Introduction

Foreign Direct Investment (FDI) is conventionally recognized as an important resource for economic development. Nevertheless, the empirical literature on the contribution of FDI to poverty reduction is not consensual. Studies generally recognize a central role of FDI in growth and development (^{Iamsiraroj, 2016}; ^{Dinh et al., 2019}). The positive effects are felt through increased supply, reduced prices, tax revenue, and government spending. Moreover, according to ^{Hayami (2001)}, FDI is an important way for the host country to break out of the vicious circle of underdevelopment. In particular, FDI offers an alternative path to combat poverty (^{Ahmad et al., 2019}; ^{Ganic, 2019}) mainly because of its effects on the demand for skilled or unskilled labor (^{Zia & Nishat, 2010}).

The literature points to an indirect impact of FDI on poverty reduction through economic growth (^{Magombeyi & Odhiambo, 2017}) and inequality (^{Suanes, 2016}; ^{Fazaalloh, 2019}). Developing countries seek to combine domestic with foreign resources to boost economic growth and societal income and thus reduce poverty (^{Do et al., 2021}).

In the case of Mexico, the literature has not sufficiently explored the relation between FDI and poverty by state. It tends to focus on its determinants or effects on growth, productivity, exports, inequality, and employment from a national or sectoral perspective and, to a lesser extent, from a regional analysis.

^{Díaz and Turner (2012)} mention that until 2000, in the northern border states, the main beneficiaries of the expo-oriented model, increasing amounts of FDI led to high growth rates and low levels of unemployment, thus containing the advance of poverty. In contrast, ^{Bravo (2015)}, using simple linear regression, finds that FDI is not necessary for poverty reduction at the national and state levels during 1990-2010 and that high GDP and FDI flows do not guarantee decreased state poverty.

Thus, according to ^{Torres et al. (2017)}, some states have been more successful than others in attracting FDI. From 2008-2018, Mexico City captured 17.8% of the total FDI in the country, followed by Nuevo Leon and the State of Mexico with just over 9% each. Conversely, Nayarit, Tlaxcala, Colima, Yucatán, and Chiapas each received less than 0.5% of the total. Thus, there is a high level of inequality in the levels of FDI captured at the state level. ^{Figueroa (2013)} explains that FDI is concentrated in cities with higher levels of infrastructure, purchasing power, and education.

At the same time, the states with the greatest FDI attraction (e.g., Mexico City, Nuevo León, Baja California Sur, Coahuila), in general, contribute with greater gross value added and have lower percentages of the population living in poverty. Similarly, the states with the lowest attraction of foreign capital have the highest poverty levels (e.g., Chiapas, Guerrero, Oaxaca). There is also a group of states with relatively high participation of FDI and value-added but high poverty levels (e.g., Veracruz, Puebla, Zacatecas). Nonetheless, despite following a growth model from trade intensification based on attracting FDI (average FDI growth of 58% between 2008 and 2018), neither the country nor the states have achieved a reduction in poverty of comparable magnitude (average decrease of 8.9% in that period).

Indeed, this paper intends to identify the effect of FDI on poverty at the state level in Mexico in 2010-2018, given the availability of data from CONEVAL, a source of state poverty data. It is recognized that poverty is a multidimensional phenomenon linked to monetary aspects and other conditions related to rights and freedoms that illustrate the ability of individuals and households to meet their basic needs. Nevertheless, the focus is on researching the effects of FDI on poverty reduction at the state level in Mexico, so it incorporates additional explanatory variables to broaden the view of the subject, such as economic, political, social, and infrastructure variables.

Section 2 reviews the theoretical and empirical literature on the FDI-poverty link. Section 3 describes the variables and the dynamic panel sys-GMM econometric methodology given the data structure. Section 4 presents the results and discussion. Finally, some reflections are presented. The paper contributes to the analysis of FDI, particularly its effect on poverty reduction at the state level in Mexico.

FDI and poverty

The literature has identified the direct and indirect impacts of FDI on poverty reduction in the foreign country in which it operates (^{Oishi & Kesebir, 2015}). The indirect impact is through economic growth, resulting in improved living standards (improved productivity and economic environment). In contrast, the direct impact is generated through increased employment and a reduction in people living below the poverty line due to increased demand for employment and improved labor force (^{Klein et al., 2001}).

Regarding the first effect, ^{Dollar and Kraay (2000)} estimate that growth tends to increase the incomes of people experiencing poverty in direct proportion. ^{Kakwani (2000)} indicates that the positive effects of FDI more than offset the negative effects-crowding out of domestic investment- resulting in net economic growth and poverty reduction. It can also reduce the adverse impact on people experiencing poverty of financial instability and improve government management capacity (^{Hung, 2005}). Likewise, it supports social safety nets based on redistributive public programs (^{Klein et al., 2001}). Moreover, as FDI drives economic growth, the local market expands (^{Jeong, 2009}), offering new income-earning opportunities for people with low incomes, i.e., the demand for labor grows from FDI itself, from new FDI, from complementary investments, from public or private investment or even from the informal market.

Additionally, FDI is associated with complementary productive investments and investments in public infrastructure (education, transportation, health, etcetera) (^{Wekesa et al., 2016}). ^{Hayami (2001)} states that FDI closes the gap between desired investment and domestic savings, increases tax revenues, and improves host economies’ management, technology, and labor skills. The net availability of capital in the domestic economic system due to FDI inflows depends on the substitution and complementarity between FDI and domestic capital (^{De Mello, 1999}). Thus, the greater the complementarity, the greater the contribution of FDI to poverty reduction.

The effects of FDI on poverty depend on various factors on both the productive capital and the host location. ^{Calvo and Hernandez (2006)} found that the benefits of FDI varied according to the initial local conditions and the orientation of the foreign subsidiary. Thus, on the one hand, the magnitude of the effect is determined by the amounts and characteristics of FDI associated with green investments, reinvestments, technological or labor intensity, industrial sector, tax exemptions, and labor hiring, among others (^{Xibao & Zhi, 2009}). Conversely, the economic, industrial, social, political, and governmental characteristics of the host location affect the scope of FDI (^{Elboiashi, 2015}), e.g., local suppliers, tax collection from FDI and its local use, community acceptance of a certain manufacturing plant, land use conversion it employs, etcetera. Collectively, all these conditions influence poverty reduction through FDI.

In particular, ^{Meyer (2004)} states that the mode of FDI entry plays a key role in the effects of FDI on poverty reduction in the host country. FDI can be in the form of mergers and acquisitions or greenfields. Mergers and acquisitions, which involve a change of ownership, bring fewer welfare benefits to the local society. In contrast, new FDI, which increases capital, brings more benefits.

Also, according to ^{Winters (2002)}, trade liberalization impacts poverty through changes in commodity prices, offering more products at competitive prices that give local society access to satisfiers. For ^{Francois et al. (2004)}, the increase in trade in developing countries, where unskilled workers tend to be in the majority, favors reducing the number of impoverished people.

The benefits of FDI in the host country depend on the orientation of the foreign subsidiary. There are three reasons for localizing FDI in a given country. Search for markets, motivating an increase in local content and employment. Search for raw materials, where employment generation may be limited. Search for a platform for exports, contributing to income and technology transfer (^{Chang 2003}). Regardless, in cases where FDI is capital-intensive or based on advanced technology, the host economy may lack skilled supporting labor, limiting the effect on employment, income, and poverty reduction (^{Calvo & Hernández, 2006}).

Similarly, the host country’s economic development level determines the potential advantages of FDI in poverty reduction (^{Meyer & Sinani, 2009}). The development level influences domestic companies’ number and capacity to insert themselves into the FDI value chain and thus benefit from inward FDI. A developing country tends to offer a poor business ecosystem, which limits the potential positive effects of FDI. Nevertheless, the impact of foreign direct investment is strong in low-income countries with high levels of social capabilities (supportive institutional framework, effective communication, a well-educated workforce, and infrastructure support) (^{Kemeny, 2010}).

To this end, ^{Klein et al. (2001)} indicate that there are preconditions for FDI to contribute to poverty reduction, among which are an adequate environment for foreign capital, an egalitarian and competitive platform that avoids protection for any capital, and regulation (restrictions and discrimination) of foreign investors. In particular, these measures favor the reduction of income poverty. Likewise, ^{Fowowe and Shuaibu (2014)} indicate that FDI inflows contribute to poverty reduction depending on institutional quality, human capital development, and financial system development.

Besides, in poverty reduction, the application of Corporate Social Responsibility (CSR) policies and guidelines by Multinational Enterprises (MNEs) in the FDI recipient country can be noted, as stated by ^{Akwaowo and Swanson (2016)}. MNEs, through their FDI, must face the important decision on how to act according to CSR guidelines, in order to, among other things, reduce poverty in the recipient countries. In this regard, ^{Tomas de Cavia (2019, P.3)} mentions that FDI, accompanied by CSR policies, measures, and regulations, is a dynamic factor of economies and a powerful tool for poverty reduction, especially when investments are conducted in developing countries.

According to ^{Kolstad and Tondel (2002)}, “The higher the level of FDI, the higher the level of economic growth, which enables better social development and other development goals.” In their line of argument, the authors assert that “the degree to which economic and social progress improves the lives of all members of society is identified through wealth distribution and poverty reduction.” Furthermore, ^{Visser (2009, P.481)} states, “There is a powerful argument that CSR in developing countries is directly linked to the socioeconomic environment in which business organizations operate and, above all, to the development priorities that this entails. This argument is linked to poverty reduction, health provision, infrastructure development, and education.”

Nevertheless, the direct effect of FDI on poverty reduction requires that the demand for labor mostly favors the unemployed poor or low-income earners. When they find work where they receive training, they become skilled laborers with aspirations for higher incomes and higher living standards, surpassing the poverty line (Sarisoy & Koc, 2012). For ^{Do et al. (2021)}, FDI directly alleviates poverty through better training of local workers.

The ^{World Bank (WB, 2019)} also considers individuals who receive less than one dollar a day to be in extreme poverty and those who receive less than two dollars a day as poor. In the case of Mexico, the National Council for the Evaluation of Social Development Policy (^{CONEVAL, 2018}) considers a person to be in poverty when they have at least one social deprivation¹ and their income is insufficient to acquire goods and services to satisfy their basic needs. Likewise, it defines a person in extreme poverty if they have three or more deprivations and an income below the minimum welfare line. This income is considered so low that it does not allow them to acquire the nutrients necessary for a healthy life.

Therefore, poverty is the lack of economic opportunities, access to basic services such as education and health care, opportunities to participate in the public decision-making process, and deprivation of social protection. According to ^{Sen (1999)}, maintaining or expanding these deprivations would limit people’s ability to maximize their potential. Thus, limited individual capabilities translate into obstacles to freely choosing the most valued and desired way of life.

Thus, through public policies, the state seeks to generate an enabling environment so that people can provide adequate nourishment for themselves and their families. In particular, it has been interested in promoting the formation of human capital and specialized personnel, given its relevance in attracting FDI (^{Asali & Campoamor, 2011}).

Following this idea, ^{Checchi et al. (2007)} estimate that FDI positively affects human capital formation, reflected in greater employment opportunities and wages, which makes it possible to access health services, social security, quality housing, basic public services, and food. ^{Jensen and Rosas (2007)} state that FDI is related to improvements in the levels of income inequality, and through this, FDI impacts poverty reduction (^{Kwasi, 2017}).

Indeed, there is no consensus in the literature regarding the effect of FDI on poverty. Studies have identified estimated negative effects of FDI. ^{UNCTAD (1999)} notes that the direction of the impact of FDI depends on the variables included in the study. For example, ^{Huang et al. (2010)} find a negative effect of these capitals on poverty in 12 Latin American and East Asian countries. ^{Mihaylova (2015)} suggests that FDI likely focuses on capital-intensive activities, leading to higher unemployment rates in traditional sectors and increased poverty. The results of ^{Osemenshan et al. (2020)} show that FDI negatively affects poverty reduction. Specifically, they suggest that the level of FDI needed to alleviate poverty has not been achieved.

On the other hand, according to ^{Do et al. (2021)}, FDI has indirectly worsened poverty through international trade. Te Velde (2003) states that international trade, linked to the effect of FDI, may increase the demand for skilled rather than unskilled labor, which does not improve poverty.

Methodology and data

A dynamic panel data econometric methodology with fixed effects is employed. In particular, a Generalized Method of Moments (sys-GMM) model is estimated, which incorporates equations in differences and levels and a specific set of instrumental variables. The procedure used by ^{Ángeles and Ramirez (2014)} is followed.

Since the states show differences in the dimensions of interest, the existence of fixed effects is accepted. This information is verified with the Breusch and Pagan Lagrange multiplier test (BPLM), which evaluates random effects using the null hypothesis where the individual specific variance equals zero (H₀: σ² _u=0).

A dynamic panel data model (DPSM) that lags the dependent variable is proposed, which enables testing the correct specification of the model and the possibility of autocorrelation (^{Ángeles & Ramírez, 2014}):

Where X is the vector of explanatory variables, and POB is some variable related to poverty -defined below-in the i-th state at time t. The variables in X are grouped into four categories: productive and globalization (PROGLO), political (POL), government spending (GP), and infrastructure (INFRA). The model is specified from the review of the literature summarized above that estimates the effects of FDI on poverty. In particular, it is an extension of ^{Khan et al. (2019)}, ^{Uttama (2015)}, ^{Ucal (2014)}, and ^{Fowowe and Shuaibu (2014)}. These models incorporate as dimensions of analysis in addition to trade openness and FDI, other economic (GDP per capita, inflation, employment), financial (domestic and external public debt), political (regional integration, corruption), and infrastructure (telephone lines, goods transported via road, rail, air, or ports) factors. ^{Aaron (1999)} and ^{Hung (2005)} classify the possible effects of FDI on poverty reduction in the host economy as direct and indirect, and they are grouped into this group of variables.

The lagged dependent variable leads to the regressor γ and the error term u_it being correlated. Therefore, the Generalized Method of Moments (GMM) proposed by ^{Arellano and Bover (1995)} is used. The method eliminates the individual effects associated with each state, η_i, by posing a dynamic equation in the first differences as follows:

^{Bun and Windmeijer (2007)} state that if the dependent variable appears as an explanatory variable, the strict endogeneity of the regressors does not hold. To control for the endogeneity of the lagged dependent variable, P_t-1, reflected in the correlation between it and the error term in the new equation, the GMM estimator in the first difference is followed, with the lagged endogenous variables as instruments.

Nonetheless, this estimator shows finite sample bias and low precision (^{Blundell & Bond, 1998}), which are explained by the weakness of the instruments given the lagged levels of the series. This weakness is explained by the variance of the individual effects, η_i, concerning the variance of u_it (^{Blundell et al., 2012}).

Thus, since lagged differences are still informative about current levels, ^{Blundell and Bond (1998)} propose to estimate a system of equations, system GMM (sys-GMM), which combines moment conditions for the model in first differences with moment conditions for the model in levels (^{Bun & Windmeijer, 2007}), each with a specific set of instrumental variables. The method uses lagged γ_it differences as instruments for the equations in levels and lagged γ_it levels as instruments for the first-difference equations (^{Angeles & Ramirez, 2014}). The sys-GMM, based on the moment condition E[ΔP_it-1(η_i+u_it)]=0, significantly improves accuracy and reduces small sample bias (^{Blundell & Bond, 2000}).

Therefore, three specification tests are employed to enable the consistency of the sys-GMM estimator to be assessed: (i) the Hansen test of overidentifying restrictions to test the joint validity of the instruments given the heteroscedastic distribution of the errors;² (ii) the difference-in-Hansen test, defined as the difference between the Hansen statistics of the differenced GMM and the sys-GMM;³ and, (iii) the Arellano-Bond autocorrelation test to determine the existence of first and second order autocorrelation, and to corroborate that [u_it-u_it-2]=0, assuring the consistency of the GMM estimator.

Given the availability of information, the study covers the years 2008, 2010, 2012, 2014, 2016, and 2018 with data from CONEVAL at the state level. Based on the review of the literature, the variables used in the estimation are classified into two groups. The first group includes five poverty indicators (POB), which show the percentage of the population living in poverty, moderate poverty, extreme poverty, income below the income poverty line, and income below the extreme income poverty line. Table 1 defines these variables and describes them statistically.

This approach to poverty differs from the usual one, which considers it the lowest quintile in a country’s income distribution (e.g., in ^{Dollar & Kraay, 2004}). It differs from studies that use GDP per capita, infant mortality rate, and household consumption, among others, as a proxy for poverty (^{Magombeyi & Odhiambo, 2017}).

Including different poverty measures makes it possible, in addition to examining the effect of FDI on poverty, to analyze in which type of poverty this effect is felt. Given the lack of consensus on the effect of FDI on poverty, this helps to clarify the estimated sign. Econometrically, this constitutes a robust analysis of the results.

Likewise, in addition to the econometric specification, the consideration of the lagged dependent variable in the model is supported by the Mexican context, characterized by the persistence of poverty as pointed out by various sources such as ^{Rodríguez (2016)}, ^{Millán (2018)}, ^{Canto (2019)}, and CONEVAL itself (2015). ^{Andrade et al. (2017)} establish that poverty is persistent, notwithstanding the economy’s capacity to generate wealth. Accordingly, as they state, it is accepted as a hypothesis that poverty tends to be self-perpetuating or, at least, to affect its own future behavior.

The second group includes the explanatory variables at the state level. PROGLO includes i) total casual insured workers, which are people insured in the Mexican Social Security Institute (IMSS) in urban and rural areas, tate; ii) productive specialization, understood as the percentage share of primary sector activities in the state GDP, ep; iii) foreign direct investment, total sum of new investments, reinvestment of profits and intercompany accounts, tiedr; and, iv) international remittances, monetary amount from abroad transferred by an individual to a beneficiary in national territory, remr.

POL captures institutional quality or governmental stability (^{Devangi et al., 2013}) and is approximated by v) perception of corruption, the percentage of the population aged 18 and over in urban areas of 100 000 inhabitants or more that consider corruption to be an important problem in their state, corr. GP, due to its link with poverty reduction, is approximated using different measures of social spending (^{Cammeraat, 2020}), such as (vi) government spending, the sum of public investment (public works and production projects and promotion measures), and transfers, allocations, subsidies and other aid, gp; (vii) participations to states from Ramo 28, resources allocated to states and municipalities under the terms established by the Fiscal Coordination Law (LCF) and the Agreements of Accession to the Fiscal Coordination System and Administrative Collaboration in Federal Fiscal Matters, rr28; viii) federal contributions for states and municipalities for Branches 25 and 33, resources to strengthen their capacity to meet demands in education, health, basic infrastructure, financial strengthening, and public safety, based on the provisions of the LC, plus the resources of the SEP in salary and economic provisions for the educational spending funds of Branch 33, rr2533; and, ix) federal spending for the Prospera Social Inclusion Program including operating costs, prospr. Finally, INFRA, because of its direct links via economic growth and in the productivity and income options of the regions (^{Setboonsarng, 2006}), is expressed as vi) road density, length of the total road network (paved, surfaced, dirt, and improved dirt roads) per thousand inhabitants, dcp.

The variables are considered realistically (2013=100) and transformed into logarithms. Coefficients with a negative sign are expected for all variables, indicating that they contribute to reducing state poverty. In addition, given the heterogeneity among states, individual effects are considered. For this purpose, a regional dummy variable, dfn, is included. This variable enables testing whether regional effects associated with attracting foreign capital show a difference between the northern border states and the rest.⁴

Results and discussion

Table 2 shows the results of the tests of the fixed effects model in the different specifications for each poverty variable. For all of them, the Breusch and Pagan Lagrange multiplier test (BPLM), which evaluates random effects, rejects the null hypothesis, enabling it to corroborate the assumption of individual effects among states. Likewise, the Hausman test yielded a meaningful result, accepting, in all cases, that the best specification is the fixed effects model. For all five specifications, the tests suggest that the data do not show cross-sectional dependence or multicollinearity. Nevertheless, there is evidence of heteroscedasticity and first-order autocorrelation.

Consequently, dynamic panel models (sys-GMM) are estimated. Regarding the consistency of the sys-GMM estimator, Table 3 shows the results of the specification tests. The p-values for the Hansen, Diff-in-Hansen, and AR (1) and AR (2) tests are reported. The tests do not reject the null hypotheses of joint validity of the instruments or validity of the additional instruments in any of the specifications. Similarly, there is no evidence of second-order autocorrelation. Thus, it is accepted that the specifications are well-defined (preferred over GMM modeling in first differences) and reflect the effects of FDI on poverty at the state level.

Moreover, the significance of the coefficients is systematically maintained, and the signs remain unchanged. Nevertheless, the effects of FDI on poverty using GMM were estimated to consider the robustness of the results. The results (Table 3) are unchanged (signs, magnitudes, and, to a lesser extent, significance). Thus, the reported results are a robust reflection of the effects of the variables on poverty.

Final thoughts

The evidence shows that the effect of FDI and the other variables on poverty is sensitive to the proxy variable used to measure the latter variable. ^{Magombeyi and Odhiambo (2017)} also estimate that the effect of FDI on poverty is sensitive to the proxy variable used to measure the latter. While the magnitudes remain in range, the significance and signs of the explanatory variables change slightly. Remittances and casual employment decrease extreme poverty to a greater extent, while productive specialization increases income poverty. Government spending reduces moderate poverty; federal participation and contributions do not show a clear pattern, but the former tends to reduce poverty and the latter to increase it. Likewise, expenditure via Prospera increases poverty, particularly extreme poverty.

Although casual employment and remittances tend to contribute mainly to reducing extreme poverty, these dimensions are manifestations of poverty. Remittances come from precarious and vulnerable migrant workers toward family members in conditions of poverty and contexts of social marginalization (^{Canales, 2007}). At the same time, temporary employment is characterized by precarious conditions (^{Cano, 2004}). The comparative lack of protection in the primary sector also reinforces poverty.

Regarding government spending, there is limited efficiency and impact on fulfilling social objectives. Whether directly, through economic support to shore up monetary income, or indirectly, through the generation of jobs that lead to higher incomes, the capacity of public resources to reduce poverty is mainly focused on moderate poverty, negatively affecting the proportion of the population living in extreme poverty. The limited effective coordination of policies and programs among governments at the three levels and the separation of the precise needs of the territories, linked to qualitative conditions, are the basis of these results and are also linked to the persistence of poverty.

Although road density contributes to reducing income poverty and moderate poverty by expanding access to markets and opportunities, corruption is a central factor in perpetuating poverty. There is also evidence that poverty is self-reinforcing, making it even more complex to combat.

Specifically, there is evidence that FDI tends to reduce moderate poverty but, at the same time, increases extreme poverty. The participation of foreign companies in the economy distorts the labor market, offering employment and income opportunities, directly and indirectly, for some, helping to escape poverty but excluding others, representing an obstacle to achieving income and employment alternatives to improve access to basic goods and services, leading them to destitution. In short, from the approach of this research, the economic model in the country is characterized by generating moderate poverty with work linked to FDI and extreme poverty without formal work, excluded from the benefits of FDI. This transnational development model focuses on improving competitiveness to attract investment, leaving aside alternatives for improving the social conditions of workers.

To change this situation, foreign capital with quality jobs and sufficient quantity to contribute to poverty reduction is required. As has been widely pointed out in the literature, one alternative is the insertion of small and medium-sized companies in the value chain of subsidiaries, moving up toward higher value stages. In any case, greater government efforts are needed to establish complementary policies and materialize the potential benefits of FDI regarding technological diffusion and adaptation, export promotion, and profit taxes, among others.