March 2003 Insttutonal Effcency, Montorng Costs, and the Investment Share of FDI Joshua Azenman and Mark M. Spegel * Abstract Ths paper models and tests the mplcatons of nsttutonal effcency on the pattern of foregn drect nvestment (FDI). We post that domestc agents have a comparatve advantage over foregn agents n overcomng some of the obstacles assocated wth corrupton and weak nsttutons. We model these crcumstances n a prncpal-agent framework wth costly ex-post montorng and enforcement of an ex-ante labor contract. Expost montorng and enforcement costs are assumed to be lower for domestc entrepreneurs than for foregn ones, but foregn producers enjoy a countervalng productvty advantage. Under these asymmetres, multnatonals pay hgher wages than domestc producers, n lne wth the nsght of effcency wages and wth the evdence about the multnatonals wage premum. FDI s also more senstve to ncreases n enforcement costs. We then test ths predcton for a cross secton of developng countres. We use Mauro s (1995) ndex of nsttutonal effcency as an ndcator of the strength of property rghts enforcement wthn a gven country. We compare nsttutonal effcency levels for a large cross secton of countres n 1989 to subsequent FDI flows from 1990 to 1999. We fnd that nsttutonal effcency s postvely assocated wth the rato of subsequent foregn drect nvestment flows to both gross fxed captal formaton and to prvate nvestment. Ths fndng s true for both smple cross-sectons and for cross-sectons weghted by country sze. Keywords: Multnatonal corporatons, corrupton, property rghts, wage premum JEL classfcaton: F21, F23, J31, P14 Joshua Azenman (correspondence) Department of Economcs, UCSC, 217 Socal Scences 1 101 Market St. Santa Cruz, CA USA 95064 San Francsco CA 94105 Tel: (831) 459-4791 Tel: (415) 974-3241 Mark M. Spegel Federal Reserve Bank of San Francsco Fax: (831) 459-5900 Fax: (415) 974-2168 Emal jazen@cats.ucsc.edu Emal mark.spegel@sf.frb.org * Edmund Chang provded excellent research assstance. Helpful comments were receved from semnar partcpants at U.C. Santa Cruz and the Federal Reserve Bank of San Francsco. The vews expressed n ths paper represent those of the authors alone and do not necessarly reflect the vews of the Federal Reserve Board, the Federal Reserve Bank of San Francsco, or the Natonal Bureau of Economc Research.
1. Introducton The large ncrease n FDI n recent decades has stmulated a growng emprcal and theoretcal lterature. 1 The salent emprcal regulartes emergng from ths lterature nclude the fndng of a hefty multnatonal wage premum multnatonals wages exceed the wages pad by domestc producers by a sgnfcant margn, and multnatonals productvty tends to be hgher than that of domestc producers. 2 The purpose of ths paper s to outlne and to test a model that provdes an nterpretaton to these fndngs. Specfcally, we dentfy stuatons where t s n multnatonal s self nterest to pay a wage premum relatve to domestc producers. A number of prevous papers have concentrated on knowledge spllovers as an argument for a multnatonal wage premum. Fosfur, et al (2001) ntroduce a model where a multnatonal pays ts traned workers a hgher wage to nduce t to resst movng to a local compettor. Our analyss focuses nstead on the role of strength n the enforcement of property rghts, as measured by the domestc level of nsttutonal effcency, on the pattern and behavor of multnatonals. Despte efforts to lmt such behavor, corrupton and brbery appear to be prevalent features of foregn drect nvestment actvtes. For example, Hnes (1995) examnes the mpact of the Foregn Corrupt Practces Act of 1977 forbddng foregn brbery by Amercan frms on subsequent FDI growth n corrupt natons orgnatng n the Unted States. Hnes fnds that the law put US frms at a compettve dsadvantage n those states as growth n FDI orgnatng n the US n corrupt states was sgnfcantly lower than n non-corrupt states subsequent to the law s passage. A number of studes [e.g. Markusen (2001), We (1997a, b), and Smarzynska and We (2000)], have posted that nsttutonal neffcences such as corrupton wll be detrmental to both FDI and domestc nvestment. The emprcal evdence concernng the mpact of nsttutonal effcency and property rghts enforcement on nward nvestment has been mxed. Wheeler and Mody (1992) estmate a cross-country panel of manufacturng and electroncs nvestment n whch a prncpal component they label Rsk ncludes such soco-economc factors as the Busness Internatonal ndcators of corrupton and bureaucratc red tape. They fnd no sgnfcant 1 See Markusen (2002) and Feenstra (2002, Chapter 11) for overvew of multnatonals, and Lpsey (2002) for a revew of the emprcal evdence. 2 See Blomström (1983b), Haddad and Harrson (1993), Okamoto and Sjöholm (1999), Lpsey and Sjöholm (2001) and the references n Lpsey (2002). 1
mpact of ths component on captal expendtures by U.S. multnatonals. Smlarly, Hnes (1995) fnds no measurable mpact of corrupton on total nward FDI n host natons after 1977. However, more recent studes fnd robust evdence that corrupton reduces the level of FDI enterng nto a country. We (2000) examnes a panel of blateral stocks of FDI from 12 source countres to 45 host countres and fnds a large and statstcally sgnfcant negatve mpact of corrupton on nward FDI. Hs pont estmates ndcate that the ncrease n corrupton from the level of Sngapore s to that of Mexco s the equvalent of a 20 percentage pont ncrease n the tax rate on multnatonals. Smlarly, We (1997) fnds that uncertanty n corrupton levels also has a measurable negatve mpact on nward FDI. Whle these later studes establsh a negatve relatonshp between corrupton and FDI, ther results do not mply that FDI flows would be more senstve to host country corrupton levels than domestc nvestment. Domestc nvestment rates are also lkely to respond negatvely to corrupton levels. However, the possblty that corrupton s especally harmful to FDI,.e. relatve to ts adverse mpact on domestc nvestment, s mportant n terms of the general consensus that FDI plays an mportant role n transferrng technology to developng countres. 3 We conjecture that FDI wll be more senstve to nsttutonal neffcences than domestc nvestment. We post that domestc entrepreneurs wll have an advantage n overcomng nsttutonal neffcences relatve to ther foregn compettors n overcomng some of the obstacles assocated wth corrupton and weak nsttutons. Ths may be due to multtude of reasons, ncludng better famlarty of the court system and the government, better knowledge of the key people that should be brbed and of local networks that help n resolvng dsputes, etc. Our model focuses on the mplcatons of ths presumpton on the employment and nvestment patterns of domestc versus foregn entrepreneurs. 4 Specfcally, we model such crcumstances n a prncple agent framework wth costly ex-post montorng and enforcement of an ex-ante contract wth domestc labor. The home advantage stems from our assumpton that the ex-post montorng and 3 For example, see Barrel and Pan (1997). However, see Atken and Harrson (1999) for an opposng vew. 4 The lterature has dealt wth other possble dmensons assocated wth home advantages and dsadvantages of domestc versus foregn entrepreneurs. For example, Razn, Sadka and Yuen (1999) studed the mplcaton of multnatonals havng access to cheaper cost credt and possbly nferor nformaton about the qualty of domestc projects relatve to domestc entrepreneurs on the patterns of FDI. Our approach abstracts away from these ssues, assumng equal fnancal costs for both domestc and foregn agents. Ths allows us to dentfy the mplcatons of the home advantage assocated wth contract enforcement on the patterns of nvestment. An mplcaton of our assumptons s that, unlke n Razn et. al. (1999), FDI unambguously mproves the host county s welfare. 2
enforcement cost of the labor contract s lower for domestc entrepreneurs than for foregn ones. Under these dsadvantages, foregn producers requre a countervalng productvty advantage to compete. Gven crcumstances where both multnatonals and domestc producers exst sde-by-sde, we show that multnatonals pay hgher wages than domestc producers, n lne wth the nsght of effcency wages and wth the evdence about the multnatonals wage premum. We also show that multnatonal nvestments are more senstve to weakness (or more costly enforcement of) property rghts. We then drectly examne the mpact of nsttutonal effcency on the share of FDI n a host country s overall nvestment portfolo. In partcular, we estmate the mpact of an ndex of nsttutonal effcency on the rato of average FDI flows to both gross fxed captal formaton and prvate domestc nvestment over the followng ten years for a cross-secton of natons. We fnd that nsttutonal effcency s robustly postvely correlated wth the rato of FDI to total domestc nvestment. Ths suggests that nsttutonal neffcency dscourages FDI more severely than t does domestc nvestment, as predcted by our theoretcal model. We then demonstrate that ths result s robust to the ncluson of a number of condtonng factors. Ths paper s organzed nto fve sectons. Secton 2 ntroduces a smple prncpalagent model of foregn drect nvestment wth mperfect property rghts protecton. Secton 3 dscusses the emprcal methodology and data used n the paper. Secton 4 revews our results and conducts some robustness testng. Secton 5 concludes. 2. A Smple Model of FDI wth Imperfect Property Rght Enforcement In ths secton, we ntroduce a smple model of FDI wth mperfect property rght enforcement. We assume that there s a sector contanng two frms, a multnatonal subsdary and a domestc frm. Both of these frms are assumed to face a prncpal-agent problem vs-à-vs ther laborers, wth costly ex-post montorng and enforcement of an exante contract. FDI s assumed to co-exst wth domestc producton, where the technologcal superorty of foregn subsdares and the relatve superorty of domestc frms concernng the agency problem lead to an nteror soluton for the share of FDI n host-country nvestment. The producton functons of the domestc and foregn frms are assumed to be Cobb- Douglas n captal, K, and labor, L. We dstngush the foregn frm wth stars. The producton functon of the domestc frm s assumed to satsfy 3
Y = zak α L β (1.1) where z s the effectve productvty shock, the outcome of labor s effort and the realzed exogenous state of nature, ε : 1 + ε wth L effort z = χ(1 + ε) no L effort (1.2) Labor s effort therefore contrbutes 1 χ to output. Smlarly, the foregn frm producton functon s assumed to satsfy where * z satsfes * * * * * Y = z A K α L β (1.3) z * * 1+ ε = χ + * (1 ε ) wth L effort no L effort (1.4) * We assume that ε and ε are ndependently dstrbuted unform on the nterval ε, ε. We start the analyss wth the smplest benchmark by gnorng the possblty of random montorng and random shrkng. In the absence of spendng montorng and verfcaton costs, the representatve entrepreneur n the domestc and foregn sector observes only the effectve productvty shocks, z and respectvely. Verfcaton of labor effort can be done only ex-post, after the realzaton of output. The cost of verfyng labor s effort s * * * assumed to equal proportons c and c of the labor nputs, cl and cl respectvely. Snce the cost s lkely to be hghly correlated wthn a country, we assume that * z * c = ψ c. Moreover, we assume that the domestc frm enjoys a low cost of verfyng and enforcng effort, such that ψ > 1. However, we assume that the foregn subsdary enjoys a countervalng productvty advantage over ts domestc counterpart, so that A* A. The opportunty cost of labor s tme s assumed to equalω. 5 There are two possble labor types, dfferng n the amount of effort e needed to yeld the hgh output [alternatvely, two possble qualtes of matches between labor and captal, dfferng n the effort nput needed]: 5 The opportunty cost of labor could be alternatvely nterpreted as lesure or as the prevalng wage n a tradtonal sector. 4
e e = e a b wth probablty wth probablty a b (1.5) where e < e b and a+ b= 1. a Labor s utlty satsfes where C s labor consumpton and ω 0 [ ω ] U = C + e L (1.6) s the shadow prce of lesure. Wth perfect nformaton, labor s pad the sum of ( ω + e ) L ( = a, b) wth effort, and ω L wth no effort. Labor knows ts type, and ts effort decson s endogenous. The entrepreneurs observe only the effectve productvty shock [ ( 1+ ε ) χ or 1 + ε, dependng on labor s effort]. Ex-post, the entrepreneur may decde to pay the verfcaton and enforcement cost n order to reveal labor s effort. In the absence of verfcaton and enforcement, labor s compensaton s not contngent on effort. dmensons The labor contract sets the compensaton rule ex-ante. It has the followng - A threshold φ of the effectve productvty shock z that wll trgger the costly verfcaton and enforcement. - In the absence of verfcaton, or f the verfcaton wll reveal no shrkng, labor would be pad w. If shrkng s detected, labor would be pad zero. 6 n L We assume that the varous parameters nduce a separatng equlbrum, where the more effcent type (a) would supply effort, and the less effcent type (b) would shrk. In ratonal-expectaton equlbrum, labor would prefer puttng effort to shrkng f the penalty for shrkng exceeds the cost of effort. Under the assumptons above, ths condton satsfes and Henceforth, we assume that φ + ε w 2ε wn n > (1.7) e ω + e (1.8) 2ε ea eb wn e a ω + > φ ε ω + + (1.9) 6 Maxmzng the penalty assocated wth shrkng (.e., payng zero when shrkng s detected), s optmal. 5
Ths mples that the densty functon of effectve productvty shock z s b for (1 ε) χ < z < 1 ε 2εχ b a f( z) = + for 1 ε < z< (1 + ε) χ 2εχ 2ε a for (1 + ε) χ < z < 1+ ε 2ε (1.10) The decson problems faced by the domestc and foregn entrepreneurs are dentcal. The domestc entrepreneur sets the contract n order to maxmze the expected profts V, where where the cost of captal s equal to whch satsfes α β V = [1 (1 χ) b] AK L (1 + ρ) K LE( lc) (1.11) 1 + ρ and E( lc ) represents the expected cost per worker, φ φ b 1 E( lc) = wn 1 + ε dz c f ( z) dz (1.12) 2 χ ε χ 2ε ( 1 ) (1 ε ) χ The frst term on the RHS of equaton (1.11) represents expected output. The second term s the cost of captal; the thrd s the expected cost of labor. The cost of employng a worker, E(lc), takes nto account that the wage payment to shrkng labor wll be zero when the worker shrks, and that employng labor s assocated wth the expected cost of montorng and enforcement [the second term of E(lc) ]. 7 Henceforth we focus on the case where the entrepreneur pays labor the reservaton wage that just nduces laborers of type a to supply effort: w n 2ε ea = φ + ε (1.13) 7 Note that the support on the second term begns at ( 1 ε ) χ montorng and enforcement costs together. However, n the range ( 1 ε ) χ z ( 1 ε). For analytc smplcty, we combne the labor, the entrepreneur would optmally choose only to pay the enforcement costs and not montor, snce montorng s not needed to establsh that a worker s of type b n that range. Smlarly, wthn the range ( ε ) 1 z φ, the entrepreneur would always need to montor to establsh labor s type, but would only face enforcement costs when the laborer turned out to be of type b. Whle ths s clearly a smplfcaton, t drves none of the qualtatve results. We return to ths smplfcaton n the concluson. 6
Optmzng V wth respect to φ, K, and L, we nfer: CLAIM 1: An nternal separatng equlbrum (.e., where type b would shrk, and type a would supply the needed effort) s characterzed by and dl dk dφ dw < 0; < 0; < 0; n > 0 (1.14) dc dc dc dc CLAIM 2: The captal labor rato and the optmal nvestment levels depend negatvely on the expected cost of labor, E (lc). Proof: see the appendx. 8 Our results follow the logc of effcency wages. Hgher montorng costs would nduce lower ncdence of montorng and enforcement, leadng the entrepreneur to pay hgher wages. The net outcome s hgher wage, needed to keep the penalty assocated wth shrkng hgh enough despte the drop n the ncdence of montorng. A by-product of t s that nvestment and employment wll drop. Fnally, the level of montorng and enforcement costs wll affect the relatve levels of domestc and foregn nvestment, as noted n the followng clam: CLAIM 3: Hgher enforcement costs (mantanng constant the relatve cost dsadvantage of the foregn producer, ψ) reduce the rato of multnatonal nvestment to domestc nvestment at a rate that ncreases wth the enforcement cost gap. Proof: Denotng the optmal stock of captal n the domestc frm by K ~, and the probablty of enforcng and montorng n the domestc ndustry by Q, we demonstrate n the appendx that cq d / dc> 0 (1.15) Elc [ ] 8 The comparatve statc are smplfed consderably by the observaton that around the optmum " " " V V = V 0. K, φ = L, φ K, c = 7
and dlog[ K% */ K% ] β 1 cq c* Q* = < 0 (1.16) dc 1 α β c E[ lc] E[ lc* ] as predcted n Clam 3. The ntuton behnd Clam 3 s that hgher enforcement costs ncrease the rato of expected enforcement costs to total worker cost, whch s cq / E ( lc ) for the domestc frm and c* Q */ E( lc *) for the foregn frm. 9 The decrease n the rato of foregn to domestc nvestment resultng from an ncrease n enforcement costs wll then be proportonal to the dfference n the montorng and enforcement cost ratos of domestc and foregn producers. The observaton that the enforcement cost rato ncreases wth the level of enforcement cost mples that the greater s the cost gap, the larger s the drop of the relatve captal share nduced by a gven ncrease n the montorng and enforcement costs, c. Our model therefore predcts that multnatonals characterzed wth hgher productvty and hgher cost of montorng and enforcement wll opt to pay hgher wages. Moreover, the greater s the cost of domestc enforcement c, the lower wll be the rato of foregn drect nvestment to domestc nvestment. In the followng secton, we test the latter emprcal predcton. 3. Emprcs 3.1 Methodology The theoretcal model above mpled that foregn drect nvestment would consttute a smaller share of the overall nvestment package n countres that had nferor property rghts protecton. In ths secton, we test ths theory emprcally for a cross-secton of countres usng data on nsttutonal effcency. where We frst estmate the followng specfcaton FDI / GCFC FDI GFCF = α + β + β + β +ε (1.17) 1 2Dev 3 Ores & Metals represents the average rato of nward foregn drect nvestment to gross fxed captal formaton from 1990 through 1999, represents the ndex of nsttutonal effcency from Mauro (1995), dscussed n more detal below, Dev represents a zero-one 9 The hgher cost c nduces a relatvely small drop n the probablty of montorng and 8
dummy ndcatng a developed country, Ores & Metals represents the share of exports comprsed of ores and metals, and ε represents a dsturbance term that s assumed to be ndependently and dentcally dstrbuted normal. We estmate equaton (1.16) wth and wthout the Ores & Metals varable, whch s ntroduced to dentfy countres that are ntensve n actvtes tradtonally assocated wth hgh levels of foregn drect nvestment. 10 We also estmate equaton (1.16) wth and wthout weghtng our observatons by country sze, as measured by gross domestc product n 1989. Fnally, we report our results wth developed and developng countres pooled wth the dummy ncluded as well as the two samples separated. Our coeffcent of nterest s β 1, the mpact of the nsttutonal effcency ndex n 1989 on subsequent nward foregn drect nvestment as a share of gross fxed captal formaton. The model s estmated usng ordnary least squares wth Whte s heteroscedastcty-consstent standard errors below. Snce heterogenety n government nvestment may add nose to the denomnator n the dependent varable n our specfcaton above, we repeat our estmaton usng the rato of average nward foregn drect nvestment to prvate domestc nvestment, Dev, from 1990-1999. Ths specfcaton results n a smaller sample, but provdes a good check of the robustness of the results we report for the larger sample. FDI / PVT 3.2. Data Insttutonal effcency data was obtaned from Mauro (1995). The data are from Busness Internatonal s ndex of nsttutonal effcency, and reflect the reports of analysts concernng the functonng of the domestc bureaucracy, wth a grade of 10 ndcatng a smoothly functonng, effcent bureaucracy whle a grade of 4 ndcates constant need for government approvals and frequent delays. Remanng data, ncludng foregn drect nvestment flows, gross fxed captal formaton, the share of prvate nvestment n total domestc nvestment, and the shares of ores and metals n total exports, were obtaned from the World Development Indcators. cq enforcement, Q, such that the net effect s ncreasng. E[lc] 10 The ol ndustry s also commonly assocated wth hgh shares of foregn drect nvestment. Usng the set of ol-exportng countres dentfed n Mankw, Romer and Wel (1992), two of the countres n our data set, Gabon and Iran, can be consdered ol-exportng. We re-ran the regressons wth these two countres omtted and obtaned very smlar results. These are avalable upon request. 9
Countres were desgnated as developed on the bass of membershp n the OECD n November 1988. Summary statstcs are shown n Table 1. Our sample ncludes 97 countres, 76 of whch are desgnated as developng and 21 of whch are desgnated as developed. It can be seen that the ncluson of the Ores & Metals varable reduces our sample sze to 71, whle usng the FDI / PVT dependent varable nstead of FDI / GCFC reduces the sample sze to 52 countres, only 10 of whch are developed. Consequently, we do not report results for the developed countres alone wth ths dependent varable. n the Unsurprsngly, the developng natons score poorly relatve to the developed natons ndex, wth the developng natons mean ndex at 4.67 whle the developed natons ndex has a mean of 8.874. Nevertheless there s a far amount of dsparty wthn both samples, wth the developng natons rangng from 0 to 8.33 whle the developed natons range from 6.67 to 10. One mght expect that the developng natons would have a hgher share of nward foregn drect nvestment. However, the data show that that s not necessarly the case. In fact, the mean rato of foregn drect nvestment to gross fxed captal formaton s slghtly larger for the developed natons. In contrast, the mean rato of FDI to prvate domestc nvestment s larger for the developng natons, as we mght expect. Nevertheless, nether dfference s statstcally sgnfcant. The smple correlatons between our ndex and levels of nvestment relatve to gross domestc product for our developng naton sample are shown n Fgure 1. It can be seen that there s a modest postve raw relatonshp between property rghts protecton and both FDI and domestc nvestment as measured by gross fxed captal formaton. Ths confrms the results found n We (2000). 11 We plot the smple correlaton between the ndex and the FDI / GCFC and FDI / PVT ratos n Fgure 2 for our developng country sample. We observe a modest postve relatonshp between protecton of property rghts protecton and these ratos, as predcted by our theory. We next turn towards testng these hypotheses formally. 12 11 The full sample also dsplayed a modest postve relatonshp. However, as developed countres tend to have much lower corrupton scores, we nclude a dummy to dentfy the developed natons n our parametrc analyss wth the full sample below. 12 Fgures 1 and 2 reveal that there s a lot of clusterng n the nsttutonal effcency ratngs. In response, we also ran the regressons below wth estmators robust to clusterng and obtaned essentally dentcal results, These are avalable from the authors on request. 10
4. Results 4.1. Base Specfcaton Results wth FDI / GCFC as the dependent varable are shown n Table 2. It can be seen that the performance of the varable s very robust. Wth the sample weghted by country sze, the varable enters postvely and sgnfcantly ether wth or wthout controllng for the share of ores and metals exports. The pont estmate of slghtly over 0.02 mples an economcally sgnfcant 2 percent ncrease n the rato of FDI to gross fxed captal formaton for each pont ncrease n the corrupton ndex. Wth the unweghted sample, the varable enters postvely and sgnfcantly after controllng for ores and metals exports, but s nsgnfcant wthout ths control. However, we focus prmarly on the weghted results to avod results domnated by small outlers. The very small R-squared results we obtan n all of the unweghted regressons reported suggest that the weghted samples contan far less nose. The control varables enter as would be expected. The varable s negatve and statstcally sgnfcant for the weghted samples. The Ore s & Metals varable enters postvely and sgnfcantly at least a ten percent confdence level wth ether the weghted or un-weghted specfcatons. Dev We then break the sample up nto ts developed and developng naton sub-samples and obtan smlar results. For both sub-samples, the varable enters postvely and sgnfcantly for both specfcatons wth a weghted sample, and after controllng for ores and metal exports wth the un-weghted sample. There s a notable dfference n the pont estmates between the two sub-samples, although ths dfference s not statstcally sgnfcant. Table 3 dsplays the results wth as the dependent varable. As noted above, the use of ths varable sgnfcantly reduces our sample sze. Indeed, the data s avalable for ths reduced sample for only ten of the developed natons, so we do not report regresson results for that sub-sample wth ths dependent varable. 13 results for the FDI / PVT Nevertheless, our varable of nterest appear to be robust n the full sample. As was the case for the rato to gross fxed captal formaton, the varable enters postvely and 13 For completeness, we dd run the specfcaton wth ths sub-sample. Unsurprsngly, everythng was very nsgnfcant, ncludng all of the control varables. 11
sgnfcantly n both of the specfcatons wth the weghted sample and wth the un-weghted sample after controllng for the share of ores and metals exports. The coeffcent values are also qute smlar to those we obtaned n the gross fxed captal formaton regressons. The results wth the developed country sample alone are also smlar. The varable enters sgnfcantly n the weghted specfcaton wthout controllng for ores and metals, and s close to a 10 percent confdence level wth the control ncluded. As above, the varable also enters sgnfcantly wth ts expected postve coeffcent wth the Ores & Metals control ncluded. 4.2. Robustness to Incluson of Condtonng Varables Because we are estmatng a cross-secton, we obvously are precluded from usng panel estmators, such as country fxed and random effects, to control for dfferences n country characterstcs outsde of our theory that may ndependently nfluence the relatve share of FDI n nvestment. To account for other possble nfluences, we ntroduce a number of condtonng varables nto our specfcaton from the Sachs and Warner (1997) data set. 14 The condtonng varables ntroduced are Sub-Sahara, a dummy ndcatng Sub- Saharan Afrcan natons, Openness, an ndcator of the degree to whch domestc polcy favors free trade, Access, a dummy ndcatng a naton havng navgable access to the sea, Tropcs, a varable measurng the share of land area subject to a tropcal clmate, Lfe, the log of lfe expectancy at brth measured between 1965 and 1970, and a measure of Ethnolngustc fractonalzaton. The latter varable measures the probablty that two randomlyselected people from a country wll not belong to the same ethnc or lngustc group. Our results for the mpact of corrupton on the rato of FDI to gross fxed captal formaton s shown n Table 4. All specfcatons are weghted by country sze. As before, we estmate the full sample wth a dummy varable to ndcate developed natons. The frst column ntroduces all of the condtonng varables smultaneously. It can be seen that the Corrupton varable s nsgnfcant after condtonng for the shares of ores and metal exports, but s sgnfcant at a ten percent confdence level wth ths condtonng varable. Some exploraton revealed that the Ethno-lngustc fractonalzaton varable dsplayed a hgh nfluence, and therefore the second column repeats the specfcaton wth 14 See Sachs and Warner (1997) for orgnal data sources. 12
ths varable dropped. The Corrupton varable s now postve and sgnfcant at a fve percent level as before wth or wthout condtonng for ores and metals export shares. 15 Fnally, we ntroduce the condtonng varables one at a tme. It can be seen that the sgnfcance of the Corrupton varable s robust to the ncluson of any of the condtonng varables ndvdually, ether wth or wthout condtonng for the ores and metals share. Moreover, the pont estmate for the varable coeffcent s smlar to that of the weghted fullsample regresson above. On ther own, the condtonng varables do not appear to be sgnfcant, wth the excepton of the nfluental Ethno-lngustc fractonalzaton varable dscussed earler. Table 5 repeats all of the specfcatons for FDI as a share of prvate nvestment. The results are qute smlar to those n Table 4. Wth the excepton of the ncluson of all of the condtonng varables, the Corrupton varable s robustly postve. As before, f we exclude the Ethno-lngustc fractonalzaton varable, the Corrupton varable enters postvely and sgnfcantly (column 2). In summary, the Corrupton term appears to be strongly robust to the ncluson of the condtonng varables, ether for FDI as a share of gross-fxed-captal formaton or for FDI as a share of prvate nvestment. The only excepton was the specfcaton that ncluded all of the condtonng varables. When the Ethno-lngustc fractonalzaton varable was excluded, the Corrupton varable always entered postvely and sgnfcantly, as predcted. 16 15 We also found that the Lfe varable was nfluental. Its excluson resulted n the Corrupton varable enterng sgnfcantly even wth the Ethno-lngustc fractonalzaton varable retaned for FDI as a share of gross-fxed captal formaton. 16 The hgh nfluence of the ethno-lngustc fractonalzaton varable s puzzlng. The correlaton coeffcent between t and the Corrupton varable s only -0.28. Moreover, the ntroducton of the varable only results n the excluson of one observaton from the sample, Chna. However, excludng Chna whle also excludng the ethno-lngustc fractonalzaton varable also results n the Corrupton varable enterng sgnfcantly postve. 13
5. Concluson Ths paper ntroduced a model of foregn drect nvestment wth costly enforcement of property rghts. We demonstrated that when foregn drect nvestment suffered from a relatve dsadvantage n property rghts protecton, t economzed on ts physcal captal nvestment and pad ts laborers a hgher wage premum. Ths premum nduced a separatng equlbrum where the relatvely productve workers refraned from shrkng, whle the less productve workers shrked. Fnally, we demonstrated that the rato of multnatonal nvestment to domestc nvestment would be ncreasng n the securty of property rghts. We then tested ths predcton for a cross-secton of countres usng data on corrupton. Our results demonstrated a robust negatve relatonshp between the level of corrupton and the rato of FDI flows to domestc nvestment flows. It should be noted that a number of our smplfyng assumptons above do not drve our results. For example, our assumpton of a unform dstrbuton for the productvty shock resulted n a wdenng of the tals of the dstrbuton, where montorng s not requred to dentfy the worker s type. Usng a more standard dstrbuton, such as a normal, the probablty space where montorng was requred would be lkely to ncrease, and thereby ncrease the property rghts advantage of the domestc frm n a corrupt envronment. Another smplfcaton noted earler was the mplct combnaton of montorng and enforcement actvtes. Whle the domestc entrepreneur s lkely to enjoy advantages n both of these actvtes, as specfed above, one could magne a stuaton where relatve advantages n montorng may dffer by ndustry. Holdng enforcement costs equal, we may see multnatonal nvestment relatvely specalzed n ndustres n whch foregn frms enjoy relatve advantages n montorng costs. For example, multnatonals may enjoy manageral advantages n some ndustres, whch may correspond to reduced montorng costs, but may suffer from the enforcement dsadvantages alluded to above. Fnally, the wage premum result above came from the specfcaton of property rghts lmtatons concernng the enforcement of labor effort. However, one could easly magne a scenaro where the employment of captal also resulted n enforcement problems. In future work, we wll also allow for property rght lmtatons to arse n ths dmenson. 14
Appendx Ths Appendx summarzes the dervaton of the clams dscussed n the paper. 1. Proof of Clams 1 and 2 The frst order condtons correspondng to the entrepreneur s problem are: (A.1) Note that around the equlbrum ( ) dv E Y = β E lc = dl L ( ) ( ) 0 V E Y = α ( 1+ ρ) = 0 K K ( ) V E lc = L = 0 φ φ " E( lc) (A2) V L, φ = = 0, φ where the last equalty follows from the frst order condton (A1) determnng the threshold n order to mnmze the expected cost of employng a worker. In addton, note that " " " ",,,, = (A3) V K c = VK φ = Vφ L = Vφ K 0. It s easy to confrm that " " ",,, < (A4) V φ φ < 0; V φ c < 0; V L c 0. The comparatve statc of the system are determned by ", 0 " V K K VK, L 0 dk / dc " " (A5) " VL, K VL, L 0 dl / dc = VL, c. " " 0 0 V φ, φ dφ / dc Vφ, c It s easy to confrm that the second order condtons for maxmzaton hold, and the determnate of the system s negatve. 15
Hence, sgn 0 V " K, L " " " " " [ dk / dc] sgn V V 0 = sgnv [ V V ] < 0. = L, c L, L φ, φ K, L L, c " " Vφ, c 0 Vφ, φ 0 Applyng smlar methodology, we nfer that part A of proposton 1 follows from (A5) and (A4). Applyng the frst order condtons (A1), and the Cobb-Douglas output specfcaton (1), t follows that the optmal captal and labor levels, denoted by K ~ and L ~, s (A6) ~ Aα K = [1 (1 χ) b] (1 + ρ) 1 β 1 β 1 β [ E( lc)] 1/(1 α β ) (1 ) L% β + ρ = K% α Elc ( ) Note that, applyng the envelope theorem, φ 1 ε de( lc) E( lc) dφ E( lc) b 1 b (A7) = + = ( + ) + > 0 a dε 2 dε. dc φ dc c χ ε χ2ε 1 ε (1 ε ) χ Hence, hgher enforcement costs would ncrease the expected cost of employng labor, reducng thereby the optmal nvestment, hence dk ~ (A8) < 0. dc dl ~ Smlar analyss mples that < 0. dc Note that Clam 2 then follows drectly from equaton (A.6). 2. Proof of Clam 3 By equaton (A6) f follows that 16
(A9) Hence, gven that c (A10) or (A11) Note that * = ψ c ~ β /(1 α β ) K [ ( )] ~ * E lc = K [ E( lc*)] K% * d log K% β Q ψq* = dc 1 α β E ( lc) E ( lc* ) K% * d log K% β 1 cq c* Q * = dc 1 α β c E ( lc) E ( lc* ) (A12) cq d E ( ) Q dc E( lc) ( ) = > E lc cq 0 2 Applyng (A12) to (A11) we can nfer that K% * d log K% (A13) < 0. dc. 17
References Atken, Bran J., and Ann E. Harrson, (1999), " Do Domestc Frms Beneft from Drect Foregn Investment? Evdence from Venezuela " Amercan Economc Revew, June, 89(3), 605-618. Barrell, Ray and Ngel Pan, (1997), Foregn Drect Investment, Technologcal Change, and Economc Growth wthn Europe, Economc Journal, 107, November, 1770-1786. Blomström, Magnus, (1983), Foregn Investment and Spllovers, Routledge, London and New York. Feenstra (2002, Chapter 11), Advanced Internatonal Trade: Theory and Evdence, forthcomng, Prnceton Unversty Press, 2003. Fosfur, Andrea, Massmo Motta, and Thomas Rønde, Foregn Drect Investment and Spllovers Through Workers Moblty, Journal of Internatonal Economcs, 53, 205-222. Haddad, Mona, and Ann Harrson (1993), Are There Postve Spllovers from Drect Foregn Investment?, Journal of Development Economcs, Vol. 42, pp. 51-74. Hnes, James R., (1995), Forbdden Payment: Foregn Brbery and Amercan Busness After 1977, NBER Workng Paper no. 5266, September. Lpsey, Robert E. (2002), Home and Host Country Effects of FDI, manuscrpt, presented at The Challenges to Globalzaton, CEPR/NBER/SNS Conference, May. Lpsey, Robert E. and Fredrk Sjöholm (2001), Foregn Drect Investment and Wages n Indonesan Manufacturng, NBER Workng Paper No. 8299, Cambrdge, MA, Natonal Bureau of Economc Research. Mankw, N. Gregory, Romer, Davd, and Davd N. Wel, (1992), A Contrbuton to the Emprcs of Economc Growth, Quarterly Journal of Economcs, May, 107(2), 407-437. Markusen, James R. (2002), Multnatonal Frms and the Theory of Internatonal Trade, MIT Press. Markusen, James R. (2001), Contracts, Intellectual Property Rghts, and Multnatonal Investment n Developng Countres, Journal of Internatonal Economcs, 53, 189-204. Mauro, Paolo, (1995), Corrupton and Growth, Quarterly Journal of Economcs, August, 681-712. 18
Okamoto, Yumko, and Fredrck Sjöholm (1999), FDI and the Dynamcs of Productvty: Mcroeconomc Evdence, manuscrpt. Razn, Assaf, Efram Sadka, and Ch-Wa Yuen, (1999), Excessve FDI Flows Under Asymmetrc Informaton, NBER Workng Paper no. 7400. Smarzynska, Beata K. and We, Shang-Jn, (2000), Corrupton and Composton of Foregn Drect Investment: Frm-Level Evdence, NBER Workng Paper no. 7969, October. Evdence We, Shang-Jn, (1997), Why s Corrupton So Much More Taxng than Tax? Arbtrarness Klls, NBER Workng Paper no. 6255, November. (2000), How Taxng s Corrupton on Internatonal Investors?, Revew of Economcs and Statstcs, 82(1), 1-11. Wheeler, Davd, and Ashoka Mody (1992), Internatonal Investment Locaton Decsons: The Case of U.S. Frms, Journal of Internatonal Economcs, August, 33(1-2), 57-76. 19
Table 1. Summary Statstcs 1. Developng Natons Mean Mnmum Maxmum Standard # of Devaton Countres 4.6684 0 8.3333 1.7595 76 FDI / GFCF 0.1049-0.0930 0.3811 0.0957 76 FDI / PVT 0.1861-0.0019 0.7056 0.1613 42 Ores & Metals 10.178 0.03 61.18 16.0499 50 2. Developed Natons Mean Mnmum Maxmum Standard # of Devaton Countres 8.8794 6.6667 10 1.0998 21 FDI / GFCF 0.1107 0.0021 0.3347 0.0900 21 FDI / PVT 0.1262 0.0027 0.3833 0.1047 10 Ores & Metals 4.8495 0.95 16.98 4.0513 21 Note: Corrupt s corrupton ndex from Mauro (2000). Note that ndex s decreasng n domestc corrupton level. FDI / GFCF represents average rato of nward foregn drect nvestment to gross fxed captal formaton, whle FDI / PVT represents rato of nward foregn drect nvestment to prvate domestc nvestment. Corr upt and Ore s & Metals values are for 1989. FDI / GFCF and FDI / PVT values are averages from 1990-1999. 20
TABLE 2. Impact of Corrupton on FDI/GFCF. I. Full Sample Weghted Unweghted α -0.029-0.052 0.071** 0.030 (0.038) (0.041) (0.029) (0.022) 0.023** 0.022** 0.007 0.014** (0.007) (0.008) (0.005) (0.005) Developed -0.098** -0.078** -0.025-0.049 (0.029) (0.029) (0.030) (0.033) Ores & Metals 0.004** 0.001* (0.002) (0.001) # of obs 97 71 97 71 R-squared 0.15 0.22 0.02 0.09 II. Developed Natons Weghted Unweghted α -0.207* -0.179-0.104-0.110 (0.108) (0.116) (0.130) (0.130) 0.032** 0.026* 0.024 0.028* (0.013) (0.015) (0.015) (0.015) Ores & Metals 0.007-0.005 (0.006) (0.004) # of obs 21 21 21 21 R-squared 0.16 0.23 0.09 0.14 III. Developng Natons Weghted Unweghted α 0.019 0.001 0.079** 0.035 (0.024) (0.029) (0.031) (0.031) 0.013** 0.014** 0.006 0.013** (0.004) (0.005) (0.006) (0.005) Ores & Metals 0.002** 0.001* (0.001) (0.001) # of obs 76 50 76 50 R-squared 0.13 0.25 0.01 0.11 Note: Estmaton by ordnary least squares. Whte s heteroskedastcty-adjusted standard errors n parentheses. ** ndcates sgnfcance at a 5 percent confdence level. * ndcates sgnfcance at a ten percent level. 21
TABLE 3. Impact of Corrupton on FDI/PVT. I. Full Sample Weghted Unweghted α -0.004-0.017 0.152* 0.077** (0.054) (0.058) (0.077) (0.036) 0.028** 0.025* 0.008 0.014* (0.012) (0.014) (0.013) (0.008) Developed -0.180** -0.157* -0.094-0.098 (0.078) (0.087) (0.063) (0.060) Ores & Metals 0.003* 0.003 (0.002) (0.002) # of obs 52 43 52 43 R-squared 0.26 0.31 0.03 0.18 II. Developng Natons Weghted Unweghted α 0.003-0.008 0.147* 0.066* (0.056) (0.067) (0.078) (0.035) 0.026** 0.026 0.009 0.016* (0.013) (0.016) (0.014) (0.008) Ores & Metals 0.002 0.003 (0.002) (0.002) # of obs 42 33 42 33 R-squared 0.18 0.20 0.01 0.21 Note: Estmaton by ordnary least squares. Whte s heteroskedastcty-adjusted standard errors n parentheses. ** ndcates sgnfcance at a 5 percent confdence nterval. * ndcates sgnfcance at a ten percent confdence level. 22
Table 4. Impact of Corrupton on FDI/GFCF (Condtonng Varables Included). Wth Ores and Metals Constant -1.871** (0.725) -1.269* (0.710) -0.052 (0.042) -0.061 (0.048) -0.058 (0.043) -0.088* (0.051) -0.884 (0.572) -0.074* (0.042) Corrupton 0.013 0.019** 0.021** 0.022** 0.022** 0.023** 0.017** 0.021** (0.009) (0.009) (0.007) (0.008) (0.007) (0.007) (0.008) (0.007) Developed -0.124* -0.122* -0.068** -0.084-0.071** -0.048-0.119** -0.066** (0.063) (0.066) (0.031) (0.052) (0.031) (0.035) (0.046) (0.033) Ores and Metals 0.003 0.004** 0.004** 0.004** 0.004** 0.003** 0.004** 0.003* (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) Sub-Saharan -0.022-0.061 0.007 Afrca (0.172) (0.181) (0.151) Openness -0.006-0.034 0.019 (0.056) (0.058) (0.049) Access -0.020-0.026-0.027 (0.044) (0.047) (0.044) Tropcs 0.075 0.064 0.062 (0.053) (0.053) (0.049) Lfe 0.454** 0.313 0.216 (0.190) (0.188) (0.148) Ethno-lngustc 0.001** 0.001** Fractonalzaton (0.000) (0.000) Inflaton -0.000-0.001 (0.001) (0.001) # of obs 60 61 67 64 67 66 66 66 63 R-squared 0.38 0.29 0.21 0.21 0.21 0.23 0.24 0.27 0.22-0.046 (0.045) 0.022** (0.007) -0.079** (0.035) 0.004** (0.002) -0.001 (0.001) Wthout Ores and Metals Constant -1.820** (0.651) -1.140* (0.651) -0.032 (0.037) -0.033 (0.041) -0.035 (0.037) -0.078* (0.044) -0.842* (0.492) -0.064* (0.037) Corrupton 0.014* 0.019** 0.022** 0.022** 0.023** 0.024** 0.018** 0.023** (0.008) (0.008) (0.006) (0.007) (0.007) (0.006) (0.007) (0.006) Developed -0.119** -0.110* -0.087** -0.093** -0.091** -0.058* -0.137** -0.085** (0.057) (0.061) (0.027) (0.046) (0.027) (0.030) (0.040) (0.027) Sub-Saharan 0.054 0.024 0.034 Afrca (0.100) (0.106) (0.085) Openness -0.005-0.036 0.006 (0.051) (0.053) (0.044) Access -0.019-0.023-0.024 (0.040) (0.043) (0.040) Tropcs 0.083* 0.078 0.080* (0.048) (0.048) (0.042) Lfe 0.440** 0.281 0.211 (0.170) (0.172) (0.128) Ethno-lngustc 0.001** 0.001** Fractonalzaton (0.000) (0.000) Inflaton 0.000-0.001 (0.001) (0.001) # of obs 74 76 87 84 87 86 85 85 79 R-squared 0.34 0.21 0.14 0.14 0.14 0.17 0.16 0.21 0.14 Note: Estmaton by weghted least squares, wth weghts by GDP. Whte s heteroskedastcty-adjusted standard errors n parentheses. See text for condtonng varable defntons. ** ndcates sgnfcance at a 5 percent confdence nterval. * ndcates sgnfcance at a ten percent confdence level. -0.032 (0.040) 0.022** (0.007) -0.085** (0.033) 0.000 (0.001) 23
Table 5. Impact of Corrupton on FDI/PVT (Condtonng Varables Included). Wth Ores and Metals Constant -3.117** (0.979) -0.762 (1.077) -0.023 (0.071) -0.006 (0.081) -0.026 (0.072) -0.048 (0.086) 0.128 (0.923) -0.032 (0.064) Corrupton -0.011 0.028* 0.025* 0.024* 0.026* 0.027* 0.026* 0.015 (0.015) (0.016) (0.013) (0.014) (0.013) (0.014) (0.014) (0.012) Developed -0.130-0.173-0.150** -0.120-0.151** -0.141** -0.141* -0.070 (0.083) (0.105) (0.055) (0.086) (0.055) (0.058) (0.079) (0.054) Ores and Metals 0.002 0.005* 0.003 0.003 0.004 0.003 0.004 0.004* (0.002) (0.003) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) Sub-Saharan 0.046-0.123-0.069 Afrca (0.298) (0.379) (0.371) Openness -0.023-0.084-0.032 (0.065) (0.082) (0.072) Access 0.029-0.030-0.024 (0.073) (0.092) (0.090) Tropcs 0.074 0.061 0.037 (0.062) (0.074) (0.072) Lfe 0.797** 0.196-0.039 (0.257) (0.286) (0.240) Ethno-lngustc 0.001** 0.001** Fractonalzaton (0.001) (0.000) Inflaton 0.000-0.002* (0.001) (0.001) # of obs 40 41 41 41 41 41 41 40 41 R-squared 0.53 0.38 0.29 0.30 0.29 0.30 0.29 0.33 0.34-0.016 (0.069) 0.030** (0.013) -0.196** (0.061) 0.005** (0.002) -0.002 (0.001) Wthout Ores and Metals Constant -3.235** (0.919) -0.721 (1.043) -0.011 (0.065) 0.007 (0.072) -0.013 (0.066) -0.053 (0.076) -0.182 (0.849) -0.034 (0.063) -0.002 (0.065) Corrupton -0.011 (0.014) 0.029* (0.015) 0.028** (0.012) 0.026** (0.012) 0.028** (0.012) 0.030** (0.012) 0.027** (0.013) 0.020* (0.011) 0.031** (0.012) Developed -0.130 (0.078) -0.169 (0.101) -0.173** (0.049) -0.138* (0.080) -0.174** (0.050) -0.153** (0.053) -0.183** (0.070) -0.102** (0.049) -0.201** (0.059) Sub-Saharan Afrca 0.117 (0.217) -0.069 (0.281) -0.016 (0.266) Openness -0.024 (0.062) -0.086 (0.078) -0.037 (0.067) Access 0.034 (0.069) -0.023 (0.090) -0.021 (0.086) Tropcs 0.077 (0.058) 0.077 (0.070) 0.065 (0.064) Lfe 0.827** (0.242) 0.186 (0.277) 0.045 (0.221) Ethno-lngustc Fractonalzaton 0.002** (0.000) 0.001** (0.000) Inflaton 0.001 (0.001) -0.001 (0.001) -0.001 (0.001) # of obs 44 46 47 47 47 47 47 45 46 R-squared 0.51 0.31 0.24 0.24 0.24 0.26 0.24 0.25 0.25 Note: Estmaton by weghted least squares, wth weghts by GDP. Whte s heteroskedastcty-adjusted standard errors n parentheses. See text for condtonng varable defntons. ** ndcates sgnfcance at a 5 percent confdence nterval. * ndcates sgnfcance at a ten percent confdence level. 24
Fgure 1. Investment Flows and Insttutonal Effcency. 0.7 0.6 0.5 0.4 0.3 FDI/GDP0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 10-0.1 INST. EFFICIENCY 0.7 0.6 0.5 0.4 0.3 GFCF/GDP 0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 10-0.1 INST. EFFICIENCY Note: Developng country sample. Insttutonal effcency ndex based on a scale of 1 to 10, wth 10 representng lowest level of nsttutonal effcency. Investment data are averages of flows from 1990-1999. GFCF represents gross fxed captal formaton. 25
Fgure 2. FDI Ratos and Insttutonal Effcency. 0.7 ( 0.6 0.5 0.4 FDI/GFCF 0.3 0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 10-0.1 INST. EFFICIENCY 0.7 0.6 0.5 0.4 FDI/PVT 0.3 0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 10-0.1 INST. EFFICIENCY Note: Developng country sample. Insttutonal effcency ndex based on a scale of 1 to 10, wth 10 representng lowest level of nsttutonal effcency. Investment data are averages of flows from 1990-1999. GFCF represents gross fxed captal formaton. PVT represents prvate nvestment flows. See text for detals. 26