The (un)lucky neighbour: di erences in export performance across Mexico's states∗†

This paper studies the reasons behind the export performance of di erent Mexican states from 1994 to 2002. Mexican exports are divided into two components: (1) foreign market potential; and (2) supply capacity. Results suggest that states that experience an increase in supply capacity are, in most cases, also those with better export expansion. However, results suggest that in most cases, export growth is due mainly to an increase in US demand rather than to an improvement in the states' competitiveness to supply this demand. When looking to the determinants of supply capacity improvements, results suggest the presence of positive neighbouring e ects. In other words, part of a state's improvement in the export sector is thanks to its neighbours.


Introduction
In the mid 1980s Mexico shifted its growth strategy from a policy of industrialization through import substitution to an export-led growth strategy. Trade liberalization policies attempted to encourage investments in the production of tradable goods to exploit Mexico's potential as an export platform to  Blecker (2006) in addition to the devaluation of the peso, other circumstances played an important role in Mexico's export performance from 1996-2000. The rst is the fact that NAFTA, apart from opening the US market to Mexican products, guarantees foreign investors' property rights in Mexico. A second circumstance is the economic boom of the US economy in the Clinton era. However, Mexican exports entered into a recession in 2001, partly due to the Mexican authorities' encouragement of a high peso as a means of controlling ination [Blecker (2006)], and partly due to a slowing down in the US growth rate, pointing to a high elasticity of exports in US income, as suggested by Pacheco-López and Thirlwall (2004). As Figure 1 illustrates, most Mexican exports go to the US market, and its growth rate seems to depend on US demand.
Since the Mexican outward oriented strategy was mainly focused on serving the US market, the reference market for many rms changed with liberalization reforms. From a theoretical model inspired by the Mexican case, Krugman and Elizondo (1996) suggest that as the economy opens, the importance of Federal District as an industrial centre decreases. In other words, liberalization policies weaken the * This is a postprint version of an article whose nal and denitive version has been published in Papers in Regional Science on 3 Novembre 2010 (http://www.blackwell-synergy.com/doi/abs/10.1111/j.1435-5957.2010.00291.x). The denitive version is available at wileyonlinelibrary.com † I would like to thank Fernando Borraz for providing useful data used on this paper. I am also grateful for comments from two anonymous referees that substantially improved the paper. forward and backward linkages formed by protection policies, and a dispersion of economic activities takes place. These theoretical ndings were supported by Hanson (1998) who studies the eects of opening-up on Mexico's regional economies. Analyzing the eects on regional labour demand, he nds evidence that the pattern of industrial concentration around the capital is changing, and new industrial centres have been created in the northern states to prot from their access to the US market.
Given these facts, the purpose of this paper is to study an important issue regarding Mexico's external sector that has not yet received much attention: the origin of disparities in export performance among Mexican states. Indeed, as Figure 2 illustrates, there are great dierences in the export share between Mexican states. Most "exporter" states are located in the north of the country. However, as suggested by Krugman and Elizondo (1996), a dispersion of the export sector activity seems to be taking place.
Comparing shares in 1993 and 2000/2002, the importance of the metropolitan region states (the Federal District and the State of Mexico) and Chihuahua as exporters has decreased. To determine the factors behind Mexican states' improvements in export performance, a theoretical model developed by Redding and Venables (2004a) based on a monopolistic competition framework is employed. This model allows for the decomposition of countries' exports into two components. The rst relates to shifts in the demand for countries' tradable goods, and the second to changes in countries' supply capacity. This paper is organized as follows. In section 2, Redding and Venables (2004a) partial equilibrium model is adapted to the Mexican case to develop an estimation framework. Assuming that all Mexican exports go to the US market, exports by Mexican state are presented as a function of their proximity to the US market and of the capacity to supply this market. Section 3 presents estimation results. Mexican states export performances are decomposed into foreign market potential and supply capacity, and dierences across Mexico's states are analyzed. Results suggest that most of the export growth can be explained by shifts in the demand side, measured as market potential. However, states having the better export performances are those that experienced the greatest increases in supply capacity. Section 4 analyses supply capacity in depth by studying its sources. A theoretical extension to the model is presented that takes into account the general equilibrium process. Then, spatial econometric techniques are introduced to control for spatial linkages and the extended model is estimated. Results suggest that supply capacity is inuenced by the economic size and the road infrastructure. Most importantly, results suggest the existence of spatial spillovers that play a positive role in determining supply capacity. Finally, section 5 presents some concluding remarks.

Theory
The previous section points out two main facts: (1) Mexican export dependency on the US economy, and (2) dierences in terms of international trade levels across states, mainly due to dierences in their access to the US market. The Dixit-Stiglitz-Krugman model of monopolistic competition and trade in a multi-region setting allows for the decomposition of exports into two components: the rst related to market potential, and the second to supply capacity. This theoretical model is employed to analyze the disparities in export performance among Mexican states. The main assumption is that an increase in the expenditure of tradable goods in US should raise the demand of Mexican tradable goods. However, states would not prot from this increase in Mexican goods demand in the same way, due to dierences in trade costs. Thus, the objective is to study which states have reacted better to shifts in the demand side.

US demand
It is assumed that US consumers have identical preferences. In addition, their utility is given by the consumption of both tradable (C T us ) and non tradable (C N T us ) goods. This utility function is then represented by a Cobb-Douglas function: We assume that the demand of tradable goods is the aggregate of the demand of tradable goods produced in region i = 1, . . . R. Moreover, region-i's rms produce under increasing returns to scale a range of symmetric dierentiated goods to provide the US demand. US consumption of tradable goods by sector is represented as standard CES setup: where n i is the set of varieties produced by region-i; x i,us is the US consumption of a single product variety produced in region-i; and σ is the elasticity of substitution across varieties.
In this framework, the US demand for each variety produced in region-i is a function of the US total expenditure on tradable goods (E us ), product price (p i,us ) and a price index given by all tradable goods sold in the US (P us ). Demand for an individual variety can then be written as: where P us can be formulated as a generalized mean of the number of rms that supply the US market, as well as a function of prices of individual varieties produced by each region: Following the literature, trade costs take the iceberg form. It is assumed that region-i's varieties by sector have the same producer price (p i ), and the transportation costs (τ ) introduce a spread between the domestic and foreign price of any variety produced in region-i. The value of region-i's exports to the US is therefore given by: Finally, we can dene the term M i = φ i,us m us as region-i foreign market access or real foreign market potential (Head and Mayer, 2004;Redding and Venables, 2004a). Then export performance depends on the supply capacity and on the real foreign market potential: 3 Export performance According to our theoretical model, two elements are necessary to determine the export performance of Following the literature, bilateral distance and common border are employed as a proxy for obtaining transport costs. More precisely, transportation costs are computed as φ ij = distance β ij ·exp( γBorder ij ). Supply capacity can be computed from state-i xed eects, s i = exp(F E i ). Finally, the US market capacity is obtained from the US importer dummy m us = exp(F E us ). Once transportation costs and US market capacity are estimated, state-i real US market potential is obtained from M i = φ i,us m us .

3.1
Data sources and estimation strategy Regression results are presented in Table 1. As expected, the coecient on distance is negative, which indicates that the further away state-i is from the US, the less it exports. We can also see that the distance coecient value initially decreases and then increases between the second and third periods, but these variations are not important. Distance coecient value is near to −1, which is a frequent nd among gravity equations literature [Disdier and Head (2003)   We have seen that market capacity can be computed from the importer xed eects as: m j = exp(Imp j ). Table 1 illustrates the xed eect evolution of Mexico's NAFTA partners as well as of the other two major world markets: Japan and Germany. We can see the big dierence in terms of market capacity between the US and the other countries. In addition, US market capacity increases by 28% between the rst and second periods, moreover it experiences an increase of more than 40% between the second and third periods.
We have seen that results look rather stable across the three time periods. In order to look for dierences across time periods, a set of tests for the equality of the coecients is computed and some of them are reported in Table 2. We can see that the null of the equality between regressions are rejected, which, according to our model, suggest that at least one of supply capacity, market capacity and trade

Exports decomposition
Gravity equation allows us to compute the US market capacity the trade costs and the Mexican states supply capacity. Once we have these elements, we can proceed to compute the US market potential for each state, and to analyze their export performance. To compute the foreign market potential for each state we use the following equation: M i = φ i,us m us . Once the foreign market potential is computed, export performance can be decomposed into supply capacity growth and foreign market potential growth.
Results are presented in Table 3, where states are sorted from best to worst export performance between periods 1 and 3. From the average performance (last row of Table 3), we can see that exports grow considerably between periods 1 and 2, but slowed considerably between periods 2 and 3. This change in growth is explained mainly by a slowdown in supply capacity. Indeed, between periods 1 and 2, the average supply capacity grew by 55.4%, while it decreased by 11.6% between periods 2 and 3.
Between the initial and nal periods of study, average exports grew by about 103%, but most of this boom can be explained by an signicant increase in foreign market potential. Moreover, supply capacity decreased for one third of the Mexican states between the initial and nal periods by about 30%. We can also see that the foreign market potential increases by 96% between the initial and the last period.
This result is two times bigger than the one reported by Redding Table 3, which highlights the weight of the US market over the overall foreign market potential for Mexico.
Looking at individual states, we can see that states with the highest levels of exports did not experience the highest average growth in exports. As Table 3 illustrates, the northern border state that experienced the best performance in exports between periods 1 and 3 is Coahuila, ranked sixth in terms of export growth. Jalisco has the best export improvement; indeed, this state increased its exports by over 500%, explained by a considerable improvement in supply capacity that places it as the states with the highest supply capacity.
At the opposite extreme, we can nd the poorest Mexican states: Oaxaca and Chiapas. Between periods 1 and 3, these states experienced a export reduction despite of the important increase in foreign market potential. It is important to point out that between the second and third periods, twenty three states (72%) showed a decrease in their supply capacity, among them the country's capital and four of six northern border states. Table 3 shows that dierences across Mexican states in export performances are mainly driven by dierences in supply capacity improvements. This opens the way for the next section which explores the determinants of supply capacity.
Results are also summarized in gures 3 and 4, which draws the spatial distribution of export performance and supply capacity improvements between periods 1 and 3. Concerning export growth, Figure   3 shows that states having a weaker export performance are usually neighbors of at least another state also experiencing a poor export performance. On the other hand, states achieving the strongest levels of export improvements share a border with at least one state with important export growth. However, in both cases, the states can be located at the north or at the south of the country. Figure 4 shows the improvements in supply capacity. We can see that this map is very similar to that of export growth.
This opens the way for the next section which explores the determinants of supply capacity. In other words, both gures show that states having the worst export performance and the lowest supply capacity growth are usually neighbors of another state with similar a performance. We can see that in most cases the opposite is also true. Thus, we suspect a neighbor's eect on a state's performance. 4 The determinants of supply capacity The main objective of this section is to clarify the main components of supply capacity. As indicated in the theoretical framework presented previously, supply capacity is a function of the number of varieties produced and of production prices. It is then expected that variables measuring economic size and factor endowments are related to supply capacity. However, as pointed out by Redding and Venables (2004a), supply capacity could also be dependent on real market potential. Indeed, an increase in foreign market potential increases the potential returns of export activities; thus an expansion of the export sector is expected, having some impact on supply capacity. To take this endogeneity into account, Redding and Venables (2004a) applied a simple general equilibrium feature that captures the opportunity costs of resources used in the export sector, summarized by the following equation:  Table notes: export growth is the growth of average exports by state between two time periods. Supply capacity and US market access growth are computed from results reported in Table 1. Mexican states sorted from high to low exports growth between period 1-3. cap superscript refers to Mexican capital and most important economic center. nb superscript refers to Northern border states.
where a i is a measure of the size of the economy; c i is a measure of comparative costs in the export sector of region-i; ω is the price elasticity of export supply; t i is the total expenditure on internal transport costs; and · denotes a proportional deviation from a reference point. It is important to point out that a less than proportional relationship between foreign market potential and supply capacity is expected.
Thus, the coecient market potential is expected to be less than one: σ > (1 + ω). Indeed, an increase in export volume requires an increase in production resources which would be drawn from non export sectors. This production factor demand increase in turn increases prices and, consequently, deteriorates supply capacity.
Other expected results are the following ones: an increase in internal trade costs have a negative impact on export performances given that σ > 1; domestic size increases exports volume; and high costs means that a lower volume of exports is supplied for a given price (Redding and Venables, 2004a).

Empirical specication and data sources
The sample subject to empirical specication is composed of a panel of annual data of the 32 Mexican states for 1994-2002. The empirical specication of equation 9 takes the following form: where the dependent variable is the log of total exports of state i = 1, ..., N for a year t = 1, ..., T . X i (t) To measure dierences in production factor costs, the average wage of the state is employed. An increase in wages is suppose to deteriorate supply capacity, but average wages can also be an indicator of skilled-labor endowments which can improve supply capacity.
At the end of the 90s, Mexican government attempted to reform the electric sector. According to President Zedillo's Mexico faced a growing demand for electrical power and a chronic insuciency of public resources to make the necessary investments to satisfy this demand (Breceda, 2000). Indeed,  A most important econometric issue is the inuence of neighboring states on export performances. More precisely, since the price index is a function of imports from all regions, the exports from one region would be inuenced by other regions exports through this index. These authors suggest that the coecient of the spatial lag dependent variable can be interpreted as a measure of spatial competition.
LeSage and Pace (2008) provide two econometrical motivations to include a spatial lag of the dependent variable in the study of trade ows. The rst one consist of considering spatial dependence as a long-run equilibrium of a dynamic spatiotemporal process. In other words, a shock in a region is transmitted directly to its neighbors, it is transmitted indirectly to its neighbors neighbors, as well as to its neighbors neighbors neighbors, and so on; moreover, shocks get reected back to the region from its neighbors. As a consequence, in the equilibrium, the dependent variable in any observation depends on the exogenous variables and error terms in all other observations. The second motivation is founded on an omitted variables argument. These authors show that omitting variables that exhibit a spatial dependence can lead to any: spatial lag model (if the omitted variable is correlated to any exogenous regressor) or spatial error model (if the omitted variable is correlated to the error term).
According to equation 9, factors are immobile across regions and resources are drawn out of non export sectors. However, in the case of regions inside the same country where labour can be a mobile factor, an export sector expansion can draw production resources from other regions via labour migration 2 . In addition, improvements in variables such as the road network can lead to improvements in export performances to neighboring regions using this road network to export. Thus, observed and unobserved neighboring characteristics might inuence the export performance of a Mexican state-i. Following LeSage and Pace (2008) advises, it should be convenient to include a spatial lag of the dependent variable.
However, in the presence of spatially dependent omitted variables correlated only with the error terms, the error terms will exhibit some form of cross-sectional correlation that has to be dealt with (Koch et al., 2007). In our case, Mexican states can face to similar unobserved shocks that lead to a cross-sectional correlation of this kind. For example, a slowdown in the demand of the automobile industry would impact states exporting this industry related products. To handle this cross-sectional correlation, we assume that the disturbance term follows a rst order spatial autoregressive process. Including both, the spatial autoregressive disturbances and the spatial lag dependent model in the model yields to a mixed-regressive-spatial autoregressive model with spatial autoregressive disturbance, commonly called the General Spatial Model (SAC) 3 : ε . This assumption implies that a shock in state-i would be transmitted to all other states aecting their export performance. LeSage and Pace (2008) propose to take into account the spatial dependences of both the origin and the destination regions. In our case, given that a common destination for all exports is assumed, spatial dependences are considered only between the exporter regions. The weight matrices W i = M i are constructed as follows. For any year t = 1, ..., T , a weight matrix W i is dened as: 2 Note that if this is the case, the coecient value of M P i may be higher than one. 3 Kelejian and Prucha (1998) refer to the SAC model as the (rst-order) spatial autoregressive model with (rst-order) autoregressive disturbances of order (1, 1), for short SARAR(1, 1). where w i,j = (1/distance i,j ) if the states i and j are neighbors, and zero otherwise. Matrix W t is standardized, so that the sum of every row of the matrix is one.
An additional econometric problem comes from the construction of the foreign market potential.
This variable was calculated for three year periods, while this section employs yearly data. In addition, rm heterogeneity literature suggests that the heterogeneity in productivity of existing rms plays an important role in the export performance. Chaney (2008) extends Melitz (2003) model to a multicountry framework. He shows that the number of rms exporting, as well as its export levels, depend on rms' productivity and on the elasticity of substitution. When trade barrier decreases, a region hosting competitive rms captures a larger market share than a region hosting less productive rms. The rms already implanted in a region inuence both present and future export performance of the region. These points suggest that there are characteristics that are transmitted over time. To take this into account, it is allowed for the innovations to be correlated over time. Following Kapoor et al. (2007), the error structure is dened as: (14) where µ i represents the N ×1 vector of state-i specic error components with zero mean and variance σ 2 µ ; and υ i (t) in an N × 1 of error components with zero mean and variance σ 2 υ that vary over both t periods and i states. The innovations vector ε i (t) thus depends on disturbances which dier for each state but are constant for the same state over dierent years (µ i ), and of disturbances that are specic to each state and to each time period (υ i (t)). In other words, the specication of the innovations corresponds to that of a classical one-way error component model (Kapoor et al., 2007). It is assumed that the processes µ i and υ i (t) are independent. Thus, ε i (t) are autocorrelated over time, but are not correlated across units: .., µ (T )] ; I T , I N and I N T are identity matrices of dimension T , N and N T , respectively; e T is a vector of ones of dimension T ; and ⊗ denotes the Kronecker product. Note that this specication allows for spatial interactions in both the error components υ i (t) and the specic error µ i components. Finally, the variance-covariance matrix of the disturbance vector ε can be represented as: The estimation approach employed in this paper is a feasible generalized spatial two stages least squares (FG2SLS) procedure developed by Kapoor et al. (2007) and Kelejian and Prucha (1998). This approach involves three steps. In the rst step, the regression model in equation 15 is estimated by two stage least squares (2SLS) to obtain the residuals ( u = X − Z β). In this step, following Kelejian and Prucha (1998), the exogenous regressors and their rst and second order spatial lags (z i (t), W i z i (t), W 2 i z i (t)) are employed as instruments for the spatial lag of the dependent variable. In the second step, the residuals are used in the generalized moments procedure suggested by Kapoor et al. (2007) to estimate the spatial autoregressive parameter ρ, σ 2 µ , σ 2 υ and Ω ε by a non-linear optimization routine. Finally, in the third 4 For more details on the properties of the matrices and assumptions see Kapoor et al. (2007). step, a spatial Cochrane-Orcutt type transformation of the model is applied to obtain spatially independent disturbances, and this model is reestimated in terms of a feasible generalized spatial least squares estimator. More specically, multiplying of equations 15 and 16 with (I N T − ρI T ⊗ W ) yields: X * = Z * δ + ε (19) where X * = [I T ⊗ (I N − ρW )]X; and Z * = [I T ⊗ (I N − ρW )]Z. Then the FG2SLS estimator is dened as: 5.1 Estimation results Table 5 presents parameter estimates of the model employing OLS which will be used as a benchmark.
Note that, to capture Mexican economy volatility (especially during the nancial crisis years), time dummies were included to the specication, but not reported. OLS estimates suggest that the size and the foreign market potential are important factors inuencing export performances, since both variables are always signicantly positively related to the dierent specications of the dependent variable. Results suggest that road network is an important factor, since it is signicant for the dierent specications.
When the dependent variable is the total exports (columns 1-2), time lagged GDP, market potential and infrastructure are signicantly positively correlated to exports. More importantly, when adding regionally xed eects to control for spatial heterogeneity, they remain signicant; see column 2 in comparison to column 1. This is not the case for the agglomeration and electricity costs variables which lose their signicativity when adding region dummies. When employing domestic absorption as a measure of economic size (columns 3-4), this variable is signicant and positively correlated to exports, as well as market potential and road network. Agglomeration variable lost again its signicativity when adding region dummies. In the third specication (columns 5-6), the market potential and the road network coecients are again positive and signicant. Given that the only measure of size is the population levels, this variable becomes signicant and positively correlated to exports. In this specication, wages become signicant and positively correlated to exports/GDP ratio. This result is not robust compared to those found in previous specications, which suggest that wages coecient could be inuenced by omitted variables bias. Table 5 also reports the Lagrange multiplier (LM) and its robust versions (RLM) to see which specication applies better to our case, spatial-error or spatial-lag dependences. We can see that we cannot reject the presence of both spatial lag dependent variable and spatial error inuences. The next step is then to control for these types of spatial autocorrelation by computing the FG2SLS estimates.
Results (reported in Table 6) suggest that state's size, road infrastructure and foreign market potential are the main determinants of export performance. The size of the state is again signicant when measured as economic size. Moreover, when omitting economic size variables, the population size reects the signicance of the size of the states. Concerning spatial dependences, results suggest the presence of these dependences across neighboring states. The spatial lag of the dependent variable is always positive and signicant, which suggest the agglomeration of export activity into a group of states. This result can also be interpreted as the result of a positive inuence of variables aecting export performance in neighboring regions such as neighbors size and road networks.
The estimate of the autoregressive parameter ρ is always signicantly related to the dependent variables, but its sign changes when including regional dummies. We can then infer that this coecient captures the eects of omitted variables that create both positive and negative spatial spillovers across Mexican states. Hence, controlling for spatial error dependence allows us to reduce the bias created by the omission of important variables. This can be seen clearly if we compare the coecients of the wages variable reported in column 6 in both Table 5 and Table 6. Indeed, introducing the error variable spatial wage loses its signicance.
The results reported in Table 6 are in line with the theoretical model and can be interpreted as follows.
Road density facilitates transportation, thus reducing trade costs improves the export performance. In addition, as suggested by the theory, the expansion of the export sector needs production factors. Thus, the availability of a signicant quantity of production factors, measured by the size of the state, would improve the export sector. In addition, improvements in the export sector would be transmitted to neighboring states, creating regions that concentrate the export activity. Finally, after controlling for supply capacity, the foreign market potential is a key element in improving exports. Moreover, the FMP  estimator is lower than one, which suggests that an increase in the demand side raises costs and prices in the export sector, discouraging supply capacity as proposed by Redding and Venables (2004a).
Given that the FMP variable was constructed from regressions involving export data, one could doubt the pertinence of employing this variable. This variable is replaced, therefore, by two alternative measures of market potential. From gravity equation we have found a distance estimator around minus one. The rst alternative measure is then the ratio of real US GDP to distance between US and state-i; the second is the ratio of total US imports minus state-i exports to distance between the US and state-i. Table 7 draws the estimates when employing alternative measures of market potential. Comparing results to those found previously, we can see that there are no signicant dierences on the signicance and coecient values for the market potential, road network and size variables. Indeed, these remain the most important factors explaining export performances. Population size is also signicant excluding when the GDP and the absorption variables are excluded. Regarding the coecient value of the three market potential variables, in all cases are lower than one as expected. Moreover, except for the population variables, after controlling for spatial dependences 5 , their coecient value become smaller. Hence, not controlling for these dependences could lead to overstate the importance of signicant variables, and even to give signicance to variables that are not (as the variable wages in our case 6 ).
Finally, note that, except for production costs which are not signicant, results are in line with equation 9. In other words, states' size, improvements in internal trade costs, and foreign market potential are the main determinants for export supply capacity.

Conclusions
This paper studies the export performance of Mexican states after liberalization reforms. The paper focuses on the role played by the transport costs determining access to foreign markets, as well as the states' competitiveness in supplying foreign market demand. Given the Mexico's location, its main foreign market in recent years has been the US market. Thus, the theoretical assumption is that Mexican exports vary in function of shifts in the US demand for tradable goods. The paper also studies the response of the dierent states to these changes in demand in order to determine which states prot better from liberalization reforms. 5 The results without spatial lag variable and spatial error not reported for the alternative measures of FMP, but available under request. 6 The wages variable has a similar behavior when using alternative measures of FMP. Indeed, this variable is signicantly positively related to the exports/GDP variable until controlling for spatial dependences.
Results also highlight the importance of states' market potential in determining export levels. Northern border regions have almost ve times more market potential than southern regions. However, when analyzing the changes in market potential and their impact on export performance, northern border states fare no better. Furthermore, results suggest that the states that have the better export performance are not necessarily those with the higher increase in market potential. An important factor in export performance is the supply capacity improvement of each state. States that experience an increase in supply capacity are, in most cases, also the states with better export performance.
Among the factors inuencing supply capacity, results indicate that economic size, road infrastructure and spatial spillovers from neighboring states' are signicant factors. Since export sector expansion needs additional resources, better endowed states react better to this expansion. Improvements in infrastructure impact directly the trade costs since it makes easier the distribution of goods and supplies. Concerning spatial dependences, result suggest the presence of positive spatial spillovers from neighboring states' export activity. In other words, changes in a state's export volume or in state's characteristics inuencing export activity (as size and road network) will indirectly have a positive impact on the supply capacity in neighboring states. According to obtained results, there are also spatial spillovers through the error term. This result implies that state level data cannot be considered as independently generated because of the presence of similarities with neighboring states. Moreover, taking into account this form of dependence allows to reduce the omitted bias variables. Results also suggest that dierences in production costs, measured by the wages and electricity prices, as well as their evolution have not aected the competitiveness of export activities.
Unfortunately for the Mexican economy, results suggest that in most cases export improvement is due mainly to an increase in the US demand for tradable goods rather than to an improvement in supply capacity. It seems, then, that an export-led growth strategy has reinforced Mexico's dependency on the US economy. As a consequence, Mexico's exports have become vulnerable to slowdowns in the United States. Thus, citing Blecker (2006), ". . . given the likelihood that the United States will have slower growth over the next several years as a result of its mammoth scal and trade decits, it is all the more important for Mexico to redouble its eorts at export diversication (perhaps more through marketing eorts rather than trade agreements)." (Blecker, 2006: pp.33-34). In addition, it is clear that there are some regions for which the US demand (measured as the market potential) is not as important as for other regions. Hence, in order to reduce disparities across regions, Mexican economic policy must also look to stimulating national demand (especially in the poorest states), rather than expecting improvements in a foreign demand that only favours a group of states.