Intergenerational Effects of Trade Liberalization

Transcription

1 Itergeeratioal Effects of Trade Liberalizatio Erha Artuç 1 March, 2009 Abstract 2002 Pew Global Attitudes survey shows that workers support for free trade decreases with age, which is a pheomeo previously uexplored by ecoomists. We explore distributioal effects of trade liberalizatio i particular age ad gais from free trade, usig a dyamic structural geeral equilibrium model. The model we develop here is i the same spirit as Artuc, Chaudhuri ad McLare (2007), but allows for a much richer treatmet of both ex-ate ad edogeous worker heterogeeity. This feature requires a completely differet estimatio strategy, which comes at a cost of more computatio time ad stroger assumptios o workers expectatios. After estimatig the structural model with U.S. data sets NLSY ad CPS, we simulate a hypothetical trade liberalizatio i metal maufacturig sector (which has bee especially vulerable to trade shocks i the past, the steel idustry i particular). We show gradual adjustmet of labor allocatio, wages ad prices i respose to this trade shock. We fid a mirror effect where very youg workers are either moderately better off or moderately worse off, while older workers are either extremely better off or extremely worse off depedig o their sectors. JEL Classificatio: F1, D58, J2, J6 Keywords: Trade Liberalizatio, Sectoral Mobility, Labor Market Equilibrium. 1 Koç Uiversity, Departmet of Ecoomics. Cotact ifo available at: eartuc. This project is sposored by a Scietific ad Techological Research Coucil of Turkey grat (TUBITAK). I would like to thak Sebem Kalemli-Ozca, Isa Tuali, Da Trefler, Shubham Chaudhuri, Steve Ster ad Chris Otrok for their commets. I am grateful to Joh McLare ad Leora Friedberg for their guidace. The views expressed here are mie ad do ot ecessarily reflect those of the affiliated istitutios ad sposors. All errors are mie. 1

2 1 Itroductio Recetly, trade ecoomists have developed dyamic structural models to aalyze effects of trade liberalizatio. Oly a few of those structural models were desiged to aalyze welfare effects of free trade o workers, while may of them, especially those which geerated a large body of reasearch, focused o firms rather tha workers. 2 Most of the research o distributioal effects of trade liberalizatio o workers have bee coducted via reduced form regressios. 3 Although reduced form regressios helped ecoomists to aswer may questios about effects of trade policy shocks, there are certai very iterestig questios that ca ot be aswered by them, such as: How log will it take to reach the ew steady state after a trade shock? What are the welfare effects o export-sector workers? How do workers adjust to free trade i case of delayed trade liberalizatio - if it is ot implemeted yet? What are the o-pecuiary welfare costs of trade liberalizatio? Artuc, Chaudhuri ad McLare (2007) developed a method to aswer such questios (heceforth ACM), withi a dyamic structural geeral-equilibrium framework. 4 The theoretical foudatio of ACM was itroduced by Camero, Chaudhuri ad McLare (2008). Although ACM s method was well suited to aalyze welfare effects of trade liberalizatio, it was ot desiged to study distributioal effects i detail. Studyig distributioal effects seriously requires extremely large data sets, whe detailed worker heterogeeity is itroduced to their model. This is because their estimatio strategy relies o aggregate mobility matrices for each observed worker type. which could easily be cotamiated with empty cells if the state-space is fiely partitioed. I this paper, we develop a differet method, complemetary to ACM, which ca be used to study distributioal effects of trade liberalizatio at a cost of loger computatio time ad stroger assumptios o workers expectatios. The model we itroduce i this paper is also based o Camero, Chaudhuri ad McLare (2008) similar to ACM, but the mai differece is we do ot use their (computatioally cheap ad compact) Euler-equatio coditio approach. Istead, we calculate expected values of workers similar to structural discrete choice models i the labor ecoomics literature, which allows us to itroduce a richer treatmet 2 Such as Berard, Eato, Jese ad Kortum (2003) ad Melitz (2003). 3 Amog others, some promiet examples are Revega (1992), Pavcik, Goldberg ad Attaasio (2004) ad Kletzer (2002). See Slaughter (1998) for a overview. 4 Amog others, we ca list Davidso ad Matusz (2001) ad Kambourov ad Maovskii (2003) as examples of structural trade models with a special emphasis o labor. 2

3 of ex-ate ad edogeous worker heterogeeity compared to ACM. With the itroductio of ex-ate ad edogeous worker heterogeeity (age, educatio ad experiece), we ca aalyze distributioal effects of trade liberalizatio i more detail i additio to dyamics of the adjustmet process. Aother differece is, we use NLSY data set alog with CPS, which provides detailed work history. Usig work history of workers, we model huma capital accumulatio process joit with sectoral mobility. By doig so, we ca calculate welfare loss of workers i import competig sector, who lose their sector specific huma capital as their sector shriks, ad welfare gais of workers i exportig sectors at the same time. The fial major differece is the sector opeig to free trade, i cotrast to their aalysis o the maufacturig sector, we focus o liberalizatio i the metal maufacturig sector. Sice metal maufacturig output is ot directly cosumed, it has o effect o cosumer price idex. Therefore, we have to allow outputs of sectors to be used as iputs i the productio fuctio (otherwise free trade of metal maufacturig product would ot directly affect other sectors). Studyig a very small sector, such as metal maufacturig, with ACM s method would require a ureasoably large data set eve without worker heterogeeity. 5 I this paper, we will particularly focus o effects of trade liberalizatio o differet geeratios. Recet Pew Global Attitudes survey, coducted i 2002, showed that youg people are more ethusiastic about free trade compared to older people. The egative correlatio betwee age ad support for free trade is previously uexplored by ecoomists. Without cosiderig ay ecoomic explaatio, oe could attribute this egative correlatio (betwee age ad supportig free trade), to older people s beig more comfortable with status-quo compared to youg. I other fields of social scieces, there are studies aalyzig age ad opeess to chage, amog others Na ad Duckitt (2003) report that youg Koreas are more ope to chage compared to old. Takig those ad other researchers seriously, oe ca claim that older people s beig more coservative ca potetially explai their attitude towards free trade. However, we show that there is ideed a ecoomic explaatio as well for this itriguig differece betwee youg ad old workers, although it was uoticed so far by ecoomists. We do ot dey the possibility of existace of psychological factors, which are out of the scople of this paper. Imagie that all workers were perfectly mobile across sectors, the all workers would be 5 Needless to say that their method also has some other advatages over ours. For example, it requires shorter time series sice their method does ot rely o calculatio of future values. 3

4 uaimously better off or worse off after trade liberalizatio, as i Heckscher-Ohli model. If all workers were absolutely immobile ad attached to certai sectors, the there would be clearly distict wiers ad losers from free trade, as i Ricardo-Vier model; i that case, workers sectors would determie their gai ad loss. I reality, mobility costs probably lie betwee these two extremes ad vary across groups. A major source of variatio i mobility has to do with the age of affected workers, causig differeces i their positio towards free trade. I order to explore the relatio betwee worker mobility, age ad welfare effects of free trade, we first estimate huma capital accumulatio process ad mobility of workers joitly with a sectoral choice model, usig NLSY ad CPS data. The, we calibrate productio, iput demad ad cosumptio demad fuctios to set a geeral equilibrium framework with the estimated sectoral choice parameters. Fially, we simulate a hypothetical trade liberalizatio i the metal maufacturig sector (which has bee especially vulerable to trade shocks i the past, the steel idustry i particular) to aalyze gradual adjustmet of labor, wages ad prices i all sectors i respose to this trade shock. The trade shock ca be cosidered simply as a exogeous reductio i the metal maufacturig product s price, as cheaper imports will be available with trade liberalizatio; everythig else will be edogeous. It should be oted that, although we put a special emphasis o huma capital accumulatio process, our ultimate goal is to simulate trade shocks, which will make our model differet from similar research i labor literature such as Keae ad Wolpi (1997) ad Lee ad Wolpi (2004). We are borrowig some isights from sectoral/occupatioal choice literature to aalyze a topic which ca ot be explored with other iteratioal trade ecoomics tools. We deviate from them by modelig sectoral choice rather tha occupatioal ad educatioal choice, sice a trade shock first hits sectors, the possibly occupatios i a less direct way. Aalyzig effects of trade liberalizatio o occupatios is out of the scope of this paper, ad a very good example o that topic is Kambourov ad Maskii (2003). I additio to this deviatio, we reduced the state space by usig a limited umber of age, educatio ad experiece groups, so that the couterfactual simulatios are computatioally feasible. This simplificatio decreases the computatioal burde sigificatly by reducig the time eeded to calculate value fuctios, which makes our model relatively easy to im- 4

5 plemet compared to other logitudial-structural models. Sice we have edogeous wages ad multiple sectors, models developed by labor ecoomists would ot be computatioally feasible for our purpose. Fially, we have icluded idiosycratic utility shocks i additio to wage shocks, similar to Sulliva (2004) ad ACM. Iclusio of these shocks is ecessary because deviatios i wages ca oly explai a very small part of labor mobility, see two papers metioed earlier for more discussio. Oe importat questio is: Why old workers are less mobile tha youg workers? Followig the previous literature we ca give several differet aswers to this questio: For example Borjas ad Rose (1980), attributed decreases i mobility with age to icreases i wages with teure. The decrease i mobility with age ca be attributed to specific huma capital as i Topel (1991), better job match as i Jovaovic (1979) or implicit cotracts as i Lazear (1979). Groot ad Verbere (1997) suggested that the decrease i mobility with age ca be partially attributed to o-fiacial reasos as well. Ufortuately, we will ot be able to icorporate all these features i our model at the same time: we assume that workers become less mobile as they get older because they become more likely to hold sector specific huma capital ad the other reasos will be captured by their implicit movig cost a la Groot ad Verbere (1997). Although models of sector specific huma capital is less commo i the literature compared to firm specific huma capital; Neal (1999) shows that it is a very importat part of huma capital. Aother related lie of research to ours is the displaced workers literature, such as Jacobso, LaLode ad Sulliva (1993). Although they also aalyze distributioal effects of trade liberalizatio, their aalysis is focused o import competig sector workers oly. They study oly import-competig sector workers with a atural experimet, therefore their results ca be cosidered as more precise, however they ca ot explai what will happe i a hypothetical sceario ad what will happe i other sectors, such as service. I the ext sectio we will preset the model, followed by estimatio results. After a sectio o couterfactual simulatio of trade liberalizatio i metal sector, we will itroduce uobserved heterogeeity to the model. The we will coclude the paper. 2 Model Cosider a ecoomy with I idustries, where workers choose a sector to work i dyamically i each period. Aggregate productio fuctios for each sector has Cobb-Douglas form, 5

6 where workers wages are margial product of labor derived from the productio fuctios. Workers prefereces are also expressed with Cobb-Douglas utility fuctios. Our goal is to simulate a hypothetical trade liberalizatio i oe of the sectors (metal maufacturig) ad see how labor allocatios, prices ad wages adjust after the trade shock. We will discuss welfare effects of this policy chage o workers from differet age, educatio ad experiece groups. The parameters of the workers problem are estimated from NLSY79 ad CPS, while the parameters of the productio fuctios are calibrated from BEA data. The idustries are aggregated ito 4 mai sectors: 1. Mauf : Maufacturig ad Agriculture (tradable sector), 2. Metal : Metal Maufacturig (sector subject to policy chage) 3. Service : Service except Trade (o-tradable sector) 4. Trade : Wholesale ad retail trade (aother o-tradable sector). The idustries are aggregated maily i two groups, tradeable ad o-tradeables. However, sice wholesale ad retail trade is a very large idustry, we decided to take it as a separate sector apart from service. 6 I the ext sub-sectio we will describe workers problem. 2.1 Workers Assume that there are N workers ad I sectors i the ecoomy. Workers choose a sector i which to work i each period. If a worker idexed by decides to work i sector i the d t = i where d t {1, 2, 3, 4}. (1) A worker,, receives wage wt i from workig i sector i. Wage of worker is defied as the price of sector specific huma capital, rt, i times uits of huma capital hold by the idividual, h i t : w i t where uits of huma capital is defied as = r i th i t, log h i t = φ i 0 +φ i 1Coll +φ i 2SecEx i t +φ i 3MktEx t +φ 4 Coll SecEx i t +φ 5 Coll MktEx t +zt i, (2) 6 A more favorable approach could be usig two digit defiitios to separate idustries which have differet characteristics tha the others such as Agriculture, Professioal, Govermet, etc. There are two reasos that prevet us from doig so: First, Icreasig umber of choices will make the problem computatioally ifeasible, secod, the mai dataset we use, NLSY79, has a fairly small sample size therefore estimates of importat sector specific parameters would ot be sigificat. 6

7 where Coll i is a dummy for college educatio, SecEx i t is years of sectoral experiece i sector i, MktEx t is years of market experiece ad zt i is a iid ormal radom shock. Sice we are iterested i fluctuatios i rt i we will ot derive the stadard Micer wage equatio from this specificatio but rather move i a slightly differet directio. For coveiece, let us defie log h i t = φ i 0 + Xt i Φ i + zt i, where Xt i is a vector of idividual characteristics, Φ i is a vector of sector specific huma capital parameters except the itercept φ i 0. We ca write wages as a fuctio of average wages: wt i = w t i h i t, (3) h i t log w i t = log w i t + log h i = log w i t + ( X i t t log h i t, X i t ) Φ i + z i t. I order to reduce the size of state space we discretize variables Age t, SecEx i t MktEx t i such a way that they ca oly take the followig values: ad, Age t {26, 33, 40, 47, 54}, (4) SecEx i t {4, 11, 18}, MktEx t {4, 11, 18}. Although estimatio of the model without discretizig these variables ca be possible, simulatio is ot feasible with edogeous wages. To accommodate for discretizatio, we assume that i each year a worker with age, A, moves to the ext possible age, A + k, with a probability 1/k, where k = 7 i this case 7. Whe a worker reaches age 54 there is a 1/k probability that she will receive a lump-sum moey ad retire, to keep the populatio costat we assume that a ew worker eters system for each retiree. Whe age icreases market experiece icreases as well util 18 ad does ot icrease further. Sectoral experiece will evolve depedig o workers decisio to stay i their curret sector, the maximum value for sectoral experiece is also Figure 16 shows how vaue fuctios of maufacturig workers would look like with k=1 (actual) ad k=7 (approximatio) usig estimates from the ext sectio. 8 Estimatig the life-time wage-teure profile is out of the scope of this paper. We were ot able to idetify teure-wage profile for workers above 40 usig NLSY79, ad work history is ot available i CPS. 7

8 All sectoral experiece will be lost if a worker chages her sector: if d t d t 1 = SecEx i t = 4 (5) else if d t = d t 1 ad Age t = Age t 1 = SecEx i t else if d t = d t 1 ad Age t = Age t = SecEx i t = SecEx i t 1, = SecEx i t 1 + 7, I additio to the wage, each worker,, receives a idiosycratic radom utility, u i t, from workig i sector i. Where u i t is distributed mea zero extreme-value with variace π2 6 ν2, see Patel, Kapadia ad Owe (1976) for properties of the extreme-value distributio. Hece the istataeous utility of beig i sector i at time t is U i t = w i t (s t,z t,ξ t )+u i t, (6) where wage, wt i (s t,zt i,ξt) i is a fuctio of the state variables, radom shock ad average wages as described above, ad ξ t = [ w t 1.. w t I h 1 t.. h I t ] is a vector of relevat aggregate state variables. The o-radom state vector s t is cosist of educatio, previous period s choice, sectoral ad market experiece: s t = [ d t 1 Coll SecEx d t 1 t MktEx t ]. (7) The o-radom state vector s t ca also be cosidered as type of a worker. Workers who move from sector i to j will icur a movig cost, C ij, if they chage their sectors, so C ij > 0 if i j ad C ij = 0 if i = j. If a worker chages her sector, the movig cost fuctio of her age, educatio ad the sector she is movig i will be a C ij t = C Age t 1 + C j 2 + Coll C 3, this movig cost should ot be take literally as fiacial cost, it will accout for all Lee successfully estimated wage-teure profile from CPS via MSM, however their method does ot allow us to idetify parameters of the idiosycratic utility shock. We simply assume that for the last two age groups wage cotributio of market ad sectoral experiece stay costat followig the geeral shape of teure-wage profile estimated i previous works, such as Micer (1958). 8

9 umodelled frictios ad psychological costs as well. For otatioal simplicity, cosider a vector of all relevat state variables for a idividual, which are s t, ξ t, zt =[zt 1..zt I ] ad u t =[u 1 t..u I t ], such that ηt =[s t ξ t zt u t ]. The objective of a idividual for ay time t =1,..., T is to maximize her preset discouted total utility followig a Bellma equatio: where sector (alterative) specific value fuctios are: ( V t (ηt ) = max V i t (ηt ) ), (8) i V i t (ηt ) = Ut i (ηt )+Emax β { ( ) V ( j t+1 η t+1 C i,j st+1)}, (9) j ( ) ( ) = Ut i (ηt )+βe s Ω i t+1 η t+1 + βeu,z,s Vt+1 i η t+1, for all periods where β is the discout factor. Thus, we ca write the optio value of movig as Ω i t+1 ( η t+1 ) = Eu,z ( max j { ( ) ( ) ( V j t+1 η t+1 V i t+1 η t+1 C i,j st+1)} ), (10) ad ote that Ω i t+1 (η t ) ca be calculated aalytically upto a certai level. {See Appedix B for details. } Timig: At ay give time period t the order of evets for a worker is as follows: 1. Pays the movig cost C > 0 if her previous sector is differet. 2. Works ad ejoys her utility: wt i +u i t, 3. Lears the ext period s radom shocks { zt+1,ut+1} j j Chooses her sector. j=1 5. Eters the ext period t+1 ad repeats steps 1-5 for t+1. Note that there is o aggregate ucertaity i the model except for the shock therapy (e.g. ξ t,ξ t+1,ξ t+2,..., ξ T,... are kow at time t). Estimatio of workers problem: Usig the equatios above we ca calculate probability of a worker s trasitio from state s to s {see Appedix B for details.} let us deote this probability with m ss, which is a fuctio of s t ad expected ext period alterative-specific values for each state. I additio to the trasitio probabilities m ss, it is also possible to calculate probability of observig wage w t give s type s t ad average wage i her sector: w i t. The estimatio 9

10 strategy is to maximize the log-likelihood fuctio Λ= N T =1 t=1 log m s t 1 s t Pr (w t ) du t dzt. z u Note that it is possible to solve the itegral over u t aalytically i a way similar to multiomial logit models, however, the itegral over zt has to be calculated umerically with a quadrature or simulatio based method. Thus, we use method of simulated maximum likelihood to estimate the parameters of iterest. { See Appedix B for details. } 2.2 Aggregate Ecoomy: Let l (s) be the ratio of workers with a give state s {1,..., S}, where s is a idex represetig type of a worker, where total umber of workers are ormalized such that 1= S s=1 l (s). There are 96 possible states thus S =96 types of workers. For ay give type s we ca calculate time t labor allocatio give time t 1 labor allocatio ad trasitio probabilities l (s t )= S s t 1 =1 m s t 1s t l (s t 1 ). (11) Let L t be a vector represetig the distributio of workers such as L t =[l (1)... l (96)]. The L t ca be expressed as a fuctio of previous period s distributio L t 1, average wages ad expected ext period alterative specific values: L t = M ( L t 1,ξ t, V 1 t+1 (1),.., V I t+1 (S) ), (12) where V i t+1 is expected alterative specific value ad ξ t is a vector of average wage ad huma capital levels. Average wages are edogeous from the aggregate perspective. Let productio fuctios be y i t = B i ( L i t h i t) b i L ( K i ) b i K I j=1 ( ) q ji b i j t, (13) where L i t is the ratio of workers i sector i, where L i t = s S i l (s) ad S i is the set of types of worker where d t = i (types that are from sector i), h i t is average huma capital, K i is capital, ad q ji t is amout of product from sector j used i productio i sector i. 10

11 We assume that capital is specific to sectors similar to Ricardo-Vier models as i ACM. (We also experimeted with perfect capital mobility simulatios ad foud that qualitative implicatios of the model, i geeral, are uchaged). Each worker will receive her real margial product, give price level p i t ad cosumer price idex ϕ t : thus w i t = pi t ϕ t y i t L i t h i t, (14) h i t w t i = pi t yt i. (15) ϕ t L i t Therefore the average wages ca be writte as a fuctio of L i t ad h i t (hece the distributio of workers L t ), p i t ( ) wt i = b i L L i b i L 1 t hi t hi t ζt i, (16) ϕ t where ζt i is a part of the Cobb-Douglas productio fuctio, ζt i = B i (K i ) bi K I ( ) j=1 q ji b i j t. Fially cosumptio prefereces are described by a simple utility fuctio: where q jc t Υ t = I j=1 ( ) q jc θj t, (17) is quatity of j cosumed at time t ad θ j s are the weights. I the ext sectio we discuss estimatio results of workers problem ad calibratio of productio fuctios. 3 Estimatio ad Calibratio of Parameters: We are iterested i estimatio to fid plausible parameters for the simulatios, thus i cotrast to papers from labor ecoomics literature, our mai focus will be couterfactual trade liberalizatio simulatios. We estimate huma capital accumulatio ad mobility parameters joitly, from each worker s simulated likelihood cotributio. The we calibrate productio fuctio parameters from Breau of Ecoomic Aalysis data, assumig Cobb-Douglas forms. Data For estimatio of the workers problem, we use 1979 cohort of the Natioal Logitudial Survey of Youth (heceforth NLSY) as our mai data set. NLSY is widely used i estimatio of occupatioal choice models, sice it follows idividuals over years ad provides 11

12 detailed iformatio o work history. The sectoral experiece variable, SecEx i t, ca be easily costructed from NLSY. Oe importat restrictio of NSLY is it follows idividuals aually util about age 40, so we ca ot idetify parameters for older idividuals i the model. I order to idetify movig cost parameter C ij for idividuals over 40 we iclude Curret Populatio Survey March sample (heceforth CPS) i our estimatio. Sice sectoral experiece is ot observed i CPS, we ca ot use it to calculate likelihood cotributio of observed wages. The average wages are also calculated usig CPS because NLSY sample size is much smaller. Iitially, NLSY has 12,686 people i the sample, cosistig of 6,403 males ad 6,283 females. Like most of the other mobility models, we oly pick males for our sample (such as Keae ad Wolpi (1997) ad ACM). Moreover, we take blacks ad Hispaics out of our sample, who are over-sampled by the NLSY, agai followig the previous research. This reduces our sample size by approximately 40%, so we are left with 3,790 idividuals. The idividuals i our sample are betwee ages of 14 ad 21 as of year We use observatios, from years 1983 to 1994, of idividuals who were at least 23 years old, worked at least 26 weeks i the observed year ad who do ot have ay missig idustry iformatio from previous years. For example if a certai idividual s data is missig for 1990, we do ot use him after 1990 sice we ca ot costruct sectoral experiece data for him after the missig observatio. If a certai idividual is observed less tha 7 years betwee 1983 ad 1994, we take him out from the sample. We do ot use observatios of idividuals whose implied full time real aual wage icome is less tha $5000, or more tha $300,000 (where the base year is 2000). We ed up with 1190 idividuals i the sample. Neal (1999) reports that there are codig errors i NLSY79 regardig occupatios. A similar error is also preset for idustry codigs. I order to miimize this problem, we use the followig method as i Neal (1999); wheever a sector chage is reported, we require that the worker has to chage his employer as well, otherwise it is cosidered as a codig error ad the origial sector is kept. Teure of workers with their curret employer is reported i NLSY. The CPS sample is from 1983 to 2001 ad costructed i a similar way: We use white ad male idividuals, who are betwee 23 ad 57, ad who worked at least 26 weeks i a give year. We have a miimum of 11,857 ad a maximum of 20,211 people i our fial 12

13 sample betwee the years 1984 ad 2001 (sample size chages every year). I CPS, reported mobility rates are 5 moths mobility rather tha aual mobility, we follow a procedure similar to ACM to correct trasitio probabilities. Table 1: Distributio of Workers. Pael A: Sectors Sector NLSY CPS Mauf 27.7% 27.2% Metal 3.8% 3.3% Service 49.5% 52.9% Trade 19.0% 16.7% Pael B: Age Age NLSY CPS 23 to % 18.6% 30 to % 25.4% 37 to 43 NA 23.4% 44 to 50 NA 18.5% 51 to 57 NA 14.3% Pael C: Sectoral Experiece Experiece NLSY CPS 1 to % NA 8 to % NA 14 to 18 NA NA Pael D: Educatio Educatio NLSY CPS No-college 60.3% 59.1% College 39.7% 40.9% Table 1 summarizes distributio of workers across sectors, age, sectoral experiece ad educatio groups i both NLSY ad CPS samples. Note that sectoral experiece is ot available for CPS sample ad NLSY sample icludes people oly up to age 40. Maufacturig ad agriculture workers (heceforth mauf.) are approximately 28%, metal workers are about 3%, service workers except trade workers (heceforth service) are about 50% ad fially 13

14 wholesale ad retail trade workers (heceforth trade) are close to 20% of the total sample (see Pael A). Ulike ACM, our aalysis does ot rely o calculatio of aggregate trasitio probabilities, which allows us to have a very small sector, such as metal. Calculatio of aggregate trasitio probabilities requires observig some workers from each sector movig to every possible directio, which is impossible whe oe of the sectors is small or whe there are may worker types. Pael B shows age distributio ad Pael C shows sectoral experiece distributio i the sample. As illustrated i Pael D, about 40% of workers have at least oe year of college educatio i the sample. Table 2: Trasitio Probabilities (CPS). Pael A: No College Educatio Age Mauf Metal Service Trade 23 to to to to to Pael B: Some College Educatio Age Mauf Metal Service Trade 23 to to to to to Pael C: Trasitio Marix Mauf Metal Service Trade Mauf Metal Service Trade Table 2 shows example trasitio probabilities from differet age groups, educatio groups ad sectors. Pael A presets probabilities of sector chage for workers with o 14

15 college educatio while Pael B shows for those with at least oe year of college educatio. The effect of educatio o probability of sector chage is ambiguous. However, it is clear that probability of sector chage is decreasig with age for both educatio groups. Pael C shows trasitio probability from oe sector to aother. As oe would expect, probability of movig out of a larger sector is lower tha probability of movig out of a smaller sector, ad probability of movig i a larger sector is higher tha probability of movig i a smaller sector. Estimatio Results Wages are deated by the CPI, ad ormalized so that over the whole sample the average aualized wage is equal to uity. Numbers reported i Table 3 are comparable to average aual wage, which is equal to uity. This is because the istataeous utility icludig the movig cost ca be expressed i a moey metric form for workers who chage sector: Ut i = wt i + u i t C ji where j is last year s sector. However, the umbers i Table 4 are ot comparable with the umbers i Table 3 because wage equatio is expressed i a log form. The estimatio results show that idiosycratic shocks play a very importat role i decisio of workers. The parameter of the extreme value distributio ν is about 1.5 (reported i Table 3 - Pael A) which meas that stadard error of the idiosycratic utility shock is approximately double of average aual wage. I the Table 4 - Pael A the stadard error of wage shock. σ z is reported as The estimated movig cost, reported i Table 3 - Pael B, starts from 4.5, icreases with age, ad ed up beig as large as 5.9 for the oldest type. These umbers are large but ot surprisig sice similar projects with idiosycratic utility shocks such as Sulliva (2006), Kea ad Walker (2003), ad ACM also report such large mobility cost. For example ACM fid a movig cost equal to 6.5 times of average aual wage. We preset estimatio results of a exteded model with uobserved heterogeeity to shed some light o possible reasos of this urealistically large movig costs i a later sectio. Pael C shows the sector specific compoet of the movig cost, which is how much more (or less) the movig cost would be depedig o the sector a perso chooses to work i. The movig costs reported i Pael B ca be 3.2 more if a worker is movig to metal sector (which is the smallest sector) or -1.5 less if a perso is movig to service sector (the largest sector). Pael D shows that people with some college educatio bear larger movig costs, this ca be attributed to 15

16 Table 3: Estimatio - Movig Cost. Pael A: Idiosycratic Shock Parameter Coefficiet t-stat ν Pael B: Age specific cost Age Coefficiet t-stat C1 1 Age C1 2 Age C1 3 Age C1 4 Age C1 5 Age Pael C: Additioal sector specific cost Sector Coefficiet t-stat C2 1 Mauf 0.00 N/A C2 2 Metal C2 3 Service C2 4 Trade Pael D: Additioal educatio specific cost Educatio Coefficiet t-stat C 3 College the fact that people with more educatio ear higher wages. Because of higher wages, their wage offers fluctuate more i levels (obviously ot ecessarily i logs), hece for a similar mobility rate, as reported i Table 2, people with more educatio should face larger movig costs. Table 4 reports estimates of the wage equatio related parameters. Pael B shows that retur o educatio is the highest for service sector ad the lowest for metal sector. This is a expected result sice sectors like professioal, fiace, public are parts of service sector. Retur o sectoral experiece for differet sectors ad educatio levels is show i Pael C. Fially, Pael D reports retur market experiece (which is almost the same as age i our model sice we do ot model labor force participatio as a alterative choice). 16

17 Table 4: Estimatio - Huma Capital. Pael A: Std. Dev. Of Wage Shock Coefficiet t-stat σ z Pael B: Educatio Parameters Sectors Coefficiet t-stat φ 1 1 Mauf φ 2 1 Metal φ 3 1 Service φ 4 1 Trade Pael C: Sectoral Experiece Parameters Sectors Coefficiet t-stat φ 1 2 Mauf φ 2 2 Metal φ 3 2 Service φ 4 2 Trade Educatio Coefficiet t-stat φ 4 College Pael D: Market Experiece Parameters Sectors Coefficiet t-stat φ 1 3 Mauf φ 2 3 Metal φ 3 3 Service φ 4 3 Trade Educatio Coefficiet t-stat φ 5 College

18 Calibratio The parameters b i L, bi K, bi j ad ζa i are calibrated from the Bureau of Ecoomic Aalysis data, they are reported i Pael A ad Pael B of Table 5. We simply assume that cost shares of labor ad iputs are parameters of the Cobb-Douglas productio fuctios. We pick ζa i s such that observed average wages are as close as possible to the implied wages, give the distributio of labor. The parameters of Cobb-Douglas utility fuctio, θ i s are calibrated from Cosumer Price Idex data, which are reported i Pael C of Table 6. ACM also follow a similar calibratio method ad provide more detail. For the estimatio ad the simulatios, we assume that discout factor β is equal to We were ot able to estimate β sice it is poorly idetified by our model. Table 5: Calibratio - Productio ad Utility Fuctios. Pael A: Productio Fuctio Iput Shares Mauf Metal Service Trade Labor Capital Mauf Metal Service Trade Pael B: Productio Fuctio Costat Mauf Metal Service Trade logζa i Pael C: Utility Fuctio Shares Mauf Metal Service Trade θ i Simulatio: 4.1 Autarky Steady State: As first step for the couterfactual exercise, we simulate autarky steady state to calculate the iitial labor distributio, L t, which will gradually coverge to the free trade distributio after the trade shock. Here we use the subscript A istead of t to refer to ay time period 18

19 before the shock. We use equatios from previous sectios to illustrate autarky steady state: p i A ( ) w A i = b i L λ i b i L 1 A hi A hi A ζa i, (18) ϕ A L A = M ( L A,ξ A, V A, 1.., V A) I, V A i (s, ξ A ) = w A i (s, ξ A )+βe s Ω i A (s i,ξ A )+βe s V A (s,ξ A ), h i A = exp ( Xi A Φ i), where w i A is the expected autarky wage for type s workers i sector i, ad ξ A = [ w 1 A.. wi A h 1 A.. h I A ] is the vector of autarky average wage ad huma capital levels, Xi A is the average of huma capital parameters used i wage equatio (3). For simplicity we assume that p i A =1 ad ϕ A = 1. Note that λ i A ad h i A ca be calculated from L A, sice they are ratio of workers ad average huma capital i sector i respectively, where L A is distributio of types i the ecoomy. Other relevat variables are output: y i A = ( λ i A h i A) b i L ζ i A, ad icome spet o i: µ i A = I j=1 [ b j i + θ ( i b j L + )] bj K y j A pj A. Defie X A =[wa 1.. wi V A A 1.. V A I l (1).. l (96)], the simulatio exercise ca be defied as a problem of fidig a fixed poit X A = F (X A ) where F (.) is a fuctio described by the set of equatios (18). The fixed poit is calculated umerically, similar to Artuc, Chaudhuri ad McLare (2008). 4.2 Trasitio: We assume that with the abolishmet of tariffs i the metal sector, the prices will decrease about 30%, thus p Metal =0.7 whe t>0. First, we have coditios for the trasitio, similar to the autarky steady state coditio (18) : 19

20 ( ) w t i = b i L λ i b i L 1 t hi t hi t ζt i p i t, (19) ϕ t L t = M ( L t 1,ξ t, V t+1, 1.., V t+1) I, V t i (s t,ξt ) = w t i (s t,ξ t )+βe s Ω i i t+1 (s t+1,ξ t+1 )+βe s V t+1 (s t+1,ξ t+1 ), h i t = exp ( Xi t Φ i). However this time we ca o loger assume that p i t = 1 or ϕ A = 1, sice prices chage durig trasitio. I additio to prices the iputs used i productio will also chage, thus the parameter ζt i will be differet from the calibrated parameter ζa i. We ormalize these parameters with their autarky values, such that x t deotes x t /x A. The chage i cosumer price idex ca be calculated with the chage i prices ϕ t = I ) i θi ( p t (20) i=1 The chage i prices (for service ad trade) ca be calculated with the chage i icome spet o each product ad quatities produced, simply by exploitig the Cobb-Douglas form of demad ad productio fuctios µ i t = I j=1 p i t = µi t µ i A [ b j i + θ ( i b j L + )] bj K y j t p j t, (21) The chage i ζ i t ca be calculated usig the chage i prices ad output 1. ỹ i t q ij t = pi t ỹt, i (22) ζ i t = p j t I j=1 ) ij b i ( q j t. Fially, the chage i output ca be writte as a fuctio of chage i total huma capital 20

21 i sectors ad the chage i ζ i t V 1 t ( ) λ ỹt i i b i L = t hi t ζi λ t. (23) i A h i A Defie vector X t =[wt 1.. wt I.. V t I ỹt 1.. ỹt I p 1 t.. p I t ], ad matrix X =[X 1 X 2.. X T ]. We assume that for t>t, X t = X t 1. Let X = G (X, L A ) be a fuctio defied by equatios (19), (20), (21), (23) ad (22). Fidig the trasitio values is also equivalet to fidig a fixed poit give the autarky labor allocatio, L A. Similar to the autarky problem, this problem ca also be solved umerically. We check if X T is ideed the free trade steady state, if ot we icrease T ad fid aother fixed poit. See Artuc, Chaudhuri ad McLare (2008) for details. 4.3 Results We first simulate the model for the autarky steady state, where all prices are ormalized to uity. The, we assume that a shock-therapy trade liberalizatio i the metal sector decreases its product s price by 30%, forcig it to be equal to the world price at t = 1. We assume that service ad trade sector outputs are o-tradable, while maufacturig sector output is tradable, which makes maufacturig price costat over time. Give the iitial autarky labor allocatio, outputs ad prices, we calculate trasitio of labor allocatio, wages, values of workers, prices, output ad demad of goods after the trade shock. Figures 1-3 show gross flows of workers, that is percetage of workers leavig their sectors. Figure 1 is for workers with average 4 years of sectoral experiece ad who are about 26 years old. Iitially, approximately 19% of trade workers leave their sector, ad 14% of other workers leave their sector every year. After the trade shock, this percetage icreases to 29% for metal workers, decreases to 13% for mauf workers ad stays approximately the same for trade ad service workers. Figure 2 presets gross flows of workers with average 4 years of sectoral experiece ad who are about 54 years old; while Figure 3 shows gross flows of 54 years old workers with at least 18 years of sectoral experiece. The geeral tred of gross flows are the same i all three figures: Sectoral mobility decreases with age ad experiece. Figure 4 shows the adjustmet of wages after the trade shock: We fid that gross flow of workers are egatively correlated with wages. The adjustmet process of metal sector is very simple but iterestig: Iitially price of metal sector product decreases, causig 21

22 the wages i metal sector to decrease. Workers start leavig metal sector, causig large out-flows, which evetually icreases wages i metal sector. The log ru free trade wage i metal sector is, however, lower tha the autarky wage. For the other sectors, the adjustmet process is quite subtle: First, it should be oted that metal sector product is ot cosumed, but used as a iput i mauf ad metal sectors. Therefore, metal price has o direct effect o CPI or real wages. However, the decrease i metal price icreases its use as iput, ad icreases the margial product of mauf workers, causig a icrease i wages. Other sectors wages are do ot chage much after the shock. Figure 5 displays chage i output, showig a otable icrease i the mauf output ad a very large decrease i metal output. Like wages, the other sectors outputs are ot sigificatly affected from the trade shock. Figure 6 shows the chage i demad (from both cosumers ad producers) for each product. As expected, demad for metal icreases sigificatly with the price cut. Demad for mauf also icreases, eve though there is o chage i its price, because of the geeral icrease i output after the shock. (Sice there are o market failures or exteralities i our model, it is safe to assume that GDP icreases after the trade liberalizatio although we do ot explicitly calculate it). The demad for service also icreases, but the chage is much smaller compared to mauf because service price icreases with demad. Note that mauf price is costat as it takes the world price. Figure 7 shows chage i prices. Goig back to Figure 4, the chages i service ad trade prices pull the service ad trade wages up, while workers movig i from metal sector push wages dow. These two opposite forces cause wages i service ad trade chage oly slightly: a small icrease for service ad a small decrease for trade. Figures 8-11 show how low-skill workers value chage right after the trade shock. 9 (The results for high-skill workers are very similar ad show i Figures 12-15). Sice metal sector output is used itesively oly i mauf ad metal ad ot cosumed, service ad trade sector workers are less affected from the trade shock compared to others. Note that size of metal sector is fairly small ad flows out of metal sector do ot sigificatly chage labor allocatios i other sectors. (We ca safely igore tariff reveues from metal 9 Although we origially have oly 96 types of workers, as show i (4), we ca calculate values for more a precisely partitioed state space, e.g. Age {23, 24, 25, 26, 27,..., 57}, usig the aggregate distributio of workers ad wages from the simulatios. 22

23 sector because of its small size). Figure 8 shows that mauf workers, i geeral, beefit from the shock, cosistet with the fact that output icreases more compared to the icrease i umber of workers i mauf. The gais icrease with age ad experiece, reachig maximum level for middle aged people who have worked i the mauf sector for their etire life. The gais decrease with age after a certai age because expected time horizo to ejoy beefits of free trade decreases. Whe a worker is sufficietly close to retiremet, she would oly care about purchasig power of her retiremet savigs. We assume that whe a worker retires, she receives a lump-sum moey which is ot a fuctio of price levels, thus retired workers are worse off after the shock sice prices of services icrease after the shock. A equally plausible assumptio would be assumig a iflatio-protected retiremet beefits scheme, the retired workers would be uaffected from the shock. Sice the differece is trivial we do o show simulatio results for the alterative approach. However, we believe that to shed more light o this issue, a more detailed retiremet model is required, which is out of the scope of this paper. Youger workers beefit less compared to middle aged, because optio value of youger people decrease after the trade shock - which is a sigificat portio of their value as show i (10). As a worker s movig cost icreases, her optio value decreases, so it becomes fairly small whe a worker gets old. After the shock metal sector becomes a uattractive alterative for workig i mauf, causig a otable drop i youg mauf workers optio value. It is very ulikely for a perso, who has bee workig i mauf for may, years to move to other sectors, so developmets i other sectors do ot chage older workers optio value much. Figure 9 shows chage i metal workers value, which is almost the opposite of previous figure. Youg workers are hurt less because of the icrease i their optio value, while older workers are hurt more. Older ad middle aged workers iability to move to other sector makes them lose more compared to youg. The story for the workers, who are close to their retiremet, is exactly the same as mauf sector. Figure 10 displays chage i values of service ad Figure 11 shows chages i values of trade workers. These two sectors are oly slightly affected from the trade shock as they are ot usig metal i their productio. Service workers are slightly better off, while 23

24 trade workers are slightly worse off. The geeral shape of the value fuctios surface is similar to other sectors: Middle aged workers beefit more if they workers beefit i geeral, ad hurt more if workers are hurt i geeral (compared to youg). We would like to call it a mirror effect. Trade shocks affect older people by a large scale both positively ad egatively, while youger people are affected usually i the same directio as older people i their sector but by a much smaller scale. (Yet, it is theoretically possible that old people are worse off i import competig sector while youg people are better off.) Our results show that there is a relatio betwee age ad supportig free trade, but they do ot explai why older people would be less supportive of free trade, as it is see i the Pew Survey. 10 The couterfactual exercise preseted here is of a very small sector, so the distributioal effects of a log term globalizatio ca ot be studied from our graphs. Aalyzig globalizatio i detail ad explaiig results of the Pew Survey is out of the scope of this paper. But just to shed some light o this iterestig issue, we simulated a hypothetical trade liberalizatio i mauf sector. We fid that all workers except middle aged ad older mauf workers beefit from trade liberalizatio, icludig youg mauf workers. So a liberalizatio i mauf sector would be perfectly cosistet with the Pew Survey. We also experimeted with perfect capital mobility ad foud that qualitative implicatios of our base simulatios are uchaged. (results available upo request). 5 Uobserved Heterogeeity: A Exteded Model As we discussed earlier, the movig cost we estimate should ot be take as a fiacial movig cost sice there are may umodelled frictios i the labor market which is captured by the movig cost we estimate. We assume there is uobserved heterogeeity, particularly two types of workers: type I - workers who ca move after payig some movig cost, ad type II - workers who ca ot move at all. Havig these two uobserved types will allow us to capture some of the umodelled frictios which would otherwise be captured by the movig cost. I the origial model we had 96 types, with iclusio of a biary uobserved heterogeeity we ed up with 192 types. We assume that ucoditioal probability of beig type II is α, probability of trasitio from type I to type II is γ 1 ad probability of trasitio from type II to type I is γ 2. Cosider this special case: If oe s type i the ext period is idepedet of her curret 10 We provide a simple aalysis of the Pew Global Attitudes Survey i Appedix C for descriptive purposes. 24

25 type the, γ 1 = α ad γ 2 =1 α. For this special case, we ca thik of type I workers as those who ca get job offers this year, ad type II workers as those who ca ot get job offers. The ay frictio captured by iclusio of uobserved types for this special case ca be cosidered as search frictios. Aother uderlyig frictio ca be time persistece of utility or wage shocks. If a worker likes oe sector better tha others, because of fiacial or o-fiacial reasos, her preferece this year is most likely correlated with her preferece last year. For example, the worker might be a very taleted sales-perso resultig i higher wages for her i trade sector, or her spouse might also be a sales-perso ad she might wat to work i the same compay with him. Both of these shocks are time persistet, icosistet with our iid assumptios. By itroducig two uobserved types, we allow a simple persistece i the idiosycratic movig cost. Probability of beig type II ucoditioal o history depeds o observed state of a idividual. Assume that probability of beig type II give the observed characteristics is α (s t ), where s t is the observed characteristics (or state) of the idividual. This probability ca easily be calculated from the aggregate distributio vector, L t. Also assume that probability that a worker will stay i her sector is ϕ (s t ). If a worker has moved i last period, we are sure that she was type I i last period, so for her to be type I agai this period is Pr(I s t,s t 1 )=1 γ 1. Similarly if a worker has moved two periods before today, she was a type I the so her probability of beig type I today is where Pr(I s t,s t 1,s t 2 )= p 11 p 11 + p 12, p 11 = γ 1 γ 2 + (1 γ 1 ) 2 ϕ (s t 1 ), p 21 = (1 γ 2 ) γ 2 + γ 2 (1 γ 1 ) ϕ (s t 1 ), p 12 = γ 1 (1 γ 2 ) + (1 γ 1 ) γ 1 ϕ (s t 1 ), p 22 = (1 γ 2 ) 2 + γ 2 γ 1 ϕ (s t 1 ). 25

26 If a worker has ot moved i the last two periods, the probability that she is a type I is Pr(I s t,s t 1,s t 2 )= {1 α (s t 2 )} ϕ (s t 2 ) p 11 + α (s t 2 ) p 21 {1 α (s t 2 )} ϕ (s t 2 ) {p 11 + p 12 } + α (s t 2 ) {p 21 + p 22 }. Fially, if a worker s history is ukow the her probability of beig type I is Pr(I s t )=1 α (s t ). We use these probabilities i maximum likelihood cotributios of idividuals. Ideally, we could go back more i history of workers, but we prefer to go back oly two periods sice it is sufficiet to idetify the parameters we are iterested i. Goig back more i history would require computig all possible paths, which is computatioally (ad aalytically) ifeasible. Tables 6 shows that movig costs are much lower with the iclusio of uobserved heterogeeity. We ow fid that movig cost varies betwee 1.4 to 3.1, while it varies betwee 4.5 to 5.9 without uobserved heterogeeity. Other results, that are preseted i Tables 6 ad 7 are very similar to Tables 3 ad 4. Thus parameter estimates, excludig mobility costs, are ot affected much from the itroductio of uobserved heterogeeity. Aother iterestig parameter is α, which is estimated as high as This ca be iterpreted as, 75 percet of workers ca ot move either because of search frictios or time persistece of shocks. Fially, we fid that γ 1 =0.47, thus it ca be iferred that uobserved types show some persistece, sice it is much smaller tha Whe we repeat the simulatio exercise for the exteded model, we fid that iclusio of uobserved heterogeeity does ot chage the qualitative implicatios of the mai model (figures available upo request). 6 Coclusio We itroduced a model which ca be used to aalyze distributioal effects of trade liberalizatio. Our iitial settig was somehow similar to ACM, but we followed a completely differet ecoometric strategy which allowed us to itroduce richer ex-ate ad edogeous worker heterogeeity. Although this strategy preveted us from usig their compact Eulerequatio coditios, we simplified the estimatio process by discretizig state-space. Usig NLSY ad CPS, we estimated mobility parameters ad huma capital accumulatio process joitly. With estimates of these structural parameters ad calibrated productio fuctios, we simulated a couterfactual trade shock i metal maufacturig sector (which was subject 26

27 Table 6: Estimatio - Movig Cost (Uobserved Heterogeeity). Pael A: Idiosycratic Shock Parameter Coefficiet t-stat ν Pael B: Age specific cost Age Coefficiet t-stat C1 1 Age C1 2 Age C1 3 Age C1 4 Age C1 5 Age Pael C: Additioal sector specific cost Sector Coefficiet t-stat C2 1 Mauf 0.00 N/A C2 2 Metal C2 3 Service C2 4 Trade Pael D: Additioal educatio specific cost Educatio Coefficiet t-stat C 3 College Pael E: Types Probability Coefficiet t-stat α Pr(II) γ 1 Pr(I to II) γ 2 Pr(II to I) 0.16 N/A 27

28 Table 7: Estimatio - Huma Capital (Uobserved Heterogeeity). Pael A: Std. Dev. Of Wage Shock Coefficiet t-stat σ z Pael B: Educatio Parameters Sectors Coefficiet t-stat φ 1 1 Mauf φ 2 1 Metal φ 3 1 Service φ 4 1 Trade Pael C: Sectoral Experiece Parameters Sectors Coefficiet t-stat φ 1 2 Mauf φ 2 2 Metal φ 3 2 Service φ 4 2 Trade Educatio Coefficiet t-stat φ 4 College Pael D: Market Experiece Parameters Sectors Coefficiet t-stat φ 1 3 Mauf φ 2 3 Metal φ 3 3 Service φ 4 3 Trade Educatio Coefficiet t-stat φ 5 College

29 to shocks recetly, steel sector i particular.) We fid that: (1) The results show that estimated movig costs are large ad icrease further with age. Preferece shocks are importat i explaiig labor mobility, therefore psychological or uobserved factors play a role i mobility decisios. (2) High movig costs foud i this paper (ad i ACM) might be partially due to omissio of uobserved heterogeeity, which may be caused by search frictios ad persistece of preferece ad wage shocks. (3) After a trade shock i the metal sector, maily metal ad maufacturig workers would be affected sice output of metal sector is ot cosumed but used as iput i maufacturig ad metal sectors. (4) Metal workers would be worse off i geeral. However, youg workers would be much less compared to middle-aged, due to their ability to move to other sectors (hece high optio values). O the other had, Maufacturig workers would be better off. Agai youg workers much less compared to middle-aged because of the drop i their optio values. The relevat figures display a mirror effect i maufacturig ad metal sectors. (5) As workers get close to retiremet they should be uaimous sice they have less time to ejoy or suffer from the effects of free trade o their wages. To aalyze effects of trade shocks o very old workers, a more detailed modellig of savigs ad retiremet beefits is required, which is left for future research. 29

30 Appedix A: Alterative Estimatios To aalyze robustess of our results we have experimeted with alterative specificatios. The presetatio of results ad discussios are kept short to limit legth of the paper. Basic Model vs. Restricted Utility Shocks I the labor literature utility shocks are ot commo; i similar discrete choice models, such as Keae ad Wolpi (1997), labor allocatios are maily drive by wages shocks. To demostrate the effect of iclusio of utility shocks we itroduce a descriptive model with a very simple movig cost ad huma capital structure: C ij t = C 1 + C j 2, log h i t = φ i 0 + φ 1 Coll + φ 2 SecEx i t + φ 3 MktEx t + zt i, we estimate this model usig oly NLSY data. restrictio ν = The, we repeat the exercise with a Table 8: Basic Model ad Restricted Utility Shocks. Pael A: Basic Model Estimates with NLSY ν C 1 C2 1 C2 2 C2 3 C2 4 Coef tstat N/A σ z Sec. Exp. Mkt. Exp. College Coef tstat Pael B: Estimates with NLSY (Restricted Nu) ν C 1 C2 1 C2 2 C2 3 C2 4 Coef tstat N/A N/A σ z Sec. Exp. Mkt. Exp. College Coef tstat We do ot set ν equal to zero because otherwise we eed to make substatial chages i the computatios. The results are show i Table 8 Paels A ad B respectively. Here, 30

31 we show the cotributio of utility shocks to large movig costs. Wage shocks ca be easily idetified sice wages are observed. We observe that wages do ot fluctuate much, but workers chage sectors very ofte, which implies very small movig costs whe preferece shocks are omitted. Alterative iitial distributios We have to guess the iitial distributio of sectoral experiece for older workers i estimatio process, because we do ot observe workers sectoral experiece i CPS. We iterate labor allocatio equatios usig year 1983 wages to calibrate iitial distributio of CPS workers. The alteratively, we iterate usig average wages over the CPS sample (year 1983 to 2001). Usig simple descriptive model (itroduced above) we show that the iitial disributio of workers does ot affect our results sigificatly. Results show i Table 9 Paels A ad B. Table 9: Alterative Iitial Distributios. Pael A: Estimates with NLSY ad CPS ν C 1 C2 1 C2 2 C2 3 C2 4 Coef tstat N/A σ z Sec. Exp. Mkt. Exp. College Coef tstat Pael B: Estimates with NLSY ad CPS (Alterative) ν C 1 C2 1 C2 2 C2 3 C2 4 Coef tstat N/A σ z Sec. Exp. Mkt. Exp. College Coef tstat Appedix B: Key Equatios The expected utility fuctio used i programmig ca be expressed as: 31

32 V i t (s t ) = E u,z V i t (s t,z t,u t ) ( = E z wt i s t,zt, w t) i + E max β { ( ) V j t+1 η t+1 s ( t C i,j ηt+1)}, j { ( } = w t i (s t,ξ t )+βω i t+1 (s t )+βe s,u,z V i t+1 s t+1,zt+1,ut+1) s t, { ( ) ( ) ( = w t i (s t,ξ t )+βω i t+1 (s t )+βe s V i t+1 s t+1 w i t+1 s t+1 + w i t+1 s t+1,zt+1, w t+1)} i dz z ( ) = w t i (s t,ξ t )+βω i t+1 (ξt i )+βe s V t+1 s t+1 where w i t (s t,ξ t )=E z w i t (s t,z t,ξ t ). Defie ( ) ( ) Ṽt i (s t,zt )=wt i s t,zt, w t i + βω i t+1 (s i t )+βe s V t+1 s t+1 gross flows from state s t to s t+1 ca be represeted as: m sts t+1 = E s z ( ) j exp {(Ṽ t+1 s t+1,zt+1 Ṽ i k exp {(Ṽ k t+1 ( s t+1,z t+1 ( t+1 s t+1,zt+1) ( C i,j st+1) ) } /ν ) ( Ṽt+1 i s t+1,zt+1) ( C i,k st+1) ) }dz, /ν if aget stays same age; or m sts t+1 = 1 7 E s z ( ) j exp {(Ṽ t+1 s t+1,zt+1 Ṽ i k exp {(Ṽ k t+1 ( s t+1,z t+1 ( t+1 s t+1,zt+1) ( C i,j st+1) ) } /ν ) ( Ṽt+1 i s t+1,zt+1) ( C i,k st+1) ) }dz, /ν if aget is older at s t.note that s t+1 should have the correct sectoral experiece give s t followig the process i (5). Fially the optio value ca be expressed as Ω(s t )= νe s z log k exp {(Ṽ k t+1 ( s t+1,z t+1 1 ) ( Ṽt+1 i s t+1,zt+1) ( C i,k st+1) ) /ν } dz. ACM preset derivatio of equatios similar to the oes above i detail. The mai differece 32

33 here is havig a additioal o-liear shock, z t, hece we take itegrals over that shock usig simulatios. Appedix C: A Brief Aalysis of the Pew Survey Pew Global Attitudes survey is cosist of iterviews coducted i 44 coutries with 38,263 idividuals i It icludes approximately 100 questios o various popular issues ad persoal backgroud. We are ot usig the survey data to estimate our mai model but to estimate a simple descriptive probit model to illustrate our motivatio for this research. A questio o iteratioal trade poits out a very iterestig yet uexplored issue. The questio is: Ad what about the differet products that are ow available from differet parts of the world - do you thik this is a very good thig, somewhat good, somewhat bad or a very bad thig for our coutry? We fid that as people get older they are less likely to aswer this questio as good ad somehow good. We set up a simple probit model to demostrate the correlatio betwee age ad probability of supportig free trade. Cosider that A = 1 if the idividual, i, aswers the questio as good ad else A = 0. We assume that A = 1 if ad oly if gais from trade u is greater tha a certai threshold ū. We use a simple liear form u i = β 0 + β 1 Age i + β 2 Age 2 i + β 3 Age 3 i + β 4 F emale i + β 5 Employed i + ε i, Where ε is a iid shock, Employed is a dummy for employmet status which is oe for employed people ad zero for uemployed, Female is a dummy variable which is oe for female ad zero for male, ad Age meas age i last birthday mius eightee. Estimates show that probability of a perso s gais from trade beig larger tha the threshold decrease with age (i a liear fashio). See Table 10 - Pael A for the estimates. The percetage of people from differet age groups supportig free trade is reported i Pael B of Table 10. I additio to Pew data, Geeral Social Survey, coducted i US startig from 1972, also has questios related to people s perceptio of free trade. The egative correlatio betwee age ad supportig free trade is also observed i GSS, but the evidece is ot coclusive sice the umber of respodets to these questios i Geeral Social Survey is much smaller compared to Pew data. For example 1348 people gave a valid respose to the questio: Does America beefit from beig a member of NAFTA? the probability of aswerig Yes agai, i geeral, decreases with age. See Table 10 - Pael C. 33

34 Table 10: Pew Statistics. Pael A: Pew Probit Results Coefficiet Std. Error Costat Age Age Age Female Employed Pael B: Probability of Supportig Free Trade i the Pew Sample Age 20 s 30 s 40 s 50 s 60 s 70 s Probability Pael C: Probability of Supportig NAFTA i the GSS sample Age 20 s 30 s 40 s 50 s 60 s 70 s Probability It should be oted that the questios asked i these surveys are very geeral; the aswers give deped o may factors ot observed from the data such as oe s perceptio of free trade, cosumptio prefereces, recet chages i trade policy i their coutries, skill level, worker s idustry, ad may other thigs which would differ from coutry to coutry ad perso to perso. The probit aalysis ad tables oly show that there is, i geeral, a egative correlatio betwee age ad supportig free trade i most coutries for most of the people. I the ext sectio we set up a model to explai why age ad gais from trade are correlated. We will show that welfare effects of trade liberalizatio deped o people s age, educatio ad experiece level as well as the type of liberalizatio. 34

35 Refereces [1] Artuc, E., S. Chaudhuri, J. McLare (2007). Trade Shocks ad Labor Adjustmet: A Structural Empirical Approach. America Ecoomic Review, Forthcomig. [2] Artuc, E., S. Chaudhuri, J. McLare (2008). Delay ad Dyamics i Labor Market Adjustmets: Simulatio Results. Joural of Iteratioal Ecoomics, 75. [3] Bellma, R. (1957). Dyamic Programmig. Priceto NJ, Priceto Uiversity Press. [4] Berard, A., J. Eato, B. Jese ad S. Kortum (2003). Plats ad Productivity i Iteratioal Trade, America Ecoomic Review, 93(4). [5] Borjas, G., S. Rose (1980). Icome prospects ad job mobility for youger me, i Ehreberg, R. (Eds),Research i Labor Ecoomics, JAI Press Ic., Greewich, CT., [6] Camero, S., S. Chaudhuri ad J. McLare (2003). Mobility Costs ad Dyamics of Labor Market Adjustmets to Exteral Shocks: Theory, Natioal Bureau of Ecoomic Research Workig Paper # [7] Davidso, C., ad S. J. Matusz (2001). O Adjustmet Costs. Michiga State Uiversity Workig Paper. [8] Groot, W., M. Verbere (1997). Agig, Job Mobility, ad Compesatio, Oxford Ecoomics Papers, 49. [9] Jacobso, L., R. LaLode ad D. Sulliva (1993). Earig Losses of Displaced Workers, AmericaEcoomic Review, 83. [10] Jovaovic, B., (1979). Job Matchig ad Theory of Turover, Joural of Political Ecoomy, 87. [11] Kambourov, G., ad I. Maovskii (2003). Labor Market Restrictios ad the Sectoral Reallocatio of Workers: The Case of Trade Liberalizatios, Uiversity of Wester Otario workig paper. 35

36 [12] Kambourov, G., ad I. Maovskii (2004). A Cautioary Note o Usig (March) CPS Data to Study Worker Mobility, Mimeo: Uiversity of Toroto. [13] Keae, M., ad K. Wolpi (1997). The Career Decisios of Youg Me. Joural of Political Ecoomy 105. [14] Kea, J., ad J. R. Walker (2003). The E?ect of Expected Icome o Idividual Migratio Decisios, NBER Workig Paper No [15] Kletzer, L. (1989). Returs to Seiority After Permaet Job Loss, America Ecoomic Review, 79. [16] Kletzer, L. G., (2002). Imports, Exports ad Jobs: What Does Trade Mea for Employmet ad Job Kalamazoo,Michiga: W. E. Upjoh Istitute for Employmet Research. [17] Lazear, E. (1979). Why is there Madatory Retiremet?, The Joural of Political Ecoomy, 87. [18] Lee, D. ad K. Wolpi (2004). Itersectoral Labor Mobility ad the Growth of the Service Sector Ecoometrica - forthcomig. [19] Melitz, M. J., (2003). The Impact of Trade o Itra-Idustry Reallocatios ad Aggregate Idustry Productivity, Ecoometrica, vol. 71(6) [20] Na, E.Y. ad J. Duckitt (2003). Value cosesus ad diversity betwee geeratios ad geders, Social Idicators Research, [21] Neal, D. (1999). The Complexity of Job Mobility amog Youg Me, Joural of Labor Ecoomics, 17. [22] Neal, D. (1995). Idustry Specific Huma Capital: Evidece from Displaced Workers, Joural of Labor Ecoomics, 13. [23] Patel, J., C. H. Kapadia, ad D. B. Owe (1976). Hadbook of Statistical Distributios, New York: Marcel Dekker, Ic. [24] Pavcik, N., O. Attaasio ad P. Goldberg (2004). Trade Reforms ad Icome Iequality i Colombia. Joural of Developmet Ecoomics, 74 (August), pp

37 [25] Pissarides, C. A. (1985). Short Ru Equilibrium Dyamics of Uemploymet, Vacacies ad Real Wages, The America Ecoomic Review, 75. [26] Revega, Aa L. (1992). Exportig Jobs?: The Impact of Import Competitio o Employmet ad Wages i U.S. Maufacturig, The Quarterly Joural of Ecoomics 107:1. (February), pp [27] Slaughter, M. J., (1998). Iteratioal Trade ad Labor-Market Outcomes, Ecoomic Joural, 108:450 (September), pp [28] Sulliva, P. (2005). A Dyamic Aalysis of Educatioal Attaimet, Occupatioal Choices, ad Job Search. Mimeo: Uiversity of Michiga [29] Topel, R. (1991). Specific Capital, Mobility, ad Wages: Wages Rise with Job Seiority, The Joural of Political Ecoomy,

38

39

40

41

42

43

44

45

46

47

48