Global Imbalances and Structural Change in the United States

Transcription

1 Global Imbalances and Srucural Change n he Uned Saes Tmohy J. Kehoe Unversy of Mnnesoa, Federal Reserve Bank of Mnneapols, and Naonal Bureau of Economc Research Km J. Ruhl Pennsylvana Sae Unversy Joseph B. Senberg Unversy of Torono Saff Repor 489 Revsed January 2018 DOI: hps://do.org/ /sr.489 Keywords: Global mbalances; Real exchange rae; Srucural change JEL classfcaon: E13, F34, O41 The vews expressed heren are hose of he auhors and no necessarly hose of he Federal Reserve Bank of Mnneapols or he Federal Reserve Sysem. Federal Reserve Bank of Mnneapols 90 Hennepn Avenue Mnneapols, MN hps://

2 Federal Reserve Bank of Mnneapols Research Deparmen Saff Repor 489 Revsed: January 2018 (Frs verson: Aug 2013) Global Imbalances and Srucural Change n he Uned Saes* Tmohy J. Kehoe Unversy of Mnnesoa, Federal Reserve Bank of Mnneapols, and Naonal Bureau of Economc Research Km J. Ruhl Pennsylvana Sae Unversy Joseph B. Senberg Unversy of Torono ABSTRACT Snce he early 1990s, as he Uned Saes borrowed heavly from he res of he world, employmen n he U.S. goods-producng secor has fallen. We consruc a dynamc general equlbrum model wh several mechansms ha could generae declnng goods-secor employmen: foregn borrowng, nonhomohec preferences, and dfferenal producvy growh across secors. We fnd ha only 15.1 percen of he declne n goods-secor employmen from 1992 o 2012 sems from U.S. rade defcs; mos of he declne s due o dfferenal producvy growh. As he Uned Saes repays s deb, s rade balance wll reverse, bu goods-secor employmen wll connue o fall. Keywords: Global mbalances; Real exchange rae; Srucural change JEL classfcaon: E13, F34, O41 *We hank Davd Back, Davd Sras, Harald Uhlg, Frank Warnock, Ke-Mu Y, Vvan Yue, Jng Zhang, and four anonymo referees for helpful dscsons and suggesons. Kehoe and Ruhl graefully acknowledge he suppor of he Naonal Scence Foundaon under gran SES We are graeful o Jack Rossbach for exraordnary research asssance. The daa and compuer code ed here can be found n he supplemenary maerals avalable a The vews expressed heren are hose of he auhors and no necessarly hose of he Federal Reserve Bank of Mnneapols or he Federal Reserve Sysem.

3 1. Inroducon Beween 1992 and 2012, he U.S. ne nernaonal nvesmen poson deeroraed by $4 rllon as hoeholds and he governmen n he Uned Saes borrowed from he res of he world. Durng hs same perod, he goods-secor share of oal employmen fell sharply. Economc heory predcs ha rade defcs drve employmen ou of domesc goods-producng secors, and a commonly held vew n polcy crcles, expressed, for example, n Sco (2015), s ha U.S. rade defcs generaed by hs borrowng have played an mporan role n he declne n U.S. goodssecor employmen and ha rebalancng U.S. rade wll reverse a large par of hs rend. In hs paper, we provde he frs model-based esmaes of he mpac of rade defcs on U.S. goodssecor employmen. We e a dynamc general equlbrum model of he Uned Saes and he res of he world o address he quesons: To wha exen are rade defcs responsble for he loss of U.S. goodssecor employmen? Wll employmen reurn o goods-producng secors when U.S. borrowng ends and rade defcs become rade surples? Our framework combnes an open-economy model of neremporal rade wh a model of srucural change he secular shf of employmen from goods-producng ndres o servcesproducng ndres. Our model feaures wo man ngredens from he srucural change leraure: ncome effecs from nonhomohec preferences, and faser labor producvy growh n he goods-producng secor compared wh he servces secor combned wh a low elascy of subsuon beween goods and servces. Typcally, srucural change has been suded n closedeconomy sengs; we exend he analyss o an open economy wh unbalanced rade. The key open economy feaure n our model s he savng glu ncreased demand for savng by he res of he world whch we calbrae o mach he U.S. rade balance. We e our model o quanavely assess he conrbuons of radonal srucural change forces and he savng glu o he declne n U.S. goods-secor employmen. Conrary o asserons by analyss such as Sco (2015), who wres ha rade [defcs], no producvy, s he culpr [behnd goods-secor job loss], we fnd ha he savng glu only explans beween 11 and 20 percen of he declne n U.S. goods-secor employmen; our preferred esmae s 15.1 percen. We calbrae our model s producon and demand srucure so ha he model replcaes he 1992 U.S. naonal accouns and a world npu-oupu marx we consruc from he World Inpu- Oupu Daabase (Tmmer e al., 2015). We feed no he model wo exogeno drvng forces: 1

4 labor producvy growh raes ha dffer across secors and demand for savng n he res of he world. The equlbrum oucome of hs model our benchmark savng-glu equlbrum accouns for 99.2 percen of he observed declne n he goods-secor employmen share from 1992 o 2012 and accouns for several oher key facs abou he U.S. economy durng hs perod. We hen compare hs equlbrum wh one n whch foregn demand for savng s consan our no-savng-glu counerfacual. The counerfacual accouns for 84.1 percen of he declne n he goods-secor employmen share. The dfference beween hese wo fgures, 15.1 percen, s he conrbuon of he savng glu o hs declne. Ths mples ha elmnang he rade defc wll no generae a sgnfcan ncrease n goods-secor employmen. In fgure 1, we plo he share of goods-producng secors agrculure, mnng, and manufacurng n oal employmen, whch has fallen as he rade defc has grown. The economc heory ha lnks hese daa depend on mpored goods beng subsuable for domescally produced goods. As he Uned Saes rades bonds for foregn goods, labor shfs away from domescally produced goods and oward servces and consrucon, whch are less subsuable for foregn goods. Ths mechansm mples ha, when he deb has o be repad, labor wll flow back no he U.S. goods secor o produce he exra goods needed o repay he deb. The quanave mporance of hs channel n drvng he declne n U.S. goods-secor employmen over he pas wo decades s he foc of our sudy. There s ample hsorcal evdence connecng secoral reallocaon and rade balance dynamcs. Afer Span enered he European Communy n 1986, s rade balance deeroraed subsanally, whle he goods-secor share of Spansh GDP fell from 40 percen o less han 30 percen (Fernandez de Cordoba and Kehoe, 1999). The Balc counres had smlar experences afer lberalzng capal markes n he 1990s (Bems and Jönsson, 2005). In emergng economes lke Mexco, sudden sops of he world s wllngness o lend are ypcally accompaned by sharp reallocaons away from servces producon and no goods (Kehoe and Ruhl, 2009). Alhough he savng glu s no as drasc as capal accoun lberalzaon or reversals due o sudden sops, s plable ha 20 years of ncreased capal nflows could have a sgnfcan effec on goodssecor employmen. In consrucng our model, we need o specfy he drvng force behnd U.S. borrowng. A common explanaon s ha foregn demand for savng ncreased, makng foregners more wllng o rade her goods for U.S. bonds. Bernanke (2005) coned he erm global savng glu o refer o 2

5 hs mechansm, and we adop Bernanke s global-savng-glu hypohess. Several explanaons have been proposed for he ncreased demand for savng n he res of he world, such as a lack of fnancal developmen n he res of he world (Caballero e al., 2008; Mendoza e al., 2009) and dfferences n bness cycle or srucural growh properes (Back e al., 2006; Perr and Fogl, 2010; Jn, 2012). We do no ake a sand on whch of hese explanaons are correc. Insead, we ake he savng glu as gven and calbrae a process for hoehold preferences n he res of he world over curren vers fuure consumpon so ha our model maches exacly he pah of he U.S. rade balance durng The drvng force behnd srucural change n he model s dfferenal labor producvy growh. As evden n fgure 2, labor producvy n he goods secor has grown a a faser pace han n oher secors over he pas wo decades. The srucural change leraure emphaszes asymmerc producvy growh as an mporan drver of he long-run reallocaon of labor across secors (Nga and Pssardes, 2007). We ake a smlar approach n an open-economy model. Several oher recen papers sudy srucural change n open economes (Echevarra, 1995; Masuyama, 1992, 2009; Spos, 2012; Uy, Y, and Zhang, 2013). Wh he excepon of Spos (2012), hese sudes e models of balanced rade, absracng from nernaonal capal flows. Our model enrches he dynamcs n Spos (2012) by dsngushng beween hoehold and governmen deb, endogenzng he nvesmen rae, and allowng counres o hold deb on he balanced growh pah. Buera and Kabosk (2009) denfy hree sources of srucural change, all of whch we nclude n our baselne model: dfferenal secoral producvy growh and a low elascy of subsuon beween secors (Nga and Pssardes, 2007); nonhomohec preferences (Kongsamu e al., 2001); and capal deepenng drven by dfferen capal shares across secors (Acemoglu and Guerrer, 2008). We fnd ha dfferenal capal shares are no sgnfcan sources of srucural change n our model. Nonhomohec preferences accoun for 5.6 percen of srucural change, conssen wh Buera and Kabosk (2009), who argue ha, becae subssence requremens are mos mporan a low ncome levels, ncome effecs are unlkely o be mporan n explanng U.S. srucural change n recen decades. Boppar (2014) and Herrendorf e al. (2013) fnd larger roles for ncome effecs over longer me horzons. We nclude several oher feaures n our model ha make well-sued o address our quesons. Frs, ncludes hree secors: goods, servces, and consrucon. Goods and servces n 3

6 our model are boh raded, n conras o he ual convenon n nernaonal macroeconomcs whch assumes only goods are radable. The ual convenon s a odds wh he daa: servces are a large componen of U.S. expors. Consrucon s he only purely nonradable secor and s ed enrely o produce nvesmen goods, whch means ha consrucon s more sensve han he oher secors o he effecs of capal flows and economc flucuaons n general. Second, our model ncludes a global npu-oupu srucure. Frms e nermedae npus of boh goods and servces, produced boh a home and abroad, and we dsngush beween rade n fnal vers nermedae goods and servces. Ths producon and demand srucure allows for subsuon beween domesc producs and her foregn equvalens, and complemenary beween producs from dfferen secors. We also allow for more subsuon beween domesc and foregn goods han servces, whch allows o mach he volales n he goods rade balance and he servces rade balance. 2. The U.S. economy, We vew he massve foregn borrowng and he dfferences n producvy growh across secors as exogeno drvng forces ha we ake as npus no he model. We presen hree key facs abou U.S. daa ha we vew as ess for our model. Our model s ably o capure hese facs n response o he exogeno drvng forces gves confdence n ng he model o perform counerfacual expermens and make predcons abou he fuure of he U.S. economy. Exogeno drvng force 1: Foregn borrowng ncreased, hen decreased Fgure 3 llraes how much borrowng hoeholds and he U.S. governmen have done. The curren accoun balance measures capal flow no he Uned Saes, bu we see ha he rade balance racks he curren accoun balance almos exacly. Snce our model s no desgned o accuraely capure he dfference beween hese wo seres, we e he rade defc as our measure of U.S. borrowng. Fgure 3 shows ha beween 1992 and 2006, he rade defc grew seadly, reachng 5.8 percen of GDP, afer whch began o shrnk. In 2012, he rade defc was 3.6 percen of GDP. We vew he pah of he rade defc as wha defnes he savng glu n our model. Capal flows are mporan o our analyss n wo ways. Frs and foremos, he U.S. rade defc, whch s composed prmarly of goods (fgure 4), drecly affecs he need for labor o produce goods domescally. Second, accumulaed deb evenually needs o be repad, whch could affec he employmen needed o produce goods n he Uned Saes n he fuure. In our model, cumulave rade balances are he measure of hs deb, whch dffers from he U.S. ne foregn 4

7 asse poson by any revaluaon effecs. These revaluaon effecs have, a mes, played a sgnfcan role n he value of he U.S. ne foregn asse poson (Lane and Mles-Ferre, 2008; Gournchas and Rey, 2007). Revaluaon has been n favor of he Uned Saes overall; he U.S. ne foregn asse poson s smaller han cumulaed rade balances n he daa. Ths mples ha fuure reallocaon of labor back o goods-producng secors caed by deb repaymen may be smaller han our model predcs. Exogeno drvng force 2: Labor producvy grew fases n he goods secor Durng , labor producvy n he goods secor real value added per worker grew a an average of 4.4 percen per year, whle grew by only 1.3 percen per year n servces and fell by 0.84 percen per year n consrucon. Wha s essenal n our model s he dfferenal beween producvy growh n goods compared wh servces. As he daa n fgure 2 show, hs dfferenal has been close o consan snce 1980, wh average producvy growh of 4.3 percen per year n goods durng , compared wh 1.1 percen n servces. Excep for he producvy slowdown of he 1970s, he dfferenal has perssed snce Fac 1: The real exchange rae apprecaed, hen deprecaed Fgure 3 presens he frs key fac ha we ask our model o mach. The fgure shows ha he U.S. real exchange rae was volale and racked he rade balance closely durng We consruc our measure of he U.S. real exchange rae by akng a weghed average of blaeral real exchange raes wh he Uned Saes 20 larges radng parners, wh weghs gven by hese counres shares of U.S. mpors n 1992 (oher weghng schemes yeld smlar resuls). The real exchange rae apprecaed by 27.9 percen beween 1992 and 2002, afer whch deprecaed by 22.1 percen beween 2002 and The nuon for why he real exchange rae and rade balance should move closely s sraghforward: as foregn goods and servces become cheaper, U.S. hoeholds buy more of hem. Noce, however, ha he mng s off: he maxmum apprecaon of he real exchange rae occurred n 2002, whereas he larges rade defc occurred n Our baselne model s unable o replcae hs paern; we revs n our sensvy analyss and concludng remarks. Fac 2: The goods secor drove aggregae rade balance dynamcs Fgure 4 presens our second key fac, plong dsaggregaed rade balances for goods and servces separaely. The goods rade balance generaes mos of he flucuaons n he aggregae rade 5

8 balance, whle he servces rade balance flucuaes beween 0.5 and 1.2 percen of GDP. The conssen U.S. servces rade surpl movaes one of he key feaures of our modelng framework. Sandard modelng convenons n nernaonal macroeconomcs lump servces ogeher wh consrucon no a nonradable secor, reang goods as he only secor ha produces oupu ha can be raded nernaonally. By conras, we allow boh goods and servces o be raded n our model. Fac 3: Employmen n goods declned seadly; consrucon employmen rose, hen fell Fgure 5 presens our hrd fac: beween 1992 and 2012, he fracon of oal labor compensaon pad o he goods secor fell from 19.7 percen o 12.5 percen. The fracon of oal labor compensaon s our preferred measure of a secor s employmen share becae maps drecly no our model, where we measure labor npus n erms of effecve hours raher han raw hours. As fgure 1 shows, hs measure moves closely wh more common measures such as he share of employees n he goods secor. The consrucon-secor share of labor compensaon rose from 4.4 percen n 1992 o 5.7 percen n 2006, as consrucon boomed pror o he fnancal crss n Employmen n consrucon hen sared o fall, and by 2012, he consrucon-secor share of labor compensaon was agan 4.4 percen. Reallocaon away from goods has, h far, been permanen, whle reallocaon no consrucon was emporary. 3. Model Our model has wo counres: he Uned Saes () and he res of he world (rw). A perod s one year. Represenave hoeholds n each counry work, consume, and save o maxmze uly subjec o a sequence of budge consrans. Compeve frms produce commodes ha serve fnal and nermedae es boh a home and abroad. Agens n our model have perfec foresgh excep for n he frs perod, We model he savng glu as an unancpaed and emporary hough decades-long change n he res of he world s wllngness o lend. Ths assumpon capures our vew ha U.S. hoeholds n he early 1990s dd no ancpae he large rade defcs her counry would run over he nex wo decades. 3.1 Gross oupu producon The Uned Saes produces goods ( g ), servces () s, and consrucon ( c ); he res of he world produces only goods and servces. Gross oupu s produced accordng o a nesed consan 6

9 elascy of subsuon (CES) echnology ha es capal, labor, and nermedae npus of goods and servces from boh counres. As n he daa, consrucon s no ed as an nermedae npu; s only ed n he producon of nvesmen goods. We e superscrps o ndex counres or fnal demand secors, desnaon frs, and subscrps o ndex producon secors, desnaon frs. To smplfy exposon, we descrbe only he U.S. economy below. Excep where noed, we model he res of he world analogoly. Gross oupu n secor, y, s an aggregae of value added, v, and an nermedae bundle, m, 1 (1) y v (1 ) m. The parameer governs he share of value added n gross oupu, and governs he elascy of subsuon beween value added and he nermedae npu bundle. facor ed o faclae calbraon. s a consan scalng Value added n secor s a Cobb-Douglas aggregae of capal, k, and labor,, (2) v A k 1, where s capal s share of value added, s labor producvy, and facor. We absrac from capal n he res of he world, (3) v A. rw rw rw rw s a consan scalng Labor producvy s exogeno and grows a dfferen raes n each secor and counry. The labor rw producvy process (, ) s where we nroduce our frs exogeno drvng force: asymmerc producvy growh. We e bars o dsngush labor producvy and oher exogeno me-varyng parameers from consan parameers and equlbrum varables. The compose nermedae bundle m s composed of nermedae purchases from each raded secor j n each counry,, rw ( j, j ) m m, 1 j j. jg, s (4), j (1 ), rw m j j mj j mj Inermedae goods are aggregaed n wo sages. Frs, nermedae npus from each counry are aggregaed whn he goods and servces secors separaely. Second, goods and servces 7

10 nermedaes are combned o form he aggregae nermedae bundle ed n ndry. Consder, for example, he e of res and seel n he producon of auomobles. Car producers frs combne domesc and foregn res no a re bundle, and do he same for seel. The re bundle and seel bundle are hen combned o form he oal nermedae bundle. The parameer s a consan scalng facor. j governs he share of secor- j nermedaes n he aggregae nermedae bundle, and j govern he domescally sourced share of secor- j nermedaes n he secor- nermedae bundle. governs he elascy of subsuon beween nermedaes from dfferen secors, whle j governs he elascy of subsuon beween secor- j nermedaes from dfferen source counres. Ths specfcaon allows o calbrae a producon srucure n whch goods and servces are srong complemens n nermedae producon (1/(1 ) 0 ), bu an npu from he res of he world s a subsue for an npu from he Uned Saes ( /(1 ) 1). 1 j Gross oupu s produced by perfecly compeve frms. A represenave frm n secor n he Uned Saes chooses capal, labor, and nermedae npus o maxmze profs,, rw, rw (5) q y w rk k qj mj qj mj, jg, s subjec o nonnegavy consrans and (1), (2), and (4). The wage s w, r s he capal renal k rae, and he prces of secor- gross oupu sourced from he wo counres are n he res of he world face a smlar problem bu do no e capal o produce. rw q and q. Frms 3.2 Fnal demand The Uned Saes has hree ypes of fnal demand: prvae consumpon ( h ), publc consumpon ( g ), and nvesmen ( x ). The res of he world has only consumpon. Hoeholds n boh counres and he U.S. governmen consume goods and servces produced a home and abroad. Invesmen n he Uned Saes s produced ng goods and servces from home and abroad, and wh local consrucon servces. Smlar o he aggregaon scheme for nermedae npus, fnal demand for oupu from secor s a compose of domesc and foregn producs. To allow for dfferences n shares and elasces across fnal es, we consruc hese composes separaely for each caegory of fnal 8

11 demand. Hoehold consumpon of oupu from radable secor g, s s an aggregae of h, h, rw purchases from home and abroad ( c, c ), h h h h, h h, rw (6) c c (1 ) c 1. The parameer governs he elascy of subsuon n fnal demand beween domesc and foregn producs. As wh nermedae npus, we allow goods and servces o have dfferen h elasces. The parameer governs he domescally sourced share of hoehold expendure on secor- oupu, and h s a consan scalng facor. The prce of secor- hoehold consumpon, p h, s gven by he sandard CES prce ndex. The governmen s consumpon of secor- oupu, g c, and he nvesmen producers e of secor- oupu, x x, are consruced ng analogo versons of (6), and her prces, g p and x p, are gven by he analogo consan elascy of subsuon prce ndces. The assocaed g x g x share parameers are (, ) and he scale parameers are (, ) Fnancal asses and exchange raes Before we can dscs nvesmen producon, hoeholds, and he governmen, we need o descrbe he bond marke. The U.S. governmen, hoeholds n he Uned Saes, and hoeholds n he res of he world have access o a one-perod, nernaonally raded bond, b, ha s denomnaed n uns of he U.S. consumer prce ndex (CPI), (7) p c p c. h h h h g g,1992 s s,1992 h h h h pg,1992 cg,1992 ps,1992 cs,1992 P U.S. hoeholds can also save by nvesng n he U.S. capal sock, bu wll be ndfferen beween holdng capal and bonds, as hey pay he same reurn n equlbrum. The real neres rae n uns of he U.S. CPI s (8) 1 r 1 P / Q, where Q s he bond prce. We compue he real exchange rae n our model ng consumer prce ndces, j as we do n he daa, rw (9) rer P / P, 9

12 where he res of he world s CPI, rw P, s defned analogoly o he U.S. CPI n (7). We model a sngle bond, bu he equlbrum of our baselne model s equvalen o one n whch boh governmens and hoeholds ssue deb. The deermnsc naure of he model gves rse o Rcardan equvalence (excep a he unexpeced onse of he savng glu), so he spl beween publc and prvae deb s essenally rrelevan. We have expermened wh a sochasc verson of he model n whch publc and prvae deb are dsnc, bu we fnd hs change o be quanavely nsgnfcan Invesmen Invesmen s produced by perfecly compeve frms operang a consan reurns o scale echnology; we foc on he represenave frm. Invesmen s an aggregae of goods and servces x x composes, ( x, x ), and he oupu of he U.S. consrucon secor, g s 1 x x x x x x x (10) x E g xg s xs c yc where, x x E s a consan scale facor, governs he share of goods, servces, and consrucon n nvesmen expendure, and governs he elascy of subsuon n nvesmen producon. The prce of nvesmen, x p, s gven by he sandard CES prce ndex Hoeholds Each counry s populaed by a connuum of dencal hoeholds. We draw a dsncon beween he oal ( n ) and workng-age ( ) populaons so ha our model capures he mpac of demographc changes, boh whn and across counres, on hoeholds ncenves o borrow or save. We normalze he me avalable for work and lesure n each perod o one. The represenave hoehold n he Uned Saes chooses consumpon of goods and servces, nvesmen, bond holdngs, and labor supply o maxmze lfeme uly, (11) 0 h h 1 h h h h 1 c h g cg h cs c s g s n n subjec o he budge consrans and he law of moon for capal, (12) p h c h p h c h p x x x Qb h h h 1 w P b (1 ) r k T, g g s s k k, 10

13 (13) k 1 (1 ) k x x, approprae nonnegavy consrans, nal condons for he capal sock and bond holdngs, and he no-ponz condon. Hoeholds pay consan proporonal axes, k, on capal ncome and receve a lump-sum ax or ransfer, T beween goods and servces n hoehold consumpon,. The parameer governs he elascy of subsuon governs he share of me devoed o h h lesure, (, ) govern he shares of goods and servces n hoehold consumpon, and g s deermnes he neremporal elascy of subsuon. Followng Buera and Kabosk (2009) and Herrendorf e al. (2013), we e Sone-Geary preferences o model nonhomoheces ha generae srucural change. The subssence requremen for goods, c h g, and he endowmen of servces, c h s, vary over me o ensure ha he model reans conssency wh balanced growh. 1 The res of he world s hoeholds solve a smlar bu smpler problem. They choose consumpon of goods and servces, labor supply, and bond holdngs o maxmze lfeme uly, (14) 0 rw rw rw rw 1 rw rw rw rw rw rw 1 c rw g c g rwcs c s ( ) g rw s rw rw n n subjec o he budge consrans rw rw rw rw rw rw rw rw (15) pgcg ps cs Qb 1 w P b, nonnegavy consrans, and a no-ponz condon. The parameer rw,, whch shfs he res of he world s neremporal demand, s he second exogeno drvng force n our model. We wll rw 2012 calbrae he seres { } so ha he U.S. rade balance n our model s savng-glu equlbrum 1992 rw maches he daa. Afer 2012, gradually revers o one, 1 Kongsamu e al. (2001) show ha, n a closed-economy model wh Sone-Geary preferences, one can ensure a balanced growh pah by choosng he subssence requremens so ha hey cancel ou n he hoehold s budge consran. In closed-economy models, balanced growh pahs are unque. Our model has a connuum of balanced growh pahs wh dfferen relave prces, so here are no subssence parameers ha genercally sasfy hs knfeedge condon. Boppar (2014) consrucs a closed-economy model ha generaes balanced growh whou hs parameer resrcon. We leave exendng ha model o an open economy wh unbalanced rade o fuure research. 11

14 rw rw (16) 1 (1 ) U.S. governmen The governmen n he Uned Saes leves axes and sells bonds o fnance exogenoly requred consumpon expendures. The budge consran s g g g g g g (17) pg cg ps cs Qb 1 k r k k T P b. Toal governmen consumpon expendures and deb as fracons of GDP follow exogeno me pahs, g c and g b. These pahs follow he hsorcal daa for , and n he subsequen years, he pahs follow he Congressonal Budge Offce (CBO) projecons (Congressonal Budge Offce, 2012). We allow he lump-sum ax, T 12, o vary as necessary o ensure ha he governmen s budge consran s sasfed. The governmen s objecve n each perod s o choose goods and servces consumpon o maxmze a CES uly funcon subjec o he consran ha oal consumpon expendures are a he requred level, g g g g max g g c c (18) c g, c g g s s s 1 g g g g g subjec o p c p c c GDP, g g s s g g where (, ) govern he shares of goods and servces n governmen consumpon and g s governs he elascy of goods and servces n governmen consumpon. Governmen deb s g g smply gven by b b GDP. The lump-sum ax mples ha our model exhbs near-rcardan equvalence, whch means ha he mng of axes and governmen borrowng s almos rrelevan. Rcardan equvalence breaks down only when we nroduce unexpeced evens such as he savng glu. In our sensvy analyss, however, we show ha devaons from Rcardan equvalence have lle mpac on our resuls. 3.3 Marke clearng Marke clearng for gross oupu of goods and servces n each counry l {, rw} requres ha gross oupu, l y, equals he sum of all nermedae and fnal demand (recall ha we do no model he governmen or he nvesmen secor n he res of he world), l, k rw, k h, k rw, k g, k x, k (19) y mj mj cj cj cj xj, g, s. k, rw jg, s

15 Consrucon marke clearng s rval snce consrucon s ed only o produce nvesmen. Marke clearng n facor markes requres (20),, k k, rw rw g,, s c g, s g,, s c and fnally, bond markes m clear, (21) b h b g b rw Equlbrum and balanced growh pahs An equlbrum n our model, gven a sequence of me-varyng parameers { rw, g, g l l l c b,( n, ) l{, rw},( ) l{, rw}, { g, s, c} } 0 and nal condons ( b h 1992, b g 1992, k 1992), s sequences of model varables such ha: hoeholds n he Uned Saes and he res of he world maxmze uly subjec o her consran consrans; prces and quanes sasfy margnal produc prcng condons for all commodes; prces and quanes are such ha all producon acves earn zero profs; all commody, facor, and bond marke clearng condons are sasfed; and he U.S. governmen solves s consumpon allocaon problem. {, c, b,( n, ) } are consan, and labor producvy n each secor rw g g When l l l{, rw} 0 l l grows a he same consan rae, { /, 1 g } 0 for l {, rw} and { g, s, c}, he model converges o a balanced growh pah. A balanced growh pah n our model s an equlbrum n whch all quanes grow a he consan rae of producvy growh, g 1 (excep for labor supply, whch s consan), and all relave prces are consan. Snce we allow for unbalanced rade, our model has a connuum of balanced growh pahs ndexed by U.S. ne foregn asse levels. The balanced growh pah o whch a parcular equlbrum converges s deermned by he equlbrum s nal condons and me-varyng parameers. Consequenally, solvng he model s no merely ransng beween known nal condons and a known balanced growh pah. We m solve for he ranson and he balanced growh pah smulaneoly. Ths complexy nfluences our calbraon approach, as we descrbe n he nex secon. To solve he model, we requre ha he model converges o a balanced growh pah by 2092, 100 years afer our sarng dae. Ths mples ha secoral producvy growh raes m evenually be equal, he oher exogeno me-varyng parameers m be consan, and hoehold preferences m be homohec. We do no wan hese end pon condons o dsor he behavor 13

16 of he model durng our perod of neres ( ). To acheve hs, we allow he me-varyng parameers o ake her calbraed values durng For he 50 years ha follow, secoral producvy growh raes converge o equaly, he oher exogeno me-varyng parameers converge o consans, and he subssence consumpon requremens converge o zero. Our resuls durng are no sensve o ncreasng he me akes o converge o he balanced growh pah beyond 100 years. We consruc wo equlbra n our model: a benchmark n whch he savng glu s followed by a gradual rebalancng of U.S. rade and a no-savng-glu counerfacual n whch he savng glu does no occur and U.S. rade s roughly balanced forever. The dfferences beween hese wo equlbra are compleely due o he pahs of { rw, g, g c b } 0 ; all oher parameers and he nal condons do no change Calbraon We calbrae he model so ha he counerfacual equlbrum (he one whou he savng glu) replcaes he U.S. daa n Our assumpon s ha he savng glu was no foreseen n We vew hs as he mos naural exercse, bu our resuls are no sensve o calbrang a benchmark equlbrum n whch agens foresee he savng glu. We ake he elasces of subsuon from he relevan leraure and calbrae he remander of he model s producon and preference parameers so ha he equlbrum replcaes he 1992 world npu-oupu marx (able 1) we have consruced from he World Inpu-oupu Daabase (WIOD). In he nex secon, we show ha, whle we have no calbraed he elasces, he model does a good job n machng observed changes n several key expendure shares evdence ha our chosen elasces are approprae. Table 2 summarzes he parameer values. More deals are provded n he onlne supplemenary maerals avalable a Gross oupu parameers We choose he scalng facors l, l A, and ( l, rw) so ha U.S. GDP n 1992 s 100 and l all gross oupu prces n 1992 are one. Followng Aalay (2014), we se he elascy of subsuon beween value added and nermedae npus, 1/(1 ), o 0.05, and he elascy of subsuon beween nermedae npus from dfferen secors, 1/(1 ), o For he whnsecor elasces of subsuon beween nermedaes from dfferen counres, 1/(1 ), we e hree for goods and one for servces. These choces are near he upper and lower bounds of he

17 aggregae Armngon elasces ed n he leraure. Gven hese values, we choose he share rw rw parameers, (, ) and (, ), so ha he equlbrum replcaes he quanes n our world npu-oupu marx. j j j j We calbrae he U.S. labor producvy growh parameers,, so ha real value added per worker n he model grows a he same rae as n he daa for each secor. Usng daa on value added and labor compensaon from he Bureau of Economc Analyss (BEA) for , we fnd ha he average annual growh raes of real value added per effecve worker are 4.4 percen n goods, 1.3 percen n servces, and 0.84 percen n consrucon. In he model s frs 50 years ( ), we se so ha real value added per worker n he model grows a he same raes as n he daa. Sarng n 2042, all of he secoral producvy growh raes begn o converge o 2 percen per year o ensure ha he equlbrum converges o a balanced growh pah. 2 Ths convergence process akes 25 years. Our resuls, whch foc on he perod , are no rw sensve o hs mng assumpon. For he res of he world, we calculae growh raes of so ha he prce of goods relave o servces follows he same pah n he res of he world as n he Uned Saes. 4.2 Fnal demand parameers Agan, we choose scalng facors l so ha 1992 fnal demand prces are one. The WIOD daa do no dsngush beween publc and prvae consumpon, so we do no have separae daa on changes n publc and prvae consumpon expendure shares. We herefore se boh 1/(1 ) and 1/(1 ) o he esmae n Aalay (2014) of 0.65 abou halfway beween he commonly ed value of 0.5 and he esmae of 0.85 by Herrendorf e al. (2013). We se he elascy of subsuon beween dfferen commodes n nvesmen, 1/(1 ), o one (Bems, 2008). The fnal demand Armngon elasces, 1/(1 ), are wo for goods and one for servces. As wh 2 In Nga and Pssardes (2007), when secors dffer permanenly n producvy growh, long-run balanced growh s possble f and only f he elascy of subsuon across secors n nermedae e s one. Snce we e a low elascy of subsuon beween nermedae goods and servces, we assume convergence across secors n producvy growh raes o ensure balanced growh. Ths assumpon has no sgnfcan mpac on he equlbrum dynamcs durng our perod of neres,

18 he gross oupu producon srucure, gven our choce of elasces, we choose values for he l share parameers,, so ha he equlbrum replcaes he world npu-oupu marx. 4.3 Hoehold parameers l We se he share parameers,, so ha hoeholds n boh counres e one-hrd of her me for work. We se he U.S. and world dscoun facors so ha he real neres rae on he balanced growh pah s 3 percen. Esmaes of he real neres rae from academcs and governmen research groups range from 0.5 o 5 percen (McGraan and Presco 2000, 2005; Kng and Low 2014; Congressonal Budge Offce, 2012, 2015). Gven hese consderaons, we chose he CBO s 2012 projecon of 3 percen becae s n lne wh esmaes from he leraure, and s conssen wh he CBO s budge projecons, whch we e n modelng governmen spendng and deb. Our man resuls are no sensve o reasonable neres rae values. For he demographc me seres l n and l, we e hsorcal daa and fuure projecons from he World Populaon Prospecs: 2010 Revson (Uned Naons, 2011). Conssen wh our measure of he real exchange rae n Secon 2, we consruc he res of he world s populaon as a weghed average of he Uned Saes 20 larges rade parners. To ensure convergence o a balanced growh pah, we assume ha populaons n he Uned Saes and he res of he world sar o converge o consan levels n 2042 and fnsh convergng 25 years laer. The subssence consumpon levels, h c, are consan durng Herrendorf e al. (2013) repor subssence consumpon levels for 1947 and We lnearly nerpolae beween hese daes o fnd he 1992 subssence parameers for he Uned Saes, and we assume ha, n per capa erms, he res of he world s subssence levels are he same as hose n he Uned c ( / ) c. To ensure a balanced growh pah n our model, preferences m rw rw h Saes:, ,1992 be homohec n he long run, so we assume he subssence parameers converge o zero durng We provde more deals n he onlne supplemenary maerals. 4.4 Governmen spendng and deb We calbrae he model under he assumpon ha agens dd no foresee he savng glu. I s lkely ha he observed pah of governmen spendng n he daa has been nfluenced by he savng glu, so we m specfy agens belefs over fuure governmen acons n he absence of he savng glu. In he no-savng-glu counerfacual, we assume ha governmen spendng remans a s

19 level (as a share of GDP) forever and ha governmen deb converges from 48 percen of GDP (he observed level n 1992) o 60 percen of GDP, conssen wh he 1992 CBO projecons. We h se he nal value of hoehold bond holdngs, b 1992, so ha oal ne foregn asses, h g b1992 b1992, are 7.8 percen of GDP (Lane and Mles-Ferre, 2007). When we solve he model wh he savng glu, he behavor of governmen spendng and deb wll dffer as dscsed below. 4.5 Capal sock, capal ax, and deprecaon Usng he mehod of Back e al. (2007), we compue U.S. capal socks for To calbrae he deprecaon rae, we e he capal sock daa, and he consumpon of fxed capal, repored by he BEA, o calculae an average annual deprecaon rae of 5.3 percen. We se he U.S. capal ncome ax, k, o 39 percen, he sauory rae repored by Devereux e al. (2002). We se he nal capal sock, k 1992, o 278 percen of GDP, as s n he daa. 5. Quanave resuls Havng calbraed he model, we now presen our quanave resuls. We sar wh a comparson of our benchmark savng-glu equlbrum o he daa, and hen examne he no-savng-glu counerfacual o answer he queson: wha would have happened o he U.S. economy and wha would happen n he fuure bu for he savng glu? 5.1 The benchmark savng-glu equlbrum We have calbraed he model o he no-savng-glu counerfacual he equlbrum n whch he savng glu dd no occur. We now e he calbraed model o sudy he mpac of he savng glu. In hs equlbrum, we keep all of he model parameers as hey are n he counerfacual, excep rw for hree: We change o creae he savng glu, and we change governmen spendng and deb rw pahs o mach he observed daa. We calbrae so ha he equlbrum replcaes he aggregae rw U.S. rade balance durng ; n he subsequen years, revers o one accordng o (16). U.S. governmen consumpon and deb (as a share of GDP) mach he daa for and follow projecons from he CBO s 2012 Long-Term Budge Oulook, whch we adj n laer years o be conssen wh balanced growh. More deals on our consrucon of projeced governmen spendng and deb are avalable n he onlne supplemenary maerals. Our resuls are 17

20 no sensve o assumpons abou he behavor of governmen spendng and deb, as well as hoeholds expecaons abou hese varables. 3 To compue he equlbrum n he model wh he savng glu, we e 1993 as he nal perod and e he 1993 equlbrum values of capal and bond holdngs from he no-savng-glu counerfacual as nal condons. Ths specfcaon embodes he assumpon ha agens n 1992 dd no foresee he savng glu and were surprsed by he ncrease n demand for savng n he res of he world, as well as by he ncrease n domesc governmen spendng and deb ha accompaned. Once agens learn of he savng glu, hey have perfec foresgh Goodness of f: Changes n expendure shares Before urnng o our hree key facs, we compare he model wh he daa along several key dmensons ha were no arges of he calbraon. These momens are eful n judgng he elasces of subsuon ha we have chosen from he leraure. We foc on changes n U.S. expendure shares beween 1995 and 2011, whch map drecly o he elasces of subsuon. We repor he changes n expendure shares n he daa and he model n able 3. The elascy of subsuon beween value added and nermedaes s an mporan deermnan n he change n he nermedae-npu share of gross oupu. We e he esmae from Aalay (2014), 1/( 1 ) 0. 05, whch makes hs share sensve o changes n he relave prce of nermedaes. As we repor n able 3, he change n he nermedae-npu share of gross oupu beween 1995 and 2011 s 1.04 percen n he daa and 1.50 percen n he model. The elascy of subsuon beween nermedae npus of goods and servces affecs he sensvy of he goods share of oal nermedae e o he relave prce of goods; we e he esmae from Aalay (2014), 1/( 1 ) In he daa, he goods share of oal nermedae e falls by 6.8 percen, compared wh 5.4 percen n he model. The elasces of subsuon beween goods and servces n publc and prvae consumpon are 1/(1 ) 0.65 and 1/( 1 ) As shown n Herrendorf e al. (2013), hese elasces m be conssen wh he underlyng model; elasces n models of gross oupu are sgnfcanly larger han hose n models of value added. We e he esmae n Aalay (2014), 3 For example: If we assume governmen spendng and deb are consan, he savng glu s responsble for 17 percen of he declne n goods-secor employmen and rses o 20 percen f we assume hoeholds perfecly foresee he acual pah of governmen spendng and deb durng he savng glu. 18

21 whch s derved from a gross oupu model wh an npu-oupu srucure and dsnc elasces of subsuon beween secoral oupu n fnal and nermedae es. To evaluae he appropraeness of our elascy, we examne he change n he goods share of aggregae consumpon. In he daa, hs share falls by 1.7 percen, compared wh 2.7 percen n he model. The elascy of subsuon beween goods and servces n nvesmen s 1/(1 ) 1, as found n Bems (2008). The change n he goods share of nvesmen n he daa s 0.2 percen, compared wh he unchanged share n our model. Lasly, he elasces beween domesc and foregn producs n nermedae and fnal es, 1/(1 ) and 1/(1 ), deermne he volaly of he secoral rade balances. The sandard devaon of he goods rade balance s 1.5 n he daa, compared wh 1.5 n he model. The lower elascy of subsuon beween servces produced a home and abroad generaes less volaly n he servces rade balance, whch maches he daa well. The sandard devaon of he servces rade balance s 0.24 n he daa, compared wh 0.20 n he model Goodness of f: Replcang he hree key facs In fgure 6, we plo he real exchange rae n he baselne model agans he daa and he no-savngglu counerfacual. Our model does a good job n machng he magnude of he apprecaon durng : The real exchange rae apprecaes by 27.9 percen n he daa and 27.1 percen n he model before begnnng o deprecae. The baselne model, however, fals o capure he mng of he deprecaon. In he daa, he real exchange rae begns o deprecae n 2002, four years before he rade defc begns o shrnk. In our model, he real exchange rae moves n andem wh he rade balance; does no begn o deprecae unl The lag beween he real exchange rae and rade balance movemens n he daa, ofen referred o as he J-curve (Back, e al., 1994), s dffcul o generae n models of aggregae rade such as ours. Alessandra and Cho (2015), for example, show ha heerogeneo frms ha prce o marke and make dynamc decsons abou enerng and exng he expor marke are crucal elemens n generang a realsc J-curve. The sor of frcons ha generae J-curves a he bness cycle frequency, however, are unlkely o generae he sor of four-year lag ha we see n he U.S. daa. To ncorporae hs lag no he model n Secon 6, we mechancally nroduce wedges so ha he model maches boh he rade balance and real exchange rae durng , and show ha hs has lle mpac on our model s predcons for secoral labor reallocaon. 19

22 In fgure 4, we plo he secor-level rade balances n he model alongsde he daa. The model closely maches he rade balances for boh goods and servces beween 1992 and Ths aspec of he model s performance s due n par o our choce of Armngon elasces. Table 3 shows ha n boh he model and he daa, he goods rade balance s more volale han he servces rade balance. Had we ed he same elasces n boh secors, he goods rade balance would no have moved enough, whle he servces rade balance would have been oo volale. In fgure 5, we plo he employmen shares for he goods and consrucon secors n he model alongsde he daa. The baselne model capures 99.2 percen of he declne n he goodssecor employmen share beween 1992 and Our model s ably o capure he full exen of he reallocaon of labor away from he goods secor makes well sued o answerng he counerfacual queson a he hear of he paper: Wha would have happened o he goods-secor employmen share n he absence of he savng glu? Fgure 5 also shows ha he model capures aspecs of he consrucon-secor employmen share beween 1992 and Durng , he consrucon-secor employmen share rses n boh he model and he daa, alhough our model generaes a larger ncrease. The subsequen b s smaller n he model han n he daa, prmarly becae we have no nroduced he fnancal crss of n any form oher han he way n whch affeced he rade balance. If we were o nroduce a more accurae model of he crss, we would undoubedly do beer n hs regard, bu hs s no he foc of our sudy. 5.2 The mpac of he savng glu We now e he calbraed model o sudy he mpac of he savng glu by comparng he baselne savng-glu equlbrum wh he no-savng-glu counerfacual. Ths comparson s nformave abou he mpac of he savng glu n boh he pas and he fuure. Our frs queson concerns he rade balance: Wha would he U.S. rade balance have been had he res of he world s effecve dscoun facor remaned consan? Fgure 7 shows ha, n he absence of he savng glu, U.S. rade would have been roughly balanced hroughou he perod; he enre cumulave U.S. rade defc durng hs perod s due o ncreased demand for savng n he res of he world. Ths s no an obvo or a rval resul nal condons and dfferences n demographc and ndral srucures across counres, coupled wh asymmerc producvy growh across secors, can generae cross-counry dfferences n savng behavor. Our resuls ell ha hese facors are no mporan n explanng he behavor of he U.S. rade 20

23 balance over he pas wo decades. Lookng no he fuure, our model predcs ha he rade balance wll be abou 1 percen of GDP hgher by 2024 han would have been f he savng glu had no occurred. Ths s drven by he repaymen of he deb ncurred by he Uned Saes durng he savng glu. Examnng he counerfacual secor-level rade balances n fgure 4, we see ha he savng glu s mpac on he rade balance has been concenraed n he goods secor; he servces rade balance would have been smlar had he savng glu no occurred. Ths s also rue n he fuure. The rade surples he Uned Saes wll run o pay back s savng-glu deb wll come almos enrely from he goods secor. The Uned Saes wll run a surpl n servces as well, bu by 2024 hs surpl would be he same as f he savng glu had no occurred. Our counerfacual real exchange rae mrrors he aggregae rade balance. As evden n fgure 6, n he counerfacual equlbrum, he real exchange rae s roughly unchanged beween 1992 and Comparng he counerfacual wh he baselne, we see ha all of he real exchange rae apprecaon beween 1992 and he md-2000s s arbuable o he savng glu, as s he subsequen deprecaon. In he fuure, as he Uned Saes repays s deb o he res of he world, s real exchange rae wll deprecae above s pah n he no-savng-glu counerfacual. Our prmary neres s he mpac of he savng glu on labor reallocaon n he Uned Saes: Wha poron of he declne n he goods-secor employmen share durng s arbuable o he savng glu? To answer hs queson, we compare he drop n he goods-secor employmen share n he counerfacual equlbrum wh he drop n he baselne equlbrum. We measure boh n he same uns: he fracon of he declne n he goods-secor employmen share n he daa. The declne n he no-savng-glu counerfacual s 84.1 percen of he observed declne, and he declne n he baselne equlbrum s 99.2 percen. The dfference beween he wo, 15.1 percen, s he decrease n he goods-secor employmen share ha s drven by he savng glu. Lookng o he fuure, fgure 5 shows ha he reversal of he savng glu wll have a neglgble mpac on goods-secor employmen. The share of employmen n he goods secor would be abou he same n 2024 f he savng glu dd no happen; he surpl n goods rade necessary o repay he deb wll have lle mpac on he allocaon of labor across secors. Elmnang he U.S. rade defc wll no brng back employmen o goods producon. Our foc s on he savng glu s mpac on srucural change, bu we can also measure he mpac of radonal srucural change forces n our model: dfferenal producvy growh and 21

24 nonhomohec preferences. When we se he subssence consumpon levels o zero, so ha preferences are homohec, he benchmark model accouns for 93.5 percen of he declne n he goods-secor employmen share, compared wh 99.2 percen n he baselne model; dfferenal producvy growh s clearly he domnan drver of srucural change n our model. Our assumpon abou preferences does no change our fndng regardng he savng glu. The savng glu s responsble for 14.4 percen of he declne n he goods-secor employmen share n he homohec model, compared wh 15.1 percen n he baselne model. 6. Sensvy analyss We have conduced a wde range of addonal expermens wh our model, and we fnd ha our man resul he modes mpac of he savng glu on U.S. srucural change s rob. We lm our dscson here o only hree ses of sensvy analyses. The frs modfes he model o mprove s ably o mach he mng of he U.S. real exchange rae. In he second, we calbrae our model o counerfacual npu-oupu ables o shed lgh on he role of nermedae npu lnkages n drvng srucural change. In he las analyss, we model rade defcs as drven by domesc facors raher han ncreased demand for savng n he res of he world. We summarze he resuls of our sensvy analyss n able 4. In he supplemenary maerals avalable onlne, we demonsrae he robness of our resuls o he hoehold s labor-lesure choce. 6.1 Trade wedges Fgure 6 shows ha our model maches he magnude of he real exchange rae apprecaon and deprecaon n he daa, bu msses on he mng. The real exchange rae begns o apprecae n 2002, bu he rade defc does no sar o shrnk unl To undersand he mporance of hs phenomenon for our resuls, we sudy a verson of our model wh rade wedges as n Alessandra e al. (2013). In hs model, shpmens of goods and servces from he Uned Saes o he res of rw rw he world are dsored by he me-varyng facor (1 ). Takng he rade wedges, m, as gven, producers n he res of he world maxmze profs, rw rw rw rw rw rw, rw rw, rw (22) q y w ( 1 m, ) qj mj qj mj jg, s Ths mples ha he prces of goods and servces n he res of he world are now funcons of he rade wedges, dsorng producers and consumers demand for producs from he Uned Saes. We calbrae hese wedges so ha he model maches exacly boh he rade balance and he real exchange rae daa durng m.

25 In fgure 8 we plo he goods-secor employmen share n he baselne model and he rade wedge model. The wo models behave almos dencally beween 1992 and 2002; he rade wedge model generaes less movemen n he goods-employmen share beween 2002 and 2016, and he wo models are almos dencal hereafer. The benchmark rade wedge model generaes 94.6 percen of he observed declne n he goods-secor employmen share beween 1992 and 2012, vers he baselne model s 99.2 percen. The no-savng-glu counerfacual s he same n boh models. Compared wh he baselne model, he rade wedge model arbues a smaller poron of he declne n he goods-secor employmen share durng o he savng glu, 10.6 percen compared o 15.1 percen. 6.2 The npu-oupu srucure Inpu-oupu lnkages quanavely affec U.S. srucural change n wo ways. Frs, servces are sronger complemens n nermedae e han n fnal e, so expendures on nermedae goods fall more han expendures on fnal goods n response o asymmerc producvy growh. Ths s evden n able 3: he goods share of U.S. nermedaes fell more han he goods share of U.S. consumpon beween 1995 and 2011, 6.8 percen vers 1.7 percen. Second, compared wh servces, goods are ed more nensvely as nermedaes. The rao of gross oupu o value added s 2.9 for goods and 1.7 for servces. We consruc wo counerfacual versons of our npu-oupu able o llrae he role of hese wo mechansms. In our frs counerfacual, we se all of he nermedae npu requremens o zero whle keepng secoral value added fxed. Ths exercse capures he full exen of he npu-oupu lnkages conrbuon o U.S. srucural change. In hs producon srucure, secoral oupu s now value added raher han gross oupu, so we m adj he elascy of subsuon beween goods and servces n consumpon (Herrendorf e al. 2013). We se he elascy of subsuon beween goods and servces n consumpon o 0.03, he same as our elascy of subsuon beween goods and servces n nermedae e. Ths s very close o he suggesed elascy of 0.02 from Herrendorf e al. (2013). In hs verson of he model, he benchmark equlbrum accouns for 66.2 percen of he declne n he goods-secor employmen share. Ths s subsanally less han he baselne model s 99.2 percen, ndcang ha here remans an mporan role for npu-oupu lnkages n drvng U.S. srucural change. Our second counerfacual npu-oupu able llraes how heerogeney n nermedae e across secors affecs srucural change. Here we rean he npu-oupu srucure, bu equalze 23