Information, Sticky Prices and Macroeconomic …frontiers.sauder.ubc.ca/Meltzer_Sticky.pdfFile: Meltzer_Sticky.wpd 3 April 10, 2002 15:52 As Keynes recognized, a principal missing

File: Meltzer_Sticky.wpd 1 April 10, 2002�15:52

Information, Sticky Prices and Macroeconomic Foundations Meltzer, Allan H

Review - Federal Reserve Bank of St. Louis; St. Louis; May/Jun 1995; Volume: 77 Issue: 3 StartPage: 101

Abstract: Recently accumulated evidence supports the proposition that nominal prices adjust with a lag.The lag is sufficiently long that real variables respond to nominal changes. Costs of acquiringinformation about the persistence of observed changes - permanent-transitory confusion - is a maincomponent of the cost and a main reason for the lag. Even where prices reflect information fully, allindividuals or firms may not have adjusted fully to available but costly information. The comment thatmacroeconomics should be built on micro-foundations is correct if and only if the micro-foundations areappropriate for the task.

For decades, macroeconomists have listened tocriticism from their professional colleaguesabout the absence of micro-foundations frommost of what they say and do. A typical commentis: "I don't understand much aboutmacroeconomics; how is it related toeconomics?" A possibly less-frivolous commentis that nothing can be said about macroeconomicpolicy until economists develop amacroeconomic theory from a microeconomicfoundation. Anything less is branded "ad hoc"and dismissed.

Much of the work in macroeconomics of the pasttwo decades has responded to these criticisms byattempting to build macroeconomics on anexplicit micro-foundation. That is a worthwhilegoal but it opens the question: Which foundationshould that be? The current generation ofacademic researchers are as divided about theappropriate analytic model as their predecessors.Analytic paradigms now include multipleequilibrium, real business cycle, neo-Keynesian,monetary-rational expectations, and eclecticmodels. Most have a micro-foundation, but it isnot the same foundation.

I do not question the presumption that amicro-foundation is useful. At issue is what thefoundation should be. I question the relevance

for monetary and macroeconomics ofmicro-foundations which feature a representativeagent who trades on a complete set ofArrow-Debreu markets. Despite some limitedsuccesses, it is time to question whether this nowwidely accepted approach is likely to be fruitfuland to suggest why I believe its success will belimited. Success is relative, of course. Realbusiness cycle theory has developed an explicitanalysis of the transmission of productivity andterms of trade shocks. These shocks, thoughwidely recognized earlier, had not been made thesubject of an explicit model. Neo-Keynesiansand others have produced some suggestionsabout pricing. Overlapping-generations modelsof money, intertemporal substitution theories ofunemployment, and productivity shock theoriesof the business cycle have not proved fruitful.These models have not produced either anaccepted foundation for macro theory or averified theory of aggregate output, prices andinterest rates. Perhaps they will in the future, butthe results to date are not promising.

There are two main ways to tie micro and macrotheories. The first is aggregation. Aggregation isan important and much neglected issue, but it isnot my main theme. Most current or recentresearch dispenses with the aggregation problemby assuming a representative individual. Here the


old and new macroeconomics are equallydeficient and open to charges of "arm waving"and "ad hocery." The representative individual isa useful working assumption for some purposes,but the representative individual discards animportant difference between markets andindividuals.

The second problem is the specificmicro-foundation used for macro theory. Most ofmy discussion is directed there. I do not attemptto survey a large literature on micro theory,uncertainty and industrial organization. Theimplications of some of this literature for pricestickiness is ably summarized in Gordon (1990).This paper is a personal statement, reflectingjoint work as noted earlier.

The problem with existing micro-foundationsthat I emphasize is the neglect of costs ofinformation. In Walrasian micro-foundations, alltrades take place at market clearing prices, andthere is no cost of acquiring information. Anauctioneer calls out the prices at zero cost anddoes not close transactions until all transactorsare at an equilibrium. There is a numeraire, butthere is no way for money to disturb the realvalue of production or purchases. Non-neutralitycannot arise. Attempts to graft a monetarydisturbance onto these micro-foundations seemmisdirected.

The hypothesis that all observed prices aremarket clearing prices does not imply that allindividuals, behaving rationally, know thoseprices and fully adjust to them. Information iscostly to acquire; time and resources must beused to collect, process and interpret--the latterespecially--new data. Even in markets dominatedby price takers, there are differences between theinformation processed and known to arbitragersand specialists and what is known tonon-specialists. The representative individualparadigm ignores this distinction.

One way to proceed is by aggregating

heterogenous individuals who face differentcosts of acquiring information, and much newwork has taken this path. As Gordon (1990)emphasizes, at any time there are many differentlayers of pricing and output included inaggregate prices and output--suppliers, suppliersof suppliers, foreign producers, and so on.Instead of trying to aggregate over these manydiverse and changing levels of decisions, it mayprove more useful to treat them as part of astochastic process.

One of the problems with the hypothesis that"sticky" prices reflect decisions at different levelsof aggregation comes out clearly in Gordon'spaper. He argues that sluggish price adjustmentreflects marginal cost pricing. "[Agents] careabout the relation of their own price to their owncosts, not to aggregate nominal demand. Unless asingle agent believes that the actions of all otheragents will make its marginal costs mimic thebehavior of nominal demand with minimal lags,the aggregate price level cannot mimic nominaldemand, and Keynesian output fluctuationsresult."

This argument has a perverse implication.Farmers or farms that operate in commoditymarkets with many competitors should beslowest to adjust to aggregate demand shocks.Yet commodity prices typically adjust quickly.Large industrial firms--General Motors,Mitsubishi--know that they are a relatively largepart of the economy, so they should adjust toaggregate demand more promptly thancommodity producers. Typically, they do not.

Missing from Gordon's analysis is the differencein information and in costs of acquiringinformation. Commodities are traded on openmarkets, so prices promptly reflect changes inthe factors affecting demand and supply. Pricesof autos, steel, heavy industrial products andconsumer durables are not set in organizedmarkets. Information on which to base prices ismore uncertain.


As Keynes recognized, a principal missingelement is uncertainty about the future.Uncertainty and its twin, costs of information,make it rational for some firms to adjust pricesslowly. Prices may be "sticky," as economistshave observed for about as long as there has beena discipline.

Some economists will scoff that sticky prices areirrational or equivalent to leaving five, 50 or 500dollar bills in the street. This is an error arisingfrom dependence on Walrasian foundations andneglect of information costs. Yet, there isnothing novel about invoking costs of acquiringinformation to explain sluggish price adjustmentor sticky prices. The positive slope of Lucas'(1972) aggregate supply curve arises fromconfusion between relative and absolute pricechanges; prices in that model do not immediatelyadjust to new information. The cost of learningwhether a change in demand is an aggregate orrelative change is infinite for one period and zerothereafter. In the neo-Keynesian modelsdiscussed in Ball and Mankiw (1994), costs ofprice adjustment--so-called menu costs--and"real rigidities" are assumed to be present. Theauthors accept that one of the principal costs ofprice adjustment is the cost of acquiringinformation relevant for a decision about howmuch to change price.

Given this widespread acceptance of informationor transaction costs, what remains to be done?Lucas' (1972) model implies more rapidadjustment of prices and output to newinformation than we observe in practice. Could itbe rational to adjust more slowly than in oneweek or month? Is the compelling fact that dataon the price level are released monthly? Or isthis information subject to error so that it takeslonger to disseminate, interpret and act on thesedata? Neo-Keynesian models build on the modelof an imperfectly competitive firm, with strongimplications for profits and excess capacity thatdo not find empirical support at the micro level.If monopolistic elements and menu costs are the

source of sticky prices, there are testableimplications for the profits of different types offirms; there is a significant problem ofreconciling continuous excess capacity withrational behavior. Further, as Gordon (1990) hasnoted, costs of adjusting output are neglected inthese models. Yet these costs may be larger formany firms than costs of price adjustment.

My principal criticism of both neoclassical andneo-Keynesian approaches is that they seek toexplain why firms delay using availableinformation. By putting the issue in thatframework, they neglect the uncertainty thatsurrounds much of the aggregate anddisaggregated data. This paper uses severalstrands of earlier work to explain rational pricesetting and gradual adjustment as a response touncertainty about what current informationimplies. Information is costly to acquire and tointerpret as in Brunner and Meltzer (1971) orAlchian (1977). As in Bomhoff (1983), aprincipal difficulty in interpreting information isuncertainty about how long changes will persist.This is the central idea developed in Brunner,Cukierman and Meltzer (1983), but we took theidea from Muth's (1961) seminal paper onrational expectations. In Meltzer (1982), I usedthese ideas to discuss price setting.

There are three separable aspects of pricing toconsider. First, (some) prices are set by firms.Second, firms choose nominal values; they donot index or set a relative price. Third, prices thatare set change less frequently than prices inauction markets.

Each aspect is important for macroeconomics. Ifsome prices were not fixed in nominal value forat least one period, relative prices would beinvariant to a monetary change. Delays inrecognizing permanent changes in money or itsgrowth rate would affect only real balances. Atbest, we would be forced to fall back on the realbalance effect in consumption to explainshort-term real effects of changes in money. A


rational reason for setting some, but not all,nominal prices permits a more direct effectthrough inventory adjustment. Firms that holdinventories can both anticipate future pricechanges and buffer current transitory pricechanges by varying inventories. Because someprices are determined in auction markets, pricesetting introduces different speeds of adjustmentand relative price changes. My extensive workwith Brunner emphasized the relative prices ofassets and output. This introduces the differencebetween the replacement cost, or the cost ofcurrent production, and the market price ofexisting assets. Asset prices adjust more rapidlythan output prices particularly for assets thattrade in organized markets. Hence, informationcosts (and transaction costs) are implicit in thatframework as an explanation of changes inrelative prices. My emphasis here is oninformation as in Brunner, Cukierman andMeltzer (1983). Relative price changes inresponse to nominal shocks are part of that story,but they remain in the background.

HOW STICKY ARE PRICES?

Postwar recessions typically last about ninemonths on average. Costs of information or otherexplanations of sluggish adjustment must be ableto explain this timing, and estimates of pricestickiness should be consistent with data oncyclical fluctuations. I digress therefore toconsider a recent attempt to measure stickiness.Blinder (1991) reports preliminary findings froma survey of pricing decisions by managers of arandom sample of corporations with annual salesof $10 million located in the northeastern UnitedStates. Blinder's survey produced two verydifferent sets of estimates.

First, Blinder reports the answer to a questionabout how frequently a firm changes its price. Hefinds that the median firm changes price aboutonce a year. Nearly 40 percent of the firms wereat the median. More than one-sixth of the firmschanged prices less frequently than once a year,

so 55 percent of the prices change no more thanonce a year.

Second, Blinder also reports the respondents'estimate of the mean lag of actual prices behindincreases and decreases in demand and cost.There is considerable uniformity in theseresponses. The mean lag for increases ordecreases in demand and cost is three to fourmonths.

The second set of data suggest considerableuniformity of response to the four shocks. Thereis little evidence of the asymmetric adjustmentassociated with Keynesian downward priceinflexibility. The data also suggest aninconsistency that Blinder does not mention.How do we reconcile a mean delay of three tofour months in response to changes in demandand cost with the report that 76 percent of thesample changes price no more than twice a yearand, as noted, 55 percent changes price no morethan once a year? Are demand and cost changesinfrequent? It is difficult to reconcile theassumption that costs change infrequently withevidence showing that commodity prices andother open market prices change daily andtypically fall in recessions and rise in expansions.

Blinder did not ask whether firms adjust theirprices fully in response to changes in demandand cost. They may adjust partially, as impliedby rational behavior under uncertainty about thepersistence or permanence of announcedchanges, or they may anticipate future changesand adjust prices more than current changes incost or demand. One way to reconcile thedifferent responses is to assume that respondentsanswer the two questions in different ways.Suppose they treat price as a scalar when askedabout the frequency of price changes but includein price adjustment more than just price setting.On this interpretation, quoted prices are onecomponent of a vector of terms and conditionsrelevant to sellers and buyers. The theoreticalterm price used in economics typically subsumes


delivery time, discounts, advertising allowances,volume rebates, payment terms and otherconditions used to adjust the buyer's cost and theseller's net receipts. Blinder's early resultsconfirm that delivery lags and service aremoderately important for 65 percent of the firmsin his sample; he reports delivery lags andservice are the most commonly cited reason forprice stickiness.(1) Delivery lags and service areone way of adjusting the theoretical term "price"while leaving the quoted price unchanged.

Assume that firms initially respond to changes incost and demand by adjusting deliveries,advertising allowances, discounts, and so onwhile leaving quoted prices unchanged forseveral months or longer. If managers areuncertain about the duration of changes indemand or cost, they can change othercomponents of the price vector to test themarket's response. By changing delivery terms,or offering or removing discounts, firms canchange their revenues or the buyers cost withoutchanging the quoted price. This pricing modelcan be used to rationalize the familiar Keynesiansupply curve--a reverse L--when quoted pricesare distinguished from other terms in the pricevector. Equally, the model can explain thedifference in response to the questions inBlinder's survey.

Figure 1 shows the initial response to an increasein demand. (Figure 1 omitted) On the left, thequoted price (p) does not respond to a perceivedchange in demand when output is belowcapacity; the firm (or industry) increases output(q) with little or no change in quoted price. Onthe right, the price vector (p) includes otherdimensions of the firm's price; supply is drawn asa positively sloped linear function. An increasein perceived demand from d sub 0 to d sub 1induces the firm to reduce advertisingallowances, remove discounts, or change someother component of the price vector, so (p)increases.

If the firm's initial response reflects uncertaintyabout the persistence of increased demand, theresponse changes as information accrues. Anincrease in the quoted price may substitute for orsupplement other components of the price vector.As perception of the magnitude of the permanentincrease becomes clearer, the firm may recognizethat the new demand curve is at, or to the rightof, d sub 1 . Or the firm may raise p, with Punchanged.

We need not explore the many possibilities. Themain point is that uncertainty about the degree ofpersistence and the use of a price vector permitthis hypothesis to account for very differentresponses within a rational expectationsframework. In particular, the firm's revenues mayrespond instantly to increases in demand, butquoted prices may adjust with a long lag, asBlinder found. The reasoning is symmetric.Perceived reductions in demand may inducefirms to offer discounts and allowances with punchanged. As information accrues andperceptions change, the actual price, p, isreduced.

The Keynesian supply function does not work forchanges in cost if the demand curve is notkinked. Changes in cost shift the supply curve, soprices change instantly, whether the supply has areverse L-shape or is monotonically increasing.Since Blinder's survey finds that price respondsabout as promptly to changes in cost as tochanges in demand, this evidence rejects theKeynesian supply curve.(2)

The distinction between price as a vector and ascalar does not reconcile the differences intiming reported in the survey. Although I believethat prices, terms and conditions change atdifferent rates when there is uncertainty about thepersistence of the market conditions inducingfirms to change prices, those differences areneglected hereafter. Price stickiness will meanthat it takes about three or four months onaverage for prices to respond to demand and cost


changes. The three-to-four month delay thatBlinder reports includes at least one quarterlyreporting period at which managementsannounce earnings and sales and, mostimportantly, observe reported earnings,inventories and sales of competing firms.(3)Knowledge of competitors' results helps the firmto supplement trade gossip and other informaldata sources to decide whether a persistentchange in market demand (or industry costs) hasoccurred.

INFORMATION AND PRICE SETTING

Anyone familiar with literature on the behaviorof firms knows that there are many rationalreasons for firms to set prices. This sectiondiscusses one reason that is rather general andcan be incorporated readily intomacroeconomics. Price setting is considered as aresponse to uncertainty or costs of acquiringinformation about the market clearing price by atleast one (large) group of market participants.(4)At times, prices convey information known tothe seller but not to the buyer. At times, neitherthe buyer nor the seller is certain about themarket clearing price. Among the possiblereasons on the cost and demand sides, thissection emphasizes rational reasons why buyersand sellers do not instantly know the permanenceof changes in demand. They develop contractsand market arrangements to deal with thisuncertainty.

This approach differs from the literature onso-called menu costs. The menu cost literatureemphasizes costs of changing prices.(5) Thesecosts are recognized as relatively small (Ball andMankiw, 1994). Moreover, the menu cost modeldoes not explain why sellers face different costsand adjust at different speeds. I do not challengethe existence of menu costs, but the emphasishere is on uncertainty and information costs--thecost of learning about current and prospectivemarket conditions. Information costs differ byfirm and industry. They depend on the way in

which markets are organized and the types ofcontracts that emerge to reduce costs ofuncertainty, information and moral hazard.

Market Organization

Frank Knight was an early expositor of theeconomics of information. Knight (1933), arguedthat decisions about how much and when toproduce require a pooling of information aboutindividual decisions to purchase. Whenaggregated, the individual decisions constitute ademand curve.

Firms reduce uncertainty for consumers bypooling information. In principle, individuals cancontract in advance for the goods and servicesthat they want. Organizing retail firms that poolinformation is an efficient alternative to futurescontracts if individuals are less certain about themagnitude and timing of their purchases thanfirms are about market demand. Knight appealsto the law of large numbers to explain firms'advantage. He compares pooling by firms toinsurance and concludes that the two differ in animportant way; a firm's pricing and outputdecisions are non-insurable because they requiremore subjective judgment, and errors are lesslikely to cancel across firms.

In Knight's view, firms produce for inventoryusing pooled information about expecteddemand. This arrangement shifts uncertaintyfrom individuals to firms and, by pooling,reduces the social cost of bearing uncertainty.

Knight's argument is one of several that linksinformation costs and uncertainty to price setting.A non-uniform distribution of information iscritical for these arguments. In an auctionmarket, all market participants must haveinformation about the qualities of the goodstraded and their prices. In the standard Walrasianmodel, this is accomplished by: (1) assuming thepresence of an auctioneer who calls out theprices; (2) allowing recontracting; and (3) letting


all trades be made simultaneously. Theseassumptions are necessary for equilibrium. Theyleave no role for monetary disturbances.

The necessary conditions are frequently violatedin practice. Some people have a comparativeadvantage in acquiring information. Somereceive information about market conditions as aby-product of other activity. For example, insecurities markets, there are brokers, dealers andmarket makers who acquire specializedinformation in the course of trading. Assemblingall or a sufficiently large number of marketparticipants also imposes severe constraints. Formany reasons, including the law of largenumbers, people choose different times for theirmarket activities. The commitment to assembleat pre-specified times for all marketing purposesis costly, if not impossible. The use of an agentintroduces costs of monitoring and supervision.

In practice, markets operate in many differentways. One alternative commonly found inoriental countries is called a bazaar. Prices arenot posted, and there is no auctioneermechanism. Buyers and sellers negotiate (haggle)until a transaction is completed or the negotiationterminates. Considering a bazaar brings out theimportance of information.

The bazaar requires an investment of time bytransactors. Where the auction market requiressimultaneous arrival of all participants or theiragents, the bazaar depends on a trickle ofarrivals. It cannot cope with large groups arrivingsimultaneously because each negotiation isseparate. It is not surprising that the bazaar isfound in comparatively simple, low-incomeeconomies. In these circumstances, the allocationof time to the bargaining process may be partly aconsumption good. With rising opportunity costsof time, the disadvantages of the bargainingprocess exceed the benefits.

The working of a bazaar restricts its application.There are severe limits on the number of

transactions per period. Sellers cannot serveseveral customers simultaneously. Delegatingbargaining to employees poses both an incentiveand a moral hazard problem. Buyers incur coststo learn about reservation prices, sinceinformation is revealed only by commitments totransact firm offers to buy or sell. An acceptedoffer to purchase may be above the seller'sreservation price; a refusal to sell may be basedon an incorrect inference that the buyer is willingto pay more. Learning the reservation price andnegotiating the market price often requires aseries of offers. There is no certainty that thereservation price is revealed. Subsequenttransactors on one side of the market do notknow the history of past transactions. They mustinvest their own time.

A seller who posts prices reduces the buyers'costs of acquiring information about prices.Buyers' costs of comparing prices by differentsellers are reduced. The social advantage of pricesetting is greatest where one party to thetransaction has more information about marketconditions. Some examples illustrate thisargument.

Information about prices conveys more than justpurchase cost. Posted prices can also reducecosts of acquiring information about quality. Forexample, a restaurant owner must decide on themarket he wants to serve. This decisioninfluences the kind and quality of food served,the services offered, and the prices charged. Byposting prices, the owner informs the buyerabout his choices. Although the buyer mustsample to judge quality, the correlation betweenquality and price helps the buyer to decidewhether to sample. A policy of frequentlychanging prices reduces information and placesthe restaurant at a competitive disadvantage.

The organization of the diamond market providesadditional evidence on the role of information inmarket organization. The wholesale diamondmarket is an auction market dominated by buyers


and sellers who are specialists. Traders rely ontheir own skill in judging quality, knowledge ofprices and other attributes. The retail market isvery different. The sellers are mainly specialists;the buyers typically have much less informationthan the sellers. By posting prices, sellers exploitthe correlation between quality and price toinform buyers. Buyers find it less costly to investin information about the seller than to invest ininformation about the quality of diamonds, sosellers use resources to build reputation. If costsof acquiring information about the quality ofdiamonds were to fall to a minimal value, thesearrangements would change. Diamonds might besold in supermarkets or in retail auction markets.

Price setting is valued by transactors even insome auction markets. In well-organized auctionmarkets, we find people willing to pay for theright to purchase or sell at fixed nominal prices.The contracts expressing these rights, known as"put" and "call" options, give the owner the rightto buy or sell at a fixed price within a fixed timeperiod. The prices of the puts and calls aredetermined in auction markets. The price of theseoptions is the cost that people pay for the right totrade in the future at prices fixed today.Similarly, in commodity markets, hedgers pay tochange uncertain future prices into knownvalues. They pay a fee for the right to buy or sellat fixed nominal prices. In such markets,information about current and currentlyanticipated future prices is available. The feepermits transactors to avoid uncertain futureprice changes.

Costs of information and transactions are notuniform across goods, so no single form ofmarket organization dominates all others. Thespecific reasons transactors are willing to pay forputs and calls differ from the reason for pricesetting in the diamond market, just as thediamond market differs from the restaurant. Eachis related, however, to costs of information. Theorganization of the diamond market reduces thecosts of bearing uncertainty about quality. The

market for puts and calls permits asset owners orspeculators to limit risk of wealth changes. Andthere are other market arrangements where pricesetting is useful. Catalogues must post prices.Contracts fix prices for a year or more onhousing rentals, magazine subscriptions, andautomobile leases. These many differentexamples suggest that there are advantages indifferent types of contracts.

The examples suggest a way to model pricesetting formally. The seller has information thatis costly for the buyer to acquire. The sellerinternalizes the cost of acquiring the information;it is part of his specialized knowledge and herevises his information in the process of buyinginputs. By posting a price, he exploits thecorrelation between price and quality. In goodsmarkets, the seller may offer a particular type ofput--an option for the buyer to return themerchandise if the quality is not as representedor, perhaps, if the buyer finds the samemerchandise at a lower price. The buyer pays forthe good and for the put, but the purchase cost islower than under alternative forms oforganization. In service markets, the buyer maypurchase increased certainty that he will not haveto relocate or search for a particular input at aninconvenient time.

Labor Markets

The conditions leading to price setting also applyto the labor market. The terms negotiated and thetime horizon built into an agreement depend onthe assessments of the parties. There are noorganized futures markets for labor. Both partiesuse all available information to form theiruncertain assessments. An assessment of themarket is a (subjective) probability distribution.The more diffuse the distribution, the shorter thetime covered by the arrangement.

Realizations often deviate from the expectationsimplied by the subjective probabilitydistributions. Both parties have to infer from


realizations whether the unexpected changes aretransitory or permanent.

A transitory change does not change the expectedvalue, so the unanticipated gains and losses donot change the information on which theagreement was based. Either party may believethat a costless revision of the bargain to adjust toa transitory change would be beneficial, butattempts at revision for each such change raisethe cost of transacting and eliminate the benefitsof a longer-term agreement.

A more permanent change in conditions poses adifferent problem. The initial assessment of atleast one party must be revised in the light of thenew information. If the stakes are sufficientlyhigh, the permanent change may justify the costof renegotiation.

Negotiations proceed more smoothly when bothparties share the reassessment of marketconditions. Differences in assessment provideevidence on the extent of uncertainty. Strikes andlockouts increase with differences in assessment,for example, at the start of a period of inflationor disinflation. If both parties could agree on theactual shocks that occurred in the past and onhow long they will persist, they could contract inadvance to compensate for unanticipated changesafter they occur. Permanent changes in nominalvalues would not be allowed to affect real wages.Permanent changes in productivity would be paidto workers if positive and by workers if negative.That we do not observe contracts of this kindsuggests that assessments differ even after theevent and that reaching a common assessment ofthe past is costly. Of course, setting nominalwages is not costless either. Bargaining ornegotiating to correct for unforeseen events canbe costly, privately and socially, if there arestrikes or layoffs.(6)

Differences in information can explain pricesetting, but they do not fully explain why firmsand employees often set nominal wages or

prices. There must be some additional cost ofsetting relative prices or benefit from settingnominal wages (or prices). One explanation isthat the parties do not agree on the interpretationof real wages and, particularly at low rates ofinflation, have difficulty agreeing on anappropriate index.

There are two different meanings of real wages.One meaning expresses real wages in terms ofthe product of the firm at which the worker isemployed. The second refers to the basket ofgoods and services that the worker can purchase.The problems of setting contract terms differ inthe two cases.

Contracts that set wages in relation toproductivity require a satisfactory solution to themeasurement of productivity. Where precisemeasurement is difficult, as in service industriesor managerial tasks, real wage contracts aredifficult to write. Even when productivity ismeasured reliably as in piece-work systems, themeasure does not translate directly into a realwage rate. Valuation is required; often some(more or less) arbitrary system must be used toimpute the price of the final product to thevarious inputs. Imputation and valuation bringtwo additional problems. One is moral hazard;the employer has some incentive to adopt a costaccounting system that benefits him. The other isthe difference between relative and absoluteprices. The employer is mainly concerned withthe relation of wages to product prices. Theemployee is concerned also, and perhaps most,about the relation of wages to the price ofconsumables--the second meaning of real wages.A frequent compromise in periods of inflation ispartial indexation to the price of consumables.

Real wage setting with full indexation is rare.This suggests that at low levels of inflation,buyers and sellers prefer ex post adjustmentthrough negotiation to reliance on an impreciseor imperfect index. As inflation increases, costsof non-indexation rise relative to costs of


indexation. More parties choose a mixed strategyof partial indexation. Experience in Israel andBrazil suggests that at relatively high rates ofinflation, indexation is nearly, but never fully,complete. There is always some lag inadjustment. Data for countries with high inflationalso suggest that workers willingly foregoindexation if offered relatively stable prices.

The choice of an index is a problem for bothparties. Some of the problem would be removedif shocks could be identified unambiguously, ifone-time price changes (transitory changes in therate of inflation) could be separated frompermanent changes in inflation, or if all shockswere of one kind--for example, permanent,nominal or real aggregative, or real allocative.The absence of reliable information preventssettlement on an optimal indexation formula.

Nominal wage contracting is also not ideal.Different types of contracts are used to adjustnominal wages for inflation. In periods of low ormoderate inflation, we observe contracts thatdiffer in duration, in the extent of formalindexation, and in the use of clauses permittingreopening of the wage agreement during the lifeof the contract. We observe also that the types ofcontracts change with the rate of inflation andthat employers can be induced to compensate for(some) past price changes when (non-indexed)contracts are renewed. At high rates of inflation,firms and other market participants monitor therate of inflation. Costs that were previouslymarginal costs of information became fixed orquasi-fixed costs of information. Nominal pricesadjust more frequently.

Retail store leases differ from wage contracts.Leases are often indexed to the volume of sales.Sales are more easily monitored and thereforeless subject to moral hazard. Valuation is basedon receipts, so measurement is not as much of aproblem as for profits or productivity. Bothparties have an interest in maintaining theproperty.

Bond contracts provide another example of theproblem of choosing an index. Private parties donot issue price-level linked bonds. Under thegold standard, however, firms offered to pay ingold. Buyers and sellers could agree on thisindex of long-term value. Once this commonmeasure, related to the value of money, becameless relevant, indexed bonds were rare. Inabilityto agree on an index left no agreement. InBritain, Israel, Brazil and a few other countries,the government resolved this problem by issuingan indexed bond.

Comparison of the choices made in markets forlabor, rental property and bonds suggests thatagreement on an index is most difficult whenprices can change because of real and nominalshocks, and changes can be permanent ortransitory. This should not suggest thatnon-indexation is optimal. Contracting partiesfind many different solutions but, as experiencein labor markets shows, full indexation is rare.

A Stylized Model

In several papers, I have used a model ofpermanent and transitory changes based onBomhoff (1983) to study the frequency of shocksand their interaction. A main conclusion of thiswork is that there is no reason to expectconstancy or even repetition in the frequencydistribution of shocks. The public cannot usedata from the past to anticipate the future relativefrequency of permanent and transitory shocks orreal and nominal shocks. These frequencieschange with public policy decisions, policy rulesat home and abroad, weather, inventions orinnovations, changes in market structure, andother factors affecting an economy's structuralrelations.

The model has three types of shocks: transitoryshocks to the level; permanent shocks to thelevel (which are transitory shocks to the growthrate); and permanent shocks to the growth rate.Let x sub t be the current value of real output and


p sub t the current value of the price level. Bothprices and output can be affected by each of thethree shocks so that

(Equations omitted)

and

(Equations omitted)

There are stochastic elements in the growth pathsof output and inflation in addition to transitoryand permanent changes in the levels of outputand prices. Much confusion in the discussion ofinflation has been caused by the use of"inflation" to refer both to level changes such asoil shocks (distributed over time) and persistentchanges in the maintained rate of change.

Suppose we now introduce a common type ofPhillips relation between p and x in theneighborhood of p sub t = 1.0.

(Equation omitted)

The way in which prices will change over timedepends on the permanence of the shocks. Ittakes time for agents to learn whether the shockschange u sub t , v sub t or z sub t and, from thesimultaneity of x and p, shocks to epsilon sub t ,gamma sub t and p sub t . The path by whichprices adjust--or the degree to which they aresticky--depends on the nature of the shocks. It isentirely rational in this framework for prices tobe sticky and for the speed of adjustment todiffer from one episode to another.

The model in Brunner, Cukierman and Meltzer(1983) conveys some of the ideas just discussed.The main idea of the model can be written in ageneral way. Consider the following system ofsimultaneous equations for a macroeconomicsystem. In the underlying micro model, firms setprice at the start of the period and holdinventories. Shocks are revealed after productiondecisions are made.

(1) f[y sub t , p sub t , Delta h sub t , h sub t-1 , isub t , r sub t , x sub t , m sub t ]=0,

where y = output, p = price level, h = stock ofinventories, i = nominal interest rate, r = realinterest rate, x = exogenous real shock, and m =nominal money stock. Under full informationabout the structure of the shocks and in theabsence of transaction costs, the price levelreflects current information about money. Moneyis neutral. Incomplete information of some sort isa necessary condition for significant monetaryeffects, but it is not sufficient.

Lucas' (1972) hypothesis about incompleteinformation restricts uncertainty to misperceptionof current shocks; people do not know whether ashock to demand is specific to their product or isa general increase in demand. It is now generallyrecognized that Lucas' hypothesis produces aresponse of real output to a nominal shock butdoes not generate as much persistence as is foundin cyclical fluctuations of output and prices.Persistence must be introduced. The problem isto introduce persistence without introducing animplausible reason for neglecting information incurrent output, prices, interest rates and othervariables.

In Brunner, Cukierman and Meltzer (1983), arepresentative producer sets price and output atthe start of each period using all availableinformation at the time. This includes, inparticular, the variables in equation 1--inventorylevels and changes, interest rates and currentpolicy. When making price and output decisions,producers are uncertain about the permanence ofobserved shocks. As in the earlier discussion,they do not respond to changes that are perceivedto be transitory.

Let x* and m* denote the perceived permanentcomponents of x and m. Knowing these values,producers set output and the price level; y* andp* denote the producers' decisions. They aredetermined from a subset of the system of


simultaneous equations:

(2) f[y sub t *, p sub t *, Delta h sub t *, h sub t-1, i sub t *, r sub t *, x sub t *, m sub t *] = 0.

The actual values are x and m, but x - x* and m -m* are ignored in adjusting prices and output.Transitory changes are not innocuous. With y*and p* adjusted to x* and m*, nominal and realinterest rates and the change in inventories adjustto the perceived transitory changes, shown asequation 3:

(3) f[y sub t *, p sub t *, Delta h sub t , h sub t-1 ,i sub t , r sub t , x sub t , m sub t ] = 0.

Output, inventories and other real values respondto both nominal and real shocks in this model byamounts that depend on the size of themisperception. As new information arrives,producers revise their beliefs about thepermanence of shocks; p*, y* and all othervariables adjust to the changed perception.

An econometrician examining the data generatedby this model would at times find seriallycorrelated changes in output and seriallycorrelated errors. Serial correlation arisesfollowing a large permanent real or nominalshock if the shock is believed for a time to betransitory. As time passes, and errors arerepeated, perceptions adjust. Even if theunconditional error in the population is seriallyuncorrelated, misperception of the permanenceof shocks can lead producers to make errors that,ex post, are serially correlated. A model of thiskind may explain why some researchers havefound ex post real effects of anticipated changesin money. See, inter alios, Mishkin (1983).

Although prices do not fully adjust to shockswhen they occur, decisions are entirely rational.Producers use all available information, but theymisinterpret the nature of the shock. Once theyperceive that the shock is permanent, prices andoutput fully reflect the information.

The length of the recognition lag depends on therelative variance of permanent and transitoryshocks. The larger the variance of permanentshocks, relative to transitory shocks, the shorterthe recognition lag. If all shocks were permanent,prices and output would adjust to p* and y*values as soon as the shocks occurred. The lag inour model would be one period, as in the Lucasmodel. If all shocks are transitory, y and Pchange, but y* and p* never adjust.

In Keynesian models, inflexible prices (orwages) and gradual adjustment are taken asevidence of disequilibrium. The informationstructure of the model here implies that thisinference is invalid. Buyers and sellers use allavailable information and adjust to a marketequilibrium.

There are three types of equilibrium. At anymoment, there is a permanent stock equilibriumcharacterized by the state variables x* and m*.This equilibrium occurs when Delta h = 0. Thevalues of all variables are adjusted to theperceived permanent shocks and the condition ofunchanging inventories. Each firm uses resourcesat the profit maximizing rate. No firm seeks toexpand or contract output or change its price andinventory position.

A permanent equilibrium is less encompassing.The state variables in this case are x*, m* and hsub t-1 . All other variables adjust to theseconditions. If x* and m* remain unchanged, thepermanent equilibrium converges over time tothe permanent stock equilibrium.

A transitory equilibrium imposes an adjustmentof the system to given values, m -= m*, x -= x*,y*, p* and h sub t-1 . Inventory adjustment andinterest rate changes produce the transitoryshort-run equilibrium.

In Brunner, Cukierman and Meltzer (1983) weshow how real variables respond to monetaryshocks in a model of this kind. All expectations


are rational. No information is wasted once it iscorrectly perceived. Misperceptions occur, so thesystem adjusts sluggishly to information that, expost, turns out to be permanent. The discussionhere, following the original, uses inventories as arepresentative real variable but the response ofreal output is similar.

Figure 2 shows the adjustment path. (Figure 2omitted) Asterisks denotes permanent values,and a superscript a denotes actual values.

"Up to period t, the economy is at anequilibrium: (equation omitted) During period t,there is an unanticipated increase in moneygrowth. Interest rates fall; with prices fixed forthe period, aggregate demand increases, andinventories are reduced. Inability to identifypermanent shocks means that the perceived valueof m changes as information about the rate ofmonetary expansion becomes available. Forecasterrors remain on one side of zero for severalperiods. Forecast errors reinforce the cyclicaldeviation of inventories. A large permanentincrease in money growth, that is notimmediately recognized as permanent, increasesm sub t , and adds to the cyclical deviation ofinventories.

"The reduction in inventories sets off a processof adjustment of output and inventories. The pathalong which inventories are expected to adjust atthe onset of period t+1 is shown by the positivelysloped line from (character omitted) to thepermanent level of inventories, H sup p . Alongthis path, a typical firm plans to produce outputin excess of expected sales and build inventories.The expected value of inventories by the end ofperiod t+1 is shown as H sub t+2 . If themonetary shock is correctly perceived astransitory and there are no further shocks, firmsadjust along the planned path and achieve thevalues of inventories, H sub t+2 , H sub t+3 , insuccessive periods until the permanent value, Hsup p , is restored.

"Suppose, however, that the increase in moneygrowth persists. In period t+2, interest rates areagain pushed below the value expected for theperiod, and actual inventories, (characteromitted) are, again, below the value expected, Hsub t+2 as shown in fig. 2. At the beginning ofperiod t+2, firms and households expect theeconomy to adjust along the new path, from(character omitted) to H sup p . The new pathreflects all the available information about theshocks, including beliefs about the permanenceof the change in money growth and knowledge ofthe structural parameters...If the variance of thetransitory component of money growth is largerelative to the variance of the permanentcomponent, adjustment to permanent changes isrelatively slow. Inventories can fall below theirexpected value for several periods and, thus,move away from H sup p .

"Additional information about the permanence ofthe shock that first occurred in period t isrevealed each period, so the path of adjustmenttoward H sup p is not smooth. As time passes,however, the addition to information is small.After t+3 in fig. 2, inventories adjust toward Hsup p unless another shock--anotherunanticipated increase in money growth--lowersinventories and starts a new process of learningand adjusting.

"Actual inventories follow the outer envelope infig. 2; expected inventories, H, follow theadjustment paths that start at the actual values foreach period. The figure shows principal featuresof our model of inventory behavior, augmentedby the effect of permanent-transitory confusion.Deviations from H sup p are on one side of Hsup p for several periods because of the slowadjustment of inventories. This feature occurseven if all shocks are white noise. In addition,information about the permanence of shocksbecomes available gradually. People use allinformation and their beliefs about permanentvalues to determine the adjustment path, but theymake unavoidable errors because they learn


about the permanence of shocks gradually."

SUMMARY

My emphasis in this section is onmicro-foundations that lead to price setting andto gradual adjustment. I do not claim to haveuncovered a unique structure that produces stickyprices under rational behavior. There are manyreasons and many valid hypotheses that makeboth price setting and gradual price adjustmentcompatible with rational behavior.

Price setting is sufficient for real effects ofmonetary shocks. I have discussed severalreasons for price setting--menu costs, Knight'suncertainty argument for the existence of firms,producers' desire to signal quality, purchasers'gains from lower cost of information andbargaining, and other differences in costs ofacquiring information.

In principle, sellers or buyers could set relative,not absolute, prices. To do so efficiently, buyersand sellers have to agree on an index. Price indexnumbers are subject to real and nominal shocksthat are sometimes permanent, sometimestransitory, and sometimes alter the rate of pricechange persistently. If these shocks could becorrectly identified ex post, and if their expectedduration were known, the parties might agree toadjust prices after shocks occur. Far more often,ex post adjustment is done by negotiation, or theparties agree to partial indexation and negotiateabout the remainder. These arrangements areconsistent with the presence of relatively largecosts of acquiring information and agreeing onwhat has occurred, whether the change ispermanent or transitory and how long and howlarge future price changes will be. Among thesecosts are the costs associated with moral hazardif one party controls relevant data.

Economic contractions typically last nine monthsto one year. Considerable evidence suggests thatprice changes may lag as much as two years

behind monetary shocks. To yield the patterns ofprice and output change observed in actualeconomies, price setting must be joined to arational reason for persistence.

Permanent-transitory confusion--uncertaintyabout the duration or persistence ofshocks--provides one such condition. Under thishypothesis, lags can be long or short and ex posterrors can be serially correlated, if a largepermanent shock is perceived as transitory. If thevariance of t he permanent shocks is high relativeto the variance of transitory shocks, the lag isrelatively short, and there is no reason forsignificant serial correlation to be observed expost. In this case, price setters believe that mostshocks are permanent, so they adjust promptly.

SOME EVIDENCE

We cannot directly observe how people decideon the degree of persistence in the rate ofinflation (or other variables). However, we canmeasure some of the errors that contribute topermanent-transitory confusion and useeconometric methods to estimate the variance ofpermanent and transitory errors in the price level(or other variables). This section considers thesesources of evidence on the relative size ofpermanent and transitory changes.

Bullard (1994) reported the size of revisions toquarterly reports of the rate of change of theGNP deflator for the years 1986 through 1992.He found that the mean revision for the 28quarters was positive in this period; early reportsunderstated the rate of inflation and laterrevisions added additional amounts. Bullard alsoreports the range of revisions for each of the fourquarters following the period considered. Thereported ranges exclude the most extreme 5percent of the revisions in each tail. Table 1reproduces Bullard's results. (Table 1 omitted)

The average rate of inflation for the period isapproximately 3 percent. The range of revisions


(excluding extreme values) is 2.3 to 2.9percentage points. The revisions are from 70percent to 130 percent of the (3 percent) averagereported change. The relatively large size of therevisions suggests that it is rational for the publicto act as if the initial announcement is a verynoisy indicator of the true rate of inflation.

In several papers summarized in Brunner andMeltzer (1993), we report forecast errors for therate of change of output or inflation usingdifferent models, methods of forecast andcountries. The forecasting methods includestate-of-the-art econometric modeling,time-series analysis, judgment and combinationsof these methods. A rule-of-thumb summary isthat mean absolute errors for output growth inthe major industrial countries is 50 percent ormore of the average rate of change one quarter orone year ahead. Inflation is less variable thanoutput over short periods, and its forecast error isa smaller fraction of the average rate of change.Still, errors in both growth and inflation forecastsare large.

The data on forecast errors make a persuasivecase that forecasters frequently misperceivefuture values. Data on revisions suggest thatcurrent reports are subject to large errors. Errorsmay be unbiased, but that is a small consolationfor those who adjust to events that are found laternot to have occurred or those who do not adjustto changes that did occur. Thepermanent-transitory distinction implies thatpartial adjustment is optimal. Ex post, it willseem that adjustment has been sluggish. Indeed,as noted, errors may appear to be seriallycorrelated following large changes.

A filter can be used to separate permanent andtransitory components.(7) Let the error term in anaggregate such as the price level or output have apermanent and transitory component,

(4) epsilon sub t = epsilon sup p sub t + epsilonsup tau sub t ,

where p and tau denote the two components. Thetransitory component is white noise. Thepermanent component has the property of zeromean and constant variance, sigma sup 2 .

(Equation 5 omitted)

At any time t, the expected change in thepermanent component is zero.

People observe prices and output. As in themodel of the previous section, they cannotseparate the levels or changes of the permanentor transitory components. The rate at which theyadjust depends on the relative variances of thetransitory and permanent components of inflationor output growth. The larger the variance of thepermanent component relative to the variance ofthe transitory component, the quicker theeconomy adjusts to permanent changes. Arelatively large transitory variance slows theresponse.

Meltzer (1986) estimated the variances of thepermanent and transitory components of inflationand growth for Canada, Germany, the UnitedKingdom, and the United States under fixed andfluctuating exchange rates. The fixed exchangerate period runs from the first quarter of 1960through the third quarter of 1971. Fluctuatingrates begin in the fourth quarter of 1971 and endin the fourth quarter of 1984.(8)

Table 2 shows the ratios computed from thesedata using the adjustment equations in Muth(1960). (Table 2 omitted) The estimated varianceratios for inflation and growth changed underfluctuating rates, but the direction of change isnot uniform across countries. The length of theadjustment lag required to distinguish permanentfrom transitory shocks changed much less. Forinflation, seven of the eight values of theadjustment lag (lambda) lie between 0.41 and0.56. These values imply that, consistent withBlinder's survey data, about half of the shock toinflation is seen in the current quarter. Between


82 percent and 96 percent of the adjustment ofpermanent inflation occurs within four quartersof the initial shock to the inflation rate. Thisadjustment is faster than the two-year average lagcommonly suggested. For real growth, seven ofthe eight values of lambda lie between 0.45 and0.67; within four quarters, 91 percent to 99percent of the adjustment occurs. The speed ofoutput adjustment is broadly consistent with thelength of post-war recessions if these recessionsare interpreted as the cumulative adjustment setoff by a monetary or real shock. Inflation adjustsmore slowly than growth as many studies haveshown.

Clarida and Gali (1994) studied the response ofreal exchange rates to nominal shocks in fourcountries. For Canada and the United Kingdom,they were unable to find any structuraleffects--real or nominal--on real exchange rates.

For Germany and Japan, the evidence suggeststhat nominal shocks explain 45 percent and 34percent, respectively, of the four-quarter-aheadforecast error variances of the log level ofbilateral real dollar exchange rates. Clarida andGali note that their estimates are consistent withthe evidence from vector autoregressionsreported in Eichenbaum and Evans (1992).

Clarida and Gali (1994) use a trivariate vectorautoregression to estimate the transitorycomponent of real exchange rates.(9) Theirmodel includes shocks to aggregate demand andsupply. The authors report the ratio of thevariance of transitory shocks to the variance ofactual shocks. Using their data, we can computethe ratio of the variance of permanent shocks tothe variance of transitory shocks. The ratiocovers a wide range in these countries--from 0.42in Germany to 3.76 in Canada. These findingssuggest that most shocks to the Canadian realexchange rate have been permanent while mostshocks to the German real exchange rate weretransitory. Hence, the Canadian real exchangerate should adjust more rapidly to shocks than

the German real rate.

Meltzer (1993) used the permanent-transitorydistinction to model the U.S. multilateral realexchange rate. For both levels and firstdifferences under fixed and fluctuating ratesfrom 1960 to 1991, the data suggest that there isa large permanent component in the change ofthe real exchange rate and a significant transitorycomponent. Further, the data suggest that themultilateral real exchange rate responds tochanges in the nominal stock of money. Theeffect eventually vanishes, but monetary changeshave real effects until prices adjust. Thesefindings are consistent with short-runnon-neutrality and long-run neutrality of moneyif permanent changes in money were perceivedas transitory at the time they occurred orconversely.

The studies of prices, output and exchange ratessupport the principal arguments of the article.They are only a small part of the evidencesupporting ex post, short-run non-neutrality.They are of interest because they attribute slowadjustment of nominal values and real effects ofnominal shocks to the difficulty of discerning thepersistence of shocks. With a non-trivial cost ofacquiring information, price setting andpermanent-transitory confusion imply thatnominal changes have real effects that persist fora time.

Other recent studies find evidence of costs ofacquiring information. Investors frequently pay apremium to buy country-specific mutual funds.The premium implies that they could buy theindividual securities at lower cost. If they areuncertain about which securities to buy and whento buy or sell, it may be rational to pay for theservices of traders who specialize in theparticular market.

Smith (1991) uses costs of acquiring informationas one reason for the absence of optimal portfoliodiversification of world market securities. The


degree of diversification depends on costs ofacquiring information. People know much moreabout values and earnings in their own marketthan in foreign markets. Prices in foreign marketsmay reflect full information, but some investorseither do not have this information or cannotassess whether changes are permanent. Hence,they do not respond promptly to informationabout each of these markets. They do not holdthe "true" equilibrium portfolio they would holdif information costs were zero. A principal costin this case, as in others, is the interpretation ofavailable information. Permanent-transitoryconfusion is one part of the interpretationproblem.

In Fuhrer and Moore (1993), the inflation rate issticky. Firms adjust relative prices to the averageof other sectors' expected relative prices over thelife of existing contracts. Firms also adjust forthe current and expected level of output. Theautocorrelation functions based on their modelhave very similar shapes to the autocorrelationsgenerated by an unconstrained vectorautoregression. In particular, they showconsiderable persistence in inflation and outputmovements and sustained effects of inflation onreal output. In short, Fuhrer and Moore provideevidence that is consistent with a model in whichthere are costs of learning about the permanenceor persistence of changes, and in which peopleadopt strategies that leave room formisperception and real effects of nominalchanges.

Earlier work by Boschen and Grossman (1982),Gordon (1982) and Mishkin (1983) also provideevidence that supports sticky prices. Indeed, theevidence of gradual adjustment of prices and ofshort-term real effects of monetary change iscommon. These studies lack micro-foundations.Price-setting in part of the economy andpermanent-transitory confusion, as in Brunner,Cukierman and Meltzer (1983), reconciles thisevidence with rational behavior.

CONCLUSION

The examples in the preceding section are asmall part of the recently accumulated evidenceshowing that there is much more than casualobservation to support the main propositions inthis paper: Nominal prices adjust with a lag. Thelag is sufficiently long that real variables respondto nominal changes. Costs of acquiringinformation about the persistence of observedchanges--permanent-transitory confusion--is amain component of the cost and a main reasonfor the lag. Even where prices reflect informationfully, all individuals or firms may not haveadjusted fully to available but costly information.

The oft-repeated comment that macroeconomicsshould be built on micro-foundations is correct ifand only if the micro-foundations are appropriatefor the task. Standard micro theory, such asArrow-Debreu, imposes complete markets andmarket clearing in each market. There is no rolefor monetary disturbances. This is not theappropriate micro-foundation formacroeconomics. No amount of squeezing,cutting and pasting will make it so.

Rational behavior and rational expectations areentirely consistent with costs of acquiringinformation and the inability to fully identifypermanent and transitory shocks either when theyoccur or for several quarters after. Indeed, Muth's(1961) initial formulation of rationalexpectations is based on the latter distinction.

One alternative explanation of sticky prices inrecent literature relies on menu costs andimperfect competition. This explanation is afoundation for the so-called L-shaped supplycurve familiar from Keynesian theories. I showthat the evidence in Blinder's (1991) surveyrejects the L-shaped supply curve. Further, theimplications of monopolistic competition, suchas widespread excess capacity, do not explaingradual price adjustment in most serviceindustries.


Gordon (1990) proposes a disaggregated systemto take account of the information at many levelsof the economy. He argues that prices respond tomarginal cost but that marginal cost for any firmdepends on the pricing strategy of its suppliers.Hence, such information enters firms' decisionsabout price and output adjustment. Informationfrom macrodata is much less relevant.

This argument captures some of the dynamics ofpricing, but it poses unresolved challenges foraggregation over different industry structures.Moreover, Gordon's framework implies thatcommodity producers should adjust slowly toaggregate shocks and that large firms in durablegoods industries should adjust promptly. Thestylized facts suggest that the opposite is true.One reason is that organized commodity marketsincrease the information available to commodityproducers. There are no comparable markets forconsumers' and producers' durable goods output.Differences in information and the costs ofacquiring information are consistent with thestylized facts on speed of adjustment.

The micro-foundations suggested in this articleuse costs of acquiring information to explainthree common observations. First, many pricesare set. Second, price setters choose nominalvalues. Third, the daily, weekly, monthly orquarterly variances of "set prices" are smallfractions of the variance of prices in auctionmarkets; set prices are stickier.

Households and firms do not operate in a worldof full information. Incomplete information andcosts of acquiring information are centralproblems of a monetary economy. Informationand transaction costs explain why people holdand use money as a medium of exchange(Brunner and Meltzer, 1971). There isconsiderable evidence that these costs are nottrivial. The article cites revisions to reporteddata, forecast errors, incomplete informationabout costs, profits and strategies of competingfirms. These examples do not exhaust the costs

that firms and individuals face.

Some of these costs can be reduced byinstitutional and contractual arrangements. Thearrangements that people choose may be optimalwhen contracts are written but, with changes inthe environment, there are unforeseen gains andlosses. If the gains and losses are transitory, theirexpected value is zero. It may not be worthwhileto change the contract or the method ofcontracting. Once the change is consideredpersistent, gains and losses are expected tocumulate. Adjustment or re-negotiation becomesmore appealing to at least one party.

Information about permanent and transitorychanges in profits, prices, wages and othervariables is costly to acquire. The distribution ofshocks between real and nominal, permanent andtransitory may differ from one sample period tothe next. People learn to monitor events orchanges that are costly to ignore. But learningrequires a continuous process of monitoring bothwhat has happened and what should be observed.This is a basic problem for firms and households.As such, it is a more appropriatemicro-foundation for macroeconomics.

1 Beaulieu and Mattey (1994) estimate theeffects on price dispersion of factors such asadvertising, transport costs, and concentrationusing cross-section data for specificcommodities. They find effects of some of theseelements on price dispersion.

2 Ball and Mankiw (1994) use the ability togenerate a reverse L-shape (Keynesian) supplycurve as evidence for a model of menu costs andimperfect competition.

3 This should not suggest that the timing isconstant. As in Muth (1961), the timing of theresponse depends on the relative variances ofpermanent and transitory components. Seebelow.


4 Much of this section is taken from Meltzer(1982), and especially Brunner and Meltzer(1993). Additional examples of rational pricesetting when information is costly to acquire canbe found in these references. A recent paper byHeaton and Lucas (forthcoming) analyzes theeffects of transaction costs, such as bid-askspreads, investor heterogeneity and persistence ofshocks. They find that many puzzles in assetpricing can be reduced or removed by usingmodels with transaction costs and persistence.Balvers and Cosimano (1990) model priceadjustment when firms must learn about thepersistence of changes in demand.

5 See McCallum (1989) for a critical discussionof menu costs as a basis for rational price-levelstickiness.

6 Large Japanese firms that offer lifetimeemployment partially index by tying bonuses tosome measure of the firm's profitability. Withlifetime employment, it may be rational toassume that errors will (approximately) cancelover time. Lifetime employment contractsimpose costs on firms (owners) if output fallsbelow some expected value and impose costs onworkers if output is above the expected value.The latter can be compensated more readily byadjusting bonuses, but negative bonuses are notobserved. An unexpectedly long period of slowgrowth eliminated some of these contracts.

7 See Bomhoff (1983) for a discussion of theBayesian multi-state Kalman filter.

8 lambda is the adjustment lag given by

(equation omitted)

and a is the ratio of the variance of the permanentcomponent to the variance of the transitorycomponent.

9 The trivariate system has four lagged values ofthe change in the log real exchange rate, the

change in the log ratio of United States--homecountry real GDP, and the difference betweenU.S. and foreign inflation. Data are quarterlyfrom mid-1970 to the fourth quarter of 1992.

9 The trivariate system has four lagged values ofthe change in the log real exchange rate, thechange in the log ratio of United States--homecountry real GDP, and the difference betweenU.S. and foreign inflation. Data are quarterlyfrom mid-1970 to the fourth quarter of 1992.

REFERENCES

Alchian, Armen A. "Why Money?" Journal ofMoney, Credit and Banking (February 1977, part2), pp. 133-40.

Ball, Laurence, and N. Gregory Mankiw. "ASticky Price Manifesto," Carnegie RochesterConference Series on Public Policy (December1994), pp. 127-52.

Balvers, R. J., and T. F. Cosimano. "Activelylearning About Demand and the Dynamics ofPrice Adjustment," Economic Journal(September 1990) pp. 882-98.

Baumol, William J. "The Transactions Demandfor Cash: An Inventory Theoretic Approach,"Quarterly Journal of Economics (November1952), pp. 545-56.

Beaulieu, Joe., and Joe Mattey. "The Effects ofGeneral Inflation and Idiosyncratic Cost Shockson Within-Commodity Price Dispersion:Evidence from Microdata," Finance andEconomic Discussion Series No. 94-12 (May1994), Board of Governors of the FederalReserve System.

Blinder, Alan S. "Why Are Prices Sticky?Preliminary Results from an Interview Study,"The American Economic Review (May 1991),pp. 89-96.


Bomhoff, Edward J. Monetary Uncertainty.North Holland, 1983.

Boschen, John F., and Herschel I. Grossman."Tests of Equilibrium Macroeconomics UsingContemporaneous Monetary Data," Journal ofMonetary Economics (November 1982), pp.309-33.

Brunner, Karl, and Allan H. Meltzer. Money andthe Economy: Issues in Monetary Analysis.Cambridge University Press, for the RaffaeleMattiali Lectures, 1993.

-- and --. "The Uses of Money: Money in theTheory of an Exchange Economy," TheAmerican Economic Review (December 1971),pp. 184-805.

-- and --. "Economies of Scale in Cash BalancesReconsidered," Quarterly Journal of Economics(August 1967), pp. 422-36.

--, Alex Cukierman and Allan H. Meltzer."Money and Economic Activity: Inventories andBusiness Cycles," Journal of MonetaryEconomics (May 1983), pp. 281-319.

Bullard, James B. "How Reliable Are InflationReports?" Federal Reserve Bank of St. LouisMonetary Trend (February 1994), p. 1-56

Clarida, Richard, and Jordi Gali. "Sources ofReal Exchange Rate Fluctuations: HowImportant Are Nominal Shocks?" CarnegieRochester Conference Series on Public Policy(December 1994), pp. 1-56.

Eichenbaum, Martin, and Charles Evans. "SomeEmpirical Evidence on the Effects of MonetaryPolicy Shocks on Exchange Rates," FederalReserve Bank of Chicago Working Paper No.92-3 (December 1992).

Fuhrer, Jeffrey, and George Moore. "InflationPersistence," Finance and Economic Discussion

Series No. 93-17 (May 1993), Board ofGovernors of the Federal Reserve System.

Gordon, Robert J. "What Is New-KeynesianEconomics?" Journal of Economic Literature(September 1990), pp. 1115-11.

--. "Price Inertia and Policy Ineffectiveness in theUnited States, 1890-1980," Journal of PoliticalEconomy (December 1982), pp. 1087-1117.

Heaton, John, and Debra Lucas. "The Importanceof Investor Heterogeneity and Financial MarketImperfections for the Behavior of Asset Prices,"Carnegie Rochester Conference Series on PublicPolicy (forthcoming).

Knight, Frank H. Risk, Uncertainty and Profit(No. 16 in reprints of Scarce Classics inEconomics). London School of Economics,1933.

Lucas, Robert E., Jr. "Expectations and theNeutrality of Money," Journal of EconomicTheory (April 1972), pp. 103-24.

McCallum, Bennett T. "New ClassicalMacroeconomics: A Sympathetic Account,"Scandinavian Journal of Economics (June 1989),pp. 223-52.

Meltzer, Allan H. "Real Exchange Rates: SomeEvidence from the Postwar Years," this Review(March/April 1993), pp. 103-17.

--. "Size, Persistence and the Interrelation ofNominal and Real Shocks: Some Evidence fromFour Countries," Journal of Monetary Economics(January 1986), pp. 161-94.

--. "Rational Expectations, Risk, Uncertainty andMarket Responses," in Paul Wachtel, ed., Crisesin the Economic and Financial Structure.Lexington Books, 1982, pp. 3-22.

Mishkin, Fredrick S. A Rational Expectations


Approach to Macroeconometrics: Testing PolicyIneffectiveness and Efficient Markets Models.University of Chicago Press, 1983.

Muth, John F. "Rational Expectations and theTheory of Price Movements," Econometrica(July 1961), pp. 315-35.

--. "Optimal Properties of ExponentiallyWeighted Forecasts," Journal of the AmericanStatistical Association (June 1960), pp. 299-306.

Smith, Clifford W., Jr. "Globalization ofFinancial Markets," Carnegie RochesterConference Series on Public Policy (spring1991), pp. 77-96.

Allan H. Meltzer is a professor of politicaleconomy and public policy at Carnegie MellonUniversity. An earlier version of this article waspresented at the 1994 Konstanz Seminar in InselReichenau, Germany, and some material isincluded from earlier work with Karl Brunnerand Alex Cukierman. The author thanks DaleHenderson and Bennett McCallum for commentson earlier drafts.

Reproduced with permission of the copyrightowner. Further reproduction or distribution isprohibited without permission.

Documents

Information, Sticky Prices and Macroeconomic …frontiers.sauder.ubc.ca/Meltzer_Sticky.pdfFile: Meltzer_Sticky.wpd 3 April 10, 2002 15:52 As Keynes recognized, a principal missing