TRADE COSTS, FINANCIAL CONSTRAINTS, AND FIRM PERFORMANCE IN DEVELOPING COUNTRIES

Transcription

1 TRADE COSTS, FINANCIA CONSTRAINTS, AND FIRM PERFORMANCE IN DEVEOPING COUNTRIES Abstract: Ths paper extends on wor done n the heterogenous-frms trade lterature by addressng both heterogenety n trade costs at the frm level as well as the exstence of fnancal constrants. These extensons to the heterogenous-frms models are also appled n the context of a developng country. Utlzng a framewor that endogenzes technologcal choce, the analyss shows that fallng trade costs and mprovng credt marets (or less fnancal constrants) mprove frm performance. Also, frm-level trade costs are shown to mpact a frm s ablty to enter the export maret, mplyng heterogenety n trade costs at the frm level. The current results show nconclusve evdence for the effect of ndustry-level trade costs and fnancal constrants on the ablty to enter the export maret, but future addtons to the robustness of the data n ths worng paper wll address ths ssue. Erc Tseng, the Oho State Unversty Selected Paper prepared for presentaton at the Agrcultural & Appled Economcs Assocaton s 204 AAEA Annual Meetng, Mnneapols, MN, July 27-29, 204. Note: Copyrght 204 by Erc Tseng. All rghts reserved. Readers may mae verbatm copes of ths document for non-commercal purposes by any means, provded ths copyrght notce appears on all such copes. All wor n ths manuscrpt s solely the author s wor, unless referenced, noted, or cted.

2 I. Introducton Internatonal economcs has long been occuped wth the queston of whether nternatonal trade leads to growth and progress. One approach n the lterature has utlzed nternatonal data to assess the mpact of trade on aggregate growth measures such as ncome. Another approach to evaluatng these gans from trade assesses the mpact of trade at the mcroeconomc level. Ths approach typcally nvolves analyzng effects of changes n trade costs on productvty. Wth the heterogenous-frms models set out n Meltz (2003), Bernard et al. (2003), and Yeaple (2005), the emphass on the mpacts of changng trade costs falls on the reallocaton of producton and productvty wthn ndustres. Fallng trade costs reallocate producton wthn ndustres, wheren the most productve frms contnue to produce and begn exportng, whle less productve frms are forced out of the maret, unable to compete. Of the sx maor mplcatons of the model, three nvolve the reallocaton of producton and productvty of frms: fallng trade costs force less productve frms to ext, let more productve frms begn exportng goods, and lead to export growth. These changes n allocaton also mply that frm-level productvty wll ncrease. Prevous studes such as Bernard, Jensen, and Schott (2006) have provded ey evdence supportng the heterogenous-frms model. Gven that ther study analyzed Unted States frms only, numerous emprcal studes have recently been done on frms n developng countres, such as Chle (Blyde and Ibert 202, Alvarez and ópez 2005) and Colomba (Eslava et al. 2009). These studes n developng countres have tested the Meltz (2003) and Bernard et al. (2003) results to varous degrees. In addton, the heterogenous-frms lterature has recently started to loo at the role of fnancal constrants n a frm s export decson-mang process. One such example s Manova (203), who apples fnancal constrants to a Meltz-le model and emprcally tests the 2

3 mplcatons of such a model. The cross-country model shows that fnancal constrants stunt the development of frms, preventng frms from enterng producton, enterng the export maret, and lowerng the sales of exporters. Fnancal constrants are lely to have an even larger effect on developng countres, gven the lac of ther fully developed credt marets. I ntegrate fnancal constrants nto my model to further examne ths effect. Therefore, I extend the Yeaple (2005) model to a developng-countres context to further test the heterogenous frms model. The model I develop follows closely, but t extends on the Yeaple (2005) model n that t accounts for frm heterogenety n transport costs and the presence of borrowng and fnancal constrants. Whle trade costs are typcally consdered as ndustrylevel trade costs, t s possble that frm-level trade costs exst and affect the export decsonmang of these frms. Falng to account for these frm-level trade costs may lead to potental ms-specfcaton and the presence of omtted varable bas n the results. Therefore, I modfy the theoretcal model to account for these frm-level trade costs and fnancal constrants, utlzng Enterprse Surveys data to provde measures of frm-level trade costs and fnancal constrants to test whether they are mportant n a frm s decson-mang. Utlzng the Yeaple (2005) model allows me to emprcally test three separate results: I test whether fallng trade costs ncrease the amount of exporters n the maret, leads to productvty growth, and leads to growth n revenue per worer. The analyss s structured as follows: frst I set out the theoretcal model, extendng on wor done n Yeaple (2005). Next, I provde bacground on the data and the constructon of ey varables for analyss. I then translate three mplcatons of the model nto emprcal specfcatons for testng. After testng the three results from the model, I summarze and ndcate the drecton of possble future wor. 3

4 II. Theoretcal Specfcaton The model lad out n ths secton extends on the model found n Yeaple (2005). I heretofore augment the Yeaple model by addng n frm-level trade costs and fnancal constrants (such as n Manova 203). A. Setup I begn by assumng that consumers have a constant elastcty substtuton ofσ = > α gven log-lnearzed Cobb-Douglas utlty, wrtten as U = ( β ) ln( Y ) + β ln( X ). Addtonally, let there be worers wth mass M, sll (or productvty) level Z, where Z s ncreasng, dstrbuted along G(Z) wth densty g(z) and support [0, ). Goods are produced wth one nput, labor, n marets wth free entry, though frms may bear a fxed cost. Two goods may be produced: a sngle-technology good (Y), or a two-technology composte good (X). Frms, therefore, operate n a monopolstcally compettve maret. Each worer produces φ ( Z ) of a good, where φ ( Z ) s ncreasng and contnuous. Followng Yeaple (2005), I defne { Y,, }, where Y s the numerare, s the lowtechnology X good, and s the hgh-technology X good. astly, assume the followng s true regardng the productvty of worers: φ (0) = φ (0) = φ (0) =, and Y φ ( Z ) φ( Z) φy ( Z) > > > 0 Z φ ( Z) Z φ ( Z) Z φ ( Z ) Y () Note that ths support, [0, ), s theoretcally fnte, though can be a very large value. For the purposes of the theoretcal porton, ths large theoretcal upper lmt s represented as. 4

5 Equaton () mposes sll-based comparatve advantage for worers n the economy. Those worers wth hgh Z values (.e. the most slled) have a comparatve advantage producng n X good wth the more productve technology, those wth lower productvty Z values have a comparatve advantage producng X wth less productve technology, whle those wth the lowest Z values (.e. the least slled) have a comparatve advantage n producng Y. Therefore, let Z be the hghest productvty assocated wth the producton of good Y, such that all worers wth sll less than or equal to Z produce good Y. et Z 2 be the lowest productvty assocated wth the producton of X wth hgh-technology, such that all worers wth productvty hgher than Z 2 produce hgh-technology X. Then, worers wth productvty n between Z and Z 2 produce the low-technology X. B. Closed Economy Equlbrum Frst, note that the revenue of any frm n the X ndustry s R = ( β EP ) p. The unt σ σ X cost of a partcular Z can be wrtten as defned as: C = W ( Z) φ ( Z), for all { Y,, }. Wages can thus be CYφY ( Z) 0 Z Z W ( Z) = Cφ ( Z) Z Z Z2. Cφ ( Z) Z Z2 (2) Wages, then, smply reflect the unt cost of producton and the sll of the worer. As long as worers wth sll Z are ndfferent between producng Y and, and worers wth sll Z are ndfferent between producng and, normalzng MR = P = mples that 2 Y Y 5

6 C C Y = φ ( Z ) φ ( Z ) Y = < (3) C φ ( Z ) φ ( Z ) = Y 2 C φ ( Z) φ ( Z2) <. Equatons (2) and (3) mply a partcular wage structure, wheren the wages move accordng to productvty but are segmented accordng to the good the worer produces. Ergo, worers producng X are pad more than those producng Y, and those n X producng at earn more than those producng at. To produce n X, frms must pay up-front fxed costs, such that F > F. Therefore, to produce n X, frms are requred to borrow to enter. Up front, frms are only requred to cover ( d ) of ther fxed costs, meanng that frms must borrow d (0,) up a collateral porton h (0,), earnng them a loan amount Φ ( Z ). In borrowng, frms put, whch s bounded by Φ( Z ) R C ( x + ( d ) F ). The Φ( Z ) constrant mples that frms can only borrow up to ther net revenue. λ addtonally represents the probablty that the contract s enforced. Therefore, the typcal frm faces ths proft maxmzaton problem: σ = ( EPX ) p C ( x + ( d ) F ) Φ( Z) ( ) hc F π β λ λ (4) s.t. Φ β + σ ( Z ) ( EPX ) p C ( x ( d ) F ) (4.) Free entry mples zero profts, whch mples that R = C ( x + ( d ) F ) + λ Φ ( Z ) + ( λ ) h C F (5) For tractablty, I assume that Φ ( Z ) = d C F. Ths assumpton s essentally one of perfect borrowng, n that frms borrow up to as much as they need to enter the maret. Ths 6

7 assumpton allows me to easly pn down the equlbrum. Note that wth the assumpton of Φ ( Z ) = d C F, σ R C δ C F = = R C δ C F, where δ = ( λ )( h d ). Utlzng δ = ( λ )( h d ), δ (0, 2) provdes a relatvely ntutve scale of fnancal constrants. Essentally, when δ (0,), such that h d > 0, a frm that proportonally borrows less than what they can provde as collateral wll fnd t easer to enter the maret. If δ (, 2), such that h d < 0, a frm that s requred to proportonally borrow more than what they can provde as collateral wll fnd t more dffcult to enter the maret. If h d = 0, thenδ =, whch reduces the case down to the Yeaple (2005) model. In the context of the developng world, the case where h d < 0 and therefore δ (, 2) s the most lely scenaro. By the mposed assumpton of comparatve advantage n () and some smplfcaton, the rato of unt costs for hgh and low technology can be expressed as: C φ ( Z2) δ F = = C φ ( Z2) δ F σ (6) Equaton (6) pns down the threshold Z2 n equlbrum. Notce that t s pnned down solely by the frm s problem (and thus the zero-proft condton). The threshold s therefore unaffected by a multtude of country attrbutes, such as the quantty of worers n the populaton and the dstrbuton of productvty Z. Note that ths threshold Z2 ncreases when δ or ncrease, pushng frms out of the technology. Conversely, f δ or F decrease, more frms wll begn producng wth the technology. F 7

8 Next, by maret clearng, Y = ( β ) E = ( β ) MW, where W s the average wage. Average wage can therefore be wrtten as: Z Z2 φ 0 Y φ Z φ Z2 W = ( Z ) dg ( Z ) + C ( Z ) dg ( Z ) + C ( Z ) dg ( Z ) (7) I more explctly defne the thresholds of Z and Z 2 (that separate Y and, and producton respectvely) as S( Z ) φ ( Z ) φ ( Z ) C ( ) = ( ) C Y 2 = C and A Z2 φ ( Z) φ Z2 (8) Recall, furthermore, thaty = ( β ) E = ( β ) MW = M φ ( Z) dg( Z) from the maret clearng condton n Y. Therefore, the second equlbrum condton s 0 Z Y β ( ) Z Z2 φ 0 Y = φ + Z 2 φ β S Z Z2 ( Z ) dg ( Z ) ( Z ) dg ( Z ) A ( Z ) ( Z ) dg ( Z ) (9) Equatons (6) and (9) defne the closed economy equlbrum. Note that nether of these equlbrum condtons are dependent on M, so the amount of worers n the economy does not nfluence the separaton of producton nto the three types. In addton, the dstrbuton of sll gven by g( Z ) affects the dstrbuton of productvtes Z, and wll therefore mpact Z gven Z 2 (as pnned down by (6)), but also the mappng of worer sll onto the wage functon W ( Z ). Totally dfferentatng equaton (9) shows that dzdz 2 < 0. Ths mples that the ndvdual thresholds move n opposte drectons. Changes that ncrease Z wll decrease Z 2, and vce versa. In addton, changes n the level of fnancal constrants wll affect the thresholds of Z and Z 2 through equaton (6), whch pns down Z 2, whch pns down Z n equaton (9). 8

9 C. Open Economy Equlbrum Recall that frms must pay fxed costs for producton n good X, where {, }, F for low-technology X and F for hgh-technology X. Now, frms can now enter the export maret, but the decson to export also carres an addtonal fxed cost F. Frms addtonally face X transport costs τ and γ n ther decson, where τ represents ndustry-level transport costs and γ represents frm-level transport costs. Wth the ntroducton of γ, my wor further departs from the Yeaple (2005) model. In the maorty of the lterature, transport costs are exogenous and occur at the ndustry level, especally n emprcal analyses. The addton of γ accounts for frm heterogenety n trade costs: whle all frms wll face dentcal ndustry-level trade costs, gven a host of factors (ncludng but not lmted to nfrastructure, government barrers, or technologcal neffcences at the frm level), a frm s true transport costs may not be accurately captured by τ alone, potentally leadng to msspecfcaton and ms-measurement. Therefore, I assume that τ > andγ 0. Ergo, revenue for exporters can be wrtten as σ R [ ( τ γ ) ] + + when servng both the local maret and foregn maret. Ths mples that addtonal revenue from exportng, R x, can be wrtten as R σ x RD / [( τ γ ) ] = +. Therefore, frms costs wll be C F for servng the local maret, and C ( F + F ) when servng the local maret and the foregn maret. I mpose the assumpton that x F > F ( τ + γ ) > F σ x (0) 9

10 Ths separaton condton ensures that there wll not only be producers of and, but also that some frms wll export and some wll not 2. Wth ths separaton condton, all producers wll choose to export, whle all producers wll not export. Equatons (2), (3), and (0) mply that exporters have a low unt cost and a hgh fxed cost, pay a hgher wage, and have a hgher value of output per worer. Ths condton mples that the rato of revenue between and technologes can now be wrtten as: R σ ( + ( τ + γ ) ) C R σ = + + C C ( δ F + FX ) = δ C σ ( ( τ γ ) ) () Of partcular nterest are the breaponts Z and Z 2, where producton shfts from Y to, and from to. e before, I defne A( Z 2) and S( Z) thresholds of Z and Z 2. These are as the functons that explctly model the C φ( Z2) δ F + F X = A( Z2) = σ C φ ( Z2) δ F ( + ( τ + γ ) ) σ φy ( Z) and S( Z) = = C φ ( Z ) (2) Equaton (9) and A( Z2) from equaton (2) defne the equlbrum for the open economy. The same prncples from the closed economy equlbrum apply here n the open economy equlbrum. The equlbrum s lewse ndependent of the mass of worers M. Addtonally, gven Z 2, Z s mpacted by potental changes n the dstrbuton of worer sll G( Z ). The mpact of fnancal constrants n the open economy equlbrum s also dentcal to the case presented n the closed economy equlbrum. 2 Note that f F ( τ + γ ) > F > F, then no frms export, whle f F F F σ x σ > > x ( τ + γ ), all frms export. These cases do not yeld nterestng economc problems, and therefore are not consdered n my analyss. 0

11 By acnowledgng that producton by frms s M Z2 φ ( ) ( ) Z dg Z, and by the zeroproft condton, t s possble to derve the number of frms producng the good. Ths can be wrtten as: N M = σ ( δ F + F ) X Z2 φ ( Z) dg( Z) (6a) Smlarly, the frms producng X wth technology can be expressed as N M Z2 = φ ( Z) dg( Z) σδ F. Z (6b) Therefore, the number of frms producng n X s N X = N + N. Therefore, gven Equatons (9) and A( Z2) n (2), t s possble to derve the effects of decreases n trade costs on the number of frms producng and (thus also dervng the change n the number of frms exportng as mpled by the condton n (4) ). These can be wrtten as: dz dτ < 0 dz dγ < 0 dz dγ > 0 dz dγ > (7) The comparatve statcs from (7) combned wth (6) yeld the man re-allocatve effect of fallng trade costs: fallng trade costs wll ncrease N. Ths sortng nto hgher producton mples ncreased productvty and revenue per worer of frms n X. These effects are drven by the ablty of frms to adopt hgher and more productve technologes wth fallng trade costs. Frms that are able to adopt technology enoy R > R, thus mplyng growth n revenue per worer. Producng wth technology addtonally mples φ > φ, such that frms become more productve. These effects are summarzed below: PROPOSITION : Wth fallng ndustry-level or frm-level trade costs, more frms enter the export maret. Ths selecton nto the export maret yelds productvty growth and growth n revenue per worer.

12 Addtonally, the model shows the effect of fnancal constrants. Gven that decreases, the overall fxed costs that frms must overcome to enter the export maret (and, moreover, enter the maret overall) decrease, thus mang t easer for frms to enter the maret and so N decreases (and vce versa). Smlar to the effect of fallng trade costs, ths ncreased ablty to enter the maret due to less fnancal constrants yelds a sortng effect that leads to δ productvty growth and growth n revenue per worer ( φ > φ and R R >, respectvely). The effect of fnancal constrants can be summarzed thusly: PROPOSITION 2: Wth less fnancal constrants (such as better credt access or more complete credt marets), frms are more lely to enter the export maret. Ths selecton nto the export maret yelds hgher revenues and productvty for frms. Conversely, tghter fnancal constrants means that frms are less lely to enter the export maret, whch mples negatve productvty growth and negatve growth n revenue per worer. The two propostons form the foundaton of the emprcal analyss n the followng sectons. Decreases n ndustry-level trade costs ( τ ) and frm-level trade costs ( γ ) should ncrease the amount of frms that export, yeld productvty growth and growth n revenue per worer. Improvng credt constrants or credt marets (through fallng δ ) should mae t easer for frms to enter the maret as exporters (wth qualty ), lead to productvty growth, and lead to growth n revenue per worer 3. III. Data Frm-level data for Argentnan manufacturers comes from Enterprse Surveys, a data set compled by the World Ban 4. The data from the Enterprse Surveys samples the entrety of an economy s prvate sector, wth data coverng aspects of busness operatons and the busness 3 These results can also be construed as movements n average productvty and average revenue per worer n ths framewor. The theoretcs motvatng ths can be found n Appendx A. 4 Enterprse Surveys can be found at 2

13 envronment such as nfrastructure, n addton to the typcal measures of frm performance (aggregated at the equvalent of the 2-dgt ISIC level 5 ). The analyss I conduct utlzes the typcal measures of frm performance, n addton to a set of busness envronment ndcators that form the bass of my barrers to trade measure. The data set s nomnally avalable n panel form (consstng of two cross-sectons contanng years 2006 and 200). Gven the lac of a more robust sample, I reduce the panel nto a cross-secton for the year 2006 to obtan measures n tme t and t+. The frm-level data tself serves as the bass for the constructon of the ey varables: the quantty exported, total factor productvty, and estmates of the frm s producton. Total factor productvty n partcular s more smply constructed as a measure of the total value of outputs over the total value of labor nputs for a gven frm. Ths devates from Bernard, Jensen and Schott (2006) and some followng wor (.e. Blyde & Ibert 202), who use more complcated superlatve productvty ndces n the ven of Caves, Chrstensen, and Dwert (982) and Good et al. (997). Gven the data constrants presented by utlzng the Enterprse Survey, I use a smplfed TFP measure nstead. The measure of TFP s essentally a measure for labor productvty, constructed n ths way to more closely algn wth the theoretcal model, wheren labor s the only nput: TFP value( outputs) value( outputs) = = value( nputs) value( labor ) Freght cost data s provded by Argentna s Natonal Insttute of Statstcs and Censuses (INDEC). Freght costs pose sgnfcant costs of trade for frms, and cannot be dsmssed from analyss as was common n the pror lterature. Ths s true especally for developng countres, 5 Aggregated at the 2-dgt ISIC level, the frm-level data provded by Enterprse Surveys the lmtng factor n my analyss. The resultng measures I construct are therefore aggregated at the 2-dgt ISIC level. 3

14 where burgeonng economes may face ndustry-level costs of entry nto the export maret. I closely follow the lterature and construct ad valorem freght rates as the value of freghts and nsurance over the value of mports 6. Addtonally, ad valorem tarff rates are obtaned from the TRAINS database, a part of the World Integrated Trade Soluton developed by the World Ban. Both ad valorem freght rates and tarff rates are aggregated at the 2-dgt ISIC level. The measure of frm-specfc costs of trade that I construct s contaned n the Enterprse Survey data. In the trade and regonal lterature, costs of trade have been represented as ceberg costs: when tradng a good, a certan quantty s lost over the course of delvery. I follow that ven of the lterature n specfyng my frm-level costs of trade varable. As part of the Enterprse Survey, frms were ased to estmate the percentage of goods n transt lost due to varous factors such as breaage, spolage and theft. These losses, at the frm level, represent costs of trade at a level that s not captured va the ndustry-level freght rates and tarff rates. Whle these frm-level costs of trade may not exst or be relevant for a developed country, t s plausble that these costs of trade are relevant and problematc n developng countres. Ths varable s the emprcal representaton of γ found n the theoretcal model, allowng for the testng of ts sgnfcance. Therefore, I nclude ths measure n my emprcal specfcatons, to test whether frmlevel costs of trade affect export decson-mang choces by frms, n conuncton wth ndustrylevel trade costs. owever, t s possble that the raw costs-of-trade measure does not effectvely capture the frm-level heterogenety. I augment the frm-level costs-of-trade varable, nstrumentng t wth a seres of multnomal choce varables measurng dffculty of trade that are also present n the data set. Ths nstrumented costs-of-trade s specfcally used n the testng 6 Ths constructon of the ad valorem freght rate follows Blyde and Ibert (202) and Bernard, Jensen, and Schott (

15 whether fallng trade costs nduce frms nto exportng, gven that ceberg costs may not completely capture the entry decson for frms 7. I capture the effect of fnancal constrants through the frm s response to the followng queston n the Enterprse Surveys: Is access to fnance, whch ncludes avalablty and cost, nterest rates, fees and collateral requrements, No Obstacle, a Mnor Obstacle, a Moderate Obstacle, a Maor Obstacle, or a Very Severe Obstacle to the current operatons of ths establshment? (200). The response to ths queston serves as a rough estmate of the extent of fnancal constrants for use n my emprcal analyss. owever, gven how mprecse ths measure s and potental endogenety ssues (such as smultanety when consderng effects of trade costs on productvty and revenue), I choose to nstrument ths measure of fnancal constrants va a set of nstruments descrbng where the frm borrows from (such as bans, lenders, for example). Table shows the breadown of trade costs by ndustry. Freght rates and tarff rates both declne overall n between 2002 and Freght rates, n general, do not declne by nearly as much as tarff rates, where tarff rates reduce by at least 50% for the ndustres n the sample. The largest tarff rate reducton occurs n the food ndustry, where tarffs are reduced by a staggerng 65.5%, whle machnery and textle ndustres also receve large reductons n tarffs of around 60%. Decreases n freght rates typcally are by about one percentage pont or less n comparson such as n textles, where the rate falls from 5.08% down to 4.52%, a small reducton n freght costs n comparson to the large tarff rate reductons. Addtonally, frm-level trade 7 More specfcally, the decson-mang rule for the frm to enter the export maret may not be the same as the decson-mang rule for quantty exported (.e. the ntensve margn). The multnomal choce varables used as nstruments as whether transport or customs and trade regulatons pose problems for the frm. 5

16 Table - Average Rates by Industry and Year, % Industry Freght Rate Freght Rate Tarff Rate Tarff Rate Frm-evel Costs Other Manufacturng Food Textles Garments Machnery & Equpment Note: Both freght and tarff rates are calculated ad valorem. Frm-level costs are the estmated percentage of goods lost n transport. costs are small effects, wheren the maxmum average frm-level trade cost s found n the food ndustry, wth a 0.737% estmated frm-level trade cost. Argentna was a developng country wth some developed export marets n ths tme frame (and stll s developng today): t s plausble that the small frm-level trade costs are a reflecton of more developed export marets. owever, gven the combnaton of freght and tarff costs, t s plausble that even these small frm-level trade costs may nfluence frm export decson-mang and other measures of frm performance. Therefore, t s mportant to test whether these frm-level trade costs matter, snce a small absolute sze of frm-level trade costs does not necessarly mply that the frm-trade costs do not matter to frms on the margn. IV. Emprcal Specfcaton, Results The emprcal analyss heren loosely follows the structure found n Bernard, Jensen, and Schott (2006), for example. I test the three results derved from the theoretcal model separately utlzng smple regressons to examne whether the comparatve statcs hold. Frst, I test whether fallng trade costs ncrease the chances of a frm enterng the export maret. Next, I test whether fallng trade costs yeld productvty growth. ast, I test whether fallng trade costs lead to growth n revenue per worer. Testng these specfcatons also lets me test the effect of 6

17 mprovng fnancal constrants n each scenaro: mprovng fnancal constrants should mprove the chances of a frm enterng the export maret, lead to productvty growth and growth n revenue per worer. The results are all condensed nto Table 2 for vewng. A. Entry nto the Export Maret To test whether fallng trade costs cause more frms to enter the export maret (.e. ncreasng N ), I follow the approach n Roberts and Tybout (997) and, smlarly, n Bernard et al. (2003). These approaches rely on the observaton that frms who export must overcome fxed costs to enter the export maret. Ths observaton s expressed n the theoretcal model as equaton (4), F > F ( τ + γ ) > F. Therefore, ths export condton s formulated nto a probt σ x equaton that tests whether trade costs affect frm export decsons: Pr( Export = Export = 0) = α τ + β X + η( τ X) + θγ + ε,t +, t + (8) Pr( Export = Export = 0) s the probablty that a frm chooses to begn exportng n,t +, t+ the perod t + (200) gven that n tme t (2006), they were not exportng. τ represents the vector of ndustry-level trade costs n between 2002 and 2006, ncludng freght rates and tarff rates, X s the vector of frm attrbutes, τ X s the vector of ndustry-level trade costs nteracted wth the vector of frm attrbutes, and γ s the frm-level trade costs. The results (found n column []) show that ndustry-level trade costs do not affect frm entry nto the export maret as suggested by the theoretcal wor n the paper. These results currently 7

18 Table 2 Summarzed Results, log(employment) Dependent Varable: Entry nto the Export Maret log(tfp) log Ω [] [2] [3] Freght Rates ( FC) *** *** (4.875) (2.34) (2.552) Tarff Rates ( TC) 2.37** -.95*** -2.23*** (.68) (0.65) (0.697) FC TFP (0.076) ( ) TC TFP ** ( ) ( ) Total Factor Productvty (TFP) ** (0.048) ( ) log(employment) *** 0.288** (0.93) (0.2) (0.3) log(abor Cost) * 0.708*** *** (0.57) (0.0928) (0.) K/ Rato 0.245** (0.25) ( ) (0.0779) Frm-evel Trade Costs -.262** -.259*** -.003** (0.603) (0.404) (0.429) Fnancal Constrants ** ** (0.589) (0.36) (0.346) N = 242 N = 22 N = 222 W = F = F =.20 R 2 =0.6 R 2 = R 2 = Note: Industry dummes are ncluded n all specfcatons. Note: Dependent varable represents ether: the chance a frm enters the export maret n between 2006 and 200, percent change n productvty between 2006 and 200, and the percent change n revenue per worer between 2006 and 200 gven yearly changes n trade costs from 2002 to The frm-level data s from the Enterprse Survey, whle the ndustry level trade costs are from the Natonal Insttute of Statstcs and Censuses (INDEC). Note: Frm-level costs of trade s nstrumented here by a seres of multnomal choce varables descrbng dffculty of trade, where F = 2.03, p = 0.356, and R 2 = n the frst stage. Fnancal constrants are also nstrumented by the sources of the frm s borrowng, where F =2.7, p = , and R 2 = Note: *, **, *** Sgnfcant at the 0%, 5%, and % level, respectvely. 8

19 llustrate that fallng ndustry-level trade costs (freght and tarff rates both) mae t more dffcult for frms to enter the export maret. The coeffcent estmate on the effect of freght rates s nsgnfcant whle the coeffcent estmate on tarff rates s strongly sgnfcant. It s currently unclear what drves these results, and further emprcal testng, prmarly wth a larger data set, would help shed lght onto these results. owever, frm-level trade costs have a strongly sgnfcant effect on entry nto the export maret. Fallng frm-level trade costs ncrease the chance that a frm wll enter the export maret. A decrease n frm-level trade costs of one standard devaton corresponds to a 9.68% ncreased chance that a frm wll begn exportng n 200 gven that they dd not n The presence of these frm-level trade costs mples that despte the low frm-level trade costs n absolute terms, the margnal mportance of these frm-level trade costs affects a frm s export decson-mang. Therefore, the results ndcate that potental frm heterogenety n trade costs should not necessarly be gnored when examnng frm behavor. Addtonally, t s nherently plausble that addng n these frm-level trade costs has soaed up some of the effect provded prevously by ndustry-level trade costs. Fnancal constrants do not play a sgnfcant role n the frm s export entry decson, accordng to the results. The coeffcent has the expected sgn, but the sgnfcance s lacng, meanng that ther effect on export entry s ndscernble. It s not altogether clear as to why ths s the case. Potentally, the ndustres contaned n the sample (as seen n the summary statstcs n Table ) may not be partcularly susceptble to fnancal constrants, gven that the sample ncludes ndustres such as food and garments. Industres such as technology, steel manufacturng are more lely to have hefty startup costs that would requre loans (especally n developng countres), so the absence of these manufacturng-heavy ndustres n the sample may 9

20 explan the lac of statstcal sgnfcance n the fnancal constrants ndcator. Adequately addressng ths emprcal result would requre further wor n the area. B. Frm Productvty To test whether fallng trade costs and mprovng credt marets (or less fnancal constrants) leads to productvty growth, I utlze a smple regresson: log( TFP) = ατ + β X + θγ + ε (9) log( TFP) s the percent change n the level of productvty of a frm between tme t + (200) and tme t (2006). τ represents the vector of ndustry-level trade costs, ncludng freght rates and tarff rates n between 2002 and X s the vector of frm attrbutes and controls, and γ s the frm-level trade costs. Results can be found n column [2]. I fnd that ndustry-level trade costs have a strongly sgnfcant effect on a frm s productvty. Fallng ndustry-level trade costs (both freght costs and tarff costs) nduce productvty growth n frms. Addtonally, I fnd that frm-level trade costs have a smlar effect on productvty, n that reductons n frm-level trade costs correspond wth ncreases frms experencng productvty growth between 2006 and 200. Overall, t appears that whle both are sgnfcant n affectng the productvty of the average frm, ndustrylevel trade costs have a larger mpact on productvty than frm-level trade costs. Ths mples that whle frm-level trade costs may be smaller than ndustry-level trade costs, they do affect frm productvty and thus should not necessarly be gnored, especally n the developngcountry context. The effect of trade costs on productvty growth (as estmated here) concdes wth the theoretcal model and the extant lterature. 20

21 Fnancal constrants do sgnfcantly nfluence frm productvty. As fnancal constrants lessen and credt marets mprove and become more complete, frms should expect to see productvty growth. Ths s n accordance wth the theoretcal model, whch states that frms who face more strngent fnancal constrants are less lely to be productve n comparson to frms who face more complete credt marets. Intutvely, more complete fnancal marets should allow more frms to partcpate n the maret and let more frms partcpate at hgher and more advanced technologes, thus mprovng productvty. C. Revenue per Worer of the Frm To test whether fallng trade costs and mprovng credt marets (or less fnancal constrants) leads to productvty growth, I utlze the followng regresson: log Ω = α τ + β X + η( τ X ) + θγ + ε (20) log Ω s the percent change n level of revenue per worer of a frm between tme t + (200) and tme t (2006). τ represents the vector of ndustry-level trade costs, ncludng freght rates and tarff rates n between 2002 and 2006, X s the vector of frm attrbutes, τ X s the vector of ndustry-level trade costs nteracted wth the vector of frm attrbutes, and γ s the frm-level trade costs. The results from testng ths porton of the model are found n column [3]. These results show that ndustry-level trade costs are strongly sgnfcant and have the correct sgn: fallng trade costs (both freght and tarff costs) ncrease revenue per worer, n lne wth both the extant lterature and the model derved n ths paper. Frm-level trade costs are lewse strongly sgnfcant. As frm-level trade costs fall, frms mprove ther revenue per worer, 2

22 followng closely to the theoretcal model derved n ths paper. The sgnfcance of the estmates of frm-level trade costs n ths specfcaton mean that they are lely mportant enough to not be gnored, especally n the developng-country context where these frm-level trade costs may be hgh, or where the frm-level trade costs have a hgh margnal effect. Fnancal constrants lewse affect revenue per worer, and the estmate n ths specfcaton s sgnfcant at the 5% level. Frms who face more strngent fnancal constrants wll have lower revenue per worer than those frms that face more complete credt marets, followng the theoretcal model set out n the paper. Thus, more complete marets wll allow more frms the ablty to partcpate n the maret, and partcpate wth hgher technologes, whch wll correspond to hgher revenue per worer. V. Summary, Dscusson Ths paper sets out a model that extends Yeaple s (2005) verson of the heterogenousfrms model. The model restrcts tself to a two-good world, wth one numerare good and a composte good. By mposng sll-based comparatve advantage onto the model, frms select a technology, ether to produce the numerare good, the low-technology X, or the hgh-technology X. Frms face fxed costs, not only to produce the good, but can also pay a fxed cost to export. In addton, exportng ncurs both ndustry-level trade costs and frm-level trade costs that frms must also pay to export. Frm-level trade costs account for potental heterogenety n trade costs, and ther ncluson s one contrbuton of ths paper. Ths leads to frms separatng nto the producton of Y, X wth low technology, and X wth hgh technology, where only the frms who produce X wth hgh technology export. The equlbrum condtons allow for the dervaton of a few comparatve statcs, showng that fallng trade costs nduce more frms nto exportng, ncrease productvty and revenue per worer. 22

23 The man contrbuton of ths paper s the ncorporaton of fnancal constrants nto the model (as smlarly seen n Manova 203). Fnancal constrants may play a crucal role n developng countres, gven that ncomplete and poorly functonng marets may prevent frms from not only begnnng to export, but may prevent frms from enterng the maret altogether. Therefore, I fully ntegrate the presence of fnancal constrants nto the Yeaple (2005) framewor to show that fnancal constrants mae t tougher for frms to enter the maret, and that they lmt productvty growth and growth n revenue per worer. The paper tests out the three comparatve statcs relatng to export entry, productvty growth, and revenue per worer n an emprcal settng. Utlzng the Enterprse Surveys n addton to trade cost data from Argentna s Natonal Insttute of Statstcs and Censuses (INDEC) and the World Integrated Trade Solutons (WITS), I construct a frm-level crosssecton for the year of Testng the three comparatve statcs provdes evdence that generally supports the model derved n ths paper. In comparson to the extant lterature, I fnd agreeng evdence that fallng trade costs lead to productvty growth and growth n revenue per worer, but I fnd contrary evdenced n the effect of ndustry-level trade costs on export entry. Ths may be due to the ncluson of frm-level trade costs, whch may be soang up some of the effect. Frm-level trade costs are found to sgnfcantly affect frm performance and frm export decson-mang. Across all specfcatons, the results show that fallng frm-level trade costs wll yeld more entrants nto the export maret, productvty growth, and growth n revenue per worer. Whle the estmated effect s not as large n magntude as changes n ndustry-level trade costs mght, a one standard-devaton change n frm-level trade costs wll ncrease the chances of a frm enterng the export maret by approxmately 9.68%, a sgnfcantly large ncrease n 23

24 probablty. Therefore, t s unlely that these frm-level trade costs should be gnored n future analyss gven that ther evdent mpact frm performance and a frm s ablty to enter the export maret. These frm-level trade costs are also more lely to be relevant n the developng-country context, where trade costs at the frm level are more lely to be prevalent. Fnancal constrants were shown to be partally supported n the emprcal wor. Fnancal constrants were found to not affect entry nto the export maret. owever, the lac of sgnfcance may be due to the ndustres used n ths sample: ndustres such as food and garments may not be as vulnerable to ncomplete credt marets as other ndustres may be, such as electroncs or heavy manufacturng. Therefore, n ths context, further testng should occur. Incomplete credt marets were shown to hnder productvty growth and revenue per worer, as was shown n the theoretcal model. Future wor could occur on both the theoretcal front and the emprcal front. On the theoretcal front, one frst extenson could be relaxng the perfect borrowng assumpton I mae. Ths assumpton has ntutve tracton and greatly smplfes the model, but t hnges on the zero-proft condton. A logcal frst extenson would be relax the perfect borrowng assumpton and to wor wth a broader borrowng functon Φ( Z) and test whether or not the results derved n ths paper are changed wth a more loosely defned borrowng functon. On the emprcal front, future wor could nvolve testng out the comparatve statcs on a larger, more robust sample. The emprcs generally support the theoretcal model, but do not completely. Addng n more countres would mprove the sample and provde an opportunty to see f some of the nconsstences n the results are due to the sample used n ths paper. Addtonally, the nstrument for frm-level trade costs s currently weaer than what would be preferred for 24

25 estmaton. Utlzng a better nstrument or an expanded data set would help strengthen the constructed measure of frm-level trade costs. 25

26 Appendx A Movements n Average Productvty and Average Revenue Per Worer Gven that Y = ( β ) E, aggregate revenue per worer n Y s gven as Z Ω Y = φ ( ) ( ) ( ) Z dg Z 0 Y G Z. Snce revenue n X s Z2 [ φ( ) ( ) + φ ( ) ( )] Z Z2, M C Z dg Z C Z dg Z also utlzng the fact that employment n X s gven by M[ G( Z)] maes t possble to also derve the aggregate revenue per worer n ndustry X, whch s gven as: β Z Ω X = φ ( ) ( ) ( 0 Y β G Z) Z dg Z Ths mples that the revenue per worer per frm n X s smply Ω Ω X X Ω = = N + N N X and Thus, fallng trade costs ncreases average revenue per worer, Ω, snce Ω X ncreases as τ and/or γ decrease. For average productvty, let Z be the most N X decreases productve Z n the maret, such that Z > Z. Therefore, the range of productvtes n the X 2 maret can be represented as the nterval Z, Z. Then, let the average productvty be Z. Gven dzdτ < 0 and dzdγ < 0 from (), then τ < 0 and γ < 0 should mply that Z > 0. Therefore, Z > 0 mples that Z > 0. The productvty of the average frm should therefore ncrease wth fallng trade costs. The emprcal wor presented n the paper supports the results of the theoretcal model when constructed n ths manner. Movements n average productvty and average revenue per worer are also tested n the specfcatons presented n secton IV. The results presented n the secton lend credence to the theoretcs lad out n ths Appendx. 26

27 References Alvarez, Roberto and Rcardo ópez Exportng and Performance: Evdence from Chlean Plants. Canadan Journal of Economcs 38: Anderson, James E. and Erc van Wncoop Trade Costs. Journal of Economc terature 42:69-75 Bernard, Andrew B., Jonathan Eaton, J. Bradford Jensen and Samuel Kortum Plants and Productvty n Internatonal Trade. Amercan Economc Revew 93: Bernard, Andrew, J. Bradford Jensen and Peter Schott Trade Costs, Frms and Productvty. Journal of Monetary Economcs 53: Bernard, Andrew B., J. Bradford Jensen, Stephen J. Reddng and Peter K. Schott. 20. The Emprcs of Frm eterogenety and Internatonal Trade. Worng Paper 7627, NBER. Blyde, Juan and Gonzalo Ibert Trade Costs, Resource Reallocaton and Productvty n Developng Countres. Revew of Internatonal Economcs 20: Caves, D.,. Chrstensen and W. Dewert Output, nput and productvty usng superlatve ndex numbers. Economc Journal 92: Dxt, Avnash K. and Joseph E. Stgltz Monopolstc Competton and Optmum Product Dversty. Amercan Economc Revew 67: Eslava, Marcela, John altwanger, Adrana Kugler and Maurce Kugler Trade Reforms and Maret Selecton: Evdence from Manufacturng Plants n Colomba. Worng Paper 4935, NBER. 27

28 Good, Davd, M. Ishaq Nadr and Robn C. Scles Index Number and Factor Demand Approaches to the Estmaton of Productvty. In ashem Pesaran and Peter Schmdt eds., andboo of Appled Econonometrcs. Vol. 2: Mcroeconometrcs, ondon: Basl Blacwell, pp elpman, Elhanan Trade, FDI, and the Organzaton of Frms. Journal of Economc terature 44: elpman, Elhanan, Mar J. Meltz and Stephen R. Yeaple Export versus FDI wth eterogenous Frms. The Amercan Economc Revew 94: Manova, Kalna Credt Constrants, eterogenous Frms, and Internatonal Trade. Revew of Economc Studes 80: Meltz, Mar J The Impact of Trade on Intra-Industry Reallocatons and Aggregate Industry Productvty. Econometrca 7: Meltz, Mar J. and Danel Trefler Gans from Trade when Frms Matter. Journal of Economc Perspectves 26:9-8. Roberts, Mar J. and James R. Tybout The Decson to Export n Colomba: An Emprcal Model of Entry wth Sun Costs. The Amercan Economc Revew 87: Tybout, James R. and M. Danel Westbroo Trade beralzaton and the Dmensons of Effcency Change n Mexcan Manufacturng Industres. Journal of Internatonal Economcs 39: Tybout, James R Manufacturng Frms n Developng Countres: ow Well Do They Do, and Why? Journal of Economc terature 38:

29 Yeaple, Stephen R A smple model of frm heterogenety, nternatonal trade, and wages. Journal of Internatonal Economcs 65: