Global Imbalances and Structural Change in the United States

Transcription

1 Federal Reserve Bank of Minneapolis Research Deparmen Saff Repor 489 Aug 2013 Global Imbalances and Srucural Change in he Unied Saes Timohy J. Kehoe Universiy of Minnesoa, Federal Reserve Bank of Minneapolis, and Naional Bureau of Economic Research Kim J. Ruhl Sern School of Biness, New York Universiy Joseph B. Seinberg Universiy of Torono ABSTRACT Since he early 1990s, as he Unied Saes has borrowed from he res of he world, employmen in U.S. goods-producing secors has fallen. Using a dynamic general equilibrium model, we find ha rapid produciviy growh in goods producion, no U.S. borrowing, has been he mos imporan driver of he decline in goods-secor employmen. As he Unied Saes repays is deb, is rade balance will reverse, bu goods-secor employmen will coninue o fall. A sudden sop in foreign lending in would cae a sharp rade balance reversal and painful reallocaion across secors, bu would no affec long-erm srucural change. Keywords: Global imbalances; Real exchange rae; Srucural change JEL classificaion: E13, F34, O41 *We hank David Back, Kei-Mu Yi, and Frank Warnock for helpful discsions. Paricipans a conferences and seminars made eful commens and suggesions. Kehoe and Ruhl graefully acknowledge he suppor of he Naional Science Foundaion under gran SES We are graeful o Jack Rossbach for exraordinary research assisance. The daa ed here are available a hp:// The views expressed herein are hose of he auhors and no necessarily hose of he Federal Reserve Bank of Minneapolis or he Federal Reserve Sysem.

2 1. Inroducion Beween 1992 and 2012, hoeholds and he governmen in he Unied Saes borrowed heavily from he res of he world, and he U.S. ne inernaional invesmen posiion deerioraed by 4.0 rillion dollars. A commonly-held view in policy circles, expressed, for example, by Bivens (2006) and Sco, Jorgensen, and Hall (2013), is ha he U.S. rade deficis generaed by his heavy borrowing have played an imporan role in he decline of employmen in he U.S. goods secor and ha an end o hese deficis will reverse a large par of his rend. We e a dynamic general equilibrium model of he Unied Saes and he res of he world o address he quesions: To wha exen are rade deficis responsible for he loss of U.S. goods-secor employmen? Will employmen reurn o goods-producing secors when U.S. borrowing ends and rade deficis become rade surples? In addiion o modeling foreign lending, we incorporae a key feaure of he srucural change lieraure: differenial produciviy growh across secors. Using he calibraed model, we find ha U.S. rade deficis accouned for 16.3 percen of he decline in he goods-producing secor s employmen beween 1992 and 2012, wih mos of he remainder aribued o faser produciviy growh in he goods secor relaive o oher secors. This implies ha eliminaing he rade defici will no generae a significan increase in goods-secor employmen. We find ha, if he rade defici gradually changes o a surpl, U.S. borrowing in he 1990s and 2000s has been welfare improving. If he rade balance reverses quickly and unexpecedly a sudden sop, like ha in Mexico in hen welfare would be greaer if he Unied Saes had no borrowed. I is easy o see why some view rade deficis as derimenal o goods-secor employmen. Figure 1 shows ha he share of employmen in he goods secor agriculure, mining, and manufacuring has fallen dramaically as he rade defici has grown. Moreover, his idea has a simple inuiion: Impored goods are subsiues for domesically produced goods. As he Unied Saes rades bonds for foreign goods, labor shifs away from domesically produced goods and is reallocaed o producing services and consrucion, which are less subsiuable for foreign goods. When he deb has o be repaid, so he idea goes, labor will flow back ino he goods secor o produce he exra goods needed o repay he deb. In consrucing he model, we m ake a sand on he driving force behind U.S. borrowing. A common explanaion is ha foreign demand for saving increased, making foreigners more willing o rade heir goods for U.S. bonds. Bernanke (2005) coined he erm 1

3 global saving glu o refer o his idea, and we adop Bernanke s global-saving-glu hypohesis. Several explanaions have been proposed for he increased demand for saving in he res of he world, such as a lack of financial developmen in he res of he world (Caballero, Farhi, and Gourinchas, 2008; Mendoza, Quadrini, and Ríos-Rull, 2009), differences in biness cycle or srucural growh properies (Back, Henriksen, Lamber, and Telmer, 2006; Perri and Fogli, 2010), and demographic differences (Du and Wei, 2013). We do no ake a sand on which of hese explanaions, if any, are correc. Insead, we ake he saving glu as given and sudy is impac on he U.S. economy by calibraing a process for he preferences of hoeholds in he res of he world over curren vers fuure consumpion so ha our model maches exacly he pah of he U.S. rade balance beween 1992 and We include four feaures in he model ha make i well suied o address hese issues. Firs, we model an economy wih hree secors: goods, services, and consrucion. Goods and services can be raded, which allows o capure he fac ha he Unied Saes consisenly runs a subsanial rade surpl in services. Inernaional macroeconomic models ually rea goods as he only radable secor, and lump all oher secors ino a single nonradable secor. This assumpion is a odds wih he daa: Services are a large componen of U.S. expors. Consrucion is he only nonradable secor and is ed almos enirely o produce invesmen goods, which means ha consrucion is more sensiive han he oher secors o he effecs of capial flows and economic flucuaions in general. Second, we build a deailed inpu-oupu srucure ino he producion side of he model. This allows o model he elasiciy of subsiuion beween goods, services, and consrucion in producion as well as consumpion. Third, we allow he elasiciy of subsiuion beween foreign and domesic inpus o differ across secors. Our calibraion assigns a higher elasiciy of subsiuion beween domesic and foreign inpus in he goods secor o mach he fac ha he goods rade balance is significanly more volaile han he services rade balance, as seen in figure 6. Fourh, and perhaps mos imporan, we allow labor produciviy o grow a differen raes across secors, which allows o mach he fac ha labor produciviy in he goods secor grew a a faser pace han in oher secors over he pas wo decades. The srucural change lieraure emphasizes asymmeric produciviy growh as an imporan driver of long-run reallocaion of labor across secors. Recen sudies embed his mechanism, originally due o Baumol (1967), ino closed-economy models ha are consisen 2

4 wih aggregae balanced growh (Ngai and Pissarides, 2007; Buera and Kaboski, 2009). We ake a similar approach in an open-economy model. Several oher recen papers sudy srucural change in open economies. (Echevarria, 1995; Masuyama, 1992, 2009; Sposi, 2012; Ui, Yi, and Zhang, 2013). Wih he excepion of Sposi (2012), hese sudies e models of balanced rade, absracing from inernaional capial flows. We place capial flows a he forefron of our analysis, and we perform a quaniaive assessmen of he relaive conribuions of radiional srucural change forces (asymmeric labor produciviy growh) and he saving glu o he decline in goods-secor employmen in he Unied Saes. We calibrae our model so ha i maches exacly he naional accouns and inpu-oupu able for he Unied Saes in 1992, and he U.S. rade balance during As figures 5 7 show, our calibraed model endogenoly generaes oucomes ha mach several key facs abou he U.S. economy during his period: We mach he magniude of he real exchange rae appreciaion and subsequen depreciaion, he dynamics of he disaggregaed goods and services rade balances, and he changes in secoral employmen shares. Afer demonsraing our model's abiliy o mach hisorical daa, we e i o conduc wo kinds of exercises: hisorical counerfacuals and predicions abou he fuure. In our counerfacual scenario, we urn off he saving glu, which allows o answer quesions abou he pah he U.S. economy would have aken had he saving glu never occurred. In our model wihou a saving glu, he goods secor s employmen share falls from 19.7 percen in 1992 o 15.3 percen in 2011, compared o 14.2 percen in he model wih he saving glu and o 12.4 percen in he daa. Looking o he fuure, our benchmark model wih he saving glu predics ha he goods secor s employmen share will coninue o decline in he long run, and ha he rade balance reversal required o repay he deb he Unied Saes incurred during he saving glu will have lile impac on goods-secor employmen. The goods secor s employmen share is 13.1 percen in 2024 in our benchmark saving-glu scenario, vers 12.6 percen in he counerfacual scenario in which he saving glu never occurs. In our baseline model, we assume ha he ransiion from a rade defici o a surpl will happen gradually. In an exension, we consider he effecs of a sharp reversal of he rade balance. The possibiliy of a U.S. sudden sop arises frequenly in policy discsions. Bernanke (2005), for example, has argued ha, while a gradual rebalancing process is likely, he canno rule ou a sudden sop: 3

5 [T]he underlying sources of he U.S. curren accoun defici appear o be mediumerm or even long-erm in naure, suggesing ha he siuaion will evenually begin o improve, alhough a reurn o approximae balance may ake some ime. Fundamenally, I see no reason why he whole process should no proceed smoohly. However, he risk of a disorderly adjmen in financial markes always exiss. Sudden sops in he pas have ofen been accompanied by severe economic disrupion in he shor run: large real exchange rae depreciaions, subsanial produciviy-driven oupu conracions, and sharp reallocaions of facors across secors (Calvo, Izquierdo, and Mejía, 2004; Calvo and Talvi, 2005; Kehoe and Ruhl, 2009). We e our model o sudy wha a sudden sop in he Unied Saes in would look like and quanify he cos of he disrupion in economic aciviy. Our resuls indicae ha a sudden sop in he Unied Saes would look similar o hisorical episodes in emerging economies. The rade balance, as a share of GDP, would rise by 4.0 percen on impac, he real exchange rae would depreciae by 10.7 percen, and employmen would shif away from nonradable secors (consrucion) o radable secors (goods and services). As in sudden sops of he pas, hese effecs would be shor-lived; several years afer he sudden sop ends, he U.S. economy would be on almos he same pah on which i would have been if he sudden sop had never happened a all. Our resuls sugges ha he impac of a sudden sop would be emporary, bu his does no mean he welfare coss of a sudden sop would be small. We consruc a measure of he real income of U.S. hoeholds in 1992, and compare he welfare of U.S. hoeholds in hree scenarios: (1) he baseline, in which he saving glu ends in gradual rebalancing, (2) an alernaive in which he saving glu ends in a sudden sop, and (3) a counerfacual in which he saving glu never happened. The saving glu iself improves welfare: The lifeime real income of U.S. hoeholds in 1992 would have been almos 700 billion dollars lower (abou 11 percen of 1992 GDP) if he saving glu had no happened. A sudden sop in , hough, would reduce real income from he perspecive of model agens in 1992 by 1 rillion dollars compared o gradual rebalancing, eliminaing he welfare gains creaed by he saving glu. In oher words, if he saving glu ends in a sudden sop, U.S. hoeholds would prefer ha he saving glu had never happened. In he sensiiviy analysis of our model, we find ha our main resuls are rob o a range of modeling choices, such as alering our assumpions abou he pah of governmen spending and abandoning perfec foresigh in favor of a sochasic model wih raional expecaions. The 4

6 inpu-oupu producion srucure and he radabiliy of services, however, play imporan quaniaive roles in our model s predicions for goods-secor employmen. Removing he inpuoupu srucure raises he elasiciy of subsiuion beween goods and services in gross oupu, which caes our model o capure a significanly smaller fracion of he decline in goods-secor employmen we observe in he daa during This change also leads our model o oversae he role of he saving glu in reducing goods-secor employmen during his period; making services nonradable has a similar effec. To jify our assumpion ha he U.S. rade defici has been driven by he res of he world s demand for saving raher han domesic facors, we sudy a version of our model in which we aler he preferences of U.S. hoeholds raher han hose in he res of he world o mach he U.S. rade balance. This domesic-saving-drough experimen does no significanly affec our main resuls abou goods-secor employmen and he impac of a sudden sop, bu i leads o predicions for invesmen and consrucion employmen ha are sharply a odds wih he daa: The foreign-saving-glu hypohesis fis he daa beer. 2. The U.S. economy, This secion presens our approach o analyzing U.S. daa over he pas wo decades. We view he massive foreign borrowing and he differenials in produciviy growh across secors as exogeno driving forces ha we hardwire ino he model. We presen hree key facs abou U.S. daa ha we view as ess for our model. I is our model s abiliy o replicae hese hree facs in response o he exogeno driving forces ha gives confidence in ing he model o make predicions abou he fuure of he U.S. economy. Exogeno driving force 1: Foreign borrowing increased, hen decreased Figure 3 illraes j how much borrowing hoeholds and he governmen in he Unied Saes have done. The curren accoun balance measures he exac magniude of capial flows ino he Unied Saes, bu we see ha he rade balance racks he curren accoun balance almos exacly. Since our model is no designed o accuraely capure he difference beween hese wo series and he rade balance has an exac model analogue, we e he rade defici as our measure of U.S. borrowing. Figure 3 shows ha beween 1992 and 2006, he rade defici grew seadily, reaching 5.8 percen of GDP, afer which i began o shrink. In 2012, he rade 5

7 defici remained a 3.6 percen of GDP. We view he pah of he rade balance as wha defines he saving glu in our model. Our hypohesis is ha he U.S. economy is currenly in he process of emerging from he saving glu. Exogeno driving force 2: Labor produciviy grew fases in he goods secor During , labor produciviy in he goods secor increased by an average of 4.6 percen per year, while i increased by only 1.4 percen per year in he services secor and fell by 1.3 percen per year in consrucion. Wha is essenial in our model is he differenial beween produciviy growh in he goods secor and ha in he services secor. As he daa in figure 2 show, his differenial has been close o consan since 1980, wih average produciviy growh of 4.4 percen per year in goods during , compared wih 1.1 percen in services. Excep for he produciviy slowdown of 1970s, he differenial has been persisen since Our hypohesis is ha his produciviy differenial persiss unil a leas Fac 1: The real exchange rae appreciaed, hen depreciaed Figure 3 presens he firs key fac ha we ask our model o replicae. The figure shows ha he U.S. real exchange rae was volaile and racked he rade balance closely during We consruc our measure of he U.S. real exchange rae by aking a weighed average of bilaeral real exchange raes wih he Unied Saes 20 larges rading parners, wih weighs given by hese counries shares of U.S. impors in (Oher weighing schemes do no significanly affec our resuls.) This approach forms he basis of our concep of he res of he world in our model. The real exchange rae appreciaed by 27.9 ( 100 (100 / ) ) percen beween 1992 and 2002, afer which i depreciaed by 22.1 percen beween 2002 and The inuiion for why he real exchange rae and rade balance should move closely is sraighfoard: As foreign goods and services become cheaper, U.S. hoeholds buy more of hem. Noice, however, ha he iming is off: The maximum appreciaion of he real exchange rae occurred in 2002, while he larges rade defici occurred in Fac 2: The goods secor drove aggregae rade balance dynamics Figure 6 presens our second key fac. Here we disaggregae he rade balance, ploing he rade balances for goods and services separaely. We see ha he goods rade balance 6

8 generaes mos of he flucuaions in he aggregae rade balance, while he services rade balance flucuaes in a band beween 0.5 and 1.2 percen of GDP. Tha he Unied Saes has consisenly run a rade surpl in services moivaes one of he key feaures of our modeling framework. Sandard modeling convenions in inernaional macroeconomics lump services ogeher wih consrucion ino a nonradable secor, reaing goods as he only secor ha produces oupu ha can be raded inernaionally. By conras, we allow boh goods and services o be raded in our model, and we calibrae home bias parameers for each secor separaely o capure he fac ha he Unied Saes consisenly runs a surpl in services and a defici in goods. Fac 3: Employmen in goods declined seadily; consrucion employmen rose, hen fell Figure 7 presens our hird fac: Beween 1992 and 2011, he fracion of oal labor compensaion paid o he goods secor fell from 19.7 percen o 12.4 percen. The fracion of oal labor compensaion is our preferred measure of he goods secor s employmen share becae i maps direcly ino our model, where we measure labor inpus in erms of effecive hours worked, raher han raw hours worked. As figure 1 shows, his measure moves closely wih more common measures like he share of employees in he goods secor; he same is rue for he consrucion secor. The consrucion secor s share of labor compensaion rose from 4.4 percen in 1992 o 5.7 percen in 2006, as consrucion boomed prior o he financial crisis in Employmen in consrucion hen sared o fall, and by 2012 he consrucion secor s share of labor compensaion was 4.4 percen. 3. Model We model an economy wih wo counries, he Unied Saes and he res of he world (RW). We e he superscrips and o denoe prices, quaniies, and parameers in he Unied Saes and he res of he world. The lengh of a period is one year. Each counry has a represenaive hoehold ha works, consumes, and saves o maximize uiliy subjec o a sequence of budge consrains. We assume ha he only inernaionally raded asses are oneperiod bonds denominaed in unis of he U.S. consumer price index (CPI). In he baseline model, hoeholds have perfec foresigh over he fuure rajecory of he world economy here is no uncerainy. Each counry produces commodiies ha serve boh inermediae and final es. We model he U.S. producion srucure in deail, ing an inpu-oupu srucure 7

9 which we calibrae o a benchmark inpu-oupu marix published by he U.S. Bureau of Economic Analysis (BEA). We model producion in he res of he world in a simpler fashion, absracing from invesmen and domesic inpu-oupu linkages. We model he U.S. governmen in a reduced-form fashion as well. The governmen s spending and deb are specified exogenoly as fracions of U.S. GDP, and he governmen levies lump-sum axes on U.S. hoeholds o ensure ha is budge consrains are saisfied. Producion The Unied Saes produces four commodiies: goods y, services g y s, consrucion y, c and invesmen y i. The res of he world produces is own goods y and services g y s. Prices are denoed similarly. All commodiies are sold in perfecly compeiive markes. Each U.S. commodiy j gs, is produced ing capial k and labor j, along wih inermediae inpus j of U.S. goods z gj, U.S. services z, U.S. consrucion sj z cj, and impors m j purchased from he same secor j in he res of he world: z z z 1 j j j. gj sj cj j 1 j (1) y j M j j min,,, Aj kj j j (1 j ) mj a a a gj sj cj This nesed producion funcion embeds a Leonief inpu-oupu srucure in an Armingon aggregaor. The parameers of he producion funcions are: M j and A j, he consan scaling facors ed o faciliae calibraion; j, which governs he share of impors in producion; j, which governs he elasiciy of subsiuion beween domesic and impored inpus; a,, gj a sj acj, which govern he shares of goods, services, and consrucion in gross oupu; j, he secorspecific labor produciviy; and j, capial s share of value added. Our assumpion of zero subsiuabiliy beween inermediae inpus is sandard in he inpu-oupu lieraure, consisen wih empirical findings ha direc requiremen coefficiens vary lile over ime (Sevaldson, 1970; Miller and Blair, 2009). This assumpion implies ha he elasiciy of subsiuion beween goods and services in gross oupu is lower han he same elasiciy in consumpion. In secion 7, 8

10 we show ha his plays an imporan role in our model s abiliy o capure he bulk of he reallocaion of labor across secors in he daa. We allow he Armingon elasiciies 1/(1 ) o differ across secors o capure he fac ha he goods rade balance is considerably more volaile han he services rade balance (see figure 6). We also allow labor produciviy j o grow a differen raes across secors o capure he fac ha produciviy in he goods secor has grown faser han in oher secors. Consisen wih our inpu-oupu able (able 1), he consrucion secor is he only purely nonradable secor. The producion funcion in he consrucion secor is a special case of (1): z z z c min,,, A c k. agc asc acc gc sc cc c (2) 1 y c c c c U.S. producers in all hree secors j { g, s, c} choose inpus of inermediaes and facors o minimize coss, which implies sandard marginal produc pricing condiions for capial and labor. The U.S. invesmen good is produced ing inpus and consrucion according o a Cobb-Douglas echnology: (3) yi G zgi zsi zci 9 j z gi, z si g s c, g s c 1., and zci of goods, services, This specificaion is consisen wih evidence repored by Bems (2008) ha expendiure shares on invesmen inpus are approximaely consan over ime across a range of counries. We model he res of he world s producion srucure in less deail, absracing from invesmen and inpu-oupu linkages. Goods and services in he res of he world are produced ing labor j and impored inermediae inpus m j from he same secor j in he Unied Saes. The producion funcions are simpler nesed Armingon aggregaors of he form 1 j j j (4) y j M j j j j (1 j ) mj Hoeholds. Each counry is populaed by a coninuum of idenical hoeholds. We draw a disincion beween he oal and working-age populaions so ha our model can capure he impac of demographic changes. We denoe he oal U.S. populaion by n and he working-age

11 populaion by. We evaluae consumpion per capia on an adul-equivalen basis, defining he U.S. adul-equivalen populaion as n ( n )/2. The res of he world s demographic variables are defined analogoly. We normalize he amoun of ime available for work and leisure by a U.S. working-age person o 1 and denoe oal U.S. labor supply by consumpion of composie goods and services, h b o maximize uiliy h cg and. U.S. hoeholds choose labor supply, h c s, invesmen i, and bond holdings (5) (1 ) 1 / h (1 ) h c h g h c s 0 n n subjec o he budge consrains h h h h (6) p c p c p i qb 1 w p ( p, p ) b (1 ) r k T, g g s s i he law of moion for capial g s k k (7) k 1 (1 ) k i, appropriae non-negaiviy consrains, iniial condiions for he capial sock and bond holdings h k0 and b 0, and a consrain on bond holdings ha rules ou Ponzi schemes bu does no oheise bind in equilibrium. We e he superscrip h for U.S. hoeholds consumpion and bond holdings o disinguish hem from hose of he U.S. governmen, for which we e he superscrip g. Bonds are denominaed in unis of he U.S. CPI, defined as (8) p c p c p ( pg, ps ). P h h g g1992 s s1992 h h g1992cg1992 Ps1992cs 1992 We model discoun bonds, so he price q represens he price in period of one uni of he U.S. CPI baske in period 1. The real ineres rae in unis of he U.S. CPI is given by (9) 1 1 p ( p, p )/ q. r g s 10

12 Hoeholds pay consan proporional axes k on capial income and a lump-sum ax or ransfer T. We e he capial income ax o obain a sensible calibraion for he iniial capial sock and depreciaion rae. We also allow he ax rae on capial income in 1993 o differ from he consan rae in order o mach he level of invesmen in During he sudden sop episode ha can occur in 2015, bond holdings are fixed and he inernal real ineres rae is deermined endogenoly in each counry separaely. The res of he world s hoeholds solve a simpler problem. We absrac from invesmen dynamics in he res of he world, so he only way he res of he world s hoeholds can save is by buying U.S. bonds. We can easily allow for res-of-he-world bonds as well wihou changing he equilibrium of he model. Labor supply is again endogeno and he res of he world s hoeholds have a uiliy funcion similar o ha of U.S. hoeholds: (10) 0 (1 ) c g c s (1 ) 1 /. n n The only differences from he U.S. hoehold s uiliy funcion are he share parameer and he uiliy weigh. The laer is an ineremporal demand shifer, which we calibrae o mach he U.S. rade balance during During his period falls, reflecing a reducion in uiliy gained from consumpion a compared wih fuure consumpion. This is he driving force behind he saving glu. The res of he world s represenaive hoehold chooses labor supply, consumpion of goods and services maximize uiliy subjec o he budge consrains, c g and r (11) w p c p c qb 1 w p ( p, p ) b, g g s s g s c s, and bond holdings b o and non-negaiviy and no-ponzi consrains similar o hose ha U.S. hoeholds face. The res of he world s CPI is defined analogoly o he U.S. CPI in (8). We hen define he real exchange rae as (12) rer p ( p, p )/ p ( p, p ). g s g s 11

13 U.S. governmen The governmen in he Unied Saes levies axes and sells bonds o finance exogenoly required expendiures on consumpion of goods and services. The budge consrain is g g g g (13) p c p c qb 1 r K T p ( p, p ) b. g g s s k k g s We specify exogeno ime pahs for governmen consumpion expendiures and deb as fracions of GDP, ing hisorical daa for and Congressional Budge Office (CBO) projecions for he fuure. We e and o denoe hese ime-series. We allow he lumpsum ax T o vary as necessary o ensure ha he governmen s budge consrain is saisfied. We specify he elasiciy of subsiuion beween goods and services in he governmen s spending rule o be 1, so ha he governmen chooses g c g and g c s o maximize a Cobb-Douglas g aggregaor wih goods-share parameer subjec o he consrain ha he governmen makes is required oal expendiures. We assume ha governmen spending does no ener he hoehold s uiliy funcion (or equivalenly, eners in a separable fashion), nor does i ener any of he producion funcions. Becae of he lump sum ax T, our model exhibis near-ricardian equivalence. Tha is, he iming of axes and governmen borrowing is almos irrelevan. Ricardian equivalence breaks down only when we inroduce unexpeced evens he saving glu and he sudden sop. Unanicipaed changes in he ime pah of governmen deb ha accompanies hese evens do affec he model s equilibrium dynamics, paricularly in he shor run. Equilibrium {,, } An equilibrium in our model for a given sequence of ime-series parameers and iniial condiions h 0 0, g 0, 0 ( b b k ) consiss of a sequence of all model variables such ha hoeholds in he Unied Saes and he res of he world maximize heir uiliies subjec o heir consrain ses, prices and quaniies saisfy marginal produc pricing condiions for all commodiies, prices and quaniies are such ha all producion aciviies earn zero profis, all commodiy, facor and bond marke clearing condiions are saisfied, and he U.S. governmen solves is consumpion-spending allocaion problem in each period. When we solve he model numerically, we require ha equilibria converge o balanced growh pahs afer

14 years. There are an infinie number of possible balanced growh pahs one for every combinaion of public and privae bond holdings. 4. Calibraion As in mulisecor, saic applied general equilibrium models like Kehoe and Kehoe (1994), we calibrae many of he model s parameers so ha he equilibrium in 1992 of he model in which he saving glu does no occur replicaes he inpu-oupu marix. There are several discrepancies beween he naional income and produc accouns and he inpu-oupu marix, so we le he NIPA ables ake precedence and perform several adjmens ing he RAS algorihm (Sone, 1961) o he inpu-oupu marix so ha i maches he relevan NIPA daa exacly. Deails are provided in he online daa appendix a hp:// Our adjed inpu-oupu marix is shown in able 1. Table 2 liss our calibraed parameer values. U.S. producion parameers We choose unis so ha U.S. GDP is equal o 100 and all prices are equal o 1 in 1992 all quaniies are expressed as percenages of 1992 GDP. We compue he parameers in he Leonief porion of he U.S. producion funcions in (1) direcly from he inpu-oupu marix. For example, o compue a gc, he amoun of goods needed o produce one uni of gross oupu in he consrucion secor, we divide he value in he goods row and consrucion column (3.7) by gross oupu in consrucion (10.7). We e a similar procedure o calculae facor shares in value added for each secor. For he Armingon aggregaors in (1), we firs specify values for he elasiciies of subsiuion beween domesic and impored inpus. There is some debae over his elasiciy becae biness cycle models end o imply low elasiciies while analysis of rade policy changes ofen suggess much higher elasiciies; see Ruhl (2008) for a deailed discsion. Becae he services rade balance is less volaile han he goods rade balance, we choose a lower elasiciy of subsiuion beween domesic and foreign inpus in he services secor. We se he elasiciy in goods 1/ (1 g ) o 3 and ha in services 1/ (1 s ) o 1. We hen e equilibrium condiions (marginal produc pricing and zero profis) o calibrae j from he 13

15 inpu-oupu marix. The scale facors M j follow immediaely. We e equilibrium condiions in a similar procedure o calibrae he invesmen secor s parameers. Hoehold and governmen parameers We se he elasiciy of ineremporal subsiuion 1/(1 ) o 0.5. We se he long-run ineres rae o 3 percen (U.S. Congress, Congressional Budge Office, 2012). We se he discoun facor so ha his ineres rae is consisen wih balanced growh. We se he elasiciy of subsiuion beween goods and services in consumpion, 1/(1 ), o 0.5, similar o h he value in Sockman and Tesar (1995). We calibrae he parameers and of he hoehold s preferences ing privae consumpion daa from our inpu-oupu able and daa in hours worked in he hoehold s firs-order condiions. A similar procedure yields he g governmen s share parameer. U.S. iniial condiions To calculae he iniial capial sock, we se he 1992 real ineres rae o 4 percen. The real ineres rae in 1992 is no an equilibrium objec in our model; i would have been deermined in 1991, bu 1992 is our iniial model year. The real ineres rae on 10-year U.S. Treasury bonds was approximaely 4 percen in Our resuls are no sensiive o alernaive approaches o calibraing he iniial capial sock. Given a ax rae on capial income k, we can compue our iniial capial sock ing daa on depreciaion from he 1992 naional income accouns and daa on paymens o capial in our inpu-oupu able. We choose 0.4, which implies a depreciaion rae of 6.6 percen per year, well wihin he sandard range of annual depreciaion raes ed in he lieraure. U.S. governmen deb was 48.1 percen of GDP in 1992, g and we e his value direcly o se he governmen s bonds in 1992, b We hen se privae h h g bond holdings, b 1992, so ha oal ne foreign asses b1992 b1992 are 7.8 percen of GDP as repored in Lane and Milesi-Ferei (2007). k Calibraing he res of he world To calibrae he remaining parameers, we need o specify wha he res of he world is in he daa. We selec he Unied Saes op 20 rading parners, ranked by average annual bilaeral 14

16 rade (expors pl impors) beween 1992 and 2012, and weigh hem by heir average share of U.S. oal annual rade during his period. We e hese counries weighs o consruc a composie rading parner, hinking of he res of he world as being composed of 20 idenical counries ha all look like his composie. To calculae he size of he res of he world relaive o he Unied Saes, we ake a weighed average of goods and services consumpion of hese 20 counries and muliply hese figures by 20 o ge oal consumpion of goods and services in he res of he world. We e equilibrium condiions in a similar manner as before o calibrae he res of he world s Armingon aggregaors and preference parameers. Exogeno processes We e hisorical daa and fuure projecions from he World Populaion Prospecs: 2010 Revision (Unied Naions, 2011) o consruc ime series for he demographic parameers for boh he Unied Saes and he res of he world (ing he same 20 counries and weighs as before). The Unied Saes and he res of he world are projeced o grow a differen raes well pas he 100-year erminal dae in our model, so o ensure balanced growh in our compuaion, we assume ha he populaions in boh counries begin o converge o consan levels afer Our model s equilibrium dynamics beween 1992 and 2024, he period on which we foc, are no sensiive o his assumpion. We calculae secor-level produciviy growh raes ing daa on value added and labor compensaion by secor from he BEA for We e hese daa o perform growh accouning by secor and find ha he average growh raes of labor produciviy over his period are 4.6 percen in goods, 1.4 percen in services, and 1.3 percen in consrucion. We e hese values in he model beween 1992 and 2030, and in he years following we assume ha all of he secor-level growh raes converge slowly o 2 percen per year, o ensure ha he equilibrium converges o a balanced growh pah. We consruc several ime series for governmen consumpion expendiure and deb. We e hisorical daa from he BEA for governmen consumpion expendiures and from he CBO for governmen deb. We e CBO projecions for governmen expendiures and deb as fracions of GDP as a saring poin, bu we make adjmens o reain consisency wih he naional income accouns and o allow for balanced growh in he long run. We e he mos recen projecions when possible. More deails are available in he daa appendix. 15

17 In he no-saving-glu counerfacual, we assume ha governmen spending as a fracion of GDP remains consan a is 1992 value and ha governmen deb gradually rises o 60 percen of GDP over ime. In modeling he rebalancing scenario, we e hisorical daa for and he CBO projecions hereafer. Some of he projecions he CBO has made for he pahs of governmen deb are implaible, however; he exended baseline scenario predics ha governmen deb will drop below zero by 2070 and he exended alernaive scenario predics ha deb will surpass 250 percen of GDP by We herefore assume ha governmen deb as a fracion of GDP will remain consan a he 2023 value of 77.0 percen of GDP in 2024 and beyond. Finally, in he sudden sop scenario, we e he same expendiure series bu assume ha he deb-gdp raio falls o 60 percen once he saving glu ends, reflecing a permanen change in U.S. governmen policy ha coincides wih he sudden sop. 5. Quaniaive resuls This secion presens he baseline quaniaive resuls and conrass he resuls beween he model in which he saving glu occurs and one in which i does no occur, o show how he model is able o replicae our hree key facs. We hen sudy he model s predicions for he fuure and how a sudden sop in foreign lending in would play ou. We finish wih an analysis of he welfare consequences of he saving glu and a sudden sop. Quaniaive sraegy Our firs sep is o calibrae he model s parameers and iniial condiions so ha he equilibrium in which neiher he sudden sop nor he saving glu occurs replicaes he benchmark inpu-oupu marix and naional accoun figures published by he BEA for We calibrae he res of he world s preference parameer 1992 o mach he U.S. rade balance in 1992 and assume ha i converges quickly o a consan value of 1 hereafer. In his sep, agens in our model do no expec he saving glu o occur. We do no calibrae our model o mach any ime series a all in his sep. We rea he model s equilibrium dynamics in his scenario as a counerfacual exercise, allowing o ask he quesion: Wha would have happened over he pas wo decades, and in he fuure, had he saving glu no happened? We could skip his sep and go righ o he nex one wihou changing our main resuls significanly, bu we view his as he mos naural way o hink abou he agens expecaions in he early 1990s: We do no hink 16

18 ha U.S. hoeholds in 1992 foresaw he kind of borrowing he Unied Saes would do over he subsequen wo decades. Our second sep is o solve for he model s dynamics in he scenario in which he saving glu acually happens. Here, we hold fixed all of he parameers calibraed during he firs sep and calibrae he values of for so ha he equilibrium replicaes exacly he aggregae U.S. rade balance during his period. As in he firs sep of our exercise, we assume ha gradually converges o 1 afer We e he equilibrium values of capial and bond holdings in 1993 from he no-saving-glu firs sep as iniial condiions in his second sep. The saving glu, which manifess in our model as emporarily reduced uiliy from consumpion and leisure in he res of he world, is an unanicipaed even: Model agens in 1992 do no expec i o occur, bu hey have perfec foresigh hereafer. We refer o he model s pos-2012 dynamics in his scenario as a gradual rebalancing, represening he oucome of a slow, orderly end o he forces driving he saving glu. In our sensiiviy analysis, we relax he perfec foresigh assumpion and allow for agens o be uncerain abou he lengh of he saving glu. Our hird, and final, sep is o solve for he model s dynamics from 2015 onward when a sudden sop occurs. We rea a sudden sop as anoher unforeseen even and rea he equilibrium sae variables from he second sep (gradual rebalancing) in 2015 as he iniial condiions. As before, once he sudden sop occurs, model agens have perfec foresigh hereafer. We model a sudden sop as a wo-year period in which he Unied Saes is resriced from borrowing furher from he res of he world, afer which foreigners are again willing o purchase U.S. bonds. We assume ha he res of he world s preference parameer converges o is long-run value of 1 more quickly han in he previo scenario, which we inend o capure he idea ha a sudden sop is associaed wih a faser end o he forces ha drove he saving glu in he firs place. Furhermore, we assume ha a sudden sop generaes a disorderly adjmen in financial markes ha caes oal facor produciviy (TFP) in all hree producion secors o fall by 10 percen in 2015, o recover half of his drop in 2016, and o fully recover in This TFP drop capures he sor of disrupion and rapid recovery ha occurred in sudden sop episodes like ha in Mexico in (Kehoe and Ruhl, 2009). We also assume ha he U.S. governmen s deb as a fracion of GDP falls o a lower long-run level han in he gradual rebalancing case, 17

19 represening he idea ha a sudden sop is associaed wih, or perhaps riggers, a long-erm change in U.S. governmen deb policy. We model he sudden sop in in he same manner as Kehoe and Ruhl (2009), who model he Mexican sudden sop of as a surprise. Agens in he model have perfec foresigh before and afer he sudden sop, bu, if and when he sudden sop occurs, i is compleely unexpeced. We model he sudden sop as a surprise becae U.S. ineres raes currenly indicae ha financial markes do no assign a significanly posiive probabiliy o a U.S. deb crisis j as hey did no assign significanly posiive probabiliies o a crisis in Mexico in 1995 (which sared during he las wo weeks of 1994) or o he currenly ongoing deb crises in he Eurozone. (See, for example, Arellano, Conesa, and Kehoe, 2012.) We hink of he possibiliy of a deb crisis sriking he Unied Saes as he possibiliy of he sor of selffulfilling crisis modeled by Cole and Kehoe (2000) and Conesa and Kehoe (2012). Replicaing he hree key facs The rade balance pos 2012 and all of he prices and quaniies pos 1992 are endogeno oucomes of he model. Here we show ha he equilibrium of he calibraed model maches he U.S. daa in he hree key dimensions laid ou in secion 1. As figure 4 shows, we calibrae our model o mach he U.S. rade balance exacly beween 1992 and In his figure we have also ploed he rade balance from he model in which he saving glu never occurs. Noice ha, in his counerfacual scenario, rade would be close o balanced hroughou he period. In he baseline model, however, he saving glu has occurred, so he Unied Saes m repay is deb in he long run. The figure shows ha, as long as he rebalancing process goes smoohly, he U.S. rade balance will swich from a defici o a surpl in 2018, and will reach a surpl of more han 1.1 percen of GDP by Figure 5 plos he model s real exchange rae in he baseline model agains he daa and he no-saving-glu counerfacual. Our model does a good job of maching he magniude of he appreciaion during : The real exchange rae appreciaed by 27.9 percen in he daa and 26.0 percen in he model before beginning o depreciae. The model, however, fails o capure he iming of he depreciaion. In he daa, he real exchange rae begins o depreciae in 2002, four years before he rade defici begins o shrink. In our model, he real exchange rae moves in andem wih he rade balance, so i does no begin o depreciae unil In our 18

20 model, consumers and firms begin o impor fewer foreign goods and services only once hey begin o become more expensive. Figure 5 also shows ha if he saving glu had never occurred, he U.S. real exchange rae would appreciae slowly over he long run, due in large par o he increase in he relaive price of services, in which he Unied Saes has a comparaive advanage. The saving glu did occur, however, and becae he Unied Saes m run a rade surpl in he long run, is goods and services m become cheaper. Our model predics ha he real exchange rae will coninue o depreciae for several years, evenually converging o a level ha is 6.3 percen depreciaed compared o ha in he no-saving-glu counerfacual. Figure 6 plos he secor-level rade balances in he model alongside he daa. The model maches boh he goods and services rade balances closely beween 1992 and This aspec of he model s performance is due o our choice of Armingon elasiciies. Had we ed he same elasiciy in boh secors, he goods rade balance would no have moved enough, while he services rade balance would have been oo volaile. The figure also shows ha in he absence of he saving glu, he services rade balance would no have been subsanially differen, while he goods rade balance, refleced in he aggregae rade balance, would have been almos fla. In he long run, he model predics ha he goods rade balance will be 1.4 percen of GDP higher by 2024 han i would have been if he saving glu had never occurred. Despie his, he model predics ha he goods rade balance will be negaive in he balanced growh pah: The services secor will be he source of he enire long-run rade surpl. Figure 7 plos he employmen shares for goods and services in he model and he daa. In he baseline gradual rebalancing scenario, our model maches he daa closely beween 1992 and 2001, afer which he goods employmen share falls more in he daa han in he model. For , our baseline model capures 75.2 percen of he decline in he goods secor s employmen share ha we observe in he daa. In he no-saving-glu counerfacual, however, he model sill capures 60.5 percen of he decline (he decline in he counerfacual is only 19.5 percen smaller han he decline in he baseline rebalancing scenario). The figure suggess ha, while he saving glu did emporarily accelerae he decline in goods-secor employmen, he bulk of he decline is aribuable o he fac ha labor produciviy has grown faser in he goods secor han in he res of he economy. 19

21 The implicaion of his resul is ha, conrary o popular belief, he end of he saving glu will have very lile impac on employmen in he goods secor. By 2024, he model indicaes ha he goods secor s share of employmen will be almos exacly he same regardless of wheher or no he saving glu had happened. Despie he fac ha he U.S. economy m run a rade surpl in he long run o repay is deb, he goods secor s employmen share will coninue o decline once he saving glu ends. There are wo main reasons for his resul. Firs, labor produciviy coninues o grow faser in goods han in oher secors in our model even afer he saving glu ends. This aspec of srucural change has been a consisen force in he U.S. economy since he 1960s, and we see no reason ha i should end when he saving glu does. Second, he Unied Saes can repay is deb wih he rade surpl generaed by he services secor; our model predics ha he Unied Saes will coninue o run a defici in goods rade once he saving glu ends and he aggregae U.S. rade balance reverses. Figure 7 also shows ha he model capures several aspecs of he consrucion secor s employmen share beween 1992 and Beween 1992 and 2006, he consrucion secor s employmen share rises in boh he model and he daa, alhough our model generaes a larger increase. Our model overexplains he boom in consrucion secor employmen j by inroducing he saving glu. The subsequen b is smaller in he model han in he daa, bu his is primarily becae we have no inroduced he financial crisis of in any form oher han he way in which i affeced he rade balance. If we were o inroduce addiional feaures o he model o more accuraely model he crisis, we would undoubedly do beer in his regard, bu his is no he foc of our paper. As he figure shows, our model suggess ha he effecs of he saving glu on he consrucion secor will largely dissipae by 2016; j like in he goods secor, he consrucion secor s long-run employmen share dynamics will be driven primarily by produciviy growh raher han he rebalancing process. A sudden sop in The resuls we have presened so far foc on he scenario in which he rebalancing process is gradual and orderly. Figures 4 7 also illrae wha our model predics if his process insead involves a sudden sop in Figure 4 illraes he dynamics of he aggregae rade balance in he sudden-sop scenario compared o he rebalancing scenario; figure 5 shows analogo resuls for he real exchange rae We see ha a sudden sop would rigger an 20

22 immediae reversal in capial flows and a large real exchange rae depreciaion. The rade balance would rise from 1.4 percen of GDP o 2.6 percen on impac, and he real exchange rae would depreciae by 10.7 percen. These effecs, however, would be shor-lived. By 2024 he rade balance and real exchange rae would be almos idenical o heir levels in he gradual rebalancing scenario. While a sudden sop involves quicker deb repaymen, he effec on he long-erm need o repay is small in our model. The Unied Saes has borrowed so much from he res of he world in he las wo decades ha wo years of rapid repaymen do no make much of a den in he overall sock of deb. Figure 6 shows ha he large, emporary rade balance reversal comes mosly from he goods secor. The goods rade balance rises by 3.4 percen of GDP on impac, while he services rade balance rises by 0.6 percen. Figure 7 shows ha, despie he large increase in he goods rade balance, we do no see a large reallocaion of labor oward he goods secor. The goods secor s labor compensaion share rises from 14.4 percen o 15.1 percen on impac, bu his is shor-lived, j like he oher effecs of he sudden sop. Once he sudden sop ends, rapid produciviy growh coninues o ph goods employmen downward, and by 2024 he goods secor s labor compensaion share is almos exacly as i would have been if he sudden sop had no occurred. As compared wih he goods secor, consrucion sees a much larger (alhough j as emporary) impac from he sudden sop. The consrucion secor s share of labor compensaion falls by more han one-hird, from 6.9 percen o 4.7 percen. There are wo channels a work. Firs, consrucion is he only nonradable secor, so i is more sensiive o changes in capial flows. Second, he bulk of he consrucion secor s oupu is ed o produce invesmen. The sudden sop caes he U.S. real ineres rae o rise by more han 260 basis poins, leading o a large decline in invesmen, which subsanially reduces demand for consrucion. In shor, a sudden sop in he Unied Saes would look, in many respecs, similar o pas sudden sop episodes in emerging economies. There would be a large rade balance reversal and real exchange rae depreciaion, and severe disrupion in he consrucion secor, bu hese effecs would be shor-lived once he sudden sop ends, he U.S. economy would reurn o roughly he same rajecory on which i would have been had he sudden sop never happened. While a sudden sop would emporarily sem he decline in goods-secor employmen, his rend would coninue unabaed once he sudden sop ends. 21

23 Welfare implicaions of he saving glu We have shown ha he saving glu has had a large impac on some aspecs of he U.S. economy and ha a sudden sop will cae subsanial shor-erm disrupion. Here we ask wo relaed quesions: Did he saving glu make U.S. hoeholds beer or worse off? How does he answer depend on wheher he saving glu ends in a gradual rebalancing or a sudden sop? To answer hese quesions, we consruc a measure of real income in 1992 denominaed in 1992 U.S. dollars. Our real income index is based on an alernaive bu equivalen specificaion of he represenaive hoehold s preferences ha is homogeneo of degree one, (14) 1992 h h c g c s (1 ) n n (1 ) 1. The cos of achieving his uiliy in unis of he U.S. CPI in 1992 is. 1 h h (15) 1992 q ( ) s 1992 s pgcg pscs w The prices and quaniies above represen equilibrium objecs in our benchmark gradual rebalancing scenario, he one in which he saving glu occurs, bu a sudden sop does no. To conver his objec o 1992 dollars, we scale i so ha consumpion expendiures in he model are equal o 1992 privae consumpion in he NIPA ables. We e he resuling scaling facor o calculae real income in alernaive scenarios, like he counerfacual in which he saving glu does no occur or he scenario in which boh he saving glu and a sudden sop occur. In our baseline model we assume ha, in 1992, model agens expec governmen consumpion expendiures o remain fixed a he 1992 level of 16.6 percen of GDP, bu when he saving glu begins, an unforeseen change in governmen spending policy occurs: Governmen spending as a fracion of GDP racks he daa beween 1993 and 2012, hen rises o 22.9 percen over ime. This reflecs policy changes ha have occurred over he pas wo decades, such as increased healh care and defense spending, ha people likely did no anicipae in he early 1990s. This increase in governmen consumpion gives U.S. hoeholds an incenive o save for he fuure. We repor our welfare resuls under an alernaive assumpion: In 1992, agens expec governmen consumpion, as a fracion of GDP, o follow he pah i acually ook 22