77494 Foreign Direct Investment in Mexico since the Approval of NAFTA Alfredo Cuevas, Miguel Messmacher, and Alejandro Werner Cross-country panel data are used to assess the effect of free-trade agreements on flows of foreign direct investment (FDI). Free-trade agreements are found to have a significant positive effect on FDI flows, and free-trade agreements are found to matter more for the smaller members of the agreement. For example, the North American Free-Trade Agreement’s (NAFTA) effect on FDI flows into Mexico is much larger than its effect on flows into the United States. These cross-country results are used to assess NAFTA’s effect on FDI flows into Mexico. After controlling for a set of other factors—such as an increase in worldwide FDI flows—the trade agreement is found to generate FDI flows nearly 60 percent higher than they would have been without the agreement. The composition of the capital account of Mexico’s balance of payments changed significantly at the same time that the North American Free-Trade Agreement (NAFTA) came into effect in 1994. In particular, there is a well- documented increase in the importance of foreign direct investment (FDI). This contrasts sharply with the dominant role of portfolio flows and external bor- rowing by the banking sector during the early 1990s (figure 1). In principle, the effect of a free-trade agreement on FDI need not be positive. In the simplest Heckscher–Ohlin world, when free trade achieves factor-price equalization, capital has no incentive to cross borders. Under this logic, a free- trade agreement could reduce the incentives for FDI if the original purpose of FDI was to bypass trade barriers to supply the protected domestic market. By contrast, when countries’ factor endowments are sufficiently unequal, capital has incentives to relocate to the labor-intensive country, and these incentives are strengthened when the flow of goods between countries becomes unimpeded. Moreover, there could be a diversion effect, as third-country corporations may desire to invest in one of the trade agreement members to take advantage of the lower tariffs imposed by the partners in the agreement. In addition, NAFTA contains specific provisions for liberalizing international investment in the region. The most important are national treatment and most- favored-nation privileges for any investor residing in North America—both of which should promote FDI. In fact, many observers argued that promoting the THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 3, pp. 473–488 doi:10.1093/wber/lhi015 Advance Access publication December 6, 2005 Ó The Author 2005. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org. 473 474 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 3 F I G U R E 1 . Main Components of Mexico’s Private Capital Flows 30,000 25,000 Commercial banks Loans to nonban k agent s Foreign direct investment Investment in the stock exchange 20,000 U.S. dollars (millions) 15,000 10,000 5,000 0 -5,000 -10,000 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Source: Bank of Mexico (various years). growth of foreign capital flows was Mexico’s main purpose in joining the agree- ment: it was not a matter of further liberalizing trade but of creating a legal and economic environment conducive to FDI. This article attempts to test that idea. The objective here is to assess whether the observed change in financing patterns, and concretely the increase in FDI, can be explained to a significant extent by NAFTA. The analysis controls for general changes in the patterns of international capital flows, because global FDI surged during the 1990s. Mexico also faced other domestic factors that are harder to control for in the analysis. It undertook substantial structural reforms during the late 1980s and early 1990s,1 and shortly after NAFTA went into effect it suffered a banking crisis, which might have left firms with no other option than to obtain foreign financing. The strategy here is to identify the typical behavior of FDI in a sample of countries and then to apply that model of behavior to Mexican data. Statistical analysis of an international data set is used to disentangle the effects of various determinants of FDI and to assess whether free-trade agreements generally promote 2 FDI. The results suggest that FDI responds positively to free-trade agreements and 1. These reforms include Mexico’s entry into the General Agreement on Tariffs and Trade (GATT) in 1986, the restructuring of Mexican sovereign debt in 1990, the privatization program, and the financial liberalization process that took place during the late 1980s and early 1990s. See Morley, Machado, and Pettinato (1999). 2. This analysis is related to the literature on gravity models, though with some important differences. Cuevas, Messmacher, and Werner 475 that the effect can be substantial. When these results are applied to Mexico, NAFTA appears to be responsible for increasing FDI by about 70 percent. However, the increase in FDI during the second half of the 1990s seemed disappointing relative to what the model predicts. Tentative explanations for this disappointing perfor- mance are offered, but the issue remains part of the agenda for future research. I. THE EFFECT OF FREE-TRADE AGREEMENTS ON FDI FLOWS This section reviews theoretical arguments about the effect of free-trade agree- ments on FDI flows and previous empirical studies that try to estimate the determinants of FDI. FDI and Free Trade: Theory Levy-Yeyati, Stein, and Daude (2001) offer a useful taxonomy of theoretical links between changes in FDI and the freeing of external trade, based on a distinction between ‘‘vertical’’ and ‘‘horizontal’’ FDI. With vertical FDI, multinational corpora- tions establish the stages of production in different countries to take advantage of specific conditions in local factor markets. This is the typical case of ‘‘outsour- cing,’’ with central management functions at the corporate headquarters and basic manufacturing in countries that offer cheap labor. This type of operation pre- sumes reasonably unimpeded trade, because the manufacturing facilities must also be able to supply the corporation’s home market. With horizontal FDI, by contrast, the multinational firm establishes similar production facilities in several countries, each one serving the local market. This is a rational strategy when obstacles to trade (natural or artificial) are significant. Horizontal FDI does not require any specific degree of similarity or difference in factor endowments among the countries, as long as production becomes feasible and economically viable after investment has taken place. Under this typology, liberalizing trade will make horizontal FDI unnecessary while facilitating vertical FDI. These arguments suggest that NAFTA must have promoted FDI for Mexico. In some cases, NAFTA may have led to some reduction in horizontal FDI, to the extent that planned or ongoing North American investment in Mexico was motivated by a desire to circumvent old trade restrictions. However, the vertical FDI argument points in the opposite direction: Mexico’s cost advantages, mainly from its low wages, may have been muted by trade barriers before NAFTA but became more obvious to investors after trade in goods was liberalized.3 Thus, 3. Krugman and Hanson (1993) explain this effect with the following example. Imagine a product with such economies of scale that it ought to be produced in only one location. Suppose that unit production and transport costs are $9 if production is located in Mexico and $10 if in the United States and that both countries apply a tariff of $1.50 per unit. If the bulk of the good is consumed in the United States, locating production there would achieve the lowest overall cost by avoiding the tariff that would fall on units imported from Mexico. But when both tariffs are removed, cost differences make production in Mexico the optimal choice. 476 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 3 vertical FDI should increase. Clearly, the difference in size between Mexico and its NAFTA partners strongly suggests that any reduction in horizontal FDI would be offset many times over by the increase in vertical FDI. The effects of trade liberalization on FDI should have been enhanced by the stability brought on by the high level of the agreement. The U.S. Congressional Budget Office (1993, p. 29) noted that as a result of NAFTA labor-intensive Mexican producers would ‘‘face lower tariffs, fewer quotas, and greater certainty that the United States would not suddenly put up protectionist barriers if they expanded exports.’’ Stability in the region’s policy toward Mexican exports was expected to make Mexico more attractive for investment in export-oriented activities. Simi- larly, policy stability in Mexico was expected to be a key benefit of NAFTA. The locking in of structural reforms begun with membership in the General Agreement on Tariffs and Trade in 1986 was widely considered a key promise of the agree- ment (Kehoe and Kehoe 1994). By strengthening the legal status of the Mexican government’s commitment to open markets, NAFTA was expected to consolidate an investment-friendly policy environment, which would ultimately boost FDI.4 In addition, the investment chapters in the agreement were expected to facilitate the decision to establish production facilities in Mexico by firms headquartered in Canada and the United States. According to Serra Puche (1992), the key principles incorporated in the free-trade agreement include the most-favored-nation principle, which ensured that no investor from outside North America would be granted benefits exceeding those available to North American investors; the national treatment principle, which guaranteed no discrimination among investors from each of the three NAFTA members; the absence of trade-related performance requirements for foreign investors; and the freedom to buy foreign exchange and to transfer funds across countries. Moreover, the Foreign Investment Law was liberalized to bring it into accord with NAFTA (Dussell Peters, Galindo Paliza, and Lorı ´a 2002). Finally, when NAFTA was launched, investors outside North America were expected to see Mexico as a more promising export platform to the United States and thus to locate some of their production facilities in Mexico. This is just an extension of the well-known tariff-jumping argument. In this variation investment locates in low-wage Mexico to ‘‘jump’’ the U.S. tariff. Given the general character of federal legislation, some of the reforms of the Foreign Investment Law induced by NAFTA would also be beneficial to investors from other regions. 4. NAFTA contains a provision that it can be canceled by any member with six months’ notice. Given the increase in trade and financial integration among the three members, it seems unlikely that the provision will ever be invoked. That said, if investors feel that reforms are still reversible, they may be more cautious than if they perceived the free trade agreement to be truly permanent. In a broad statistical study of FDI, the inability to measure this intensity of belief may lead to the underestimation of strong free trade agreements’ general effect on FDI, and in a particular country where policies may seem reversible to the overprediction of FDI. Cuevas, Messmacher, and Werner 477 Existing Empirical Research Recent cross-country panel models of FDI have been estimated by, among others, Levy-Yeyati, Stein, and Daude (2001) and Albuquerque, Loayza, and Serven (2003). Waldkirch (2003) uses panel data by source country of FDI stocks in Mexico to estimate NAFTA’s effect. His approach and results are discussed in detail in the next section. Levy-Yeyati, Stein, and Daude (2001) fitted a gravity model to bilateral FDI flows data. In their basic model, the stock of direct investment by residents of country A in country B depends on each country’s GDP, geo- graphic distance, and economic and cultural proximity, represented by dummy variables for common membership in a free-trade agreement and common language. In addition, the model includes two variables that mea- sure extended market effects. The first, expected to make investment in country B more attractive, is the total GDP of all countries that have free- trade agreements with country B. The second is the total GDP of countries that have free-trade agreements with country A. This last variable, therefore, represents the menu of options open to investors residing in country A, and its growth is expected to divert investment away from country B. The model’s results support the idea that regional integration and membership in a free-trade agreement promote bilateral FDI. Albuquerque, Loayza, and Serven (2003) differentiate between global and local drivers of investment. Global drivers simultaneously affect decisions concerning investment in many receiving countries. These factors include, for example, global financing conditions, represented in the empirical ana- lysis by a rich structure of financial indicators. Local factors, by contrast, affect investment only in specific receiving countries and are exemplified by expropriation risk, local macroeconomic stability, and other idiosyncratic variables. II. EMPIRICAL ANALYSIS OF THE EFFECT OF FREE-TRADE AGREEMENTS ON FDI This section uses regression techniques to identify the effect of membership in a free-trade agreement on FDI and then uses those estimates to gauge the relative contribution of regional integration, globalization, and other factors to the evolution of FDI in Mexico. The model used here combines features of two existing models. As in Levy-Yeyati, Stein, and Daude (2001), one of the main forces of attraction for FDI is represented by an extended market variable that reflects the overall economic size of the free-trade area. The model used here is not strictly a gravitational model, however, in that it is not restricted to bilateral investment flows and it does not explicitly incorporate a measure of distance. As in Albuquerque, Loayza, and Serven (2003), the model used here takes into account both local and global factors to explain FDI flows into a country. 478 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 3 Panel Regressions of FDI A panel model is estimated to explain the behavior of total inflows of FDI into each country in the sample. This differs from previous panel studies, which have focused mostly on bilateral stocks. In addition, as in Albuquerque, Loayza, and Serven (2003), indicators of macroeconomic stability and a direct measure of the globalization process are included. The data set is a panel of 44 countries with observations over 1980–99.5 The free-trade agreements considered for the construction of these variables are largely the same as those in other studies of regional integration, such as Frankel and Wei (1998): the Association of South- east Asian Nations, the European Free-Trade Association, the European Union, NAFTA, the Group of Three, the Andean Group (in its recent revival), and Mercosur.6 The dependent variable is the natural logarithm of total net inflows of FDI into a country.7 As in Levy-Yeyati, Stein, and Daude (2001), extended market variables were constructed to capture the positive and negative effects of inte- gration on FDI mentioned above. FTAGDP is the sum of the total GDP of all countries with which a country has free-trade agreements and the GDP of the index country, divided by the index country’s GDP. This is to capture the fact that the effect of a free-trade agreement on the FDI received by a country varies with the size of the country relative to its trading partners—for example, NAFTA could have a large effect on FDI flows to Mexico but is likely to have a much smaller effect on FDI flows to the United States. Naturally, FTAGDP is expected to have a positive coefficient. To capture the investment diversion effect, an overall measure of integration, INTEGRATION, was constructed as the weighted total GDP of all countries in the sample that belong to a free-trade arrangement. The weights are the share of total sample GDP covered by each country’s trade agreements. Thus, Chile’s GDP is included in this measure because, toward the end of the sample period, it joined free-trade agreements with Mexico and other Latin American countries. But if it had had an agreement with the United States, its weight would have been higher. INTEGRATION is defined so that if all countries in the sample signed free-trade agreements with one another, it would equal the value of aggregate GDP in the sample. For an individual country, an increase in INTEGRATION with 5. The countries included in the estimation are Argentina, Australia, Austria, Belgium, Bolivia, Brazil, Canada, Colombia, Chile, China, Denmark, Ecuador, Egypt, Finland, France, Germany, Hungary, Ice- land, India, Israel, Italy, Japan, the Republic of Korea, Malaysia, Mexico, Netherlands, New Zealand, Norway, Paraguay, Peru, Poland, Portugal, Singapore, South Africa, Spain, Sweden, Switzerland, Thai- land, Turkey, Uganda, United Kingdom, United States, Uruguay, and Venezuela. 6. The European Free Trade Association includes Iceland, Lichtenstein, Norway, and Switzerland. The Group of Three includes Colombia, Mexico, and Venezuela. And the Andean Group includes Bolivia, Colombia, Ecuador, Peru, and Venezuela. In addition, the free trade agreement between Chile and Mexico was included. 7. The ‘‘net’’ in ‘‘net inflows’’ indicates the subtraction of repatriations of capital by foreign investors, not the investment abroad undertaken by domestic investors. Cuevas, Messmacher, and Werner 479 FTAGDP held constant would signal stronger competition from other countries, which have become better integrated with one another. This should, other things being equal, make the concerned country less attractive to global inves- tors.8 Therefore, the coefficient of INTEGRATION is expected to be negative.9 It is well known that the bulk of FDI flows among rich countries, perhaps reflecting the quality of their institutions and their high financial development. The ratio of a country’s per capita GNI to that of the United States, RELGNIPH, was included to control for this effect.10 Size, measured by GDP and EXPORTS (which also captures outward orientation), should have a positive impact on FDI inflows thanks to a simple scale effect and the possibility of economies of scale or transport costs. These variables represent the ‘‘gravitational’’ element in the model. The ratio of the external current account balance to GDP (CURRENT), the ratio of the budget surplus to GDP (BUDGETBAL), consumer price inflation (INFLA- TION), and real growth of output (GDPGRWTH) are meant to capture such elements of the investment environment as macroeconomic stability and good economic policy management. The average years of education of the population aged 25 or older, EDUCATION, was included as a regressor because FDI might complement a highly skilled labor force. Finally, the ratio of mining to GDP, MIN/GDP, was also included, because FDI may be higher in countries with important mineral resources, irrespective of their institutional environment. The rate of real growth in global output, WORLDGRTH, and the one-year Treasury bill rate in the United States, US1YTBILL, are used as controls for the state of the world in a given year. The aggregate level of inward FDI flows in the world, FDIWORLD, is included to control for the increasingly important forces of globalization. (Sum- mary statistics for all the variables are provided in the appendix.) All the regressions yield positive, significant, and fairly stable coefficients on FTAGDP, strongly supporting the notion that entering a regional trade block should lead to higher FDI inflows (table 1). This suggests that a free-trade agreement has a positive effect on FDI and that the magnitude of the effect depends on the size of the trading partners. For NAFTA in 1994, the regression coefficients imply an increase of 17–26 percent in FDI flows to Mexico—but an increase of only 0.05–0.07 percent for the United States. Once time trends are 8. This variable should handle some cases that might trouble the investment diversion variable in Levy-Yeyati, Stein, and Daude (2001). Their model has the difficulty capturing the adverse competitive effects that might arise among third-party countries. For example, when Argentina and Brazil joined Mercosur in 1991, they became more attractive to third-party investors who might also have been considering investing in other countries in the region (say, Chile). The specification here would show no change in Chile’s FTAGDP, while reporting an increase in INTEGRATION. 9. All variables with a monetary dimension, such as INTEGRATION and others mentioned later, are measured in constant dollars and expressed in natural logs. 10. Given that what constitutes a ‘‘rich country’’ changes over time as the level of GNI per capita in rich countries fluctuates, GNI per capita is expressed relative to that of the United States. In a sense, this variable is really an index of how well off a country is at a given point in time by comparing the country’s performance with that of the leading market economy. T A B L E 1 . Fixed-Effects Regressions of the Log of FDI Against Membership in a Free-Trade Agreement and Other Variables Variable/Model Regression 1 Regression 2 Regression 3 Regression 4 RELGNIPH 0.452 (0.420) 2.779 (1.420) 5.253 (2.910) 1.421 (0.530) GDPGRWTH 0.030 (2.940) 0.018 (1.850) 0.016 (1.620) INFLATION 0.000 (À0.330) 0.000 (À0.070) 0.000 (À0.080) 0.000 (0.100) BUDGETBAL À0.015 (À1.130) 0.030 (1.990) 0.035 (2.380) 0.025 (1.390) CURRACCT À0.038 (À3.870) À0.023 (À1.89) À0.031 (À2.410) WRLDGRWTH À0.043 (À1.150) À0.019 (À0.55) 0.014 (0.430) À0.046 (À1.180) US1YTBILL 0.026 (1.400) 0.037 (1.450) 0.055 (2.360) 0.050 (1.710) EXPORTS 0.659 (3.470) 0.737 (2.270) 0.904 (2.650) FDIWORLD 0.642 (5.690) 0.424 (2.950) 0.356 (2.520) 0.522 (3.180) 480 GDP 0.113 (0.530) 0.404 (1.650) 0.709 (3.460) 0.462 (1.750) FTAGDP 0.009 (2.800) 0.010 (2.150) 0.011 (2.460) 0.013 (2.420) INTEGRATION 0.047 (0.350) À0.157 (À1.200) À0.010 (À0.080) À0.241 (À1.650) EDUCATION 0.505 (1.050) 4.451 (3.530) 5.272 (4.230) 6.209 (4.510) MIN/GDP 0.045 (2.260) 2 R 0.8329 0.8807 0.8801 0.8945 F-test regressors 59.09 46.31 48.37 40.82 Degrees of freedom 56,664 99,621 97,623 98,472 Number of observations 721 721 721 571 Note: Numbers in parentheses are t-statistics. Regressions 1–3 exclude the ratio of mining to GDP because only few observations are available. Regressions 2–4 include country-specific time trends because of the possibility of unit roots in several variables. Regression 3 excludes GDP growth, current account balance, and exports because of the possibility of reverse causation. Source: Authors’ analysis based on World Bank (2002) and global FDI data from UNCTAD (various years). Cuevas, Messmacher, and Werner 481 included, the coefficients on INTEGRATION become negative, though they are not significant. This suggests that if there is any diversion effect from free-trade agreements, it is quite small. The sign of the coefficient on RELGNIPH is positive, as expected, but generally not significant. It is significant when GDP growth is excluded (regression 3). This suggests that a high degree of collinearity exists between the two variables in a panel setting and, in a broad sense, that being rich may be seen as the result of possessing other characteristics that make a country attractive to investors (such as a relatively stable economic environment and high levels of education), and once these are controlled for it ceases to be a key factor. Globalization forces have the expected positive sign and are significant; in fact, FDIWORLD is one of the key determinants of net FDI inflows into any given country in any given year, and the elasticity of net FDI with respect to globalization forces is 0.35–0.64 in all regressions. The elasticity of FDI inflows with respect to exports is 0.65–0.9 in all regres- sions. This is consistent with expectations, but a simultaneity problem may affect exports if FDI targets traded sectors and improves export performance. The use of annual data reduces this risk because there is likely a long gestation period between new investment and increased exports. The purchase of existing production facilities through mergers and acquisitions accounts for a significant share of FDI. As Caldero ´n, Loayza, and Serve´n (2002) note, this accounts for the majority of FDI in developed countries and a substantial share of FDI in develop- ing countries, and it is only with a lag that such operations tend to lead to new investments in additional physical capital. This extra lag further reduces the risk of simultaneity. Finally, excluding EXPORTS does not affect the significance of FTAGDP, the main focus of this study. GDPGRWTH is more likely to experience a simultaneity problem. But the vari- able’s significance tends to decrease once a time trend and the share of mining in GDP are included, and its exclusion does not affect the significance of FTAGDP. Most coefficients of the control variables for economic environment have the expected signs. High economic growth seems to facilitate FDI. The budget balance seems to have a positive effect on FDI, controlling for time trends, but the coefficient is not significant when MIN/GDP is included. With a negative coefficient, the current account balance seems to represent financing need rather than an unstable environment in all four regressions. The coefficient on the rate of economic growth of the world is insignificant, and its negative sign is opposite to the one expected. This may be due to episodes in which capital flowed out of the largest countries during slowdowns, though the coefficient is clearly unstable. A more educated workforce seems to have a very significant positive effect on FDI flows, suggesting strong complementarities between FDI and a skilled workforce. Finally, some regressions included other variables that turned out not to be significant and thus are not included in the reported results. In particular, to capture the possibility that free-trade agreements may have a stronger effect 482 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 3 when they occur between contiguous countries, a dummy variable (equal to 1 when a free-trade agreement included countries with a common border) was interacted with FTAGDP. Unfortunately, many free-trade agreements include at least one bordering country; thus, the correlation of the dummy variable with FTAGDP is high, and its inclusion renders both variables insignificant. The real effective exchange rate was also included in the regressions to assess whether FDI flows increase when the real exchange rate is at a depreciated level. The coefficient has the expected sign in the regressions with time trends, but it is never significant, and its inclusion does not affect the level or significance of the other variables. To test whether a free-trade agreement has a one-time effect on the desired stock of FDI and thus only a temporary effect on FDI flows, a dummy variable was created for each age of the trade agreement and multiplied by FTAGDP—that is, a dummy variable for all first free-trade agreement years, a dummy variable for all second free-trade agreement years, and so on. These dummy variables are also insignificant, suggesting that the effect of a free-trade agreement is not pure stock adjustment but a permanent increase in FDI flows. To test whether the effect of a free-trade agreement is modified by cultural affinity, a dummy variable equal to 1, if another member of the agreement has the same official language, was interacted with FTAGDP, but its coefficient was not significant. Finally, some regressions included World Bank indexes of government stability, law and order, and the bureaucracy’s quality. The coefficients of these variables have positive signs as expected, but they are not significant. The effect of these variables may be captured by other variables—they have a very high correlation with EDUCATION and show less variability over time. Application of Results from FDI Regressions to the Mexican Case The estimates of the relationship between FDI and the explanatory variables of the regressions are now combined with Mexican data to determine the con- tribution of various factors to the growth of FDI inflows in Mexico. The main conclusion of this analysis: participating in NAFTA during the second half of the 1990s may have resulted in FDI inflows to Mexico roughly 60 percent higher than would have been received without participating. A second conclusion: factors not included in the regressions caused exceedingly slow growth of FDI inflows in Mexico in the late 1990s. Two possible explanations of this disap- pointing performance are a low level of mergers and acquisitions activity in Mexico’s FDI and the halt in economic reforms in Mexico since the mid-1990s, which has placed the country at a disadvantage relative to countries that have continued reforming. The fit of the predicted values from regressions 3 and 4 with the actual Mexican data is quite good (figure 2). The correlation coefficient is above 0.85. Because the Mexican data are only a subset of the data used to estimate the regressions, the variance of ln(FDI) cannot be decomposed into the variances of the fitted values and of the estimated errors to compute a sort of R2 for the Cuevas, Messmacher, and Werner 483 F I G U R E 2 . Actual and Predicted FDI in Mexico 10 9.5 9 8.5 Log of FDI 8 7.5 7 Actual FDI Regression 3 6.5 Regression 4 6 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 Source: Authors’ analysis based on the regressions reported in table 1. Mexican data. The estimated error term and the fitted values are not orthogonal for any given subset of the observations, although they are mutually orthogonal by construction for the whole set of observations.11 Between 1991 and 1999, real FDI inflows in Mexico approximately doubled. With the coefficients estimated in regressions 3 and 4 and the data for Mexico, the different forces behind the overall increase in FDI can be distinguished. Regression 3 is used because it presumably avoids endogenous effects that may be present in the other regressions, and regression 4 is used because it enables the inclusion of the potential effect of NAFTA on exports and a possible indirect effect of the greater integration on FDI. Overall, the regressions clearly tend to overpredict FDI in Mexico in 1999 (table 2). This is simply a numerical representation of the fact that the regressions overpredict FDI toward the end of the sample (see figure 2)—that is, the error term turns large and negative.12 With regression 3, FDI after joining NAFTA was nearly 25 percent higher than it 11. The estimated error and the fitted values are almost orthogonal for regressions 2–4, with the correlation coefficient between À0.09 and 0.03. The correlation coefficient for regression 1 is À0.45. 12. The calculations in table 2 are illustrative, since the regressions estimated, being linear in logs, are multiplicative in levels, and thus the order in which each independent effect is presented influences the apparent (additive) magnitude of its contribution to the change in FDI. 484 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 3 T A B L E 2 . Illustrative Decomposition of the Rise in FDI in Mexico between 1991 and 1999 (millions of 1995 U.S. dollars) Concept FDI (Regression 3) FDI (Regression 4) Predicted FDI in 1991 4,724 4,724 Predicted FDI in 1999 with change in country 5,842 6,748 factors Predicted FDI in 1999 with increase due to 10,724 14,018 global processes Predicted FDI in 1999 with increase due to NAFTA 13,394 18,262 Actual FDI in 1999 10,782 10,782 Note: The difference between lines 4 and 5 is the regression error. Source: Authors’ analysis based on World Bank (2002) and global FDI data from UNCTAD (various years). would have been if Mexico had remained outside any regional arrangements; with regression 4, it is 30 percent higher. FDI was 60 percent higher once NAFTA-induced exports were taken into account.13 This estimate is similar to that obtained by Waldkirch (2003), who concluded that FDI from North America would have been some 42 percent lower without NAFTA (that is, NAFTA led to a 72 percent increase in FDI from Canada and the United States). Echoing other researchers such as Borja Tamayo (2001), Waldkirch (2003) argues that NAFTA has not brought about a significant increase in FDI from outside North America.14 This rough similarity is interesting, given that Waldkirch’s approach is quite different from the one here. He uses panel data on FDI stocks in Mexico by source country to assess NAFTA’s effect on the FDI stocks owned by nationals of NAFTA and non-NAFTA countries. By contrast, this article uses a cross-country panel data to assess the effect of all free-trade agreements in general and applies the results to the particular quantification of NAFTA’s effect in Mexico. The results here are not strictly comparable to Waldkirch’s (2003) because the model used here cannot distinguish NAFTA’s 13. The idea is to assess the effect of export growth due to NAFTA and then transfer this quantity to the estimation of the free trade agreement effect. NAFTA’s effect on exports is assumed to have stimulated the growth of nonoil exports beyond its traditional rate of growth. Thus, the 9.6 percent growth rate of real nonoil exports between 1980 and 1991, just before NAFTA negotiations started, is used to project a NAFTA- free level of exports in 1999. This amount is $35 billion less than the actual level of exports observed in 1999. The additional, NAFTA-induced exports’ effect on FDI is then calculated using the coefficient from regression 4. Adding this effect to the direct effect of the change in FTAGDP yields a total increase of 53–65 percent associated with NAFTA. 14. The maquiladora program, used to give U.S. firms an advantage over other foreign firms, predates NAFTA by decades. NAFTA might have been expected to reduce the relative advantage of maqui- ladora establishments, since other firms would be able to enjoy similar benefits. But investment in the sector has significantly increased since the mid-1990s. Most maquiladora establishments are close to the Mexico–U.S. border, which suggests that just-in-time production and distribution are important factors driving investment in Mexico, given that other countries have much lower labor costs in manufacturing. However, as explained, contiguity effects are hard to identify in the sample. Cuevas, Messmacher, and Werner 485 effects on FDI from different regions. In sum, according to the regression results here, FDI would grow by a factor of 3, with a 20 percent increase due to domestic factors, a 100 percent increase due to Mexico’s participation in global currents of investment, and a 67 percent increase due to joining NAFTA. The level of FDI forecasted by the regressions for 1999 is substantially higher than the actual level of FDI (see figure 2 and table 2). Clearly, the model is not a dynamic one, and it is not fitted using only Mexican data, which could lead to autocorrelation in the error terms during certain peri- ods. But factors not included in the regressions might explain why FDI grew less than predicted. Several events could help explain the seemingly lackluster performance of FDI during the last few years of the 1990s. First, there could be lingering effects from volatility occasioned by the Tequila crisis. Second, as noted by Caldero ´n, Loayza, and Serve ´n (2002), during the 1990s a growing share of global FDI took the form of mergers and acquisitions of existing firms. The share of mergers and acquisitions investment in Mexico’s FDI was atypically low, but this indicates only which form of FDI Mexico failed to receive; it does not explain why Mexico did not receive more FDI in the first place. A possible explanation of the low level of mergers and acquisitions is the limited size and early end of the Mexican government’s privatization program.15 Third, even though the free-trade agreement age dummy variables included in some speci- fications (but not reported) were not significant, NAFTA may still be responsible for only a brief stock adjustment starting in 1994, in which case that adjust- ment is over and FDI is settling toward more ‘‘normal’’ levels. Bajo-Rubio and Lo´pez-Pueyo (1997) and Ferna ´ndez de Cordoba and Kehoe (2000) suggest that this was the case after Spain’s entry into the European Community. The free-trade agreement age dummy variables here may be too simple to capture this nonlinear process. Another possibility is that the regressions do not handle country contiguity correctly for Mexico. Gravity models that take explicit account of distance, such as that of Levy-Yeyati, Stein, and Daude (2001), seem to suggest that free-trade agreements with immediate neighbors are likely to be more meaningful than agreements with geographically distant partners. Also, because those models focus on bilateral flows, they handle contiguity more naturally. Most free-trade agreements comprise nearby countries. However, the variable FTAGDP makes no 15. The strengthening of the pro-competition laws in the mid-1990s may have affected the level of mergers and acquisitions activity in Mexico. These laws limit mergers between companies to prevent excessive concentration, and perhaps their more competitive environment reduced the expected profit from buying a Mexican company. By contrast, restrictions on domestic alliances could have led firms to seek alliances with foreign companies. While the planned acquisition of Bancomer by Banamex in 2000 may have been impeded by concerns about industry concentration, this was not the case in the acquisition of these banks by foreign institutions. In fact, the peak in the series of FDI in Mexico reflects Citicorp’s acquisition of Mexico’s Banamex in 2001. 486 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 3 distinction between contiguous and noncontiguous partners. This is not a systema- tic problem, and so there is no reason to regard the parameters of the regressions as suspect. Moreover, the dummy variable for common border in a free-trade agree- ment was not significant because almost all the sample countries have a border with at least another member of the agreement. There are too few agreements between noncontiguous countries to draw any conclusions from. However, it could be an issue in Mexico’s case because the United States, which is very large geographically, is located between Mexico and Canada. When considering only the United States as a partner of Mexico’s, the prediction error of the late 1990s was in fact reduced, but only marginally. This likely reflects the fact that Canada is much smaller than the United States and thus that the distance between Mexico and Canada is only a very small part of the story. In addition to the arguments presented here, however, a policy issue also exists. The weakening growth of FDI flows into Mexico seems to have coincided with the perceived decline of that country’s competitiveness. Between 1999 and 2001, in fact, Mexico’s position in the World Economic Forum’s (2000, 2001) global competitiveness ranking dropped from 31 to 43. A partial list of the emerging market countries that overtook Mexico by 2001 is quite suggestive: Hungary, Poland, the Czech Republic, Turkey, and the Slovak Republic were all upgrading their institutions with a view to joining the European Union, and Greece was trying to strengthen its macroeconomic policies to meet the Maastricht criteria for monetary integration. These countries were working on structural reforms and deepening their regional bonds. Mexico offers a clear contrast: the pace of structural reform slowed drastically after 1997 (when a Congress in which no political party had enough votes to pass any law by itself was elected), and NAFTA is and will remain for the foreseeable future a trading bloc. III. CONCLUSIONS NAFTA’s arrival coincided with a transformation of the capital account of the balance of payments in Mexico. Initially, a boom in portfolio flows into the stock exchange was observed during the agreement’s negotiation. It is difficult to gauge, however, the extent to which these inflows were motivated by inves- tors’ desire to take positions in anticipation of the agreement, given the coin- cidence of a strong privatization program at the time and the normalization of the Mexican government’s borrower status. Later, FDI became the most dynamic element of the capital account. In fact, the FDI series shows at least two clear jumps. The first occurs in the early 1990s and reflects the extensive reform process that Mexico followed from the mid-1980s to the mid-1990s. Among the reforms is a change in the investment law that allowed greater FDI in several economic sectors and the beginning of NAFTA negotiations. The second occurs when the agreement goes into effect, and since then FDI has been by far the most important type of capital inflow in Mexico. Cuevas, Messmacher, and Werner 487 Analysis of a multicountry panel data set indicates that free-trade agreements seem to have a strong positive influence on FDI. In addition to membership in a free-trade agreement, a set of global variables, namely world FDI flows, and a set of domestic variables seem to determine FDI. Among the domestic variables, having a more open economy, as measured by the level of exports, increases FDI. A more stable macroeconomic environment, as measured by the ratio of the budget balance to GDP, also increases FDI. Finally, a more educated workforce also attracts higher flows, possibly due to complementarities between FDI and a skilled workforce. FDI in Mexico under NAFTA was an estimated two-thirds higher than it would have been without NAFTA, despite the winding down of Mexico’s privatization program and the banking and currency crises. The effect is so large because trade liberalization’s effect on FDI depends on the relative sizes of the trading partners—there is a large asymmetry between Mexico and its NAFTA partners. A somewhat puzzling result of the model is that in the late 1990s FDI in Mexico is lower than expected. This was likely partly the result of the apparent halt in the country’s structural reform process. Nevertheless, a fuller explana- tion of this shortfall could not be offered here, and it remains an important area for additional research. APPENDIX T A B L E A - 1 . Summary Statistics of Variables Used in Estimation Variable Mean Standard Deviation Minimum Maximum FDI (log) 7.19 1.94 0.03 12.44 RELGNIPH 0.50 0.28 0.03 1.07 GDPGRWTH 3.52 3.72 À11.89 14.82 INFLATION 55.21 253.68 À0.13 3,763.31 BUDGETBAL À3.21 4.13 À21.85 15.53 CURRACCT À1.25 4.54 À16.35 25.02 WRLDGRWTH 2.75 1.12 0.60 4.80 US1YTBILL 6.65 2.74 3.02 14.08 EXPORTS (log) 10.11 1.52 5.60 13.46 FDIWORLD (log) 12.26 0.68 11.24 13.80 GDP (log) 11.93 1.48 8.32 15.94 FTAGDP 11.97 23.19 1.00 138.05 INTEGRATION 14.08 0.75 12.77 14.86 EDUCATION (log) 1.92 0.37 0.79 2.51 MIN/GDP 12.33 4.80 0.00 30.39 Note: The number of observations for MIN/GDP is 571. The number of observations for all other variables is 721. Source: Authors’ analysis based on World Bank (2002) and global FDI data from UNCTAD (various years). 488 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 3 REFERENCES Albuquerque, Rui, Norman Loayza, and Luis Serven. 2003. ‘‘World Market Integration through the Lens of Foreign Direct Investors.’’ Policy Research Working Paper 3060. World Bank, Washington, D.C. Bank of Mexico. Various years. The Mexican Economy. Mexico City. Bajo-Rubio, O., and C. Lo´pez-Pueyo. 1997. ‘‘An Industry Analysis of Foreign Direct Investment in Spanish Manufacturing, 1986–1992.’’ Documento de Trabajo 9804. Universidad de Navarra, Pam- plona, Spain. Borja Tamayo, A. 2001. ‘‘El TLCAN y la inversio ´n extranjera directa: el nuevo escenario.’’ In Para evaluar el TLCAN. Tizapa ´ ngel, Me ´n-San A ´xico: Miguel Angel Porru´a. ´n, C. N. Loayza, and L. Serve Caldero ´n. 2002. ‘‘Greenfield FDI vs. Mergers and Acquisitions: Does the Distinction Matter?’’ World Bank, Washington, D.C. Dussell Peters, Enrique, Luis Galindo Paliza, and Eduardo Lorı ´a. 2002. ‘‘Visio´n microecono´mica de los impactos de la integracio ´n regional en la inversio´n inter e intraregional: el caso de la inversio ´n extranjera directa en Me ´xico.’’ Universidad Nacional Auto´noma de Me ´xico, Facultad de Economı ´a, Mexico City. ´ndez de Cordoba, G., and T. Kehoe. 2000. ‘‘Capital Flows and Exchange Rate Fluctuations Ferna Following Spain’s Entry into the European Community.’’ Journal of International Economics 51(1):49–78. Frankel, J. A., and S. Wei. 1998. ‘‘Regionalization of World Trade and Currencies: Economics and Politics.’’ In J. A. Frankel, ed., The Regionalization of the World Economy. Chicago, IL: University of Chicago. Greene, W. 2003. Econometric Analysis. 5th edn. New York: Prentice Hall. Kehoe, P., and T. Kehoe. 1994. ‘‘Capturing NAFTA’s Impact with Applied General Equilibrium Mod- els.’’ Federal Reserve Bank of Minneapolis Quarterly Review 18(2):17–34. Krugman, P., and G. Hanson. 1993. ‘‘Mexico–US Free Trade and the Location of Production.’’ In P. Garber, ed., The Mexico-US Free Trade Agreement. Cambridge, MA: MIT Press. Levy-Yeyati, E. E. Stein, and C. Daude. 2001. ‘‘Regional Integration and the Location of FDI.’’ Universidad Torcuato Di Tella, Buenos Aires; Inter-American Development Bank, Washington, D.C.; University of Maryland, College Park. Morley, S., R. Machado, and S. Pettinato. 1999. ‘‘Indexes of Structural Reform in Latin America.’’ Economic Reforms 12. Economic Commission for Latin America and the Caribbean, Santiago. ´ n de la Negociacio Serra Puche, J. J. 1992. Conclusio ´ xico ´ n del Tratado de Libre Comercio entre Me Canada´ y Estados Unidos. Me ´a de Comercio y Fomento Industrial. ´xico City: Secretarı UNCTAD (United Nations Conference on Trade and Development). Various years. World Investment Report. Geneva (http://www.unctad.org). U.S. Congressional Budget Office. 1993. ‘‘Estimating the Effects of NAFTA: An Assessment of the Economic Models and Other Empirical Studies.’’ Washington, D.C. Waldkirch, A. 2003. ‘‘The New Regionalism and Foreign Direct Investment: The Case of Mexico.’’ Journal of International Trade and Economic Development 12(2):151–84. World Bank. 2003. ‘‘The Impact of NAFTA on Foreign Investment in Third Countries.’’ Washington, D.C. World Economic Forum. 2000. Global Competitiveness Report 2000–2001. Geneva. ———. 2001. Global Competitiveness Report 2001–2002. Geneva.