The better you are the stronger it makes you: Evidence on the asymmetric impact of liberalization

Journal of Development Economics 99 (2012) 474–485

Contents lists available at SciVerse ScienceDirect

Journal of Development Economics journal homepage: www.elsevier.com/locate/devec

The better you are the stronger it makes you: Evidence on the asymmetric impact of liberalization☆ Leonardo Iacovone ⁎ The World Bank, 1818 H Street NW, Washington DC, 20433, USA

a r t i c l e

i n f o

Article history: Received 4 March 2009 Received in revised form 31 May 2012 Accepted 1 June 2012 JEL classification: D22 F14 F15 L22 L25 L60 O12 O14 O31

a b s t r a c t This paper shows how trade liberalization can have an asymmetric effect on heterogeneous firms. It develops a neo-Schumpeterian growth model predicting that the impact of liberalization on economic performance is positive “on average”, but more advanced firms benefit more. These predictions are tested using Mexican plant-level data confirming that, under NAFTA, the liberalization spurred productivity growth on average. However, the empirical analysis goes beyond estimating the average effect of liberalization and shows that more advanced firms benefited disproportionately more from the liberalization. Focusing on the mechanisms explaining these results, the paper shows that the results are not just driven by an increase in input usage and investments, but rather by innovative and managerial efforts as they are significantly stronger in those sectors where the scope for innovative activities is larger. © 2012 Elsevier B.V. All rights reserved.

Keywords: Liberalization NAFTA Innovation Heterogeneous firms Neo-Schumpeterian models

1. Introduction The relationship between liberalization and economic performance has been debated since Adam Smith and it is still today at the center of

☆ The author is thankful to Gerardo Leyva and Abigail Durán for granting access to INEGI data at the offices of INEGI in Aguascalientes under the commitment of complying with the confidentiality requirements set by the Mexican Laws. I also wish to thank each one of the INEGI's employees that helped me during the work at Aguascalientes, in particular Gabriel Romero and Alejandro Cano. I am especially indebted to Alan Winters and Gustavo Crespi for their support and guidance. Additionally, special thanks go to Rafael De Hoyos, Gerardo Esquivel, Caroline Freund, Beata Javorcik, Hiau Kee, Eric Verhoogen, Carolina Villegas, the participants at the conference on “Welfare Effects of Trade Liberalization in Less Developed Countries” at Anáhuac University, ELSNIT 2007 conference in Barcelona, World Bank Research Group Trade seminar, and the participants to the conference on “Organization and Performance: Understanding the Diversity of Firms” in Tokyo for their useful comments. Furthermore, I thank two anonymous referees and the editor, Gordon Hanson, for the extremely helpful comments. Finally I gratefully acknowledge the financial support from ESRC and LENTISCO. The views expressed in this paper are those of the author and should not be attributed to the World Bank, its Executive Directors or the countries they represent. ⁎ Tel.: + 1 202 458 4982. E-mail address: [email protected]. 0304-3878/$ – see front matter © 2012 Elsevier B.V. All rights reserved. doi:10.1016/j.jdeveco.2012.06.001

disputes. Even more than economists, policy makers consider the subject central to their concerns. In particular developing countries, which during the 1990s have undergone profound liberalizations with the objective of improving their economic performance and accelerating their growth rates, have recently started to put these policies under discussions because of mixed results.1 In this study we will try to explain why liberalization can result in outcomes that are different depending on the context and, more specifically, unevenly affect different firms. The paper develops a model explaining how the impact of liberalization can be asymmetric for different types of firms and test its main predictions focusing on the Mexican liberalization under the North America Free Trade Agreement (NAFTA). NAFTA provides an excellent opportunity to study a deep process of liberalization and integration with the global economy as it increased the pressures from import competition and entry threat of foreign competitors. The study shows that, consistent with a simple theoretical model, the impact of NAFTA was highly 1 “Some trade critics are bothered by the disappointing performance of Latin America since it slashed tariffs in the 1980s and 1990s while more protectionist China and Southeast Asia sped ahead” (“Pains from free trade spurs second thoughts”, Wall Street Journal, 30 March 1997).

L. Iacovone / Journal of Development Economics 99 (2012) 474–485

asymmetric on different types of firms and, while productivity growth “on average” is larger for those firms facing lower tariffs, this effect is weaker for plants more distant from the production technology frontier. A number of previous studies using firm-level 2 data have already shown that an increase in import competition tends to have a positive impact on economic performance (De Hoyos and Iacovone, 2011; Fernandes, 2007; López-Córdova, 2003; Nickell, 1996; Pavcnik, 2002; Tybout and Westbrook, 1995). However, these studies mainly focused on the “average” impact of liberalization and implicitly assume that the effect of liberalization is homogeneous across different firms. 3 Differently, some recent work from Aghion et al. (2004a) shows that the effect of competition is non-linear and heterogeneous depending on the productivity level of the firm. Only “good firms” (i.e. close to the productive frontier) are positively affected by competition as their innovative effort is enhanced in response to the increased entry threat of foreign competitors. Vice-versa “bad firms” (i.e. far from the productive frontier) are negatively affected because the intensified entry threat only reduces their expected profits as their efficiency is too low to allow them to compete successfully with foreign entrants (Aghion et al., 2004b). This paper contributes to the existing literature in two ways. Firstly, on the theoretical front it develops a neo-Schumpeterian model that extends the previous work by Aghion et al. (2004a) and still reaches consistent results with their model at industry-level, while generating novel results at the firm-level. Results that are consistent with many recent firm-level studies showing that trade liberalization tends to have a positive impact on firm performance “on average”. Secondly, on the empirical front, different from much of the previous literature that analyzed the effect of liberalization in a cross-country setting focusing mostly on outcomes (e.g. relationship between growth and openness), our approach is much more micro and concentrates on the specific mechanisms through which liberalization may influence within-plant firms' responses. Our empirical approach, while closely related to previous studies (Fernandes, 2007; Schor, 2004) retains various relevant innovations. First, similar to Fernandes (2007) and Schor (2004) we principally focus on one specific channel through which liberalization affects productivity, namely the “asymmetric” impact of increasing product market competition. However, as shown by Amiti and Konings (2007) and Schor (2004), another important channel to take into account is the increased access to imported inputs. Similarly, as recently shown by various studies, another channel to take into account, that can also affect asymmetrically heterogeneous firms, is the increased access to foreign markets (Bustos, 2011; Lileeva and Trefler, 2010; Verhoogen, 2008). For this reason, an important difference with respect to previous studies is that we control for both these channels, and confirm that both the effects of lower tariffs on intermediate inputs as well as that of lower tariffs in the export market are highly asymmetric. Second, in this paper we go beyond the identification of the “channels” through which liberalization affects productivity and focus on the mechanisms. With this objective, we first build a theoretical model focusing on innovative effort as the key mechanism behind the asymmetric effect of liberalization, and then empirically show how our results are consistent with this model as the asymmetric effect of liberalization is significantly stronger in sectors where this effort is likely to matter more. The paper is divided into five sections. Following this introductory section, the following one discusses the existing literature related to our study. Section 3 presents a set of stylized facts and develops the theoretical model. In Section 4, the predictions from the model are tested,

2 In the paper we refer interchangeably to “firm” and “plant”, however it is important to emphasize that in our empirical analysis the unit of observation is the plant consistently with the unit of observation in our data. 3 Schor's (2004) study on Brazil, and Fernandes' (2007) on the case of Colombia constitute two notable exceptions.

475

together with an explicit search to identify the potential mechanisms driving our results. The last section concludes and highlights some avenues for further research. 2. Existing literature This paper is related to various strands of the literature, in particular the recent works of Aghion and Bessonova (2006) and Aghion et al. (2004b) arguing that liberalization boosts innovative efforts, but only for those firms that are closer to the productive frontier. These models are characterized by three main elements. First, innovation is the main driver of firm-level growth. Second, a firm innovates as long as its post-innovation profits are larger than the pre-innovation ones. Third, the effect of a successful innovation is limited and a firm can only advance “one step at the time” over the “productivity ladder”. Therefore only a firm that is close to the productive frontier can prevent the potential competitor's entry threat by innovating. In such a model, tariff reduction increases the entry threat and boosts the incentives of “advanced” firms to innovate in order to preempt the potential foreign entry. For these firms, the increased competition reduces pre-innovation profits more than the post-innovation ones because if the firm does not innovate it risks losing all its market due to the entry of the foreign competitor. The consequences for less productive firms are very different, and the increased entry threat reduces their post-innovation profits more than the pre-innovation ones. In fact, the “laggard incumbents” (i.e. less productive) cannot prevent the entry of foreign competitors, even when able of successfully innovating because they are too far from the technology frontier. Also, related to this paper is the large set of studies arguing that increased competition puts pressures on “slacking managers” and pushes them to reduce the X-inefficiency (Leibenstein, 1978; Martin, 1978; Martin and Page, 1983). However, all these studies rely on a set of restrictive assumptions and normally assume that firms are homogeneous (Rodrik, 1988). Our work is clearly linked to previous empirical studies analyzing the effect of competition on innovative activities and productivity growth even if the earlier studies could only rely on industry-level data (Gerosky, 1990; Haskel, 1992). Even more closely related to our study are the studies using firm-level data to evaluate the impact of competition on productivity and innovation (Disney et al., 2003; Nickell, 1996) as well as the studies evaluating directly the impact of trade liberalization on firm-level productivity (De Hoyos and Iacovone, 2011; Pavcnik, 2002; Topalova, 2004; Tybout and Westbrook, 1995). However, as previously mentioned, these studies mostly focus on the average effect of liberalization and do not account for the possibility that this effect could be different across heterogeneous firms.4 Two notable exceptions, that we are aware of, explicitly allow for the impact of liberalization to be heterogeneous across different firms but differ from the present study for various reasons (Fernandes, 2007; Schor, 2004). First, while both studies mainly focus on the “asymmetric effect” of liberalization through the “import competition channel”, these do not take into account also two other channels: the expanded access to foreign markets5 and the intermediate input liberalization.6 This is important because, similar to the effect of the increased import competition, these channels could also affect productivity asymmetrically with more productive firms benefiting disproportionately more. Hence, it becomes important to control for both 4 In some additional robustness checks Topalova (2004) evaluates if different types of firms are asymmetrically affected by the unilateral trade liberalization in India and shows that the positive effect of tariff liberalization on productivity is significant only for privately owned companies but not the state owned ones. 5 This is especially important as various recent influential studies showed how the expanded market access can asymmetrically affect productivity and upgrading in a context where firms are heterogeneous (Bustos, 2011; Lileeva and Trefler, 2010; Verhoogen, 2008). 6 Schor (2004) does include the intermediate input channel and confirms that this channel also affects firms asymmetrically, however she does not control for the “expanded export market access” channel.

476

L. Iacovone / Journal of Development Economics 99 (2012) 474–485

these channels as omitting them may bias the results if tariff reduction on intermediate inputs, foreign markets and domestic goods are correlated. Finally, another important difference with these two studies lays in the fact that in the present paper we try to move beyond the identification of the “channels” through which liberalization can affect productivity and aim at capturing the specific mechanisms through which liberalization affects productivity. 7 This paper is very much part of the growing literature on heterogeneous firms showing how firm heterogeneity interacts with external policy changes generating dynamics that fit much better to the empirical evidence (Bernard et al., 2004; Griffith et al., 2003; Helpman et al., 2004; Hopenhayn, 1992; Jovanovic, 1982; Melitz, 2003; Melitz and Ottaviano, 2003; Sabirianova et al., 2005; Yeaple, 2005). In particular these studies show how the impact of liberalization and FDI depends on the initial productivity of the firms, with more productive firms benefitting disproportionately more from globalization than the less productive ones. 3. Trade liberalization and productivity 3.1. The facts Before discussing more in details the NAFTA reform and its impact it is useful to draw attention on the importance of micro-level heterogeneity. Productivity differences between firms, within narrowly defined industries (i.e. 6 digit CMAP), are not smaller than the average differences between industries (see Fig. 1 in the online appendix). We are certainly not the first to discover and emphasize the importance of plant-level heterogeneity. In fact, in a review of various plant-level studies, Tybout (2000) noticed how one of the distinctive features of LDC's manufacturing sector is its dualism where “large numbers of microenterprises and a handful of modern, large-scale factories produce similar products side by side”. Similarly, previous empirical studies have confirmed the persistence of a high degree of heterogeneity even after a profound liberalization process. For example, in the case of Chile during the 1980s and 1990s Crespi (2005) found that the productivity of firms in the top decile was 256% larger than that of those in the bottom decile. Analogously, Disney et al. (2003) found a difference of “only” 155% in UK after important liberalization reforms. Having stressed the importance of plant-level heterogeneity it is important to notice that after the implementation of NAFTA we observe a remarkable increase in industrial inequality with good firms getting better, and larger firms getting larger. Analyzing the evolution of productivity distribution during the period 1993–2002 we observe two fundamental trends: a shift in the mean, as Mexican firms became on average more productive, and an increase in the spread of the productivity distribution.8 This result is evident when plotting the distribution of the productivity gap between each individual firm-level productivity and the “productive frontier”. This distribution moves rightward, indicating an expansion of the gap between the average Mexican firm and the production technology frontier, while the right tail becomes 7 An additional difference with Schor (2004) consists in the fact that our results about the “heterogeneous” effects of own tariffs are substantially different. Similar to the findings of Fernandes (2007) and consistent with our “neo-Schumpeterian” model we find that “the more distant” a firm is from the production frontier, the less its productivity grows in response to the liberalization shock. Differently Schor (2004) finds that least productive firms, conditional on survival, respond more to tariff cut and explains that this is driven mainly by a selection mechanism with liberalization pushing low productivity plants either to exit or to make substantial improvements. Our findings, are similar when evaluating the “heterogeneous effects” through the “import tariff channel” with more productive firms benefiting more from the increased availability of cheaper inputs. 8 We formally tested the hypothesis that the productivity spread increased by regressing the coefficient of variation, calculated within each narrowly defined sector at six digits, on a linear trend. We find that the coefficient on the linear trend estimated with a FE model is positive and statistically significant confirming that dispersion of productivity among firms even within narrowly defined sectors got larger during the period under analysis.

fatter, implying an increase in the density of firms with larger gaps (see Fig. 2 in the online appendix). This expansion in productivity inequality is even more interesting if we consider that during this period we observe the exit of a significant number of less productive firms. In fact, Fig. 1 depicts the number of exiting plants in panel (a), and shows in panel (b) that the average productivity of exiting plants (dashed line) is always substantially smaller than the average productivity of surviving ones (continuous line). In a nutshell, after NAFTA was implemented besides a substantial exit of less productive plants, which should compress the productivity distribution by trimming its left tail, we observe an increase in productivity inequality among surviving firms. A final fact to discuss is the importance of the liberalization process under NAFTA. During the 1990s Mexico underwent a process of deep integration with its North-American neighbors. This process marked the completion of a liberalization already started during the second part of the 1980s. However, it is important to notice that, different from the unilateral liberalization of the 1980s, the liberalization under NAFTA locked in Mexican policy makers much more than the previous reforms because of the credibility imposed by an agreement with a powerful counterpart. Therefore, when considering the scope of the liberalization under NAFTA we need to take into account also the importance of this credibility effect (Tomz, 1997; Tornell and Esquivel, 1995). In this perspective, NAFTA implied a deeper liberalization than the drop in the average tariffs would suggest. Furthermore in Fig. 2 we show that the reduction of NAFTA tariffs, applied to products manufactured in US or Canada, was not negligible. While the average tariff went down from about 16% in 1993 to less than 5% in 2002, the tariff peak was reduced from about 70% in 1993 to about 20% in 2002. 3.2. The model In order to explain the facts presented in the previous section, in particular the expansion of the within industry inequality after a profound liberalization, we develop a model with heterogeneous firms based on the neo-Schumpeterian growth models. The key intuition driving the results of these classes of models is that liberalization can generate at the same time contrasting effects on the incentives to innovate. On one side, higher competition can limit the incentives to innovate of incumbent firms by reducing their expected profits (defined as the “Schumpeterian effect” by Aghion et al. (2002)). At the same time, the liberalization may lead the incumbent to increase its innovative efforts in order to “escape competition”. The synthesis provided by the neo-Schumpeterian growth models consists exactly in allowing these two mechanisms to coexist and interplay, and provides the theoretical basis to explain the stylized facts presented in Section 3.1. Our model builds on the work of Aghion et al. (2002, 2004a, 2004b) and introduces a key innovation by relaxing the assumption that the impact of increased competition on backward firms is invariably negative. Instead, we also allow backward firms to be able to catch-up with the productive frontier, even if this will be less likely because of their initially disadvantaged condition. For this reason, different from Aghion et al. (2004b), in our model the impact of increasing competition on the innovative effort will be positive on average for all plants. While this may appear not plausible in certain contexts, it is not unreasonable to accept that having incurred sunk set-up costs even weaker companies will “fight until death” before accepting exiting from the market. 9 We characterize this model by discussing separately (1) domestic production, (2) innovation decision, (3) foreign competition, and (4) equilibrium innovation. 9 In a context where entrepreneurs that fail have an especially difficult time at re-starting a new business this is a common attitude as confirmed during interviews with various companies in Mexico.

477

.6 0

.2

.4

Density

.8

1

L. Iacovone / Journal of Development Economics 99 (2012) 474–485

−1

−2

0

1

2

Log Ratio of Median Sectoral Labour Productivity to Economy−Wide Median Labour Productivity

0

.5

Density

1

1.5

(a) Heterogeneity Between Sectors

−.5

0

.5

1

1.5

2

Log Ratio of Plant Labour Productivity to Median Sectoral Labour Productivity within 6−digits Sector

(b) Heterogeneity Within Sectors Fig. 1. Firm demography. (a) Number of exiting firms. (b) Average Productivity of exitors vs stayers. Source: INEGI, Aguascalientes.

3.2.1. Domestic production In the economy we have a final good y that is produced using a continuum of intermediate goods v ∈ [1, 0]. The production is described by Eq. (1) where xt(v) is the quantity used of the intermediate

input v and At(v) measures its productivity at time t. Finally α is a parameter varying between 0 and 1. This final good can be consumed, used to produce intermediate inputs or invested in innovation.

1 1 1−α α ∫ A ðvÞxt ðvÞdv α 0 t

ð1Þ

.2

.4

.6

Each intermediate good is produced by a monopolist 10 at a constant marginal cost equal to one unit of the final good. The monopolist maximizes its profits but its monopoly power is restricted by the existence of a set of “fringe firms” that do not operate in equilibrium but could produce the same input using χ unit of output. Basically χ is a parameter capturing the competition intensity and is larger than 1. Therefore the maximum price that the monopolist can charge for the intermediate good v is pt(v) and can be expressed in terms of the output that is the numeraire

0

Kernel Density Distribution

.8

yt ¼

0

1

2

3

4

Log of Distance from Frontier 1993

1998

Fig. 2. Import tariff drop under NAFTA. Source: Secretaria de Economia, Mexico.

5

pt ðvÞ ¼ χ:

ð2Þ

2002

10 This assumption can be changed and has two producers competing under the Bertrand competition.

478

L. Iacovone / Journal of Development Economics 99 (2012) 474–485

Because the final-good producing sector is perfectly competitive the price of the intermediate good v must be equal to its marginal product MP t ðvÞ ¼

dyt α−1 ¼ α ðxt ðvÞ=At ðvÞÞ dxt ðvÞ

ð3Þ

therefore from Eqs. (3) and (2) we obtain that 1

xt ðvÞ ¼ ðχ=α Þα−1 At ðvÞ:

ð4Þ

The model can be easily solved as shown by Aghion and Griffith (2005) and the profit function is inversely correlated with the strength of the competition, proxied by χ, and positively correlated with the productivity of the firm producing the intermediate input v as described by Eq. (5) below. πt ðvÞ ¼ At ðvÞδðχ Þ

ð5Þ

where 1 χ α−1 : δðχ Þ ¼ ðχ−1Þ α

ð6Þ

3.2.2. Innovation decision In each period the technological frontier evolves exogenously at a rate g as described by Eq. (7) A t ¼ A t−1 ð1 þ g Þ

ð7Þ

and there are two types of firms as explained by the following equation ( firm type ¼

advanced or type 1 if at the end of t−1 At−1 ¼ A t−1 : backward or type 2 if at the end of t−1 At−1 ¼ At−2

In order to characterize the innovation decision we will assume that the advanced firm will successfully innovate, and catch up, with probability z, where z is exactly measuring its innovative effort. Analogously, the backward firm is able to innovate, however, because of its relative position with respect to the technological frontier, when it does innovate and move “one step forward” with probability z this is not enough to catch up with the frontier. In fact, in order to catch up the backward firm needs to move “two steps ahead” over the technology ladder and to do so it needs to make an extra innovation effort. This extra effort will be successful only with probability s. In the original version of the model presented by Aghion and Griffith (2005) the backward firm does not have any chance of catching up (i.e. moving “two steps ahead”) therefore the introduction of this extra effort and the fact that the backward firm is able to catch up with probability s is the principal innovation of our model. However, in order to make this more realistic we impose the condition that s is smaller than z (i.e. it is more likely that a firm is successful in moving “one step” ahead rather than “two steps”). To simplify we can also assume that s ¼ θz where θ is a parameter that we assume smaller than one but not smaller than g. Basically we want to allow the backward firm to be able to catch up with the technological frontier. g≤θ≤1

ð8Þ

Following Aghion et al. (2004b) we assume that the cost of innovating is quadratic in the research effort and linear in the current technological level as in Eq. (9). However, for backward firms we need to take into account also the cost of the extra-effort. So the innovation cost function is c1t for the advanced firm and c2t for the backward one. 8 1 > < c1t ¼ z2 At−1 ðvÞ 2 innovation cost ¼ > : c ¼ 1 z2 A ðvÞ þ 1 s2 A ðvÞ 2t t−2 t−2 2 2

ð9Þ

At the end of every period, all firms that remain backward will then be upgraded automatically, assuming there are spillovers from mature technology, and move up one step in the technology ladder. This simplifies the model by allowing us to have only two types of firms to deal with. Finally, at any moment t there is an exogenous probability h that any firm may exit and be replaced by a new advanced firm at time t + 1. 3.2.3. Foreign competition In every period foreign competitors can enter the domestic market, and their decision is done after having observed the outcome of the innovation effort of the domestic firms. A foreign company needs to incur in a sunk cost equal to ξ to enter the domestic market. Once it has paid this, the firm will be able to successfully penetrate the domestic market with probability μ, which captures how difficult it is to enter the domestic market (i.e. proxy for tariffs and regulations). Because, the foreign firm is assumed to be at the technological frontier, if entry is successful and it faces a backward domestic firm, then it gains the entire market, if instead it faces a domestic firm at the frontier then they engage in Bertrand competition and both firms will see their profits going to zero. Consequently the entry threat simplifies into the following condition

entry threatt ¼

8 0 > > < > > :

μ

if domestic firm is at frontier in t−1 and innovates succesfully in tor is backward in t−1 but able to reach the frontier in t otherwise:

3.2.4. Equilibrium innovation The solution of the expected profit maximization problem for the backward firm is obtained by solving Eq. (10) h i max E½π 2t  ¼ δðχ Þ zð1−μ ÞA t−1 þ ð1−zÞð1−μ ÞA t−2 þ sA t þ ð1−sÞA t−2 ð1−μ Þ ð10Þ z  1 2 2 − z þ s A t−2 2

because when, with probability z, this firm is successful in moving one step ahead and obtain a productivity A t−1 , it will retain its domestic market only if the foreign firm does not enter with probability 1 − μ. Similarly, when it is unsuccessful at innovating with probability 1 − z and maintains productivity A t−2 , it will retain its market only if the foreign firm does not enter, with probability 1 − μ. The backward firm also engages in an extra-effort to catch-up with productive frontier and, if successful, with probability s, obtains a productivity equal to A t and retains the market. While if unsuccessful, with probability 1 − s, it maintains productivity A t−2 and retains its domestic market as long as the foreign firm is unable to enter with probability 1 − μ. Solving this ∗ maximization problem we obtain the optimal innovative effort z2t of the backward firm as described by Eq. (11) below. 

z2t ¼

δ ½g ð1 þ 2θ þ θg Þ þ μ ðθ−g Þ 1 þ θ2

ð11Þ

L. Iacovone / Journal of Development Economics 99 (2012) 474–485

At the same time the advanced firm chooses its optimal innova∗ tion effort z1t maximizing its expected profits π1t as in Eq. (12) h i 1 2 max E½π1t  ¼ δðχ Þ zA t þ ð1−zÞð1−μ ÞA t−1 − zt A t−1 z 2

ð12Þ

because it retains the market when it successfully innovates, with probability z, and obtains a productivity A t , in which case the foreign firm stays out of the market. The advanced firm also keeps its market if unsuccessful at innovating, with probability 1 − z, and a resulting productivity of A t−1 as long as, with probability 1 − μ, the foreign firm is unable to enter. Therefore the optimal innovation effort of this firm is equal to 

z1t ¼ δðg þ μ Þ:

ð13Þ

After having derived the optimal innovative effort for both firms, we can now determine what is the effect of a reduction in import barriers which increases entry threat. In particular we derive two predictions from Eqs. (11) and (13). dz2t δ ¼ ðθ−g Þ dμ 1 þ θ2

ð14Þ

dz1t ¼δ dμ

ð15Þ

Prediction 1. Liberalization by increasing entry threat (μ) increases the optimal innovative effort of all types of firms. Prediction 2. Liberalization increases the optimal innovative effort of advanced firms more than that of backward firms. The first prediction derives immediately from the fact that both δ θ−gÞ ðθ−gÞ and ð1þθ 2 are positive. The second prediction is a consequence of 1þθ2

being smaller than one. The predictions of our model are consistent with those of Aghion et al. (2004b) at the industrial level because an increase in the foreign competition (increase in μ) expands the industrial inequality as advanced firms react to it with a larger innovative effort than the backward ones. However, when we analyze the model prediction at the firm level, our model implies that the effect of liberalization is, on average, positive for both advanced and backward firms, while in Aghion et al. (2004b) the impact on backward firms' innovative effort is always negative because the Schumpeterian effect dominates. The predictions of our model are consistent with the basic intuition from the previous literature on X-efficiency suggesting that increased competition is expected to spur efforts (Leibenstein, 1978), as well as various empirical studies analyzing the impact of trade reforms on productivity. These studies normally found that the “average” effect of increased competition is positive (De Hoyos and Iacovone, 2011; Fernandes, 2007; Nickell, 1996; Tybout and Westbrook, 1995). In other words, our model extends that of Aghion and Griffith (2005) and Aghion et al. (2004b) by making their predictions more in line with previous empirical studies but maintains its central feature intact, which is that liberalization has an unequal effect and advanced firms benefit more from it. However, it is important to acknowledge one important shortcoming of our model. In the absence of threat from foreign competition, for certain ranges of our parameters, the equilibrium innovative effort of less competitive firms is larger than that of the more competitive ones. Given that our focus here is on the impact of increasing external competitive pressures, we consider that a situation where there is no threat from external competition lies outside of the domain of applicability of this model.

479

4. Empirical analysis The data used in our analysis are collected by INEGI and cover the entire period of NAFTA reforms, 1993–2002. The data are collected at the plant level, and individual establishments are identified by a unique identifier which allows us to build a panel. After having cleaned the dataset the number of establishments varies between 6500 and about 5000 because of attrition. The sampling structure of the survey is such that the data tend to be more representative of larger firms, overall the survey covers 85% of industrial output excluding the “maquiladoras”. The survey frequency is yearly and the variables collected cover various aspects of the firm's operations: workers, wages, electricity usage, intermediate inputs, production, inventory, domestic sales, exports, investment in different types of physical assets, investment in R&D and in technology transfers (for more details on the data see the data description in the online appendix or refer to Iacovone (2008)). In our empirical analysis we address three questions to test the predictions of our model. First, what is the average impact of increased import competition on productivity growth? Second, is this impact the same for all firms? Consistent with our model, the implicit assumption that we initially make here is that productivity growth is mainly driven by innovative efforts. Third, in the last part of our analysis, we explicitly assess this assumption and focus on the mechanisms driving our results. 4.1. The asymmetric impact of liberalization Based on the theoretical model developed in the previous section, we would expect the impact of increased foreign competition to spur productivity growth on average, but do so differently across firms with different productivity levels (i.e. distance from production technology frontier). To test this, we first calculate the domestic production frontier defining it as the average value added labor productivity (i.e. deflated value added divided by the number of workers) of the top five firms in each sector, 11 and for every firm we measure the distance from its domestic frontier as in Eq. (16). Specifically, for firm i belonging to sector j this is equal to the ratio between the productive frontier πjtF and its own labor productivity πit. As in Griffith et al. (2003), the wider the gap between the productivity of a firm and that of the top firms in its sector, the larger this distance index is.

DLF ijt ¼

πFjt πijt

ð16Þ

We then estimate Eq. (17), where our explanatory variable is the growth of the value added labor productivity of firm i between t and t + 1, and our main explanatory variables are the distance from the frontier (DLFijt), the tariffs faced by its foreign competitors in sector j and their interaction. We also add year dummies, to control for the effect of macroeconomic shocks, as well as location and industry dummies (at 6 digit CMAP). Furthermore, in some specifications, we control for size, measured by the number of employees, capital intensity, measured by the ratio of capital stock to number of workers, and average wages as a proxy for the quality of human capital. Finally, we also control for the plant investments in R&D and technological transfers. All variables, except tariffs, are in logarithm and we estimate the model using both OLS and FE estimators. Because we have a lagged dependent variable in this equation appearing on the denominator of the distance index our estimates will be biased (Cameron and Trivedi, 2005, p.764).

11 Each sector has on average between 98 and 132 plants, depending on the years, with the largest one has 470 plants and the smallest 11.

480

L. Iacovone / Journal of Development Economics 99 (2012) 474–485

Table 1 Liberalization and competition: multiple outcomes. (1)

(2)

Total sales MX NAFTA tariffs (lagged) Distance from the frontier

0.009⁎ (0.005) − 0.753⁎⁎⁎ (0.031)

Distance × tariffs (lagged) N r2 Plant FE Location FE Year FE

47,212 0.089 Yes Yes Yes

(3)

(4)

Employment − 0.032⁎⁎⁎ (0.007) − 0.883⁎⁎⁎ (0.037) 0.022⁎⁎⁎ (0.003) 47,212 0.094 Yes Yes Yes

0.000 (0.001) − 0.066⁎⁎⁎ (0.009)

46,791 0.029 Yes Yes Yes

(5)

(6)

Profits − 0.012⁎⁎⁎ (0.002) − 0.106⁎⁎⁎ (0.011) 0.007⁎⁎⁎ (0.001) 46,791 0.033 Yes Yes Yes

0.002 (0.004) − 0.956⁎⁎⁎ (0.027)

41,202 0.130 Yes Yes Yes

(7)

(8)

PCM − 0.040⁎⁎⁎ (0.006) − 1.101⁎⁎⁎ (0.032) 0.023⁎⁎⁎ (0.002) 41,202 0.136 Yes Yes Yes

− 0.008 (0.008) − 0.235⁎ (0.131)

39,947 0.001 Yes Yes Yes

(9)

(10)

Exit − 0.040 (0.028) − 0.348⁎ (0.202) 0.017 (0.011) 39,947 0.001 Yes Yes Yes

− 0.002⁎⁎⁎ (0.000) 0.027⁎⁎⁎ (0.003)

43,036 0.039 Yes Yes Yes

0.001⁎ (0.001) 0.037⁎⁎⁎ (0.003) − 0.002⁎⁎⁎ (0.000) 43,036 0.041 Yes Yes Yes

Note: ⁎⁎⁎ denotes significant at 1%, ⁎⁎ at 5%, ⁎ at 10%.

In particular the OLS coefficient will be downward biased while the FE coefficients will be upward biased (Arellano, 2003). Δπ ijt ¼ β0 þ β1 DLF ijt−1 þ β2 T jt−1 þ β3 T jt−1  DLF ijt−1 þ β4 X ijt−1 þ β5 Year þ μ it ð17Þ Based on our model we expect β2 to be negative as lower tariffs should promote higher productivity growth spurred by the increased competitive pressure. We also expect β3 to be positive, because, as discussed in Section 3.2 the impact of increased liberalization is less positive for firms that are more distant from the technological frontier. Before presenting our main results, it is important to confirm that indeed there is evidence that NAFTA increased the degree of competition in Mexico, as this is a crucial assumption in our model. Following Head and Ries (1999) who show that Canadian tariff reduction decreases the sales of Canadian firms and Trefler (2004) who shows that Canadian tariff reduction decreased employment of Canadian firms, we analyze the impact of NAFTA tariff reduction on sales and employment, as well as on profits, markups (proxied by price–cost margins) and probability of exiting. The results presented in Table 1 confirm that reduction in tariffs is associated with increasing competitive pressures as it caused a reduction in sales, employment, profits, markups, and increased the likelihood of exiting. However, what is especially interesting is that also these results are highly asymmetric and less productive firms hit harder.12 We then move to our baseline results, reported in Table 2. We first present a set of regressions where we only include our main variables of interest and the fixed effects, models (1) and (2), and then include also the plant-level controls, models (3) and (4). Consistent with our expectations on the bias induced by the lagged dependent variable, the coefficient on the variable measuring the distance from the frontier, which contains a lagged dependent variable in its denominator, appears to be larger in the FE (model 2) than in the OLS (model 1). However it is important to stress that the qualitative results do not change between the two models, in fact, in both models these variables are significant and have the same sign. Our main finding is that the productivity of firms facing lower tariffs tends to grow faster but this productivity effect is lower for firms with a larger distance from the productive frontier as indicated by the positive coefficient on the interaction term. The coefficients of the remaining variables are consistent with what we would have expected a priori. In particular, we find evidence of catching up as distance from the frontier positively affects subsequent productivity growth. It is also interesting to notice that the productivity of larger and more capital intensive firms tends to grow faster. Also, as expected, the productivity of firms paying higher wages grows faster. And finally, higher expenditures in R&D and technology transfers promote faster productivity growth. 12 In Table 5 of the online appendix we report similar estimates also including firm-level controls and the results are substantially the same.

However, it is important to stress that these control variables are potentially endogenous, therefore we do not want to push too much on their interpretation. The principal justification of introducing them is to control for the firms' characteristics that may be correlated with both distance and productivity growth and therefore, if omitted, could bias our principal coefficients of interest. However, their potential endogeneity is the reason why in columns (1) and (2) of Table 2 we have excluded these regressors, in order to confirm that their inclusion is not crucial for our results.13 Because of the presence of the interaction we can calculate the impact of a percentage change in tariffs on productivity growth for different types of firms depending on their distance from the productive frontier. In Fig. 3 we show these marginal effects based on the results from columns (3) and (4) of Table 2. The bottom line is that for most of the plants, except those in the top decile, i.e. the ones that are more distant from the production frontier, the tariff reduction spurs their productivity growth. 4.2. Robustness checks After having presented our main results, we will now discuss a set of potential concerns that may invalidate our findings and assess if these are truly robust. To begin, an important concern to be addressed is the possibility that tariffs could be endogenous. If we analyze the way NAFTA negotiations were carried out we can strengthen our confidence about the Mexican tariff phase out being exogenous for various reasons. First, the NAFTA agreement placed Mexico in bilateral negotiations with partners that had more influence and negotiating power. Additionally, the negotiations were carried out in a relatively short period of time which reduced the possibility of external interventions and interferences from individual firms. Moreover, the high degree of uncertainty surrounding the US Congress approval suggests that Mexican negotiators focused their minds mainly on supporting the negotiating process to increase the chances of NAFTA being approved, rather than trying to defend the interests of individual firms.14 These historical remarks are consistent with the findings of Kowalczyk and Davis (1996) who found that while for US tariffs there is evidence that sectors with higher duties and lower intra-industry trade were characterized by slower phase out, there is not such evidence when analyzing the phase out of Mexican tariffs except that this appear correlated with the US phase-out.15 13 As a robustness check we re-estimated the model and instead of labor productivity we use a TFP index, similar to the one used by Aw et al. (2001). The results presented in Table 6 of the online appendix confirm that our findings are robust to the use of this alternative performance measure. 14 NAFTA was only approved by the US Congress only on December 8th, 1993, by a narrow majority of 234–200 votes and it took effect after less than 1 month on January 1st, 1994. 15 Kowalczyk and Davis (1996) argue that an empirical analysis at the five-digit SITC level “suggests that from the perspective of Mexican negotiations, many other issues, including the overriding one of obtaining free trade in the near future with its northern neighbors, were given higher priority than the questions of how to phase duties out”.

L. Iacovone / Journal of Development Economics 99 (2012) 474–485 Table 2 Asymmetric effect of liberalization.

Domestic frontier growth MX tariffs NAFTA (lagged) Distance from the frontier (lagged) Distance × tariffs (lagged)

Marginal Effect of % Tariff Change on % Productivity Growth 1.5

OLS

FE

OLS

FE

(1)

(2)

(3)

(4)

0.240⁎⁎⁎ (0.015) − 0.013⁎⁎⁎ (0.002) 0.112⁎⁎⁎

0.392⁎⁎⁎ (0.014) − 0.017⁎⁎⁎ (0.002) 0.500⁎⁎⁎

0.260⁎⁎⁎ (0.015) − 0.012⁎⁎⁎ (0.002) 0.173⁎⁎⁎

0.391⁎⁎⁎ (0.014) − 0.015⁎⁎⁎ (0.002) 0.499⁎⁎⁎

(0.007) 0.003⁎⁎⁎

(0.011) 0.006⁎⁎⁎ (0.001)

(0.008) 0.002⁎⁎⁎ (0.001) 0.045⁎⁎⁎ (0.003) 0.021⁎⁎⁎ (0.002) 0.083⁎⁎⁎

(0.011) 0.006⁎⁎⁎ (0.001) 0.086⁎⁎⁎ (0.013) 0.004 (0.003) − 0.026 (0.016) 0.006⁎⁎⁎ (0.002) 0.004⁎⁎⁎ (0.002) No No Yes Yes Yes 44,176 0.217

(0.001) Employment (lagged) Capital intensity (lagged) Average wages (lagged) R&D (lagged) Technology transfers (lagged) Industry FE (6 digits) Location FE Year FE Plant FE Plant-level controls N r2

Yes Yes Yes No No 44,948 0.056

481

No No Yes Yes No 44,948 0.214

(0.009) 0.004⁎⁎⁎ (0.001) 0.006⁎⁎⁎ (0.001) Yes Yes Yes No Yes 44,176 0.075

Note: ⁎⁎⁎ denotes significant at 1%, ⁎⁎ at 5%, ⁎ at 10%.

Additionally, to dispel any remaining doubt, the analysis of the evolution of Mexican tariffs under NAFTA confirms that, even at a rather disaggregated level (i.e. 6 digits), the rankings of the Mexican tariffs under NAFTA remained fundamentally stable. This is confirmed both by evaluating a transition matrix where tariffs are split into deciles and transition probabilities along the diagonal can be shown to vary between 50 and 86% (see Table 7 in the online appendix), as well as by calculating the Spearman rank correlation between tariffs. No matter which pair of years we consider, we always reject the null hypothesis that the two tariff schedules are independent (see Table 8 in the online appendix). This suggests that whatever political economy existed in the pre-NAFTA period, this did not change much during the NAFTA period. Therefore, following the same empirical approach adopted by Goldberg and Pavcnik (2005) and Schor (2004), using the 6-digit industry fixed effects is sufficient to control for these time-invariant industry characteristics affecting the political economy of tariff liberalization. However, having shown that the NAFTA tariff reduction was not affected by Mexican firms, or that at least the political economy forces behind tariff setting in Mexico did not change through time, does not rule out entirely the possibility that tariffs could still be endogenous due to the presence of omitted variables correlated with both the degree of tariff reduction at the industry level and industry characteristics. In particular, we would be especially concerned if the NAFTA liberalization was skewed toward “export-intensive”16 sectors where Mexico had a comparative advantage (i.e. unskilled-labor intensive sectors). In such case, the demand for firms in the industries experiencing the tariff reduction might rise because of the Heckscher–Ohlin type of reallocation across industries. In order to address this concern we proceed in three steps. First, we re-estimate our main baseline model and include a proxy for skill intensity and capital intensity at the industry level (6 digits), respectively measured as a ratio of white collars over the total workers and the amount of capital per worker. The results are reported in Table 3 and show that indeed firms from industries characterized by lower skill intensity experienced faster productivity growth, however the introduction of these new controls does not affect our main results. 16 As shown by Hanson and Harrison (1999) and Revenga (1997) this was indeed the case for the initial patterns of protection before the liberalization in the 1980s.

1 0.5 0 -0.5 -1 OLS

FE

-1.5 1%

5%

10%

25%

50%

Mean 75%

90%

95%

99%

Percentile - Distance from Frontier Fig. 3. Marginal effect of tariffs on productivity growth.

Second, we run a simple regression to evaluate the correlation between the NAFTA total tariff cuts between 1994 and 2002 and average measures of capital or skill intensity at the 6-digit industry level in 1994. The results suggest that while there is a positive correlation between tariff cuts and skill intensity this is not statistically significant (see Table 9 in the online appendix). Third, we argue that if the HO reallocation was at play we should observe sales increasing in sectors characterized by tariff reductions. However, in columns (1) and (2) of Table 1, we exactly showed that firms facing lower tariffs tend to reduce their sales, even if this effect is asymmetric and more productive firms are less affected. Given that NAFTA did change the geographical production patterns of Mexico, shifting much of the industrial production toward the northern region, we could be concerned that our results may be biased by these time-varying geographical trends. As in Hanson (2004) we control for this geographic trends by including region-year fixed effects. The results, presented in the online appendix (see Table 10), confirm that our key findings are not affected by the inclusion of these additional controls.17 The recent work by Amiti and Konings (2007) as well as the findings by Schor (2004) points toward the crucial importance of input liberalization to explain the evolution of firm productivity. If input and output tariffs are correlated, we could be concerned about the potential omitted variable bias when just the final product tariffs are included. In Table 4 we re-estimate our baseline model and include tariffs on intermediate inputs as well as their interaction with distance from the frontier.18 The results confirm that our key findings about the asymmetric impact of liberalization are robust. Interestingly, consistent with the findings of Schor (2004), we also find that the intermediate input liberalization operates as an additional channel through which productivity is affected, and this channel too has a highly asymmetric impact with firms that are closer to the production technology frontier benefiting the most. 19 In all our discussion we have assumed that increased competition and entry threat, captured by a reduction in Mexican tariffs under NAFTA, are having an uneven effect on firms with different productivity levels. However, it is also possible that Mexican tariffs may be capturing some other effects. For example, if Mexican tariffs and US tariffs under NAFTA are correlated the reduction in Mexican tariffs may be identifying the impact of enhanced market access under NAFTA as we have 17 Our results do not change if we use state-year fixed effects instead of using region-year fixed effects and are available upon request. 18 Intermediate input tariffs are calculated using the input–output tables provided by INEGI. 19 As discussed by López-Córdova (2003) another channel that would be important to take into account operates through the FDI spillovers however, because of data limitations, we cannot pursue this idea further as the EIA did not collect information on foreign ownership, except for the year 1993.

482

L. Iacovone / Journal of Development Economics 99 (2012) 474–485

Table 3 Controlling for industry level skills and capital intensity. (1)

(2)

(3)

(4)

(5)

(6)

Distance × tariffs (lagged)

0.278⁎⁎⁎ (0.016) − 0.010⁎⁎⁎ (0.002) 0.167⁎⁎⁎ (0.009) 0.003⁎⁎⁎

0.428⁎⁎⁎ (0.015) − 0.015⁎⁎⁎ (0.003) 0.542⁎⁎⁎ (0.014) 0.007⁎⁎⁎

0.260⁎⁎⁎ (0.015) − 0.012⁎⁎⁎ (0.002) 0.173⁎⁎⁎ (0.008) 0.002⁎⁎⁎

0.392⁎⁎⁎ (0.014) − 0.015⁎⁎⁎ (0.002) 0.499⁎⁎⁎ (0.012) 0.006⁎⁎⁎

0.278⁎⁎⁎ (0.016) − 0.010⁎⁎⁎ (0.002) 0.167⁎⁎⁎ (0.009) 0.003⁎⁎⁎

0.428⁎⁎⁎ (0.015) − 0.015⁎⁎⁎ (0.003) 0.542⁎⁎⁎ (0.014) 0.007⁎⁎⁎

Employment (lagged)

(0.001) 0.045⁎⁎⁎ (0.003) 0.013⁎⁎⁎ (0.002) 0.082⁎⁎⁎ (0.009) 0.005⁎⁎⁎

(0.002) 0.074⁎⁎⁎ (0.016) 0.004 (0.003) − 0.051⁎⁎⁎ (0.019) 0.004⁎⁎

(0.001) 0.045⁎⁎⁎ (0.003) 0.021⁎⁎⁎ (0.002) 0.083⁎⁎⁎ (0.009) 0.004⁎⁎⁎

(0.001) 0.086⁎⁎⁎ (0.014) 0.004 (0.003) − 0.026 (0.017) 0.006⁎⁎⁎

(0.001) 0.045⁎⁎⁎ (0.003) 0.013⁎⁎⁎ (0.002) 0.082⁎⁎⁎ (0.009) 0.005⁎⁎⁎

(0.002) 0.073⁎⁎⁎ (0.016) 0.004 (0.003) − 0.051⁎⁎⁎ (0.019) 0.004⁎⁎

(0.001) 0.006⁎⁎⁎ (0.001) − 0.189 (0.138)

(0.002) 0.004⁎⁎ (0.002) − 0.390⁎⁎⁎ (0.139)

(0.001) 0.006⁎⁎⁎ (0.001)

(0.002) 0.004⁎⁎ (0.002)

− 0.019⁎ (0.011) 44,178 0.075 Yes Yes Yes No

0.006 (0.012) 44,178 0.217 No Yes Yes Yes

(0.001) 0.006⁎⁎⁎ (0.001) − 0.183 (0.138) − 0.020 (0.012) 38,040 0.070 Yes Yes Yes No

(0.002) 0.004⁎⁎ (0.002) − 0.394⁎⁎⁎ (0.139) 0.012 (0.012) 38,040 0.223 No Yes Yes Yes

Domestic frontier growth MX tariffs NAFTA (lagged) Distance from the frontier (lagged)

Capital intensity (lagged) Average wages (lagged) R&D (lagged) Technology transfers (lagged) Product median skill intensity Product median capital intensity N r2 Industry FE (6 digits) Location FE Year FE Plant FE

38,040 0.070 Yes Yes Yes No

38,040 0.223 No Yes Yes Yes

Note: ⁎⁎⁎ denotes significant at 1%, ⁎⁎ at 5%, ⁎ at 10%.

not included US tariffs in our baseline regressions. To address this problem we include US tariffs and their interaction with distance from the frontier. The results are reported in Table 5, where in columns (1) and (2) we exclude the plant-level controls while in columns (3) and (4) we also include them. For clarity we will concentrate on the results reported in columns (1) and (2) as these are substantially similar to the ones where we include the plant-level controls, but are not affected by the potential endogeneity of these additional regressors. Table 4 Controlling for tariffs on intermediate inputs.

Domestic frontier growth MX tariffs NAFTA (lagged) Input tariffs NAFTA (lagged) Distance from the frontier (lagged) Distance × tariffs (lagged) Distance × input tariffs (lagged) Employment (lagged)

(1)

(2)

(3)

(4)

0.236⁎⁎⁎ (0.016) − 0.017⁎⁎⁎ (0.002) − 0.007⁎⁎⁎ (0.001) − 0.050⁎⁎⁎

0.398⁎⁎⁎ (0.015) − 0.013⁎⁎⁎ (0.003) − 0.003⁎⁎⁎ (0.001) 0.483⁎⁎⁎

0.238⁎⁎⁎ (0.016) − 0.017⁎⁎⁎ (0.002) − 0.008⁎⁎⁎ (0.001) − 0.028⁎⁎⁎

0.396⁎⁎⁎ (0.015) − 0.012⁎⁎⁎ (0.003) − 0.002⁎⁎ (0.001) 0.485⁎⁎⁎

(0.006) 0.004⁎⁎⁎

(0.012) 0.003⁎⁎

(0.007) 0.004⁎⁎⁎

(0.012) 0.003⁎⁎

(0.001) 0.004⁎⁎⁎ (0.000)

(0.001) 0.002⁎⁎⁎ (0.001)

(0.001) 0.004⁎⁎⁎ (0.000) 0.043⁎⁎⁎

(0.001) 0.002⁎⁎⁎ (0.000) 0.090⁎⁎⁎

(0.003) 0.016⁎⁎⁎ (0.002) 0.025⁎⁎⁎ (0.009) 0.002 (0.001) 0.004⁎⁎⁎

(0.014) 0.003 (0.003) − 0.028 (0.018) 0.006⁎⁎⁎ (0.002) 0.004⁎⁎⁎

(0.001) 42,211 0.062 Yes Yes Yes No

(0.002) 42,748 0.218 No Yes Yes Yes

Average wages (lagged) R&D (lagged)

42,940 0.053 Yes Yes Yes No

Table 5 Controlling for US tariffs.

Domestic frontier growth

Capital intensity (lagged)

Technology transfers (lagged) N r2 Industry FE (6 digits) Location FE Year FE Plant FE

The principal conclusion is that the inclusion of US tariffs does not alter our previous results, and the coefficients on Mexican tariffs and their interaction with distance from the frontier are substantially unchanged. Further, it is interesting to notice that the increased access to US

43,493 0.214 No Yes Yes Yes

Note: ⁎⁎⁎ denotes significant at 1%, ⁎⁎ at 5%, ⁎ at 10%.

MX tariffs NAFTA (lagged) US NAFTA tariff (lagged) Distance from frontier (lagged) Distance × MX tariffs (lagged) Distance × US tariffs (lagged) Exiting plant

OLS

FE

(1)

(2)

FE

(3)

(4)

0.243⁎⁎⁎ 0.387⁎⁎⁎ (0.016) (0.014) − 0.013⁎⁎⁎ − 0.015⁎⁎⁎ (0.002) (0.003) − 0.007⁎ − 0.015⁎⁎⁎

0.262⁎⁎⁎ 0.386⁎⁎⁎ (0.016) (0.014) − 0.012⁎⁎⁎ − 0.014⁎⁎⁎ (0.002) (0.003) − 0.005 − 0.013⁎⁎⁎

(0.004) 0.141⁎⁎⁎ (0.007) 0.003⁎⁎⁎

(0.005) 0.516⁎⁎⁎ (0.012) 0.005⁎⁎⁎

(0.004) 0.200⁎⁎⁎ (0.008) 0.003⁎⁎⁎

(0.001) 0.003 (0.002) − 0.714⁎⁎⁎ (0.040)

(0.001) 0.005⁎⁎

Employment (lagged) Capital intensity (lagged)

(0.001) 0.002 (0.002) (0.002) − 0.692⁎⁎⁎ − 0.677⁎⁎⁎ (0.038) (0.040) 0.046⁎⁎⁎ (0.003) 0.018⁎⁎⁎

Average wages (lagged) R&D (lagged) Technology transfers (lagged) Industry FE (6 digits) Location FE Year FE Plant FE Plant-level controls N r2

OLS

Yes Yes Yes No No 41,480 0.098

No No Yes Yes No 41,480 0.245

Note: ⁎⁎⁎ denotes significant at 1%, ⁎⁎ at 5%, ⁎ at 10%.

(0.005) 0.515⁎⁎⁎ (0.012) 0.004⁎⁎⁎

(0.002) 0.085⁎⁎⁎

(0.001) 0.004⁎ (0.003) − 0.671⁎⁎⁎ (0.039) 0.080⁎⁎⁎ (0.013) 0.001 (0.003) − 0.034⁎⁎

(0.009) 0.004⁎⁎ (0.001) 0.006⁎⁎⁎

(0.017) 0.006⁎⁎⁎ (0.002) 0.004⁎⁎⁎

(0.001) Yes Yes Yes No Yes 40,760 0.115

(0.002) No No Yes Yes Yes 40,760 0.247

L. Iacovone / Journal of Development Economics 99 (2012) 474–485 Table 6 Controlling also for exchange rate. OLS

FE

OLS

FE

(17)

(18)

(19)

(20)

0.243⁎⁎⁎

0.389⁎⁎⁎

0.262⁎⁎⁎

(0.016) − 0.012⁎⁎⁎ (0.002) − 0.006 (0.004) 0.177⁎⁎⁎

(0.014) − 0.009⁎⁎⁎ (0.003) − 0.008 (0.005) 0.666⁎⁎⁎

(0.016) − 0.011⁎⁎⁎ (0.002) − 0.003 (0.004) 0.240⁎⁎⁎

0.387⁎⁎⁎ (0.014) − 0.009⁎⁎⁎ (0.003) − 0.007 (0.005) 0.652⁎⁎⁎

(0.027) − 0.001a (0.000) 0.002⁎⁎ (0.001) 0.002 (0.002) − 0.714⁎⁎⁎ (0.040)

(0.031) − 0.001⁎⁎⁎ (0.000) 0.002 (0.001) 0.002 (0.003) − 0.688⁎⁎⁎ (0.039)

Capital intensity (lagged)

(0.028) − 0.001a (0.000) 0.002⁎ (0.001) 0.001 (0.002) − 0.677⁎⁎⁎ (0.040) 0.046⁎⁎⁎ (0.003) 0.018⁎⁎⁎

Average wages (lagged)

(0.002) 0.085⁎⁎⁎

(0.032) − 0.001⁎⁎⁎ (0.000) 0.002 (0.001) 0.001 (0.003) − 0.669⁎⁎⁎ (0.039) 0.074⁎⁎⁎ (0.013) 0.001 (0.003) − 0.035⁎⁎

(0.009) 0.004⁎⁎ (0.001) 0.006⁎⁎⁎ (0.001) Yes Yes Yes No Yes 40,760 0.115

(0.017) 0.006⁎⁎⁎ (0.002) 0.004⁎⁎⁎ (0.002) No No Yes Yes Yes 40,760 0.248

Domestic frontier growth MX tariffs NAFTA (lagged) US NAFTA tariff (lagged) Distance from frontier (lagged) Distance × exchange rate (lagged) Distance × MX tariffs (lagged) Distance × US tariffs (lagged) Exiting plant Employment (lagged)

R&D (lagged) Technology transfers (lagged) Industry FE (6 digits) Location FE Year FE Plant FE Plant-level controls N r2

Yes Yes Yes No No 41,480 0.098

No No Yes Yes No 41,480 0.246

Note: ⁎⁎⁎ denotes significant at 1%, ⁎⁎ at 5%, ⁎ at 10%. a Indicates a coefficient that is smaller than 0.0005.

markets has a similar positive effect on productivity as the increase of domestic competition, in fact the size of the coefficients on US tariffs is very close to that on Mexican tariffs. Additionally, it is intriguing to notice that, consistent with the findings of Verhoogen (2008), also the effect of expanded market access asymmetrically affects Mexican firms, with the more productive ones benefitting more. As a further robustness check we test if our results are influenced by the 1994 peso devaluation. It is fair to argue that, simultaneous to the NAFTA reforms, Mexican firms were also affected by the devaluation shock which could also influence different plants asymmetrically. In order to control for this we introduce the interaction between real exchange rate and the distance from the frontier as additional control. The results are presented in Table 6 and show that the original impact of tariffs is somewhat attenuated and while in the OLS estimation the interaction between tariffs and distance from frontier is still positive and significant, in the case of the fixed-effect estimation this remains positive but becomes marginally insignificant. 20 Finally, we briefly discuss three additional robustness checks which results are also reported in the online appendix. First, we exclude from our analysis the “top plants” used to calculate the productive frontier as their inclusion could lead to spurious correlation by construction (columns 1–4 of Table 11 in the online appendix). Second, because our dataset is an unbalanced panel and includes plants that at some point close down and exit from the sample we are concerned that exiting could influence our results. In order to

20 This reduced significance of our key coefficient of interest may generate some concerns about the robustness of our results, however as we will show in the last part of our empirical analysis, this may be driven by the fact that in sectors where there is more limited scope for innovative activities our results become weaker, when controlling also for the exchange rate, but this is not the case for sectors where the scope for innovative activities is larger.

483

address this problem we first introduce a dummy variable that is equal to one when a firm will be exiting in the year and zero otherwise (columns 5–6 of Table 11 in the online appendix), and we then repeat our estimations by using a balanced panel and excluding altogether exiting plants (columns 7–8 of Table 11 in the online appendix). As a final robustness check instead of using the lagged value of distance from the frontier we define this distance in the initial year of the sample (i.e. 1993) and use this initial value instead than the time-varying lagged distance from the frontier (Table 12 in the online appendix). While this is a time-invariant measure of the distance from the frontier, its advantage is that it can be considered predetermined prior to NAFTA. Our main results are robust to all these robustness checks. 4.3. Focus on mechanisms After having shown that, consistent with the predictions of our model, the NAFTA liberalization increased competition and asymmetrically influenced productivity growth with firms closer to the production technology frontier benefiting disproportionately more. In the last part of our empirical analysis we focus on the mechanisms driving this relationship. In fact, productivity growth could be driven by various mechanisms such as reduction in X-inefficiencies, increased use of capital, increased investment in technology, and expanded innovation effort to improve either products or organizational processes. The latter mechanism being the one especially emphasized in our model is consistent with the neo-Schumpeterian literature such as in Aghion et al. (2004b). In order to shed light on the mechanisms driving our results we proceed in three steps. First, we present some additional regressions where we evaluate the effect of tariff liberalization on a set of variables proxying for some of the mechanisms discussed above. Second, we will discuss more in details some results presented earlier where we included, as firm-level controls, variables that could be capturing some of these mechanisms: capital intensity, R&D and technology transfer expenditures. Finally, we re-estimate our model and evaluate if our effects are significantly different for those sectors characterized by a larger scope for product or process innovation. One mechanism to increase labor productivity is to use more, or improved, inputs and assets. Therefore, in Table 7 we analyze if in response to tariff liberalization Mexican companies increased investments in assets, R&D investments, technology transfer investment and what we defined as “total technology investment” equal to the sum of expenditures for R&D and technology transfers. For each mechanism we estimate our regressions using industry fixed effects (at the 6 digit CMAP) or firm fixed effects. The results provide some limited evidence, mostly when just industry fixed effects are included, that indeed liberalization promoted investments in technology and fixed assets, and it did so especially for more productive companies. The results for total technology investments are the more robust as they remain significant, even if just marginally significant, also when controlling for firm fixed effects. To understand the role of these investments further we now focus again on Table 2, our baseline results, where the main difference between the first two columns and the last ones is due to the inclusion of some lagged firm characteristics such as capital intensity, expenditure for R&D and technology transfers. If the mechanisms driving the productivity changes during the liberalization were driven by an increase in capital intensity, or an increase in the expenditures on technology developed in-house (through R&D) or externally purchased (through technology transfers) we should observe a reduction in our main coefficient of interests. However, we only observe a small reduction of our main coefficients of interest pointing toward the fact that while additional investments in inputs, both physical and technological, seem to be partially responsible for our results, at the same time these are not driven by just more inputs, but something that goes beyond it such as innovative and managerial efforts.

484

L. Iacovone / Journal of Development Economics 99 (2012) 474–485

Table 7 Asymmetric effect of liberalization: potential mechanisms. (1)

(2)

(3)

Net investment per worker

R&D

Distance × tariffs (lagged)

− 0.006⁎ (0.003) 0.052⁎⁎⁎ (0.011) 0.002⁎⁎

− 0.010⁎⁎⁎ (0.004) − 0.289⁎⁎⁎ (0.012) 0.004⁎⁎⁎

N r2 Industry FE (6 digits) Location FE Year FE Plant FE

(0.001) 44,782 0.011 Yes Yes Yes No

MX NAFTA tariffs (lagged) Distance from the frontier (lagged)

− 0.002 (0.002) 0.158⁎⁎⁎ (0.014) − 0.001 (0.001) 44,782 0.010 No Yes Yes Yes

(0.001) 47,211 0.117 Yes Yes Yes No

(4)

− 0.006 (0.005) − 0.078⁎⁎⁎ (0.017) 0.002 (0.002) 47,211 0.004 No Yes Yes Yes

(5)

(6)

(7)

(8)

Technology transfers

Total technology investment

− 0.022⁎⁎⁎ (0.006) − 0.761⁎⁎⁎ (0.019) 0.012⁎⁎⁎

− 0.030⁎⁎⁎ (0.006) − 0.894⁎⁎⁎ (0.021) 0.013⁎⁎⁎

− 0.010⁎ (0.005) − 0.164⁎⁎⁎ (0.024) 0.004⁎

(0.002) 47,211 0.216 Yes Yes Yes No

(0.002) 47,211 0.003 No Yes Yes Yes

(0.002) 47,211 0.185 Yes Yes Yes No

− 0.002 (0.005) − 0.110⁎⁎⁎ (0.021) 0.002 (0.002) 47,211 0.003 No Yes Yes Yes

Note: ⁎⁎⁎ denotes significant at 1%, ⁎⁎ at 5%, ⁎ at 10%.

If innovative and managerial efforts are driving our findings we would expect that our results would be stronger in sectors where the scope for product and process innovation is larger. In order to provide more evidence to support this conjecture we use Rauch's classification that splits products into “differentiated” and “homogenous” ones (Rauch, 1999). Homogenous products are those commodities traded on exchanges or with set reference prices, while differentiated ones are those for which quality differences, reflecting multidimensional characteristics relevant to consumers, are particularly important and therefore cannot be traded on exchanges nor reference prices can be easily set. The intuition is that for “homogenous products” we expect that what matters is mostly cost-cutting innovation, and the scope for product innovation is therefore more limited. Alternatively for “differentiated” products we expect that the “potential quality differences” between products give to producers more scope, and incentives, to Table 8 Focus on mechanisms: triple interaction for differentiated and R&D intensive sectors. (1)

Domestic frontier growth MX tariffs NAFTA (lagged) Differentiated × MX tariffs (lagged) R&D intensive × MX tariffs (lagged) US NAFTA tariff (lagged)

(2)

(3)

Differentiated sectors

R&D intensive sectors

0.221⁎⁎⁎ (0.016) − 0.010⁎⁎

0.219⁎⁎⁎ 0.364⁎⁎⁎ (0.016) (0.015) − 0.020⁎⁎⁎ − 0.010⁎⁎⁎ (0.002) (0.004)

0.372⁎⁎⁎ (0.016) − 0.007 (0.004) (0.009) − 0.012⁎⁎⁎ − 0.032⁎⁎⁎ (0.004)

(0.009)

− 0.006⁎⁎⁎ (0.002) − 0.019⁎⁎⁎ − 0.045⁎⁎⁎ − 0.014⁎⁎⁎ (0.004) (0.007) (0.004) Distance from the frontier − 0.061⁎⁎⁎ 0.064⁎⁎⁎ − 0.061⁎⁎⁎ (lagged) (0.007) (0.011) (0.007) Distance × exchange rate 0.001⁎⁎⁎ 0.002⁎⁎⁎ 0.001⁎⁎⁎ (lagged) (0.000) (0.000) (0.000) Distance × MX tariffs (lagged) − 0.000 0.008⁎ 0.007⁎⁎⁎ (0.002) (0.004) (0.001) Differentiated × distance × MX 0.008⁎⁎⁎ 0.011⁎⁎ tariffs (lagged) (0.002) (0.005) 0.004⁎⁎⁎ R&D intensive × distance × MX tariffs (lagged) Distance × US tariffs (lagged) Exiting plant N r2 Industry FE Location FE Year FE Plant FE

(4)

(0.001) 0.010⁎⁎⁎

0.022⁎⁎⁎ 0.007⁎⁎⁎ (0.002) (0.004) (0.002) − 0.707⁎⁎⁎ − 0.697⁎⁎⁎ − 0.706⁎⁎⁎ (0.040) (0.045) (0.040) 40,739 40,739 40,739 0.097 0.226 0.097 Yes No Yes Yes No Yes Yes Yes Yes No Yes No

Note: ⁎⁎⁎ denotes significant at 1%, ⁎⁎ at 5%, ⁎ at 10%.

− 0.007⁎⁎⁎ (0.003) − 0.005 (0.005) 0.637⁎⁎⁎ (0.035) − 0.001⁎⁎⁎ (0.000) 0.002 (0.002)

0.005⁎⁎⁎ (0.001) 0.001 (0.003) − 0.677⁎⁎⁎ (0.043) 40,739 0.241 No No Yes Yes

introduce not only cost-cutting process innovation but also product innovation. Accordingly, in Table 8 we extend our baseline regressions, following the same specification as in Table 6, and include a triple interaction of our main variables of interest with a dummy equal to one if a product is “differentiated” using Rauch's classification. The results reported in columns 1–2 show that the asymmetric effect of the liberalization is especially strong in “differentiated” good sectors, and these differences are statistically significant. To further strengthen our confidence about the importance of innovative efforts as key drivers behind our findings, we split our sample based on the average sectoral R&D intensity in the pre-NAFTA period (i.e. 1993) and identify with a dummy those sectors having an R&D intensity above the economy-wide median. We then present new results introducing a triple interaction between our main variables of interest and the “innovation intensive” sector dummy in columns 3–4 of Table 8 to confirm that in more R&D intensive sectors the asymmetric effect of competition is significantly stronger. 21 Finally, in the online appendix, we report additional results where we interact our key variables of interest with a proxy for the degree of dependence from external finance as defined by Rajan and Zingales (1998). This index is a proxy to characterize the degree of dependence from external funds of a specific sector (at 4 digits) and it is equal to the average ratio of capital expenditures not financed with internal funds over the total capital expenditures. The results, reported in Table 13 of the online appendix, confirm that our main results appear not to be driven by financial constraints as the coefficient of the triple interaction is not statistically different from zero. That is the asymmetric effect of competition is not significantly different in sectors characterized by a higher dependence from the financial sector. This result is interesting because it is consistent with the conjecture that the mechanisms driving our findings are not driven by access to finance. 5. Conclusions In this paper we have shown that it is important, when analyzing the impact of exogenous policy shocks, to take into account microlevel heterogeneity. Starting from the stylized fact that industrial inequality increased after NAFTA liberalization, we proposed an explanation based on a simple neo-Schumpeterian growth model. The model predicts that the impact of liberalization is asymmetric across different types of firms, with “good firms” benefitting more from the increase in competitive pressures than the “bad ones”. In this 21 It is important to underline that these results also include the exchange rate interacted with the distance from the frontier, as in Table 6, which generates additional confidence in the robustness of our results, at least for those sectors characterized by larger opportunities for innovative efforts.

L. Iacovone / Journal of Development Economics 99 (2012) 474–485

model, the liberalization tends to generate two competing effects: on one side it spurs more innovative efforts, because of the increased entry threat by foreign competitors, on the other the enhanced competition curtails expected profits and reduces the resources available to finance innovative activities. The “pro-competitive effect” is weaker for firms more distant from the frontier, as it is harder for them to catch-up. We tested the predictions from the model and confirmed that liberalization affected asymmetrically the different types of firms. In our baseline results, a 10 percent reduction in tariffs spurred productivity growth between 4 and 8% on average. However, while for backward firms this effect is much weaker if not close to zero, different for more advanced ones this effect is stronger leading to a productivity growth between 11 and 13%. In addition to assess the key predictions from our model, we tried to identify the mechanisms driving our results. In fact, increases in labor productivity could just be driven by additional investments and input usage rather than innovative efforts to improve products and processes. We showed that even if investments, especially in technology, are partially driving our results, these cannot account for the entire story. Instead, consistent with the conjecture that our findings are driven by additional innovative and managerial efforts, we find that our results are significantly stronger in sectors where the scope for innovative activities is larger. These findings have various implications. In the context of the empirical debate on the relationship between trade liberalization and growth they suggest that we should not be surprised by the evidence that the effects of liberalization vary across countries or sectors because this could just be a consequence of the differences in the specific distribution of firm productivity before the reforms. Furthermore, for policy makers being aware of the possible heterogeneous effects of the reforms is potentially even more important than knowing about the average impact of the reforms in order to devise appropriate complementary policies and anticipate the political economy of the responses to the proposed reforms. Finally, these findings open the avenue to further research to better understand why identical policies may have asymmetric impact on heterogeneous firms. Appendix A. Supplementary data Supplementary data to this article can be found online at http:// dx.doi.org/10.1016/j.jdeveco.2012.06.001. References Aghion, P., Bessonova, E., 2006. On entry and growth: theory and evidence. Revue de l'OFCE, Special Number on Industrial Dynamics, Productivity and Growth, 259. Aghion, P., Griffith, R., 2005. Competition and Growth: Reconciling Theory and Evidence, Zeuthen Lecture Book Series. MIT Press. Aghion, P., Bloom, N., Blundell, R., Griffith, R., Howitt, P., 2002. Competition and innovation: an inverted U relationship. Working Paper 9269, NBER. Aghion, P., Blundell, R., Griffith, R., Howitt, P., Prantl, S., 2004. “Firm Entry, Innovation and Growth: Theory and Micro Evidence,” Mimeo, Department of Economics, Harvard University. Aghion, P., Burgess, R., Redding, S., Zilibotti, F., 2004. “Entry Liberalization and Inequality in Industrial Performance,” Mimeo, Department of Economics, Harvard University. Amiti, M., Konings, J., 2007. Trade liberalization, intermediate inputs and productivity. The American Economic Review 97 (5), 1611–1638. Arellano, M., 2003. Panel Data Econometrics. Oxford University Press. Aw, A.Y., Chen, X., Roberts, M.J., 2001. Firm-level evidence on productivity differentials and turnover in Taiwanese manufacturing. Journal of Development Economics 66 (1), 51–86. Bernard, A.B., Redding, S., Schott, P.K., 2004. Comparative advantage and heterogenous firms. Working Paper 10668, NBER. Bustos, P., 2011. Trade liberalization, exports and technology upgrading: evidence on the impact of MERCOSUR on Argentinean firms. The American Economic Review 101 (1), 304–340. Cameron, C., Trivedi, P., 2005. Microeconometrics: Methods and Applications. Cambridge University Press. Crespi, G., 2005. Productivity and Firm Heterogeneity in Chile.

485

De Hoyos, R., Iacovone, L., 2011. Impact of NAFTA on economic performance: a firmlevel analysis of the trade and productivity channels. Policy Research Working Paper Series 5661. The World Bank. Disney, R., Haskel, J., Heden, Y., 2003. Restructuring and productivity growth in UK manufacturing. The Economic Journal 113 (489), 666–694 ISSN: 0013‐0133 Available From: Publisher's URL Journal Article. Fernandes, A.M., 2007. Trade policy, trade volumes and plant-level productivity in Colombian manufacturing industries. Journal of International Economics 71 (1), 52–71. Gerosky, P., 1990. Innovation, technological opportunity and market structure. Oxford Economic Papers 42, 586–602. Goldberg, P.K., Pavcnik, N., 2005. Trade, wages, and the political economy of trade protection: evidence from the Colombian trade reforms. Journal of International Economics 66 (1), 75–105. Griffith, R., Redding, S., Simpson, H., 2003. Productivity convergence and foreign ownership at the establishment level. Discussion Paper dp0573, CEP, LSE. Hanson, G., 2004. What has happened to wages in Mexico since NAFTA? In: Estevadeordal, T., Rodrick, D., Taylor, A., Velasco, A. (Eds.), FTAA and Beyond: Prospects for Integration in the Americas. Harvard University Press, Cambridge. Hanson, G.H., Harrison, A., 1999. Trade liberalization and wage inequality in Mexico. Industrial & Labor Relations Review 52 (2), 271–288. Haskel, J., 1992. Imperfect competition, work practices and productivity growth. Oxford Bulletin of Economics and Statistics 53, 265–279. Head, K., Ries, J., 1999. Rationalization effects of tariff reductions. Journal of International Economics 47 (2), 295–320. Helpman, E., Melitz, M.J., Yeaple, S.R., 2004. Export versus FDI with heterogeneous firms. The American Economic Review 94 (1), 300–316. Hopenhayn, H., 1992. Entry, exit, and firm dynamics in long-run equilibrium. Econometrica 60 (5), 915–938. Iacovone, L. 2008. “Exploring Mexican Firm-Level Data”, Mimeo. Jovanovic, B., 1982. Selection and the evolution of industry. Econometrica 50 (3), 649–670. Kowalczyk, C., Davis, D., 1996. Tariff phase-outs: theory and evidence from GATT and NAFTA. NBER Working Papers 5421. National Bureau of Economic Research. Leibenstein, H., 1978. On the basic proposition of X-efficiency theory. The American Economic Review 68 (2), 328–332. Lileeva, A., Trefler, D., 2010. Improved access to foreign markets raises plant-level productivity … for some plants. Quarterly Journal of Economics 125 (3), 1051–1099. López-Córdova, E., 2003. NAFTA and manufacturing productivity in Mexico. Economia: Journal of the Latin American and Caribbean Economic Association 4 (1), 55–88. Martin, J.P., 1978. X-inefficiency, managerial effort and protection. Economica 45 (179), 273–286. Martin, J.P., Page, J.M., 1983. The impact of subsidies on X-efficiency in LDC industry: theory and an empirical test. The Review of Economics and Statistics 65 (4), 608–617. Melitz, M.J., 2003. The impact of trade on intra-industry reallocations and aggregate industry productivity. Econometrica 71 (6), 1695–1725. Melitz, M., Ottaviano, G., 2003. “Market Size, Trade and Productivity,” Mimeo, Department of Economics, Harvard University. Nickell, S.J., 1996. Competition and corporate performance. Journal of Political Economy 104 (4), 724–746. Pavcnik, N., 2002. Trade liberalization, exit and productivity improvements: evidence from Chilean plants. The Review of Economic Studies 69 (1), 245–276. Rajan, R.G., Zingales, L., 1998. Financial dependence and growth. The American Economic Review 88 (3), 559–586. Rauch, J.E., 1999. Networks versus markets in international trade. Journal of International Economics 48 (1), 7–35. Revenga, A., 1997. Employment and wage effects of trade liberalization: the case of Mexican manufacturing. Journal of Labor Economics 15 (3), S20–S43. Rodrik, D., 1988. Imperfect competition, scale economies and trade policy in developing countries. In: Baldwin, R. (Ed.), Trade Policy Issues and Empirical Analysis. University of Chicago Press and N.B.E.R. Sabirianova, K., Terrell, K., Svejnar, J., 2005. Distance to the efficiency frontier and foreign direct investment spillovers. Journal of the European Economic Association 3 (2–3), 576–586. Schor, A., 2004. Heterogenous productivity response to tariff reduction. Evidence from Brazilian manufacturing firms. Journal of Development Economics 75 (2), 373–396. Tomz, M., 1997. Do International Agreements Make Reforms More Credible? The Impact of NAFTA on Mexican Stock Prices,” Presented at 1997 Annual Meeting of the American Political Science Association. Topalova, P., 2004. Trade liberalization and firm productivity: the case of India. IMF Working Papers 04/28. International Monetary Fund. Tornell, A., Esquivel, G., 1995. The political economy of Mexico's entry to NAFTA. NBER Working Papers 5322. National Bureau of Economic Research, Inc. available at http://ideas.repec.org/p/nbr/nberwo/5322.html. Trefler, D., 2004. The long and short of the Canada–U. S. Free Trade Agreement. The American Economic Review 94 (4), 870–895. Tybout, J.R., 2000. Manufacturing firms in developing countries: how well do they do, and why? Journal of Economic Literature 38 (1), 11–44. Tybout, J.R., Westbrook, M.D., 1995. Trade liberalization and the dimensions of efficiency change in Mexican manufacturing industries. Journal of International Economics 39 (1–2), 53–78. Verhoogen, E.A., 2008. Trade, quality upgrading, and wage inequality in the Mexican manufacturing sector. Quarterly Journal of Economics 123 (2), 489–530. Yeaple, s. r, 2005. A simple model of firm heterogeneity, international trade, and wages. Journal of International Economics 65 (1), 1–20.