Exports and Income Inequality in Mexico

Abstract

What is the effect of exports on local income inequality in developing countries? To respond to this question, we combine data on exports with a panel of welfare indicators for 2000 Mexican municipalities, and implement an instrumental variable approach to address endogeneity concerns. Our results show that a 10 per cent increase in the exports per worker reduces local income inequality, measured by the Gini coefficient, by 0.17 points (using a 0 to 100 scale). This is driven by income growth among households at the bottom of the income distribution. We also find that although exports do raise the total amount of labour incomes at the municipal level, average labour incomes do not change. This occurs because municipalities with growing exports also experience an increase in employment and working-age population driven by inflows of returning migrants and lower emigration outflows. Remittances also decline in response to rising exports.

Keywords:

• international trade
• exports
• inequality
• local impacts
• migration

Acknowledgements

The authors wish to thank Raymundo M. Campos-Vasquez, Christoph Lakner, Gladys C. Lopez-Acevedo, and Ambar Narayan for their comments and suggestions. They are also grateful to participants at the World Bank authors workshop on the distributional effects of trade, the World Bank workshop on poverty and trade, and the 5th Congress Sobre Mexico organised by the Universidad Iberoamericana. The authors are indebted to Natalia Volkow and Liliana Martinez of the Microdata Center of the National Institute of Statistics and Geography, Mexico, for their support for data use in compliance with the confidentiality requirements set by Mexican law. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organisations, or those of the Executive Directors of the World Bank or the governments they represent.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 In contrast, recent empirical evidence points to the negative impact import competition from developing countries had on workers in developed economies (for example, see Autor, Dorn, & Hanson, 2013; Balsvik, Jensen, & Salvanes, 2015; Hakobyan & McLaren, 2016).

2 Mexico joined the forerunner of the World Trade Organization, the General Agreement on Tariffs and Trade, in 1986 and the North American Free Trade Agreement (NAFTA) in 1994, leading to substantial tariff reductions globally and regionally, greater export orientation, and diversification away from oil. Between 2004 and 2014, poverty declined in Mexico by 4.0 percentage points, from 37.6 per cent to 33.6 per cent, while, in Latin America and the Caribbean, it declined by nearly 17.0 percentage points, from 41.3 per cent to 24.4 per cent (US$5.50-a-day per capita poverty line in 2011 purchasing power parity). Moreover, almost half the decline in poverty in Mexico is explained by redistribution (because of the shift from general subsidies to targeted and conditional transfers) rather than to economic growth, despite the expansion in trade over the period. In Latin America, in contrast, economic growth accounted for 80 per cent of the reduction in poverty, while the rest was due to redistribiution.

3 Most of Mexico’s exports – about 70 per cent to 80 per cent since 2004 – go to one high-income economy, the United States.

4 That increasing exports did not translate into lower poverty and higher income growth is not yet fully understood. Lederman, Maloney, and Servén (2005) argue that NAFTA had positive impacts on foreign direct investment (FDI), trade, and productivity in Mexico, but only modest impacts on wages and the convergence of income per capita. Barriga Cabanillas, Rodriguez-Castelan, and Lopez-Calva (2020) find that local exposure to trade is associated with higher productivity in the manufacturing sector which in turn declines with higher industry concentration. Accordingly, suggestive research points out that negative aggregate shocks in Mexico tend to have important negative effects in regional growth, but that positive shocks, such as a rise in export growth, are not typically reflected in large positive effects on growth nor, to a certain extent, on household welfare. In most states in Mexico between 2005 and 2014, as Campos-Vázquez and Monroy-Gómez-Franco (2016) observe, negative shocks increased poverty more than positive shocks reduced it. The dynamics behind this observation is, up to a point, what this paper aims to test, that is, to identify how and whether a positive trade shock – growth in exports – benefits (or not) the population at the local level.

5 See UNDESA (2008); HS (Harmonized Commodity Description and Coding Systems) (database), Statistics Division, Department of Economic and Social Affairs, United Nations, New York, https://unstats.un.org/unsd/tradekb/Knowledgebase/50018/Harmonized-Commodity-Description-and-Coding-Systems-HS.

6 While there are sources of information that would allow for more frequent (for example annual) measures of household incomes and Gini coefficients at the subnational level in Mexico, they are not suitable for the purposes of this study. In particular, the most disaggregated level at which the ENIGH (‘Encuesta Nacional de Ingreso y Gasto de los Hogares’) survey – the main survey to monitor poverty and inequality in Mexico – allows to estimate representative statistics is the State (‘Entidad’), and also State-urban and State-rural (that is at most 64 geographies). That is, it does not allow to obtain representative statistics at the municipal level. This issue coupled with the finding that often the local impacts of trade tend to be slow (Autor, Dorn, & Hanson, 2016) makes the more frequent data not very useful. In contrast, having household data for about 2000 municipalities helps improve the precision of our estimates (rather than using only 64 geographies) while also allowing to detect long-term impacts.

7 Even though the 2015 population count is from a survey, the sample size is sufficiently large to provide statistics representative at the municipality level. See Enamorado et al. (2016).

8 Because the 2005 data do not include detailed information on labor market outcomes, 2000 data are used instead as the initial year for these indicators.

9 These include the following sectors: agriculture, mining, food products, textiles, wood, paper, printing, chemicals, rubber and plastics, metal, electronics, machinery, automotive, other manufacturing, utilities, construction, education, and other nonmanufacturing.

10 To prevent losing municipalities from the sample that have zero exports in one or both years because of the logarithmic transformation, a constant equal to 1 is added to the exports per worker variable so that this is never zero.

11 For example, Autor, Dorn, and Hanson (2013) estimate the impacts of import competition from China on local US labor markets by combining the initial local distribution of employment across sectors and nationwide changes in imports from China by sector. They instrument this variable using changes in Chinese imports by other high-income countries.

12 Goldsmith-Pinkham et al. (2020) shows that the Bartik estimator can be decomposed into a weighted sum of the instrumental variable estimators that use each industry share as a separate instrument. Such weights are the Rotemberg weights.

13 Goldsmith-Pinkham et al. (2020) show as an example that an R2 of 0.46 is ‘very high’ and thereby the variables that help predict the outcome can also help predict the instrument.

14 Since in the case of Agriculture we find that all of the initial characteristics are statistically significant, we run an additional test. To rule out the hypothesis that the results are driven by the Agricultural sector (because this would raise concerns about the validity of the instrument), we remove agricultural exports from both the instrument and the endogenous variable, and re-estimate the main results. We find that the results are not affected by the exclusion of the agricultural sector. These results are available upon request.

15 In a lin-log model, to calculate the expected change in the dependent variable associated with a 10 per cent rise in the independent variable, the estimated coefficient should be multiplied by log(110/100) = log(1.1) = .095.

16 This is not driven by the fact that plant-level studies focus on wages while we focus on household income. As as seen in Table A3 in the Appendix, we do not find that exports have an unequalizing impact on wages and labor income.

17 The list of municipalities excluded due to zero or very high exports per worker is presented in Table A4 in the Appendix.

18 See ‘U.S. Immigration Trends’, Migration Policy Institute, Washington, DC, https://www.migrationpolicy.org/programs/data-hub/us-immigration-trends#source.

19 These data are not available for 2005 and 2015.