pareto-efficient point in the wholepayoff space [property
(iv)].
Finally, property (v) means that there is amethod for finding
the equilibrium that worksexactly in the same way in the real as
well as inthe nominal frame. Note that the framing ofpayoffs has no
impact at all on whether a par-
ticular strategy is (weakly) dominated. In thereal frame a
(weakly) dominated strategy Pi has(weakly) smaller real payoff
numbers at anylevel of P# 2i. In the nominal frame a
(weakly)dominated strategy Pi has (weakly) smallernominal payoff
numbers at any level of P# 2i.Thus, to eliminate (weakly) dominated
strate-gies in either frame, subjects only need to elim-inate those
strategies that have (weakly) smaller(real or nominal) payoff
numbers at any givenlevel of P# 2i. Since, in the condition with
humanopponents, the best-reply function and, hence,the number of
(weakly) dominated strategies isexactly the same under the real and
the nominalframe, there is, in the absence of money illu-sion, no
reason why adjustment should differacross the RH and the NH.
C. Experimental Procedures and Parameters
All major experimental parameters are sum-marized in Table 2.
The experiment was con-ducted in a computerized laboratory with
agroup size of n 5 4. The group composition didnot change
throughout the whole experiment,i.e., for 2T periods. In each group
there weretwo types of subjects: Subjects of type x andsubjects of
type y. Payoff functions differedamong the types. This difference
implied thatx-types had to choose a relatively low price
inequilibrium while y-types had to choose a rel-atively high price
(see Table 2 for details).There is no particular reason for our
choice ofthe group size because there are no strong con-jectures
about the net effects of a differentgroup size. On the one hand, a
larger group sizeenhances the chances that there are
individualswith money illusion in a group. On the otherhand, the
relative impact of an individual onaverage prices becomes smaller.
With regard tothe heterogeneity of the players’ payoff func-tions,
the case of four different payoff functionswould be the most
realistic but also the mostcomplicated case. Therefore, we went for
anintermediate solution with only two types ofplayers.6
In the pre-shock phase of each treatment the
6 The payoff functions of the two types were the same upto a
parallel shift. Except for P*k and P# *k all parameters of
thepayoff function specified in the Appendix are the same forboth
types.
1246 THE AMERICAN ECONOMIC REVIEW DECEMBER 2001
-
money supply was given by M0 5 42 while inthe post-shock phase
it was given by M1 5M0/3 5 14. In the pre-shock equilibrium
theaverage price over all n group members is givenby P# *0 5 18
while in the post-shock equilib-rium it is P# *1 5 6. In the
treatments with humanopponents both the pre- and the
post-shockphase consists of T 5 20 periods while in thetreatments
with computerized opponents T 510. The reason for this difference
was that weexpected that adjustment would take longer inthe
presence of a coordination problem. For thepurpose of comparing
post-shock nominal iner-tia across treatments it is crucial that
the re-quired price adjustment, i.e., the differencebetween actual
nominal prices in the final pre-shock period and the new post-shock
equilib-rium price, is roughly the same. To ensurecomparable
adjustment requirements acrosstreatments we gave players more time
to reachthe equilibrium in the treatments with a coordi-nation
problem.
In each decision period subjects had tochoose an integer Pi [
{1, 2, ... , 30}. Inaddition, they had to provide an
expectationabout P# 2i which we denote by P# 2i
e . Finally,subjects indicated their confidence about their
expectation P# 2ie by choosing an integer from 1
to 6 where 1 indicated that the subject is “not atall confident”
while 6 indicated that he or she is“absolutely confident.”7 This
measure of confi-dence can be interpreted as an indicator of
sub-jects’ perceived uncertainty about the otherplayers’ choices.
Note that this uncertainty is aninevitable component of the
coordination prob-lem that subjects face in the condition withhuman
opponents. At the end of each periodeach subject was informed about
the actual re-alization of P# 2i and the actual real payoff pi ona
so-called outcome screen. In addition, theoutcome screen provided
information about thesubject’s past choices of Pi , past
realizations ofP# 2i , and past real payoffs pi.
Subjects received the payoff information inmatrix form.8 The
payoff matrix shows the realand the nominal payoff, respectively,
for each
7 The detailed meaning attached to the numbers is: 1 5not at all
confident; 2 5 not much confidence; 3 5 not quiteconfident; 4 5
quite confident; 5 5 very confident; 6 5absolutely confident.
8 Appendix C of Fehr and Tyran (2000) contains thepayoff
matrices of x- and y-types for all treatment condi-tions. See also
footnote 5.
TABLE 2—EXPERIMENTAL DESIGN
Panel A: All PeriodsRepresentation of payoffs in the nominal
frame P# 2ipiRepresentation of payoffs in the real frame piGroup
size n 5 4Information feedback in period t P# 2i , piReal
equilibrium payoff 40Choice variable Pi [ {1, 2, ... , 30}Length of
pre- and post-shock phase in treatment with computerized opponents
T 5 10Length of pre- and post-shock phase in treatment with human
opponents T 5 20
Panel B: Pre-Shock ValuesMoney supply M0 42Average equilibrium
price P# * and average equilibrium expectation for the whole group
18Equilibrium price for type x 9Equilibrium expectation P# 2i
e for type x 21Equilibrium price for type y 27Equilibrium
expectation P# 2i
e for type y 15
Panel C: Post-Shock ValuesMoney supply M1 14Average equilibrium
price P# * and average equilibrium expectation for the whole group
6Equilibrium price for type x 3Equilibrium expectation P# 2i
e for type x 7Equilibrium price for type y 9Equilibrium
expectation P# 2i
e for type y 5
1247VOL. 91 NO. 5 FEHR AND TYRAN: DOES MONEY ILLUSION
MATTER?
-
feasible integer combination of (Pi , P# 2i). Toinform subjects
about the payoffs of the othertype, each subject also received the
payoff ma-trix of the other type. This information condi-tion was
common knowledge. The presentationof payoffs in the form of a
matrix made itparticularly easy to find the best reply for anygiven
P# 2i: The subject just had to look for thehighest real or nominal
payoff in the columnassociated with P# 2i.
9 In fact, one of the firstthings most subjects did after we
distributed theinstructions was to mark the best replies in
thepayoff tables.
After subjects had read the instructions, butbefore the start of
the experiment, each subjecthad to solve a series of exercises (see
the Appen-dix in Fehr and Tyran, 2000). These exercisesinvolved the
computation of their own payoff andthe payoff of their opponents
for given hypothet-ical strategy profiles. In the nominal
treatments, inparticular, subjects had to compute the real pay-offs
from their nominal payoff tables for givenhypothetical strategy
profiles. The subjects knewthat we did not start the experiment
until everyparticipant in a session had solved all
exercisessuccessfully. All subjects were in fact able to solvethe
exercises. By this training procedure wewanted to rule out that
subjects do not know howto properly deflate nominal values. It is
quitelikely that this procedure diminished the amountof
individual-level money illusion in our experi-ment. It was
motivated by the question whether asmall amount of individual-level
money illusionwill cause long-lasting nominal inertia in the
NHbecause of the indirect effects of money illusion.Obviously, the
case for the relevance of moneyillusion is stronger if we observe
large indirecteffects.
In the treatments with computerized oppo-nents, subjects
received the same instructionsand payoff tables as in the
treatments with hu-man opponents. In addition, subjects were
in-formed that the decisions of the other threeplayers in the group
would be made by prepro-grammed computers. Subjects received an
in-
formation sheet that informed them about theP# 2i response of
the three computers to eachprice choice Pi [ {1, 2, ... , 30}.
Fifty percentof the human subjects in these conditions wereendowed
with the payoff function of anx-player, the other 50 percent had
the payofffunction of a y-player.
At the end of the final pre-shock period thenominal shock was
implemented in the follow-ing way: Subjects were publicly informed
thatx- and y-types received new payoff tables.These tables were
based on M1 5 M0/3. Againeach type received the payoff table for
his ownand the other type. Subjects kept the pre-shocktables and
were encouraged to compare the pre-and post-shock tables. They were
told that, ex-cept for payoff tables, everything else wouldremain
unchanged. They were given enoughtime to study the new payoff
tables and tochoose Pi for the first post-shock period.
10 Thisprocedure ensured that in the first post-shockperiod
subjects faced an exogenous and fullyanticipated negative nominal
shock. At the be-ginning of this period it was also commonknowledge
that the experiment would last foranother T periods.
Before we proceed to the experimental re-sults, it needs to be
emphasized that in a givenphase the number of dominated price
choices isidentical across all treatments. It is, however,not
identical between the pre- and the post-shock phase. Since the
money supply is lower inthe post-shock phase the number of
dominatedstrategies is also lower in this phase. Note thatthe
smaller number of dominated strategies in thepost-shock phase is an
inevitable result of thefact that the money supply is reduced while
thenominal strategy space and the nominal ac-counting unit is kept
constant.11 Due to the
9 If a subject is uncertain about the true value of P# 2i ,
thecalculation of the best reply requires, of course, to take
intoaccount the subjective distribution of P# 2i and not onlythe
expectation of P# 2i. However, for simplicity, in thefollowing we
will use the term “best reply” in the sense ofa best reply to the
expectation of P# 2i.
10 Subjects were told that they had ten minutes to studythe new
payoff tables and, in addition, three minutes tomake a decision for
the first post-shock period. Yet, almostall subjects made their
decision well before the 13 minuteshad elapsed. In the subsequent
periods subjects also rarelyexhausted their time limits.
11 A change in the nominal price in the post-shock phase(i.e.,
at M0/3) by one unit has the same real effects as achange in the
nominal price by three units in the pre-shockphase (i.e., at M0).
This means that if a nominal price isstrictly dominated in the
post-shock phase there will, ingeneral, be three nominal prices
that are strictly dominatedin the pre-shock phase.
1248 THE AMERICAN ECONOMIC REVIEW DECEMBER 2001
-
differences in the number of dominated strat-egies a comparison
of the adjustment speedacross phases must take this difference
intoaccount. The higher number of dominatedstrategies in the
pre-shock phase means, inparticular, that the indirect effects of
moneyillusion are likely to be smaller in this phase.This is so
because, if a strategy is dominated,it is optimal to not play this
strategy irrespec-tive of the expectations about other
players’behavior. Thus, expectations about otherplayers’ money
illusion necessarily have lessimpact and, as a consequence, one
wouldexpect a quicker adjustment towards equilib-rium in the
pre-shock phase. Note also thatthe different number of dominated
strategiesacross phases is not a problem for the mainpurpose of our
research. We are not interestedin comparing adjustment speed across
phasesbut across treatments in the post-shock phase.For our
purposes the crucial point is that inthe post-shock phase the
number of domi-nated strategies is identical across
treatmentsbecause the only difference in the payoff ta-bles
concerns the framing of the payoffs.
III. Results
In total, 130 subjects participated in theexperiments described
in Table 1.12 Twenty-two subjects participated in the real
treatmentwith computerized opponents (RC) and 24subjects in the
nominal treatment with com-puterized opponents (NC). Eleven groups
offour human subjects participated in the nom-inal treatment with
human opponents (NH)and ten groups in the real treatment withhuman
opponents (RH). No subject partici-pated in more than one
treatment. Subjectswere undergraduate students from
differentdisciplines at the University of Zurich, Swit-zerland.
They were paid a show-up fee ofCHF 15 (approx. $12 at that time)
and theirtotal earnings from the experiment were onaverage CHF 35
(approx. $28) (including theshow-up fee). On average, an
experimentalsession lasted 90 minutes.
A. Nominal Price Adjustment as anIndividual Optimization
Problem
In this section, we address the questionwhether individual-level
money illusion andother individual-level irrationality contribute
tonominal inertia. Therefore, our discussion isconstrained to the
RC and the NC, where ad-justment to the post-shock equilibrium is a
pureindividual optimization problem. Our first mainresult is that
in the RC all subjects instanta-neously adjust to the new
post-shock equilib-rium, i.e., nominal inertia is completely
absent.Support for this claim is provided by column 1of Table 3 and
by Figure 1. Both the table andthe figure show the pre- and
post-shock path ofthe average price of all human subjects in theRC.
What is remarkable here is that, except fora few periods, the
average price is exactly equalto the equilibrium price of P# *0 5
18 in the pre-and P# *1 5 6 in the post-shock period. More-over, it
is not just the average that coincideswith equilibrium. In most
periods literally allsubjects play the equilibrium. This result
con-trasts with what we observe in the nominalframe. In the NC
there is a small amount ofnominal inertia since some subjects do
not fullyadjust prices to the new post-shock equilibrium.This claim
is supported by Table 3 (column 2)and Figure 1. Both the table and
the figure showthat the evolution of average prices is, in
gen-eral, more volatile relative to the RC. This sug-gests that at
least some subjects in the NC haveproblems in finding the optimal
solution to theirmaximization problem. Moreover, while in theRC all
subjects instantaneously adjust theirprices fully to the post-shock
equilibrium, in theNC only 80 percent of the subjects do so.
Therest of the subjects choose prices above theequilibrium so that
in the first post-shock periodthe average price is by 2.1 units too
high.Throughout the whole post-shock phase the NCmost of the time
is close but never exactly inequilibrium which contrasts again with
the RCwhere after the second post-shock period allsubjects are
exactly in equilibrium almost all ofthe time.
These differences in post-shock adjustmentalso give rise to
differences in the real incomelosses across RC and NC. Nominal
inertia in theNC causes small but nonnegligible real incomelosses
in the post-shock phase. In contrast, in
12 In follow-up experiments with a positive moneyshock,
described in detail in Section IV, another 96
subjectsparticipated.
1249VOL. 91 NO. 5 FEHR AND TYRAN: DOES MONEY ILLUSION
MATTER?
-
the RC there are no or only extremely small realincomes losses
in the post-shock phase. To ver-ify this claim we calculate by how
much actualreal income of player i, pi , falls short of realincome
in equilibrium p*. For this purpose wehave computed «it [ (p* 2
pit)/p* for allplayers in each period t. «it is a measure of
theincome loss relative to the equilibrium payoff asa percentage of
the equilibrium payoff. Since
the equilibrium is efficient it is also a measureof the
efficiency loss. Columns 5 and 6 of Table3 present the evolution of
the average value of«it over all players in the RC and in the NC.
Thetwo columns indicate that after the shock theaverage efficiency
loss is most of the time zeroin the RC and always lower than in the
NC.
Taken together, the results of the treatmentswith computerized
opponents indicate that there
TABLE 3—EVOLUTION OF PRICES AND EFFICIENCY LOSSES OVER TIME
Period
Average Price Average Efficiency Loss (Percent)
Computerizedopponents Human opponents
Computerizedopponents Human opponents
Real(RC)
Nominal(NC)
Real(RH)
Nominal(NH)
Real(RC)
Nominal(NC)
Real(RH)
Nominal(NH)
220 17.6 18.5 14.4 19.0219 18.2 19.3 21.5 14.6218 17.8 19.1 14.1
10.2217 17.7 19.4 9.5 11.7216 17.9 19.2 8.8 6.8215 18.3 19.1 10.8
13.2214 17.6 18.2 8.0 9.9213 17.9 18.6 8.2 4.2212 17.9 18.7 6.3
3.1211 17.6 18.3 5.5 7.5210 17.9 15.2 17.8 18.4 1.0 16.4 9.4 3.429
18.1 17.0 17.5 18.2 0.5 12.6 3.6 1.628 17.8 17.2 17.6 19.0 1.6 9.0
3.3 6.027 18.0 18.0 17.7 18.3 0.5 3.0 2.4 1.826 17.6 17.2 17.6 18.2
2.4 10.4 10.9 1.325 18.0 17.7 18.1 18.3 0.3 5.4 7.0 2.724 18.0 18.1
18.1 18.4 0.0 3.5 7.3 2.523 17.8 16.1 17.6 18.6 1.3 12.6 3.7 2.822
18.4 18.3 17.9 18.2 2.3 1.9 2.2 0.721 18.0 17.0 18.0 18.2 0.0 5.3
0.9 0.9
1 6.0 8.1 9.1 13.1 0.0 10.4 51.8 65.12 7.0 7.4 7.7 12.9 3.6 8.2
20.0 47.53 6.0 6.8 7.4 11.4 0.0 4.4 15.0 34.84 6.0 6.4 6.9 10.4 0.6
6.5 9.1 27.45 6.0 6.9 7.0 9.9 0.0 8.0 14.8 17.46 6.0 6.8 6.6 10.2
0.0 15.6 7.7 15.97 6.0 7.5 6.3 9.7 0.0 9.3 4.5 16.48 6.0 6.8 6.4
9.1 0.0 15.5 4.6 10.79 6.0 6.5 6.3 8.7 0.0 4.3 3.8 9.5
10 5.9 6.5 6.8 8.6 1.6 3.8 11.0 13.811 6.1 8.1 4.6 8.212 6.2 7.6
3.3 6.413 6.2 7.2 2.1 6.214 6.2 6.9 2.8 4.615 6.1 6.7 2.6 2.616 6.1
7.3 2.1 9.617 6.0 6.8 0.9 5.218 6.1 7.2 1.8 14.219 6.1 7.5 1.4
12.520 6.2 7.0 3.0 2.4
1250 THE AMERICAN ECONOMIC REVIEW DECEMBER 2001
-
is a small amount of money illusion at theindividual level but
beyond that there is noindividual irrationality. The small amount
ofindividual-level money illusion is suggested bythe small price
differences between the NC andthe RC after the shock. The absence
of otherforms of individual irrationality is suggested bythe
perfect adjustment to the shock and thegenerally high incidence of
equilibrium play inthe RC.
B. Nominal Price Adjustment as aCoordination Problem
The fact that in the RC the adjustment to thepost-shock
equilibrium is perfect makes the in-terpretation of the deviation
of prices from thepost-shock equilibrium in the RH
particularlyeasy. It means that the whole deviation is due tothe
fact that subjects in the RH face a relativelycomplex coordination
problem. The major factsabout price adjustment in the RH are
displayedin Table 3 and Figure 1. Column 3 of Table 3
shows that in the first post-shock period averageprices in the
RH are 3.1 units above the averageequilibrium price of P# *1 5 6.
This deviationquickly decreases to 1.4 units in period 3 andafter
period 4 the deviation is never larger thanone unit. This pattern
of average behavior is notan artifact of aggregation but is also
revealed atthe level of individual choices. In the final pre-shock
period 93 percent of the subjects in theRH play exactly their
equilibrium strategies. Inthe first post-shock period only 35
percent of thesubjects play the new equilibrium and 23 per-cent of
the subjects are only one or two priceunits above the equilibrium.
The other 42 per-cent are more than two units above the
equilib-rium. Yet, after only three periods thedistribution of
individual price choices hasmoved much closer to the equilibrium.
In period4, 45 percent of all subjects play exactly theequilibrium,
48 percent are one or two unitsabove and only 7 percent are more
than two unitsabove the equilibrium. This post-shock evolu-tion of
prices indicates that the coordination
FIGURE 1. EVOLUTION OF AVERAGE PRICES
1251VOL. 91 NO. 5 FEHR AND TYRAN: DOES MONEY ILLUSION
MATTER?
-
problem initially causes considerable nominalinertia but that
after a few periods this effect israther small because prices are
again close tothe equilibrium.
Our description of the pattern of nominalinertia in the RH is
also supported by formalstatistical tests. To check how long
averagegroup prices in the RH and the NH deviatesignificantly from
the equilibrium we ran thefollowing regression for the post-shock
phase:
(3) P# it 2 P# *1 5 Ot 5 1
19
a t d t 1 Ot 5 1
20
b t ~1 2 d t !
where P# it denotes the average price of group iin period t. dt
5 1 if the price observation inperiod t comes from the RH. The
coefficients atmeasure the deviation from equilibrium in theRH
while the coefficients bt measure the devi-ation in the NH.13 The
results of regression (3)are summarized in Table 4. The table
showsthat, at the 5-percent level, average prices in theRH deviate
significantly from the equilibriumfor two periods. Yet, from period
3 onwards, thehypothesis that average prices are in equilib-rium
can no longer be rejected.14
To what extent is nominal inertia in the RHassociated with real
income losses? Column 7of Table 3 indicates that in the first
post-shockperiod the real income loss resulting from
dis-equilibrium is quite considerable (52 percent).Yet, due to the
relatively quick adjustment ofnominal prices after this period, the
real incomeloss declines substantially and after the
fifthpost-shock period it is—except for period 10—always below 10
percent. In the final periods thereal income loss is always rather
small whichreflects the high incidence of equilibrium play.
The key difference between the RC and theRH is the presence of a
relatively complexcoordination problem in the RH. If subjects
perceive coordination as a difficult problem thisshould be
reflected in subjects’ confidence inP# 2i
e . In the first few pre-shock periods, sub-jects’ average
confidence is at a level of 4 whichmeans that they are, on average
“quite confi-dent.” The high frequency of equilibrium playbefore
the shock then causes a general increasein the confidence level. In
the last five pre-shockperiods, subjects exhibit, on average, a
confi-dence level between 5 and 5.5. This means thatmost subjects
are “very confident” (5 level 5)or even “absolutely confident” (5
level 6) thatthey have correct expectations. The
anticipatednegative money shock causes, however, a con-siderable
decrease in subjects’ confidence. Inthe first post-shock period,
subjects are on av-erage “not quite confident” (level 3) or
“quiteconfident” (level 4) that their expectations willbe correct.
It takes about eight periods untilpre-shock confidence levels are
again estab-lished. This indicates that the money shock in-deed
causes a considerable coordinationproblem for the subjects.
Taken together, the evidence suggests that
13 To prevent linear dependence among the set of regres-sors, we
included no dummy variable for period 20 ofthe RH.
14 We also examined the null hypothesis that prices inthe RH
differ from prices in the RC by means of nonpara-metric tests with
individual data. The null hypothesis ofequal price distributions
and of equal average prices can berejected for the first four
post-shock periods at the 10-percent level (Kolmogorov-Smirnov Test
and Mann-Whitney Test).
TABLE 4—DEVIATION FROM POST-SHOCK EQUILIBRIUM INTREATMENTS WITH
HUMAN OPPONENTS
Post-ShockPeriod
Real Treatment withHuman Opponents(RH) Coefficient at
Nominal Treatmentwith Human Opponents
(NH) Coefficient bt
1 3.10*** 7.14***2 1.68** 6.86***3 1.43 5.43***4 0.90 4.41***5
1.00 3.86***6 0.55 4.18***7 0.25 3.77***8 0.35 3.05***9 0.25
2.70***
10 0.83 2.59***11 0.13 2.05***12 0.23 1.61**13 0.18 1.1814 0.18
0.8915 0.10 0.7016 0.13 1.2517 0.03 0.8018 0.13 1.2019 0.05 1.4520
— 0.95
Notes: P# it 2 P# *1 5 ¥t 5 119 atdt 1 ¥t 5 1
20 bt(1 2 dt).dt 5 1 if price observation in period t is from
RH.***Denotes significance at the 1-percent level.**Denotes
significance at the 5-percent level.
1252 THE AMERICAN ECONOMIC REVIEW DECEMBER 2001
-
the introduction of a coordination problem inthe real treatment
causes initially a nonnegli-gible amount of nominal inertia that is
associ-ated with considerable real effects. Yet, nominalinertia
vanishes relatively quickly so that al-ready after a few periods
prices are quite closeto the equilibrium.
C. Coordination in the Presenceof Money Illusion
Nominal inertia in the RH has nothing to dowith money illusion
but is caused by the prob-lem to coordinate expectations and
actions onthe new equilibrium. From the comparison be-tween the RC
and the NC we already know thatindividual-level money illusion has
a small pos-itive effect on nominal inertia. In the NH asmall
amount of individual-level money illusionmay, however, cause
important indirect effects.These indirect effects can arise because
thepresence of individual-level money illusion islikely to affect
subjects’ expectations, which inturn affect their behavior. If
money illusionindeed causes such indirect effects we shouldobserve
that the introduction of the nominalframe has a larger effect in
the setting withhuman players than in the setting with
comput-erized players. We should, in addition, also ob-serve that
in the setting with human players thenominal frame gives rise to an
increase in thestickiness of subjects’ price expectations.
Figure 1 and Table 3 (columns 3 and 4)provide the relevant
information regarding theimpact of the nominal frame. They show
thatnominal prices are indeed much stickier in theNH compared to
the RH. In the final pre-shockperiod the overwhelming majority of
the sub-jects play exactly the equilibrium both in the RH(93
percent) and the NH (80 percent). There-fore, average prices are
very close to thepre-shock equilibrium P# *0 5 18. In the
firstpost-shock period, however, only 11.5 percentof all subjects
in the NH play exactly the equi-librium and 73 percent of the
subjects are threeor more price units above the equilibrium.
Incontrast, in the RH, 35 percent play exactly theequilibrium and,
in addition, 23 percent are onlytwo or less price units above the
new equilib-rium. These treatment differences in
individualadjustment behavior also give rise to large dif-ferences
in the average price level. In the first
post-shock period the average price in the NH is7.1 units above
the equilibrium while in the RHthe deviation is only 3.1 units (see
Table 3). Ittakes eight periods in the NH until the deviationof
average prices from equilibrium decreases to3.1 units. These large
differences in price ad-justment speed are also confirmed by
formalstatistical tests. Table 4 reveals that in the NHthe
hypothesis of equilibrium play can be re-jected at the 5-percent
level for the first 12post-shock periods while in the RH it can
onlybe rejected for two periods. Similar resultsemerge when we
examine the null hypothesis ofequal average prices in the NH and
the RH bymeans of a nonparametric Mann-Whitney Testwith individual
data. For the first nine post-shock periods the null hypothesis can
be re-jected already at the 2-percent level. For thenext three
post-shock periods it can be rejectedat the 10-percent level.
To what extent is nominal inertia in the NHassociated with real
income losses? Column 8 ofTable 3 indicates that shortly before the
shock,subjects in the NH achieve almost full efficiency.The
monetary shock leads, however, to a substan-tial real income loss.
In the first period after theshock the average income loss is 65
percent andduring the first ten post-shock periods the loss isnever
below 9.5 percent. Note also that throughoutthe whole post-shock
period the income loss is ingeneral much higher in the NH than in
the RHwhich is a consequence of the much stickier pricesin the NH.
For example, in the first ten post-shockperiods of the NH, the
aggregate real income lossis roughly twice as large as the loss in
the RH. Intotal, groups in the NH lose 26 percent of thepotential
payoff in the first ten post-shock periods.In the RH, the
respective losses are slightly lessthan 14 percent. Thus, the
evidence clearly indi-cates two results: (i) In the setting with
humanplayers the introduction of a nominal frame haslarge and
long-lasting effects on price stickiness.(ii) This increase in
price stickiness is associatedwith a considerable increase in the
real incomeloss caused by the anticipated money shock.
From Figure 1 and Table 3 we also can inferthat the nominal
frame causes much stickier priceswhen money illusion can have
indirect effects.Throughout the first ten post-shock periods
theadjustment difference in average prices betweenthe NH and the
RH, DPNH 2 DPRH 5 PNH 2PRH, is between two and 13 times larger than
the
1253VOL. 91 NO. 5 FEHR AND TYRAN: DOES MONEY ILLUSION
MATTER?
-
adjustment difference between the NC and theRC, DPNC 2 DPRC 5
PNC 2 PRC. It is, forexample, easy to infer from Table 3 that, in
thesecond post-shock period, the adjustment differ-ence between the
NH and the RH is 12.9 2 7.7 55.2 price units, while the difference
between theNC and the RC is only 7.4 2 7.0 5 0.4 units.Hence, in
this period the impact of the nominalframe is 13 times larger in
the setting with humanplayers compared to the setting with
computerizedplayers. In the tenth post-shock period the adjust-ment
difference is still 1.8 units in the setting withhuman players and
only 0.6 units in the settingwith computerized players. To
substantiate theindirect effects of money illusion we also
con-ducted period-wise t-tests for the null hypothesisthat the
adjustment difference between the NHand the RH is bigger than
between the NC and theRC. In five of the first ten post-shock
periods thedifference between the NH and the RH is signif-icantly
larger at the 5-percent level. In view of theconsiderable variance
across the four conditionsthis is quite remarkable.15 Thus, the
implementa-tion of the nominal frame has a much larger im-pact in
the setting where money illusion can alsohave indirect effects.
If money illusion has indirect effects weshould also observe
that expectations are stick-ier in the NH compared to the RH.
Figure2 shows the evolution of the average price ex-pectations over
time in both treatments. Thefigure shows that in the last few
pre-shock pe-riods, expectations are in equilibrium in
bothtreatments. In the post-shock phase there are,however, striking
differences. While expecta-tions are very sticky in the NH they are
far lesssticky in the RH. To provide statistical evidencefor this,
we ran the same regression as in Table4 with expectations data. It
turns out that, in theNH, expectations differ significantly from
theequilibrium (at p , 0.05) for 13 periods whilein the RH they
differ only for three periods.Thus, there can be little doubt that
the nominalframe causes a large increase in the stickiness ofprice
expectations. The next question then is, towhat extent this
difference in expectationscauses differences in subjects’ price
choices. Or
put differently, to what extent did subjects playa best reply to
their expectations. The vast ma-jority of subjects in both
treatments indeedplayed best replies to P# 2i
e . During the first tenpost-shock periods 84 percent of the
subjects inthe RH choose exactly the payoff-maximizingprice in
response to P# 2i
e and the rest of thesubjects chooses prices that were close to
thebest reply. In the NH there are slightly fewersubjects (80
percent) who chose exact best re-plies during the first ten
post-shock periods.Yet, as in the RH, the deviations from the
exactbest reply were in general very small. The factthat most
subjects responded to P# 2i
e with apayoff-maximizing price choice suggests thatthe greater
stickiness of the expectations in theNH also caused a greater
stickiness of actualprices in the NH.
IV. Nominal Inertia after
a Positive Money Shock
A. The Relevance of a Positive Money Shock
Our results so far indicate that the directeffects of
individual-level money illusion arerelatively small. The
introduction of the nomi-nal frame in the setting with computerized
play-ers leads only to a small increase in nominalinertia. Nominal
inertia is much more pro-nounced, however, when money illusion
canalso affect players’ expectations and can, thus,also have
indirect effects. In the NH, subjects’expectations are much
stickier and, as a conse-quence, prices are much stickier. This
raises thequestion of why expectations are so sticky inthe NH
compared to the RH. We believe thatthe answer to this question can
be found in theexistence of subjects who take nominal payoffsas a
proxy for real payoffs. Subjects who applythis rule of thumb
mistakenly believe that if allplayers choose relatively high
prices, all willreap high real payoffs because they all reap
highnominal payoffs. They mistakenly believe thatthere are real
gains from jointly setting highprices. Such subjects will,
therefore, be reluc-tant to cut their nominal prices after the
negativemoney shock in the NH. Moreover, if the pres-ence of
subjects who are reluctant to cut pricesis anticipated by other
subjects, others will beinduced to cut their price insufficiently
also.
It is important to note that the above rule of
15 There are two subjects in the NC condition who couldwell be
classified as outliers. If we run the tests withoutthese two
subjects the indirect effects are highly significantin the first
nine post-shock periods.
1254 THE AMERICAN ECONOMIC REVIEW DECEMBER 2001
-
thumb cannot become effective in the RH. Inthe RH the numbers in
the payoff tables repre-sent real payoffs which makes it
completelytransparent that at high nominal prices real pay-offs are
not generally higher. This means thatthe presence of subjects who
take nominal pay-offs as a proxy for real payoffs causes no
reluc-tance to cut nominal prices after the negativeshock in the
RH. These differences between theNH and the RH in the reluctance to
cut nominalprices also provide a rationale for the muchstickier
price expectations in the NH.
Yet, if the above explanation for the stickierexpectations in
the NH is correct, we shouldalso observe that after a positive
money shock,prices and expectations adjust more quickly tothe
equilibrium than after a negative shock. Thisis so because after a
positive shock, adjustmenttowards equilibrium means adjustment
towardshigher prices and, hence, higher nominal pay-offs. A quicker
adjustment after a positiveshock, however, does not mean that money
il-lusion is absent when a positive shock occurs. It
only means that money illusion has a differentimpact after a
positive shock compared to anegative shock. After a negative shock
the ruleof thumb mentioned above causes a reluctanceto adjust
prices (downwards) while after a pos-itive shock it does not cause
a reluctance toadjust prices (upwards). Note further that whilethe
rule of thumb implies a quicker adjustmentto equilibrium after a
positive shock in the NH,the adjustment speed in the RH should not
differacross positive and negative shocks. The reasonis that the
rule of thumb cannot become opera-tive in the RH.
To test these implications of our explanationfor the much
stickier expectations in the NH weconducted additional experiments
with a posi-tive money shock. Forty-eight subjects (12groups)
participated in the RH and another 48subjects (12 groups)
participated in the NH withthe positive money shock. The easiest
way toimplement a positive shock would be a reversalin the sequence
of the money supply in ourprevious design. Unfortunately, this
approach is
FIGURE 2. EVOLUTION OF AVERAGE EXPECTATIONS
1255VOL. 91 NO. 5 FEHR AND TYRAN: DOES MONEY ILLUSION
MATTER?
-
not reasonable because the number of domi-nated strategies is
much larger in the pre-shockphase than in the post-shock phase.
Therefore,the indirect effects of money illusion can play amuch
smaller role in the pre-shock phase. Thefact that prices in the NH
adjust much morequickly to the equilibrium in the pre-shockphase
than in the post-shock phase (see Figure1) is consistent with this
argument. Therefore, ifwe just reversed the sequence of the
moneysupply, we would probably observe that adjust-ment is indeed
quicker after the positive shock.Yet, this increase in the
adjustment speed wouldnot count as evidence for our explanation of
thestickier expectations in the NH.
What is, therefore, needed, is an experi-mental design in which
the number of domi-nated strategies is roughly the same after
thenegative and after the positive shock. Ourparameterization of
the design with the posi-tive shock serves this purpose. Except
forthree aspects, all experimental details in thepositive-shock
design are identical to thenegative-shock design. In particular,
all fivefeatures of the payoff functions, as describedin Section
II, subsection B, are also present inthe positive-shock design. The
differences arethe following: (i) We did not implement
com-puterized players in the positive-shock designbecause the main
purpose of this design wasto observe whether the expectations of
humanplayers and, hence, also prices adjust morequickly to the
equilibrium after a positiveshock compared to the negative shock.
(ii) Inthe positive-shock design the pre- and thepost-shock phase
consisted of 15 instead of20 periods. This shortening of the phases
wasimplemented because in the negative-shockdesign reliable
equilibration was alreadyachieved after 10 –15 periods. (iii) To
achieveroughly the same number of dominated strat-egies in the
post-shock phase, equilibriumprices for x- and y-types in the
positive-shockdesign were as follows: The pre-shock equi-librium
price for x-types (y-types) is P*x 5 11(P*y 5 14) and the
post-shock equilibriumprice is P*x 5 22 (P*y 5 28). As a
conse-quence, the average pre-shock equilibriumprice in a group is
P# *0 5 12.5 while in thepost-shock equilibrium it is P# *1 5 25.
Thus,the difference in average prices between pre-and post-shock
equilibrium is 12.5 in the positive-
shock design while it is 12 in the negative-shockdesign. This
slightly bigger adjustment require-ment in the positive-shock
design is, however,not a problem. If adjustment to equilibrium
inthe NH is faster after the positive shock, this iseven more
remarkable because it occurs despitethe slightly bigger adjustment
requirement inthe positive-shock design.
B. Prices and Expectations after the PositiveNominal Shock
Table 5 shows the evolution of pre- andpost-shock average prices
in the RH and theNH. In the NH pre-shock prices convergefrom above
to the equilibrium P# *0 5 12.5 andas in the negative-shock design
the vast ma-
TABLE 5—EVOLUTION OF PRICES AND EFFICIENCY LOSSESOVER TIME:
POSITIVE SHOCK
Period
Average PriceAverage Efficiency
Loss (Percent)
Real(RH)
Nominal(NH)
Real(RH)
Nominal(NH)
215 13.0 14.9 14.7 26.3214 13.0 14.7 18.2 24.7213 12.7 14.6 10.7
20.7212 12.7 14.3 5.3 13.6211 12.7 14.3 6.1 20.5210 12.5 14.1 1.6
9.129 12.5 13.6 2.1 10.928 12.5 13.4 0.3 11.327 12.4 13.7 1.2
14.826 12.5 13.8 0.6 13.225 12.5 13.8 1.6 8.424 12.5 13.9 0.3
10.423 12.5 13.6 0.9 7.022 12.6 13.1 4.7 6.921 12.5 13.1 1.9
1.0
1 22.5 20.5 22.3 24.02 24.3 22.8 3.9 7.23 24.8 24.1 1.2 4.24
24.9 24.8 0.7 1.45 25.0 25.0 0.2 0.96 25.0 25.1 0.1 0.37 25.0 25.2
0.1 0.48 25.0 25.1 0.1 0.19 25.0 25.0 0.1 0.1
10 25.0 25.2 0.1 0.311 25.0 25.2 0.2 0.112 25.0 25.0 0.1 0.113
25.0 25.0 0.1 0.114 24.3 24.5 6.3 5.915 24.6 24.9 4.0 1.4
1256 THE AMERICAN ECONOMIC REVIEW DECEMBER 2001
-
jority of individuals plays exactly the equilib-rium in the
final pre-shock period. Then, inthe first post-shock period prices
make a bigjump upwards to 20.5 and already in period 4after the
shock, average prices are almostexactly at the new equilibrium of
P# * 5 25.From that period onwards prices remain veryclose to the
equilibrium. This contrastssharply with the adjustment process
after thenegative shock where, throughout the wholepost-shock
period, average prices never cameso close to the equilibrium. This
difference inNH adjustment paths after the negative andthe positive
shock is depicted in Figure 3. Theheavy line in Figure 3 shows the
difference inthe post-shock deviations of average pricesfrom the
equilibrium between the positive andthe negative shock.16 The graph
reveals towhat extent in the NH the adjustment gap,i.e., the
deviation of average prices from theequilibrium, is larger after
the negative shock
than after the positive shock. It shows that thedeviation from
equilibrium is substantiallylarger after the negative shock.
Between pe-riod 2 and 7, e.g., the adjustment gap is fouror more
units bigger after the negative shock.Even in period 10 the
adjustment gap is stillalmost three units bigger.
The impression conveyed by Figure 3 is con-firmed by a more
formal statistical analysis. Ifwe perform regression (3) with the
data after thepositive shock, it turns out that in the NH the
16 Let P# *11 be the post-shock equilibrium price in
thepositive-shock design and let P# 1 be the actual
post-shockaverage price. Analogously, let P# 2 be the actual
post-shockaverage price in the negative-shock design and denote
thepost-shock equilibrium price in this design by P# *12. Thenthe
heavy line in Figure 3 is given by (P# 2 2 P# *12) 2(P# *11 2 P#
1), which measures the difference in the averagedeviations from
equilibrium across the positive and thenegative shock for the first
15 periods of the post-shockphase in the NH.
FIGURE 3. DIFFERENCES IN DEVIATIONS FROM EQUILIBRIUM ACROSS THE
NEGATIVE AND THE POSITIVE SHOCK
1257VOL. 91 NO. 5 FEHR AND TYRAN: DOES MONEY ILLUSION
MATTER?
-
hypothesis of equilibrium play can only be re-jected for the
first three periods (at the 5-percentlevel). Remember that after
the negative shock,group prices were significantly above the
equi-librium for 12 periods. Thus, the evidence un-ambiguously
indicates that adjustment in theNH is much quicker after the
positive shock,which is consistent with our hypothesis thatthere is
less reluctance against adjustment afterthe positive shock.
If there is indeed less reluctance against ad-justment after the
positive shock, at least somesubjects should anticipate this.
Therefore weshould also observe that expectations are lesssticky
after the positive shock. The dashedheavy line in Figure 3 shows
the difference inthe post-shock deviations of the average
expec-tations of P# 2i from the equilibrium between thepositive and
the negative shock. This graph isconstructed analogously to the
heavy line inFigure 3 except that we used the expectationsP# 2i
e to construct it. Thus the dashed heavy lineshows to what
extent the adjustment gap in theexpectations, i.e., the deviation
of average ex-pectations from equilibrium, is larger after
thenegative shock than after the positive one. Thegraph indicates
that the adjustment gap in theexpectations is much larger after the
negativeshock for many time periods. Interestingly, thegraph is
hump-shaped, i.e., the relative sticki-ness of expectations after
the negative shockincreases in the first few periods. This is due
tothe fact that between period 2 and 5 after thepositive shock,
expectations rapidly converge toequilibrium while they are very
sticky after thenegative shock.
Finally, since the rule of thumb of takingnominal payoffs as a
proxy for real payoffscannot be operative in the RH, we
shouldobserve no differences in price adjustment inthe RH across
negative and positive shocks.Table 5 shows the evolution of average
pricesin the RH after the positive shock and Figure3 illustrates
the differences in average pricesand average expectations across
shocks. Ta-ble 5 indicates that in the pre-shock phase ofthe RH the
average price is very close to theequilibrium P# *0 5 12.5 already
after threeperiods. Immediately after the positive shockthere is a
big upward jump in prices to 22.5,only 2.5 units below the new
equilibrium.Already in the third post-shock period the
average price is again very close to the equi-librium. This
indicates that price adjustmentafter the positive shock is rather
quick in theRH—similar to the pattern after the negativeshock. This
similarity is also displayed inFigure 3 and by formal statistical
analysis.The thin line in Figure 3 is constructed anal-ogously to
the heavy line except that we usethe price data from the RH. It
shows that priceadjustment in the RH is only slightly fasterafter
the positive shock. If we perform regres-sion (3) with the
post-shock data from thepositive-shock design we get the
followingresults: The hypothesis that average prices inthe RH are
in equilibrium can only be rejectedfor the first two periods (at
the 5-percentlevel). Note that this is exactly the same num-ber of
out-of-equilibrium periods as after thenegative shock. This
suggests that the differ-ences in the price adjustment across
shocks inthe RH are indeed negligible. The dashed thinline in
Figure 3, which is constructed analo-gously to the dashed heavy
line except thatwe use the expectations data from the RH,indicates
that we can basically make a similarconclusion with regard to the
differences inthe adjustment of expectations across shocks.While in
the NH there are large differences inthe stickiness of expectations
across shocks,in the RH the differences in expectations arerather
small.
Thus all major regularities are consistentwith our hypothesis
that there are beliefs thatsome subjects take nominal payoffs as
aproxy for real payoffs. Nonetheless, it wouldbe reassuring if
subjects themselves ex-pressed such a belief. To check to what
extentsubjects indeed believed this they could indi-cate their
degree of agreement with the fol-lowing statement after the
experiment: “Ibelieved that the other subjects would inter-pret
high nominal payoffs as an indicator forhigh real payoffs.”
Participants could indicatewhether they weakly (dis)agreed,
whetherthey strongly (dis)agreed or whether they to-tally
(dis)agreed with this statement. Thirtypercent of the subjects in
the NH agreed ei-ther “strongly” or “totally” and further 25percent
indicated a weak agreement. In ourview, this can be taken as direct
evidence thata majority of the subjects believed that othersubjects
were affected by money illusion. In
1258 THE AMERICAN ECONOMIC REVIEW DECEMBER 2001
-
any case, these answers nicely fit with ourexplanation for the
large amount of nominal iner-tia observed in the NH after the
negative shock.
V. Summary and Concluding Remarks
Most economic transactions are representedin nominal terms.
Therefore, it seems likely thatpeople often perceive and think
about economicproblems in nominal terms which may inducemoney
illusion. However, for several decadesmoney illusion has been
considered as largelyirrelevant for the nominal inertia of
aggregateprice levels. Instead, most economists have fo-cused on
informational frictions, costs of priceadjustment, and staggered
contracts. This papershows, however, that even in the absence
ofthese factors a fully anticipated negative nomi-nal shock can
cause long-lasting nominal inertiathat is associated with large
real income lossesduring the adjustment phase. Our results
indi-cate that a large part of this nominal inertia canbe
attributed to the direct and indirect effects ofmoney illusion. The
experiments in the settingwith computerized opponents show that the
di-rect effects of money illusion in the form ofindividual
optimization mistakes are not veryfrequent: The introduction of the
nominal framein the setting with computerized opponentscauses only
a small amount of nominal inertia.However, the combined direct and
indirect ef-fects of money illusion generate a very largeincrease
in nominal inertia. This is indicated bythe fact that the
introduction of the nominalframe in the setting with human
opponentscauses a huge increase in the sluggishness ofprices.
Instead of two it takes 12 periods untilaverage prices reach the
post-shock equilibriumin this setting.
The major cause for nominal inertia after thenegative shock is
that subjects’ expectations arevery sticky. In our view this
stickiness of priceexpectations is related to the nature of
moneyillusion in our experiment, i.e., to the belief thatthere are
subjects who take nominal payoffs asa proxy for real payoffs. This
conjecture issupported by direct questionnaire evidence andby the
results of further experiments with a fullyanticipated positive
nominal shock. It turns outthat price sluggishness is much smaller
after apositive nominal shock than after the negativeshock. This
result is also interesting insofar as
there is field evidence indicating that positiveand negative
money shocks have asymmetriceffects. While negative shocks have an
output-reducing effect, positive shocks do not seem toaffect output
(James Peery Cover, 1992; J.Bradford DeLong and Lawrence H.
Summers,1988). The asymmetric effects of money illu-sion on price
sluggishness can be considered asa potential microfoundation for
this result.
Finally, another interesting result of our ex-periments is that
we isolate—in addition tomoney illusion—a further source of
nominalinertia. This source is related to the fact that ina
strategic situation subjects do not merely facean individual
optimization problem but thatthey also have to predict other
agents’ behavior.After any shock, the new equilibrium can onlybe
achieved if subjects have coordinated (equi-librium) expectations.
The comparison of ad-justment paths in the real treatments
withcomputerized and with human opponents showsthat after a fully
anticipated nominal shock, itcannot be taken for granted that
subjects instan-taneously succeed in solving this
coordinationproblem. They will, in general, go through aperiod of
disequilibrium that is associated withnominal inertia. Note,
however, that the coordi-nation problem alone causes substantially
lessnominal inertia than money illusion. It also doesnot cause
asymmetric effects: In the real treat-ment with human opponents the
extent of nom-inal inertia is very similar after the positive
andthe negative nominal shock.
These results show that experiments can be auseful tool for the
examination of the nature, theextent, and the impact of money
illusion instrategic economic interactions. There are manyother
questions that could be usefully tackled byexperimental methods.
One open question ishow subjects who are familiar with moneyshocks
in our pricing game will respond to newshocks. Future research
should thus examinehow the adjustment pattern varies when
expe-rienced subjects face a series of differentshocks. Other
interesting questions concern thenature of nominal wage rigidity
and the inter-action of price and wage policies during adjust-ment.
To answer these questions it is necessaryto introduce workers as
separate players in thegame. The analysis of the interaction
betweenwage and price setting may prove particularlyuseful because
there are likely to be strong
1259VOL. 91 NO. 5 FEHR AND TYRAN: DOES MONEY ILLUSION
MATTER?
-
natural complementarities. For example, iffirms anticipate that
workers will resist wagecuts after a negative money shock they
willprobably be reluctant to cut prices becausethis would reduce
their profits. Yet, if pricesstay high, workers may feel justified
in resist-ing wage cuts. Thus, the reluctance to cutwages and
prices could be mutually reinforc-ing for an extended period of
time. Finally,another interesting question concerns how theimpact
of money illusion varies with the de-gree of strategic
complementarity. Is a lowerdegree of complementarity associated
withless aggregate nominal inertia or not? Or,more fundamentally,
does the impact ofmoney illusion vanish in an environment
withstrategic substitutability?
In our view the results of our experimentsindicate that money
illusion should be con-sidered as a serious candidate in the
explana-tion of nominal inertia and the real effects ofnominal
shocks. Paraphrasing Abraham Lin-coln,17 one can say that, to
render money
illusion behaviorally relevant, it is not neces-sary to fool all
the people some of the time,not to speak of fooling all the people
all thetime. All that is needed is an environmentwith strategic
complementarity and the pres-ence of a small amount of money
illusion atthe individual level—a presupposition thatseems quite
plausible.
APPENDIX: PAYOFF FUNCTIONS
As explained in detail in Section II, subsec-tions B and C,
payoffs were presented to subjectsin payoff tables. These tables
were calculatedfrom the payoff functions explained below. A fullset
of payoff tables is contained in Fehr andTyran (2000) which can be
downloaded fromhttp://www.iew.unizh.ch/wp/iewwp045.pdf. Thepayoff
tables are also available from the authorsupon request.
The real payoff for agent i of type k 5 x, yis given by:
p ik 5
V z F1 1 a z D2
1 1 b z D2G1 1 c z FSP ikM 2
P*k
MD 2 d z D 1 e z arctan~ f z D!G
2 .
P# 2ik is the actual average price of the othern 2 1 players
from the viewpoint of player iwho is of type k. P# *k is the
equilibrium averageprice of the other n 2 1 players from
theviewpoint of a player of type k. Pik is the actualprice of i who
is of type k. P*k is the equilib-rium price of a player of type k.
Finally,D 5 (P# 2ik/M) 2 (P# *k/M). The parameters forM and all
equilibrium values can be found inTable 2.
In all periods and all experimental sessions
the parameters a, b, c, d, e, f, and V were thesame. They were
given by a 5 0.5, b 5 0.6,c 5 27, d 5 1, e 5 0.05, f 5 20 and V 5
40.
REFERENCES
Abbott, Michael and Ashenfelter, Orley. “LabourSupply, Commodity
Demand and the Alloca-tion of Time.” Review of Economic
Studies,October 1976, 43(3), pp. 389–411.
Akerlof, George A.; Dickens, Williams T. and
Perry, George L. “The Macroeconomics ofLow Inflation.” Brookings
Papers on Eco-nomic Activity, 1996, (1), pp. 1–59.
Akerlof, George A. and Yellen, Janet L. “A Near-rational Model
of the Business Cycle, with
17 In his speech on September 8, 1858, Lincoln said:“You can
fool all the people some of the time, and some ofthe people all the
time, but you cannot fool all the people allthe time.”
1260 THE AMERICAN ECONOMIC REVIEW DECEMBER 2001
-
Wage and Price Inertia.” Quarterly Journalof Economics, 1985,
Supp., 100(5), pp. 823–38.
Belongia, Michael T. “Measurement Matters: Re-cent Results from
Monetary Economics Re-examined.” Journal of Political
Economy,October 1996, 104(5), pp. 1065–83.
Bernanke, Ben S. and Carey, Kevin. “NominalWage Stickiness and
Aggregate Supply in theGreat Depression.” Quarterly Journal
ofEconomics, August 1996,