Spillovers, foreign investment, and export behavior

INTEMATIONAL ECONOMICS Journal of International Economics 43 (1997) 103-132

EL‘SVIER

Spillovers, foreign investment, and export behavior Brian Aitken”,

Gordon H. Hansonb’d.*, Ann E. HarrisonCSd

“International Monetary Fund, 700 19th Street, NW, Washington, DC, USA hDepartment of Economics, University of Texas, Austin, TX 78712. USA ‘Columbia Business School, 615 Uris Hall, New York, NY 10027, USA ‘NBER, Cambridge, MA, USA

Abstract Case studies of export behavior suggest that firms that penetrate foreign markets reduce entry costs for other potential exporters, either through learning effects or establishing commercial linkages. We examine whether spillovers associated with one firm’s export activity reduce the cost of exporting for other firms. We identify two sources of spillovers: export production in general and the specific activities of multinationals. From a simple model of export behavior we derive a probit specification for the probability that a firm exports. Using panel data on Mexican manufacturing plants, we find evidence of spillovers from multinational enterprises but not from general export activity. Keywords:

Multinational

enterprises; Geographic concentration of industry; Export pro-

duction JEL class$cation:

F14; F23

1. Introduction What explains the export behavior of firms? Trade theory has little to say about which firms in a country export and which produce for the domestic market, and there has been little formal empirical work on the subject.’ Anecdotal evidence, mostly derived from case studies in developing countries, suggests that the process *Corresponding author. Tel.: 5 17-475-8522;email: [email protected]+ utexas.edu ‘Two exceptions are Roberts and Tybout (1993) and Bernard and Jensen (1995). both of which examine the effect of sunk costs on the export-supply decision. 0022-1996/97/$17.00 0 1997 Elsevier Science B.V. All rights reserved PII SOO22- 1996(96)01464-X

104

B. Aitken

et al. I Journal

of International

Economics

43 (1997)

103-132

of breaking into foreign markets is far from trivial (Morawetz, 1981; Kessing, 1983). One of the most spectacular case studies of export growth, the development of garment exporters in Bangladesh, suggests that informational externalities are likely to be very important. The entry of one Korean garment exporter in Bangladesh lead to the establishment of hundreds of exporting enterprises, all owned by local entrepreneurs. Garment exports, which accounted for a negligible percentage of total export earnings, became the single largest source of foreign exchange earnings after the entry of one multinational firm. The idea we pursue in this paper is that there are localized spillovers associated with exporting by one firm that reduce the cost of foreign market access for nearby firms. We consider two versions of the export-spillovers hypothesis: (1) all export activity generates spillovers, and (2) only the export activity of multinational enterprises generates spillovers. Spillovers may take a variety of forms. The geographic concentration of exporters may make it feasible to construct specialized transportation infrastructure, such as storage facilities or rail lines, or may improve access to information about which goods are popular among foreign consumers. Whatever their source, spillovers imply that there are location-specific external economies associated with exporting. This means that firms are more likely to export the larger is the local concentration of export activity. The potential for spillovers from multinational enterprises (MNEs) derives from the fact that foreign firms have a multi-market presence. MNEs are a natural conduit for information about foreign markets, foreign consumers, and foreign technology, and they provide channels through which domestic firms can distribute their goods. To the extent that MNEs directly or indirectly provide information and distribution services, their activities enhance the export prospects of local firms. There is a growing body of empirical work that claims to find evidence consistent with spillovers.’ A standard approach for identifying the effects of spillovers on the production decisions of individual firms is to use a measure of aggregate local industrial activity, such as employment or value added, as an independent variable. The reasoning is that the larger is the local concentration of economic activity, the larger are the benefits from spillovers. Carlton (1983) finds that the geographic concentration of firms in an industry makes a location more attractive to entering firms; Nakamura (1985) and Henderson (1986) find that geographic concentration is positively correlated with total factor productivity; and Wheeler and Mody (1992) and Head and Ries (1994) find the local stock of foreign investment is a positive factor in the location decision of MNEs. These findings are interpreted as evidence of spillovers, but it is just as plausible to interpret them as evidence of unobserved fixed factors that firms value. The ?he externalities we take as our focus are static in nature. For empirical literature on dynamic externalities see Glaeser et al. (1992) and Henderson (1995).

B. Aitken et al. I Journal of International

Economics 43 (1997) 103-132

105

confusion is due to a basic identification problem: geographic concentration arising from location-specific external economies is observationally equivalent to geographic concentration arising from exogenous site-specific characteristics. One contribution of our approach is that we address the identification problem by decomposing the geographic concentration of industry into different elements. We are able to independently measure for each location the overall concentration of economic activity, the concentration of general export activity, and the concentration of MNE export activity. We control for unobserved fixed factors that might affect the export behavior of firms by controlling for the overall concentration of activity in a location. If the local concentration of export activity is a byproduct of overall industry concentration, then we will find that local export activity and local MNE activity do not affect the firm export decision. If, on the other hand, there are spillovers specific to exporting or to MNEs, then we will find that the local activities of other exporters or of MNEs positively influence the firm export decision3 A further contribution of our approach is that we treat local export activity and local MNE activity as endogenous variables. In this way we control for the existence of region-industry specific shocks that make all firms in a particular region and industry more likely to export. The role of MNEs in developing countries has received a great deal of attention. Most recent literature considers MNEs as a direct or indirect source of foreign technology. The proposed benefits of foreign direct investment (FDI) include technological diffusion (Davies, 1977; Teece, 1977), worker training, and the expansion of producer services (Rivera-Batiz and Rivera-Batiz, 1991). Empirical studies that use aggregate data (Caves, 1974; Blomstrom and Persson, 1983; Blomstrom, 1986) find a positive correlation between sectoral productivity and the sectoral level of FDI; but studies that use plant-level data (Aitken and Harrison, 1992; Haddad and Harrison, 1993) find either a negative correlation or no relationship between the presence of MNEs and the productivity of domesticallyowned manufacturing plants. The hypothesis we test is that MNEs act as export catalysts. Case studies suggest that the export activities of MNEs often produce externalities that enhance the export prospects of domestic firms (Rhee and Belot, 1990). By providing inputs not available in the local market, MNEs link local firms to foreign buyers. MNEs may internalize some of the benefits of their activities through subcontracting or licensing arrangements, though other effects, such as those related to demonstration or expanded cross-border trade, are more difficult to internalize and may spill over onto neighboring firms. There is much anecdotal evidence about spillovers from MNEs, but there has been no study which formally examines the role of foreign firms as export catalysts or assessesthe importance of this role in the economy at large. While we present results on the export decision for all firms, ‘Our approach is similar in spirit to that of Head et al. (1995). who study the location decision of Japanese multinationals.

106

B. Aitken et al. I Journal of International

Economics 43 (1997) IO3- I32

both domestic and foreign owned, we focus on the effects of MNE activity on domestic firms. The basis for our study is panel data on 2104 Mexican manufacturing plants. The sample covers the period 1986-1990, which follows Mexico’s liberalization of trade in 1985.4 We observe the period when many Mexican manufacturers first began to insert themselves into foreign markets. We develop a simple model of the firm production decision and derive a reduced form for the probability that a firm exports. If there are spillovers associated with exporting, a firm will be more likely to export the larger is the local concentration of export activity; if there are spillovers associated with the specific activities of MNEs, a firm will be more likely to export the larger is the local concentration of exports by MNEs.

2. An empirical model 2.1. Theory

In the production decision we model, firms choose whether to serve the domestic market, the foreign market, or both. Although firms have the option of exporting to multiple markets, for simplicity we treat foreign markets as a single entity. We assume that the distribution costs of producing for the domestic market differ from those of producing for the foreign market. For a representative firm, total costs are given by (1) where d indexes the domestic market, f indexes the foreign market, h() is the production cost function, mi() is the distribution cost function for market i, and h() and mi() are increasing and convex in their arguments. Separating production and distribution captures the idea that some costs, such as those related to design, advertising, and transportation, are market specific. The central hypothesis in our empirical work is that market-specific costs in exporting are a decreasing function of the local concentration of export activity. Very simply, we suppose that exporters benefit from being near other exporters. Let 4, be total export activity and r,,, be the export activity of MNEs in the regional industry to which a firm belongs. Positive export spillovers imply that for the representative firm, a%qf)

ar,,


- 7

(2)

4Trade liberalization in Mexico ended four decades of import-substitution industrialization. The country initiated reform in 1985 and by 1988 had greatly reduced most trade barriers.

B. Aitken et al. I Journal of International

Economics 43 (1997) 103-132

am&qf1 arMNE5 cl

107

(3)

We leave to Section IV the discussion of how we measure r,, and r,,,. For empirical tractability, we assume the following simple functional forms for the production and distribution cost functions: b

h(q,+qf)=Y(q,+q,)*+R(q,+qf),

m,(ch)=+qffc,qiy

where a, g, b,, and ci (i = d, f) are scalar parameters.5 We assume that g and c, are functions of cost variables the firm takes as given in making its output decision: g = g(x), cd = c,(x, z,),

C, = qx

.q r,,, r,,,),

where X represents cost variables that are common to production for both markets, and Z, represents cost variables that are specific to production for market i. This specification implies that export spillovers do not affect the costs of serving the domestic market. The production decision for the representative firm is given by the solution to max yd,yfp,q, + pfsf - Mq, + 4,) - m,(q,) - m,(s,)

s.t. qd3qf 2 0.

(4)

In theory, the optimal choice of output may be zero in either market. In practice, some firms in our sample produce zero exports, but all firms produce positive quantities for the domestic market. Accordingly, we only consider the possibility of a corner solution for the variable q,. To do so, we define the latent variable 47, such that ifq,>O

4*f = q,j * =() 1 qi

Using (5), we derive the following conditions to (4):

q, = &(Pdd sl” = &

(5)

OIW

f

system of equations from the first-order

- %I::- g(X)- c,,K Z,,>)?

(61

(pf- aq, - g(x) - +x9 zf, I;,, r,,,)).

‘We treat the firm output decision as static. It is conceivable that there are fixed costs to exporting. Roberts and Tybout (1993) find evidence of large sunk costs in the export-supply decision of Columbian manufacturing plants; Bernard and Jensen (1995) find evidence of small sunk costs in the export-supply decision of U.S. manufacturing plants. Given the time series in our data is two years, such dynamic considerations are beyond the scope of this paper.

B. Aitken et al. I Journal of International

108

Economics 43 (1997) 103-132

For purposes of estimation, we assume that Eqs. (6) and (7) can be rewritten as qdj = ff,Pd + a*q;

+ ffiZdj + a:xj

+ Udj’

(8)

where j indexes the firm, Zij is a 1 X K vector of cost variables specific to market i, Xj is a 1 X J vector of cost variables common to both markets, cy3and & are 1 X K vectors of coefficients, CY~and p, are 1 X J vectors of coefficients, and uij is a normally distributed error term for market i and firm j, which has mean zero and variance ~2. From Eqs. (6) and (7), az and /?, are negative. Eqs. (8) and (9) represent a simultaneous-equation model with a censored variable. Given our interest is in the firm export decision, we choose to focus the estimation on the probability that a firm exports. By estimating the dummy variable yj,

I

yj = 1 yj = 0

ifq,>O OIW

(10)

which indicates whether or not a firm has positive exports, we obtain consistent estimates of the parameters in Eqs. (8) and (9). One advantage of this approach is that it is straightforward to address certain estimation issues, such as the endogeneity of regressors. It follows from Eqs. (8)-( 10) that the probability firm j has positive exports is given by '
= l) = Pr[P(Pf + P~
P*Gij

+

&'MNEj

+

aiZdj) 'j

+ PJzfi +
P:>'j

(11)

where vj = &udj + ufi. From (1 l), the probability of exporting is a function of domestic and foreign final-goods prices, cost variables common to both markets (Xi), market-specific cost variables (Zdj and Z,), and the local concentration of export activity (r, and r,,,). Eqs. (2) and (3) imply Pr(yj= 1) is increasing in L and he- The distributional assumptions on udj and ufi imply vj is normally distributed, which permits estimation of Eq. (11) as a binary probit. 2.2. Estimation

issues

Several estimation issues merit further discussion. The first is whether we can identify the effects of spillovers on the export decision in Eq. (11). Suppose either that (i) all economic activity generates spillovers, or that (ii) spillovers associated with exporting also reduce domestic production and distribution costs. In the case of (i), firms benefit from being near other firms in their industry, regardless of which markets these firms serve. If I& and r,,, are positively correlated with the local concentration of economic activity, then the coefficients p, and & will

B. Aitken

et al. I Journal

of International

Economics

43 (1997)

103-132

109

capture the effects of general spillovers and not the effects of export spillovers. To control for this possibility, we include a measure of the geographic concentration of overall economic activity in the region as a regressor. In the case of (ii), export spillovers reduce costs for both foreign and domestic markets; the “true” expression for qd in Eq. (8) contains the additional term, (ygl&. + a6rMNE, and the “true” expression for export decision in Eq. (11) contains the additional term, p2(~5&+ c+,TMNE).The coefficient on I&,,, becomes & +/$LY~. If there are positive spillovers from MNE export activity, cr, and & will be positive, but, since p, is negative, the estimated coefficient on r,,,,, will be positive only if & > &a6, which holds as long as the cost-reducing effects of MNE activity on export production exceed the cost-reducing effects of MNE activity on domestic production by a sufficient amount. There is an analogous condition for r,,. A second issue is the possibility that the variables, I& and rMNE, are determined simultaneously with the export decision of individual firms. This will be the case, for instance, if output-supply shocks are common to firms that share a given industry and region, or if MNEs are attracted to regions and industries that offer favorable conditions for exporting. The latter point suggests that we should exercise caution in interpreting the estimation results on Eq. (11) for foreign firms. For this reason, we emphasize results on the export decision of domestic firms. To address the simultaneity issue, we treat r,, and I& as endogenous variables. Let TExj be the level of local export concentration that firm j perceives and LvEi be the level of local MNE export concentration that firm j perceivesP We specify r,, and rMNEj as (12) (13)

where 7, and wj are normally distributed random errors with mean 0 and variances ~2, and u;, respectively, and 3. is a 1xH vector of variables that includes the vectors X,, Zf,, and Zdi, plus additional variables that do not appear in (11). Define cov(~~, vj) = a7,, cov(w,, v~)= a-,, and cov(qj, wj) = a,,“. To estimate Eqs. (1 l13), we use the two-stage conditional maximum likelihood estimator (2SCML) proposed by Rivers and Vuong (1993). This estimator produces strongly consistent coefficient estimates, as do other limited-information estimators. The advantage of 2SCML is that it has desirable small-sample properties and is well-suited for tests of exogeneity. As a first stage, OLS is performed on Eqs. (12) and (13). The residuals from these regressions are then included as regressors in probit estimation of Eq. (11). Under the null hypothesis of exogeneity (07,, =a;, =O), the coefficient estimates on the OLS residuals in the probit equation equal zero. Hence ‘These measures vary across firms in our data since they are based on the aggregate activity of all other firms that occupy a given industry in a given region (see Section 1V.A).

110

B. Aitken et al. I Journal of International

Economics 43 (1997) 103-132

we can perform an exogeneity test on the variables r,, conditional likelihood ratio test.’

and rMNEj using a

3. The data 3.1. Data sources Data are available from two sources. We have annual data on 2104 Mexican manufacturing plants for the period 1986 to 1990 from the Secretariat of Trade and Industrial Promotion (Secofi).’ The variables include factor usage, domestic and foreign shipments, equity ownership positions by country of origin, price indices on inputs and outputs, four-digit industry classification, and the state in which the plant is located. Mexico contains 32 states and the plants are drawn from 129 four-digit (MEXSIC) industriesP Since the focus of our empirical work is the relationship between spillovers and exports, we need to reliably measure the geographic concentration of economic activity in Mexico. To do so, we supplement the Secofi data with data from the 1986 and 1989 Mexico Industrial Census. The Census data include employment, number of establishments, and average wages by state and four-digit (ISIC) industry.” Combining the data sets requires us to limit our study to the two years the Secofi sample overlaps with the Census, 1986 and 1989. Table 1 shows total manufacturing employment in the Census and in the Secofi sample. The plants in the Secofi sample account for a large share of Mexican manufacturing activity; in 1989, they employ 29 percent of all manufacturing workers. Plants in the Secofi sample are considerably larger than the plants in the

‘A further issue is whether there are unobserved firm-specific characteristics that affect the export decision. Butler and Moffit (1982) discuss a Hermite integration formula that can be used to estimate probit models in the presence of random effects. There are two disadvantages to this approach. The first is that Guilkey and Murphy (1993) find that in small samples with time series of length two, which corresponds to our case, numerical integration techniques offer little improvement over standard probit estimation. The second disadvantage is that random-effects estimation requires the assumption that the regressors are uncorrelated with firm random effects. There is evidence that in our data this assumption is unwarranted. In unreported results, we performed random-effects estimation of Eq. (I 1) as a linear probability model. Using a Hausman specification test, we reject the null hypothesis of no correlation between the regressors and firm random effects in all regressions. While there are well-known problems with the linear probability model, this is suggestive that random-effects estimation is inappropriate for our data. ‘The Secofi sample is limited to plants that survive the sample period. There are unfortunately no data on the exit or entry of manufacturing plants in Mexico. A further limitation of the data is that information is available on plants but not on the firms to which the plants pertain. ‘The Secofi data use an industrial classification code that is unique to Mexico. The code, which we call MEXSIC, is a close cousin of the SIC code. ‘“The Mexico Industrial Census uses the United Nations Revision II ISIC code.

B. Aitken et al. I Journal of International

111

Economics 43 (1997) 103-132

Table 1 Total Employment Year

Mexico Industrial Census

Secofi Sample of Plants

1986 1989

2 529 121 2 472 826

682 336 702 786

Census; in 1986, average workers per establishment is 318 in the Secofi sample and 67 in the Census. Table 2 shows that the distribution of employment across two-digit (ISIC) industries in the Secofi sample and in the Census are very similar. The Secofi sample does, however, somewhat overestimate employment in the states that contain Mexico’s largest cities and underestimate employment in border states. Two factors likely account for this fact. The Secofi sample, by design, targets medium and large plants, which happen to be located in Mexico’s largest cities, and excludes in-bond foreign assembly plants (such as plants that produce 8061807 exports), known as maquiladoras, which are concentrated in export processing zones located in border states.” In order to study the effect of the local concentration of export activity on the probability a plant exports, we need measures by industry of the geographic concentration of exports, MNE activity, and overall economic activity. We use the Census to estimate the geographic distribution of overall manufacturing activity in Mexico; this is appropriate as the Census is comprehensive. Since the Census does not contain information on export activity or on FDI, we use the Secofi sample to Table 2 Industry Employment Shares 1986 Industry

31 32 33 34 35 36 37 38 39

Mexico Industrial Census

Secofi Sample of Plants

Number of Establishments

Share of Employment

Number of Establishments

Share of Employment

46 866 15 794 15 275 6793 4688 9320 1015 28 417 1237

0.20 I 0.153 0.049 0.05 1 0.158 0.054 0.047 0.278 0.010

395 305 67 167 480 149 62 474 27

0.204 0.137 0.015 0.056 0.181 0.066 0.046 0.289 0.007

’ ‘We regard this feature of the data as a virtue, given the hypotheses we test. Maquiladoras import all inputs from parent firms abroad and export all output (until 1988, they were required to do so by law). Clearly, the export decision we posit does not characterize maquiladoras.

112

B. Aitken et al. I Journal of International

Economics 43 (1997) IO3- I32

estimate the geographic distribution of exports and MNE activity in Mexico. To the extent medium and large plants account for most export activity in Mexico and to the extent that most MNEs are medium and large firms, the Secofi sample is a representative sample of plants that are likely to export and plants that are likely to be foreign owned. The data support this claim; the plants in the Secofi sample accounted for 43 percent of Mexican manufacturing exports in 1986 and 45 percent in 1989. 3.2. Ownership

and export behavior

We define an MNE to be a plant with positive foreign equity ownership, even though the plant may not be majority foreign owned. Table 3 provides summary statistics for all plants, plants that export, and foreign-owned plants in 1986 and 1989. The share of all plants that export over the two years was 27 percent. While there was an increase in the number of exporters between 1986 and 1989, the majority of plants that exported in 1989 also exported in 1986. MNEs compose a large fraction of the sample; 28 percent of the plants had positive foreign equity ownership. Table 3 also shows the distribution of plants by region. There is a high degree of concentration in central Mexico and along the border; the Mexico City region contains 56 percent of the plants, the central states contain 20 percent, and the border states contain 17 percent. On average, exporters employ more workers than do non-exporters. Foreign shipments average 15 percent of the total shipments of exporters. The United States is the major export market; exporters on average send 56 percent of their foreign shipments to their northern neighbor. The geographic distribution of exporters, at least at the regional level, is similar to that for all plants. The fact that exporters are not overwhelmingly concentrated along the border suggests that transport costs are not a dominant factor in the plant-location decision. Not surprisingly, foreign-owned plants are more likely to export than are domestically-owned plants; over the two years, 46 percent of foreign plants export. The United States is the most common source of foreign capital and Germany is the next most common. Of plants with foreign equity ownership, 60 percent have U.S. equity participation and 15 percent have German equity participation. Foreign-owned plants on average employ more workers than do domesticallyowned plants and are more likely to be located in Mexico City. 3.3. The geography

of exports

If there are localized externalities associated with exporting then we expect the plants that export in an industry to be geographically concentrated. The geographic concentration of export activity in Mexico is considerable, but this by itself is not evidence of spillovers. Many factors cause firms to agglomerate, including regional variation in factor abundance. To isolate the effect of export spillovers on

B. Aitken et al. I Journal of International

113

Economics 43 (1997) 103-132

Table 3 Summary Statistics for Plants, 1986 and 1989 Variables (all values are sample means)

All Plants 1986

Employment Export Plant FDI Local FDI Export Share Export USA Export Canada Export Other % Export USA % Export Canada % Export Other Plant FDI: USA Plant FDI: Canada Plant FDI: Japan Plant FDI: Germany Plant FDI: Spain Plant FDI: Other Border North Central Mexico City south Number of Plants

318.4 0.218 0.278 4.16

0.169 0.037 0.204 0.557 0.033 2104

1989 327.3 0.320 0.278 4.16

0.169 0.037 0.204 0.557 0.033 2104

Plants that Export

Foreign Plants

1986

1989

1986

1989

597.6

555.0

470.2 0.382

485.7 0.527

5.80

5.80

0.598 0.045 0.034 0.146 0.060 0.286 0.144 0.022 0.178 0.632 0.024 584

0.598 0.045 0.034 0.146 0.060 0.286 0.144 0.022 0.178 0.632 0.024 584

0.486 5.33 0.157 0.712 0.094 0.588 55.53 2.36 41.46

0.187 0.035 0.214 0.538 0.026 459

0.457 5.04 0.139 0.746 0.117 0.604 58.48 1.92 39.60

0.194 0.034 0.218 0.531 0.022 674

Variable Definitions: Export= 1 if plant has positive exports. Plant FDI= 1 if plant has foreign ownership (US=U.S. ownership, etc.). Local FDI: number of foreign plants in plant’s state industry. Export Share: plant exports/plant shipments. Export USA= 1 if plant exports to U.S. (Can=Canada, etc.). % Export USA: % of exports to U.S. (Can=Canada, etc.). Border, North, Central, Mexico City, South= 1 if plant is located in indicated region

the probability a plant exports, we need to control for the overall geographic concentration of industry activity. In this way, we hold constant other factors that contribute to industry agglomeration. Before turning to the empirical estimation we provide data on the geographic concentration of export activity in Mexico. Localized export spillovers imply there will be an excess geographic concentration of export activity-a geographic concentration of exporters beyond that which exists for the industry as a whole. A simple way to identify excess export concentration for a given industry is to compare the distribution of domestic shipments across states with the distribution of exports across states. Consider a histogram of state-industry shares of national-industry domestic shipments. If

114

B. Aitken et al. I Journal of International

Economics 43 (1997) 103-132

shipments are geographically concentrated, there will be a large number of states whose share of national-industry shipments is near zero and a few states whose share of national-industry shipments is large. If, on the other hand, shipments are evenly spread across states, the distribution of state-industry shipments shares will be centered at 3 percent (l/32). Let sij be state i’s share of domestic shipments of three-digit industry j and let xii be state i’s share of national exports of three-digit industry j. Exports and shipments data are from the Secofi sample. Fig. la and Fig. lb show histograms for sij in 1986 and 1989.” In many industries a single state accounts for half or more of national industry shipments. This, in combination with the fact that most state industries have a shipments share near zero, is consistent with the idea that domestic production is geographically concentrated. Given some states are underrepresented in the Secofi sample, the distribution of shipments across state industries is not very informative. What is informative is to compare Fig. la and Fig. lb with similar figures for exports. If exports are more geographically concentrated than shipments, then compared to state-industry shipments shares, the distribution of state-industry shares of national industry exports will have a larger mass near zero and a larger mass with very high shares, but a lower mass at intermediate levels. Fig. 2a and Fig. 2b show histograms for xii in 1986 and 1989. In 1989, the fraction of state industries with export shares near zero is 60 percent, while the fraction of state industries with shipments shares near zero is only 35 percent. Also, the distribution of export shares has a smaller mass of state industries with shares between 5 percent and 25 percent, and a larger mass of state industries with shares between 75 and 100 percent. This is consistent with the idea that exports are more localized than shipments. A higher concentration of exports than domestic shipments is not sufficient evidence of spillovers. Transport-cost considerations give exporters locational incentives that are different from those of domestic producers. It would not be surprising to find, for instance, that the states that do all the exporting happen to have large ports or share a border with the United States. To identify the effects of export spillovers, it is necessary to control for other factors that affect the probability of exporting.

4. Empirical

results

In this section, we estimate the probability that a plant has positive exports. Two general results are robust across samples and specifications. After controlling for

“Figs. la-2b pool industries together. Figures for individual industries are similar.

B. Aitken

et al. I Journal

of International

Economics

43 (1997)

I15

103-132

.6 -

(a) .5 -

.4 s4 8%

.3 -

k .2 .1 I I 0.03 State-Industry

I .25 Share

I .5 of National

Industry

--I .75 Oomestlc

I Sales,

.6 I

15lElk

04

.4 .3 .2 .1 -I I 0.03 State-Industry

I .25 Share

1 of Nat&:

/

Industry

llom~~tic

I Sales,

1&l&3

Fig. 1. Geographic distribution of sales.

other factors that affect the decision to export, we find that the probability of exporting is (1) positively correlated with the local concentration of MNE activity, and (2) uncorrelated with the local concentration of overall export activity. While there are no apparent spillovers from locating near other exporters, there do seem to be benefits from locating near multinational firms.

116

B. Aitken et al. I Journal of International

Economics 43 (1997) 103-132

(4

-L

1

----

,

I

.25

0.03 State-Industry

Share

of

Nat&al

Industry.7;jxparts.

I

1986

1

@I

.6 -

.5 -

.4 -

.3 -

.2 -

.l

-

-I I

I

0.03

.25 State-Industry

Nat& Share

I

Induetry'$mrts,

of

I

1

1 1989

Fig. 2. Geographic distribution of exports.

4.1. Model specification

Eq. (11) describesthe decision to export as a function of domestic and foreign output prices, factor prices, and geographic concentration measures.Table A.1 in Appendix A defines the regressors and the additional instruments we use to estimate Eqs. (12) and (13). We measure geographic concentration at the level of the state and three-digit

B. Aitken

et al. I Journal

of International

Economics

43 (1997)

103-132

117

industry. To control for variation in the size of industries at the national level, we measure the geographic concentration of industry as the state-industry share of national industry activity. Naturally, we want to control for situations where the state-industry is large purely because the state is large. To do so, we normalize by the state share of national manufacturing activity. The resulting measure of geographic concentration is the ratio, State-industry share of national industry activity State share of national manufacturing activity ’ This measure is similar to those used in recent empirical literature on economic geography (e.g., Glaeser et al., 1992; Henderson, 1995). The three geographic-activity measures of interest are those for the overall concentration of economic activity, the concentration of export activity, and the concentration of MNE export activity. We define overall industry concentration as the state-industry share of national industry employment, relative to the state share of national manufacturing employment.13 We use this measure to control for the effects of general spillovers and exogenous site-specific factors of production, which allows us to identify the effects of different types of export spillovers. We define local export concentration as the state-industry share of national industry exports, relative to the state share of national manufacturing exports. All measures of export concentration exclude the plant on which the observation is taken. Our measure of local export concentration controls for situations where a state industry has high exports purely because the state has high exports, such as might be the case in border states or states with port facilities. For a state industry to show a high level of export concentration, the state must be relatively specialized in exporting the product of that industry. If export spillovers exist, we expect the probability a plant has positive exports to be increasing in the level of local export concentration. We measure MNE export activity as the share of state-industry MNE exports in national industry exports, relative to the state share of national manufacturing exports. Our measure of local export concentration encompasses MNE exports.‘4 By including an independent measure MNE export activity, we allow for the possibility that MNEs, rather than domestic exporters, are the source of spillovers. If MNEs generate spillovers for local exporters, we expect the probability a plant exports to be increasing in the local concentration of MNE export activity. 13As mentioned in Section III, we calculate overall industry concentration with data from the Industrial Census and export concentration with data from the Secofi sample. We measure overall industry concentration at the three-digit ISIC level and export concentration at the three-digit MEXSIC level. The latter is more disaggregated than the former, as the sample contains 27 three-digit ISIC industries and 54 three-digit MEXSIC industries. We are required to use data at the three-digit level in order to combine the Census with the Secofi sample. 14Again, we exclude the plant on which the observation is taken.

118

B. Aitken et al. I Journal of International

Economics 43 (1997) 103-132

Other explanatory variables follow the specification in Eq. (11). We measure domestic prices using four-digit producer-price indices, deflated by the wholesaleprice index.15 We measure labor costs as average remuneration per worker by state and three-digit industry. We measure non-labor factor costs using four-digit input-price indices for construction, machinery, transportation, office equipment, raw materials, and electricity. We include as regressors average tariffs and import-license requirements by four-digit industry. In addition to its effects on the output price, trade policy affects export behavior through its effects on input prices. As additional control variables we include dummy variables for the foreign-ownership status of the plant, a dummy variable for whether the plant subcontracts for other plants, employment in the plant relative to industry average employment, value-added tax payments as a share of sales, royalty payments as a share of sales, distance from a plant’s state capital to the Mexico-U.S. border and to Mexico City, and dummy variables for industry, region, and year. 4.2. Estimation results In the tables described below, we present estimation results on four different subsamples. First, we estimate Eq. (11) for the entire sample.i6 Second, we estimate Eq. (11) for the sample of domestically-owned plants, excluding all plants with some foreign ownership. Third, we redo the analysis for domestic and foreign plants but exclude natural-resource-intensive industries, for which geographic concentration is likely due to unobserved site-specific characteristics. Finally, we redo the analysis for domestically-owned plants, excluding resource-intensive industries. Observations are pooled over 1986 and 1989. We direct the discussion towards the results on local export concentration and local MNE export activity. Since the model we estimate is a reduced form, the results on input prices and trade protection variables are difficult to interpret. From Section II, these variables affect production for both domestic and foreign markets. Hence, we cannot identify the parameters that are specific to foreign production and distribution costs. 4.2.1. B.1 Standard probit results Table 4 reports estimation of Eq. (11) using standard probit techniques. Column (1) in Table 4 shows results for the full sample of plants. The coefficients on local export concentration and MNE export activity are positive and statistically

?he specification in Eq. (11) includes export prices as an independent variable. Unfortunately, no data for export prices exist. It is possible to construct unit values for exports (defined as sales divided by quantities of exports), but unit values do not offer an economically meaningful measure of export prices since they are not comparable to domestic prices. 16All regressions reported in Tables 4-6 include dummy variables for the region, industry, and year, which, for expositional ease, are not shown.

B. Aitken

et al. I Journal

of International

Economics

43 (1997)

119

103-132

Table 4 Probit Specification for Decision to Export (asymptotic t-statistics in parentheses) All Industries

Overall Industry Concentration Local Export Concentration MNE Export Activity Plant FDI: USA Plant FDI: Europe Plant FDI: Japan Plant FDI: Other Domestic Price Tariffs on Output (TR 1) Quotas on Output (TR2) Tariffs on Inputs (TR3) Quotas on Inputs (TR4) TR 1*TR2 TR3*TR4 Construction Price Index Machinery Price Index Transportation Price Index Office Equip. Price Index Raw Material Price Index Electricity Price Index State-Industry Wage

Excluding Natural Resources

All Plants (1)

Domestic Only

-0.0145 (-0.5 19) 0.0043 (2.964) 0.0353 (5.080) 0.4823 (8.008) 0.1814 (1.753) 0.7198 (2.918) 0.3729 (4.548) 0.3032 (2.697) -0.0119 (-2.475) 0.0075 (2.861) 0.0141 ( 1.485) -0.0150 (-4.367) -0.0002 (-2.847) 0.0008 (3.833) -3.3618 (-1.153) 0.6164 (-2.487) -0.3277 (-1.113) -3.6622 (-6.102) 0.1646 (1.345) 2.7901 ( 1.826) 0.0003 (2.637)

-0.0002 (-0.006) 0.0047 (2.565) 0.0450 (3.885)

(2)

0.0 139 (0.099) -0.0109 (- 1.820) 0.0060 (1.770) 0.0058 (0.503) -0.01436 (-3.622) -0.0001 (-1.443) 0.0007 (2.998) 0.8594 (0.203) -0.6649 (-2.137) -0.0862 (-0.247) -3.7582 (-5.187) 0.2799 (1.717) 0.4772 (0.218) 0.0003 (1.922)

All Plants (3)

Domestic Only (4)

-0.0294 (-0.994) 0.0012 (0.369) 0.0343 (4.184) 0.5171 (8.355) 0.2048 (1.923) 0.7739 (3.153) 0.3873 (4.498) 0.2798 (2.387) -0.0138 (-2.602) 0.0068 (2.333) 0.0086 (0.833) -0.0273 (-5.339) -0.0002 (-2.599) 0.0012 (4.325) -3.4430 (-1.151) -0.6339 (-2.322) -0.3 loo (-0.948) -2.9737 (-4.302) -0.08915 (-0.676) 2.7803 (1.779) 0.0004 (3.579)

-0.033 1 (-0.832) 0.0007 (0.122) 0.0642 (3.379)

-0.006 1 (-0.041) -0.0122 (-1.805) 0.0078 (2.093) 0.0027 (0.212) -0.0308 (-4.818) -0.0002 (-1.921) 0.0013 (3.627) I .3529 (0.309) -0.609 1 (-1.716) -0.1040 (-0.262) -2.98 19 (-3.531) -0.0447 (-0.241) 0.1475 (0.065) 0.0004 (2.853)

120

B. Aitken et al. I Journal of International

Economics 43 (1997) 103-132

Table 4. Continued All Industries

Proximity: Mexico City Proximity: Mexico City* Proximity: USA Proximity: USA* Relative Plant Size Maquila Dummy Value Added Taxes Royalty Payments Observations Log Likelihood

Excluding Natural Resources

All Plants (1)

Domestic Only

-0.0004 (-1.388) o.oooo (1.044) 0.0006 ( 1.036) 0.0000 (-0.0560) 0.2519 (10.866) 0.0698 ( 1.066) 0.4513 (1.151) 6.3671 (4.272) 4073 -1893.06

(2)

All Plants (3)

Domestic Only (4)

-0.0002 (-0.705) 0.0000 (0.674) 0.0008 (1.187) o.oooo (-0.571) 0.2676 (9.403) 0.005 1 (0.064) 0.7097 (1.425) 3.8937 (2.033) 2940 -1225.83

-0.ooo5 (-1.582) 0.0000 ( 1.446) 0.0004 (0.576) o.oooo (-0.444) 0.2781 (11.058) 0.0339 (0.496) 0.3645 (0.920) 6.2450 (4.090) 3728 -1708.53

-0.0003 (-0.883) 0.0000 (1.204) 0.0004 (0.421) 0.0000 (0.020) 0.3088 (9.707) -0.0223 (-0.264) 0.5624 (1.128) 3.0045 (1.462) 2648 -1069.88

significant, indicating that a higher level of these variables is associated with a higher probability that a plant exportsL7 This result is consistent with positive localized spillovers associated with exporting. Plants with foreign equity participation are also more likely to export, with the effect depending on the country of origin. Plants under Japanese or North American ownership are more likely to export than are domestic plants, while plants under European ownership are not. Other statistically significant variables include relative plant size, the stateindustry wage, and royalty payments as a share of sales. For a given industry, plant size is positively correlated with the probability that a plant exports. This finding is consistent with declining per unit foreign distribution costs, in which case larger plants are able to spread the fixed costs of entering foreign markets over more production. It is also consistent with the existence of fixed startup costs in exporting or the possibility that large plants are low marginal cost producers and face a higher return to exporting than other plants. An increase in the average wage in a given region is associated with a higher probability of exporting. This is consistent with export production being relatively skill-intensive, where high

“The threshold level of statistical significance we adopt is 5 percent.

B. Aitken et al. I Journal of International

Economics 43 (1997) 103-132

121

wages reflect high levels of location or individual-specific human capital. Higher royalty payments are also associated with a higher export probability, which may reflect the use of more sophisticated production technologies or more product differentiation in plants that export. The nonlinearity of the probit specification makes the economic significance of the coefficients in Table 4 difficult to interpret. We also compute the marginal effect of a change in the independent variables on the probability of exporting. At sample means for the independent variables, the predicted probability of exporting is .223. A marginal increase in MNE export activity raises the probability that a plant exports by .01.18 A marginal increase in the local wage increases a plant’s export probability by less than .Ol, and a marginal increase in relative plant size increases the export probability by .075. Given our concerns about the endogeneity of MNE export activity, we reestimate the model for a sample restricted to domestic plants. Column (2) of Table 4 shows the results, which are quite similar to those in column (1). In particular, local export concentration and MNE export activity are again positive and statistically significant. The significant coefficient on the MNE coefficient suggests that foreign firms do convey benefits to domestic plants, providing support for the concept of spillovers. One potential problem with our results is that the measure of overall industry concentration may not adequately control for unobserved site-specific characteristics, in which case the significance of local export concentration and MNE export activity may be merely capturing the fact that exporters locate in regions whose natural endowments make exporting more feasible. Our measure of overall industry concentration is at the three-digit level (see note 13), which groups together industries whose subindustries have highly dissimilar factor intensities. One example is worth noting. Nearly all exports of seafood packing in our sample come from two states, Baja California and Tamaulipas. Both are coastal states and home to Mexico’s largest shrimping fleets; hence it is not surprising that most shrimp-packing plants are located in these states. Shrimp packing is a four-digit industry, within the three-digit industry of food products. Since our measure of overall industry concentration is at the three-digit level, it may fail to control for the fact that these two states have large concentrations of shrimp packers. Columns (3) and (4) of Table 4 show results for regressions in which we exclude two sets of industries: those that are relatively intensive in the use of natural resources, such as food products, petroleum refining, and tequila, and those that have high transport costs and hence tend to produce for the local market, such

‘“We calculate the effect of a marginal increase in an independent variable on the probability of exporting as the height of the normal density at mean values for all independent variables times the regression coefficient of the variable of interest.

122

B. Aitken et al. I Journal of International

Economics 43 (1997) 103-132

as cement, bread, and tortillas.” These are industries for which site-specific factors are likely to be relatively important in the export decision. While the signs, magnitudes, and patterns of significance on MNE export activity in columns (3) and (4) are similar to those in columns (1) and (2), local export concentration is statistically insignificant in both cases. Coefficient estimates for other variables are largely unaffected. The lack of robustness on the results for local export concentration is striking. Rarely are empirical studies of spillovers able to distinguish between different types of geographic concentration. The fact that local export concentration is so sensitive to the set of industries that are included suggests caution in inferring the existence of spillovers, without first adequately controlling for exogenous sitespecific characteristics. In our case it is apparent that a relatively small set of natural-resource-intensive industries are responsible for the positive correlation between local export concentration and the probability of exporting. When resource-intensive industries are excluded, the likelihood of exporting continues to be positively correlated with the local concentration of MNEs. In other words, evidence for spillovers from MNEs is robust to the exclusion of resource-intensive industries. It is worth considering the factors that make MNEs distinct from domestic plants. MNEs have foreign parent firms, which may make them a better source of information about foreign markets than domestic exporters. MNEs may also attract specialized input suppliers to their locations, which may in turn attract domestic producers that desire to export. This latter channel has been formalized by Rivera-Batiz and Rivera-Batiz (1991) and is consistent with Head and Ries (1994) finding that foreign firms locating in China are attracted to cities with relatively large concentrations of input suppliers. 4.2.2. B.2 Two-stage probit results To address the possibility of simultaneity, we treat I& and r,,, as endogenous variables. As discussed earlier, we use the two-stage conditional maximum likelihood (2SCML) estimator proposed by Rivers and Vuong (1993). As a first stage, OLS is performed on Eqs. (12) and (13). The residuals from these regressions are then included as regressors in probit estimation of Eq. (11). Under the null hypothesis of exogeneity, the coefficient estimates on the OLS residuals in the probit equation equal zero. A valid instrument for the first-stage estimation of Eqs. (12) and (13) is a variable that is correlated with the concentration of local export activity and the “The excluded industries and their MEXSIC classification codes are the following: fruit and vegetable products (2011, 2012), wheat and corn milling (2022, 2023), coffee and tea (2025, 2026, 2027), sugar (2031), meat products (2041, 2049), milk products (2051, 2052, 2053, 2054, 2059), seafood packing (2060). bread (2071), tortillas (2093), animal feed (2098), fermented cactus (2115). tobacco products (2201, 2202, 2209), cotton packing (2311), lumber (261 l), petroleum refining (311 l), bricks and clay (3331, 3332), cement (3341), rail manufacturing (3820). shipbuilding (3831), and aircraft manufacturing (3832).

B. Aifken et al. I Journal of international

Economics 43 (1997) 103-132

123

concentration of local multinational export activity and uncorrelated with unobserved shocks that affect the export decision of individual firms. Such instruments may difficult to find, since shocks that affect one firm may also affect other firms in the same region-industry. We take advantage of an important difference between domestic and multinational plants: multinational plants are subject to cost shocks in foreign markets that are likely to have little direct effect on the production decisions of domestic plants. The instruments we use are the exogenous independent variables included in the probit estimation and variables that are likely to be correlated with the activities of multinational firms in an industry. Thus, the instruments should be considered valid only for the sample of domestic plants. While we present two-stage probit results for both the full sample of plants and the sample of domestic plants, we again emphasize results for domestic plants. A first set of instruments are from the U.S. Annual Survey of Munufuctures. We include for each industry the U.S. real wage, the U.S. labor share of value added, and the U.S. labor share of output. Relative labor costs in the United States are likely to be a determinant of U.S. foreign direct investment in Mexico. A second set of variables are constructed from Mexican trade data and the Secofi sample. For each industry, we include the ratio of the real capital stock to total employment and the ratio of imports to total consumption. Both variables provide information on factor intensity, which is likely to be correlated with the attractiveness of an industry to foreign investors. Finally, we include data from the Venezuelan Zndustrial Census on the ratio of foreign investment to total investment by industry, each industry’s share of total foreign investment in the country, and each industry’s share of national industry sales. Given similarities in the two countries’ factor endowments, the pattern of foreign investment and exports in Venezuela is likely to mirror that in Mexico. Table A. 1 (Appendix A) describes the variables. Table 5 shows estimation results for 2SCML. For the sake of presentation, we do not show results on the first-stage estimation of Eqs. ( 12) and ( 13). The 2SCML probit regressions include the residuals from these first-stage regressions as additional explanatory variables, labeled VHATEX and VHATMNE. Under the null hypothesis that local export concentration and MNE export activity are exogenous, the regression coefficients on VHATEX and VHATMNE are equal to zero. We reject this null at the 1% level of significance in all regressions, using the conditional likelihood ratio test proposed by Rivers and Vuong (1993). Unlike the results reported in Table 4, we find that local export concentration is negative in all four specifications and statistically significant in two. The pattern in Tables 4 and 5 suggests that proximity to other exporters does not, in general, raise the likelihood that a plant exports. MNE export activity, however, is positive and statistically significant at the ten-percent level in three regressions and at the one-percent level in two regressions. MNE export activity is only insignificant in the regression on the full sample, which includes foreign plants and resource-

124

B. Aitken et al. I Journal of Inrernarional Economics 43 (1997) 103-132

Table 5 Two-Stage Probit Specification for Decision to Export (asymptotic r-statistics in parentheses) All Industries All Plants (1) Overall Industry Concentration Local Export Concentration MNE Export Activity VHATEX VHATMNE Plant FDI: USA Plant FDI: Europe Plant FDI: Japan Plant FDI: Other Domestic Price Tariffs on Output (TRl) Quotas on Output (TR2) Tariffs on Inputs (TR3) Quotas on Inputs (TR4) TRl *TR2 TR3*TR4 Construction Price Index Machinery Price Index Transportation Price Index Office Equip. Price Index Raw Material

-0.222 (-0.711) -0.0600 (-3.934) -0.0275 (-0.305) 0.0652 (4.253) 0.0604 (0.669) 0.5187 (6.441) 0.3657 (3.330) 0.6837 (2.771) 0.4369 (3.595) 0.1887 (1.647) -0.0050 (-0.908) 0.0072 (2.741) 0.0052 (0.502) -0.0278 (-5.595) -0.0003 (-3.557) 0.0021 (4.930) -2.7598 (0.85 1) -0.1596 (-0.506) -0.5952 (-1.730) -4.4996 (-6.940) 0.4249

Excluding Natural Resources Domestic Only

(2) -0.0926 (-0.943) -0.0068 (-0.331) 0.2866 (1.744) 0.0125 (0.608) -0.2468 (-1.501)

-0.0387 (-0.267) -0.0139 (-1.875) 0.0066 (1.921) 0.0146 (1.035) -0.0138 (-1.989) -0.0001 (-1.312) 0.0007 (1.166) 5.6661 (1.170) -0.6961 (-1.829) 0.1133 (0.310) -3.4973 (-3.703) 0.4793

All Plants (3)

Domestic Only (4)

-0.0154 (-0.228) -0.0880 (-1.234) 0.3806 (5.023) 0.0895 (1.254) -0.3497 (-4.602) 0.3369 (4.600) 0.5782 (2.367) 0.7061 (2.861) 0.05777 (0.498) 0.2078 (1.714) -0.0069 (-1.032) 0.0042 (1.021) -0.0243 (-0.980) -0.0336 (-4.122) -0.0002 (-2.025) 0.0021 (2.152) 4.1602 (0.993) -1.0621 (-3.410) 0.2208 (0.625) -3.1337 (-4.350) 0.0641

-0.2246 (-3.981) -0.2002 (-3.098) 0.5402 (5.120) 0.2033 (3.140) -0.4849 (-4.576)

-0.1487 (-0.97 1) 0.0007 (0.093) -0.0001 (-0.012) -0.0798 (-2.930) -0.0469 (-5.745) -0.0001 (-0.826) 0.0039 (3.994) 18.4318 (3.021) -0.8096 (-2.201) 0.7962 (1.768) -2.9785 (-3.267) -0.1720

B. Aitken et al. / Journal of International

Economics 43 (1997) 103-132

125

Table 5. Continued All Industries

Price Index Electricity Price Index State-Industry Wage Proximity: Mexico City Proximity: Mexico City’ Proximity: USA Proximity: USA’ Relative Plant Size Maquila Dummy Value Added Taxes Royalty Payments Observations Log Likelihood Conditional LR Test*

Excluding Natural Resources

All Plants (1)

Domestic Only

(2)

All Plants (3)

Domestic Only (4)

(3.161) 2.8375 (1.629) 0.0001 (1.058) -0.0009 (-3.075) 0.0000 (3.132) -0.0014 (- 1.276) 0.0000 (1.297) 0.2238 (8.612) 0.2079 (2.393) 0.4151 (1.043) 6.924 (4.530) 4073 - 1879.36 27.40

(2.356) -2.1095 (-0.817) 0.0002 (1.137) -0.0002 (-0.429) O.CQOO (0.833) 0.0010 (0.736) 0.0000 (-0.060) 0.2776 (8.468) -0.0531 (-0.425) 0.7874 (1.507) 3.3507 (1.699) 2940 -1218.97 13.72

(0.335) -1.0068 (-0.493) o.ooo2 (1.290) -0.0015 (-3.303) o.ooao (3.263) 0.0027 (2.750) O.OCKlO (-1.904) 0.2634 (7.432) -0.250 (-0.350) 0.2052 (0.503) 5.4547 (3.540) 3728 1697.58 21.90

(-0.759) -7.9450 (-2.672) 0.0003 ( 1.934) -0.0011 (-2.320) 0.0000 (2.685) -0.0011 (- 1.050) 0.0000 (1.653) 0.2312 (5.590) -0.0064 (-0.07 1) 0.3146 (0.599) 2.2099 (1.066) 2648 -1057.89 22.98

* Tests the null hypothesis that Local Export Concentration and MNE Export Activity are exogenous (which is same as null that estimated coefficients on VHATEX and VHATMNE are zero). Under the null, the test statistic is distributed as ~~(2). For values greater than 9.21, the null is rejected at a 1% level of significance.

intensive industries. The most notable difference between the standard probit results and the 2SCML results is that the magnitude of the coefficient estimate on MNE export activity is considerably larger in the two-stage probit. To compare our results with the standard probit, we again compute the marginal effect of a change in the independent variables on the probability of exporting. At sample means for the independent variables, the predicted probability of exporting varies from .153 for domestic plants in all industries to .141 for domestic plants in non-resource-intensive sectors. Using 2SCML results, a marginal increase in MNE export activity raises the probability of exporting by .077 (for all domestic plants) to .121 (for domestic plants in non-resource intensive sectors). These effects are considerably larger than those computed using standard probit estimates.

126

B. Aitken et al. I Journal of International

Economics 43 (1997) 103- 132

4.2.3. B.3 Extensions In Table 6, we report a number of extensions on the two-stage probit. First, we add additional plant characteristics, which could affect the decision to export. Second, we replace our measures of industry concentration, export concentration and MNE activity with measures calculated using numbers of firms instead of total employment or total exports. Third, we add a measure of MNE export activity at the national level, to allow for the fact that spillovers could occur nationally as well as locally. Finally, we replace our measure of MNE export activity with a measure of MNE domestic production. Table A.1 (Appendix A) defines the variables.

Table 6 Extensions on Two-Stage Probit Specification for the Decision to Export (asymptotic f-statistics in parentheses) All Specifications: Domestic Plants Only, Natural-Resource Industries Excluded

(1) Overall Industry Concentration Local Export Concentration MNE Export Activity Skilled Wages/ Unskilled Wages Capital /Labor Ratio Total Factor Productivity Alternative Overall Industry Concentration Alternative Local Export Concentration Alternative MNE Export Activity MNE National Export Index MNE Domestic Activity Observations Log Likelihood

(2) -

-0.2248 (-3.706) - 1.925 (2.941) 0.4019 (3.451) -0.0177 (-0.187) 20.6829 (3.924) -0.0089 (-0.118)

(3)

(4) -0.2536 (-4.105) 0.2262 (-3.246) 0.5988 (4.981)

-0.0674 (-1.647) -0.0768 (1.276)

-0.1732 (3.190) 0.1065 (0.570) 0.7854 (3.916) -1.1841 (-0.588)

2604 -1016.21

2648 -1060.52

2648 -1051.67

0.8584 (4.490) 2648 -1064.23

The regressions reported in Table 6 also include the other independent variables in the regression reported in column (4) of Table 5. For expositional ease, the results for these variables are not shown.

B. Aitken

et al. I Journal

of International

Economics

43 (1997)

103-132

121

We perform all the extensions on the restricted sample of domestic plants, excluding natural resource-intensive industries and foreign-owned plants. Regionspecific factors that may also affect the probability of exporting include access to skilled labor, technology, and capital inputs.*” If these factors are accumulating relatively rapidly in a region, then plants that locate in the region may be more likely to export. We redo the analysis by adding three additional variables: the log ratio of white-collar to blue-collar wages, the log capital-labor ratio, and lagged total factor productivity growth. The results are reported in column (1) of Table 6. While the probability that a plant exports is positively correlated with the capital-labor ratio, our main findings are unaffected. In column (2) of Table 6 we report the coefficient estimates on alternative measures of overall industry concentration, local export concentration, and MNE export activity, where we construct each variable using numbers of firms (see Table A.1 in Appendix A). This specification allows for the possibility that a concentration of many firms, rather than high export volume by a few large firms, is critical for exporting. The results are very similar to those in Table 5. In column (3) of Table 6 we include a measure of the MNE share of national industry exports. This additional measure allows for the possibility that spillovers from MNE activity may occur nationally rather than locally. MNE export activity at the local level is again statistically significant, but MNE export activity at the national level is not. This suggests that the benefits conferred by MNE activity are local, and not national, in nature. Finally, in column (4) we replace MNE export activity with an analogous measure of MNE domestic production-the share of MNE domestic shipments in state-industry domestic shipments, normalized by the state share of national domestic shipments. We find that this variable is positive and statistically significant. This last finding merits further discussion. The export decision of domestic plants appears to be positively correlated with the local concentration of MNE activity, regardless of whether MNEs serve local or export markets. This suggests that proximity to MNEs, in general, provides domestic plants with access to foreign markets. One possibility is that domestic firms do not learn from the specific production activities of MNEs but benefit from general linkages that MNEs maintain with parent firms, or other firms, abroad. Another possibility is that the positive correlation between the probability a plant exports and MNE domestic production is merely a byproduct of the statistical correlation between MNE export activity and MNE domestic production. The two

*“For further work linking higher wages with multinational firms, see Aitken et al. (forthcoming). For an analysis of the relationship between relative wages, foreign investment, and export behavior, see Hanson and Harrison (1995).

128

B. Aitken et al. I Journal of International

Economics 43 (1997) 103-132

variables have a correlation of .70. For this reason it is important to exercise caution in interpreting this result.

5. Concluding

remarks

Anecdotal evidence suggests that the process of breaking into foreign markets can be very difficult. Potential exporters must obtain information about foreign tastes and establish distribution channels in foreign markets. One obvious way for firms to learn about export markets is to observe exporters that have already acquired experience selling abroad. Those exporters may be other domestic firms or multinationals. In this paper, we examine whether other exporters can reduce the cost of foreign market access for a firm contemplating the jump into export markets. In particular, we examine whether locating near other exporters increases the probability of exporting. The basis for the study is 2104 Mexican manufacturing plants over the period between 1986 and 1990. Following Mexico’s trade reform in 1985, many Mexican manufacturers turned away from the previously protected domestic market towards outside markets. The changes during the 1980s allow us to identify the kinds of firms most likely to become exporters. Ours is the first study which provides statistical evidence consistent with the role of foreign firms as “catalysts” for domestic exporters. Using a two-stage probit specification, we find that the probability a domestic plant exports is positively correlated with proximity to multinationaZ firms. This result is robust to the inclusion of other measures, such as overall industrial activity in a region, proximity to the capital city and border regions, and price and cost variables. Yet the plant export decision is uncorrelated with the local concentration of overall exporters. This suggests that export spillovers are restricted to multinational activity. Foreign-owned enterprises are a natural conduit for information about foreign markets and technology, and a natural channel through which domestic firms can distribute their goods. To the extent that foreign investors directly or indirectly provide information and distribution services, their activities enhance the export prospects of local firms. One implication is that firms wishing to export will tend to locate in areas where multinational firms are concentrated. Another implication is that governments may wish to encourage potential exporters to locate near each other. One policy option adopted by governments is the creation of export processing zones (EPZs), special economic zones reserved for exporting firms. These zones confer special benefits to exporters, such as duty-free imported inputs, tax holidays, or subsidized infrastructure. Our research suggests that if EPZs encourage domestic exporters to locate near multinationals, such zones could in principle help reduce the costs of breaking into foreign markets. In practice, however, EPZs have often been located in regions that are removed from domestic production centers. This may provide

B. Aitken et al. I Journal of International

Economics 43 (1997) 103-132

incentives for would-be exporters to isolate themselves from multinationals other exporting firms.

129

and

Acknowledgments The authors thank seminar participants at Princeton University, the University of Colorado, the University of Michigan, the University of Texas, and the NBER Summer Institute for valuable comments. This research was supported by World Bank research project 678-29, “Technology Spillovers, Agglomeration, and Direct Foreign Investment.”

Appendix

A

Table A.l: Variable definitions Dependent

Variable

(Probit

Regressions)

Export Independent

Equals one if plants has positive exports. Variables

Overall Industry Concentrationa Local Export Concentrationb.c MNE Export Activityb.” Plant FDI

Domestic Priced3e Region Wage”3d Plant Sizeb Royalty Payments Proximity USA

(Probit

Regressions)

State industry share of national industry employmerit/ State share of national manufacturing employment. State-industry share of national industry exports/ State share of national manufacturing exports. (State-industry MNE exports/National industry exports) / State share of national manufacturing exports. Equals one if plant has positive foreign equity participation (Plant FDI: USA, Plant FDI: Japan, etc., equal one if plant has positive equity ownership from indicated country). Producer price index. Average compensation per worker in state industry. Employment in plant/Industry average plant employment. Plant royalties paid/Plant shipments. Distance in kilometers from state capital to nearest major U.S. border crossing.

130

B. Aitken et al. I Journal of International

Proximity Mexico City Maquila Dummy Value Added Taxes Output tariff” Input tariff” Output quotae Input quotae Input pricesd3e

Economics 43 (1997) 103-132

Distance in kilometers from state capital to Mexico City. Equals one if plant has positive income from subcontracting. Plant value-added tax payments/Plant shipments. Average tariff (%) on outputs. Average tariff (%) on inputs. Average share of outputs subject to import-license requirement. Average share of inputs subject to import-license requirement. Price indices for construction, machinery, transportation, office equipment, raw materials, and electricity.

Additional Independent Variables used in First-Stage OLS Regressions for Tables 5 and 6 Compensation per worker/Industry producer price US Wage” index for U.S. industry. Labor compensation/Value added for U.S. indusUS Labor Share (1)” &Y* Labor compensation/Value of shipments for U.S. US Labor Share (2)” industry. Real capital stock/Total employment for Mexican Capital/Labor Ratiob industry. Imports /( Imports plus domestic shipments) for Import Penetration Mexican industry. Ratio b Foreign direct investment/Total investment for Venezuela FDI (1)” Venezuelan industry. Industry foreign direct investment/National Venezuela FDI (2)” foreign direct investment for Venezuela. Industry shipments/National shipments for VenVenezuela Sales Share” ezuela. Additional Independent Variables used in Table 6 Regressions Log (Plant average white collar annual earnings/ Skilled Wages/ Plant average blue collar annual earnings). Unskilled Wages Lagged change in Solow residual for the plant. Total Factor Productivity State industry share of national industry establishAlternative Overall ments / Industry Concentrationa.f State share of national manufacturing establishments. State-industry share of national industry exporters/ Alternative Local

B. Aitken

et al. I Journal

Export-Concentrationb”.’ Alternative MNE .Export Activityb.‘.’ MNE National Export Indexh,c MNE Domestic Activityb.’

of Internationul

Economics

43 (1997)

103- 132

131

State share of national manufacturing exporters. (State-industry MNE exporters/national industry exporters) / State share of national manufacturing exporters. MNE share of national industry exports. (State-industry MNE domestic shipments/National industry domestic shipments)/ State share of national domestic manufacturing shipments.

a Industry refers to three-digit ISIC (Mexican data are from Mexico Industrial Census). b Industry refers to three-digit MEXSIC classification (Mexican data are from Secofi sample). ’ Calculated excluding the plant on which the observation is taken. d Deflated by the Mexico wholesale price index. e Industry refers to four-digit MEXSIC classification. f Measures are numbers of plants.

References Aitken, B., A.E. Harrison and R. Lipsey, Wages and foreign ownership: a comparative study of Mexico, Venezuela, and the United States, Journal of International Economics (forthcoming). Aitken, B. and A.E. Harrison, 1992, Does proximity to foreign firms induce technology spillovers’?, World Bank and International Monetary Fund. Bernard, A.B and J.B. Jensen, 1995, Why firms export: experience, entry costs, spillovers, and subsidies, MIT and U.S. Bureau of the Census. Blomstrom, M., 1986, Foreign investment and productive efficiency: the case of Mexico, Journal of Industrial Economics 35, 97- 1 IO. Blomstrom, M. and H. Persson, 1983, Foreign investment and spillover efficiency in an underdeveloped economy: evidence from Mexican manufacturing industry, World Development 11, 493-501. Butler, J.S. and R. Moffit, 1982, A computationally efficient quadrature procedure for the one-factor multinomial probit model, Econometrica 50, 761-765. Carlton, D.W., 1983, The location and employment choices of new firms: an econometric model with discrete and continuous endogenous variables, Review of Economics and Statistics 65, 440-449. Caves, R.E., 1974, Multinational firms, competition, and productivity in host-country markets, Economica 41, 176- 193. Davies, H., 1977, Technological transfer through commercial transactions, Journal of Industrial Economics 26, 165-171. Glaeser, E., H. Kallal, I. Sheinkman and A. Shleifer, 1992, Growth in cities, Journal of Political Economy 100, 1126-1152. Guilkey, D.K. and J.L. Murphy, 1993, Estimation and testing in the random effects probit model, Journal of Econometrics 59, 301-317.

132

B. Aitken et al. / Journal of International

Economics 43 (1997) 103-132

Haddad, M. and A.E. Harrison, 1993, Are there positive spillovers from direct foreign investment? evidence from panel data for Morocco, Journal of Development Economics 42, 51-74. Hanson, G.H. and A.E. Harrison, 1995, Trade, technology, and wage inequality, NBER Working Paper Number 5 110. Head, K. and J. Ries, 1994, Inter-city competition for foreign investment: static and dynamic effects of China’s incentive areas, University of British Columbia. Head, K., J Ries and D. Swenson, 1995, Agglomeration benefits and location choice: evidence from Japanese manufacturing investments, Journal of International Economics 38, 223-248. Henderson, J.V., 1986, Efficiency of resource usage and city size, Journal of Urban Economics 19, 47-70. Henderson, J.V., 1995, Industrial development in cities, Journal of Political Economy 103, 1067-1090. Kessing, D.B., 1983, Linking up to distant markets: south to north exports of manufactured consumer goods, American Economic Review 73, 338-342. Morawetz, D., 1981, Why the emperor’s new clothes are not made in Colombia (University of Oxford Press, London). Nakamura, R., 1985, Agglomeration economies in urban manufacturing industries: a case of Japanese cities, Journal of Urban Economics 17, 108-124. Rhee, Y.W. and T. Belot, 1990, Export catalysts in low-income countries, World Bank Discussion Papers Number 72. Rivera-Batiz, F.L. and L.A. Rivera-Batiz, 1991, The effects of direct foreign investment in the presence of increasing returns due to specialization, Journal of Development Economics 34, 287-307. Rivers, D. and Q.H. Vuong, 1988, Limited information estimators and exogeneity tests for simultaneous probit models, Journal of Econometrics 39, 347-366. Roberts, M. and J. Tybout, 1993, An empirical model of sunk costs and the decision to export, Pennsylvania State University and Georgetown University. Teece, D.J., 1977, Technology transfer by multinational firms: the resource cost of transferring technological knowhow, Economic Journal 87, 242-261. Wheeler, D. and A. Mody, 1992, International investment location decisions, Journal of International Economics 33, 57-76.