Determinants of Vertical Integration: Trade Policy, Contracts and Technology Luigi Pascaliy Pompeu Fabra University and Barcelona GSE Current Version: 22 December 2012 First version: 11 January 2009 Abstract This paper studies the e¤ects of international openness and contracting institutions on vertical integration. It …rst derives a number of predictions regarding the interactions among trade barriers, contracting costs, technology intensity, and the extent of vertical integration from a simple model with incomplete contracts. Subsequently, it investigates these predictions using a unique dataset of more than 20,000 …rms operating in 82 countries. Consistent with theory, the e¤ect of the technology intensity of domestic producers on their likelihood to integrate vertically is decreasing in the quality of domestic contracting institutions and in international openness. Contract enforcement costs are particularly high in developing countries, and their e¤ects on the vertical structure of technologically intensive …rms may have signi…cant welfare costs. If improving domestic contracting institutions is not feasible, then an equivalent solution is to increase openness to international trade, which would discipline domestic suppliers and thereby reduce the need for vertical integration. JEL: D23, F13, L23 Keywords: Vertical integration, Hold-up, Incomplete contracts, International Trade An earlier version of this paper was circulated under the title "Contract Incompleteness, Globalization and Vertical Structure: an Empirical Analysis". y I would like to thank James Anderson, Susanto Basu, Kit Baum, Matteo Cacciatore, Fabio Ghironi, Simon Gilchrist, Fabio Schiantarelli, Luis Serven, Maurizio Zanardi and partecipants at ISNIE 2008 and EEA 2009 conferences, and seminars at Bocconi University, Boston College, Boston University, CREI, Pompeu Fabra University and University of Pavia. Contacts: Luigi.Pascali@upf.edu 1 1 Introduction Anecdotal evidence suggests the presence of substantial heterogeneity in the vertical structure of production across countries. Some observers relate this phenomenon to di¤erences in institutional environments and trade openness. For example, Khanna and Palepu (2000) provide evidence that companies in India are larger and more vertically integrated than those in the US and suggest that this di¤erence occurs because trading at arm’s length is more costly in developing countries, in which contract enforcement is weaker. The Economist (1991) notes that the Japanese companies are less vertically integrated than Western companies, although an increase in foreign competition is leading to a "Japanization" of the Western companies1 . The primary aim of this paper is to o¤er an initial attempt to analyze how domestic contract enforcing institutions and openness to international trade jointly a¤ect the vertical structure of …rms, using a novel theoretical model and testing its predictions on a unique cross-country …rmlevel data set. Two well-established theories o¤er predictions regarding how di¤erences in contracting institutions among countries could a¤ect the vertical organization of …rms. Both of these theories relate the vertical structure of …rms to the "hold-up" problem of underinvestment. Consider the common situation in which aggregate pro…ts depend on the investment of each parties and in which these investments are relationship-speci…c in the sense that they are sunk outside of the business relationship. If these investments are not contractible once they have been made, then opportunism may arise. This possibility can lead ex-ante to under-investment and ex-post to ine¢ cient economic performance. Transaction costs economics (TCE) theories assume that vertical integration solves the hold-up problem at a …xed cost and that this integration should thus be prevalent when contracts are more di¢ cult to write. In contrast, property rights theories (PRT) emphasize that vertical integration does not automatically solve the under-investment problem because employees must also be given incentives to invest and because their lack of ownership of tangible assets may weaken their incentives to invest even in a vertically integrated entity2 . 1 Other studies have emphasized the di¤erences between Emilia Romagna, an Italian region, and the rest of Europe (Johnston and Lawrence, 1998) and between South Korea and Taiwan (Levy,1991). Fan e al. (2009) documented di¤erences across Chinese regions. 2 TCE theories were pioneered by the recently Nobel laureate Oliver Williamson (1975, 1985), while PRT theories were originally developed by Grossman and Hart (1986) and Hart and Moore (1990). 2 Also the theoretical literature on the e¤ects of international trade on vertical integration is inconclusive. For example, in McLaren (2000) buyer-supplier pairs are located in the same country and simultaneously choose whether to vertically integrate or outsourcing. The integration of a pair produces a negative externality because it thins the secondary market and reduces the outside options for non-integrated …rms. In this world, trade openness partially increases incentives for outsourcing by thickening the secondary market. Market thickening is also cited by Grossman and Helpman (2002) as a reason that trade openness increases the advantages of outsourcing. In their model, thickened secondary markets imply lower matching costs between producers and suppliers3 . On the other side, Ornelas and Turner (2008) show that by increasing the gains from becoming a multinational corporation with respect to domestic outsourcing, trade openness may actually increase the vertical integration of domestic …rms4 . Also in Conconi, Legros and Newman (2012) the e¤ect of international trade on vertical integration is ambiguous. Managers decide whether to integrate or not their respective plants trading o¤ pecuniary bene…ts from coordinating production decision with the private bene…ts of operating in their preferred way their own plant. In this setting, the price of output is a crucial determinant of the organizational choice: at higher prices managers value the pecuniary bene…ts from the increase in output brought by integration more than the private costs in terms of the lost organizational control. This design is embedded in a standard two-countries speci…c factor model: under general conditions, moving from autarky to free trade increases the price of output in one country and reduces it in the other one, implying opposite e¤ects on the integration of …rms in the two countries. In sum, the e¤ects of both contracting institutions and trade openness on vertical integration are potentially ambiguous and a better understanding of these relationships requires an empirical 3 Antras (2003) and Antras and Helpman (2004) embeds a property right approach in a general equilibrium, factor proportion model of international trade with imperfect competition and product di¤erentiation. The model pins down the boundaries of multinational …rms as well as the international location of production. A reduction of tari¤s increases the propensity to international outsourcing relative to multinational vertical integration. Also in Mendi (2011), a reduction in tari¤s decreases the propensity of …rms to vertically integrate. In this work, forward vertical integration occurs for strategic reasons, namely to create a mechanism that allows a domestic upstream monopolist to discipline non-integrated domestic downstream oligopolists and sustain pro…table collusion. Since the world market is assumed to be competitive, an import tari¤ imposes a cap on the domestic …nal good price. In this setting, a reduction in the import tari¤ decreases the monopoly pro…t and therefore the incentives to vertically integrate to sustain collusion. 4 Ornelas and Turner (2008) adapt a TCE approach to a particular international context and consider a classical hold up situation between a domestic downstream …rm and a foreign upstream …rm. Under vertical integration, investment is higher and, as consequence, the expected trade volumes are larger. In this context, trade liberalization might be more bene…cial to vertically integrated …rms than to …rms that operate at arm’s length. 3 investigation. Before proceeding to the empirical analysis, in the …rst part of the paper, I present a theoretical model in the spirit of TCE theories that examines how these forces contribute to shaping the governance structure of …rms. The model uses the canonical "hold-up" framework but it extends it to introduce competitive foreign markets and to produce a set of testable predictions. A …nal good producer makes certain investments that become fully productive, depending on whether the domestic supplier decides to collaborate. Ex-ante the two parties can either vertically integrate at a …xed cost or sign a contract that will be enforced ex-post with a certain probability. In case it is not enforced, the two parties can renegotiate the contract or the investor can turn to a foreign supplier at the cost of paying an import tari¤. In the data, …rms are distributed according to a continuous spectrum of levels of vertical integration. To capture this feature, my model assumes that the investor has access not to a single investment but to a continuous set of investments, characterized by di¤erent levels of asset speci…city, which is modeled as the part of the investment that becomes unproductive without the collaboration of the respective supplier. The model produces a set of predictions that can be tested in the data. First, the combination of greater asset speci…city and lower contract enforcement implies underinvestment under outsourcing and therefore increases the incentive to vertically integrate. In addition, the model predicts another interaction e¤ect of asset speci…city and trade barriers. The ability of an investor to …nd other partners in other countries limits the ability of his domestic partner to hold him up. Therefore, lower trade barriers discipline the partners of investors and attenuate the distortions that are generated by the low quality of domestic institutions. The model di¤ers from the previous literature on several dimensions. First, the speci…city of investments, the level of vertical integration, the quality of contracting institutions and the openness to international trade are classi…ed according to a continuous measure to facilitate comparative statics on this variables. Second, it predicts that an increase in trade openness can lead to changes in the ownership structure within countries, even if …rms do not relocate across countries and even if international trade does not change, by simply disciplining domestic partners. I test the predictions of the model using detailed data on more than 20,000 …rms operating in 82 countries. This data set is obtained by aggregating a large number of country-speci…c ICA World Bank Surveys and provides, among the other, information on the vertical structure and technology 4 intensity of …rms. The main dataset is then merged with the Doing Business Database, which provides country data on institutional features, and the TRAIN Database, which provides data on tari¤s. The regressions show that vertical integration is less likely when capital intensity is associated with trade openness and high-quality contracting institutions; thus the predictions of the theoretical model are con…rmed. This result suggests that opening up to international trade may produce the same attenuation to the hold-up distortions as improving domestic contract enforcement. In the most credible speci…cation, the reduction of input tari¤s by 1% produces the same e¤ects on the vertical structure of a …rm, characterized by the median machinery intensity, as a reduction by 3.06% in the cost necessary to solve a commercial dispute in local courts or a reduction by 1.4 in the number of days necessary to close the dispute. These results are practically una¤ected when controlling for …rm size, foreign ownership and market power. To cope with potential endogeneity issues in the capital intensity of …rms, I use an IV approach that exploits exogenous variations in the level of asset speci…city that is generated by technological di¤erences across sectors. Speci…cally, I use six di¤erent measures of asset complexity within the same industry in the United States as instruments for capital intensity. In addition, I conduct a number of robustness checks and …nd that the results are robust to a wide variety of speci…cations. To the best of my knowledge, this is the …rst study that tests the e¤ect of trade openness on vertical integration in a cross-country empirical analysis5 . Moreover, this paper contributes to a recent stream of empirical research on the e¤ects of the institutional environment on the vertical boundaries of …rms. In particular, Macchiavello (2012) uses the UNIDO industry-level database to study the e¤ects of …nancial development on the average vertical integration in the industry, while Acemoglu, Johnson, Mitton (2009) use a business contact dataset, to study how …nancial development and contract enforcing costs interact to shape the vertical structure of …rms. A potential bene…t of using the ICA World Bank surveys with respect to the datasets used by these two studies is that they provide …rm-level information on the asset speci…city of …rms and, thus, allow to capture heterogeneous e¤ects of the institutional environment on the vertical boundaries 5 In a parallel contribution, Chungvilaivan and Hur (2012) examine the relationship between trade openness and the pattern of vertical integration using US manufacturing data. In particular, they document a negative relationship between import penetration and the propensity of …rms to vertically integrate, particularly pronounced in more concentrated industries. 5 of …rms operating in the same industry. Two other interesting works that test the institutional determinants of vertical integration, but in a single-country analysis are those of Fan et al (2009), which exploits institutional variation across di¤erent Chinese provinces, and Acemoglu et al. (2010), which investigates the validity of PRT theories using data on the census of manufacturing plants in Great Britain. In sum, my work provides an empirical analysis of the institutional determinants of vertical integration using a cross-country database. This study is the …rst one to use a cross-country database to evaluate the role of tari¤s and it adds further evidence on the role of contract enforcement institutions. Policy advice emerges from both the theoretical and empirical analyses of the paper. When associated with speci…c assets, poor contract enforcement can distort the vertical structure of …rms and thus have signi…cant welfare costs. If improving home institutions is not feasible, then an equivalent solution is to reduce the trade barriers to the import of intermediates. This reduction would discipline domestic suppliers and increase the incentives of producers to invest in speci…c assets. An alternative way to interpret this result is that reducing trade barriers is a way of "importing" foreign institutions, as domestic …rms will relate with one another as though the relevant contracting institutions were those of the countries in which alternative suppliers are operating. This paper is organized as follows. Section 2 details the theoretical framework and derives some testable implications. Section 3 presents the main empirical results and several robustness checks. Some concluding remarks close the paper. 2 The Model 2.1 Basic Structure A …nal good producer (P) at Home (H) seeks to buy an input that enhances the productivity of his investments. There is a speci…c supplier (HS), whose characteristics are most suitable for providing the input to …rm P and that is also located at H. P could either outsource to HS or become vertically integrated with her. Under outsourcing, the two parties write a contract on the price of the intermediate good before the speci…c investment is realized. However, because of contract incompleteness, there is a probability that this agreement will be broken after the speci…c investment has been realized. At 6 this point, a new agreement must be reached. However, the bargaining power of HS is much higher than before because the producer’s speci…c investment is partially sunk without the intermediate good. The amount with which HS can hold-up P depends on the possible alternatives that the latter has to buy a similar intermediate elsewhere. I assume that P can purchase the same intermediate from a foreign supplier (FS) located in a competitive market. Ex-ante (e.g., before the speci…c investment has been undertaken), the price of the foreign intermediate, pF , is a random variable with randomness that re‡ects both shocks to the productivity of FS and shocks to the exchange rates. Moreover, when importing an intermediate, the producer must pay a trade cost t. Ex-ante, the probability of …nding an alternative intermediate depends on the trade costs. If this probability is low, for example because trade barriers are excessively high and buying in another country is not feasible, then the producer knows that most of the revenues that are derived from his investment can be expropriated by HS. Therefore, the producer would have lower incentives to invest, which would imply a suboptimal level of investments and ex-post ine¢ cient economic performance. Under vertical integration, the two parties merge into a single …rm. As in Hart and Tirole (1990), this option "permits pro…t-sharing between upstream and downstream units so all con‡icts of interest about prices and trading policies are removed". The advantage of this option is that an e¢ cient level of speci…c investments is realized; the disadvantage is that it requires a …xed cost. The presence of a …xed cost related to the vertical integration choice is a common feature in this body of literature (see Hart and Tirole, 1990; McLaren, 2000; Ornelas and Turner, 2008) and can be interpreted as a means of capturing all of the legal, …nancial and organizational costs that are involved when merging two …rms. In sum, the choice between vertical integration and outsourcing solves the trade-o¤ between the …xed cost that arises under vertical integration and an ine¢ cient level of speci…c investments that arises under outsourcing. Better contract enforcement and lower trade barriers attenuate the relevance of the hold-up problem and the related investment distortions and thus increase incentives for outsourcing. The timing of the events in the model is summarized in Figure 1. At period 1, HS and P decide whether to integrate; if they do not choose to integrate, then they sign an ex-ante contract that de…nes the price of the intermediate input6 . At period 2, P makes its relationship-speci…c 6 Notice that I have ruled out the possibility that the producer could outsource to FS in the …rst stage of the game. 7 Home supplier and producer decide whether to vertically integrate and reach an agreement over the price of the intermediate input If vertical integration was chosen Stage 1 If outsourcing was chosen The producer chooses I and the home supplier provides the intermediate input The producer chooses I Contract not enforced Stage 2 Contract enforced Home supplier and producer bargain over the price of the intermediate input Home supplier provides the specific input Stage 3 If bargaining breaks down The producer chooses whether to buy the intermediate input from a foreign supplier Stage 4 Figure 1: The sequence of events investments, I. At the beginning of period 3, the state of nature is revealed: the price of the foreign intermediate (pF ) becomes public, and parties are informed as to whether the initial contract will be enforced. If outsourcing was chosen, then with probability (1 ) the initial contract is not enforced and the two parties must bargain again over a new price7 . At period 4, P can decide whether to buy the intermediate input from the foreign supplier FS. The production technology of the producer has the following form: f ( ; I; x) = (1 )g(I) + g(I)x (HP1) where I is the producer’s investment, and x is an indicator variable that is equal to one if the home supplier provides the intermediate good that increases productivity and zero otherwise. (0; 1) corresponds to the share of the investment that is unproductive without the intermediate good and captures the speci…city of the investment. The …rst term of the production function is the output that the producer can eventually generate without any intermediate good. The second term is the additional output that is generated by the producer conditional on the supplier providing the intermediate good. Assume that: There is no loss of generality in doing this since this alternative would be strictly dominated by the alternative to outsource to HS (because outsourcing to FS has a …xed trade cost). 7 A similar way of modelling contract incompleteness as a determinant of vertical integration has been used by Acemoglu, Johnson and Mitton (2009). 8 g 00 (I) < 0 (HP2) Normalize the cost of one unit of speci…c investment to 1, and assume that HS can provide the intermediate at no cost. The game is solved by backward induction. In stage 4, if the producer still does not have the intermediate input, then he will buy it from the foreign supplier if: pF < g(I) where t is a proxy for the appropriateness of the foreign intermediate input to the speci…c in- vestment made by the producer. Let us consider a situation in which (e.g. the domestic intermediate is at least as e¤ective as the foreign one). 2.2 Stage 3: Expected Pro…ts under Outsourcing In the third stage of the game, the producer has already made the investments and is outsourcing the production of the intermediate good to the home supplier. Suppose that the initial contract cannot be enforced and that the two parties must bargain over the price of the input. In the event of disagreement, the two parties receive their outside option. The home supplier would make zero pro…ts, whereas the producer could still …nd it pro…table to use the intermediate input that is produced by the foreign supplier. Denote the outside option of party i under outsourcing by OiO . Then: OpO (I) = (1 ) g(I) + [M axf0; g(I) OsO = 0 t pF g] I (1) (2) Denote the ex post payo¤s of party i by u0i . Then: f u0p (I) = g(I) I p u0s = p where p denotes the new price of the intermediate when bargaining is successful. According to 9 the Nash bargaining solution, the price p satis…es the following: p = Arg max p0:5 [ g(I) M axf0; g(I) p t pF g p]0:5 (3) which implies: p= 1 g(I) 2 1 M axf0; g(I) 2 t pF g (4) Hence, under the symmetric Nash equilibrium, the surplus that the producer accrues under outsourcing conditional on the lack of enforcement of the initial contract is as follows: u0p (I) = 2.2.1 1 1 g(I) + M axf0; g(I) 2 2 t pF g I (5) Stage 2: Choosing the Optimal Investments In the second stage of the game, the producer chooses the optimal investments. If the producer and the home supplier are vertically integrated, then the producer will select the level of investments I to maximize the joint variable pro…ts: VI = g(I) I (6) The optimal level of investments under vertical integration I V I is de…ned by the …rst-order condition g 0 (I V I ) = 1. If the producer is outsourcing to the home supplier, then the producer will choose the level of investments I O to maximize the expected pro…ts: E O p = [g(I) P ]+(1 ) 1 2 g(I)+(1 1 )M axf0; 2 g(I) Z t ( g(I) t pF )dF (pF )g I (7) 0 where P is the price of the intermediate good as in the initial contract. Intuitively, the …rst term represents the revenues that the producer would obtain if the initial contract is enforced; the second term represents the revenues that the producer would obtain under autarky if the initial contract is not enforced; the third term represents the additional revenues that the producer would receive in 10 the presence of international trade if the initial contract is not enforced (because of the improvement in his outside option in the bargaining game with the domestic supplier); the last term represents the investment costs. The optimal level of investments under outsourcing, I o ( ; ; t), is the level of O. p investment that maximizes E In general this function has more than one local maximum. The following hypothesis limits the number of local maxima to two. g 00 (I) < g 0 (I)2 1 ) 12 2 f ( g (I) t) ) 12 ( F ( g (I) t)) (1 (1 (HP3) Intuitively g (I) must be su¢ ciently convex (e.g., the marginal productivity of investments must decrease more rapidly than the hazard rate of the price of the foreign intermediate). To discuss the local maxima of the function E (1 ) (I; T ) where Q( ; ; I) M axf0; 12 g(I) R t ( g (I) 0 t is convenient to rewrite it as: E O P = Q( ; ; I) + h i P C] + (1 ) 1 2 g (I) I and (I; t) O P [g (I) pF ) dF pF g. De…ne I (t) the minimum investment for which: (I (t) ; t) 0 and I ( ; ) the investment that maximizes Q( ; ; I). In other words, I (t) is the minimum investment necessary to make credible the threat of buying the intermediate from a foreign supplier and I ( ; ) the optimal investment under autarky. Lemma 1 If I ( ; ) > I (t), the pro…t function has a single local maximum in I ( ; ; t) iden- ti…ed by the following equation: 1 If I ( ; ) (1 ) 1 ( 2 F ( g (I ) t)) g 0 (I ( ; ; t)) 1 0 (8) I (t), the pro…t function can have an additional single local maximum in I ( ; ) identi…ed by the following equation: 1 and such that: I ( ; ) < I (1 ) 2 g 0 (I ( ; )) 1 0 (9) ( ; ; t) Thus, the pro…t function has at most two local maxima I ( ; ) and I convexity of g(:) it is easy to verify that both I ( ; ) and I ( ; ; t). Given the ( ; ; t) are lower than I V I (e.g. investments are always lower under outsourcing rather than under vertical integration). The en11 h tity of underinvestment under outsourcing is proportional to 1 prohibitive (e.g. if I o ( ; ; t) = I ( ; )) and to (1 not (e.g. I o ( ; ; t) = I ) 21 (1 (1 ) 2 i if trade barriers are ) 12 F ( g (I) t) if they are ( ; ; t)). The last expression is intuitive. The …rst term represents the classical hold-up distortion that we …nd in the transaction cost literature. The interaction between contract incompleteness and asset speci…city distorts the investment incentives to invest of the producer because a part of the surplus that is generated by the investments can be appropriated by the supplier, and this situation produces suboptimal investments. The second term represents the e¤ect of opening the intermediate market to international trade and is the main novelty of the model. The ability of the producer to buy the same intermediate input with a certain probability from a foreign supplier limits the possibility of holding him up and de facto attenuates the distortions that are created by low-quality domestic institutions. In the limit, if trade barriers and foreign prices are su¢ ciently low (e.g., 2 F( g(I O ) t) = 1), then the hold-up problem disappears. In this sense, opening a country with bad contracting institutions to trade is a means of "importing" good institutions. This logic leads to our …rst proposition (see the Appendix for the complete proof). Proposition 2 Under outsourcing, the producer’s optimal investment is not increasing in t. Notice that by applying the implicit function theorem on equations 8 and 9, one can prove that both I ( ; ) and I ( ; ; t) are increasing in contract enforcement, , and decreasing in the speci…city of the asset, . This result provides the intuition for the following propositions (see the Appendix for the complete proof): Proposition 3 Under outsourcing, the producer’s optimal investment is increasing in . Proposition 4 Under outsourcing the producer’s optimal investment is decreasing in : 2.2.2 Stage 1: Choosing the Governance System Because both parties have access to ex-ante transfers, the subgame perfect equilibrium will always select the organizational form that maximizes their joint surplus. Consistent with the transaction cost approach, it is hypothesized that vertical integration has a …xed cost . Let S V I ( ) IV I refer to the joint surplus under vertical integration, and let S O ( ; ; t) 12 g(I V I ) g(I O ( ; ; t)) I O ( ; ; t) refer to the joint surplus under outsourcing. The comparison of these values yields the following proposition. Proposition 5 Vertical integration is more likely when assets are speci…c ( is high), contracts are incomplete ( is low) and trade barriers are high (t is high). = SV I ( ) Proof. The two parties will vertically integrate as long as S O ( ; ; t) is positive. To obtain an expression for the impact of higher asset speci…city on the governance of the …rms consider the derivative of the latter with respect to . d d Notice that since 1 dI O ( ; ;t) d = dI O ( ; ; t) [1 d g 0 (I O ( ; ; t))] is not positive by proposition 4 and [1 (10) g 0 (I O ( ; ; t))] is also not positive g 0 (I V I ) = 0 and I O ( ; ; t)) < I V I (together with the convexity of g). Thus Analogously it can be proven that d dt 0 and d d d d 0. 0. The intuition behind the last proposition is straightforward. Higher levels of asset speci…city, contract incompleteness and trade barriers tend to distort investments under outsourcing and thus increase the e¢ ciency of vertical integration. One problem when confronting this theoretical model with the data is the di¢ culty of classifying …rms according on whether they are vertically integrated or not because, in the data, …rms are distributed according to a continuous spectrum of levels of vertical integration. The model can capture this feature if the producer has access not to a single investment but to a continuum of investment opportunities I(k) with k [0; 1] ; each of which is characterized by a di¤erent level of speci…city k. Without a loss of generality, assume that these investment opportunities are ordered according to an increasing level of speci…city, such that simplifying assumption that speci…cally, assume that k k k0 > k if and only if k 0 > k. Make the is a continuous di¤erentiable function of k such that: d k dk > 0. More = (Ak); where A > 0 is a shifter and represents the sophistication of the producer’s assets (i.e., higher values are associated with a higher average level of speci…city of the investments). Finally, let k denotes the investment opportunity for which the producer is indi¤erent as to whether to outsource the production of the input that renders the investment fully productive. Based on the previous discussion, k is de…ned from the following identity: 13 SV I ( ) S O ( ; (Ak ); t) 0 (11) The producer will undertake all of the investments that are available and outsource the production of the intermediate inputs that are necessary to render them fully productive only for the investments such that k < k . Rather, for those investments such that 1 > k > k , the …rm will choose to produce them internally. For these reasons, V I 1 k can be considered a valid continuous measure of the level of vertical integration of the …rm. Proposition 6 Firms are more vertically integrated when they use more sophisticated assets, contracts are more di¢ cult to enforce and trade barriers are high. Proof. Using the implicit function theorem on equation 11, together with propositions 2, 3 and 4, we dV I dk k dV I dI O dI O 0 dV I dI O dI O 0 get: = = > 0; = =[ (Ak )A] < 0 and = =[ (Ak )A] > dA dA A d d d dt dt d 1 dV I dV I 2 1 dV I dV I 2 = < 0 and = > 0: 0; dAd A d dAdt A dt Proposition 7 The e¤ ects of the average level of asset sophistication of a …rm (A) on its vertical structure are magni…ed when contracts are di¢ cult to enforce and trade barriers are high. Proof. Using the implicit function theorem on equation 11, together with propositions 2, 3 and 4, we 1 dV I 1 dI O dI O 0 dV I 2 1 dV I 1 dI O dI O 0 dV I 2 = = =[ (Ak )A] < 0 and = = =[ (Ak )A] > get: dAd A d A d d dAdt A dt A dt d 0: The traditional IO literature has emphasized the distorting e¤ects of asset speci…city on the governance system of …rms in an institutional environment characterized by incomplete contracts. Thus, it should not come as a surprise that di¢ cult contract enforcement ampli…es the distortive e¤ects of asset speci…city. The contribution that the last proposition o¤ers to this literature is that it proposes an escape clause. In fact, the distortive e¤ects of asset speci…city on the vertical structure of a …rm in the context of bad domestic contracting institutions can be dampened if the …rm has access to foreign markets for intermediates. Note that there is no international trade in this model: in equilibrium the producer will always buy the widget from the domestic supplier. However, the threat of being replaced by a foreign supplier assists in disciplining the domestic supplier. As the latter can no 14 longer hold-up the producer, outsourcing does not produce distorted investments, and the two parties are less likely to become vertically integrated. In the next sections, the last two propositions will be tested empirically. 3 Empirical Analysis 3.1 Data and Measurement My …rm-level data are obtained from the "Investment Climate Assessments" (hereafter ICA), a set of …rm-level surveys administered by the World Bank in a large number of developed and developing countries. Each survey contains questions regarding the characteristics of …rms (e.g., industry, ownership and business age), measures of economic performances (e.g., balance sheets and information pertaining the number and quali…cation of workers) and measures of the business climate (e.g., trade costs, lobbying activity, bureaucratic delays, infrastructure, product and labor market regulations). I have used a limited subset of the information that is provided in these large surveys. In particular, I have used data pertaining to output value and intermediate costs to measure the degree of vertical integration of the …rms in the sample, information regarding the net book value of machinery and equipment to measure their capital intensity and information regarding the number of workers, the ownership structure, the share of the domestic market and the industry in which they operate. I have rearranged these variables in each country survey to render them comparable across countries8 . Ultimately, the …nal dataset covers 20,216 …rms in 82 countries. New additional country surveys are implemented each year; thus the data cover di¤erent periods for di¤erent countries, ranging from 2003 to 20099 . Vertical integration is proxied by the value added over sales ratio, which should move consistently with the number of processes performed by the …rm10 . This ratio takes out the monetary dimension and is comparable across countries without running into exchange-rate and purchasing 8 Details are available in the Appendix. In the ICA data, the great majority of …rms are surveyed only once. For a very limited number of countries, however, some …rms are surveyed up to three years. In this study, I have chosen to disregard the panel dimension and conduct a purely cross-sectional study. Whenever more observations of the same …rm are available, only the most recent one is considered. 10 A large empirical literature has already used this measure for vertical integration (see for instance, Adelman (1955), Gort (1962), Nelson (1963), La¤er (1969), Tucker and Wilder (1977), Levy (1985), Barney, Edwards and Ringleb (1992), Shin (2001), Bender (2002), Macchiavello (2008)). 9 15 power parity problems in terms of comparability. It will be close to one in the case of high vertical integration and close to zero for …rms that add only marginal value to purchased inputs. Maddigan (1981) pointed out that this measure has two mayor drawbacks. The …rst is that it is in‡uenced by forces other than vertical integration, especially pro…tability. I will control for this in the empirical analysis. However, it is reassured to notice that Tucker and Wilder (1977) compared a value-added ratio corrected for pro…ts in the numerator and in the denominator with one not corrected: the use of one series instead of the other did not a¤ect any result. The second drawback of value added ratio is that is susceptible to a bias when it is used along an industry value chain because the ratio is decreases along the value chain from upstream to downstream. However, as Levy (1985) and Bender (2002) pointed out, this bias is signi…cantly reduced when restricting the sample to manufacturing …rms. Asset speci…city is proxied by the machinery intensity of the …rm, computed as the ratio of the value of the machinery and equipment used in the production to total sales. Alternative measures, used in the 2SLS analysis, are imputed from the Compustat database using the average intensity of …xed assets and intangible assets in the corresponding industry in the United States11 . Country-level data pertaining to the institutional environment are provided by the Doing Business database (hereafter designated DB). This database represents my primary source of data regarding the quality of contract-enforcing institutions and …nancial development. More speci…cally, DB provides me with three di¤erent variables to proxy for the e¢ ciency of the judicial system in resolving a commercial dispute: the time that is necessary to enforce a contract when it is being disputed in courts, the costs that are associated with this enforcement and the average number of documents that are needed. All of these variables refer to a claim that is assumed to be equivalent to 200 percent of the average income per capita and have been inferred through studies of the codes of civil procedure and other court regulations as well as surveys completed by local litigation lawyers and judges. Moreover, the DB database reports two measures on the level of development of national …nancial institutions that are used in my analysis: an index on the strength of legal rights in the …nancial sectors and an index on the depth of credit information. The …rst one measures the de11 Machinery intensity and capital intensity have been largely used as proxies for asset speci…city in several empirical studies on the determinants of the …rm boundaries (see for instance Nunn (2007), Acemoglu, Johnson and Mitton (2009), Almeida, Campello and Hackbarth (2011), Chongvilaivan and Hur (2012)). 16 gree to which collateral and bankruptcy laws protect the rights of borrowers and lenders and thus facilitate lending; the second one evaluates the rules and practices a¤ecting the coverage, scope and accessibility of credit information available through either a public registry or a private credit bureau. Trade barriers to the imports of intermediate goods are measured using data on tari¤s from the UNCTAD TRAINS database in addition to the input/output accounts that are published by the Bureau of Economic Analysis. More speci…cally, the TRAINS database provides data on the average (e¤ectively applied) tari¤s in each country at the 4-digits SIC code. These SIC codes are subsequently matched with the appropriate 6-digits IO codes using the BEA’s concordance guide. The input tari¤ in a certain industry is then constructed as a weighted average of the output tari¤s in those industries from which it receives its inputs using cost shares as weights. Table 1 sums up how theory is related with data and presents the proxies for vertical integration and its main determinants. Table 1: Measures used in the empirical analysis Variable Vertical integration Proxy Value added / Total sales (ICA) Asset speci…city Machinery,equipment/ Total sales(ICA) Contract incompleteness Contract enforcement costs (DB) Contract enforcement procedures (DB) Contract enforcement time (DB) Trade barriers Average tari¤s (TRAINS) Average tari¤s on inputs (TRAINS+BEA) Table 2 summarizes the distribution of observations across countries. Most of the countries that are considered are low-income countries, but there are a number of exceptions (for instance, the Czech Republic, Slovakia and Croatia). Income per capita ranges from $249 in Congo to $25553 in the Czech Republic. The sample size varies considerably across countries: for example, there are 22 observations in Montenegro compared with 1172 observations in Thailand. Limited information is provided regarding how the sample is selected in each country, but in general the typical ICA respondent is a large, mature business relative to the representative business in the 17 country (Haltiwanger and Schweiger, 2004). Table 3 provides some descriptive statistics for the variables that are used in the empirical analysis. The …rst …ve rows consider …rm-level variables that are constructed from the ICA database. Row 1 reports descriptive statistics for the vertical integration index at the …rm level. Observations are fairly distributed around 0.6 (the mean and median have same value) from values that are close to zero to values that are close to one. Row 2 reports the proxy for asset speci…city, namely, the ratio of the net book value of machinery and equipment over the total sales of the …rms. The average ratio is 0.22, which is less than half of the analogous US average in the manufacturing sector, which is 0.45. This con…rms that using US data to impute the capital intensity of …rms in other countries, as in Acemoglu, Johnson and Mitton (2009), may lead to misleading estimates. Row 3 reports the average number of permanent workers at the …rm level. Both the mean and median (127 and 30, respectively) show that the typical ICA respondent is large relative to a representative business for a country. This is con…rmed in raw 4, which reports …rm’s share of national market of the main product line (on average above 18%). Row 5 reports the share of the …rm that is foreign owned. More than 14% of the …rms in the sample have some foreign participation (above 1%) in the ownership structure and half of them are 100% owned by foreign privates. The following three rows report the three di¤erent measures of the quality of contract-enforcing institutions coming from the Doing Business database. The average commercial sale dispute in local courts in my sample has an average cost of more than one-third of the claim and requires a total of 37 legal procedures and 600 days to resolve. These numbers are high compared with their US counterparts (14.4%, 32 and 300, respectively). Moreover, although the number of procedures does not di¤er greatly across countries, there is substantial heterogeneity in the other two measures. Rows 9 and 10 report the two measures of the quality of the …nancial system. Row 11 is the GDP per capita in 2009. The GDP estimates are PPP adjusted and are obtained from the Penn World Table database. The following three rows report the average tari¤s on respectively intermediate inputs imported from all countries, intermediate inputs imported from high-income OECD countries and …nal goods imported from all countries. Finally, rows 15 through 20 report six di¤erent measures of capital intensity at the industry-level computed from the US Compustat data set. 18 3.2 Empirical Results Proposition 5 predicts that opening up to international trade and improving contract enforcement institutions should reduce the likelihood of …rms to vertically integrate. Unfortunately, testing these predictions is a di¢ cult task since a long list of potential omitted country-level variables correlated with contract enforcement and trade openness may a¤ect the governance of the …rm. However, the model provides a possible methodology to handle these unobservables that consists in focusing on the interaction e¤ects of contracts and trade barriers with the machinery intensity of a …rm and adding a set of country dummies in the regression. In this respect, proposition 7 predicts that the e¤ects of asset speci…city on the vertical structure of …rms are magni…ed by bad contracts and high trade barriers. The following regression tests this proposition: V If sc = 1 ASf + 2 CIc ASf + 3 T Bcs ASf + X 0 + Sc + Ss + "f sc (12) where V If sc and ASf sc are the vertical integration and machinery intensity of …rm f in country c in sector s, CIc is the cost to enforce a contract in country c, T Bcs is the tari¤ on intermediates in country c in sector s and X 0 is a vector of control variables. Finally, Sc and Ss are the country and industry …xed e¤ects, respectively. Column 1 of Table 4 reports the OLS estimates of the following equation: note that all of the signs are consistent with the theoretical predictions. For an interpretation of the magnitude of the coe¢ cients, Figure 2 depicts the marginal e¤ect of machinery intensity on the vertical structure of a …rm as a function of trade barriers and contract enforcement costs. The …rst graph shows how the marginal e¤ect of asset speci…city on vertical integration changes as importing costs change (with contract enforcement costs …xed at their mean level). For sectors in which there are no tari¤s on the import of intermediates, I cannot reject the hypothesis that asset speci…city has no signi…cant e¤ect on vertical integration. However, as trade barriers become higher, the e¤ect of asset speci…city becomes positive and signi…cant. The second graph depicts the marginal e¤ect of asset speci…city as the cost of enforcing contracts varies. As expected, the e¤ect is much larger as these costs increase. This result suggests that improving the judicial system and reducing import barriers are two substitute policies to attenuate hold-up distortions. More speci…cally, a reduction in tari¤s by one 19 percentage is associated with the same e¤ect on the vertical structure of a …rm as a reduction in the cost of judicial disputes by 12.7 percentage points. However, two potential concerns apply to the OLS estimates that we have seen so far. First, several omitted …rm-level variables could be driving both the asset speci…city and vertical integration of the …rms in my sample. Second, the estimates may su¤er from a potential reverse causality problem. For example, …rms that are more vertically integrated may be more likely to perform primary activities that are less technologically intensive and require less speci…c assets. In both cases, the error term would be correlated with the regressors, biasing the estimates. A more satisfactory approach would involve conducting an instrumental variable analysis. The following results exploit an exogenous source of variation in the level of asset speci…city that results from technological di¤erences across sectors. More speci…cally, I instrument the machinery intensity of a …rm in a certain industry using six di¤erent measures of average asset sophistication within the same industry in the United States: the ratio of …xed assets and of the sum of …xed and intangible assets to sales, total assets and number of employees. These variables are computed from the US Compustat dataset12 . The following exclusion restrictions are implied by the IV strategy: (1) conditional on the controls that are included in the regression, capital intensity in the US has no e¤ects on the vertical structure of non-US …rms, if not through its correlation with their machinery intensity, and (2) no unobservables are systematically correlated with capital intensity in a certain US industry and the vertical structure of non-US …rms operating in the same industry. A possible concern is that the latter condition may be violated if there are common industry-speci…c unobservables across-countries. To address this issue, I will use industry …xed e¤ects throughout the remainder of the analysis. This approach implies that it is impossible to identify the marginal e¤ect of asset speci…city on the vertical structure of …rms; rather, it is only possible to determine how this e¤ect varies with the contractual institutional environment and the trade openness of each country. The …rst-stage equations for the model in equation 12 are as follows: ASf CIc = Z 0 CIc 11 + Z 0 T Bcs 12 + X0 13 + Sc + Ss + u1f (13) 12 Acemoglu, Aghion, Gri¢ th and Zilibotti (2010) were the …rst to propose this instrument for asset speci…city in a …rm-level analysis limited to UK manufacturing plants. 20 ASf T Bcs = Z 0 CIc 21 + Z 0 T Bcs 22 + X0 23 + Sc + Ss + u2f (14) where Z is the vector of instruments for asset speci…city (in other words, capital intensity in the same industry in the US). Column 2 in Table 4 reports the IV estimates of equation 12. The instrumental variable strategy con…rms that the e¤ect of machinery intensity on the vertical structure of a …rm is increasing in the cost of enforcing contracts and in the tari¤s on intermediates. Note, however, that the magnitude of the coe¢ cients on the two interaction terms increases substantially with respect to the previous OLS estimates. In particular, an increase of one standard deviation in the cost of enforcing contracts (holding asset speci…city constant at its mean) leads to an increase in vertical integration of more than a one-fourth of a standard deviation. Moreover, decreasing the tari¤s on intermediates by 1% produces the same attenuation of the hold-up distortions on the organization of production as a 3.06% reduction in the cost of enforcing contracts. For instance, suppose that Burundi (in which CI is 39.31%, 75th percentile) could reduce the cost of enforcing a contract to the same level as in Estonia (in which CI is 19.96%, 25th percentile). This reduction would produce the same e¤ect on the vertical structure of the …rms operating in Burundi as reducing the average level of tari¤s on intermediate inputs from 8.14% to 1.82%. Several authors have argued that, within industries, vertical integration is highly correlated with the size of …rms (see for instance, Whinston (2003), Banerjee and Munshi (2004) and Macchiavello (2012)). Moreover, …rm size is a¤ected by the level of competition in local markets and thus by local institutions and trade policies (see for instance Nocke and Yeaple (2008), Constinot (2004)). Therefore, the previous regressions may not capture the e¤ects of contract enforcement costs and intermediate tari¤s on the hold-up distortions in the vertical structure of …rms but rather their e¤ects on …rm size. To ensure that this is not the case, I add the number of workers among the regressors as a proxy for …rm size13 . It is reassuring to observe that the coe¢ cients on the interaction terms are practically unchanged. As already discussed, the ratio between value added and shipments is sensible not only to the …rm’s vertical integration but also to the …rm’s market power. This would bias the results if for example, trade barriers provide protection for monopolists or weak contract enforcement is 13 I have also experimented with regressions controlling for second, third and fourth order polinomial in …rm size and found very similar results. 21 likely associated with weak antitrust policies. The use of country and industry …xed e¤ects should alleviate the problem. As a further robustness check, in column 4, I have added a variable that could eventually capture …rm’s market power: the market share for main product line. Again, the main results are not a¤ected. The quality of the judicial system and the openness to international trade are highly positively correlated with the stage of development of a country. To exclude the possibility that CIc ASf and T Bc ASf are merely serving as proxies for other factors associated with economic development, in column 5 I add the interaction between per-capita GDP and asset speci…city among the regressors. This addition does not alter the estimates on the two coe¢ cients of interest. It is interesting that the coe¢ cient on this new regressor is negative and statistically signi…cant. Thus, when accounting for tari¤s and contract enforcement, more machinery-intensive …rms tend to be relatively more integrated in richer countries. Finally, a potential concern regarding these regressions is prompted by multinational …rms because they may choose their organizational structure independently from the business climate of the country in which some of their activities are located. In the last two columns, I add a dummy that identi…es …rms that are completely or partially owned by foreign parties. As expected, these …rms are less vertically integrated; however, controlling for ownership does not a¤ect the estimates for the coe¢ cients on the interaction terms. In table 5, I replicate the estimates in the previous table using two alternative proxies for the quality of contract-enforcing institutions: the number of procedures and the time that is necessary to solve a commercial dispute in courts. From a qualitative perspective, the results are similar to the previous ones. Reducing input tari¤s by one percentage point produces the same e¤ects on the vertical structure of …rms as reducing the number of procedures by 1.4 or alternatively the number of days necessary to solve the dispute by 7.5. These results are practically una¤ected when controlling for the size of the …rms, their domestic market share and their ownership. A large body of literature has emphasized the role of …nancial intermediaries in shaping the degree of vertical integration of …rms. The predictions from theoretical models are mixed. On one side, well-functioning …nancial intermediaries should increase the number of potential entrepreneurs and, by fostering entry, reduce the average size of …rms. Because smaller …rms are less likely to produce inputs on their own, …nancial development should then reduce vertical integration. 22 Moreover, this e¤ect should be disproportionately larger in sectors that are more capital intensive which usually rely more on external …nance. On the other side, however, the lack of …nancial development may prevent …rms that would like to vertically integrate from doing so (see Acemoglu, Johnson and Mitton (2009) or McMillan and Woodru¤ (1999)). To test these two competing theories and, most importantly, to exclude the possibility that the previous empirical results are not merely capturing the e¤ect of …nancial development on the vertical structure of …rms, I add the interaction between machinery intensity and …nancial development to the regressors in table 6. I use two measures of …nancial development: an index of the strength of legal rights in the …nancial sectors and an index of the depth of credit information. Both the OLS and IV estimates con…rm the previous results regarding the e¤ects of contract enforcement and tari¤s on the vertical structure of …rms. Moreover, the IV estimates appear to validate the view that imperfections in the …nancial markets prevent …rms that would prefer to vertically integrate to do so: in the data, more developed …nancial markets increases the propensity of …rms to vertically integrate, disproportionately more for …rms that are more capital intensive. The most interesting prediction of the model is the idea that opening to international trade is a means of disciplining domestic suppliers and thus reducing the need for vertical integration. This e¤ect should be larger when opening to countries that are characterized by well-functioning contract-enforcing institutions because this action enables domestic investors to …nd potential foreign partners that they can trust. This intuition is con…rmed by the data. In the …rst two columns of Table 6, the average tari¤s on intermediates are computed, respectively, on imports from the entire world and on imports from high-income OECD countries. The coe¢ cient on the interaction term between tari¤s and asset speci…city increases by nearly one-fourth in the second column; this result validates the view that reducing tari¤s toward countries with better institutions is more e¤ective in attenuating hold-up distortions. A possible concern with these results is that a country’s level of openness to international trade may be correlated with omitted country characteristics that have di¤erential e¤ects on …rms characterized by di¤erent levels of asset speci…city. To ensure that this is not the case, columns 3 and 4 of table 6 control for output tari¤s. The point estimates are only slightly a¤ected. As can be observed, only tari¤s on intermediates have di¤erential e¤ects on machinery-intensive …rms, whereas the e¤ect of tari¤s on output is not signi…cant from both an economic and a statistical 23 point of view. This is a reassuring result since it is a further validation of theories according to which vertical integration is a response to hold-up distortions rather than being originated by a collusive motive. For instance, in Mendi (2011) an upstream monopoly uses vertical integration to discipline oligopolist downstream …rms. In this setting, tari¤s on output should increase the monopoly pro…ts and the incentives to sustain collusion by vertical integration while tari¤s on inputs should produce the opposite e¤ect. Finally, the last two columns of the table report the e¤ect of trade barriers on the vertical structure of …rms by separating countries according to whether the cost of enforcing a contract is below or above the sample median. Opening up to trade is more e¤ective in counteracting hold-up distortions in countries with more expensive contract enforcement. In conclusion, both the OLS and IV estimates are consistent with the theoretical predictions of the model regarding the e¤ect of the speci…city of assets on the vertical structure of the …rms that owned them. The e¤ect of the asset speci…city of …rms on their likelihood to vertically integrate is increasing according to the cost, time and number of procedures that are necessary to enforce contracts and according to the tari¤s for the import of intermediate inputs. This result does not change when controlling for potential confounding factors, such as the economic development, …nancial development and trade openness of a country. 4 Conclusion This paper investigates the cross-country determinants of vertical integration using a new …rm-level cross-country dataset. In particular, it revisits the e¤ects of the interaction between the technology intensity of …rms with some speci…c institutional features on the vertical integration decisions. This focus is motivated by both theory and anecdotal evidence. First, I develop a simple model, in the spirit of TCE theories, to study the e¤ects of contract enforcement and international trade on the vertical structure of …rms. The model suggests that technology intensity should have greater e¤ects on the vertical structure of …rms when combined with low-quality contracting institutions and high trade barriers. The empirical results are consistent with these predictions and are robust to di¤erent econometric speci…cations and techniques. I conclude that poor contract enforcement can distort the vertical structure of …rms in the 24 presence of speci…c assets. This distortion can have signi…cant welfare costs, especially in developing economies. If improving home institutions is not feasible, then an equivalent solution is to open to international trade. This solution would attenuate the hold-up distortions by disciplining domestic suppliers and thus reducing the need for vertically integrated organizations. 25 References [1] Acemoglu, D. , Aghion, P., Gri¢ th, R. and Zibilotti F. (2010), "Vertical integration and technology: theory and evidence", Journal of the European Economic Association 8, pp. 9891033. [2] Acemoglu, D. , Johnson S. and Mitton T. (2009), "Determinants of vertical integration: …nance contracts and regulation", Journal of Finance 63, pp. 1251-1288. [3] Adelman, M. A. (1995), "Concept and statistical measurement of vertical integration", Business Concentration and Price Policy, Princeton University Press, Princeton. [4] Almeida H., Campello, M. and Hackbarth, D. (2011), "Liquidity Mergers," NBER Working Papers 16724. [5] Antras, P. (2003), "Firms, contracts and trade structure", The Quarterly Journal of Economics 118, pp. 1375-1418. [6] Antras, P. and Helpman, E. (2004), "Global Sourcing", Journal of Political Economy 112, pp. 552-580. [7] Barnerjee, A. and Munshi, K. (2004), "How E¢ ciently is Capital Allocated? Evidence from the Knitted Garment Industry in Tirupur", Review of Economic Studies 71, pp. 19-42. [8] Barney, J. B., Edwards, F. L. and Ringleb, A. H. (1992), "Organizational responses to legal liability", Academy of Management Journal 35, pp. 328-349. [9] Bender, C. (2002), "The theory of the …rm revisited: changing …rm boundaries in a new information and communication environment", Working Paper, Department of International Business, University of Muenster. [10] Chongvilaivan A. and Hur, J. (2012), "Trade Openness and Vertical Integration: Evidence from the US Manufacturing Sector", Southern Economic Journal 78, pp. 1242-1264. [11] Conconi, P., Legros, P. and Newman, A. (2012), "Trade Liberalization and Organizational Change", Journal of International Economics 86, pp. 197-208. 26 [12] The Economist. "Manufacturing, the ins and outs of outing", 1991. [13] Fan, J., Huang, J., Morck, R., Yeung, B. (2009), "Institutional determinants of Vertical Integration: Evidence form China", NBER Working Paper 14650. [14] Gort, M. (1962), "Diversi…cation and integration in American industry", Princeton University Press, Princeton. [15] Grossman, G. M. and Helpman, E. (2002), "Integration versus outsourcing in industry equilibrium", Quarterly Journal of Economics 117, pp. 85-120. [16] Grossman, G. and Hart, O. (1986), "The Costs and Bene…ts of Ownership: a Theory of Vertical and Lateral Integration", Journal of Political Economy 94, pp. 691-719. [17] Hart, O. and Tirole, J. (1990), "Vertical Integration and Market Foreclosure", Brooking Papers on Economic Activity: Microeconomics, pp. 205-286. [18] Johnston R. and Paul L. R. (1988): “Beyond the vertical integration: the rise of the valueadding partnership”, Harvard Business Review 6, pp. 94-101. [19] Khanna, T. and Palepu, K. (2000), "Is group a¢ liation pro…table in emerging markets? An Analysis of diversi…ed Indian business groups, Journal of Finance 55, 867-891. [20] Haltiwanger, J. and Schweiger, H. (2004), "Firm Performance and the Business Climate: Where does ICA Fit in?", mimeo. [21] Hart, O. and Moore, J. (1990), " Property rights and the nature of the …rm", Journal of Political Economy 86, pp. 1119-1158. [22] La¤er, A. B. (1969), "Vertical integration by corporations, 1929-1965", The Review of Economics and Statistics 51, pp. 91-93. [23] Levy, D. T. (1985), "The transaction cost approach to vertical integration: an empirical examination", The Review of Economics and Statistics 67, pp438-445. [24] Levy, B. (1991), “Transactions Costs, the Size of Firms and Industrial Policy: Lessons from a Comparative Case Study of the Footwear industry in Korea and Taiwan,” Journal of Development Economics 34, pp. 151-178. 27 [25] Macchiavello (2012), "Financial development and Vertical integration: Theory and Evidence", Journal of European Economic Association, forthcoming. [26] McLaren, J. (2000), "Globalization and vertical structure", American Economic Review 90, pp.1239-1254. [27] McMillan, J. and Woodru¤, C. (1999), "Dispute Prevention Without Courts in Vietnam", Journal of Law, Economics and Organization 15, pp. 637-658. [28] Mendi, P., Moner-Colonques, R. and Sempere-Monerris, J. (2011), "Vertical Integration, Collusion and Tari¤s", SERIEs 2, pp. 359-378. [29] Nelson, R. (1963), "Concentration in the manufacturing industries of the United States", New Haven, Yale University Press. [30] Nocke, V. and Yeaple, S. (2008), "Globalization and Endogenous Firm Scope", NBER Working Paper 12322. [31] Nunn, N. (2007), "Relationship-Speci…city, Incomplete Contracts and the Pattern of Trade", Quarterly Journal of Economics 132, pp. 569-600. [32] Ornelas, E. and Turner, J. (2008), "Trade liberalization, outsourcing and the hold-up problem", Journal of International Economics 74, pp. 225-241. [33] Tucker, I. B. and Wilder, R. P. (1977), "Trends in vertical integration in the US manufacturing sector", Journal of Industrial Economics 26, pp. 81-94. [34] Whinston, M. (2003), "On the Transaction Costs Determinants of Vertical Integration", Journal of Law, Economics and Organizations 19, pp. 1-23 [35] Williamson, O. (1975), "Markets and Hierarchies: an Analysis and Antitrust Implications", Free Press, New York. [36] Williamson, O. (1985), "The Economic Institutions of Capitalism", Free Press, New York. 28 A Derivations We can rewrite the pro…t function as E Proof of Lemma 1. and E O 2 = Q( ; ; I) + (1 ) (I; t) for I O 1 = Q( ; ; I) for I < I (t) I (t). The producer chooses investment in order to maximize this pro…t function. For I < I (t), the pro…t function is E O 1 = Q( ; ; I) and, by HP2, it is locally convex. Thus, if a local maximum exists in this range, then it is de…ned by the FOC 1 (1 ) 2 dQ( ; ;I) dI = 0, which can be rewritten as: g 0 (I ( ; )) 1 0 . For I I (t), the pro…t function becomes E O 2 = Q( ; ; I) + (1 ) (I; t) and, by the HP2 and HP3 is locally convex. Thus, if a local maximum exists in this range, then it is de…ned by the FOC dQ( ; ;I) dI )d + (1 1 (I;t) dI (1 = 0, which can be rewritten as: ) 1 ( 2 F ( g (I) t)) g 0 (I ( ; ; t)) 1 0 Notice that if I ( ; ) > I (t), there cannot be a local maximum for I < I (t) :In fact, the function Q( ; ; I) is strictly increasing in I for I < I (t). Therefore, the pro…t function will have a single local maximum in I = I ( ; ; t). In sum: If I ( ; ) > I (t), the pro…t function has a unique local maximum in I = I ( ; ; t). If I ( ; ) < I (t), the pro…t function can have at most two local maxima respectively in I ( ; ) and I ( ; ; t). Proof of Proposition 2. The producer will choose the level of investment that maximizes the pro…t function de…ned by equation 7. Given Lemma 1, the pro…t function has at most two local maxima: E O 1 ( ; ) = Q( ; ; I ( ; )) and E O 2 ( ; ; t) = Q( ; ; I ( ; ; t))+(1 ) (I ( ; ; t) ; t). In order to prove that optimal investment under outsourcing is decreasing in trade barriers t, we will consider three di¤erent cases: 1. E O 1 ( ; ) is the global maximum for every t. In this case the optimal investment is I ( ; ) and does not depend on trade barriers. 29 2. E O 2 ( ; ; t) is the global maximum for every t. In this case the optimal investment is I ( ; ; t). Applying the implicit theorem function to equation 8, is possible to verify that I ( ; ; t) is a strictly decreasing function of t. 3. E O 1 ( ; ) is the global maximum for some values of t while E O 2 ( ; ; t) is the global max- imum for some others. Notice while the …rst one is not a¤ected by t, the second one is decreasing and continuos in t. Therefore, in this case it exists a unique b t at which the producer is indi¤erent between I ( ; ) and I E O 1 ( ; ) E ( ; ; t). This tari¤ is implicitly de…ned by: ; ;b t : Consider an increase in trade costs dt. For t < b t, the optimal in- O 2 ( ; ; t), which is decreasing in t. For t > b t; the optimal investment is I ( ; ), vestment is I which is not a¤ected by t. Finally, when t increases from b t dt to b t + dt, the optimal investment drops from I ( ; ; t) to I ( ; ). Thus, also in this case investment is not decreasing in t. O 1 If the pro…t function has two local maxima E Lemma 2. ( ; ) and E O 2 ( ; ; t), then it should be that: O 1 dE ( ; ) d < dE O 2 ( ; ; t) d Proof. Using the envelope theorem on equations 8 and 9: dE O 1 ( ; ) d dE O 2 ( ; ; t) @E = d O 2 ( ; ; t) = @ = @E O 1 ( ; ) @ = P+ 1 ( ; ; t)) 2 P + g(I 2 2 g(I ( ; )) g(I Z(t; )) t ( g (I ( ; ; t)) t pF ) dF pF 0 dE De…ne = O( 2 ; ;t) d 2 (g(I ( ; ; t)) dE O( 1 d ; ) . Using the two equations above, we get: g(I ( ; )) 1 2 g(I Z(t; )) t ( g (I ( ; ; t)) t pF ) dF pF 0 Given Lemma 1, the pro…t function can have two local maxima as long as I 30 ( ; ; t) > I (t) > I ( ; ). The fact that I ( ; ; t) > I (t) implies that: g(I 2 (g(I 1 2 ( ; ; t)) Z( ; ;t)) t ( g (I ( ; ; t)) t pF ) dF pF 0 g (I(t)) t > 2 Z 1 2 (g(I (t)) g I (t) t pF dF pF = 2 (g(I (t)) 0 where the last equality comes from the de…nition of I (t). Finally, using the equation above together with the fact that I (t) > I ( ; ), we get: > 2 (g(I ( ; ; t)) g(I ( ; )) Hence, if the pro…t function has two local maxima then dE O( 2 ; ;t) d O 1 dE O( 1 d ; ) < . Proof of Proposition 3. E > 0, which implies that: Given Lemma 1, the pro…t function has at most two local maxima: ( ; ) = Q( ; ; I ( ; )) and E O 2 ( ; ; t) = Q( ; ; I ( ; ; t)) + (1 ) (I ( ; ; t) ; t). In order to prove that optimal investment is increasing in contract enforcement, we can divide our analysis in three cases: 1. E O 1 ( ; ) is the global maximum for every : In this case the optimal investment is I ( ; ) for every . Applying the implicit theorem function to equation 9, is possible to verify that I ( ; ) is a strictly increasing function of . 2. E I O 2 ( ; ; t) is the global maximum for every ( ; ; t) for every . Applying the implicit theorem function to equation 8, is possible to verify that I 3. E . In this case the optimal investment is O 1 ( ; ; t) is a strictly increasing function of . ( ; ) is the global maximum for some values of while E O 2 ( ; ; t) is the global max- imum for some others. De…ne some in ( ; ; t) E O 1 ( ; ) E O 2 ( ; ; t). In this case ( ; ; t) takes positive values for and negative values for others. Together with the fact that (because sum of continuous function) and strictly decreasing in 31 ( ; ; t) is continuous (by claim 2), this observation implies that there exists a unique b such that if > b then = b we have ( ; ; t) < 0 and for O 1 the global maximum is E O 1 ( ; ; t) > 0, if ( ; b; t) = 0. In other words, if > b global maximum is E ( ; ), if the pro…t function has two global maxima E O 2 ( ; b) = E O 2 ( ; ; t) and if the optimal investment is I <b = b, ( ; b; t). Consider an increase < b, the optimal investment is I ( ; ) and hence is increasing in ; for in . For from b < b then ( ; ; t) and hence is increasing in . Finally when d to b + d , the optimal investment jumps up from I ( ; ) to I > b, increases ( ; ; t): thus also in this case the optimal investment in increasing in . Given Lemma 1 the pro…t function has at most two local maxima: Proof of Proposition 4. E O 1 O 2 ( ; ) and E O 1 1. E ( ; ; t) for every . Three subcases are possible: ( ; ) is the global maximum for every . In this case the optimal investment is I ( ; ) for every . Applying the implicit theorem function to equation 9, is possible to verify that I ( ; ) is a continuous and strictly decreasing function of . O 2 2. E ( ; ; t) is the global maximum for every . In this case the optimal investment is I ( ; ; t) for every . Applying the implicit theorem function to equation 8, is possible to verify that I O 1 3. E ( ; ; t) is a continuous and strictly decreasing function of . ( ; ) is the global maximum for some values of while E O 2 ( ; ; t) is the global maximum for some others. Consider ( ; ; t) values for some in E ( ; ) E O 2 ( ; ; t) and notice that, in this case, it takes positive and negative for others. Together with the fact that (because sum of continuous function) and strictly increasing in (1 ) 12 [g(I ( ; )) b such that if b; ; t O 1 is I b; g(I < b then =E O 2 O 1 ( ; ; t) is continuous (notice that: d ( ; ;t) d = ( ; ; t)]), this observation implies that there exists a unique ( ; ; t) < 0, if = 0. In other words, if global maximum is E E O 1 ( ; ) and if > b then ( ; ; t) > 0 and for < b the global maximum is E O 2 ( ; ; t), if = b, > b = b , the pro…t function has two global maxima b; ; t . Consider an increase in . For ( ; ; t) and hence is decreasing in ; for 32 < b, the optimal investment > b, the optimal investment is I ( ; ) and hence is decreasing in . Finally when jumps down from I increases from b d to b +d , the optimal investment ( ; ; t) to I ( ; ): thus also in this case the optimal investment in decreasing in . 33 B Data sources and construction B.1 Industry Each ICA survey reports the industry of the …rms in the sample according to a di¤erent classi…cation system and it is not clear what are the criteria used to choose a particular classi…cation system over a di¤erent one. I have used the UN correspondence tables to convert all the industries into ISIC Rev 3.1 or NACE Rev 1.1.14 B.2 Value of Output The ICA surveys collects information on "Total market value of the production" (c274c1y) and "Total sales" (c274a1y) but there is a high number of missing values. Therefore I have used the following strategy: Compute the number of observations about the value of the production and the total sales in each country. Generate the variable output which, in each country, equals the value of the production when the number of observations about the value of production is at least 125 percent the number of observations about total sales. Otherwise it equals the total sales. Notice that I could have also adjusted total sales by subtracting the variation in the variable "Inventories and stock" (281k1y). The problem is that inventories data are questionable: too many …rms report zero while for Brazil and Ethiopia, when output is computed in this way, it is mostly negative. B.3 Cost of Intermediate Goods I consider two di¤erent measures of raw material costs: "Raw material costs (excluding fuel)" (c274b1y) and "Total purchase of raw material (excluding fuel)" (c274d1y). I use the former variable when available and the latter otherwise. The cost of energy is computed by summing up the variables "Consumption of electricity" (c274f1y) and "Consumption of fuel" (c274g1y) when both are available and using the variable 14 The correspondence tables used for creating the dataset for this paper can be downloaded at: http://unstats.un.org/unsd/cr/registry/regot.asp. 34 "Consumption of energy" (c274e1y) otherwise. For the remaining missing values, I impute the share of energy in each sector over the raw material cost. Finally I compute the cost of intermediate production goods by summing up the cost of material and the cost of energy. B.4 Vertical integration Vertical integration is measured by the ratio of value added to sales (e.g.: (Total Output- Cost of intermediate)/Total output). This measure has been used in many previous studies but, as already discussed above, is susceptible to bias. This bias increases with the amount of value added by downstream …rms. For this reason my analysis is limited to …rms producing primarily in manufacturing industries. The observations in the …rst and the last percentile have been dropped in order to correct for outliers. B.5 Number of workers The ICA survey collects information on "Average number of permanent workers" (c262a1y) and "Average number of temporary workers" (c263a1y). It is not clear whether missing values for temporary workers indicate that there are no temporary workers in that …rm or that the respondent simply gives the total number of workers under the voice permanent workers. I choose to totally disregard data about temporary workers and consider permanent workers as the only measure of the labor used in the production process. No information on hours per worker are collected. The observations in the …rst and the last percentile have been dropped. 35 Table 2: Descriptive statistics per country Country Albania Angola Argentina Armenia Azerbaijan Belarus Bhutan Bolivia Bosnia and Herzegovina Botswana Brazil Bulgaria Burkina Faso Burundi Cameroon Cape Verde Chile Colombia Congo, Dem. Rep. Costa Rica Croatia Czech Republic Cote d’Ivoire Ecuador El Salvador Estonia Gambia Georgia Ghana Guatemala Guinea Guinea-Bissau Honduras Hungary Indonesia Kazakhstan Kosovo Kyrgyz Republic Laos Latvia Lebanon GDP 6433 3838 11148 5645 10951 14241 4571 3781 6328 10072 10521 11083 962 389 1970 3344 11288 7394 249 9949 15971 25553 1491 6227 5835 18194 1205 4549 1292 5945 902 801 3489 18001 4381 13032 9444 2064 2621 13205 10481 Obs 37 194 407 56 82 52 90 191 143 98 966 615 80 90 166 51 990 492 142 245 380 47 83 254 452 79 30 156 277 162 131 41 125 78 612 116 56 127 409 147 110 Country GDP Obs Lithuania Macedonia Madagascar Malawi Mauritania Mauritius Mexico Moldova Mongolia Montenegro Mozambique Namibia Nepal Nicaragua Niger Oman Panama Paraguay Peru Philippines Poland Romania Russia Rwanda Senegal Serbia Slovak Republic Slovenia South Africa Swaziland Tajikistan Tanzania Thailand Timor-Leste Turkey Uganda Ukraine Uruguay Uzbekistan Vietnam 13963 8664 810 644 1781 9133 12922 2422 3798 8026 775 5148 1316 2116 545 14194 8494 3784 6818 3216 13714 9024 15704 956 1565 9444 21414 28131 7220 3489 1671 1007 6820 1293 10447 1063 7343 9562 2027 2973 188 53 238 70 76 204 675 135 115 22 298 92 130 185 27 38 105 127 231 439 716 381 320 52 253 102 42 82 1136 60 217 239 1172 30 1019 283 413 163 160 826 GDP is per-capita PPP converted GDP in 2005 at current prices (in International Dollars). Obs is the number of …rms included in the sample. 36 37 Industry level data Fixed Assets/Sales Fixed + Intangible Assets/Sales Fixed Assets /Assets Fixed Assets + Intangible Assets/Assets Fixed Assets/Workers Fixed Assets + Intangible Assets/Workers 0.86 1.28 0.27 0.41 151.80 210.52 4.15 4.40 4.19 38.41 37.80 600.37 4.92 2.87 15,688 Country level data Enforcing contracts: costs Enforcing contracts: procedures Enforcing contracts: days Legal Right Index Credit Information Index GDP per capita (PPP, I$) Country-industry level data Tari¤s on intermediates Tari¤s on intermediates (HI OECD) Tari¤s on …nal goods 0.58 0.22 127.86 18.18 10.61 Firm level data Value added/Sales Machinery and equipment/Sales Number of workers Market Share Foreign ownership (%) Mean 2.32 2.47 0.13 0.12 1,117.41 1,146.61 2.83 3.05 3.01 29.84 5.92 290.99 2.19 2.04 37,704 0.22 0.23 429.24 11.73 28.66 St.Dev 0.26 0.44 0.17 0.34 37.49 74.77 2.14 2.25 2.03 21.25 33.00 405.00 3.00 1.00 1,122 0.42 0.05 11.00 9.53 0 25th Pctile Table 3: Descriptive statistics 0.38 0.78 0.23 0.39 53.24 102.21 3.53 3.81 3.54 29.97 38.00 547.14 5.00 3.00 2,987 0.60 0.14 30.00 16.33 0 50th Pctile 0.68 1.10 0.35 0.46 80.10 157.29 5.66 6.15 5.88 41.70 41.00 745.00 6.67 5.00 9,022 0.77 0.32 96.50 23.39 0 75th Pctile 402 402 402 402 402 402 7,409 7,409 7,409 82 82 82 82 82 82 20,216 20,216 20,088 18,929 19,741 N Table 4: Main Results (1 ) (2 ) (3 ) OLS 2SLS 2SLS D e p e n d e nt va ria b le is: Va lu e A d d e d / To ta l P ro d u c tio n (V I) 0 .0 2 1 9 (0 .0 1 5 2 ) AS (4 ) 2SLS (5 ) 2SLS (6 ) 2SLS (7 ) 2SLS AS TB 0 .0 0 4 5 6 * (0 .0 0 2 4 4 ) 0 .0 2 2 2 * * (0 .0 1 1 0 ) 0 .0 2 3 3 * * (0 .0 1 0 8 ) 0 .0 2 0 3 * (0 .0 1 1 6 ) 0 .0 2 3 7 * * (0 .0 1 1 4 ) 0 .0 2 2 2 * * (0 .0 1 1 0 ) 0 .0 2 2 3 * * (0 .0 1 1 0 ) AS CI 0 .0 0 0 3 6 0 * (0 .0 0 0 2 0 4 ) 0 .0 0 7 1 8 * * (0 .0 0 2 9 9 ) 0 .0 0 7 1 6 * * (0 .0 0 3 0 4 ) 0 .0 0 7 1 8 * * (0 .0 0 2 9 1 ) 0 .0 0 7 9 1 * * * (0 .0 0 2 9 8 ) 0 .0 0 7 1 8 * * (0 .0 0 2 9 9 ) 0 .0 0 7 2 4 * * (0 .0 0 2 9 8 ) # Wo rke rs -0 .0 1 1 2 * * * (0 .0 0 3 0 4 ) M a rke t S h a re AS 0 .0 0 1 5 7 * * (0 .0 0 0 7 6 6 ) -0 .2 6 4 * * (0 .1 1 0 ) GDP Fo re ig n (>1 % ) -0 .0 1 0 6 * * (0 .0 0 4 7 5 ) Fo re ig n (1 0 0 % ) IN D U S T RY D U M C O U N T RY D U M YEAR DUM sa rg a n p N r2 F YES YES YES 20231 0 .2 6 8 YES YES YES 0 .1 9 4 20216 YES YES YES 0 .1 9 5 20088 YES YES YES 0 .2 8 1 18929 YES YES YES 0 .1 9 6 20216 YES YES YES 0 .2 0 0 20216 -0 .0 1 2 7 * (0 .0 0 7 4 6 ) YES YES YES 0 .1 9 8 20216 1 7 .8 8 1 8 .0 6 3 1 .0 4 1 8 .3 8 1 7 .8 2 1 7 .8 5 The table reports OLS and 2SLS estimates for the years 2003-2009. The unit of observation is the …rm. The left hand side variable VI is the ratio of the value added produced by the …rm to the total sales. AS is the ratio of the value of machinery and equipment to the total sales. Size is the number of workers in the plant. TB is average tari¤s on intermediate inputs in the 4-digit ISIC industry code in the country where the plant is located. CI is the cost to resolve a commercial dispute in courts on a claim equivalent to 200 percent of the average income of the country. GDP is per-capita Gross Domestic Product at PPP international prices (in thousands international dollars). Sarganp reports the p-value of the Hansen-Sargan overidenti…cation test. F is the F statistics for weak identi…cation. Standard errors (robust and clustered at the level of the industry) are reported in parentheses. *** signi…cant at less than 1 percent; ** signi…cant at 5 percent; * signi…cant at 10 percent. 38 Table 5: Alternative measures of contract enforcement (1) 2SLS (2) 2SLS (3) (4) (5) (6) (7) 2SLS 2SLS 2SLS 2SLS 2SLS Dependent variable is: Value Added/Total Production (V I) AS T B 0.0110** (0.00513) 0.0111** (0.00524) 0.00990* (0.00531) 0.0110** (0.00513) AS CI2 0.00784* (0.00415) 0.00861** (0.00421) 0.00802** (0.00403) 0.00790* (0.00417) AS CI3 # Workers 0.0119** (0.00493) 0.0102** (0.00504) 0.0115** (0.00486) 0.000324* (0.000176) 0.000344** (0.000174) 0.000355** (0.000179) 0.0003427* (0.000177) -0.0115*** (0.00333) 0.00154** (0.000781) Foreign (100%) INDUSTRY DUM COUNTRY DUM YEAR DUM sarganp N r2 F 0.0114** (0.00486) -0.0115*** (0.00346) Market Share (8) 2SLS 0.00151* (0.000805) YES YES YES 0.161 20216 YES YES YES 0.188 20088 0.228 18929 -0.0119 (0.00798) YES YES YES 0.167 20216 18.10 18.05 15.13 18.31 YES YES YES 0.134 20216 YES YES YES 0.153 20088 YES YES YES 0.198 18929 -0.0135* (0.00744) YES YES YES 0.138 20216 14.78 14.65 14.91 15.01 The table reports OLS and 2SLS estimates for the years 2003-2009. The unit of observation is the …rm. The left hand side variable VI is the ratio of the value added produced by the …rm to the total sales. AS is the ratio of the value of machinery and equipment to the total sales. TB is average tari¤s on intermediate inputs in the 4-digit ISIC industry code in the country where the plant is located. CI2 and CI3 are respectively the number of procedures and the time (in days) to resolve a commercial dispute in courts on a claim equivalent to 200 percent of the average income of the country. Sarganp reports the p-value of the Hansen-Sargan overidenti…cation test. F is the F statistics for weak identi…cation. Standard errors (robust and clustered at the level of the industry) are reported in parentheses. *** signi…cant at less than 1 percent; ** signi…cant at 5 percent; * signi…cant at 10 percent. 39 Table 6: Robustness Checks: Adding Financial Development (1) (2) (3) (4) (5) 2SLS 2SLS 2SLS 2SLS 2SLS Dependent variable is: Value Added/Total Production (V I) (6) 2SLS AS AS T B 0.0240** (0.0122) 0.0257** (0.0122) 0.0217* (0.0128) 0.0243** (0.0122) 0.0242** (0.0121) 0.0250** (0.0121) AS CI 0.00668** (0.00270) 0.00651** (0.00270) 0.00677** (0.00266) 0.00672** (0.00269) 0.00668** (0.00270) 0.00716** (0.00304) AS F D -0.0547** (0.0267) -0.0559** (0.0272) -0.0533** (0.0272) -0.0515** (0.0255) -0.0554** (0.0266) -0.0461** (0.0226) AS F D2 # Workers -0.0124*** (0.00318) Market Share (0.0272) (0.000833) AS GDP INDUSTRY DUM COUNTRY DUM YEAR DUM sarganp N F YES YES YES 0.149 20216 15.52 YES YES YES 0.144 20088 15.42 YES YES YES 0.198 18929 33.29 -0.378 (0.208) YES YES YES 0.148 20216 15.63 YES YES YES 0.156 20216 15.51 YES YES YES 0.150 20216 11.20 The table reports 2SLS estimates for the years 2003-2009. The unit of observation is the …rm. The left hand side variable VI is the ratio of the value added produced by the …rm to the total sales. AS is the ratio of the value of machinery and equipment to the total sales. TB is average tari¤s on intermediate inputs in the 4-digit ISIC industry code in the country where the plant is located. CI is the cost to resolve a commercial dispute in courts on a claim equivalent to 200 percent of the average income of the country. FD and FD2 are respectively an index of the strength of legal rights and an index on the depth of …nancial information of the country. GDP is per-capita Gross Domestic Product at PPP international prices. Sarganp reports the p-value of the Hansen-Sargan overidenti…cation test. F is the F statistics for weak identi…cation. Standard errors (robust and clustered at the level of the industry) are reported in parentheses. *** signi…cant at less than 1 percent; ** signi…cant at 5 percent; * signi…cant at 10 percent. 40 Table 7: Alternative measures of trade openness AS CI AS T B (1) (2) (3) (4) (5) 2SLS 2SLS 2SLS 2SLS 2SLS Dependent variable is: Value Added/Total Production (V I) 0.00718** 0.00678** 0.00879** 0.00792** (0.00299) (0.00288) (0.00354) (0.00326) 0.0222** (0.0110) 0.0194* (0.0118) 0.0272*** (0.00944) AS T B2 AS T Bout INDUSTRY DUM COUNTRY DUM YEAR DUM sarganp N F YES YES YES 0.194 20216 17.88 YES YES YES 0.186 20216 17.69 (6) 2SLS 0.0115** (0.00516) 0.0280*** (0.0102) YES YES YES 0.183 8972 52.79 YES YES YES 0.797 10805 19.31 0.0286** (0.0142) 0.000183 (0.00406) YES YES YES 0.373 18754 25.82 -0.00248 (0.00372) YES YES YES 0.303 18754 25.47 The table reports 2SLS estimates for the years 2003-2009. The unit of observation is the …rm. The left hand side variable VI is the ratio of the value added produced by the …rm to the total sales. AS is the ratio of the value of machinery and equipment to the total sales. TB is average tari¤s on intermediate inputs in the 4-digit ISIC industry code in the country where the plant is located. TB2 is average tari¤s on intermediate inputs imported from highincome OECD countries. TBout is average tari¤s on …nal goods. CI is the cost to resolve a commercial dispute in courts on a claim equivalent to 200 percent of the average income of the country. Sarganp reports the p-value of the Hansen-Sargan overidenti…cation test. F is the F statistics for weak identi…cation. Standard errors (robust and clustered at the level of the industry) are reported in parentheses. *** signi…cant at less than 1 percent; ** signi…cant at 5 percent; * signi…cant at 10 percent. 41 Figure 2: The central line depicts the estimated marginal e¤ect of vertical integration (measured as the ratio of value added to total sales) on asset speci…city (measured as the ratio of the value of machinery and equipment to total sales). The other two lines de…ne the 5 percent con…dence boundaries. 42