The Shimer Puzzle and the Identification of Productivity Shocks

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. The Shimer puzzle and the Identification of Productivity Shocks Regis Barnichon 2009-04 NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

The Shimer puzzle and the Identi(cid:133)cation of Productivity Shocks (cid:3) Regis Barnichon Federal Reserve Board 25 October, 2008 (First Draft: April 2007) Abstract Shimer(2005)arguesthattheMortensen-Pissarides(MP)modelofunemploymentlacks anampli(cid:133)cationmechanismbecauseitgenerateslessthan10percentoftheobservedbusinesscycle(cid:135)uctuationsinunemploymentgivenlaborproductivityshocksofplausiblemagnitude. Thispaperarguesthatpartoftheproblemlieswiththeidenti(cid:133)cationofproductivity shocks. Because of the endogeneity of measured labor productivity, (cid:133)ltering out the trend component as in Shimer (2005) may not correctly identify the shocks driving unemployment. Using a New-Keynesian framework to control for the endogeneity of productivity, this paper estimates that the MP model can account for a third, and possibly as much as 60 percent, of (cid:135)uctuations in labor market variables. JEL classi(cid:133)cations: E32, E37, J63, J64 Keywords: Unemployment Fluctuations, Labor productivity, Search and matching model, New-Keynesian model (cid:3)IwouldliketothankStephanieCheng,JordiGali,WouterdenHaan,BarbaraPetrongolo,ChrisPissarides, Silvana Tenreyro and seminar participants at the 2008 EEA/ESEM conference in Milan for helpful comments. Any errors are my own. E-mail: <r.o.barnichon@lse.ac.uk>. 1

1 Introduction In a very in(cid:135)uential paper, Shimer (2005) argues that the Mortensen-Pissarides (MP) search model of unemployment lacks an ampli(cid:133)cation mechanism because it generates less than 10 percent of the observed business cycle (cid:135)uctuations in unemployment given labor productivity shocks of plausible magnitude. In this paper, I argue that Shimer(cid:146)s (2005) estimate may be biased downward because of the endogeneity of labor productivity, and I estimate that a third, and possibly as much as 60 percent, of the Shimer puzzle is simply due to the misidenti(cid:133)cation of productivity shocks. The Shimer puzzle has attracted a lot of interest in the literature, and a number of researchers have focused on ways to create more ampli(cid:133)cation so that small exogenous productivity movements generate large (cid:135)uctuations in unemployment.1 However, there is substantial evidence that, perhaps due to labor hoarding and variable capacity utilization, some of the movements in productivity are in fact endogenous.2 For example, when the (cid:133)rm is demand constrained in the short-run, (cid:133)rms can respond to changes in demand by adjusting their level of capacity utilization of inputs (capital or labor), and measured labor productivity (cid:135)uctuates endogenously with aggregate demand and hence unemployment.3 By (cid:133)ltering out the trend component of output per hour to identify productivity shocks, Shimer (2005) may not identify thetrueproductivityshocksbutrathertheendogenousresponseofproductivitytounobserved disturbances. Andbecausethisendogenousresponseissmall,thismayexplainwhythecyclical component of measured labor productivity (cid:135)uctuates less than unemployment. To estimate the impact of exogenous changes in productivity on labor market variables, I imposelong-runrestrictionsinastructuralVARmodelalongthelineofGali(1999), andI(cid:133)nd thatapermanentproductivityincreasetemporarilylowerslabormarkettightness(thevacancy- 1See, among others, Hagedorn and Manovski (2005) , Hall (2005), Hall and Migrom (2005), Shimer (2004), and Mortensen and Nagypal (2005) for a review of recent e⁄orts. 2See, among others, Bils and Cho (1994), Burnside, Eichenbaum and Rebelo (1993), Burnside and Eichenbaum (1996) and Basu and Kimball (1997). 3This idea is given empirical support in Barnichon (2008), following Gali (1999). 2

unemployment ratio), while the MP model implies the opposite.4 Hence, before assessing the ampli(cid:133)cation properties of the MP model, I embed the search and matching model in a New Keynesian framework. In this set-up, a permanent increase in productivity (i.e. a positive productivity shock) may temporarily raise unemployment and lower labor market tightness because aggregate demand does notadjust immediately to the new productivitylevel in the presence of nominal rigidities, and hence (cid:133)rms use less labor. The model also generates endogenous movements in productivity. Because hiring (cid:133)rms are demand constrained, an aggregate demand shock generates a transitory movement in productivity as (cid:133)rms vary their level of capacity utilization. To estimate the proportion of Shimer(cid:146)s puzzle due to the endogeneity of productivity, I use a calibrated version of the model to control for endogenous productivity movements unrelated to productivity shocks, and I reproduce Shimer(cid:146)s (2005) exercise on data simulated from my model. With a standard calibration, simulated labor market tightness is 9 times more volatile than the cyclical component of labor productivity, while the ratio comes at about 26 in US data. I conclude that the MP model can account for about a third, rather than 10 percent, of labor market tightness (cid:135)uctuations, and a sensitivity analysis suggests that this share could be as high as 60 percent. The remainder of the paper is organized as follows: Section 2 discusses Shimer(cid:146)s (2005) puzzle; Section 3 presents and calibrates a New-Keynesian model with search unemployment and replicates Shimer(cid:146)s (2005) exercise on model generated data; and Section 4 o⁄ers some concluding remarks. 4Barnichon (2008)and Canova, Lopez-Salido and Michelacci(2008)come to similarconclusions, albeitwith di⁄erent labor market variables. 3

2 The Shimer puzzle 2.1 Shimer(cid:146)s (2005) evidence In this section, I reproduce Shimer(cid:146)s exercise (2005), and Table 1 presents summary statistics for unemployment, vacancies, labor market tightness and productivity.5 As originally argued by Shimer (2005), the volatility of productivity is only a fraction (here less than 4%) of the volatility of labor market tightness. Turning to the correlation matrix, unemployment and labormarkettightnessareweaklycorrelatedwithproductivitywithcorrelationsofrespectively 0:23 and 0:19. (cid:0) In the context of a standard MP model where productivity movements are the central driving force of unemployment (cid:135)uctuations, Shimer (2005) shows that the standard deviations of unemployment, vacancies and productivity are of the same order of magnitude, i.e. (cid:27)(u) (cid:25) (cid:27)(v) (cid:27)(p):Byestimatingthatproductivityshocksareonly10%asvolatileasunemployment (cid:25) (cid:135)uctuations, Shimer (2005) concludes that the MP model can only account for less than 10% of unemployment (cid:135)uctuations. Furthermore, Shimer (2005) notes that the MP model exhibits virtually no propagation as it implies a contemporaneous correlation between unemployment and productivity of 1 when the data show a contemporaneous and peak unemployment- (cid:0) productivity correlation of respectively only 0:23 and 0:50. (cid:0) (cid:0) 2.2 Fixing the model to add more ampli(cid:133)cation One way to reconcile the MP framework with the data is to modify the model so that it generates more ampli(cid:133)cation, i.e. that a given shock to productivity has a larger impact 5I use quarterly data taken from the U.S. Bureau of Labor Statistics (BLS) covering the period 1951:Q1 to 2005:Q4. Labor productivity is measured as real average output per hour in the non-farm business sector, and unemployment is the quarterly average of the monthly unemployment rate series constructed by the BLS fromtheCurrentPopulationSurvey. Labormarkettightnessisde(cid:133)nedasthevacancy-unemploymentratioand vacancies are the quarterly average of the monthly Conference Board help-wanted advertising index. I remove low-frequency movements using a standard HP-(cid:133)lter with (cid:21)=1600. Alternatively, using (cid:21)=105 as in Shimer (2005a) does not change any of the results presented in this paper. I measure productivity as output per hour as in Shimer (2004) instead of output of worker as in Shimer (2005). The Shimer puzzle is present with the same magnitude using both measures, and all the results in this paper hold for both measures. 4

on unemployment. Mortensen and Nagypal (2006) provide a detailed review of the current e⁄ort in that direction, and I will only emphasize two in(cid:135)uential examples. A (cid:133)rst possibility, suggested by Hall (2005) and Shimer (2005), is to introduce real wage rigidity. In the standard MP model, the Nash bargaining real wage responds so much to movements in productivity that it e⁄ectively absorbs most of the changes in productivity. As a result, the surplus of the match responds only weakly to (cid:135)uctuations in productivity. By introducing a degree of real wagerigidity,movementsinproductivityhaveamoresubstantialimpactonthematchsurplus, on the incentives of (cid:133)rms to post vacancies and hence on equilibrium unemployment. Another possibility, suggested by Hagedorn and Manovskii (2004), does not rely on real wage rigidity but uses a standard MP model with a di⁄erent calibration than the one used in Shimer(cid:146)s. Hagedorn and Manovskii (2004) show that when the opportunity cost of employment is high, the job (cid:133)nding rate becomes very responsive to changes in productivity, and the MP model can quantitatively account for the magnitude of unemployment (cid:135)uctuations. While this approach is di⁄erent from the one proposed by Hall (2005) and Shimer (2005), the underlying philosophy is the same: one needs to modify the MP model (either its equations or its calibration) so that the surplus of the match becomes more responsive to changes in productivity. 2.3 The conditional volatilities of productivity and labor market tightness Theaforementionedliteraturegenerallyconsidersproductivitymovementsasexogenous. However, there is substantial evidence that, perhaps due to labor hoarding and variable capacity utilization, some of the movements in productivity are in fact endogenous.6 To identify the impact of exogenous changes in productivity, I follow Gal(cid:237) (1999) and Blanchard and Quah (1989) and impose long-run restrictions in structural VAR models to identify technological disturbances. Technology shocks are the only shocks with a permanent impact on productivity, and I interpret transitory productivity movements as variations in 6See, among others, Bils and Cho (1994), Burnside, Eichenbaum and Rebelo (1996) and Basu and Kimball (1997). 5

capacity utilization. Speci(cid:133)cally, I am interested in estimating the system (cid:1)ln yt "a ht = C(L) t 0 1 0 1 ln (cid:18) "m t t B C B C @ A @ A where yt is labor productivity de(cid:133)ned as output per hours, (cid:18) the vacancy-unemployment ht t ratio,C(L)aninvertiblematrixpolynomialandthevectorofstructuralorthogonalinnovations comprises technology shocks "a and non-technology shocks "m.7 t t Figure 1 presents the impulse response functions. The Shimer puzzle is clearly apparent for both shocks: the standard deviation over the (cid:133)rst two years after a technology shock is 16 times larger for labor market tightness than for output per hour, and after a non-technology shock, the ratio is 21. However, as similarly emphasized in Gali (1999), Barnichon (2008) and Canova, Lopez-Salido and Michelacci (2008) a positive technology shock temporarily lowers labor market tightness, while the MP model implies the opposite.8 This implies that it is di¢ cult to draw conclusions regarding the ampli(cid:133)cation properties of the baseline MP model since its transmission mechanism is likely to be incomplete. 3 The Shimer puzzle in a New-Keynesian setting To reassess the extent of Shimer(cid:146)s puzzle, it is important to extend the search and matching model so that it can (i) rationalize endogenous productivity movements, and (ii) account for the fact that permanent productivity increases temporarily lower labor market tightness. To do so, I follow Gali (1999) and Barnichon (2008), and I extend the MP model so that hiring 7I use quarterly data taken from the U.S. Bureau of Labor Statistics (BLS) covering the period 1951:Q1 to 2005:Q4. Laborproductivityx ismeasuredasrealaverageoutputperhourinthenon-farmbusinesssector,and t labormarkettightness(cid:18) =v =u istheratioofthequarterlyaverageofthemonthlyunemploymentrateseries t t t constructedbytheBLSfrom theCurrentPopulationSurveyovertheConferenceBoardhelpadvertisingindex. FollowingFernald(2007),Iallowfortwobreaksin(cid:1)ln y ,1973:Q1and1997:Q1,andI(cid:133)ltertheunemployment h t series with a quadratic trend. Fernald (2007) showed that the presence of a low-frequency correlation between (cid:0) (cid:1) laborproductivitygrowthandunemployment,whileunrelatedtocyclicalphenomena,couldsigni(cid:133)cantlydistort the estimates of short run responses obtained with long run restrictions. 8SeeBarnichon(2008)foradiscussionaboutthepositiveimpactoftechnologyshocksonunemploymentand its implications for the modeling of unemployment (cid:135)uctuations. 6

(cid:133)rms are demand constrained in a New-Keynesian fashion. In a neoclassical setting, (cid:133)rms post vacancies depending on the return of the match. However, this needs not be the case when (cid:133)rms have to satisfy a given level of demand for their products. In a New-Keynesian setting with monopolistically competitive (cid:133)rms and nominal rigidities, (cid:133)rms may have to hire more workers when demand is unexpectedly high even if productivity (and hence the match surplus) does not increase. Put di⁄erently, the number of posted vacancies could increase without any change in productivity. In practice, (cid:133)rms also respond to higher demand by increasing capacity utilization of inputs (capital or labor). As a result, measured labor productivity (cid:135)uctuates with aggregate demand and hence unemployment. A permanent increase in productivity (i.e. a technology shock) may temporarily raise unemploymentbecausewithnominalrigidities,aggregatedemanddoesnotadjustimmediately to the new productivity level, and (cid:133)rms use less labor. In the next subsections, I present and calibrate a New-Keynesian model with search unemployment, and I replicate Shimer(cid:146)s exercise on model generated data. 3.1 A New-Keynesian model with search unemployment FollowingBarnichon(2008)andKrauseandLubik(2007), IextendtheMPmodelbyintroducing nominal frictions so that hiring (cid:133)rms are demand constrained in a New-Keynesian fashion. In addition, I make a distinction between the extensive (number of workers) and the intensive (hours and e⁄ort) labor margins. In this framework, unemployment (cid:135)uctuations are the productoftwodisturbances: technologyshocksandmonetarypolicy(oraggregatedemand)shocks. A positive technology shock permanently raises productivity but may also temporarily raise unemployment and lower labor market tightness. A positive monetary policy shock decreases unemployment and increases measured productivity temporarily, because (cid:133)rms increase labor e⁄ort to satisfy demand in the short run. As a result, measured labor productivity is the product of two components: permanent and temporary disturbances. 7

The main ingredients of the model are monopolistic competition in the goods market, hiring frictions in the labor market and nominal price rigidities. There are three types of agents: households, (cid:133)rms and a monetary authority. 3.1.1 Households Iconsideraneconomypopulatedbyacontinuumofhouseholdsofmeasureoneandacontinuum of(cid:133)rmsofmeasureone. Withequilibriumunemployment,ex-antehomogenousworkersbecome heterogeneous in the absence of perfect income insurance because each individual(cid:146)s wealth di⁄ersbasedonhisemploymenthistory. Toavoiddistributionalissues,IfollowMerz(1995)and Andolfatto (1996) in assuming that households form an extended family that pools its income andchoosespercapitaconsumptionandassetsholdingtomaximizeitsexpectedlifetimeutility. There are 1 n unemployed workers who receive unemployment bene(cid:133)ts b in units of utility t (cid:0) of consumption, and n employed workers who receive the wage payment w from (cid:133)rm i for t it providing hours h and e⁄ort per hour e .9 Denoting g(h ;e ) the individual disutility from it it it it working, the representative family seeks to maximize E 1 (cid:12)t ln(C )+(cid:21) ln( M t ) n 1 g(h ;e )di 0 t m t it it P (cid:0) t=0 (cid:20) t Z0 (cid:21) X subject to the budget constraint 1 1 P C dj +M = n w di+(1 n )bC +(cid:5) +M jt jt t t it t t t t 1 Z0 Z0 (cid:0) (cid:0) with (cid:21) a positive constant, M nominal money holdings, (cid:5) total transfers to the family and m t t " C the composite consumption good index de(cid:133)ned by C = 1 C " (cid:0)" 1 di " (cid:0) 1 where C is the t t 0 it it (cid:18) (cid:19) quantity of good i [0;1] consumed in period t and P is theRprice of variety i: " > 1 is the it 2 elasticity of substitution among consumption goods. The aggregate price level is de(cid:133)ned as 9I introduce variable e⁄ort per hour in order to generate procyclical productivity movements. 8

1 1 1 " (cid:0) P = P1 "di . The disutility from supplying hours of work h and e⁄ort per hour e is t 0 it(cid:0) 1 t t Z 0 the su@m of the dAisutilities of the members who are employed. Following Bils and Cho (1994), the individual period disutility of labor takes the form: (cid:21) (cid:21) g(h ;e ) = h h1+(cid:27)h +h e e1+(cid:27)e it it 1+(cid:27) it it 1+(cid:27) it h e where (cid:21) ; (cid:21) ; (cid:27) and (cid:27) are positive constants. The last term re(cid:135)ects disutility from exerting h e h e e⁄ort with the marginal disutility of e⁄ort per hour rising with the number of hours. An in(cid:133)nite value for (cid:27) generates the standard case with inelastic e⁄ort. e 3.1.2 Firms and the labor market Each di⁄erentiated good is produced by a monopolistically competitive (cid:133)rm using labor as the only input. There is a continuum of large (cid:133)rms distributed on the unit interval. At date t, each (cid:133)rm i hires n workers to produce a quantity y = A n L(cid:11) where A is an aggregate it it t it it t technologyindex,L thee⁄ectivelaborinputsuppliedbyeachworkerand0 < (cid:11) < 1.10 Ide(cid:133)ne it e⁄ective labor input as a function of hours h and e⁄ort per hour e such that L = h e . it it it it it Total e⁄ective labor input can be adjusted through three channels: the extensive margin n , it and the two intensive margins: hours h and e⁄ort per hour e . With variable e⁄ort, the it it model will be able to generate endogenous procyclical movements in productivity. Being a monopolistic producer, the (cid:133)rm faces a downward sloping demand curve yd = it (Pit) "Y and chooses its price P to maximize its value function given the aggregate price Pt (cid:0) t it level P and aggregate output Y . As is standard in New-Keynesian models, (cid:133)rms are subject t t to Calvo-type price setting and can only reset their price at random dates. Each period a fraction (cid:23) of randomly selected (cid:133)rms cannot reset its price. 10This production function can be rationalized by assuming a constant capital-worker ratio and a standard Cobb-Douglas production function y it =A t (nL it )(cid:11)K i 1 t(cid:0) (cid:11). Note however, that the main message of the paper doesnotrelyonthisparticularchoiceoftheproductionfunction,andthatthemodelcouldaccommodateother functional forms. 9

In a search and matching model of the labor market, workers cannot be hired instantaneously and must be hired from the unemployment pool through a costly and time-consuming jobcreationprocess. Firmspostvacanciesataunitarycost, c = cA , andunemployedworkers t t search for jobs. I assume that the matching function takes the usual Cobb-Douglas form so that the (cid:135)ow m t of successful matches within period t is given by m t = m 0 u (cid:17) t v t 1 (cid:0) (cid:17) where m 0 is 1 a positive constant, (cid:17) (0;1), u denotes the number of unemployed and v = v di the total 2 t t 0 it number of vacancies posted by all (cid:133)rms. Accordingly, the probability of a vacRancy being (cid:133)lled in the next period is q((cid:18) ) m(u ;v )=v = m (cid:18) (cid:17) where (cid:18) vt is the labor market tightness. t (cid:17) t t t 0 (cid:0) t (cid:17) ut Similarly, the probability for an unemployed to (cid:133)nd a job is m(u t ;v t )=u t = m 0 (cid:18) t 1 (cid:0) (cid:17) . Matches are destroyed at a constant rate (cid:21), and the law of motion for a representative (cid:133)rm is given by n = (1 (cid:21))n +q((cid:18) )v . it+1 it t i;t (cid:0) When a (cid:133)rm and a worker meet, they must decide on the allocation of hours and e⁄ort to satisfy demand. I assume that both parties negotiate the hours/e⁄ort decision by choosing the optimal allocation and set hours and e⁄ort per hour to satisfy demand at the lowest utility cost for the worker. More precisely, they solve (cid:21) (cid:21) min h h1+(cid:27)h +h e e1+(cid:27)e hit;eit 1+(cid:27) h it it 1+(cid:27) e it subject to satisfying demand A n h(cid:11)e(cid:11) = yd at date t, and this implies that e⁄ort per hour t it it it it is a function of total hours (cid:27)h e = e h1+(cid:27)e it 0 it 1 where e = 1+(cid:27)e(cid:21)h 1+(cid:27)e is a positive constant. Thus, changes in hours can proxy for changes 0 (cid:27)e (cid:21)e (cid:16) (cid:17) in e⁄ort, and I can write a reduced-form relationship between output and hours ’ y = y A n h it 0 t it it 10

with y = e(cid:11) and ’ = (cid:11) 1+ (cid:27)h . For ’ > 1, the production function displays short 0 0 1+(cid:27)e (cid:16) (cid:17) run increasing returns to hours, and endogenous labor productivity (i.e. output per hour) movements are procyclical. 3.1.3 Wage bill setting As is usual in the search literature, (cid:133)rms and workers bargain individually about the real wage and split the surplus in shares determined by an exogenous bargaining weight (cid:13). Denoting J (w ) the value of a matched worker to (cid:133)rm i at date t, and W (w ) and U(w ) the value i it i it it for a worker of being respectively employed by (cid:133)rm i and unemployed, the equilibrium wage w satis(cid:133)es w =argmax (W (w ) U(w ))(cid:13)(J (w ))1 (cid:13) and is a solution of the (cid:133)rst-order it it i it it i it (cid:0) wit (cid:0) di⁄erential equation h @w c b g(h ;e ) it it t t t t w = (cid:13) + (cid:18) +(1 (cid:13)) + (1) it t ’ @h (cid:21) (cid:0) (cid:21) (cid:21) it t t t (cid:18) (cid:19) (cid:18) (cid:19) with (cid:21) = 1 .11 A solution is given by t Ct c b h1+(cid:27)h w it = (cid:13) t (cid:18) t +(1 (cid:13)) t +(1 (cid:13)){ it (2) (cid:21) (cid:0) (cid:21) (cid:0) (cid:21) t t t with { = (cid:21) 1 (cid:0) h( 1 ’ (cid:13) 1 + + ( (cid:27) 1 (cid:27) h + h + (cid:27) ) (cid:27) (cid:27) h e e ) > 0.12 3.1.4 The (cid:133)rm(cid:146)s problem Given the market wage and aggregate price level, (cid:133)rm i will choose a sequence of price P it f g andvacancies v tomaximizetheexpectedpresentdiscountedvalueoffuturepro(cid:133)tssubject it f g 11Whilethewageequation(1)isaweightedaverageofbothpartiessurplusesandissimilartootherbargained wages derived in e.g. Trigari (2004), Walsh (2004) or Krause and Lubik (2007), the (cid:133)rm(cid:146)s surplus is not given by the marginal product of labor. Indeed, once the (cid:133)rm has chosen its price, it is demand constrained and a marginalworkerwillnotincreasethe(cid:133)rm(cid:146)srevenue. Instead,the(cid:133)rstterm of(1)isgivenby @wit = hit@wit, (cid:0)@nit ’ @hit the change in the wage bill caused by substituting the intensive margin (hours and e⁄ort) with the extensive one (employment). See Barnichon (2008) for more details. 12The model is well behaved only if { >0. This imposes that 1 (cid:0) ’ (cid:13) (1+(cid:27) h )>0, which will be veri(cid:133)ed by the calibrated parameters. 11

to the demand constraint, the Calvo price setting rule, the hours/e⁄ort choice and the law of motion for employment. Formally, the (cid:133)rm maximizes its value u(C ) P c E (cid:12)j 0 t+j i;t+j yd n w v t u(C ) P i;t+j (cid:0) i;t+j i;t+j (cid:0) (cid:21) i;t+j j 0 t (cid:20) t+j t+j (cid:21) X subject to the demand constraint P yd = y A n h ’ = ( i;t ) "Y it 0 t it it P (cid:0) t t the law of motion for employment n = (1 (cid:21))n +q((cid:18) )v it+1 it t it (cid:0) and the bargained wage c b h1+(cid:27)h w it = (cid:13) t (cid:18) t +(1 (cid:13)) t +(1 (cid:13)){ it : (cid:21) (cid:0) (cid:21) (cid:0) (cid:21) t t t 3.1.5 Technological progress and the central bank Consistent with the long run identifying assumption made in Section 2, the technology index series is non-stationary with a unit root originating in technological innovations. Hence, technology is comprised of a deterministic and a stochastic component: A = ea:t+at with t a = a +"a and "a N(0;(cid:27)a) is a technology shock with a permanent impact on product t 1 t t (cid:0) (cid:24) tivity. Consistent with a growing economy and zero in(cid:135)ation in (cid:147)steady-state(cid:148), the money supply evolves according to M = ea:t+mt with (cid:1)m = (cid:26) (cid:1)m + "m + (cid:28)cb"a, (cid:26) [0;1] and t t m t 1 t t m (cid:0) 2 "m N(0;(cid:27)m): I interpret "m as an aggregate demand shock. t t (cid:24) 12

3.1.6 Closing and solving the model Averaging(cid:133)rms(cid:146)employment,totalemploymentevolvesaccordington = (1 (cid:21))n +v q((cid:18) ): t+1 t t t (cid:0) The labor force being normalized to one, the number of unemployed workers is u = 1 n t t. (cid:0) Finally, as in Krause and Lubik (2007), vacancy posting costs are distributed to the aggregate households so that C = Y in equilibrium. To solve the model, I log-linearize the (cid:133)rst-order t t conditions around the (zero-in(cid:135)ation) long run equilibrium.13 3.2 Calibration I now discuss the calibration of the parameters of the model, and Table 2 lists the parameter values. Whenever possible, I use values typically used in the literature. I set the quarterly discount factor (cid:12) to 0:99 and the returns to e¢ cient labor (cid:11) to 0:64: I assume that the markup ofpricesovermarginalcostsisonaverage10percent, whichamountstosetting"equalto11. I choose (cid:23) = 0:5 so that (cid:133)rms reset their price every 2 quarters, consistent with Bils and Klenow (2004). I set the growth rate of technology (and money supply) to a = 0:5% a quarter so that the economy is growing by 2% on average each year. I use a money growth autocorrelation parameter (cid:26) of 0:5 following Krause and Lubik (2007). Turning to the labor market, I use a m middle value for the matching function elasticity (cid:17) = 0:5 and set the bargaining weight (cid:13) = (cid:17) following the Hosios (1990) condition. The scale parameter of the matching functions m is 0 chosen such that, as reported in den Haan, Ramey and Watson (2000), a (cid:133)rm (cid:133)lls a vacancy with a quarterly probability q((cid:18)) = 0:7 and, as used in Thomas (2008), a worker (cid:133)nds a job with probability (cid:18)q((cid:18)) = 0:6. Following Shimer (2005), the separation rate is 10% so jobs last forabout2.5yearsonaverage, andtheincomereplacementratioissetto40%. Ichoose(cid:27) = 2 h (i.e. an hours per worker elasticity of 0:5) and need to decide on (cid:27) to (cid:133)x a value for ’. Bils e and Cho (1994) build a model to account for the procyclicality of labor productivity. In doing so, they allow for variable e⁄ort and variable capital utilization. The present model does not consider capital explicitly but implicitly if one assumes a constant capital-labor ratio. A key 13The equations are presented in the Appendix. 13

hypothesis of Bils and Cho (1994) is that the capital utilization rate is proportional to hours. If a worker works longer hours and at a more intense pace, the utilization of the capital he operates will also tend to increase. As a result, changes in hours per worker proxy not only for variations in e⁄ort but also for unobserved changes in capital utilization. In that case, Schor(cid:146)s (1997) estimate for the elasticity of e⁄ort with respect to hours (cid:27)h = 0:5 delivers a value 1+(cid:27)e for ’ of 1:5. I set (cid:27) accordingly in order to match this estimate.14 Finally, and consistent e with the aim of the paper to reassess Shimer(cid:146)s puzzle while controlling for the endogeneity of productivity, I set the standard deviations of technology and monetary policy shocks (cid:27)a and (cid:27)m equal to the standard deviations of technology and non-technology shocks estimated with the structural VAR.15 3.3 Simulation Figure 2 and 3 show the impulse response functions after technology shocks and monetary policy(oraggregatedemand)shocks. A(cid:133)rstobservationisthatthisNewKeynesianMPmodel ful(cid:133)lls the two necessary conditions to reassess Shimer(cid:146)s puzzle: it is successful at replicating the productivity responses to both shocks (or put di⁄erently, it can be used to control for the endogeneity of productivity), and it gets the sign of labor market tightness responses right. Nonetheless, the Shimer puzzle is apparent after both shocks: model labor market tightness moves a lot less than its empirical counterpart. However, after a non-technology shock, the standard deviation of model labor market tightness over the (cid:133)rst two years after a technology shock is almost 9 times larger than for model output per hour. Since the empirical ratio is 21, the MP model explains in fact 40% of labor market tightness (cid:135)uctuations following an aggregate demand shock. This back-ofthe-envelope calculation suggests that the misidenti(cid:133)cation of productivity shocks and the endogeneity of productivity may be responsible for some of the Shimer puzzle. Usingacalibratedversionofthemodel,Isimulate50yearsofdata,andIrepeattheexercise 14This calibration is consistent with Basu and Kimball (1997) evidence that ’ ranges between 1:28 to 1:6. 15With this calibration, the model matches the persistence and volatility of the US output per hour series. 14

5000 times. Following Shimer (2005), I detrend the model generated productivity series, and in Table 2, I report the summary statistics for the simulated labor market variables. Despite a baseline Mortensen-Pissarides structure of the labor market and a standard calibration, simulated (cid:18) is 9 times more volatile than the cyclical component of labor productivity, while the ratio comes at about 26 in US data. I conclude that the MP model can account for about a third, rather than 10 percent, of labor market tightness (cid:135)uctuations. In other dimensions, the model performs remarkably well as the cross-correlations have the right signs and are not far o⁄the true values. In particular, unemployment is only weakly correlated with productivity ( 0:24) and matches quite closely its empirical counterpart ( 0:23). (cid:0) (cid:0) However, the autocorrelation of model vacancies is 0:42 instead of 0:90 for US data. This is due to the excessively rapid response of vacancies. This problem was already pointed out by Fujita and Ramey (2004) and incorporating sunk costs for vacancy creation as in Fujita and Ramey (2004) would presumably correct this shortcoming. Similarly, this excess sensitivity of vacancies can explain the slightly too high vacancy-productivity and labor market tightnessproductivity correlations (both 0:49, compared with empirical values of 0:25 and 0:19). 3.4 Robustness Since the main result of this paper comes out of a calibration exercise, I present in Table 4 the in(cid:135)uence of key parameters on the ability of the extended MP model to generate (cid:135)uctuations in labor market variables. First, I span the range of plausible values for the elasticity of the matching function from (cid:17) = 0:24 (Hall, 2005) to 0:72 (Shimer, 2005) and (cid:133)nd that the MP model explains between roughly 25 and 50 percent of (cid:135)uctuations in (cid:18).16 The return to hours coe¢ cient is also an important parameter, and Basu and Kimball (1998) estimates that it ranges between 1:3 and 1:6. Within this interval, the MP model accounts for between 30 and 60 percent of unemployment (cid:135)uctuations. With a higher degree of price stickiness (one year), 16It is important to note that the elasticity of the matching function has a large impact on the performance of the New Keynesian MP model. However, Mortensen and Nagypal (2005) show that this is not the case for the baseline MP model: a value (cid:27) =0:44 instead of (cid:27) =0:72 barely changes the elasticity of market tightness with respect to productivity and does not help to overturn Shimer(cid:146)s conclusion. 15

the MP model accounts for almost 50% of (cid:135)uctuations in (cid:18). Finally, varying the value of the income replacement ratio from 0:4 (Shimer, 2005) to the high value used in Hagedorn and Manovskii (2005) of b = 0:9 improves the "ampli(cid:133)cation properties" of the New Keynesian MP model so much that it generates too much volatility in (cid:18): Similarly, lowering the worker(cid:146)s bargaining weight improves the performance of the MP model, and a low value (cid:13) = 0:05 as in Hagedorn and Manovskii (2005) allows the MP model to account for 40% of labor market tightness (cid:135)uctuations.17 4 Conclusion In a very important paper, Shimer (2005) argues that the Mortensen-Pissarides search model of unemployment lacks an ampli(cid:133)cation mechanism because it cannot generate the observed business cycle (cid:135)uctuations in unemployment given labor productivity shocks of plausible magnitude. Inthispaper, Ishowthatbecauseof theendogeneityofmeasuredlaborproductivity, (cid:133)ltering out the trend component of output per hour as in Shimer (2005) may not correctly identify the shocks driving unemployment. In fact, using long-run restrictions in a structural VAR model to isolate exogenous productivity shocks, I (cid:133)nd that a permanent increase in productivitylowersthevacancy-unemploymentratio, whiletheMPmodelimpliestheopposite. Iembed the MP model in a New-Keynesian framework to (i) account for this empirical evidence, and (ii) control for the endogeneity of productivity, and I estimate that the MP model can account for a third, and possibly as much as 60 percent, of (cid:135)uctuations in labor market variables. Interestingly, this (cid:133)nding is in line with the work by Pissarides (2007) who reconsiders the Shimer puzzle in the context of an MP model with endogenous job destruction. Pissarides (2007) reestimates the unemployment volatility puzzle downwards and claims that (cid:147)with endogenousjobdestruction,themodelfailstoaccountforabouthalftotwothirdsofthevolatility 17Interestingly,thisimpliesthatHagedornandManovskii(cid:146)s(2005)calibrationwithahighincomereplacement ratio and low worker(cid:146)s bargaining weight generates too much volatility in (cid:18). 16

in unemployment(cid:148)instead of the 90% originally estimated by Shimer (2005). If a third of the Shimer puzzle is due to the misidenti(cid:133)cation of productivity shocks and another 30 to 50 percent is due to the omission of endogenous job destruction, the low volatility of unemployment relative to that of productivity may be less of a problem than originally thought. 17

Appendix Log-linearized equilibrium dynamics To analyze the behavior of the economy, I log-linearize the (cid:133)rst-order conditions around the (zero-in(cid:135)ation) long run equilibrium. The optimal vacancy posting condition takes the form c c t t+1 = E (cid:12) (cid:31) + (1 (cid:21)) (3) q((cid:18) ) t t+1 it+1 q((cid:18) ) (cid:0) t t+1 (cid:20) (cid:21) with (cid:31) , the shadow value of a marginal worker, given by it 1+(cid:27) h1+(cid:27)h (cid:31) it = (cid:0) w it +(1 (cid:0) (cid:13)){ ’ h i (cid:21) t t Since 1 is the expected duration of a vacancy, equation (3) has the usual interpretation: q((cid:18)t) each (cid:133)rm posts vacancies until the expected cost of hiring a worker ct equals the expected q((cid:18)t) discounted future bene(cid:133)ts (cid:31) it+j 1 j=1 from an extra worker. Because the (cid:133)rm is demand constrained, the (cid:135)ow value(cid:8)of a m(cid:9)arginal worker is not his contribution to revenue but his reduction of the (cid:133)rm(cid:146)s wage bill. The (cid:133)rst term of (cid:31) is the wage payment going to an extra it worker, while the second term represents the savings due to the decrease in hours and e⁄ort achieved with that extra worker. Log-linearizing the vacancy posting condition equation around the (zero-in(cid:135)ation) steady state, I get for any t > 0 c(cid:17) c(1 (cid:21))(cid:17) q((cid:18) ) ^(cid:18) t = E t (cid:12) (cid:31) (cid:3) (cid:31)^ it+1 + q( (cid:0) (cid:18) ) ^(cid:18) t+1 (cid:3) (cid:3) (cid:20) (cid:21) with the value of a marginal worker (cid:31)^ given by it+1 1 1+(cid:27) (cid:31) (cid:31)^ = (cid:13)c(cid:18)^(cid:18) + ( h 1)(y^ n^ ) (cid:3) it+1 (cid:0) t n(cid:22) ’ (cid:0) it+1 (cid:0) it+1 18

With Calvo-type price setting, a (cid:133)rm resetting its price at date t will satisfy the standard Calvo price setting condition: E 1 (cid:23)j(cid:12) P i(cid:3)t (cid:22)s Y P" = 0 t j P (cid:0) it+j t+j t+j t+j j=0 (cid:20) (cid:21) X where the optimal mark-up is (cid:22) = " and the (cid:133)rm(cid:146)s real marginal cost " 1 (cid:0) 1+(cid:27) Y s it = ’ h (1 (cid:0) (cid:13)){ A t h 1 it +(cid:27)h (cid:0) ’ t The (cid:133)rm will choose a price P that is, in expected terms, a constant mark-up (cid:22) over its real i(cid:3)t marginal cost for the expected lifetime of the price. ToderivetheNew-KeynesianPhillipscurve,Ilog-linearizearoundthezeroin(cid:135)ationequilibrium. However,becauseof(cid:133)rms(cid:146)ex-postheterogeneity,thederivationisnotasstraightforward as with costly price adjustment. I follow Woodford(cid:146)s (2004) similar treatment of endogenous capital in a New-Keynesian model with Calvo price rigidity. In my case, employment is the state variable and plays the role of capital in Woodford(cid:146)s model. The price-setting condition becomes 1 ((cid:23)(cid:12))kE^i[p~ s^ ] = 0 (4) t it+k it+k (cid:0) k=0 X with 1+(cid:27) h s^ = n^ + (y^ n^ ) y^ +y^ (5) it+k it+k it+k it+k it+k t+k ’ (cid:0) (cid:0) The notation E^i denotes an expectation conditional on the state of the world at date t but t integrating only over future states in which (cid:133)rm i has not reset its price since period t: p~ it (cid:17) log Pit is the (cid:133)rm(cid:146)s relative price. Pt (cid:16) (cid:17) Denoting log prices by lower-case letters and p the optimal (log) price for (cid:133)rm i at t, the (cid:3)it demand curve for (cid:133)rm i at date t+1 can be written y^ = y^ "(p p ) if it cannot it+1 t+1 it t+1 (cid:0) (cid:0) reset its price at t+1 and y^ = y^ "(p p ) if it can reset its price. it+1 t+1 (cid:0) (cid:3)it+1(cid:0) t+1 19

Averaging across all (cid:133)rms, I get 1 1 1 y^ di = y^ " (cid:23)( p di p )+(1 (cid:23))( p di p ) it+1 t+1 it t+1 (cid:3)it+1 t+1 (cid:0) 2 (cid:0) (cid:0) (cid:0) 3 Z Z Z 0 0 0 = y^ "4(cid:23)(p p )+(1 (cid:23))(p p ) 5 (6) t+1 t t+1 (cid:3)t+1 t+1 (cid:0) (cid:0) (cid:0) (cid:0) (cid:2) (cid:3) 1 where p = p di is the average price chosen by all price setters at date t+1. (cid:3)t+1 (cid:3)it+1 Z 0 With Calvo price-setting, I can write 1 p = (1 (cid:23))p 1 "+(cid:23)p1 " 1 " t+1 (cid:0) t(cid:3)+(cid:0)1 t(cid:0) (cid:0) (cid:0) (cid:1) or 1 = (1 (cid:23)) p (cid:3)t+1 1 (cid:0) " +(cid:23) p t 1 (cid:0) " : (cid:0) p p t+1 t+1 (cid:18) (cid:19) (cid:18) (cid:19) Log-linearizing around the zero-in(cid:135)ation equilibrium gives (cid:23)(p p ) = (1 (cid:23))(p p ) (cid:0) t+1 (cid:0) t (cid:0) (cid:3)t+1(cid:0) t+1 1 1 and combining with (6) gives y^ di = y^ . Further, n^ di = n^ . it+1 t+1 it t Z Z 0 0 Averaging (5) across all (cid:133)rms, I can rewrite the real marginal cost as 1+(cid:27) h s^ = s^ + 1 ( "p~ n~ ) (7) it+k t+k it+k it+k ’ (cid:0) (cid:0) (cid:0) (cid:18) (cid:19) where n~ = n n is the relative employment of (cid:133)rm i. it+k it+k t+k (cid:0) Using that E^ip~ = p E p and (7) in (5) yields t it+k it t t+k (cid:0) 1+" 1+(cid:27) h 1 p = (1 (cid:23)(cid:12)) 1 ((cid:23)(cid:12))kE^i s^ + 1+" 1+(cid:27) h 1 p 1+(cid:27) h 1 n~ ’ (cid:0) (cid:3)it (cid:0) t t+k ’ (cid:0) t+k (cid:0) ’ (cid:0) it+k (cid:18) (cid:18) (cid:19)(cid:19) k=0 (cid:20) (cid:18) (cid:18) (cid:19)(cid:19) (cid:18) (cid:19) (cid:21) X (8) 20

Moreover, subtracting (??) from its average, I get n~ = E (y^ y^ ) (9) it+1 t it+1 t+1 (cid:0) = "E (cid:23)(p p )+(1 (cid:23))(p p t it t+1 (cid:3)it+1 t+1 (cid:0) (cid:0) (cid:0) (cid:0) (cid:2) (cid:3) = "(cid:23)p~ "(1 (cid:23))(p p ) it (cid:3)it+1 (cid:3)t+1 (cid:0) (cid:0) (cid:0) (cid:0) since p = (cid:23)p +(1 (cid:23))p . t+1 t (cid:0) (cid:3)t+1 The(cid:133)rm(cid:146)spricingdecisiondependsonitsemploymentlevelandtheeconomy(cid:146)saggregatestate. Buttoa(cid:133)rstorder,thelog-linearizedequationsarelinearsothatthedi⁄erencebetweenp and (cid:3)it p , the average price chosen by all price setters, is independent from the economy(cid:146)s aggregate (cid:3)t state and depends only on the relative level of employment n n = n~ . So as in Woodford it t it (cid:0) (2004), I guess that the (cid:133)rm(cid:146)s pricing decision takes the form p p = (cid:15)n~ (10) (cid:3)it (cid:3)t it (cid:0) (cid:0) with (cid:15) a constant to be determined. Hence, (9) becomes "(cid:23) n~ = (cid:0) p~ = f((cid:15))p~ it+1 it it 1 "(1 (cid:23))(cid:15) (cid:0) (cid:0) (cid:0) Since this was shown for any t > 0, I also get n~ = f((cid:15))p~ , k > 0 so that I can it+k it+k 1 (cid:0) (cid:0) 8 rewrite (8) as (cid:30)p = (1 (cid:23)(cid:12)) 1 ((cid:23)(cid:12))kE^i s^ + 1+" 1+(cid:27) h 1 p (1 (cid:23)(cid:12)) 1+(cid:27) h 1 n~ (cid:3)it (cid:0) t t+k ’ (cid:0) t+k (cid:0) (cid:0) ’ (cid:0) it k=0 (cid:20) (cid:18) (cid:18) (cid:19)(cid:19) (cid:21) (cid:18) (cid:19) X (11) with (cid:30) = 1+" 1+(cid:27)h 1 (cid:23)(cid:12) 1+(cid:27)h 1 f((cid:15)) . ’ (cid:0) (cid:0) ’ (cid:0) (cid:16) (cid:16) (cid:17) (cid:16) (cid:17) (cid:17) Subtracting (11) from its average, I obtain 1+(cid:27) h (cid:30)(p p ) = (1 (cid:23)(cid:12)) 1 n~ : (12) (cid:3)it (cid:0) (cid:3)t (cid:0) (cid:0) ’ (cid:0) it (cid:18) (cid:19) 21

This equation is of the conjectured form (10) if and only if (cid:15) satis(cid:133)es (1 (cid:23)(cid:12))1+(cid:27)h 1 (cid:15) = (cid:0) ’ (cid:0) : (13) 1+" 1+(cid:27)h 1 (cid:23)(cid:12) 1+(cid:27)h 1 f((cid:15)) ’ (cid:0) (cid:0) ’ (cid:0) (cid:16) (cid:17) (cid:16) (cid:17) Finally, averaging (11) and using (cid:25) = 1 (cid:23)(p p ), I obtain the New-Keynesian Phillips curve t (cid:0)(cid:23) (cid:3)t (cid:0) t (cid:25) = (cid:14):s^ +(cid:12)E (cid:25) t t t t+1 (1 (cid:23))(1 (cid:23)(cid:12)) with (cid:14) = (cid:0) (cid:0) : (cid:23)(cid:30) 22

References [1] Andolfatto, D. (cid:147)Business Cycles and Labor-Market Search,(cid:148)American Economic Review, 86(1), 1996. [2] Barnichon, R. (cid:147)Productivity, Aggregate Demand and Unemployment Fluctuations,(cid:148)FinanceandEconomicsDiscussionSeries2008-47, BoardofGovernoroftheFederalReserve System, 2008. [3] Basu, S. and M. Kimball. (cid:147)Cyclical Productivity with Unobserved Input Variation,(cid:148) NBER Working Papers 5915, 1997. [4] Bils, M. and J. Cho. (cid:147)Cyclical Factor Utilization,(cid:148)Journal of Monetary Economics, 33, 319-354, 1994. [5] Burnside, C., Eichenbaum, M. (cid:147)Factor-Hoarding and the Propagation of Business-Cycle Shocks,(cid:148)American Economic Review, 86(5), pp. 1154-1174, 1996. [6] Burnside, C., Eichenbaum, M. and S. Rebelo. (cid:147)Labor Hoarding and the Business Cycle,(cid:148) Journal of Political Economy, 01(2), pp. 245-73, 1993. [7] Canova, F., C. Michelacci and D. L(cid:243)pez-Salido, (cid:147)Schumpeterian Technology Shocks,(cid:148) mimeo CEMFI, 2007. [8] Fujita, S. and G. Ramey. (cid:147)Job Matching and Propagation,(cid:148)Journal of Economic Dynamics and Control, Forthcoming. [9] Gal(cid:237), J. (cid:147)Technology, Employment and The Business Cycle: Do Technology Shocks Explain Aggregate Fluctuations?,(cid:148)American Economic Review, 89(1), 1999. [10] Gal(cid:237), J and P. Rabanal. (cid:147)Technology Shocks and Aggregate Fluctuations: How Well Does the RBC Model Fit Postwar U.S. Data?,(cid:148)NBER Macroeconomics Annual, 2004. 23

[11] Hagedorn, M. and I. Manovskii. (cid:147)The Cyclical Behavior of Equilibrium Unemployment and Vacancies Revisited,(cid:148)Univerity of Pennsylvania mimeo. [12] Hall, R. (cid:147)Employment Fluctuations with Equilibrium Wage Stickiness,(cid:148)American Economic Review, 95(1), pp. 50-65, 2005. [13] Hall, R. (cid:147)Job Loss, Job Finding, and Unemployment in the U.S. Economy over the Past Fifty Years,(cid:148)NBER Macroeconomics Annual, pp. 101-137, 2005. [14] Hall,R.andP.Milgrom.(cid:147)TheLimitedIn(cid:135)uenceofUnemploymentontheWageBargain,(cid:148) American Economic Review, 98(4), pp. 1653-1674, 2008. [15] Krause, M and T. Lubik. (cid:147)The (Ir)relevance of Real Wage Rigidity in the New Keynesian Model with Search Frictions(cid:148), Journal of Monetary Economics, 54(3), 2007. [16] Mertz,M.(cid:147)SearchintheLaborMarketandtheRealBusinessCycle,(cid:148)JournalofMonetary Economics, 49, 2002. [17] Mortensen, D. and E. Nagypal. (cid:147)More on Unemployment and Vacancy Fluctuations,(cid:148) NBER Working Papers 11692, 2005. [18] Mortensen, D and C. Pissarides. (cid:147)Job Creation and Job Destruction in the Theory of Unemployment,(cid:148)Review of Economic Studies, 61, 397-415, 1994. [19] Pissarides, C. (cid:147)The Unemployment Volatility Puzzle: Is Wage Stickiness the Answer?,(cid:148) The Walras-Bowley lecture, North American Summer Meetings of the Econometric Society, Duke University, June 21-24, 2007 [20] Schor J., (cid:147)Does Work Intensity Respond to Macroeconomic Variables? Evidence from British Manufacturing,(cid:148)Working Paper, 1987. [21] Shimer,R.(cid:147)TheConsequencesofRigidWagesinSearchModels,(cid:148)Journal of the European Economic Association (Papers and Proceedings), 2: 469-479, 2004. 24

[22] Shimer, R. (cid:147)The Cyclical Behavior of Equilibrium Unemployment and Vacancies,(cid:148)American Economic Review, 95(1), pp. 25-49, 2005. 25

Table1:Summary Statistics, QuarterlyUS Data, 1951 2005 u v θ p Standard deviation 0.125 0.139 0.257 0.010 Quarterly 0.87 0.90 0.89 0.69 autocorrelation u 1 0.95 0.97 0.23 v 1 0.98 0.25 Correlation matrix θ 1 0.19 p 1 Notes:Seasonally adjusted unemploymentuis constructed by the BLS from the Current Population Survey (CPS). The seasonally adjusted help wanted advertising indexvis constructed by the Conference Board.Labor market tightness is the vacanc unemployment ratio. Average labor productivitypis seasonally adjusted real average output per person in the non farm business sector, constructed by the Bureau of Labor Statistics (BLS) from the National Income and Product Accounts and the Current Employment Statistics.All variables are reported in logs as deviations from an HP trend with smoothing parameter 1600. Table 2:Calibration Discount rate β=0.99 Matching function elasticity σ=0.5 Bargaining weight γ=σ Hosios (1990) den Haan, Ramey and Probability vacancy is filled q(θ)=0.7 Watson (2000) Job finding probability θq(θ)=0.6 Thomas (2008) Separation rate λ=0.1 Shimer (2005) Income replacement ratio b=0.4 Shimer(2005) Returns to scale to efficient α=0.64 hours Disutility of hours σ=2 h Short run increasing returns φ=1.5 Schor (1997) to hours Growth rate of 2% a year a=0.5% 0 ν=0.5 Degree of price stickiness Bils and Klenow (2004) (2 quarters) Mark up of 10% ε=11 AR(1) process for money ρ =0.5 Krause and Lubik (2007) growth m Standard deviation of σ =0.0226 monetary policy shock m Standard deviation of σ=0.006 technology shock a 26

Table3: Summary Statistics,Model u v θ p 0.042 0.065 0.089 0.010 Standard deviation (0.004) (0.004) (0.008) (0.001) 0.85 0.42 0.74 0.69 Quarterly autocorrelation (0.03) (0.06) (0.05) (0.05) 0.35 0.72 0.24 u 1 (0.07) (0.04) (0.10) 0.90 0.49 v 1 (0.01) (0.07) Correlation matrix 0.49 θ 1 (0.08) p 1 Notes:Standard errors the standard deviation across5000 model simulations are reported in parentheses. Table4:Robustness Share of(cid:190)US(θ) Parameter (cid:190)MP(θ) explained by value MP model Matching function 0.24 0.11 47% η elasticity 0.72 0.61 25% 1.3 0.15 60% Return to hours ’ 1.6 0.07 27% Degree of price stickiness ” 0.75 0.12 47% 0.2 0.61 25% Income replacement ratio b 0.9 0.29 116% Workers’ bargaining 0.5 0.85 33% (cid:176) weight 0.05 0.10 40% Notes:(cid:190)MP(θ) isthe model generated standard deviation of labor market tightness and(cid:190)US(θ) is its empirical counterpart. 27

3 x 10 8 6 4 2 0 0 2 4 6 8 10 12 14 16 18 20 Output per Hour kcohS ygolonhceT 0.05 0 0.05 0.1 0.15 0 2 4 6 8 10 12 14 16 18 20 Labor Market Tightness 3 x 10 10 5 0 5 0 2 4 6 8 10 12 14 16 18 20 Output per Hour kcohS ygolonhceT noN 0.2 0.15 0.1 0.05 0 0.05 0 2 4 6 8 10 12 14 16 18 20 Labor Market Tightness Figure 1: Impulse response functions to one s.d. shocks. Dashed lines represent the 95% con(cid:133)dence interval. 28

3 x 10 10 0.02 0.015 5 0.01 0 0.005 5 0 0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 Output per hour Output 0.2 0 0.15 0.005 0.1 0.01 0.05 0.015 0 0.02 0.05 0.025 0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 Labor Market Tightness Unemployment Figure 2: Model (dotted line) and Empirical (plain line) impulse response functions to a positive monetary policy shock. Dashed lines represent the 95% con(cid:133)dence interval. 29

3 3 x 10 x 10 8 6 6 4 4 2 2 0 0 0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 Output per Hour Output 3 x 10 0.05 4 0 3 0.05 2 0.1 1 0.15 0 0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 Labor Market Tightness Unemployment Figure 3: Model (dotted line) and Empirical (plain line) impulse response functions to a positive technology shock. Dashed lines represent the 95% con(cid:133)dence interval. 30