<<

Journalof Economic Perspectives—Volume 16,Number 4—Fall 2002—Pages 115– 136

TheNAIRU inTheory and Practice

LaurenceBall and N.GregoryMankiw

AIRU stands forthe nonaccelerating in ation rateof .It isbeyond disputethat this acronym isan uglyaddition tothe English N language. Thereare, however, two issues that failto command consensus among , which weaddress inthis essay. TheŽ rstissue is whetherthe concept ofthe NAIRU isa usefulpiece of business cycletheory. We believeit is, and webegin this paperby attempting to explain why. In ourview, the NAIRU isapproximately a synonym forthe natural rateof unemployment.This concept followsnaturally from any theorythat says that changes inmonetarypolicy, and aggregatedemand moregenerally, push ination and unemploymentin opposite directions in the short run. Once this short-run tradeoffis admitted, there must besome level of unemployment consistent with stablein ation. Thesecond issueis why the NAIRU changes overtime and, inparticular,why itfellin thesecond half ofthe1990s. This questionis more difŽ cult, and theanswer isopen todebate. Most likely,various factors areat work,including demographics and governmentpolicies. Yet one hypothesis stands out as particularlypromising: uctuations inthe NAIRU appear relatedto  uctuations inproductivity. In the 1970s,the NAIRU rosewhen productivitygrowth slowed. In the1990s, the NAIRU fellwhen productivitygrowth sped up. Developingand testingmodels that explain thelinks among ination, unemploymentand productivityremains a challengefor students ofbusiness cycletheory.

y Laurence Ball isProfessorof ,Johns Hopkins University, Baltimore, Maryland. N.Gregory Mankiwis Professor of Economics, Harvard University, Cambridge, Massachusetts. 116Journal of Economic Perspectives

The Role of NAIRU

The word “NAIRU” enteredthe language of macroeconomicsin the1970s, a periodof rapid and risingin  ation. Yet,in a deepersense, theconcept has been thereall along.

ABuildingBlock of Macroeconomic Theory Alongtradition in economicsemphasizes that thesupply ofmoney in  uences both in ation and unemployment.In his classic 1752essay “Of ,” David Humewrote about theeffects of monetary injections, such as golddiscoveries: “It iseasyto tracethe money in itsprogress through thewhole commonwealth; where we shall Ž nd that itmust Ž rstquicken the diligence of every individual, before it increasesthe of labour. ” This insighthas motivatedmuch ofmodern macro- economictheory. Two prominent examples are ’s(1968)presi- dentialaddress tothe American Economic Association and RobertLucas ’s (1996) Nobelprize lecture. Lucas quotesexactly these words from Hume. Attimes,some economists have questionedHume ’sinsight. Thereal business cycletheorists of the 1980s, for example, suggested that business cycleswere technologicallydriven and that moneyhad no rolein explaining production and employment  uctuations (Prescott,1986; Long and Plosser,1983). But this viewis aminorityposition, both historicallyand today. Thereis wideagreement about the fundamental insightthat monetary  uctuations push in  ation and unemployment inopposite directions. That is, societyfaces atradeoff,at leastin the short run, between in ation and unemployment. Accordingto conventional macroeconomic theory, the in  ation-unemployment tradeoffis centralto understanding not onlythe effects of but also otherpolicies and eventsthat in  uencethe forgoods and services.But most of these other events and policiescan potentiallyhave effects through otherchannels as well.For example, tax policy in  uences both aggregate demand through disposableincome and aggregatesupply through workincentives. Bycontrast, beliefthat monetarypolicy has employmenteffects is inextricably tied tobelief in the in  ation-unemploymenttradeoff. Twocenturies have passed sinceHume penned thewise words quoted above, but theeconomics profession has yetto reacha consensus about whythis tradeoff arises.In classical theory,money is neutral. It isonlythe numeraire in which arequoted. Changes inits quantity should affectthe overall , but not relativeprices, production oremployment. The key question facing theoristsis why this classical theoremof monetary neutrality fails to hold inthe world. Many answers have beenproposed. Short-run nonneutralityhas beenblamed on imperfectionsof information (Friedman, 1968; Lucas, 1973;Mankiw and Reis, 2001);long-term labor contracts (Fischer,1977; Gray, 1976; Taylor, 1980); the costs ofprice adjustment (Rotemberg,1982; Mankiw, 1985;Blanchard and Kiyotaki, Laurence Ball andN. Gregory Mankiw117

1987;Ball and Romer,1990); or departures from full rationality (Akerlof and Yellen,1985). Each ofthese approaches raisesits own setof dif Ž culttheoretical and empiricalquestions, which arebeyond thescope ofthis essay. Thereis, however,a common theme:because ofsome imperfection absent fromthe classical model,changes inthe of the unit ofaccount matter.Monetary neutrality breaksdown, and at leastin theshort run, monetarychanges have oppositeeffects on in ation and unemployment. Without much loss ofgenerality, we can writethe short-run tradeoffbetween in ation and unemployment U as follows:

5 k 2 aU, where k and a . 0areparameters. This equationdoes not reallysay much, other than that and U arenegatively related. One fact about this relationshipis clear: itcannot beconstant overtime. If itwere,the data onin  ation and unemployment wouldtrace a nice,stable, downward-sloping Phillipscurve. There once was atime when someeconomists took this possibilityseriously, but data sincethe early 1970s have madethis simpleview untenable. Theinstability of this relationshipis hardly a surprise.Even Samuelson and Solow’s(1960)classic discussion ofthe suggested that theshort-run menu of in ation-unemploymentcombinations wouldlikely shift over time. Skep- ticsare sometimes tempted to usethe shifting Phillips curve as evidenceto denythe existenceof a short-run tradeoff.This ispuresophistry. It wouldbe like observing that theUnited States has moreconsumption and investmentthan does India to denythat societyfaces atradeoffbetween consumption and .The situation isnot hard tounderstand and, infact, arisesfrequently in economics. At any point intime,society faces atradeoff,but thetradeoff changes overtime. The nextquestion is what factors cause thetradeoff to shift.

Expectations,the Natural Rate and SupplyShocks SinceFriedman (1968) and Phelps ’s(1967,1968) seminal contributions, one variablehas playedcenter stage in explainingshifts inthe in  ation-unemployment tradeoff:expected in  ation. Otherthings equal,an increasein expectedin  ation isassociated withan equalincrease in actual in  ation. Thereason whyexpected in ation plays such aroledepends on thetheory of short-run nonneutrality. Moreover,the choice of theorywill in  uencethe timing of when expectationsare formed.But from a bird ’s-eyeview, the similarity of thetheories is moresigni Ž cant than theirdifferences. In moststandard theories,we can writethe in  ation- unemploymenttradeoff as

5 e 2 a~U 2 U*!, where e isexpected in  ation and U*isa parametercalled the “natural rateof 118Journal of Economic Perspectives

unemployment. ” Thenatural rateis the rate of unemploymentthat prevailswhen in ation expectationsare con Ž rmed.Seen in another light,the parameter U* embedsall shifts inthein  ation-unemploymenttradeoff previously represented by theparameter k,otherthan shifts arisingfrom expected in  ation. Thenatural ratecan beviewed as theunemployment rate that theeconomy reachesin the long run. This interpretationarises from imposing a modicumof rationalityto expectations. Over any longinterval of time,the average of expected in ation should equalthe average of actual in  ation; otherwise,forecasts are systematicallybiased. Thus, overthe same long interval, average unemployment should equalthe average natural rate.In thelong run, U cannot deviatefrom U*. Noneof this means that thenatural rateof unemployment is immutable, or eventhat itmoves only slowly over time. In principle, U*can exhibitsubstantial high- frequencyvariation, so any othershift in thein  ation-unemploymenttradeoff can bedescribed as ashiftin U*.As apracticalmatter, however, the literature on in ation-unemploymentdynamics has traditionallyused an amendedversion of the aboveequation:

5 e 2 a~U 2 U*! 1 v, where v isdubbed the “. ” Tosomeextent, the distinction between U* and v isarbitrary:both thenatural rate U*and thesupply shock v representshifts inthe in  ation-unemployment tradeoff.But many economistsview these two variables as measuringdifferent kinds ofshifts. Thenatural rate U*isthought tore  ecthow wellthe labor market matches workersand jobs. It isaltered,for instance, bychanges indemographicsor labor-marketinstitutions and isthought tomove slowly over time. By contrast, the supply shock v re ectsdisruptions inthe normal in  ation process, such as that caused byan oilembargo or a change inthe exchange rate. The supply shock is thought toexhibit more high-frequency variation than thenatural rate. 1 Toimplementthis equation, somethinghas tobe said about how expectations areformed. One approach istoassume adaptiveexpectations, according towhich expected in ation isaweightedaverage of past in  ation. Thesimplest version is to e positthat expectedin  ation equalslast period ’s in ation: 5 2 1 . The in ation- unemploymenttradeoff then becomes

5 21 2 a~U 2 U*! 1 v.

Therational expectations revolution was founded preciselyon criticizingthis approach (Lucas, 1972;Sargent, 1971). And surely,it would be indefensible to accept adaptiveexpectations as apreciseand immutabledescription of theworld,

1 Forour attempt toderive a theoryof the supplyshock v,seeBall and Mankiw(1995). TheNAIRU inTheory and Practice 119

regardlessof the monetary regime. But over the past fourdecades, theassumption ofadaptive expectations may not beso bad. In  ation has beenclose to a random walkduring this period(Barsky, 1987; Ball, 2000). In such asetting,forecasting future in ation withpast in  ation, as isassumed byadaptiveexpectations, is not far fromrational. In this environment, U*can beviewed as theNAIRU, theunem- ploymentrate at which in  ation willbe stable, absent thehigh-frequency shocks representedby v. One implicationof this analysis isthat thevalue of the NAIRU concept depends on themonetary regime. If welived in a worldwhere in  ation was close towhite noise, ratherthan highlypersistent, then adaptiveexpectations would be abad approximationto optimalbehavior. The early part ofthe twentieth century, when theUnited States operated under a goldstandard, mayhave beensuch a regime(Barsky, 1987). In that world,expected in  ation wouldbe closer to a constant ofzero, and thenatural rate U*wouldbe associated withstable prices ratherthan stablein  ation. In theU.S. monetaryregime of recentdecades, however,the NAIRU concept isuseful,and itissynonymous withthe natural rateof unemployment.In his classic paperintroducing the natural ratehypothesis, Friedman(1968) described the situation as follows: “Thereis always a temporarytradeoff between in  ation and unemployment;there is no permanenttradeoff. The temporary tradeoff comes not from in ation perse, but fromunanticipated in  ation, which generallymeans, froma risingrate of in  ation.” Friedmandidn ’tusethe term “NAIRU,” but the concept isimplicit in his analysis.

Hysteresis Someeconomists have suggestedthat thelabor market exhibits a formof “” (Blanchard and Summers,1986). In physics, hysteresisrefers to the failureof an objectto return to itsoriginal value after being changed byan external force,even after the force is removed.In thelabor market, a similarphenomenon wouldarise if the natural rate U*dependedon past unemployment U.In this case, achange inaggregatedemand would Ž rst in uenceunemployment by causing U todeviate from U*,but then wouldhave apersistenteffect on unemploymentas U* changed. Severaltheories have beenproposed toexplainwhy this mightbe the case. The mostpopular emphasizelong-lasting damage suffered by workerswho experience unemployment.These workers lose human capital, becomeless attractive to em- ployersand reducetheir job search as theybecome accustomed tobeing unem- ployed(Layard, Nickelland Jackman, 1991).All these effects make workers less likelyto be employed in the future. A recessionthat raisesunemployment leaves a permanentscar on theeconomy, as U*ishigher even after the initial shock that caused therecession has disappeared. Thesetheories of hysteresis were Ž rst devel- opedto explain the large rise in the NAIRU inEurope during the 1980s: The 120Journal of Economic Perspectives

increase in U*cameimmediately after the disin  ationaryrecession that startedthe decade.2 Thevalidity of hysteresis theories is a subject ofsome controversy, and wewill not takeup that debatehere. Regardless of how this debateis resolved,the concept ofNAIRU remainsvalid. At any point intime, there will be an unemploymentrate consistent withstable in  ation, which can becalledthe NAIRU. Hysteresistheories merelygive one reason toexpect the NAIRU tochange overtime. As wediscuss below,there are many otherreasons toexpect that theNAIRU willnot bea constant.

Two EconometricDifŽ culties Letusnow turn fromtheory to econometric implementation. A largeliterature has attemptedto estimate in  ation equations ofthis form:

5 21 2 a~U 2 U*! 1 v.

Often, thestudies include additional lagsof in  ation orunemployment. Some- times,rather than leavingthe supply shock v entirelyin aresidual,control variables areincluded, such as food and oilprices, exchange rates and dummiesfor - pricecontrols. 3 One difŽ cultissue that this literaturehas triedto skirt is the identi Ž cation problem.If themacroeconometrician assumes that U*isconstant overthe interval beingstudied and that v iscontemporaneously uncorrelated with U, then this equation can beconsistentlyestimated with ordinary least squares. Thevalue of the NAIRU, U*,can then beinferred from the estimated parameters. These identi Ž - cation assumptions arenot at allinnocuous. It iseasy to imagine that thesupply shocks representedby v arecorrelated with unemployment. For example, a burst inproductivitygrowth, such as that experiencedduring the late 1990s, might well lower in ation and unemployment.The textbook solution to this problemis to Ž nd instrumentalvariables that arecorrelated with unemployment but uncorrelated withthe supply shock. In practice, Ž nding validinstruments is hard todo and rarely done. Notethat otherstrands ofthe literature make somewhat different identi Ž ca- tionassumptions. Lucas ’s(1973)classic paperon in  ation-output tradeoffsused nominal GDPgrowthas theright-hand-side variablein aregressionestimated with ordinaryleast squares. Theimplicit assumption was that thesupply shocks inthe residualdo not in  uencenominal GDP, but can in  uenceboth realGDP and the pricelevel in opposite directions. Similarly, Barro ’s(1977)classic workon unan- ticipatedmoney implicitly assumed that supply shocks do not in  uence money

2 Fora recentstudy usinghysteresis theories to explain the increasein the EuropeanNAIRU, seeBall (1999). Foran attempt toexplain the EuropeanNAIRU basedon labormarket institutions and supply sideshocks, seeBlanchard and Wolfers(2000). 3 Fortwo examples from this largeliterature, see Gordon (1998) and Staiger,Stock and Watson (1997). Laurence Ball andN. Gregory Mankiw121

growth.These identi Ž cation schemescan also bequestioned.Below we follow the traditionalidenti Ž cation assumption, according towhich thesupply shock v is contemporaneouslyuncorrelated with unemployment U.Dealingwith the identi- Ž cation problemin amoresatisfactory way seems an importantavenue for future research. Asecond, moretractable econometric issue is the computation ofstandard errors.Until recently, the empirical literature on thePhillips curve rarely provided standard errorsfor estimates of theNAIRU. This odd oversightwas correctedin an importantpaper by Staiger, Stock and Watson (1997).Using a conventional speciŽ cation, theyestimated the NAIRU in1990 to be 6.2 percent, with a 95per- cent conŽ denceinterval from 5.1 to 7.7percent. This isalargerange. In principle, bettermeasures of supply shocks can reducethe residual variance and improvethe precisionof NAIRU estimates.But Staiger, Stock and Watson showed that given standard speci Ž cations used inthe literature, the NAIRU isnot estimatedprecisely.

ItsUse in Policy How should monetarypolicymakers use the NAIRU? Most obviously,it is a forecastingtool. When unemploymentis below the NAIRU, in  ation can be expectedto rise, and when itisabovethe NAIRU, in  ation can beexpected to fall. Thus, evenif the policy regime were one of in  ation targeting,monetary policy- makersshould keepan eyeon unemploymentand theNAIRU. It maybe tempting to point totheexperience of the 1990s to suggest that this viewis obsolete.Indeed, as wediscuss below,there is evidence that thelate 1990s weredifferent: the NAIRU declinedsubstantially. Butit would be rash tosuggest that theNAIRU isobsolete as aforecastingtool. Stock and Watson (1999)offer a comprehensivestudy ofvarious methods forforecasting in  ation. Despitethe Ž nding ofStaiger, Stock and Watson (1997)that theNAIRU isimprecisely esti- mated, Stockand Watson (1999)report, “In ation forecastsproduced bythe Phillipscurve generally have beenmore accurate than forecastsbased on other macroeconomicvariables, including rates, money, and commodityprices. ” Nonetheless,it also makessense for monetary policymakers to give some weightto other forecasting tools. When lookingahead tofuture in  ation, they should also lookat, forexample, the consensus ofprivate forecasters and the spread betweenreal and nominalbond yields.Of course, thesetools themselves re ectthe NAIRU concept, because privateforecasts of in  ation areoften based on it.Using such privateforecasts of in  ation forpolicymaking can beviewed as away todecentralize the decision making over how theNAIRU ischanging overtime.

The U.S. NAIRU, 1960 –2000

Somuch fortheory. Let ’snow turn tothepractical question: what isthelevel ofthe NAIRU forthe U.S. economy? 122Journal of Economic Perspectives

An Approach Toseehow onemight estimate the NAIRU, rewritethe Phillips curve equation as

D 5 aU* 2 aU 1 v.

If oneassumes that U*isconstant and that U isuncorrelatedwith v,then thevalue of U*can beestimatedby regressing the change inin  ation D onaconstant and unemployment U.Theratio of the constant term( aU*)totheabsolute value of the unemploymentcoef Ž cient (a)isan estimateof U*.When weperformthis exercise forannual U.S. data from1960 to 2000, measuring in  ation withthe consumer priceindex, we obtain aconstant termof 3.8and an unemploymentcoef Ž cient of 20.63.This yieldsa NAIRU estimateof 6.1 percent. However,many economistshave questionedthe assumption ofa constant NAIRU underlyingthis calculation, especiallysince the apparent fallin the NAIRU inthelate 1990s. There is agrowingliterature that seeksto estimatethe path ofa time-varyingNAIRU. This literatureis based on theidea, discussed above, that movementsin U*arelong-term shifts inthe unemployment-in  ation relation, whilethe shock v captures short-run  uctuations. Authors such as Staiger,Stock and Watson (1997)and Gordon (1998)estimate U*bypositing a stochastic process for U*(such as arandom walk)and astochastic process for v (such as whitenoise) and then using astatisticalprocedure that separatesshifts ofthe Phillips curve into thesetwo kinds ofshocks. To buildintuition, we usean approach that issimplerbut yieldssimilar results. Suppose forthe moment that weknow the value of the parameter a, which givesthe slope of the unemployment-in  ation tradeoff.We can then rearrangeto obtain theequation

U* 1 v/a 5 U 1 D /a.

Theright-hand sidecan becomputed fromthe data, yieldingan estimateof U* 1 v/a,which measuresthe shifts inthePhillips curve. Within this sum, U*represents thelonger-term trends, and v/a isproportional to theshorter-term supply shocks. It istherefore natural totry to extract U* from U* 1 v/a using astandard approach toestimating the trend in aseries. Weusethe Hodrick-Prescott Ž lter(Hodrick and Prescott,1997). The HP Ž lter isageneralizationof alineartime trend that allowsthe slope of the trend to change graduallyover time. Formally, the HP Ž lterminimizes the sum ofsquared devia- tions betweenthe trend and theactual series,with a penaltyfor curvature that keepsthe trend smooth. If therewere no penalty, the Ž lterwould yield the original series;if the penalty were very high, itwould yield a lineartime trend. Toimplementthis procedure,we must choose twoparameters. The Ž rst is the Phillipscurve slope, a.In ourresults below, we use an a of0.63, the slope coefŽ cientobtained from regressing D on unemploymentand aconstant. This TheNAIRU inTheory and Practice 123

valueis consistent withconventional wisdomabout thecosts ofdisin  ation (it impliesthat reducingin  ation byone percentage point produces 1/0.63 5 1.6point-years of unemployment). Reasonable variation in the assumed coef Ž cient has littleeffect on ourconclusions. Theother parameter is thesmoothing parameterin the HP Ž lter—the weight that theprocedure gives to keeping the estimated U*smooth ratherthan Ž tting everymovement in U* 1 (v/a).Thechoice of this parameteris largelyarbitrary. In someways, this isnot surprising:as wenoted earlier,the distinction between U* and v isnot wellde Ž ned. Most economistshave theintuition that movementsin U* are “smooth” and that v representsa differentkind of high-frequency shift in the Phillipscurve, but this intuitionis too vague to have much practicalimport. In theanalysis below,we experiment with alternative values of the HP smoothing parameter.

Results Figure1 presentsestimates of theU.S. NAIRU overthe last 40 years. The solid linegives the values of U* 1 (v/a)computed as describedabove; this represents thesum oflong-termand transitoryshifts inthein  ation-unemploymenttradeoff. Thetwo dashed linesgive smoothed versionsof the series that serveas ourestimates of U*.Thetwo versions correspond todifferent values of the HP smoothing parameter:one value is 100,the most commonly used valuewith annual data, and theother is 1000, which imposesgreater smoothing as advocated bysome research- ers(for example, Roberts, 1998). Thetwo smoothed seriestell broadly similar stories. The NAIRU has followed ahump-shaped path: ittrendedup fromthe 1960s until about 1980,then peaked and has declinedsince then. Withthe smaller smoothing parameter, there is a smalldip in theearly 1960s before U*starts torise, but this wiggledoes not survive withgreater smoothing. Moregenerally, the movements in U*aresmaller with thehigher HP parameter.With a parameterof 1000, the estimated NAIRU is 5.4percent in 1960, peaks at 6.8percent in 1979and fallsto 4.9 percent in 2000. Theseresults are broadly similar to those ofGordon (1998)and Staiger,Stock and Watson (2001).The apparent increasein the NAIRU before1980 and decline thereafterhas beenwidely recognized. These movements have motivatedpapers withtitles such as “Why isUnemployment So Very High NearFull Employment? ” inthe1980s (Summers, 1986) and “Why istheU.S. UnemploymentRate So Much Lower?” morerecently (Shimer, 1999). Whilethere is aconsensus that theNAIRU fellduring the 1980s and 1990s,this consensus tooksome time to develop. The falling NAIRU was initiallyobscured by therun-up ofactual unemploymentin the of the early 1990s. Starting in themid-1990s, many authors pointedout arun offavorable shifts inthe Phillips curve,but thesewere sometimes interpreted as transitorysupply shocks —that is, decreasesin U* 1 (v/a)wereinterpreted as movementsin v rather than U*. This interpretationwas supported bydirect evidence of favorable shocks duringthe period 1995–1998,such as afallin energy prices and astrengtheningof the 124Journal of Economic Perspectives

Figure 1 TimeVarying NAIRUs, 1960 –2000

exchangerate, which reducedimport prices (for example, Gordon, 1998).Yet the periodafter 1998 did not seeadditional favorableshocks, and indeed,energy prices movedback up. Becauseunemployment was lowthrough 2000without accelerating in ation, aconsensus emergedthat theNAIRU had fallen. On theother hand, themagnitude of theNAIRU decreaseis hard toestimate. As illustratedabove, itdepends on an arbitrarydecision about how much tosmooth theNAIRU series.The precise timing of movements in theNAIRU isalso unclear. Our estimatedmovements are smooth, withthe decrease occurring slowly over almosttwo decades. Yetthis isan artifactof oursmoothing procedure. A number ofauthors have suggestedthat theNAIRU was fairlyconstant fromthe 1980s to the mid-1990sand then fellsharply inthe late 1990s “NewEconomy. ” Perhaps this is true,and ourprocedure arti Ž ciallysmoothes out thefall in U*.Thereare limits to how much onecan learnabout theNAIRU fromunemployment and in  ation data alone.

The Falling NAIRU: AMore Employable Labor Force?

Many authors have sought toexplainthe movements in the U.S. NAIRU. This sectionand thenext review some of theleading hypotheses, witha focus on those that mightexplain the declining NAIRU ofthe 1990s. Some of these theories also helpexplain the earlier NAIRU increase. Webegin in this sectionby reviewing stories that focus on thechanging compositionof the labor force. Economists have longrecognized that unemploy- mentrates are different for different kinds ofworkers,depending, forexample, on Laurence Ball andN. Gregory Mankiw125

theirskills and theirintensity of job search. Thus, changes inthe sizes of groups withrelatively high orlow rates of unemployment can change theaggregate unemploymentrate, even without changes intherate for any individualgroup. In recentyears, a numberof authors havesuggested changes inthe labor force that reduceaggregate unemployment by reducing the sizes of high-unemployment groups.

OlderWorkers Themost obvious reason thelabor force changes isdemographics. In seeking toexplain the evolution of the NAIRU, anumberof authors point toaparticular typeof shift: thechanging agestructure as thebaby boomgeneration has moved through thelabor force. The proportion of the labor force aged 16 –24 rose from 17percent in 1960 to 24 percent in 1978 as thebaby boomersentered the labor forceas young workers,and this percentagefell to 16 percent in 2000 as the boomershave aged. Thesetrends are potentially important because young workers have higherunemployment rates than olderworkers: over 1960 –2000,the average unemploymentrate was 12.2percent for workers 16 –24and 4.4percent for workers 251.Gordon (1998)has arguedthat theincrease in young workersaccounts for much oftheincrease in the NAIRU before1980, and Shimer(1999) argues that the recentdecrease explains much ofthe NAIRU fall. Theclassic methodfor measuring the effects of demographic changes isto compute a “Perry-weighted ” unemploymentrate (Perry, 1970; Katz and Krueger, 1999).This isaweightedaverage of unemployment rates for different demographic groups with Ž xedweights; by contrast, theusual aggregateunemployment rate has weightsequal to labor-force shares, which change overtime. A timeseries for Perry-weightedunemployment shows what wouldhave happened totheunemploy- mentrate given the evolution of each group ’sunemploymentif thesizes of groups didnot change. FollowingStaiger, Stock and Watson (2001),we compute Perry-weighted unemploymentbased on 14age-sex groups, withweights based on averagelabor forceshares over1960 –2000.We then computeour estimates of the time-varying NAIRU fromunemployment and in  ation data using thesame method as in Figure 1—but using thePerry-weighted unemployment series. Figure 2 shows the resultingseries (based on an HPsmoothing parameterof 1000), along withthe correspondingseries based on thestandard unemploymentrate; the differences betweenthe two series show theimpact of demographics. Figure2 shows that this impacthas beenmodest. Thehump-shaped patternof the NAIRU remainsafter Perry-weighting,although itis dampened: theincrease from 1960 to the peak and the decreaseto 2000 are 0.9 and 1.3percentage points, respectively,compared with 1.4 and 1.9percentage points withthe standard unemploymentrate. Thus, thebroad trendsin theNAIRU remain to be explained even after one adjusts fordemographics. 4

4 ThePerry-weighting procedure assumes that demographicsaffect laborforce shares but not the unemploymentrates ofindividual groups. This assumptionhas beenquestioned by Shimer (1999, 2001), whodiscusses a numberof channels through which changing suppliesof old and youngworkers 126Journal of Economic Perspectives

Figure 2 TheEffects of Demographic Adjustment

Disabilityand Incarceration Thelabor force can also change ifgovernmentpolicies cause peopleto leave it.The aggregate unemployment rate falls if the labor-force leavers are workers who otherwisewould have high unemploymentrates. Recent work has noted twopolicy shifts that workin this direction:the rising rate of incarceration (Katz and Krueger, 1999)and thegreater generosity of disability insurance (Autorand Duggan, 2001). Peoplewho areremoved from the labor force by being locked up orthrough certiŽ cation ofdisability are likely to have experiencedhigh unemploymentrates whilein the labor force. Of thesetwo factors, disabilityappears moreimportant. The percentage of nonelderlyadults receivinggovernment disability insurance has risensteadily from 3.1percent in 1984to 5.3 percent in 2000.Autor and Duggan (2001)attribute this riseto reduced stringency in the screening of applicants and toa higherincome replacementratio. They estimate the impact on unemploymentby examiningthe effectsof variation in thedisability program across states. They Ž nd that thetotal effectof changes inthe program has beento reduceaggregate unemployment by 0.65percentage points from1984 to 2000. Katzand Krueger(1999) have observedthat lowerunemployment can re  ect greaterincarceration. However, while incarceration rates rose dramatically in the 1990s,the effect on aggregateunemployment was modest. Katzand Krueger

can affect theirunemployment rates. Shimer ’s1999paper argues that ayoungerlabor force raises unemploymentamong the young, buthis 2001paper argues that itreducesunemployment for both age groups.If the laterpaper is correct, then differencesbetween Perry-weighted and standard unemploy- mentrates givean upperbound on the effects ofdemographics. TheNAIRU inTheory and Practice 127

estimatethat this factorproduced atotaldecrease in unemployment of 0.17 per- centagepoints. Adding theeffects of disability and incarcerationyields a totalreduction in unemploymentof roughly 0.8 percentage points. This isabitmore than half ofthe decreasein the NAIRU inFigure 2 when unemploymentis Perry-weighted. How- ever,recall that theestimated fall in the NAIRU islarger if the HP smoothing parameteris set lower than 1000,in which case disabilityand incarcerationcan explaina smallerfraction of the decline. Thelikely role of a changing laborforce in explaining the NAIRU decrease depends on thetiming of the decrease. As wediscussed above, someeconomists suggestthat theNAIRU fellsharply since1995, although theaggregate data arealso consistent witha gradualdecrease since the early 1980s. If therewas infact asharp shiftfrom 1995 to 2000, the factors discussed so farcannot bethe main explana- tion. Thechanges indisabilitybene Ž tsand incarcerationare long-term trends, and onlya smallpart ofthe changes have occurredafter 1995. The aging of the labor forcewas almostcomplete by 1995:the percentage aged 16 –24reacheda troughof 15.8percent in 1997 and has sincerisen slightly. The difference between Perry- weightedand standard unemploymentrates fell only 0.2 points from1992 to 2000. If theNAIRU fellsigni Ž cantly inthe late 1990s, we must lookbeyond thenature of thelabor force to Ž nd theexplanation. This bringsus toanother setof theories.

The Falling NAIRU: ANew Economy?

TheNAIRU can change not onlybecause ofchanges inthe labor force, but also becauseof broaderchanges intheeconomy. In thesecond half ofthe 1990s, many observersalleged the advent ofa “new economy”— onewith new technolo- gies,higher productivity growth, increased “competitiveness ” and so on. If one believesthat theNAIRU fellsigni Ž cantly inthe period after 1995, it is natural to suspect alinkbetween this fact and thebroader changes intheeconomy. Wenow discuss severalleading stories along these lines. 5

GreaterOpenness to One storyabout thefavorable Phillips curve shift is that itresulted from the “globalization ” ofthe U.S. economy —thegreater openness toforeign trade (for example,Thurow, 1998).This argumentstarts withthe fact that foreigntrade has becomemore important in the United States in recent decades: theratios of

5 In additionto the storieswe mention, some people have noteddeclines in unionizationand the real minimumwage and welfarereform. There appears to be aconsensus,however, that theseare not major factors. Changesin unionization and minimumwages were modest in the 1990s.Welfare reform affected asizablenumber of workers —roughlyone million women have left the welfarerolls since 1994. However,most of thesewomen were out of the laborforce while on welfare.By joiningthe laborforce, they arelikely to have raised the unemploymentrate slightly, becausethe incidenceof unemployment ishigher for them than forthe averageworker. 128Journal of Economic Perspectives

importsand exportsto GDP have trendedup. Someobservers argue that this integrationinto the world economy has subjected U.S. Ž rmsto greater competi- tion. This inturn isanti-in  ationary: evenif unemployment is low, Ž rms cannot raiseprices aggressively because consumers willswitch toforeign suppliers. In mainstreamterminology, this means that theNAIRU has fallen. Many journalists have pickedup on this idea,but ithas largelybeen ignored byacademic economists. And theyhave ignoredit, we believe, for good reason. The theoreticallogic of the story is questionable, but themain problemis empirical. TheUnited States has becomemore open inthe last decade, withthe import-GDP ratiorising from 11 percentin 1990to 15percent in 2000.But this isnot afeature ofa neweconomy, but rathera continuation ofa trendthrough mostof theperiod sinceWorld War II. Theimport-GDP ratio was 5percentin 1950. If greater openness produces lowerunemployment, we should have seena steady downward trendin the NAIRU for the last 50years, and this hasn ’toccurred.Indeed, the decadewith the largest increase in the import-GDP ratio was the1970s, and as shown inFigure 1, this decadesaw asubstantial increase inthe NAIRU.

BetterJob Matching One reason forunemployment is job turnover.When workersmove from jobs that disappear tothose that open up, theprocess createsunemployment because it takesworkers time to Ž nd newjobs. Severalauthors suggestthat this process improvedin the1990s, leading to lower unemployment. Themost common versionof this storyfocuses on thegrowth in thetemporary helpindustry (for example, Katz and Krueger,1999; Cohen, Dickensand Posen, 2001).The percentage of workers employed by temporary help Ž rms, such as ManpowerInc., doubledfrom 1.1 percent in 1989 to 2.2 percent in 1998. This suggeststhat an increasingnumber of workerswho arebetween permanent jobs are employedas tempsrather than unemployed.In addition, tempjobs sometimesturn intopermanent jobs, so tempagencies help speed up theprocess ofpermanent job matching. However,when researcherstry to quantify theeffects of temp agencies on unemployment,the results are disappointing. Both Katzand Krueger(1999) and Staiger,Stock and Watson (2001)examine the relation across states between unemploymentand thesize of the temp industry. Katzand Kruegerestimate that thegrowth of the temp industry in the1990s reduced aggregate unemployment by anywherefrom zero to 0.4 percentage points. Staiger,Stock and Watson failto Ž nd arobust relationbetween the temp industry and unemploymentrates. Thus, stories about thefalling NAIRU based on thetemp industry remain speculative at best. It ispossible that theprocess ofjob matching improvedin ways beyond the growthof the temp industry. Cohen, Dickensand Posen (2001)suggest that the neweconomy features production processesthat put agreateremphasis ongeneral ratherthan speci Ž cskills.As aresult,workers have becomemore interchangeable, makingit easier to match workersand jobs and therebyreducing unemployment. As evidencefor this idea,Cohen, Dickensand Posen citethe management litera- Laurence Ball andN. Gregory Mankiw129

tureand interviewswith human resourcemanagers. It isan open question, how- ever,whether this phenomenon has had asizableeffect on theaggregate unem- ploymentrate.

TheProductivity Acceleration Acentralfeature of the New Economy of the late 1990s was arisein thegrowth rateof labor productivity. Average annual growthin output perhour ofwork was 1.5percent over 1974 –1995and roseto 2.6 percent over 1996 –2000.Most expla- nations ofthis change focus on theincreased use ofcomputers and theInternet (forexample, see the Symposium on Computersand Productivityin theFall 2000 issueof this journal). Forour purposes, akeyfact about theproductivity acceler- ation isthat itstarted in the mid-1990s, around thesame time that researchers starteddetecting a declinein the NAIRU. This coincidencesuggests a linkbetween thetwo phenomena. Such alinkis also suggestedby the experience of the 1970s. This was the beginningof the infamous “productivityslowdown ”:averageannual productivity growthfell to its 1974 –1995average of 1.5 percent after an averageof 3.3percent over 1948–1973.As discussed above, the1970s were also aperiodof a rising NAIRU. If thereis alinkbetween shifts inproductivitygrowth and inthe NAIRU, itmayhelp explain both therising NAIRU ofthe 1970s and thefalling NAIRU of the 1990s. Such alinkwas suggestedby students ofthe rising NAIRU inthe1970s, notably Grubb,Jackman and Layard(1982) and Braun (1984).These authors presenta particularexplanation for the link, one resting on theidea that “wageaspirations ” adjust slowlyto shifts inproductivity growth. The concept ofwage aspirations isa departurefrom the neoclassical theory of the labor market, but itbuilds on researchby psychologists and industrialrelations specialists. The story goes as follows.6 In asteadystate with constant growthof labor productivity, the growth of real wagesis determined by the growth of productivity, as suggestedby neoclassical theory(and empiricalevidence). In such asituation, workerscome to viewthe rate ofreal wage increase that theyreceive as normaland fairand toexpect it to continue. If productivitygrowth falls, as inthe1970s, fundamentals dictatethat real wagegrowth must fallas well.Workers resist this decrease,however; they try to maintain thewage increases to which theyare accustomed. To theextent that workershave somein  uenceover , this means that wagesetters will try to achievereal wage increases above the level that can besustained byproductivity growth.This mismatch betweenreal wage aspirations and productivitygrowth worsensthe in  ation-unemploymenttradeoff. In otherwords, theNAIRU rises. This storyreceived attention in the early 1980s and then faded frompromi- nence. It has beenresurrected in the last fewyears, as many economistshave

6 Of course,a moreclassical story linkingproductivity and employmentis the realbusiness cycle theory ofLongand Plosser(1983) and Prescott(1986). Fora critique,see Mankiw (1989). 130Journal of Economic Perspectives

Figure 3 TheNAIRU and ProductivityGrowth

noticedthe parallel between the 1970s and the1990s. Today ’sversionof thestory reversesthe signs. Productivityhas accelerated,but workershave becomeaccus- tomedto the slow since the 1970s. A mismatch ofproductivity and wageaspirations inthis directionshifted the Phillips curve favorably. This storyis told, forexample, by Blinder (2000), De Long (2000) and the2000 Economic Reportof the President. 7 FollowingStaiger, Stock and Watson (2001)and Balland Mof Ž tt(2001), we examinedata on unemploymentand productivitygrowth to seewhether they Ž t the story.Figure 3 shows theNAIRU seriesfrom Figure 1; again, thereare two versions correspondingto differentsmoothing parameters.Figure 3 also shows thetrend in productivitygrowth, obtained with the HP Ž lterand withthe same smoothing parametersused tocreate the NAIRU series.Productivity growth is shown on an invertedscale to make it easier to see the negative comovement between the two trends. One can seebroadly similar patterns inthe two trends, although thematch betweenthem is far from perfect. One importantsubtlety is that therate of productivitygrowth is not exactlythe relevantvariable in the story discussed above. In asteadystate, wageaspirations adjust toany growthrate. What causes ashiftin the Phillips curve is a change in productivitygrowth, because aspirations aretied to wage growth and hencepro- ductivitygrowth in the past. Therefore,following Ball and Mof Ž tt(2001), we examinethe gap betweencurrent productivity growth and alongmoving average ofpast growth(one that depends on productivitygrowth into the distant past, but

7 Theshifts inthe Phillipscurve that occurin this story areeventually reversed when wage aspirations adjust tothe newrate of productivity growth. This createssome ambiguity about the right way to describethe shifts. As wediscussed earlier, the Phillipscurve can movebecause of either a transitory “supplyshock ” ora changein the NAIRU, and the distinctionbetween the twois basedon afuzzy notion ofpersistence.Since a Phillipscurve shift causedby a productivityspeedup eventually goes away, one mightcall it asupplyshock. We prefer to call it achangein the NAIRU, however,because the shift can last formany years. In particular,it lasts longenough to in  uencethe NAIRU seriesin Figure 1, which Ž ltersout the year-to-yeareffects ofsupplyshocks. TheNAIRU inTheory and Practice 131

Figure 4 TheNAIRU and theChange in Productivity Growth

a Productivityacceleration is the differencebetween current productivity growth and an averageof past productivitygrowth (see note 8 inthe text). withgreater weight on recentobservations). This gap, likeother variables in the Ž gures,is smoothed withthe HP Ž lterto extract a trend. Figure4 graphs this trend along withour NAIRU series.Here, the comovement is closer than when we examinethe pure productivity growth rate. 8 Twodetails of these graphs deservenotice. First, our inverted gap variable peaks intheearly 1980s and starts declining,as does theNAIRU. This occurs even though, as shown inthe previous Ž gure,actual productivitygrowth does not accelerateuntil the 1990s. This suggestsan effectdiscussed byStiglitz (1997): a catch-up ofwage aspirations tothe productivity slowdown. In Stiglitz ’s story, the ongoingexperience of the productivity slowdown caused wageaspirations tofall slowly,so thegap betweenaspirations and productivitynarrowed over the slow growthera. This narrowingcaused theNAIRU tostart falling; the fall was then magniŽ edwhen productivitygrowth accelerated. Arelatedpoint isthat thetrend in ourgap variablefalls to itslowest level at the end ofour sample —as does theNAIRU, enhancing the Ž tofthe two series. In contrast, productivitygrowth rises at theend ofthesample, but isstillbelow its level inthe1960s. That is, what isspecial about thenew economy of the late 1990s is not therate of productivitygrowth, which was higher30 years before, but theincrease relativeto the recent past. Thehigh productivitygrowth of the 1960s was a continuation ofhigh growthsince World War II; wageaspirations had largely adjusted, so therewas littleeffect on thePhillips curve. In contrast, thePhillips curveshifted favorably in the late 1990s because ofthe combination ofhigh contemporaneous growthand lowgrowth in the preceding two decades. Theproductivity-based explanation for the declining NAIRU is related to a

8 If wedenote productivity growth by g,the productivityvariable in Figure4 is g 2 (1 2 b)[ g(21) 1 bg(22) 1 b2 g(23) 1 ...].That is, the weightedaverage of past productivitygrowth in the expression has exponentiallydeclining weights. Theparameter b,whichgives the rateof decline, is set at 0.95. 132Journal of Economic Perspectives

common explanationin the popular press. In explainingwhy in  ation failedto acceleratein the late 1990s despite low unemployment, many journalists citethe productivityacceleration. Their story goes as follows.According to the Phillips curve,low unemployment puts upward pressureon wagegrowth, which feedsinto in ation. Lowunemployment has ledto more rapid wage growth. However, the productivityacceleration has reduced Ž rms’ costs, offsettingthe increases from rapidwage growth. Because overall costs have not accelerated,in  ation has not had to rise.9 This storyhas commonsense appeal. It does not contain any explicitrole for slowadjustment ofwage aspirations, but such aroleis in fact implicit.In a neoclassicalworld, a risein productivitygrowth has no obviouseffect on in  ation, becausehigher productivity is re  ectedfully in higher real wages. The idea that a productivityacceleration reduces Ž rms’ costs depends ontheimplicit assumption that wagesdo not adjust fullyto productivity movements. Thus, theidea of slowly adjusting wageaspirations providesan underpinning fora common journalistic explanationfor the recent experience. 10

The Beveridge Curve

In analyzingthe labor market, a complementto the Phillips curve is the Beveridgecurve, which has recentlybeen emphasized by Blanchard and Diamond (1989).The Beveridge curve shows therelationship between unemployment (work- erswithout jobs) and vacancies (jobs withoutworkers). The Beveridge curve slopes downward inunemployment-vacancy space because an economicexpansion that reducesunemployment also raisesvacancies, as Ž rmshave trouble Ž nding workers inatighterlabor market. Likethe Phillips curve, the Beveridge curve appears toshiftover time. Figure 5 plotsunemployment and job vacancy ratesfor annual U.S. data from1960 through 1998.The vacancy seriesis taken from Cohen, Dickensand Posen (2001),who, followingAbraham (1987),estimate the level of vacancies based on helpwanted advertisingin newspapers. In the Ž gure,there appear tobe stable Beveridge curves withdifferent intercepts in differentperiods. The Beveridge curve shifted outward fromthe period 1960 –1969to the period 1975 –1985and shiftedsharply inward

9 Forexample, Louis Uchitelle (2000) discusses “the concernthat lowunemployment drives up wages and, inturn, prices. ” In explainingwhy in  ation has not risen,he points to “improvementsin produc- tivity, effectivelygiving employers more revenue to pay forraises without raisingprices. ” Uchitelle attributes this ideato Alan Greenspan. 1 0 Theproductivity hypothesis is alsosomewhat related to another popular story: Alan Greenspan(1997) has suggestedthat workers,cowed by job insecurity, lacked aggressiveness in wage negotiations. As discussedby Katz and Krueger(1999), thereis no evidence to support an exogenousshift inworkers ’ perceptionsof jobsecurity. But what matters isaggressiveness of wage seekers relative to productivity growth. Failureto increase aggressiveness when productivity accelerates has the sameeffect onthe NAIRU as an exogenousdecrease in aggressiveness. Laurence Ball andN. Gregory Mankiw133

Figure 5 TheBeveridge Curve

after1990. This patternof an unfavorableshift in the1970s and afavorableshift in the1990s corresponds tothe broad patternof Phillips curve shifts, as measuredby timevarying NAIRU estimates. 11 Thesefacts suggestthat movementsin the Phillips curve and theBeveridge curveare linked. A numberof authors, includingKatz and Krueger(1999) and Cohen, Dickensand Posen (2001),argue that such alinkhelps isolate the right explanationfor the recent fall in the NAIRU —inparticular, that itpoints toward storiesabout improvedjob matching. In theoreticalwork, the Beveridge curve is oftenderived from search modelsof thelabor market, where frictions in matching jobs and workersproduce unemploymentand vacancies (forexample, Pissarides, 2000).In thesemodels, improvements in the matching technologycause the Beveridgecurve to shift in. Thus, therecent behavior of the Beveridge curve is consistent withthe existence of such improvements,arising, for example, from the growthof the temporary help industry. Yet,we doubt that theBeveridge curve is informative about thesources of NAIRU movements.Although theshift in the curve is consistent withimproved matching technology,it is also consistent withother explanations forthe falling NAIRU. Forexample, suppose theNAIRU fallsbecause workerswho do not search hard forjobs becomeincarcerated or receive disability bene Ž ts and, therefore, drop out ofthe labor force. This reducesunemployment but has littleeffect on vacancies, because theseworkers were unlikely to Ž lljobs anyway. Orsuppose the NAIRU fallsbecause wageaspirations fallrelative to productivity growth. This makesworkers more willing to take jobs when wagesare a givenlevel relative to

1 1 Cohen,Dickens and Posen(2001) stopin 1998 because the riseof Internet advertisingmakes the newspaperhelp wanted indexan unreliablemeasure of vacancies in recent years. 134Journal of Economic Perspectives

productivity.When workerstake jobs morereadily, both unemploymentand vacancies fall,and again theBeveridge curve shifts in. Astheseexamples illustrate, mostplausible stories about ashiftingPhillips curve can explaina shiftingBever- idgecurve as well.Thus, thefact that theBeveridge curve shifted inward after 1985 says littleabout whyNAIRU fell. This argumentis strengthened by the fact that theBeveridge curve shifted outward inthe1970s, when theNAIRU rose.While some authors suggestthat the matching technologyhas improvedrecently, to ourknowledge no onehas argued that itdeteriorated in the 1970s. The relationship between Phillips curve and Beveridgecurve shifts appears tohold consistentlyover time, but itdoes not tellus much about whythese shifts occur.

Conclusion

The NAIRU— orits approximate synonym, thenatural rateof unemploy- ment—isan importantbuilding block of business cycletheory. Few economists woulddeny that shifts inaggregate demand, such as those drivenby monetary policy,push in  ation and unemploymentin opposite directions, at leastin the short run. That isall one needs to believe to accept theNAIRU concept. Thepractical application ofthis concept, however,is lessstraightforward. The valueof NAIRU ishard tomeasure, largely because itchanges overtime. The economyexperiences many kinds ofshocks that in  uence in ation and unemploy- ment.In lightof this fact, itwould be remarkable if the level of unemployment consistent withstable in  ation wereeasy to measure. Thereis no shortageof hypotheses toexplain what causes theNAIRU to change overtime and, inparticular, why it fell during the 1990s. The available evidenceis too weak to establish decisively which hypothesis isright, but the literatureon theNAIRU has madeprogress. Demography and governmentpolicy both playsome role. In addition, changes inproductivitygrowth appear toshift the in ation-unemploymenttradeoff. In thepast, mostmacroeconomists studying the Phillipscurve have concentrated theirattention on thedynamic relationship between in ation and unemployment.In thefuture, they should expand their scope tobuild and testmodels of in  ation, unemploymentand productivity. y Wearegrateful for research assistance from Robert Tchaidze and Gergana Trainor and for editorialsuggestions from J. BradfordDe Long,Timothy Taylor and Michael Waldman. TheNAIRU inTheory and Practice 135

References

Abraham,Katherine. 1987. “Help-WantedAd- gregateEvidence. ” EconomicJournal. March, 110: vertising, JobVacancies, and Unemployment. ” 462, pp. C1–C33. Brookings Paperson Economic Activity. 1, pp. 207– Blinder,Alan. 2000. “TheInternet and the 48. NewEconomy. ” BrookingsInstitution Policy Akerlof,George A. andJanet L.Yellen. 1985. Brief60, June. “ANear-RationalModel of the BusinessCycle, Braun,Steven N. 1984. “Productivityand the with Wageand PriceInertia. ” QuarterlyJournal of NIIRU (and OtherPhillips Curve Issues). ” Economics. 100:5, Supplement,pp. 823 –38. WorkingPaper No. 34, FederalReserve Board Autor,David H. andMark Duggan. 2001. “The EconomicActivity Section. Risein Disability Recipiency and the Declinein Cohen,Jessica, William T. Dickensand Adam Unemployment. ” MIT WorkingPaper #01-15, Posen. 2001. “Havethe NewHuman-Resource May. ManagementPractices Lowered the Sustainable Ball,Laurence. 1999. “AggregateDemand UnemploymentRate? ” in TheRoaring Nineties: and Long-RunUnemployment. ” Brookings Papers CanFull Employment BeSustained? Alan B. onEconomic Activity. 2, pp. 189 –251. Kruegerand RobertSolow, eds. New York: The Ball,Laurence. 2000. “NearRationality and RussellSage Foundation and TheCentury Foun- In ation inTwo Monetary Regimes. ” NBER dationPress, pp. 219 –59. WorkingPaper No. 7988. Councilof EconomicAdvisors. 2000. Economic Ball,Laurence and N. GregoryMankiw. 1995. Reportof the President. Washington, D.C.: U.S. “Relative-PriceChanges as AggregateSupply GovernmentPrinting Of Ž ce. Shocks.” QuarterlyJournal of Economics. February, DeLong, J. Bradford. 2000. “What Went 110, pp. 161–93. Right inthe 1990s?Sources of American and Ball,Laurence and Robert Mof Ž tt. 2001. “Pro- Prospectsfor World , ” in The ductivity Growthand the PhillipsCurve, ” in The AustralianEconomy inthe 1990s. DavidGruen and RoaringNineties: Can Be Sus- SonaShresha, eds.Reserve Bank ofAustralia, tained? Alan B.Kruegerand RobertSolow, eds. p. 13. NewYork: The Russell Sage Foundation and Fischer,Stanley. 1977. “Long-TermContracts, TheCentury Foundation Press, pp. 61 –90. ,and the OptimalMoney Ball,Laurence and David Romer. 1990. “Real SupplyRule. ” Journalof . Febru- Rigiditiesand the Nonneutralityof Money. ” Re- ary, 85:1, pp. 191 –205. viewof Economic Studies. April,57, pp. 539 –52. Friedman,Milton. 1968. “TheRole of Mone- Barro,Robert. 1977. “Unanticipated Money tary Policy. ” AmericanEconomic Review. March, 58, Growthand Unemploymentin the United pp. 1–17. States.” AmericanEconomic Review. March, 67:2, pp. 101–15. Gordon,Robert J. 1998. “Foundationsof the Barsky, RobertB. 1987. “TheFisher Effect GoldilocksEconomy: Supply Shocks and the and the Forecastabilityand Persistenceof In  a- Time-VaryingNAIRU. ” Brookings Paperson Eco- tion.” Journalof . January, 19, nomicActivity. 2, pp. 297–333. pp. 3–24. Gray, Jo Anna. 1976. “WageIndexation: A Blanchard,Olivier Jean and . MacroeconomicApproach. ” Journalof Monetary 1989. “TheBeveridge Curve. ” Brookings Paperson Economics. April,2:2, pp. 221 –35. EconomicActivity. 1, pp. 1– 60 Greenspan,Alan. 1997. “TheFederal Re- Blanchard,Olivier Jean and Nobuhiro Kiyo- serve’sSemiannualMonetary Policy Report. ” taki. 1987. “MonopolisticCompetition and the Testimonybefore the U.S. HouseCommittee on Effects ofAggregate Demand. ” AmericanEco- Banking, Housingand UrbanAffairs, February nomicReview. September,77, pp. 647 –66. 26. Blanchard,Olivier Jean and Lawrence H. Grubb,Dennis, Richard Jackman andRichard Summers. 1986. “Hysteresisand the European Layard. 1982. “Causesof the CurrentStag  a- UnemploymentProblem, ” in NBERMacroeco- tion.” Reviewof Economic Studies. SpecialIssue, nomicsAnnual. S.Fischer,ed. Cambridge, Mass.: 49:5, pp. 707 –30. MIT Press,pp. 15 –78. Hodrick,Robert J. andEdward C. Prescott. Blanchard,Olivier Jean and Justin Wolfers. 1997. “PostwarU.S. BusinessCycles: An Empiri- 2000. “TheRole of Shocks and Institutions in calInvestigation. ” Journalof Money, Credit and the Riseof European Unemployment: The Ag- Banking. February,29:1, pp. 1 –16. 136Journal of Economic Perspectives

Hume,David. 1752. “Of Money,” in Essays. Roberts,John M. 1998. “In ation Expecta- London:George Routledge and Sons. tions and the Transmissionof MonetaryPolicy. ” Katz, LawrenceF. andAlan B. Krueger. 1999. Boardof Governors of the FederalReserve, Fi- “TheHigh-Pressure U.S. LaborMarket of the nanceand EconomicsDiscussion Series Paper 1990s.” Brookings Paperson Economic Activity. 1, No.1998-43. pp. 1–65. Rotemberg,Julio. 1982. “MonopolisticPrice Layard,Richard, Steven Nickell and Richard Adjustment and AggregateOutput. ” Review of Jackman. 1991. Unemployment:Macroeconomic Per- EconomicStudies. October,49:4, pp. 517 –31. formanceand the Labour Market. Oxford:Oxford Samuelson,Paul A. andRobert M. Solow. University Press. 1960. “Analytical Aspects ofAnti-In  ation Poli- Long,John B. andCharles I. Plosser. 1983. cy.” AmericanEconomic Review. May, 50, pp. 177 – “RealBusiness Cycles. ” Journalof Political Econ- 94. omy. February,91, pp. 39 –69. Sargent,Thomas J. 1971. “ANoteon the Ac- Lucas,Robert E. Jr. 1972. “EconometricTest- celerationistControversy. ” Journalof Money, ingof the Natural Rate Hypothesis, ” in The Creditand Banking. 8:3, pp. 721–25. Econometricsof Price Determination. Otto Eckstein, Shimer,Robert. 1999. “Why isthe U.S. Unem- ed.Washington, D.C.: Boardof Governors of the FederalReserve System, pp. 50 –59. ploymentRate SoMuch Lower? ” NBER Macro- Lucas,Robert E. Jr. 1973. “SomeInterna- economicsAnnual. 13, pp. 11– 61. tionalEvidence on Output-In  ation Tradeoffs. ” Shimer,Robert. 2001. “TheImpact ofYoung AmericanEconomic Review. June,63, pp. 326 –34. Workerson the AggregateLabor Market. ” Quar- Lucas,Robert E. Jr. 1996. “NobelLecture: terly Journalof Economics. August, 116,pp. 969 – MonetaryNeutrality. ” Journalof Political Economy. 1008. August, 104:4, pp. 666 – 82. Staiger,Douglas, James H.Stockand Mark W. Mankiw,N. Gregory. 1985. “SmallMenu Costs Watson. 1997. “HowPrecise are Estimates of the and LargeBusiness Cycles: A Macroeconomic Natural Rate ofUnemployment? ” in Reducing Modelof . ” QuarterlyJournal of Econom- Ination: Motivation and Strategy. C.D.Romer and ics. May, 100,pp. 529 –37. D.H. Romer,eds. Chicago: University ofChicago Mankiw,N. Gregory. 1989. “RealBusiness Cy- Press,pp. 195 –246. cleTheory: A NewKeynesian Perspective. ” Jour- Staiger,Douglas, James H.Stockand Mark W. nalof Economic Perspectives. Summer,3:3, pp. 79 – Watson. 2001. “Prices,Wages, and the U.S. 90. NAIRU inthe 1990s, ” in TheRoaring Nineties: Can Mankiw,N. Gregoryand Ricardo Reis. 2001. Full Employment BeSustained? Alan B.Krueger “StickyInformation versus Sticky Prices: A Pro- and RobertSolow, eds. New York: The Russell posalto Replace the NewKeynesian Phillips SageFoundation and TheCentury Foundation Curve.” NBERWorking Paper No. 8290. Quar- Press, pp. 3–60. terly Journalof Economics. Forthcoming. Stiglitz, Joseph. 1997. “Re ectionson the Nat- Perry,George. 1970. “ChangingLabor Mar- uralRate Hypothesis. ” Journalof Economic Perspec- kets and In ation.” Brookings Paperson Economic tives. Winter, 11:1, pp. 3 –10. Activity. 3, pp. 411– 48. Stock,James H.andMark W. Watson. 1999. Phelps,Edmund S. 1967. “PhillipsCurves, Ex- “ForecastingIn  ation.” Journalof Monetary Eco- pectationsof In  ation, and OptimalUnemploy- nomics. October,44:2, pp. 293 –335. mentover Time. ” . 2:3, pp. 22– 44. Phelps,Edmund S. 1968. “Money-WageDy- Summers,Lawrence H. 1986. “Why isthe Un- namicsand LaborMarket Equilibrium. ” Journal employmentRate SoVery High Near Full Em- ofPolitical Economy. July/August, Part 2, 76, pp. ployment?” Brookings Paperson Economic Activity. 678–711. 2, pp. 339– 83. Pissarides,Christopher A. 2000. Equilibrium Taylor,John B. 1980. “AggregateDynamics UnemploymentTheory. Cambridge,Mass.: MIT and StaggeredContracts. ” Journalof Political Press. Economy. February,88:1, pp. 1 –22. Prescott,Edward C. 1986. “TheoryAhead of Thurow,Lester. 1998. “TheDisappearance of BusinessCycle Measurement. ” Carnegie-Rochester In ation.” FinancialTimes. December10. ConferenceSeries on Public Policy. Autumn, 25, pp. Uchitelle,Louis. 2000. “Greenspan’s Way of 11–44. PullingHis Punches. ” NewYork Times. July 16.