- Email: [email protected]
Foreign Direct Investment and mark-up dynamics: Evidence from Spanish firms
Journal of International Economics 76 (2008) 107–115
Contents lists available at ScienceDirect
Journal of International Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e c o n b a s e
Foreign Direct Investment and mark-up dynamics: Evidence from Spanish firms☆ Alessandro Sembenelli a,b, Georges Siotis c,d,⁎ a b c d
Dipartimento di Scienze Economiche e Finanziarie, Facoltà di Economia, Università di Torino, Corso Unione Sovietica 218bis, 10134 Torino, Italy Collegio Carlo Alberto, Italy Departamento de Economía, Universidad Carlos III de Madrid, Calle Madrid 126, 28903 Getafe, Madrid, Spain CEPR, United Kingdom
a r t i c l e
i n f o
Article history: Received 18 March 2004 Received in revised form 8 February 2008 Accepted 9 May 2008 Keywords: Foreign Direct Investment Technology transfer Spillovers Efficiency Competition Panel data GMM
a b s t r a c t A review of the literature indicates that Foreign Direct Investment has the potential to increase the intensity of competition and to act as a channel for technology transfers. Using a Spanish firm level data set, we disentangle these effects by estimating a dynamic model of firm level performance, which we proxy by mark-ups. We find that FDI has a positive long-run effect on the mark-ups of targets, but this is limited to firms in R&D intensive sectors. In addition, we find weak evidence that foreign presence dampens margins. However, this effect appears to be more than compensated by positive spillovers in the case of knowledge intensive industries. © 2008 Elsevier B.V. All rights reserved.
JEL codes: F23 L40 L60
1. Introduction In an early pioneering contribution, Caves (1974) put forward the idea that Foreign Direct Investment (FDI) influenced host country conditions through two main channels. On the one hand, FDI ought to result in technology transfers to host country firms. This effect was conjectured to be both direct—Multinational Corporations (MNCs) providing subsidiaries an efficiency advantage— and indirect—MNCs generating positive spillovers. On the other hand, an important foreign presence could also increase the intensity of competition in the recipient country.1 While the latter conjecture is intuitively appealing, direct empirical evidence of
☆ We would like to thank Editor Jim Tybout, two anonymous referees, Luigi Benfratello, Pedro L. Marín, and Reinhilde Veugelers for helpful comments. This paper was begun while Siotis was a Research Fellow of the Bank of Spain’s Research Department and he expresses his gratitude to the staff of the Bank of Spain for their help with the data. Siotis acknowledges partial support for this project from the EU financed TMR pro ject “FDI and the Multinational Corporation” and from Ministerio de Educación y Ciencia (SEJ2007-66268). Previous versions of this paper were circulated under the title “Foreign Direct Investment, Competitive Pressure, and Spillovers. An Empirical Analysis on Spanish Firm Level Data”. ⁎ Corresponding author. Departamento de Economía, Universidad Carlos III de Madrid, Calle Madrid 126, 28903 Getafe, Madrid, Spain. Tel.: +34 91 624 93 12; fax: +34 91 624 98 75. E-mail addresses: [email protected] (A. Sembenelli), [email protected] (G. Siotis). 1 Caves' insights can be summarised by the following sentences: “The host nation’s private sector does not benefit directly because the foreign subsidiary is efficient, or brings to its shores skilled entrepreneurship or productive knowledge. Rather its gains depend on spill-overs of productivity that occur when the multinational corporation cannot capture all quasi-rents due to its productive activities, or to the removal of distortions by the subsidiary’s competitive pressure”, Caves (1974, p. 176). 0022-1996/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.jinteco.2008.05.003
108
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
the pro-competitive effect of FDI is limited, if not non-existent.2 By and large, the empirical literature has focused on the technological effect, possibly because unearthing the pro-competitive effect of FDI is not trivial.3 The purpose of this paper is to attempt to disentangle empirically the efficiency, spillovers, and competition effects of FDI on firms' mark-ups. We chose mark-ups as a proxy for firm performance for three reasons. First, mark-ups are a natural choice when attempting to gauge changes in competitive pressure. Second, this measure captures the evolution of price as well as costs, both of which will be affected by the FDI induced effects identified above. Third, this variable is less prone to measurement problems stemming from the use of industry-wide deflators, as it directly reflects firms' pricing behaviour. The empirical analysis is carried out with Spanish firm level data over the period 1983–1996. Potentially, Spain's experience represents an interesting case to test the conjectures presented above as a number of factors contributed to a large increase in competitive pressure and a surge in FDI.4 The increase in competitive pressure is documented in Siotis (2003), where a drop in (econometrically estimated) industry mark-ups is identified as a result of Spain's entry into the EU. The main innovations contained in this paper are as follows. First, we use price-cost margins rather than productivity (in the remainder of the paper, we use mark-ups and price-cost margins interchangeably). Second, we focus on the dynamic responses of the dependent variable. In turn, these responses form our main identification argument. Concretely, we conjecture that the effects identified above will work their way through at a different pace. Third, we split our sample using R&D intensity, which provides an additional identifying hypothesis. Fourth, we use the identity of firms as a further check on the robustness of our results. Since Spanish owned firms typically lag foreign subsidiaries along the technological dimension, domestic firms (as opposed to foreign subsidiaries) are likely to be the main beneficiaries of spillovers. Fifth, we apply the Generalised Method of Moments (GMM) in order to deal with endogeneity biases. This also allows us to properly account for industry and firm level fixed effects. We find that after controlling for potential endogeneity biases and economy-wide effects, FDI has a positive long-run effect on the mark-ups of target firms, but this is limited to R&D intensive sectors. In addition, domestically owned firms are the main recipient of spillovers in knowledge intensive industries. Last, the results weakly indicate that an important foreign presence dampens margins, at least in the short-run. However, in the case of R&D intensive industries, this appears to be more than compensated by positive spillovers. The rest of the paper is organised as follows. The next section indicates why, under fairly general conditions, FDI could be expected to act as a channel for technology transfers as well as influence the intensity of competition. Section 3 presents the data and describes how we constructed the variables. Section 4 contains the econometric specification as well the main results. Section 5 concludes. 2. Motivation and testable conjectures 2.1. Technology transfers and competition The fact that MNCs possess firm-specific assets that confer them a competitive edge is well established in the literature (Markusen, 1995). If these assets are transferred to the subsidiary, the latter will be more efficient and, as a general rule, enjoy higher margins (the “direct” effect). For instance, this obtains in the case of Cournot or a Bertrand model with product differentiation, constant marginal cost and with firms facing a linear residual demand curve.5 However, the existence of important costs associated with a change of management may temporarily lead to lower margins, particularly if it involves a foreign firm. This may result because of differences in culture, language, or an inadequate knowledge of consumer preferences, that is, there may be a short-run drop in mark-ups driven by “teething problems compounded by foreignness” (Harris and Robinson, 2002).6 It is also a common observation that MNCs have the potential to generate positive spillovers in the host location (the “indirect effect”; see Barba Navaretti et al. (2004) for an extensive analysis). In the event that these spillovers are large and market-wide, host country competitors are likely to experience an increase in mark-ups (for instance, this is the case in the models mentioned above with complete spillovers). However, even within the simple class of models we refer to, the predictions regarding industrywide margins are less clear-cut when spillovers are limited and/or only affect a subset of firms. The latter may occur if only those firms which are sufficiently “close” to the source (with “closeness” defined as technological, product, or geographical proximity) can appropriate the positive externalities generated by MNCs. 2 In his industry level study, Caves (1974) found that the profitability of Canadian domestic plants during the period 1965–67 was negatively correlated with the average share of foreign plants in industry sales. This finding was interpreted as evidence of the pro-competitive effect of FDI. Clearly, this finding cannot be given a causal interpretation. In a similar vein, Kokko (1996) simultaneously estimates the labour productivity determinants of foreign owned and domestic plants in Mexico for the year 1970. He reports that labour productivity in domestic plants is positively correlated with productivity in foreign plants, but this holds only for a subset of industries. This finding is interpreted as pointing to the existence of competition related spillovers. 3 A recent exception is provided by Bloom Schankerman and Van Reenen (2005) who show that R&D activity generates two types of externalities on rivals: technological spillovers and an increase in product market rivalry. 4 In 1986, the country joined the European Union (EU) which led to the progressive opening of the Spanish economy and triggered a wave of domestic liberalisation meant to bring the Spanish economy into the European mainstream. Moreover, entry into the EU coincided with the most important liberalisation exercise in Europe since the 1960s, namely the implementation of the Single Market Programme. 5 A drop in costs accompanied by a downward adjustment in prices (leading to larger output) typically results in a fall in the elasticity of the residual demand faced by the firm (and therefore, a higher mark-up). Clearly, the particular case of an iso-elastic residual demand is an exception. 6 In addition, a fall in the target firm’s profitability may also occur when FDI is driven by a technology sourcing. This conjecture has received both theoretical and empirical support (for theoretical results, see Siotis (1999), and for empirical evidence Neven and Siotis (1996), Driffield and Love (2003)). However, while sourcing may be a realistic motive, its importance is likely to be very limited compared to “traditional” FDI, particularly in the context of Spain.
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
109
Table 1 Number of observations by sector Sector
Observations
Of which foreign firms
Food, beverages and tobacco Chemicals Mineral products Metal goods Mechanical engineering Electrical and instrument engineering Transport equipment Textiles and clothing Leather and leather products Wood and wooden products Paper, printing and publishing Rubber and plastics Total
5833 3832 2355 2314 2596 1945 1563 3206 907 911 2373 1483 29,318
587 (10.1%) 1312 (34.2%) 256 (10.9%) 253 (10.9%) 579 (22.3%) 459 (23.6%) 483 (30.9%) 199 (6.2%) 31 (3.4%) 17 (1.9%) 231 (9.7%) 265 (17.9%) 4672 (15.9%)
Conceptually, identifying the pro-competitive effect of FDI is more complex, since “competitive pressure” (and changes thereof) is a somewhat nebulous concept that has been proxied by a variety of market characteristics such as the degree of product substitutability, ease of entry, market size, or the number of competitors (Vives, 2006). In our context, we need to identify an explicit mechanism through which the change in ownership leading to cost reductions for the subsidiary may affect pricing behaviour. For instance, imagine that there existed some degree of explicit or tacit collusion within the industry prior to FDI. The change of ownership resulting in a foreign subsidiary enjoying lower costs (and thus increasing asymmetry among agents) may disrupt this pre-existing market equilibrium, at least initially. Indeed, the literature on infinitely repeated games indicates that cost asymmetries make it more difficult to pick a focal equilibrium and therefore may hinder tacit collusion (Tirole, 1988, pp. 250–251). Clearly, the emergence of cost asymmetries is not the only potential channel through which FDI may affect competitive pressure, as becoming part of a multinational may also modify a firm's product mix, both in terms of quality and variety. This conjecture accords well with empirical evidence, as the bulk of FDI is “horizontal” and concentrated in industries were product differentiation is pervasive. 2.2. Testable conjectures The arguments put forward above suggest that a change in ownership from domestic to foreign that is accompanied by the transfer of proprietary assets should bring long-run efficiency and therefore increases in margins. This is more likely to hold in industries where assets such as technology and other intangibles are important. However, there may also exist transient costs associated with the transfer of ownership. In addition, the arrival of a MNC may disrupt tacit collusion, at least in the short-run. Thus, the sign of the short-run effect of a change in ownership is potentially ambiguous. Nonetheless, if the effects described above are at work and our assumptions on the dynamics are correct, we expect the adjustment process for mark-ups to be described by an upward sloping function, possibly steeper for those industries where proprietary assets are more important. The long-run and industry-wide impact on mark-ups of MNC activity in an industry is also ambiguous.7 On the one hand, the existence of large and market-wide positive FDI related spillovers should increase the average mark-ups of host country firms. By contrast, the pro-competitive effect of FDI ought to depress the margins of firms that operate in industries that are characterised by an important foreign presence. Thus, in terms of mark-ups, the competitive effect and the spillovers effect operate in opposite directions. Furthermore, these effects may not necessarily be felt contemporaneously and are both unlikely to have only a transitory impact. However, it seems reasonable to assume that the competitive effect is likely to become effective quickly after the change in ownership, whereas spillovers are more likely to take time to materialise.8 Again, this suggests that for mark-ups, the transition to the long-run impact (which can be positive or negative) should follow an upward sloping adjustment process. Furthermore, this slope should be steeper for industries where proprietary assets are important, at least to the extent that these assets cannot be fully internalized by foreign affiliates. 3. Data and variable definition 3.1. Data Our results are obtained by making use of an extensive survey of firms carried by the Bank of Spain since 1983, gathered in the database Central de Balances. The data collected is comprehensive, each annual cross-section exceeds three thousand observations, and it covers all sectors of economic activity (except for financial institutions). The original data file contains more than 91,000 firm-year observations for the time period 1983–1996. Given sample size, it is possible to impose strict filters, aimed at eliminating extreme observations (replies), or questionnaires for which some of the essential data is missing.9
7
See Barba Navaretti et al. (2004), Chapter 7, for a detailed analysis of the channels through which FDI may affect host country firms. Common examples of spillovers found in the literature include: movement of skilled personnel, MNC subsidiaries acting as “role models” that are emulated by domestic firms, spillovers via common input suppliers, etc... All of the above are likely to take time before their effect can be discerned in the data. 9 A data appendix describing the filtering procedure is available at: http://www.eco.uc3m.es/siotis/investigacion#main. 8
110
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
Table 2 Descriptive statistics Variables
Domestic firms
Foreign subs.
Number of observations
24,646
4672
Price-Cost Margins (PCM) 3-digit Market Share (MS) Intangibles over Sales (ITGS) Foreign Ownership (FO50) 3-digit Foreign Presence (FP)
Mean
Median
Mean
Median
0.102 0.012 0.014 0.015 0.257
0.096 0.003 0.000 0.000 0.248
0.115 0.030 0.011 0.880 0.377
0.111 0.012 0.000 1.000 0.377
In this paper, we focus exclusively on manufacturing and our results are obtained at Central de Balances ' three-digit level. In addition, we apply panel data techniques that require a minimum of four consecutive observations which results in a reduction in sample size. We dropped the few firms that changed sectorial affiliation, the ones belonging to the residual classification “not elsewhere classified”, as well as observations pertaining to 3-digit sectors with less than 100 observations for the entire time period, yielding a total of 32 three-digit industries.10 Table 1 provides the sectorial distribution of the final sample of 29,318 observations of which 4672 (15.9%) refer to foreign firms.11 In particular, inspection of Table 1 reveals—in accordance with the internalisation theory of FDI—that foreign ownership is concentrated in more advanced sectors: chemicals (34.2% of all observations in the sector), transport equipment (30.9%), electrical and instrument engineering (32.6%) and mechanical engineering (22.3%). 3.2. Variable definition Our dependent variable is mark-ups, proxied by accounting price-cost margins. Accounting price-cost margins have been heavily criticised (Bresnahan, 1989). Nonetheless, there is increasing evidence indicating that this measure is not so flawed after all (Martin, 2002). Moreover, in a panel dataset, one can control for all sources of distortions that are either firm-specific but time-invariant, or time varying and common to all firms. Last, we are confident that for our dataset, accounting price-cost margins are a reasonable proxy for economic mark-ups. Siotis (2003) estimates sectorial mark-ups by applying a modified version of Hall (1986) to this dataset. He reports that, apart from a scaling factor, sector wide accounting margins are very similar to mark-ups that are econometrically estimated. Typically, the correlation between sectorial accounting margins and estimated industry mark-ups stands above 0.8. Price-cost margins are defined as: p−c Value added−payroll ¼ PCMit ¼ ð1Þ p it Value added þ net cost of materials it Central de Balances includes data on foreign ownership. More precisely, for each firm in each time period, we know the percentage of equity held by non-residents. We label this ratio F OC (“C ” indicates that this variable is continuous, and ranges from 0 to 1). Since there are not compelling theoretical arguments which suggest to use a continuous variable to measure the “degree” of foreign ownership, we also defined another variable that takes value zero if foreign equity stands below 50%, and 1 if it is equal or above this threshold. We label this variable FO50. In order to proxy for the degree of MNC activity in a given sector, we also define the following variable that we label “foreign presence”: FPit ¼
∑nh¼1;h≠i Sht ⁎FOC ht ∑ni¼1 Sit
ð2Þ
where i, t respectively index firm and time and S denotes nominal sales. The sums in Eq. (2) are computed only over the firms operating in the same 3-digit sector as firm i. In words, FP measures the proportion of output that is foreign controlled within a sector. This implies that we are only able to gauge the importance of intra-industry spillovers. In the empirical specification (see next section), we also introduce market share and a proxy for the degree of product differentiation as control variables. According to most oligopoly models, size differences within an industry reflect differential efficiency. Market share is defined as: MSit ¼
Sit ∑ni¼1 Sit
ð3Þ
where the sum in Eq. (3) is computed over all the firms operating in the same 3-digit industry as firm i. Central de Balances also provides data on the book value of intangible assets held by firms. We therefore construct the variable ITGS as follows: ITGSit ¼
ITGit Sit
ð4Þ
where ITG represents the book value of intangibles. 10 11
For more details, see the data appendix: http://www.eco.uc3m.es/siotis/investigacion#main. For presentational purposes, the sectorial breakdown in Table 1 is at the two-digit level. However, all our estimations are carried out at the 3-digit level.
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
111
Table 3 Basic mark-ups equations PCMt − 1 MSt ITGSt FO50t FO50t − 1 FO50t − 2 FPt FPt − 1 FPt− 2 M1 M2 Sargan Test on joint sig. of ID Test on joint sig. of TD FO50 LR effect FP LR effect
0.426(0.00) 0.282(0.07) 0.102(0.00) −0.108(0.00) 0.093(0.00) 0.013(0.07) −0.019(0.49) 0.004(0.87) 0.032(0.05) − 15.17(0.00) 1.52(0.13) 129.50(0.12) 31.75(0.43) 163.73(0.00) −0.003(0.91) 0.030(0.60)
Note: Time dummies (TD) and three-digit industry (ID) dummies are included as regressors and instruments. Additional instruments are: PCM(2,2), MS(2,3), HHI (2,3), ITGS(2,3), FO50(2,3), FP(2,3). HHI stands for the 3-digit level Hirschman–Herfindahl index of industry concentration: HHIit ¼ ∑ni¼1 ðMSit Þ2 P-values in round brackets. The null hypothesis that each coefficient is equal to zero is tested using one-step robust standard errors. m1(m2) is a test of the null hypothesis of no first (second) order serial correlation. Sargan is a test of the validity of the over-identifying restrictions based on the efficient two-step GMM estimator.
Table 2 presents descriptive statistics pertaining to the variables that we use in the estimation of our mark-ups equations separately for domestic firms and for foreign subsidiaries. On average, foreign firms enjoy higher price-cost margins (11.5% versus 10.2%) and larger market shares (3.0% versus 1.2%). 4. Econometric specification and results 4.1. Specification In order to be able to recover both the short-run and long-run effects of our variables of interest, we estimated the following autoregressive distributed lag model: PCMit ¼ ρPCMit−1 þ xit β þ ∑ps¼0 γ s FO50it−s þ ∑rs¼0 δs FP it−s þ α t þ α i þ υit
ð5Þ
where xit is the vector composed of our control variables (MS and IT GS). αt represents shocks to mark-ups common to all firms in a given year, αi is a firm-specific time-invariant component (possibly correlated with the other right hand-side variables) measuring among other things unobservable management quality, and υit is a random disturbance. All equations are estimated in first-differences to remove the firm-specific effect αi. A set of three-digit industry dummies is however kept in the estimation to allow for industry specific linear time trends in the levels of the dependent variable.12 Estimation is carried out by the Generalized Method of Moments (GMM) proposed by Arellano and Bond (1991). Since all regressors in our model are likely to be correlated with the idiosyncratic component of the error term (υit) OLS (as well as GLS) estimates would be biased and inconsistent, while the GMM methodology provides consistent estimates of the parameters by making use of appropriate instruments. In particular, under the assumption that the idiosyncratic component of the error term, υit is serially uncorrelated in the level equations, an error with a moving structure of order one is generated in the equations in differences, so that once-lagged variables are also correlated with the transformed error term. However, regressors lagged twice or more will be legitimate instruments. To check the validity of this assumption we report tests on both first and second order correlation on the residuals (m1 and m2). If our assumption is correct we expect first but not second order serial correlation in transformed residuals. As a more general test of mis-specification, we also report the Hansen–Sargan test for over-identifying restrictions. The specification presented in Eq. (5) has some attractive characteristics. First, estimation by first-differences ensures that the results are not driven by unobservable firm and industry level characteristics, potentially correlated with some of the observables.13 Second, the availability of potentially valid instruments provides an adequate answer to the standard criticism directed at the “old” cross-sectional empirical literature, namely that mark-ups, market shares and other firm-specific variables
12 This specification ensures that policy changes that affect all firms in an industry (such as the elimination of tariffs and the removal of non-tariff barriers to trade) are controlled for, at least partially. 13 Note that αi accounts for all time-invariant individual effects. The latter may be idiosyncratic to the firm (e.g., management quality), or common to all firms within an industry (e.g., entry barriers or the elasticity of demand).
112
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
Fig. 1. Aggregate dynamic effect on mark-ups of foreign ownership (dotted line) and foreign presence (continuous line).
(such as the degree of product differentiation) are simultaneously determined in oligopoly models. Third, the dynamic structure allows for a distinction between short-run and long-run effects. As argued in Section 2, this dichotomy is important in the context of the research issues addressed in this paper. 4.2. Results 4.2.1. Pooled sample Results for the estimation of Eq. (5) for the full sample (with p = r = 2) are reported in Table 3. Comfortingly, the test statistics indicate that there is evidence of first but not of second order serial correlation. As for the Sargan test, the validity of the instrument set is not rejected at conventional significance levels. The estimates for the time dummies are not reported, but they are jointly highly significant. They indicate a clear downward economy-wide trend in margins.14 Note that by including time dummies, we control for year-specific changes that affect margins across the manufacturing sector (e.g., entry into the EU or broad-based legislative initiatives). As for the 3-digit industry dummies, they are not jointly significant. The coefficients associated with MS and IT GS are both positively signed, although the market share variable is only significant at the 10% level. With respect to our central regressors, the following picture emerges.15 The variable associated with the degree of foreign ownership (FO50) indicates that in the short-run FDI has a negative and significant (at the 1% level) effect on mark-ups. There are a number of non-competing explanations for this finding. First, it may be the case that becoming a MNC's subsidiary involves real costs, particularly in the short-run. The latter may be the result of a re-organisation process and/or of differences between the MNCs' management style (so called “teething cum foreignness” problems). Second, it could also reflect the fact that the arrival of foreign firm that enjoys a cost advantage disrupts tacit collusion, leading to more aggressive pricing behaviour until collusion is reestablished. Third, the initial drop in margins may reflect the tendency to endogenously inflate one's results in the face of a realistic prospect of being taken over by a multinational. Whatever the reason, as can be seen in Fig. 1 where the adjustment path implied by our dynamic specification is pictured against time, this negative effect tends to vanish over the years, indicating that the fall in mark-ups is transient. Indeed a non-linear Wald test confirms that the long-run multiplier turns out to be not significantly different from zero. Taken at their face values, our overall results point out that a change in ownership from domestic to foreign has no longrun effect on mark-ups, a finding in line with those reported by Benfratello and Sembenelli (2006) for Italy. This probably goes against the common wisdom which associates foreign ownership with higher levels of productivity or mark-ups. However, it must be borne in mind that our results do not say that foreign controlled firms do not enjoy higher mark-ups than their domestic counterparts. This may still be the case if, for instance, foreign owners tend to pick up the best domestic firms and concentrate in high mark-up sectors. What our results suggest is that existing descriptive evidence (as well as econometric findings that do not address properly all endogeneity issues) should not be interpreted as a causal relation. As for foreign presence, we find weak evidence that a larger foreign presence dampens margins in the short-run, possibly because it enhances the stance of competition. The coefficient on the contemporaneous variable for foreign presence (FP) is negative, but it is not significant. In addition, as it can be seen from Fig. 1, this effect does not persist over time. Indeed, even if not significantly different from zero, the long-run multiplier is positive. These results are not inconsistent with the conjecture that MNC activity at the industry level generates effects on mark-ups that go in opposite directions, and that are therefore difficult to unearth empirically.16 4.2.2. R&D versus non-R&D intensive industries In what follows, we shed further light on the issues at stake by splitting the sample according to priors. Concretely, we expect direct technology transfers and spillovers to be particularly strong in knowledge intensive sectors. After all, these industries are the ones where spillovers are more likely to materialise, and where multinationals may be expected to transfer intangibles to their subsidiaries. In Central de Balances, only a subset of firms report their R&D spending, and the series is not available before 1986. 14
This is consistent with the findings of Siotis (2003) indicating an increase in competitive pressure during the same sample period. In all specifications we set both p and r equal to 2. We also experimented with alternative lag structures. Additional lags are not significant and all our main results, including the dynamic adjustment paths, are virtually unaltered. 16 We have also experimented with additional controls such concentration, the capital to output ratio and proxies for differences in input costs and quality between domestic firms and foreign subsidiaries. Introducing these variables in the estimation does not affect the essence of our findings. These additional results can be found at: http://www.eco.uc3m.es/siotis/investigacion#main. 15
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
113
Table 4 Extended mark-ups equations (i) PCMt − 1 MSt ITGSt FO50t ⁎ RD FO50t − 1 ⁎ RD FO50t − 2 ⁎ RD FO50t⁎ NRD FO50t − 1 ⁎ NRD FO50t − 2 ⁎ NRD FPt ⁎ RD FPt − 1 ⁎ RD FPt − 2 ⁎ RD FPt ⁎ RD ⁎ FO50t FPt − 1 ⁎ RD ⁎ FO50t − 1 FPt − 2 ⁎ RD ⁎ FO50t − 2 FPt ⁎ RD ⁎ (1 − FO50)t FPt − 1 ⁎ RD ⁎ (1 − FO50)t − 1 FPt − 2 ⁎ RD ⁎ (1 − FO50)t − 2 FPt ⁎ NRD FPt − 1 ⁎ NRD FPt − 2 ⁎ NRD M1 m2 Sargan Test on joint sig. of ID Test on joint sig. of TD RD FO50 LR effect Non-RD FO50 LR effect Test on FO50 restriction RD FP LR effect Non-RD FP LR effect Test on RD FP restriction RD FP (1 − FO50) LR effect RD FP FO50 LR effect Test on RD FP restriction
0.422(0.00) 0.261(0.10) 0.098(0.00) 0.037(0.34) 0.031(0.41) 0.028(0.08) −0.113(0.01) 0.091(0.01) 0.009(0.22) 0.040(0.32) 0.019(0.56) 0.072(0.12)
−0.046(0.09) 0.012(0.66) 0.017(0.22) −14.98(0.00) 1.43(0.15) 187.68(0.02) 32.83(0.38) 168.27(0.00) 0.166(0.05) −0.022(0.54) 0.188(0.04) 0.227(0.12) −0.029(0.54) 0.256(0.10)
(ii) 0.419(0.00) 0.293(0.053) 0.099(0.00) 0.056(0.27) 0.019(0.74) 0.091(0.01) −0.111(0.01) 0.090(0.01) 0.001(0.22)
0.021(0.65) 0.005(0.92) −0.027(0.52) 0.052(0.33) 0.027(0.56) 0.113(0.05) −0.049(0.07) 0.012(0.64) 0.016(0.22) −15.10(0.00) 1.42(0.16) 201.51(0.04) 32.96(0.37) 167.75(0.00) 0.286(0.02) −0.034(0.57) 0.320(0.02) −0.036(0.45) 0.330(0.08) −0.002(0.99) 0.332(0.06)
Note: All estimates include time dummies (TD) and three-digit industry (ID) dummies as regressors and instruments. Additional instruments are: in column (i) PCM(2,2), MS(2,3), HHI(2,3), ITGS(2,3), FO50 ⁎ RD(2,3), FO50 ⁎ NRD(2,3), FP ⁎ RD(2,3), FP ⁎ NRD(2,3); in column (ii) PCM(2,2), MS(2,3), HHI(2,3), ITGS(2,3), FO50 ⁎ RD (2,3), FO50 ⁎ NRD(2,3), FP ⁎ RD ⁎ FO50(2,3), FP ⁎ RD ⁎ (1 − FO50)(2,3),FP ⁎ NRD(2,3). See the note to Table 3 for the interpretation of reported test statistics.
Nevertheless, this data enables us to construct an annual proxy for R&D intensity at the 3-digit level for the period 1986–1996. Yearly sectorial R&D intensity is defined as: RDIjt ¼
∑ni¼1 RDEit ∑ni¼1 Sit
ð6Þ
where RDE is R&D expenditure at the firm level. Thus, for each year, RDIjt takes a single value for each 3-digit sector. We have then computed the average over the time period for which R&D expenditure is available: RDIj ¼
∑ki¼1 RDIjt k
ð7Þ
where k denotes the years for which R&D data is available (1986–1996, i.e.11 years). The splitting criteria that we applied to define R&D intensive sectors is an intensity greater or equal than 2%.17 Four industries fall in this category (pharmaceuticals, electronics, precision instruments, and aerospace), and they account for 2184 observations. Two dummies were constructed accordingly:
RD ¼ 1 if RDIj z0:02 NRD ¼ 1−RD if RDIj <0:02
ð8Þ
Both RD and NRD were interacted with FO50 and FP. 17 The choice of the 2% cut-off point is mainly motivated by a close inspection at the empirical distribution of sector-level R&D intensities. We computed average R&D intensities for all sectors. 26 out of 32 turn out to have average research intensities below 1% and only 4 above 2%, the two remaining sectors respectively scoring 1.6% (naval construction) and 1.1% (agricultural and industrial machinery). What the data seem to suggest us is that 2% is a sort of “natural” cut-off point. Potential alternatives would be to choose 1.5 % or 1% cut-off points thus including “naval construction” and possibly “agricultural and industrial machinery” as additional R&D intensive industries (although casual empiricism suggests that firms operating in these industries are not big R&D spenders). As robustness checks we re-ran all our equations by using 1.5% or 1.0% as alternative cut-off points. Results are qualitatively similar but point coefficients are less precisely estimated, especially with the 1% cut-off.
114
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
Fig. 2. Dynamic effect on mark-ups of foreign ownership—R&D (continuous line) versus non-R&D (dotted line) intensive industries.
Fig. 3. Dynamic effect on mark-ups of foreign presence—R&D (continuous line) versus non-R&D (dotted line) intensive industries.
Results are presented in column (i) of Table 4. As before, there is no evidence of second order auto-correlation of the errors and, the Sargan mis-specification test does not reject the null, tough only at the 1% level. Both ITGS and MS maintain their sign. The latter variable is less precisely estimated, while the former remains highly significant. As for the variables of interest, in non-R&D intensive sectors, a change in ownership has no long-run effect on mark-ups. However, the pattern described in the previous section is maintained: margins initially fall, and this effect is significant at the 1% level. After one lag, margins significantly recover. With the 2-period lag also being positive (though not significant), the long-run effect is negative, but not significantly different from zero as indicated by the Wald test and the dynamic adjustment path depicted in Fig. 2. Regarding foreign presence in nonR&D intensive sectors, the contemporaneous effect is negative and marginally significant. We take this as evidence that an increase in MNC presence increases competitive pressure in the short-run. However, this effect vanishes over time, as indicated by the adjustment path and the Wald test. This is consistent with the conjecture that FDI both increases competitive pressure and generates positive externalities for host country firms (see Fig. 3). The results for the four R&D intensive industries are significantly different from the results we find for the sample which includes all other industries. The tests on the equality of the long-run effects between the two samples of industries reject the restrictions of equal coefficients both for the foreign ownership (FO50) and for the foreign presence (FP) variables. With regard to foreign ownership, we find a statistically significant positive effect that takes time to materialise (the 2-period lag coefficient is the most significant). This is in line with the idea that MNCs do transfer firm-specific assets to their subsidiaries in these industries, and also that there is a learning period before these assets are successfully exploited (see also Fig. 2). While not individually significant at standard confidence levels, the coefficients associated with foreign presence indicate that the long-run effect on mark-ups of MNC presence is always positive and marginally significant. This finding suggests that the positive spillover effect dominates the procompetitive effect in the case of R&D intensive industries.18 Furthermore, the adjustment path pictured in Fig. 3 is upward sloping, thus pointing out that it indeed takes time for spillovers to materialise. 4.2.3. Competition versus spillovers in R&D intensive industries We further explore these issues by exploiting an additional identification condition. If technological spillovers are indeed present in R&D intensive industries, we would expect domestic firms to be the main beneficiaries. This hypothesis is motivated by the fact that Spanish firms are more likely to lag foreign subsidiaries. While the origin country of MNCs is unavailable in Central de Balances, aggregate FDI figures indicate that the main investors come from the US and more advanced Western European economies (e.g., from France, Germany and the UK). It is therefore likely that spillovers will primarily stem from subsidiaries to domestic entities. Clearly, this does not preclude positive technological externalities flowing across foreign subsidiaries; our conjecture is simply that the spillover effect will be felt more acutely by domestic firms. Operationally, we interacted the foreign ownership dummy (FO50) with our measure of foreign presence (FP) in R&D intensive sectors. These additional results are presented in column (ii) of Table 4. As before, the traditional determinants of mark-ups, namely 18 We are faced with a potential identification problem in interpreting these findings. In the four R&D sectors that we identified, it could be the case that the increase in profitability experienced by domestic firms is a by-product of an increase in R&D effort on their part, partially spurred by the increased presence of foreign firms. Data limitations prevent us from directly tackling this issue. However, OECD data (ANBERD database, 2000 release) do not suggest that Spanish firms belonging to the four sectors labelled as R&D intensive significantly increased their R&D outlays during our sample period. The same holds true for firms that reported R&D expenditure in Central de Balances.
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
115
Fig. 4. Dynamic effect on mark-ups of foreign presence in R&D intensive industries—Domestic Firms (continuous line) versus Foreign Firms (dotted line).
market share and product differentiation, have the expected sign. As for our variables of interest, their sign, significance, and long vs. short-run effects are the same as before in non-R&D intensive sectors.19 In R&D intensive sectors, the direct effect of foreign ownership (FO50 ⁎ RD) continues to be positive and significant in the long-run. Regarding foreign presence, we find that its long-run effect is positive and significant only for domestic firms. Therefore, it appears that domestic firms are the main beneficiaries of an increase in multinational activity (see also Fig. 4 where the adjustment process based on the results in column (ii) is plotted for domestic firms and foreign subsidiaries). These results are consistent with the conjecture that, as compared to their foreign-owned counterparts, domestic firms belonging to R&D intensive sectors have been the main beneficiaries of spillovers. 5. Conclusions In this paper, we have attempted to disentangle some of the effects usually attributed to FDI in the spirit of the conjectures first put forward by Caves (1974). On the one hand, the fact that MNCs possess firm-specific advantages that can be transferred back and forth across locations suggest that subsidiaries ought to enjoy greater levels of efficiency, and therefore mark-ups. Overall, we find support for this conjecture, but this is limited to R&D intensive sectors. For the rest of manufacturing, the long-run effect of a change from domestic to foreign ownership is nil. Furthermore, and in line with the existing literature, we find evidence of transient costs associated with a change in ownership. With regard to the impact of foreign presence on mark-ups, the dichotomy between R&D and non-R&D sectors is also present. For non-R&D sectors, our results indicate that increased multinational presence dampens margins in the short-run. However, this effect vanishes over time, a finding that is consistent with MNCs also generating positive externalities for host country firms. This conclusion is further supported by the results pertaining to the impact of foreign presence in R&D intensive sectors. In the latter case, the positive spillover effect dominates. Finally, we find evidence consistent with the idea that domestic firms belonging to R&D intensive sectors are the main beneficiaries of spillovers. This should come as no surprise, given that Spanish entities are likely to lag their foreign-owned counterparts. References Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies 58, 277–297. Barba Navaretti, G., Venables, A., et al., 2004. Multinational Firms in the World Economy. Princeton University Press. Benfratello, L., Sembenelli, A., 2006. Foreign direct investment and productivity: is the direction of causality so obvious? International Journal of Industrial Organization 24 (2), 735–751. Bloom, N., Schankerman, M., Van Reenen, J., 2005. Identifying technology spillovers and product market rivalry. CEPR Discussion Paper #4912, February. Bresnahan, T., 1989. Empirical studies of industries with market power. In: Schmalensee, R., Willig, R. (Eds.), Handbook of Industrial Organisation, Vol. 2. North Holland, Amsterdam, pp. 1011–1057. Caves, R., 1974. Multinational firms, competition, and productivity in host-country markets. Economica 41 (162), 176–193. Dri eld, N., Love, J., 2003. Foreign direct investment, technology sourcing and reverse spillovers. The Manchester School 71 (6), 659–672. Hall, R., 1986. Market structure and macroeconomics fluctuations. Brookings Papers on Economic Activity 285–338. Harris, R., Robinson, C., 2002. The impact of foreign acquisitions on total factor productivity: plant-level evidence form UK manufacturing, 1987–1992. The Review of Economics and Statistics 84 (3), 562–568. Kokko, A., 1996. Productivity spillovers competition between local firms and foreign affiliates. Journal of International Development 8 (4), 517–530. Markusen, J., 1995. The boundaries of multinational enterprises and the theory of international trade. Journal of Economic Perspectives 9, 169–189. Martin, S., 2002. Advanced Industrial Economics. Blackwell Publishers, Oxford, UK. Neven, D., Siotis, G., 1996. Technology sourcing and FDI in the EC: an empirical evaluation. International Journal of Industrial Organization 14 (5), 543–560. Siotis, G., 1999. FDI strategies and firms' capabilities. Journal of Economics and Management Strategy 8 (2), 251–270. Siotis, G., 2003. Competitive pressure and economic integration: an illustration for Spain, 1983–1996. International Journal of Industrial Organization 21 (10), 1435–1459. Tirole, J., 1988. The Theory of Industrial Organization. MIT Press. Vives, X., 2006. Innovation and competitive pressure. IESE SP-SP Working Paper #634.
19 We also re-ran our estimations by interacting the foreign and domestic dummy variables in non-R&D intensive sectors. No marked difference emerged between domestically owned concerns and foreign subsidiaries.
Contents lists available at ScienceDirect
Journal of International Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e c o n b a s e
Foreign Direct Investment and mark-up dynamics: Evidence from Spanish firms☆ Alessandro Sembenelli a,b, Georges Siotis c,d,⁎ a b c d
Dipartimento di Scienze Economiche e Finanziarie, Facoltà di Economia, Università di Torino, Corso Unione Sovietica 218bis, 10134 Torino, Italy Collegio Carlo Alberto, Italy Departamento de Economía, Universidad Carlos III de Madrid, Calle Madrid 126, 28903 Getafe, Madrid, Spain CEPR, United Kingdom
a r t i c l e
i n f o
Article history: Received 18 March 2004 Received in revised form 8 February 2008 Accepted 9 May 2008 Keywords: Foreign Direct Investment Technology transfer Spillovers Efficiency Competition Panel data GMM
a b s t r a c t A review of the literature indicates that Foreign Direct Investment has the potential to increase the intensity of competition and to act as a channel for technology transfers. Using a Spanish firm level data set, we disentangle these effects by estimating a dynamic model of firm level performance, which we proxy by mark-ups. We find that FDI has a positive long-run effect on the mark-ups of targets, but this is limited to firms in R&D intensive sectors. In addition, we find weak evidence that foreign presence dampens margins. However, this effect appears to be more than compensated by positive spillovers in the case of knowledge intensive industries. © 2008 Elsevier B.V. All rights reserved.
JEL codes: F23 L40 L60
1. Introduction In an early pioneering contribution, Caves (1974) put forward the idea that Foreign Direct Investment (FDI) influenced host country conditions through two main channels. On the one hand, FDI ought to result in technology transfers to host country firms. This effect was conjectured to be both direct—Multinational Corporations (MNCs) providing subsidiaries an efficiency advantage— and indirect—MNCs generating positive spillovers. On the other hand, an important foreign presence could also increase the intensity of competition in the recipient country.1 While the latter conjecture is intuitively appealing, direct empirical evidence of
☆ We would like to thank Editor Jim Tybout, two anonymous referees, Luigi Benfratello, Pedro L. Marín, and Reinhilde Veugelers for helpful comments. This paper was begun while Siotis was a Research Fellow of the Bank of Spain’s Research Department and he expresses his gratitude to the staff of the Bank of Spain for their help with the data. Siotis acknowledges partial support for this project from the EU financed TMR pro ject “FDI and the Multinational Corporation” and from Ministerio de Educación y Ciencia (SEJ2007-66268). Previous versions of this paper were circulated under the title “Foreign Direct Investment, Competitive Pressure, and Spillovers. An Empirical Analysis on Spanish Firm Level Data”. ⁎ Corresponding author. Departamento de Economía, Universidad Carlos III de Madrid, Calle Madrid 126, 28903 Getafe, Madrid, Spain. Tel.: +34 91 624 93 12; fax: +34 91 624 98 75. E-mail addresses: [email protected] (A. Sembenelli), [email protected] (G. Siotis). 1 Caves' insights can be summarised by the following sentences: “The host nation’s private sector does not benefit directly because the foreign subsidiary is efficient, or brings to its shores skilled entrepreneurship or productive knowledge. Rather its gains depend on spill-overs of productivity that occur when the multinational corporation cannot capture all quasi-rents due to its productive activities, or to the removal of distortions by the subsidiary’s competitive pressure”, Caves (1974, p. 176). 0022-1996/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.jinteco.2008.05.003
108
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
the pro-competitive effect of FDI is limited, if not non-existent.2 By and large, the empirical literature has focused on the technological effect, possibly because unearthing the pro-competitive effect of FDI is not trivial.3 The purpose of this paper is to attempt to disentangle empirically the efficiency, spillovers, and competition effects of FDI on firms' mark-ups. We chose mark-ups as a proxy for firm performance for three reasons. First, mark-ups are a natural choice when attempting to gauge changes in competitive pressure. Second, this measure captures the evolution of price as well as costs, both of which will be affected by the FDI induced effects identified above. Third, this variable is less prone to measurement problems stemming from the use of industry-wide deflators, as it directly reflects firms' pricing behaviour. The empirical analysis is carried out with Spanish firm level data over the period 1983–1996. Potentially, Spain's experience represents an interesting case to test the conjectures presented above as a number of factors contributed to a large increase in competitive pressure and a surge in FDI.4 The increase in competitive pressure is documented in Siotis (2003), where a drop in (econometrically estimated) industry mark-ups is identified as a result of Spain's entry into the EU. The main innovations contained in this paper are as follows. First, we use price-cost margins rather than productivity (in the remainder of the paper, we use mark-ups and price-cost margins interchangeably). Second, we focus on the dynamic responses of the dependent variable. In turn, these responses form our main identification argument. Concretely, we conjecture that the effects identified above will work their way through at a different pace. Third, we split our sample using R&D intensity, which provides an additional identifying hypothesis. Fourth, we use the identity of firms as a further check on the robustness of our results. Since Spanish owned firms typically lag foreign subsidiaries along the technological dimension, domestic firms (as opposed to foreign subsidiaries) are likely to be the main beneficiaries of spillovers. Fifth, we apply the Generalised Method of Moments (GMM) in order to deal with endogeneity biases. This also allows us to properly account for industry and firm level fixed effects. We find that after controlling for potential endogeneity biases and economy-wide effects, FDI has a positive long-run effect on the mark-ups of target firms, but this is limited to R&D intensive sectors. In addition, domestically owned firms are the main recipient of spillovers in knowledge intensive industries. Last, the results weakly indicate that an important foreign presence dampens margins, at least in the short-run. However, in the case of R&D intensive industries, this appears to be more than compensated by positive spillovers. The rest of the paper is organised as follows. The next section indicates why, under fairly general conditions, FDI could be expected to act as a channel for technology transfers as well as influence the intensity of competition. Section 3 presents the data and describes how we constructed the variables. Section 4 contains the econometric specification as well the main results. Section 5 concludes. 2. Motivation and testable conjectures 2.1. Technology transfers and competition The fact that MNCs possess firm-specific assets that confer them a competitive edge is well established in the literature (Markusen, 1995). If these assets are transferred to the subsidiary, the latter will be more efficient and, as a general rule, enjoy higher margins (the “direct” effect). For instance, this obtains in the case of Cournot or a Bertrand model with product differentiation, constant marginal cost and with firms facing a linear residual demand curve.5 However, the existence of important costs associated with a change of management may temporarily lead to lower margins, particularly if it involves a foreign firm. This may result because of differences in culture, language, or an inadequate knowledge of consumer preferences, that is, there may be a short-run drop in mark-ups driven by “teething problems compounded by foreignness” (Harris and Robinson, 2002).6 It is also a common observation that MNCs have the potential to generate positive spillovers in the host location (the “indirect effect”; see Barba Navaretti et al. (2004) for an extensive analysis). In the event that these spillovers are large and market-wide, host country competitors are likely to experience an increase in mark-ups (for instance, this is the case in the models mentioned above with complete spillovers). However, even within the simple class of models we refer to, the predictions regarding industrywide margins are less clear-cut when spillovers are limited and/or only affect a subset of firms. The latter may occur if only those firms which are sufficiently “close” to the source (with “closeness” defined as technological, product, or geographical proximity) can appropriate the positive externalities generated by MNCs. 2 In his industry level study, Caves (1974) found that the profitability of Canadian domestic plants during the period 1965–67 was negatively correlated with the average share of foreign plants in industry sales. This finding was interpreted as evidence of the pro-competitive effect of FDI. Clearly, this finding cannot be given a causal interpretation. In a similar vein, Kokko (1996) simultaneously estimates the labour productivity determinants of foreign owned and domestic plants in Mexico for the year 1970. He reports that labour productivity in domestic plants is positively correlated with productivity in foreign plants, but this holds only for a subset of industries. This finding is interpreted as pointing to the existence of competition related spillovers. 3 A recent exception is provided by Bloom Schankerman and Van Reenen (2005) who show that R&D activity generates two types of externalities on rivals: technological spillovers and an increase in product market rivalry. 4 In 1986, the country joined the European Union (EU) which led to the progressive opening of the Spanish economy and triggered a wave of domestic liberalisation meant to bring the Spanish economy into the European mainstream. Moreover, entry into the EU coincided with the most important liberalisation exercise in Europe since the 1960s, namely the implementation of the Single Market Programme. 5 A drop in costs accompanied by a downward adjustment in prices (leading to larger output) typically results in a fall in the elasticity of the residual demand faced by the firm (and therefore, a higher mark-up). Clearly, the particular case of an iso-elastic residual demand is an exception. 6 In addition, a fall in the target firm’s profitability may also occur when FDI is driven by a technology sourcing. This conjecture has received both theoretical and empirical support (for theoretical results, see Siotis (1999), and for empirical evidence Neven and Siotis (1996), Driffield and Love (2003)). However, while sourcing may be a realistic motive, its importance is likely to be very limited compared to “traditional” FDI, particularly in the context of Spain.
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
109
Table 1 Number of observations by sector Sector
Observations
Of which foreign firms
Food, beverages and tobacco Chemicals Mineral products Metal goods Mechanical engineering Electrical and instrument engineering Transport equipment Textiles and clothing Leather and leather products Wood and wooden products Paper, printing and publishing Rubber and plastics Total
5833 3832 2355 2314 2596 1945 1563 3206 907 911 2373 1483 29,318
587 (10.1%) 1312 (34.2%) 256 (10.9%) 253 (10.9%) 579 (22.3%) 459 (23.6%) 483 (30.9%) 199 (6.2%) 31 (3.4%) 17 (1.9%) 231 (9.7%) 265 (17.9%) 4672 (15.9%)
Conceptually, identifying the pro-competitive effect of FDI is more complex, since “competitive pressure” (and changes thereof) is a somewhat nebulous concept that has been proxied by a variety of market characteristics such as the degree of product substitutability, ease of entry, market size, or the number of competitors (Vives, 2006). In our context, we need to identify an explicit mechanism through which the change in ownership leading to cost reductions for the subsidiary may affect pricing behaviour. For instance, imagine that there existed some degree of explicit or tacit collusion within the industry prior to FDI. The change of ownership resulting in a foreign subsidiary enjoying lower costs (and thus increasing asymmetry among agents) may disrupt this pre-existing market equilibrium, at least initially. Indeed, the literature on infinitely repeated games indicates that cost asymmetries make it more difficult to pick a focal equilibrium and therefore may hinder tacit collusion (Tirole, 1988, pp. 250–251). Clearly, the emergence of cost asymmetries is not the only potential channel through which FDI may affect competitive pressure, as becoming part of a multinational may also modify a firm's product mix, both in terms of quality and variety. This conjecture accords well with empirical evidence, as the bulk of FDI is “horizontal” and concentrated in industries were product differentiation is pervasive. 2.2. Testable conjectures The arguments put forward above suggest that a change in ownership from domestic to foreign that is accompanied by the transfer of proprietary assets should bring long-run efficiency and therefore increases in margins. This is more likely to hold in industries where assets such as technology and other intangibles are important. However, there may also exist transient costs associated with the transfer of ownership. In addition, the arrival of a MNC may disrupt tacit collusion, at least in the short-run. Thus, the sign of the short-run effect of a change in ownership is potentially ambiguous. Nonetheless, if the effects described above are at work and our assumptions on the dynamics are correct, we expect the adjustment process for mark-ups to be described by an upward sloping function, possibly steeper for those industries where proprietary assets are more important. The long-run and industry-wide impact on mark-ups of MNC activity in an industry is also ambiguous.7 On the one hand, the existence of large and market-wide positive FDI related spillovers should increase the average mark-ups of host country firms. By contrast, the pro-competitive effect of FDI ought to depress the margins of firms that operate in industries that are characterised by an important foreign presence. Thus, in terms of mark-ups, the competitive effect and the spillovers effect operate in opposite directions. Furthermore, these effects may not necessarily be felt contemporaneously and are both unlikely to have only a transitory impact. However, it seems reasonable to assume that the competitive effect is likely to become effective quickly after the change in ownership, whereas spillovers are more likely to take time to materialise.8 Again, this suggests that for mark-ups, the transition to the long-run impact (which can be positive or negative) should follow an upward sloping adjustment process. Furthermore, this slope should be steeper for industries where proprietary assets are important, at least to the extent that these assets cannot be fully internalized by foreign affiliates. 3. Data and variable definition 3.1. Data Our results are obtained by making use of an extensive survey of firms carried by the Bank of Spain since 1983, gathered in the database Central de Balances. The data collected is comprehensive, each annual cross-section exceeds three thousand observations, and it covers all sectors of economic activity (except for financial institutions). The original data file contains more than 91,000 firm-year observations for the time period 1983–1996. Given sample size, it is possible to impose strict filters, aimed at eliminating extreme observations (replies), or questionnaires for which some of the essential data is missing.9
7
See Barba Navaretti et al. (2004), Chapter 7, for a detailed analysis of the channels through which FDI may affect host country firms. Common examples of spillovers found in the literature include: movement of skilled personnel, MNC subsidiaries acting as “role models” that are emulated by domestic firms, spillovers via common input suppliers, etc... All of the above are likely to take time before their effect can be discerned in the data. 9 A data appendix describing the filtering procedure is available at: http://www.eco.uc3m.es/siotis/investigacion#main. 8
110
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
Table 2 Descriptive statistics Variables
Domestic firms
Foreign subs.
Number of observations
24,646
4672
Price-Cost Margins (PCM) 3-digit Market Share (MS) Intangibles over Sales (ITGS) Foreign Ownership (FO50) 3-digit Foreign Presence (FP)
Mean
Median
Mean
Median
0.102 0.012 0.014 0.015 0.257
0.096 0.003 0.000 0.000 0.248
0.115 0.030 0.011 0.880 0.377
0.111 0.012 0.000 1.000 0.377
In this paper, we focus exclusively on manufacturing and our results are obtained at Central de Balances ' three-digit level. In addition, we apply panel data techniques that require a minimum of four consecutive observations which results in a reduction in sample size. We dropped the few firms that changed sectorial affiliation, the ones belonging to the residual classification “not elsewhere classified”, as well as observations pertaining to 3-digit sectors with less than 100 observations for the entire time period, yielding a total of 32 three-digit industries.10 Table 1 provides the sectorial distribution of the final sample of 29,318 observations of which 4672 (15.9%) refer to foreign firms.11 In particular, inspection of Table 1 reveals—in accordance with the internalisation theory of FDI—that foreign ownership is concentrated in more advanced sectors: chemicals (34.2% of all observations in the sector), transport equipment (30.9%), electrical and instrument engineering (32.6%) and mechanical engineering (22.3%). 3.2. Variable definition Our dependent variable is mark-ups, proxied by accounting price-cost margins. Accounting price-cost margins have been heavily criticised (Bresnahan, 1989). Nonetheless, there is increasing evidence indicating that this measure is not so flawed after all (Martin, 2002). Moreover, in a panel dataset, one can control for all sources of distortions that are either firm-specific but time-invariant, or time varying and common to all firms. Last, we are confident that for our dataset, accounting price-cost margins are a reasonable proxy for economic mark-ups. Siotis (2003) estimates sectorial mark-ups by applying a modified version of Hall (1986) to this dataset. He reports that, apart from a scaling factor, sector wide accounting margins are very similar to mark-ups that are econometrically estimated. Typically, the correlation between sectorial accounting margins and estimated industry mark-ups stands above 0.8. Price-cost margins are defined as: p−c Value added−payroll ¼ PCMit ¼ ð1Þ p it Value added þ net cost of materials it Central de Balances includes data on foreign ownership. More precisely, for each firm in each time period, we know the percentage of equity held by non-residents. We label this ratio F OC (“C ” indicates that this variable is continuous, and ranges from 0 to 1). Since there are not compelling theoretical arguments which suggest to use a continuous variable to measure the “degree” of foreign ownership, we also defined another variable that takes value zero if foreign equity stands below 50%, and 1 if it is equal or above this threshold. We label this variable FO50. In order to proxy for the degree of MNC activity in a given sector, we also define the following variable that we label “foreign presence”: FPit ¼
∑nh¼1;h≠i Sht ⁎FOC ht ∑ni¼1 Sit
ð2Þ
where i, t respectively index firm and time and S denotes nominal sales. The sums in Eq. (2) are computed only over the firms operating in the same 3-digit sector as firm i. In words, FP measures the proportion of output that is foreign controlled within a sector. This implies that we are only able to gauge the importance of intra-industry spillovers. In the empirical specification (see next section), we also introduce market share and a proxy for the degree of product differentiation as control variables. According to most oligopoly models, size differences within an industry reflect differential efficiency. Market share is defined as: MSit ¼
Sit ∑ni¼1 Sit
ð3Þ
where the sum in Eq. (3) is computed over all the firms operating in the same 3-digit industry as firm i. Central de Balances also provides data on the book value of intangible assets held by firms. We therefore construct the variable ITGS as follows: ITGSit ¼
ITGit Sit
ð4Þ
where ITG represents the book value of intangibles. 10 11
For more details, see the data appendix: http://www.eco.uc3m.es/siotis/investigacion#main. For presentational purposes, the sectorial breakdown in Table 1 is at the two-digit level. However, all our estimations are carried out at the 3-digit level.
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
111
Table 3 Basic mark-ups equations PCMt − 1 MSt ITGSt FO50t FO50t − 1 FO50t − 2 FPt FPt − 1 FPt− 2 M1 M2 Sargan Test on joint sig. of ID Test on joint sig. of TD FO50 LR effect FP LR effect
0.426(0.00) 0.282(0.07) 0.102(0.00) −0.108(0.00) 0.093(0.00) 0.013(0.07) −0.019(0.49) 0.004(0.87) 0.032(0.05) − 15.17(0.00) 1.52(0.13) 129.50(0.12) 31.75(0.43) 163.73(0.00) −0.003(0.91) 0.030(0.60)
Note: Time dummies (TD) and three-digit industry (ID) dummies are included as regressors and instruments. Additional instruments are: PCM(2,2), MS(2,3), HHI (2,3), ITGS(2,3), FO50(2,3), FP(2,3). HHI stands for the 3-digit level Hirschman–Herfindahl index of industry concentration: HHIit ¼ ∑ni¼1 ðMSit Þ2 P-values in round brackets. The null hypothesis that each coefficient is equal to zero is tested using one-step robust standard errors. m1(m2) is a test of the null hypothesis of no first (second) order serial correlation. Sargan is a test of the validity of the over-identifying restrictions based on the efficient two-step GMM estimator.
Table 2 presents descriptive statistics pertaining to the variables that we use in the estimation of our mark-ups equations separately for domestic firms and for foreign subsidiaries. On average, foreign firms enjoy higher price-cost margins (11.5% versus 10.2%) and larger market shares (3.0% versus 1.2%). 4. Econometric specification and results 4.1. Specification In order to be able to recover both the short-run and long-run effects of our variables of interest, we estimated the following autoregressive distributed lag model: PCMit ¼ ρPCMit−1 þ xit β þ ∑ps¼0 γ s FO50it−s þ ∑rs¼0 δs FP it−s þ α t þ α i þ υit
ð5Þ
where xit is the vector composed of our control variables (MS and IT GS). αt represents shocks to mark-ups common to all firms in a given year, αi is a firm-specific time-invariant component (possibly correlated with the other right hand-side variables) measuring among other things unobservable management quality, and υit is a random disturbance. All equations are estimated in first-differences to remove the firm-specific effect αi. A set of three-digit industry dummies is however kept in the estimation to allow for industry specific linear time trends in the levels of the dependent variable.12 Estimation is carried out by the Generalized Method of Moments (GMM) proposed by Arellano and Bond (1991). Since all regressors in our model are likely to be correlated with the idiosyncratic component of the error term (υit) OLS (as well as GLS) estimates would be biased and inconsistent, while the GMM methodology provides consistent estimates of the parameters by making use of appropriate instruments. In particular, under the assumption that the idiosyncratic component of the error term, υit is serially uncorrelated in the level equations, an error with a moving structure of order one is generated in the equations in differences, so that once-lagged variables are also correlated with the transformed error term. However, regressors lagged twice or more will be legitimate instruments. To check the validity of this assumption we report tests on both first and second order correlation on the residuals (m1 and m2). If our assumption is correct we expect first but not second order serial correlation in transformed residuals. As a more general test of mis-specification, we also report the Hansen–Sargan test for over-identifying restrictions. The specification presented in Eq. (5) has some attractive characteristics. First, estimation by first-differences ensures that the results are not driven by unobservable firm and industry level characteristics, potentially correlated with some of the observables.13 Second, the availability of potentially valid instruments provides an adequate answer to the standard criticism directed at the “old” cross-sectional empirical literature, namely that mark-ups, market shares and other firm-specific variables
12 This specification ensures that policy changes that affect all firms in an industry (such as the elimination of tariffs and the removal of non-tariff barriers to trade) are controlled for, at least partially. 13 Note that αi accounts for all time-invariant individual effects. The latter may be idiosyncratic to the firm (e.g., management quality), or common to all firms within an industry (e.g., entry barriers or the elasticity of demand).
112
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
Fig. 1. Aggregate dynamic effect on mark-ups of foreign ownership (dotted line) and foreign presence (continuous line).
(such as the degree of product differentiation) are simultaneously determined in oligopoly models. Third, the dynamic structure allows for a distinction between short-run and long-run effects. As argued in Section 2, this dichotomy is important in the context of the research issues addressed in this paper. 4.2. Results 4.2.1. Pooled sample Results for the estimation of Eq. (5) for the full sample (with p = r = 2) are reported in Table 3. Comfortingly, the test statistics indicate that there is evidence of first but not of second order serial correlation. As for the Sargan test, the validity of the instrument set is not rejected at conventional significance levels. The estimates for the time dummies are not reported, but they are jointly highly significant. They indicate a clear downward economy-wide trend in margins.14 Note that by including time dummies, we control for year-specific changes that affect margins across the manufacturing sector (e.g., entry into the EU or broad-based legislative initiatives). As for the 3-digit industry dummies, they are not jointly significant. The coefficients associated with MS and IT GS are both positively signed, although the market share variable is only significant at the 10% level. With respect to our central regressors, the following picture emerges.15 The variable associated with the degree of foreign ownership (FO50) indicates that in the short-run FDI has a negative and significant (at the 1% level) effect on mark-ups. There are a number of non-competing explanations for this finding. First, it may be the case that becoming a MNC's subsidiary involves real costs, particularly in the short-run. The latter may be the result of a re-organisation process and/or of differences between the MNCs' management style (so called “teething cum foreignness” problems). Second, it could also reflect the fact that the arrival of foreign firm that enjoys a cost advantage disrupts tacit collusion, leading to more aggressive pricing behaviour until collusion is reestablished. Third, the initial drop in margins may reflect the tendency to endogenously inflate one's results in the face of a realistic prospect of being taken over by a multinational. Whatever the reason, as can be seen in Fig. 1 where the adjustment path implied by our dynamic specification is pictured against time, this negative effect tends to vanish over the years, indicating that the fall in mark-ups is transient. Indeed a non-linear Wald test confirms that the long-run multiplier turns out to be not significantly different from zero. Taken at their face values, our overall results point out that a change in ownership from domestic to foreign has no longrun effect on mark-ups, a finding in line with those reported by Benfratello and Sembenelli (2006) for Italy. This probably goes against the common wisdom which associates foreign ownership with higher levels of productivity or mark-ups. However, it must be borne in mind that our results do not say that foreign controlled firms do not enjoy higher mark-ups than their domestic counterparts. This may still be the case if, for instance, foreign owners tend to pick up the best domestic firms and concentrate in high mark-up sectors. What our results suggest is that existing descriptive evidence (as well as econometric findings that do not address properly all endogeneity issues) should not be interpreted as a causal relation. As for foreign presence, we find weak evidence that a larger foreign presence dampens margins in the short-run, possibly because it enhances the stance of competition. The coefficient on the contemporaneous variable for foreign presence (FP) is negative, but it is not significant. In addition, as it can be seen from Fig. 1, this effect does not persist over time. Indeed, even if not significantly different from zero, the long-run multiplier is positive. These results are not inconsistent with the conjecture that MNC activity at the industry level generates effects on mark-ups that go in opposite directions, and that are therefore difficult to unearth empirically.16 4.2.2. R&D versus non-R&D intensive industries In what follows, we shed further light on the issues at stake by splitting the sample according to priors. Concretely, we expect direct technology transfers and spillovers to be particularly strong in knowledge intensive sectors. After all, these industries are the ones where spillovers are more likely to materialise, and where multinationals may be expected to transfer intangibles to their subsidiaries. In Central de Balances, only a subset of firms report their R&D spending, and the series is not available before 1986. 14
This is consistent with the findings of Siotis (2003) indicating an increase in competitive pressure during the same sample period. In all specifications we set both p and r equal to 2. We also experimented with alternative lag structures. Additional lags are not significant and all our main results, including the dynamic adjustment paths, are virtually unaltered. 16 We have also experimented with additional controls such concentration, the capital to output ratio and proxies for differences in input costs and quality between domestic firms and foreign subsidiaries. Introducing these variables in the estimation does not affect the essence of our findings. These additional results can be found at: http://www.eco.uc3m.es/siotis/investigacion#main. 15
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
113
Table 4 Extended mark-ups equations (i) PCMt − 1 MSt ITGSt FO50t ⁎ RD FO50t − 1 ⁎ RD FO50t − 2 ⁎ RD FO50t⁎ NRD FO50t − 1 ⁎ NRD FO50t − 2 ⁎ NRD FPt ⁎ RD FPt − 1 ⁎ RD FPt − 2 ⁎ RD FPt ⁎ RD ⁎ FO50t FPt − 1 ⁎ RD ⁎ FO50t − 1 FPt − 2 ⁎ RD ⁎ FO50t − 2 FPt ⁎ RD ⁎ (1 − FO50)t FPt − 1 ⁎ RD ⁎ (1 − FO50)t − 1 FPt − 2 ⁎ RD ⁎ (1 − FO50)t − 2 FPt ⁎ NRD FPt − 1 ⁎ NRD FPt − 2 ⁎ NRD M1 m2 Sargan Test on joint sig. of ID Test on joint sig. of TD RD FO50 LR effect Non-RD FO50 LR effect Test on FO50 restriction RD FP LR effect Non-RD FP LR effect Test on RD FP restriction RD FP (1 − FO50) LR effect RD FP FO50 LR effect Test on RD FP restriction
0.422(0.00) 0.261(0.10) 0.098(0.00) 0.037(0.34) 0.031(0.41) 0.028(0.08) −0.113(0.01) 0.091(0.01) 0.009(0.22) 0.040(0.32) 0.019(0.56) 0.072(0.12)
−0.046(0.09) 0.012(0.66) 0.017(0.22) −14.98(0.00) 1.43(0.15) 187.68(0.02) 32.83(0.38) 168.27(0.00) 0.166(0.05) −0.022(0.54) 0.188(0.04) 0.227(0.12) −0.029(0.54) 0.256(0.10)
(ii) 0.419(0.00) 0.293(0.053) 0.099(0.00) 0.056(0.27) 0.019(0.74) 0.091(0.01) −0.111(0.01) 0.090(0.01) 0.001(0.22)
0.021(0.65) 0.005(0.92) −0.027(0.52) 0.052(0.33) 0.027(0.56) 0.113(0.05) −0.049(0.07) 0.012(0.64) 0.016(0.22) −15.10(0.00) 1.42(0.16) 201.51(0.04) 32.96(0.37) 167.75(0.00) 0.286(0.02) −0.034(0.57) 0.320(0.02) −0.036(0.45) 0.330(0.08) −0.002(0.99) 0.332(0.06)
Note: All estimates include time dummies (TD) and three-digit industry (ID) dummies as regressors and instruments. Additional instruments are: in column (i) PCM(2,2), MS(2,3), HHI(2,3), ITGS(2,3), FO50 ⁎ RD(2,3), FO50 ⁎ NRD(2,3), FP ⁎ RD(2,3), FP ⁎ NRD(2,3); in column (ii) PCM(2,2), MS(2,3), HHI(2,3), ITGS(2,3), FO50 ⁎ RD (2,3), FO50 ⁎ NRD(2,3), FP ⁎ RD ⁎ FO50(2,3), FP ⁎ RD ⁎ (1 − FO50)(2,3),FP ⁎ NRD(2,3). See the note to Table 3 for the interpretation of reported test statistics.
Nevertheless, this data enables us to construct an annual proxy for R&D intensity at the 3-digit level for the period 1986–1996. Yearly sectorial R&D intensity is defined as: RDIjt ¼
∑ni¼1 RDEit ∑ni¼1 Sit
ð6Þ
where RDE is R&D expenditure at the firm level. Thus, for each year, RDIjt takes a single value for each 3-digit sector. We have then computed the average over the time period for which R&D expenditure is available: RDIj ¼
∑ki¼1 RDIjt k
ð7Þ
where k denotes the years for which R&D data is available (1986–1996, i.e.11 years). The splitting criteria that we applied to define R&D intensive sectors is an intensity greater or equal than 2%.17 Four industries fall in this category (pharmaceuticals, electronics, precision instruments, and aerospace), and they account for 2184 observations. Two dummies were constructed accordingly:
RD ¼ 1 if RDIj z0:02 NRD ¼ 1−RD if RDIj <0:02
ð8Þ
Both RD and NRD were interacted with FO50 and FP. 17 The choice of the 2% cut-off point is mainly motivated by a close inspection at the empirical distribution of sector-level R&D intensities. We computed average R&D intensities for all sectors. 26 out of 32 turn out to have average research intensities below 1% and only 4 above 2%, the two remaining sectors respectively scoring 1.6% (naval construction) and 1.1% (agricultural and industrial machinery). What the data seem to suggest us is that 2% is a sort of “natural” cut-off point. Potential alternatives would be to choose 1.5 % or 1% cut-off points thus including “naval construction” and possibly “agricultural and industrial machinery” as additional R&D intensive industries (although casual empiricism suggests that firms operating in these industries are not big R&D spenders). As robustness checks we re-ran all our equations by using 1.5% or 1.0% as alternative cut-off points. Results are qualitatively similar but point coefficients are less precisely estimated, especially with the 1% cut-off.
114
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
Fig. 2. Dynamic effect on mark-ups of foreign ownership—R&D (continuous line) versus non-R&D (dotted line) intensive industries.
Fig. 3. Dynamic effect on mark-ups of foreign presence—R&D (continuous line) versus non-R&D (dotted line) intensive industries.
Results are presented in column (i) of Table 4. As before, there is no evidence of second order auto-correlation of the errors and, the Sargan mis-specification test does not reject the null, tough only at the 1% level. Both ITGS and MS maintain their sign. The latter variable is less precisely estimated, while the former remains highly significant. As for the variables of interest, in non-R&D intensive sectors, a change in ownership has no long-run effect on mark-ups. However, the pattern described in the previous section is maintained: margins initially fall, and this effect is significant at the 1% level. After one lag, margins significantly recover. With the 2-period lag also being positive (though not significant), the long-run effect is negative, but not significantly different from zero as indicated by the Wald test and the dynamic adjustment path depicted in Fig. 2. Regarding foreign presence in nonR&D intensive sectors, the contemporaneous effect is negative and marginally significant. We take this as evidence that an increase in MNC presence increases competitive pressure in the short-run. However, this effect vanishes over time, as indicated by the adjustment path and the Wald test. This is consistent with the conjecture that FDI both increases competitive pressure and generates positive externalities for host country firms (see Fig. 3). The results for the four R&D intensive industries are significantly different from the results we find for the sample which includes all other industries. The tests on the equality of the long-run effects between the two samples of industries reject the restrictions of equal coefficients both for the foreign ownership (FO50) and for the foreign presence (FP) variables. With regard to foreign ownership, we find a statistically significant positive effect that takes time to materialise (the 2-period lag coefficient is the most significant). This is in line with the idea that MNCs do transfer firm-specific assets to their subsidiaries in these industries, and also that there is a learning period before these assets are successfully exploited (see also Fig. 2). While not individually significant at standard confidence levels, the coefficients associated with foreign presence indicate that the long-run effect on mark-ups of MNC presence is always positive and marginally significant. This finding suggests that the positive spillover effect dominates the procompetitive effect in the case of R&D intensive industries.18 Furthermore, the adjustment path pictured in Fig. 3 is upward sloping, thus pointing out that it indeed takes time for spillovers to materialise. 4.2.3. Competition versus spillovers in R&D intensive industries We further explore these issues by exploiting an additional identification condition. If technological spillovers are indeed present in R&D intensive industries, we would expect domestic firms to be the main beneficiaries. This hypothesis is motivated by the fact that Spanish firms are more likely to lag foreign subsidiaries. While the origin country of MNCs is unavailable in Central de Balances, aggregate FDI figures indicate that the main investors come from the US and more advanced Western European economies (e.g., from France, Germany and the UK). It is therefore likely that spillovers will primarily stem from subsidiaries to domestic entities. Clearly, this does not preclude positive technological externalities flowing across foreign subsidiaries; our conjecture is simply that the spillover effect will be felt more acutely by domestic firms. Operationally, we interacted the foreign ownership dummy (FO50) with our measure of foreign presence (FP) in R&D intensive sectors. These additional results are presented in column (ii) of Table 4. As before, the traditional determinants of mark-ups, namely 18 We are faced with a potential identification problem in interpreting these findings. In the four R&D sectors that we identified, it could be the case that the increase in profitability experienced by domestic firms is a by-product of an increase in R&D effort on their part, partially spurred by the increased presence of foreign firms. Data limitations prevent us from directly tackling this issue. However, OECD data (ANBERD database, 2000 release) do not suggest that Spanish firms belonging to the four sectors labelled as R&D intensive significantly increased their R&D outlays during our sample period. The same holds true for firms that reported R&D expenditure in Central de Balances.
A. Sembenelli, G. Siotis / Journal of International Economics 76 (2008) 107–115
115
Fig. 4. Dynamic effect on mark-ups of foreign presence in R&D intensive industries—Domestic Firms (continuous line) versus Foreign Firms (dotted line).
market share and product differentiation, have the expected sign. As for our variables of interest, their sign, significance, and long vs. short-run effects are the same as before in non-R&D intensive sectors.19 In R&D intensive sectors, the direct effect of foreign ownership (FO50 ⁎ RD) continues to be positive and significant in the long-run. Regarding foreign presence, we find that its long-run effect is positive and significant only for domestic firms. Therefore, it appears that domestic firms are the main beneficiaries of an increase in multinational activity (see also Fig. 4 where the adjustment process based on the results in column (ii) is plotted for domestic firms and foreign subsidiaries). These results are consistent with the conjecture that, as compared to their foreign-owned counterparts, domestic firms belonging to R&D intensive sectors have been the main beneficiaries of spillovers. 5. Conclusions In this paper, we have attempted to disentangle some of the effects usually attributed to FDI in the spirit of the conjectures first put forward by Caves (1974). On the one hand, the fact that MNCs possess firm-specific advantages that can be transferred back and forth across locations suggest that subsidiaries ought to enjoy greater levels of efficiency, and therefore mark-ups. Overall, we find support for this conjecture, but this is limited to R&D intensive sectors. For the rest of manufacturing, the long-run effect of a change from domestic to foreign ownership is nil. Furthermore, and in line with the existing literature, we find evidence of transient costs associated with a change in ownership. With regard to the impact of foreign presence on mark-ups, the dichotomy between R&D and non-R&D sectors is also present. For non-R&D sectors, our results indicate that increased multinational presence dampens margins in the short-run. However, this effect vanishes over time, a finding that is consistent with MNCs also generating positive externalities for host country firms. This conclusion is further supported by the results pertaining to the impact of foreign presence in R&D intensive sectors. In the latter case, the positive spillover effect dominates. Finally, we find evidence consistent with the idea that domestic firms belonging to R&D intensive sectors are the main beneficiaries of spillovers. This should come as no surprise, given that Spanish entities are likely to lag their foreign-owned counterparts. References Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies 58, 277–297. Barba Navaretti, G., Venables, A., et al., 2004. Multinational Firms in the World Economy. Princeton University Press. Benfratello, L., Sembenelli, A., 2006. Foreign direct investment and productivity: is the direction of causality so obvious? International Journal of Industrial Organization 24 (2), 735–751. Bloom, N., Schankerman, M., Van Reenen, J., 2005. Identifying technology spillovers and product market rivalry. CEPR Discussion Paper #4912, February. Bresnahan, T., 1989. Empirical studies of industries with market power. In: Schmalensee, R., Willig, R. (Eds.), Handbook of Industrial Organisation, Vol. 2. North Holland, Amsterdam, pp. 1011–1057. Caves, R., 1974. Multinational firms, competition, and productivity in host-country markets. Economica 41 (162), 176–193. Dri eld, N., Love, J., 2003. Foreign direct investment, technology sourcing and reverse spillovers. The Manchester School 71 (6), 659–672. Hall, R., 1986. Market structure and macroeconomics fluctuations. Brookings Papers on Economic Activity 285–338. Harris, R., Robinson, C., 2002. The impact of foreign acquisitions on total factor productivity: plant-level evidence form UK manufacturing, 1987–1992. The Review of Economics and Statistics 84 (3), 562–568. Kokko, A., 1996. Productivity spillovers competition between local firms and foreign affiliates. Journal of International Development 8 (4), 517–530. Markusen, J., 1995. The boundaries of multinational enterprises and the theory of international trade. Journal of Economic Perspectives 9, 169–189. Martin, S., 2002. Advanced Industrial Economics. Blackwell Publishers, Oxford, UK. Neven, D., Siotis, G., 1996. Technology sourcing and FDI in the EC: an empirical evaluation. International Journal of Industrial Organization 14 (5), 543–560. Siotis, G., 1999. FDI strategies and firms' capabilities. Journal of Economics and Management Strategy 8 (2), 251–270. Siotis, G., 2003. Competitive pressure and economic integration: an illustration for Spain, 1983–1996. International Journal of Industrial Organization 21 (10), 1435–1459. Tirole, J., 1988. The Theory of Industrial Organization. MIT Press. Vives, X., 2006. Innovation and competitive pressure. IESE SP-SP Working Paper #634.
19 We also re-ran our estimations by interacting the foreign and domestic dummy variables in non-R&D intensive sectors. No marked difference emerged between domestically owned concerns and foreign subsidiaries.