FDI and Export Participation of Local Firms in the Kenyan Garment Industry: Why did not incumbent firms start exports? *

Transcription

1 FDI and Export Partcpaton of Local Frms n the Kenyan Garment Industry: Why dd not ncumbent frms start exports? * Takahro Fukunsh Insttute of Developng Economes Abstract FDI n the garment sector has been the sngle case of large-scale manufacturng nvestment n Afrcan low-ncome countres snce the 990s. Whle FDI has trggered the development of local ndustres n many developng countres through facltatng export to the markets n OECD countres, t has not yet been realzed n Afrca. Moreover, unlke predctons of a standard model of export partcpaton, newly establshed frms wth no equpment and lttle experence responded more postvely. Ths paper nvestgates the background of local frms behavor n Kenya usng frm-level data n Kenya and Bangladesh, where the latter represents exporters. It shows that credt constrant s a prmary source of nactve partcpaton n export opportunty. Only frms whch afford addtonal producton facltes wthout sacrfcng stable domestc supply may be motvated to start exportng. However, n comparson wth Bangladesh exporters, those frms were not as motvated as them due to the large gap n expected profts. Key words: FDI spllover, Manufacturng, sub-saharan Afrca JEL Classfcaton: F, L67, O * The author would lke to thank the frms for ther understandng and cooperaton wth regards to our survey and ntervews. I am also grateful for the research assstance provded by Isabel Munand and Paul Kamau, and comments to the draft verson by Machko Nssanke and the partcpants of the Clmbng up the Global Value Chans project and the nternal semnar at Insttute of Developng Economes.

2 . Introducton Manufacturng sector has been stagnated n most of sub-saharan Afrcan countres, and n partcular manufacturng exports contrbute smaller share n total exports than n other developng countres (some data from WDI). Lterature argues several possble sources of the stagnaton, whch ncludes lack of foregn drect nvestment (FDI) n the manufacturng sector (Lall [999], Bggst et al [995], Pack [99]). FDI s regarded as one of the mportant channels of technology transfer from developed to developng countres, and t s documented that such transfer led to the growth of local ndustres ncludng textle/garment, motor bcycle, automoble and electrc applances n many developng countres (Lall and Urata [00], UNCTAD [00], Ernst et al. [998]). Laggng several decades to Asa, FDI n garment ndustry started to flow nto several poor sub-saharan countres, n partcular Lesotho, Madagascar, Kenya and Swazland, around the year 000 due to provson of exclusve preferental access to the US market by the Afrcan Growth and Opportunty Act (AGOA). Around the world, garment FDI has facltated local frms partcpaton to export market n developng countres, and recently success s seen even n low ncome countres wth poor local ndustry lke Bangladesh and Vetnam. It s argued n the lterature that technology and knowledge possessed by foregn frms effectvely transferred to local frms through demonstraton by foregn frms, transactons between them and labour turnover from foregn to local frms (Sagg [000]), or presence of foregn frms reduces the fxed costs assocated wth exportng, such as establshng dstrbuton networks, learnng about consumer demand and buldng transportaton nfrastructure (Atken et al. [997], Greenway et al. [004]). In contrast to success n the world, export partcpaton by local frms s very lmted n Afrcan countres. It s partly due to slowdown of export growth from Afrca after termnaton of Mutfber Arrangement (MFA) n 005, whch resulted n weakenng the advantage of Afrcan garments n export market. However, local partcpaton was small even before 005 and local frms dd not know the MFA termnaton untl t happened accordng to the author s ntervews wth Kenyan local garment frms. Unqueness of local partcpaton to export market s not only ts small sze, but characterstcs of local frms engaged n exportng. Emprcal studes suggested that productvty and sunk costs assocated wth exportng matters export partcpaton. Most of studes on learnng-by-exportng found that frms wth hgh productvty tend to start exportng (e.g. Delgado et al. [00], van Besebroeck [005]). And studes on export partcpaton suggested to consder sunk costs needed for exportng, and showed that frms that have experence of export (and hence pad sunk costs) are more lkely to export (e.g. Roberts and Tybout [997], Clerdes et al. [998], Bernard and Jensen [999]). These evdences mply that new entrants wth lttle experence n the ndustry and no producton equpment are less lkely to start exportng than old frms, snce they are less productve and need to bear large sunk costs. In contrast to t, majorty of local frms started exportng n Kenya are, n fact, newly establshed frms by the owner wthout experence of garment ndustry. Also, ths s not consstent wth the evdences ndcated by FDI spllover lterature that local frms wth rcher absorptve capacty are more lkely to beneft from FDI spllover (Crespo and Fontoura [006], Sagg [000]).

3 Response to the emergence of FDI observed n Kenyan garment ndustry ndcates that frm behavor dffers from one supported by many emprcal studes n other developng regons. Ths may mply that emergence of FDI does not necessarly lead to development of local frms, though lterature argued lack of FDI s one of causes of the stagnaton n Afrcan manufacturng sector. In the prevous work, the author nvestgated absorptve capacty of local frms n Kenya, and found that techncal and market knowledge of garment export was entaled to sklled worker, and hence, by employng expatrates who used to work n FDI frms, local frms was able to start as a subcontractor regardless of ts experence and sklls (Fukunsh [00]). However, t stll does not necessarly support that new entrants are more postve n exportng. Usng orgnal frm data, the present paper attempts to nvestgate frm response to export opportunty brought by FDI nflow based on the Kenyan case. Consderng the market characterstcs n Kenya, credt constrant and sunk costs assocated wth entry to domestc market (nstead of export market) are ncorporated n a standard analytcal model of export partcpaton. Emprcal analyss s conducted through estmaton of credt accessblty and expected profts of exportng usng frm data n Kenya and Bangladesh, where the latter s used as representatve exporters. The results are generally consstent wth mplcatons of the model, and n partcular t suggested that credt constrant dscouraged local garment frms to start exports. However, they also demonstrated that Kenyan frms are less motvated to exportng than Bangladesh frms due to lower proftablty, controllng credt constrant. Snce FDI n garment ndustry s a sngle case of large scale manufacturng FDI n poor sub-saharan Afrca after the 980s, nvestgaton demonstrates unque behavor of Afrcan frms. In the next secton, FDI and export partcpaton of local frms n the Kenyan garment ndustry s descrbed. A modfed model of a local frm s export partcpaton s constructed n the thrd secton. The fourth secton nvestgates credt accessblty of local frms. Based on the result, the frms ncentve of exportng s nvestgated n the ffth secton. The last secton concludes the dscusson.. FDI and Export Partcpaton by Local Frms. Overvew n the LICs The assembly process of garment producton s characterzed by relatvely smple technology compared to other manufacturng actvtes. The sewng machne s the man equpment and the use of a smple sewng machne remans cost effectve when combned wth adequate worker skll and organzaton (Lall and Wgnaraja [994]). In contrast, a hgh amount of barrers exst n marketng due to the wde varety and quck change of consumer tastes. The latest market nformaton s assembled through retalers n developed countres, and takng that advantage, they provde full specfcaton of products wth manufacturers (Gereff and Memedovc [00]). Whle lnkage wth retalers s mportant, t s restrcted wth manufacturers and tradng companes n developed countres and East Asa, whch have establshed a lnkage through long-standng busness relatons wth retalers. Hence, the nvolvement of foregn frms s essental for the start of garment exportng from LICs, whch takes the form of FDI or subcontracts from foregn tradng companes or manufacturers.

4 Producton technology and know-how s obtaned by sendng sklled workers to manufacturers n a host country, or sendng local workers to a developed country for tranng. Technology and knowledge accumulated n local exportng frms s further transferred to other local frms through turnover of experenced local workers as well as subcontractng wth them. Rhee and Belot [989] and Mostafa and Klepper [009] documented that producton technology was transferred to Bangladesh workers through tranng n South Korea and further splled over to other frms through labor turnover. Export partcpaton by local frms was most mpressve n Bangladesh. Fve years after the frst garment exports to the US market, about 700 local exporters were n operaton. In Maurtus, FDI from Hong Kong frst came n 975 seekng for a non-quota status n the US market and the preferental trade agreement wth EU, and from that tme, garment exports had been expandng untl 000. Followng the emergence of FDI, local entrepreneurs (.e., local sugar corporatons, small busnesses and even ndvduals) had nvested n the garment ndustry and t was estmated to have contrbuted more than 50% of the total nvestment (Bowman [99]). In Sr Lanka, garment exports were trggered by FDI manly from East Asa n the late 970 s, but local frms also ventured nto exports and recorded a performance on par wth foregn frms (Athukorala and Rajapatrana [000], Lall and Wgnaraja [994]).. Garment Export n Kenya Export opportunty was brought about through the enactment of the Afrcan Growth Opportunty Act (AGOA) n 000. AGOA s a US domestc act devsed to remove tarffs on a broad range of products mported from SSA countres satsfyng certan poltcal and economc condtons. Ths new trade scheme has had a drastc mpact on the Afrcan garment ndustry. Several Afrcan countres have been rapdly ncreasng garment exports to the US market, and n Kenya, exports grew by 600% between 999 and 004 (Fgure ). Rapd growth of exports n Kenya s largely brought about by the frms regstered n Export Processng Zones (EPZs), whch accounted for 85% of the exports to the US n 00. After the enactment of AGOA, new nvestments n the garment ndustry have flown nto EPZs, and n 004, 0 garment frms produced. mllon US dollars, and employed 4,600 workers (Table ). EPZ frms produce manly low-prced basc wear ordered from US buyers. All frms are funded wth foregn captal from the Mddle East (Bahran, UAE), South Asa (Inda, Sr Lanka) and East Asa, whle some frms are jont ventures wth domestc captal. The growth trend was dsrupted n 005 followng the termnaton of the MFA. The abolton of the export quota mposed on man exporters resulted n a massve ncrease of exports from compettve countres lke Chna and Inda, and consequently, exports from Afrca have dropped by 6%. Although the adverse effect was relatvely small n Kenya (-0.8%, Fgure ), the growth trend dsappeared and several EPZ frms were closed down.. Local exportng frms We have dentfed that at least 9 local frms started or sgnfcantly ncreased exports to the US and EU market after the arrval of FDI (Table A n appendx ). Though many frms exported to Afrcan markets, 4

5 they are not ncluded n FDI spllover due to ther havng less relevance to FDI. As Afrcan markets dffer wdely from the US/EU markets n terms of the type and qualty of products, the volume of orders, and competton n the market, exports to the Afrcan market unlkely resulted from the transfer of technology and knowledge from foregn frms. Although suffcent nformaton was not obtaned from some frms, we beleve that our estmaton of the number of local exporters s farly accurate. Among the 9 local exporters, 5 frms were newly establshed as an exporter (new exporter), whle 4 frms used to supply to the domestc market and then, have added or swtched to the US market (swtched exporter). Most of the new exporters were set up after 00. All local exporters from whch we managed to get nformaton (6 out of 9) produced export garments on subcontract order (CMT) as a major part of sales, whle several frms were takng orders drectly from foregn buyers (FOB) as well as subcontractng. Subcontract orders were mostly from EPZ frms n Kenya. We have ntervewed 7 new exporters and swtched exporters among the 9 local exporters. The number of employees of the ntervewed frms range from to 800 and the average s, whch s.9 tmes larger than the average of local frms and about 0% of the average EPZ frms. Turnover also has large varaton from 5.9 mllon to 65.0 mllon KShs and the average s 60.0 mllon KShs, whch s 9% of the average EPZ frms (Table ). There s clearly a mnmum scale n export producton; except for one small frm, all local exporters employ more than 70 workers and most of them have 0 workers (Table A). Although the local exporters are larger than the average local frms, they are much smaller than the average EPZ frms n terms of employment and turnover. The mtaton of the export busness by local frms s surprsngly small sze compared wth the other garment exportng countres. The nactve response by local frms s possbly related to the termnaton of MFA, snce they may have expected sgnfcant change n the export market after 005 and suspended nvestment untl they knew how the market would be. If part of the captal s sunk, decson makng wll be forward-lookng and ths can be consdered ratonal behavor regardless of a frm s rsk preference (Dxt and Pndyck [994]). However, our ntervew wth local frms ndcates that almost all of them were unaware of these nsttutonal uncertantes n 00. Though our ntervew was conducted after those events, only 4 out of 8 local non-exportng frms ntervewed repled that they had known of the MFA termnaton and had antcpated the shrnkage of exports n It was also revealed that even many local exporters were unaware of the termnaton of MFA when they started. Ths may be because EPZ frms dd not have any ncentve to tell local frms about these uncertantes and few local non-exportng frms had We manly based ths on the lst of frms regstered as Manufacturng under Bond (MuB) to dentfy a local exporter, because they are lkely to regster as MuB, whch allows them to keep mported materal bonded. However, we were not able to confrm 4 frms on the lst, manly because the regstered phone number was not n servce. It s lkely that most of undentfed 4 frms dd not start an operaton. One small frm (frm C n Table A) specalzed n prntng. There was another uncertanty n regard to the AGOA. Whle the concessonal rule of orgn was crucal for Afrcan exporters to reman compettve n the US market, t was scheduled to be revsed n July 004 and just before the termnaton date, no decson regardng the change had been reached (the rule was fnally mantaned). 4 Snce those four frms repled they dd not know AGOA change, ther reples that they were aware of termnaton of MFA are reserved. 5

6 connectons wth foregn frms. 5 Ths evdence suggests that the nsttutonal uncertantes were unlkely to have affected the local manufacturers choce of market. It s also noted that the volume of FDI and the growth of garment exports n Kenya was no less than that of the other garment exportng countres. Employment by EPZ frms n 004 was as large as that of Maurtus n 984 and larger than that of Sr Lanka n 985, whch s about 0 years after the start of exports n both countres. The annual growth rate of exports between 000 and 004 n Kenya was 58.5%, whch s much faster than the growth rate n Maurtus from 985 to 990 (0.0%) when the fastest growth was recorded. Reflectng the growth of exports, EPZ frms offered subcontracts to many local frms. Among the local frms n whch the author conducted ntervews, 7.7% were contacted by them about subcontracts. 6 Although there were about 0-50 garment frms wth more than 0 employees n Kenya based on our estmaton (Table ), only 4 frms swtched to the export market. In contrast, the newly establshed frms have been more postve about startng an export busness. Owners of the new export frms were manly from another ndustry and nvested n garment exports as a dversfcaton of busness. As far as we know, four owners have experence n the textle or garment ndustry, whle eght owners have experence n another ndustry or publc sector, whch ncludes the export of hortcultural products, transportaton, supermarkets and hotels. An nvestment semnar held by the government and World Bank n 00 facltated the dssemnaton of nformaton regardng nvestment opportuntes. Although these busness owners had very lttle experence and knowledge n garment producton, t s found that all new exporters employed expatrates who used to work n EPZ frms. They have substantal knowledge n marketng as well as producton technology for garment exportng, and owners of new exporters expressed that they had no serous problem n starng exportng busness (Fukunsh [00]). Though sample s small, new exporters performed as effcent as Kenyan EPZ frms and Bangladesh frms. 7 Transfer of knowledge through labour turnover s commonly seen n other developng countres, where the Bangladesh case s most famous, and t mples that absorptve capacty s not a sgnfcant barrer of local partcpaton nto export market.. Analytcal Framework and Methodology. Analytcal Framework Enterng the export market entals nvestment n most cases. If a frm starts producton for an export market, t needs to nvest n physcal captal and possbly n the creaton of dstrbuton channels, logstcs nfrastructure, human captal and knowledge of market demand. Gven the uncertanty regardng the future proftablty of the export market, whether or not such nvestment s sunk becomes a crucal ssue n makng the decson to export. As Dxt and Pyndck [994] argued, f uncertanty s a Markov process and nvestment s rreversble, a frm may have ncentve to refran from nvestment even when expected future 5 Ths evdence s based on the author s ntervew. 6 New exporters are not ncluded. 7 See Table 8 for comparson of techncal effcences. 6

7 profts are greater than the nvestment value. Standard model assumes that a frm needs to nvest sunk cost I when t partcpates n the export market f t does not export durng the perod rght before, and proft from the export market, π f t, s serally correlated. Let the proft from the domestc market be π d t, dscount rate be ρ, and the decson of export partcpaton be s t where s t = when a frm starts to export. A frm s value functon s descrbed as V t f d [ s ( π ( s ) I ) + π ] τ max Et ρ t+ τ t+ τ t+ τ t τ, st + τ τ = 0 = + whch leads to Bellman s equaton V t f d [ s ( ( s ) I ) + π E [ V s ] ( E ) = π ρ. max t t t t t + t t+ st Ths characterzaton mples that a frm partcpates n exportng whenever [ V s = ] ( s ) I + E [ V s 0] π. f t + ρet t+ t t ρ t t+ t = Ths mples that non-exportng frms start exportng when expected future profts earned from startng to export at that tme are greater than sunk costs plus expected future profts should they decde to wat durng ths perod. Snce a frm can start exportng after t+, the second term n the RHS contans not only future profts of domestc supply but that of export supply, whch s called the opton value (Dxt and Pyndk [994]). As opton value s greater than or equal to zero, a frm has an ncentve to wat even f future expected profts n export market exceed fxed costs plus future domestc profts. 8 We make some modfcatons to the standard model so that our model fts wth the realty of the Kenyan garment frms. Though the standard model does not ncorporate t, credt constrant s sgnfcant among Kenya frms (Isaksson and Whlborg [00]), and has receved consderable attenton n the FDI spllover lterature. A frm wth credt constrants may fnd t dffcult to fnance nvestment to supply to the export market or to multnatonal frms (Javorck and Spatareanu [009]). In the case of the garment ndustry, ntal nvestment s needed mostly for expanson of physcal captal, gven that subcontractng from foregn frms does not requre a long-dstance logstcs factor, dstrbuton channels n foregn countres or customs clearance as we wll see n secton 4. However, as mentoned, there s a mnmum producton scale for export producton, whch s larger than the average scale of local frms. A frm that s unable to fnance at a mnmum scale has no possblty of partcpatng n the export market. And gven the exchangeablty of equpment between domestc and export supply, moderately credt-constraned frms may manage to prepare the mnmum captal by utlzng the current captal used to supply n the domestc market. In contrast, those wth good credt access can fnance export producton facltes n addton to domestc ones as assumed n the standard model. Therefore, the degree of credt access substantally affects the export decson problem, and for convenence, we call the frm that s not able to fnance mnmum captal a Type t 8 On the other hand, the above condton mples ncumbent exporters contnue to export under the less restrctve condton as they do not consder fxed cost I anymore. So sunk cost leads to a dfference n export decson between current exporters and non-exporters. Robert and Tybout [997] showed emprcal evdence of effect of sunk cost on export decson (as dd some other studes,.e., Bernard and Jensen 999, Clerdes, Lach and Tybout 998). 7

8 frm, the moderately constraned frm a Type frm and the frm wth good credt access a Type frm. For Type frms, the export decson problem becomes a choce between partcpaton n the domestc or the export market. Another mportant characterstc n the Kenyan context s that once a frm wthdraws from the domestc market, re-entrance to t necesstates sunk cost to rebuld the relatonshp wth buyers due to the strong lnkage between buyers and supplers. Because of the low number of supplers n the Kenyan garment market and the fact that the man products of local frms, unforms, requre frequent contact wth buyers to satsfy customers exact specfcatons, lnkages between buyer and suppler are relatvely stable. In contrast, the nvestment for exports s less lkely to be sunk, snce the nvestment s manly for physcal captal and there s a secondhand market n Kenya. 9 Thus, for the Type frm, the decson problem s dynamc because of the sunk cost of the domestc market, whle t s more of a statc problem for Type frms gven the substantal reversblty of nvestment. Let us assume a postve sunk cost for re-entrance to the domestc market, W>0, no sunk cost for the export market, and reversble physcal captal. Now the cost of captal s ncorporated n proft as a rental cost, and then, Bellman s equaton for a Type frm s V t f d ( E [ s + ( s )( π s W ) E [ V s ]) = π ρ. max t t t t t t + t t+ st And a frm decdes to export when the followng condton s satsfed: f d π π s W + ρ( E V s = E V s ). () t [ ] [ ] t t t t+ t 0 t t+ t = Ths condton dffers from the one based on the standard model n several aspects. Frstly, gven that sunk cost apples to the domestc market rather than the export market, the crtcal proft level that a frm chooses f * d an export market for the exporter s π W + ρ( E [ V s = ] E [ V s = ] ) E t π and they now = t t t+ t 0 t t+ t consder sunk cost W, whle the one for non-exporters s f * d π π + ρ( E V s = E V s = ]) and they do not. Secondly, as the problem s makng N [ ] [ = t t t+ t 0 t t+ t the choce between the two markets, the proft of exports s compared wth the proft from the domestc market. It s noted that the thrd term n the RHS of () s the dfference of expected future proft when s t = 0 and s t =, and t s necessarly postve for non-exporters at t. By remanng n domestc supply at t, a frm can avod the possble loss that an exporter ncurs at t+ n case π f t+ < π f* E, whle t can swtch to the export market wthout sunk cost whenever t s more proftable. Therefore, E [ V s ] > E [ V s ] t t+ t = 0 t t+ t = 0 holds and the last term n () s postve. The reservaton of ths statement would be n the case where future proft (π f t) has an upward trend. Learnng-by-exportng s a typcal example; frms supplyng the export market necessarly mprove productvty faster than non-exportng frms, and hence, future profts grow faster. 0 Then, the last term n () can be negatve. Hence, f the learnng-by-exportng effect s not 9 Secondhand machnes were used n many factores. Most respondents repled to the queston about resale value of equpment n our survey. 0 Emprcal evdence of learnng-by-exportng s mxed. Some emprcal studes support the lnk between exports and productvty growth through learnng-by-exportng, compettveness pressure and ncreasng returns 8

9 substantally large, the partcpaton condton () ndcates π f t > π d t, that s, a non-exporter does not swtch to the export market unless the current export proft s greater than current domestc proft. On the other hand, the decson problem of Type frms s statc as long as no sunk cost for export partcpaton, and they start exportng when the current export proft s postve. The above model assumes rsk neutral frms, but n the context of Afrca, lterature ndcates that frms are rsk averse because of poor access to credt (Coller and Gunnng [999], Bgsten et al. [00]). Due to stronger lnkages between buyers and supplers n the domestc market, t s reasonable to assume that domestc profts are more stable than those of exports, and rsk-averse frms prefer the domestc market f expected proft s the same. In that case, crtcal profts trggerng export partcpaton (π f* N) rses by rsk premum, whch s determned by dfference n perceved rsks n the two markets and degree of rsk averson of ndvdual frms. Ths analytcal framework s consstent wth the result of the ntervew wth local exporters. Table ndcates that 0 frms among 8 samples named dffculty of physcal nvestment as a reason not to start exportng. Sx frms repled that the export market s rsky manly because of the volatlty of demand. The proftablty of the export market s questoned by 0 frms (ncludng those that repled that the current domestc busness s proftable) n comparson wth the domestc market. Ths mples that they compare the export and domestc markets rather than vewng the export market ndependently. Many frms explaned that low expectatons of the export market are manly due to uncertanty of order and the relatvely large physcal nvestment requred. Though the above model assumes unbased nformaton about proftablty of exportng, one may thnk that there s uncertanty and frms need to guess based on the avalable knowledge, as dscussed n socal learnng lterature. We rule out the possblty of socal learnng n our analyss due to the followng evdence. Most of the local frms communcated wth EPZ and got to know the detals of subcontract orders such as product specfcaton, quantty and order prce. The ntervews wth managers revealed that for local frms wth experence n garment assembly, t was not dffcult to guess how proftable they were.. Strategy for Identfcaton Though standard methodology to dentfy determnants of export partcpaton s an econometrc approach usng the probt or logt model (.e., Roberts and Tybout [997], Bernard and Jensen [998], Javorck and Spatareanu [009]), t s not applcable n our case due to the small number of frms enterng the export market n the sample as well as n the populaton. Our approach s to drectly nvestgate the structure of frm s market choce problem usng the qualtatve and quanttatve nformaton of ndvdual frms. Based on the above framework, local frms non-partcpaton n the export market s attrbuted to that they were ether not able to do so due to lack of credt access, or they were not motvated due to unattractve proftablty and/or hgh rsk n the export market. In ths paper, these two factors are to scale (Van Besebroeck [005], DeLoecker [007], Bgsten et al. [004], Grma, Greenway and Kneller [004]), whle some of them support self-selecton explanatons (Clerdes, Latch and Tybout [998], Bernard and Jensen [999], Delgado, Farnas and Ruano [00]). 9

10 approached separately. Frms credt access s nvestgated through ntervews wth local frm managers. Credt access wll be estmated from local frms credt use. Also, by estmatng the mnmum captal value from captal demand functon and comparng t wth local frms current captal value, we wll determne the nvestment necessary for an ndvdual frm to start exportng. In prncple, these estmates wll tell us whether a frm can start to export or not, but n practce, t s very dffcult to know precsely how much credt a frm can access. Therefore, we wll at least dentfy a frm wthout access to formal credt, and f ts captal value s less than the mnmum scale, we deem that the frm s Type and not able to partcpate n the export market. For moderately credt-constraned frms (Type ), our model ndcates that to satsfy the partcpaton condton (), export profts have to be larger than domestc ones plus rsk premum. Snce Type frms are serously constraned from expanson of producton capacty, they are motvated to exportng when there s a sgnfcant gap n proftablty n export and domestc markets or learnng-by-exportng works. Usng frm data of Kenyan and Bangladesh ndustry, producton functons of exporters and domestc supplers are estmated and learnng-by-exportng effect s tested. For Type frms whch starts exportng when expected profts are postve, expected profts are smulated based on the estmates of producton functon. If estmaton results are consstent wth ther response, our model explans the ncentves structure of local frms. In addton to those exercse, expected profts of exporters n Kenya and Bangladesh are smulated and compared to those of local exporters. Ths ndcates whether our model can explan the heterogeneous response to export opportunty. Ths approach has advantages of nvestgatng the structure of the decson problem. In most econometrc approaches, the reduced form representng the relatonshp between a frm s characterstcs and realzed choce s estmated, yet the true structure s that characterstcs affect choce through a frm s expectaton on proft earned n a new market. The reduced form relatonshp may ncorrectly estmate determnants f omtted varables and/or endogenety problems are sgnfcant. By drectly lookng at expected proft, our methodology avods msdentfcaton of determnants. On the other hand, dffcultes le n the collecton of measurable data related to decson makng, such as precse nformaton on credt access and rsk preference. In partcular, lack of knowledge about rsk preference of ndvdual frms caused ambguty n emprcal test of the theoretcal model. In case that the evdence s not consstent wth the model predcton, we cannot systematcally nvestgate whether t s caused by frms rsk preference or assumptons on whch our model based. Ths arses n the analyss of Type frms n secton 6., but fortunately the result dened effect of our assumpton.. Estmaton of Producton Functon, Productvty and Expected Profts Gven the small number of exporters n Kenya, we added Bangladesh frms to the sample to estmate Conley and Udry [00] also estmated expected profts n a study of farmers learnng of new technology n Ghana. They used neghborng farmers realzed profts (wth control of producton characterstcs) for expected profts, and t s bascally same methodology as ours. 0

11 producton functon so as to have robust estmates. Bangladesh frms are exportng low-prced garments, whch are n the same market segment as the products of Kenyan exporters. Gven ther success n the export market for more than 0 years, t s reasonable to regard them as a representatve exporter n a low-ncome country. Furthermore, the addton of Bangladesh frms allows us to compare expected profts between local frms n Kenya and a successful exportng country, and to nvestgate the dfference of local frms responses to export opportunty. OLS and stochastc producton fronter model are used for the estmaton of producton functon. The endogenety problem on nput choce may arse f a frm determnes the amount of nput, partcularly labor, knowng ts own productvty whch s unobservable for us. The fxed effect model and some estmaton procedures, such as those by Olley and Pakes [996] and Levnson and Petrn [00], have been suggested, but they are not applcable to cross-sectonal data. Stochastc fronter model can avod ths problem by makng assumptons on the dstrbuton of productvty. In ths methodology, a frm s productvty s measured as techncal effcency whch represents dsperson from the producton fronter ndcatng the greatest output gven nputs among the samples. Specfcally, t assumes a producton functon Ls Lu Y = α K TE error, () where Y: output, K: utlzed captal, Ls: sklled labor, Lu: sem-sklled labor, TE: techncal effcency wth value between 0 to, error: stochastc errors wth mean at one, and represents an ndvdual producer. TE s estmated by separatng regresson resdual to TE and random error based on the assumpton on dstrbuton of TE (Jondrow et al. [98]), though we do not know t. As choce of dstrbuton affects estmate of parameters as well as techncal effcency (Kumbhaker and Lovell [000]), we have used several dstrbutons to check senstvty. The learnng-by-exportng effect s tested based on the cross-sectonal varaton of techncal effcency accordng to export experence. To see senstvty of dstrbutonal assumpton to techncal effcency, we estmated parameters by OLS wthout dstrbutonal assumpton, and then, techncal effcency s separated by method of moment wth dstrbutonal assumpton followng Olson et al. [980]. Further, alternatve productvty estmates are obtaned by the ndex number method, whch s free from the arbtrary assumpton on dstrbuton and endogenety problem of nput choce. Followng Caves et al. [98], productvty of an ndvdual frm s measured relatve to a hypothetcal average frm wth average nputs, output, and factor shares by the followng formula. n n ( ) ( ) s + s lntfp lntfp = lny lny ( ln xn ln xn ) n n n ( )( ) s + s ξ + ( ln xn ln xn ) n where x n s nput (n = K, Ls, Lu), s n s the factor share of each nput, ξs returns to scale, and the varables wth upper bar (.e., ln Y ) are sample averages. The thrd term s added to control returns to scale, so that estmates can be compared wth techncal effcency whch does not nclude the returns to scale effect on,

12 productvty. The estmaton of expected profts s based on a producton functon estmate. The use of producton functon nstead of proft functon s to avod bas stemmng from the use of rental prce that s not clearly observable for us. In many cases, frm owners provde ther own land or money for ther frms but dvdend for ther contrbuton s not clearly shown n an accountng book. Therefore, captal servce cost n our frm data can be wrongly measured and, consequently, so can the rental prce. If we assume the Cobb-Douglas producton functon, then dualty allows the dentfcaton of cost functon from producton functon estmates. Wth the producton functon (), a frm mnmzes cost, C = r K + ws Ls + wu Lu, where r s rental prce of captal, ws s the wage for a sklled worker and wu s the wage for a sem-sklled worker. It s assumed that the frm may msallocate nputs, and then, actual cost becomes greater than the mnmum cost (allocatve neffcency). The frst order condtons of cost mnmzaton wth allocatve neffcency are expressed as AE ws wu Lu Ls AE r wu Lu K AE r ws Ls K = = =, where AE n >0 for all n, and t s equal to one when factor allocaton s optmal, gven factor prce ratos. From () and FOCs of cost mnmzaton, condtonal nput demand functons are gven by α α α = = = AE AE error TE Y wu ws r Lu AE AE error TE Y ws wu r Ls AE AE error TE Y r wu ws K, () where = + +. Multplyng respectvely by a factor prce, the cost functon s gven by AE TE Y wu ws A r Lu wu Ls ws K r C ˆ = + + =, where α = n n n A n=,,, (predcted output), and TE Lu Ls K Y ˆ = α + AE AE + = AE AE AE AE AE. The frst through ffth terms on the rght hand sde compose the cost fronter functon, and the last two terms represent the dsperson of actual

13 cost from the fronter; they are the costs of techncal neffcency and allocatve neffcency respectvely. Note that the above cost functon accounts only for utlzed nputs, snce captal n the producton functon s adjusted by the utlzaton rate. Addng the cost of dle captal, η, n multplcatve form, the actual cost s descrbed as TC = A r ws wu Yˆ TE AE η, (4) where η. Expected proft s obtaned by subtractng expected cost from sales n the export market, ( r, ws, wu, Y, TE AE η ) π ˆ, ˆ py TC, =. (5) Estmates of expected proft wll be gven by nsertng an ndvdual frm s factor prces, producton sze, neffcences and share of dle captal. It s noted that our approach can avod bas due to measurement error of rental prces not only n parameter estmates but n the estmaton of expected profts gven by (5). Though rental prce enters nto the equaton (5) drectly, measurement error s offset by AE, snce AE and AE ncorporate the error of rental prce as shown n the FOCs of cost mnmzaton.. Source of Informaton Two types of nformaton were collected by the author and collaborators. Frm data of the Kenyan and Bangladesh garment ndustres were collected n 00 by the Insttute of Developng Economes, Unversty of Narob and Unversty of Dhaka. The survey ncludes 7 frms n Kenya and frms n Bangladesh, of whch 47 and 65 frms were used for the analyss after the elmnaton of the samples of poor qualty. The number of samples reflects the sze of the ndustres, where the Bangladesh ndustry has more than 000 frms and the Kenyan ndustry s estmated to consst of 0-50 frms. The sample was selected usng the stratfed samplng method n Bangladesh, whle the Kenyan sample s the result of an exhaustve survey based on several ncomplete frm lsts due to the non-exstence of a complete lst. The Kenyan sample conssts of local exporters, 5 foregn exporters and 9 local frms supplyng to domestc and Afrcan markets (Table 4). On the contrary, all Bangladesh frms n the sample are exporters and only two of them are foregn owned; the rest are domestcally owned. Frm ntervews were conducted for Kenyan local frms by the author n 005 and 006 n order to collect qualtatve and quanttatve nformaton about the adopton process of local exporters, and the absorptve capacty, credt access and ncentves of local non-exporters to start exportng. It ncludes 0 local exporters and 8 local non-exportng frms (Table 4). For supplementary nformaton, 5 EPZ frms, Export Processng Zones Authorty, Mnstry of Trade and Industry and Kenyan Assocaton of Manufacturers (ndustral assocaton) were ntervewed. Informaton obtaned through the frm ntervew s manly used for the analyss of credt access, whle AE and equalty holds when AE n = for all n; the cost of allocatve neffcency s null when there s no neffcency n nput allocaton. See Appendx for detals of the samplng method and data constructon.

14 that of the frm survey s used n the estmaton of producton functon and smulaton of profts. It s noted that the two are not perfectly matched; the survey sample s larger. Therefore, qualtatve nformaton about credt access obtaned through the ntervews was generalzed to the smulaton samples and appled to the smulaton. In the process of generalzaton, we have been careful regardng the possble dfference of frms characterstcs between the two samples. As for credt access, we used frm sze as a key by whch to apply the fndngs of the ntervews to smulaton exercse gven the clear relatonshp between sze and credt use. 5. Credt Constrant and Export Opportunty The ntal nvestment requred for a garment assembler s relatvely small because of ts labor ntensveness. The most crucal equpment s sewng machnes; machnes for cuttng fabrcs and washng and pressng fnal products may also be needed, dependng on the product. Whle Kenyan local manufacturers have 5.6 sewng machnes on average, the average number n Bangladesh frms s 7 machnes, and even the 5 percentle frm equps machnes. Therefore, many of the Kenyan local frms needed to expand ther capacty. Mnmum captal sze s estmated by condtonal captal demand functon shown n equaton () wth assumpton on mnmum output. 4 We refer to the actual output of the relatvely small local exporter, whch employs 84 workers, as the mnmum scale. Frm s characterstcs, such as factor prces and effcences, are entered nto the functon, whch gves an estmated captal demand for an ndvdual frm. Based on the result of the prevous secton on absorptve capacty, local frms are assumed to mantan the same techncal and allocatve effcency as they dd n the domestc market. Thrty-nne local non-exportng frms n the survey were used for smulaton. The smulated value for a frm wth average characterstcs s 8,87 US$. Comparng the estmates wth the current captal value, necessary nvestment s estmated for the ndvdual frm. Table 5 ndcates the rato of necessary nvestment to current captal value by frm sze. It shows that frms have suffcent captal, whle 6 frms need expanson and of them need to ncrease by more than double. Credt accessblty s nvestgated through ntervews. Access to formal credt clearly dffered accordng to the sze of the frm. Wth the excepton of one case, none of the frms wth less than 49 workers had used formal credt for last 5 years, whle 75% of those wth more than 50 workers have used formal credt (Table 6). The manager s judgment of credt accessblty almost always corresponded wth credt use (rght hand sde of Table 6). Then, we set a boundary for credt access at 50 workers. Combnng ths nformaton and Table 5, t has been dentfed that frms wth less than 49 workers and less than the mnmum captal sze cannot start to export due to lack of credt (Type frm) and account for 7.8% of local non-exportng frms n our sample (shaded area of Table 5). Assumng our sample represents the populaton, the smulaton results suggested that about 7% of local frms were not fnancally feasble to enter the export market. The other frms are possbly Type or. Snce most of fnancal nsttutes requre collateral n Kenya, 4 The smulaton does not nclude land and buldng as these can be rented. See appendx for detals of the smulaton method. 4

15 the amount of credt depends on the frm s assets. Hence, unless they assume assets other than producton equpment, a frm cannot make an nvestment greater than the value of ts current equpment. Wth ths conservatve assumpton, all frms can start to export by usng ther current equpment, gven that needed expanson of equpment s smaller than current equpment value n all cases. Wth the same assumpton, only three frms can nvest n an addtonal producton ste for export supply, as the rest of frms equpment value s less than the mnmum scale and s not large enough for collateral. Then, only three frms are canddates for Type frms, and they account 7.7 % of our sample. 6. Expected Proftablty of Local Frms 6. Producton Functon Estmaton To nvestgate the dfference of producton characterstcs for export and domestc markets, a separate producton functon s estmated. Estmatons use OLS and the stochastc fronter model, whch s descrbed as ln 0 + 0Sewng + ln K + ln Ls + Y = ln Lu u + v, where Sewng s a dummy varable dscernng frms wth only a knttng process (=0) and those wth a sewng process (=), = exp(α), u = -lnte, u >0, and v = ln(error ). Ineffcency, u, s assumed to follow a half-normal dstrbuton, N + (0, σ u ), or exponental dstrbuton, N + (μ, σ u ), and the random error component, v, s assumed to be normally dstrbuted wth mean zero, N (0, σ v ). Heteroskedastcty on random errors s consdered, snce group-wse heteroskedastcty around process dummy (Sewng) was ndcated (results not reported). Specfcally, auxlary model, lnσ v =δ(, Sewng) was added to estmate σ v. The frst set of models ncorporates dfferent parameters for exporters and non-exporters to reflect ther heterogenety by addng a non-exporter dummy and ts nteracton terms wth nputs (columns and of Table 7). They show that all nteracton terms are statstcally nsgnfcant. The second set ncorporates only a non-exporter dummy (no nteracton term), and no sgnfcant dfference of a constant by market orentaton s ndcated n ether model (columns and 4). Estmates based on stochastc fronter wth exponental dstrbuton assumpton show the same result (not reported). Those results ndcate that parameters are homogenous between exporters and non-exporters, and then, a model wthout a non-exporter dummy s estmated (column 4-6). Excluson of the dummy does not lead to a drastc change of parameter estmates, whle the parameter estmate for captal becomes smaller and that of labor becomes larger. Estmates of the nput coeffcent are sgnfcant except for a captal coeffcent n the OLS model. As for the economes of scale, aggregaton of parameters s greater than one n all the three cases, but the hypothess of constant returns to scale s not rejected at the 0% level except one case. These exercses show that there s no sgnfcant change n producton functon by market orentaton, and only weak support s found for ncreasng returns to scale. Therefore, shftng from the domestc to the export market does not brng substantal ncrease n profts wthout a large expanson of scale or productvty 5

16 mprovement. The relatonshp between exportng and productvty s nvestgated to examne the learnng-by-exportng effect. To get an overvew of the relatonshp, techncal effcency s compared wth market orentaton. The results of estmaton are n lnes 4 and 5 of Table 8. Although the level of the averages dffers by estmaton model, all estmates show that the average of exporters s not hgher than that of non-exporters. To form a more rgorous nvestgaton, effect of export status on techncal effcency s estmated. Followng Kumbhakar, Gosh and McGuckn [99], export status and export years as well as other exogenous varables are assumed to correlate wth techncal effcency through the mode of ts dstrbuton (μ) as lny = 0 + ln K + ln Ls + ln Lu u + v, μ = θ Export + θ Age + θ Age * Export + φw where u ~ N + (μ, σ u ), v ~ N (0, σ v ). Export s a dummy varable takng one for exporters, Age s frm age, Age*Export s nteracton term, and W s a vector of the varables related to a manager s characterstcs and busness envronment. As all exporters n our sample have been servng the export market snce ther establshment, the nteracton term (Age*Export) pcks up the effect of export experence, whle the effect of general operaton experence s controlled by Age. On the other hand, Export wll capture the effect of export status regardless of length of experence. As frequently mentoned, the relatonshp between export status and effcency can be two ways, and thus our estmates ndcates only assocaton. However, estmated assocatons of Export and Age*Export are not sgnfcant regardless of ncluson of other varables (Table 9). There may be another possblty of learnng-by-exportng that export wll mprove allocaton of factors. Then, effect on allocatve effcency s nvestgated. Allocatve effcency enters nto cost functon as AE shown n equaton (4). Log of AE s regressed on exogenous varables, assumng a proportonal effect of export years. Table 0 shows that no sgnfcant assocaton of export status and experence, whle excludng export dummy, export years sgnfcantly reduces AE. Hausman s test does not reject the null that OLS estmator s consstent, and thus, we accept the sgnfcant and negatve coeffcent of export experence. 5 One year of experence reduces 0.49% of the cost of allocatve effcency, whch leads to a 0.4% ncrease n proft. Though ths evdence s based on a cross-sectonal sample, learnng-by-exportng s also confrmed n the panel data of the Moroccan garment ndustry (Clerdes, Latch and Tybout [998]). These exercses ndcate that whle shftng from the domestc to the export market does not ental a structural change of producton functon and does not lead to the mprovement of productvty, t does brng about the reducton of allocatve neffcency accordng to years of export experence. The learnng-by-exportng effect s supported, but relatve to the sze of dscount rate, and the effect s so small 5 To control endogenety of export years, IV estmaton was carred out usng the average tenure of sklled workers and that of sem-sklled workers as an nstrumental varable. Whle average tenure tends to be correlated wth frm s age, t s unlkely to have causaton wth cost of allocatve effcency that s fundamentally related to manageral capacty. 6

17 that t s unlkely to have a sgnfcant mpact on a frm s decson on market choce. Though expected future profts wll ncrease by 0.4% every year, t s also dscounted by 0.67% f the real nterest rate s used. 6 Therefore, our exercse ndcates that no sgnfcant proft change s expected for local frms by smply changng from the domestc to export market, and ths leads to the followng three mplcatons. Frst, n the absence of any sgnfcant learnng-by-exportng, the partcpaton condton () holds. A local frm swtches to the export market when export proft at ths perod s greater than domestc proft plus loss of future profts by choosng to export, whch results from the sunk cost of re-entry nto the domestc market. Second, to realze greater proft from the export market, a local frm needs to expand ts producton capacty or mprove ts effcency through the frm s unque effort. Thrd, as the export market s large enough to allow a frm to freely expand ts producton capacty, FDI may contrbute to the development of local ndustry through an ncrease of producton scale rather than through productvty mprovement. Yet, credt access wll be a constrant for expanson. 6. Smulaton of Expected Profts of Type Frms Based on the producton functon estmates (OLS n Table 7), cost functon was drawn as the equaton (5) and proft s smulated by the equaton (6) for Type frms. We assume that local manufacturers wll perform n the export market as effcently as they do n the domestc market, based on the result that local exporters performances were no less effcent, and that no sgnfcant learnng-by-exportng effect was dentfed. Also, t s assumed that local manufacturers can employ labor for the same wages they were payng, gven the substantal pool of sem-sklled workers resultng from the shrnkage of the ndustry after the trade lberalzaton. In terms of rental prce, ntervews demonstrated that local frms wth more than 50 employees were able to borrow from fnancal nsttutons at a rate of between 4 and 0%, and n the smulaton, rental prce was estmated assumng a nomnal nterest of 0%, the maxmum n the above range. 7 Three frms wth 50 workers and a mnmum producton scale are used for smulaton. The producton scale s set to the mnmum. The result ndcates that expected proft s postve for all three frms (lne n Table 0). And n two out of three frms, proft per captal value s larger than one; that s, one year of operaton wll cover captal nvestment. Ths smulaton result ndcates that the export market s expected to be very proftable. 8 Therefore, wthn our framework, non-partcpaton by those local frms can be attrbuted to ther rsk-averseness; that s, expected proftablty was not large enough to cover the rsk premum that local frms requre. It s noted that ths dscrepancy between the model mplcaton and evdence may be resulted from our assumpton of no sunk entry cost n export market, but the result that expected profts s larger than captal nvestment denes effect of sunk cost on non-partcpaton. 6 Average from 999 to 00 based on the World Bank [006]. 7 We chose 0% consderng nformaton from World Bank RPED data, whch showed that the nterest rates of bank loans used by garment frms are between 7 and %. See appendx for the estmaton method of rental prces. 8 Ths tendency does not alter even f we nclude Type frms n the smulaton. For local frms wth more than 50 workers, the medan s.0. Hgh proft-captal rato s a feature of exporters n our survey sample. The same trend was observed n Bangladesh garment frms n Bakht et al. [009]. 7