FDI, human capital and economic performance in Mexico : An ARDL cointegration and Granger causality approach

(1)

FDI, human capital and

economic performance

in Mexico

MASTER THESIS WITHIN: Economics NUMBER OF CREDITS: 30 ECTS

PROGRAMME OF STUDY: Economic Analysis AUTHOR: Tilda Fredriksson

JÖNKÖPING May 2020

(2)

Master Thesis in Economics

Title: FDI, human capital and economic performance in Mexico Author: Tilda Fredriksson

Tutor: Charlotta Mellander

Date: 2020-05-27

Key terms: Foreign direct investment, Human capital, Economic performance, ARDL

cointegration, Granger causality

Abstract

The nexus among foreign direct investment (FDI) inflows and the Mexican economic growth has been the subject of a number of recent papers. Yet, previous studies frequently overlook its relationship to human capital and consequently ignore potential interlinkages between the variables. By running an ARDL model and thereafter applying the Granger causality technique derived by Toda and Yamamoto (1995) and Dolado and Lütkepohl (1996) this paper investigates the relationship among FDI and economic performance in Mexico during 1970-2018 after incorporating human capital into the framework. When including human capital, measured as gross enrolment ratio in tertiary education, FDI inflows and real GDP per capita have an insignificant long-run relationship. However, this paper finds a Granger-causal relationship running from FDI inflows to human capital. Human capital, on the other hand, precedes real GDP per capita and the main implication is thus that FDI may not spur economic performance directly, but indirectly through its significant effect on the enrolment ratio in tertiary education. Therefore, to ignore the influence of human capital may result in deceptive conclusions regarding the Mexican FDI-growth nexus.

(3)

List of Abbreviations

ADF Augmented Dickey-Fuller

AIC Akaike Information Criteria

ARDL Autoregressive Distributed Lag

CUSUM Cumulative Sum of Recursive Residuals

CUSUMSQ CUSUM of Squares

ECM Error Correction Model

FDI Foreign Direct Investment

GASME Global Alliance on Small and Medium Enterprises GATT General Agreements of Tariffs and Trade

GDP Gross Domestic Product

GI Greenfield Investment

ISI Import Substituting Industrialisation KPSS Kwiatkowski-Phillips-Schmidt-Shin M&A Mergers & Acquisitions

MNE Multinational Enterprises

NAFTA North American Free Trade Agreement

OECD Organisation for Economic Co-operation and Development R&D Research & Development

TYDL Toda & Yamamoto and Dolado & Lütkepohl

UN United Nations

UNCTAD United Nations Conference on Trade and Development

UNESCO United Nations Educational, Scientific and Cultural Organisation

VAR Vector Autoregressive

WDI World Development Indicators

(4)

1 Introduction

_____________________________________________________________________________________

This section presents the purpose, research questions and contributions of this paper. Additionally, it shortly describes the methodology used and the main implications from the empirical results.

______________________________________________________________________ The relationship between FDI and economic performance has been both heavily researched and debated for several decades (Gupta & Singh, 2016). Consequently, studies investigating the determinants of FDI (Agarwal, 1980; Bevan & Estrin, 2004), the various cost and benefits of FDI (Dunning, 1994; Alfaro, 2017) and the direction of the effects among FDI and economic growth (Nair-Reichert & Weinhold, 2001; Hansen & Rand, 2006) have been conducted but with inconclusive results. Despite that numerous researchers find a positive causal one-way relationship running from FDI to economic growth or at least bi-directional, more than a few studies find no significant results or even negative relationships (Gupta & Singh, 2016). Additionally, the results seem to differ depending on the characteristics of the area (country) examined; developed countries tend to depict a positive significant relationship whereas the influence on emerging countries is more unconvincing and repeatedly negative or insignificant (Blomström, Lipsey & Zejan, 1992; Jyun-Yi & Chih-Chiang, 2008; Alvarado, Iniguez & Ponce, 2017).

Among the emerging countries, Mexico is one of the most open nations to FDI by an inflow of 31.6 billion US$ in 2018 with the United States, Spain and Canada being the main investors (UNCTAD, 2019). However, the research on FDI in Mexico is not unexplored. Multiple studies aim to define the main determinants of the Mexican FDI inflows (Thomas & Grosse, 2001; Jordaan, 2008; De Castro, Fernandes & Capos, 2013) and another body of literature is debating how the North American Free Trade Agreement (NAFTA) contract signed in 1994 influenced the direct investments (Cuevas, Messmacher & Werner, 2005; Waldkirch, 2010). There are also scholars examining the role of FDI inflows on Mexico´s economic performance and concur that FDI positively influences the Mexican economy (Adames, 2000; Ramírez, 2000; Zhang, 2001; Alguacil, Cuadros & Orts, 2002; Lal, 2017; Romero, 2019). However, these studies can be unified in two ways; first, they do not apply the autoregressive distributed lag (ARDL) model to address the problem but instead generally employ the test by Johansen (1988) or Engle and Granger (1987) to examine the nexus. The ARDL technique has several advantages over other cointegration tests (Pesaran, Shin &

(7)

2

Smith, 2001) and is commonly used in studies with similar focus but targeting other countries (Belloumi, 2014; Sunde, 2017; Kalai & Zghidi, 2019). Second, they tend to ignore the possible influence of human capital on both economic growth and FDI. Human capital is, both according to theory (Lucas, 1988; Romer, 1990) and more contemporary empirical research (Matousek & Tzeremes, 2019) a key driver of economic development and should therefore, at least, be included as a control variable. However, research by Borensztein, De Gregorio and Lee (1998) and more recent by Su and Liu (2016) also stress how it may exist strong relationships among the level of human capital and FDI inflows. Thus, to disregard human capital in the context of FDI and economic development may not only lead to omitted variable bias but also to ignorance of dynamic interrelationships among FDI, human capital and economic growth.

Moreover, much of the literature discussing the connection between FDI and the Mexican economic development examines the period around its large reforms in the 1980s and the NAFTA signature in the 1990s and do not assess more recent upswings. According to Waldkirch (2010), there are reasons to expect that the FDI inflows in Mexico has changed characteristics from merely market-seeking to efficiency-seeking after the NAFTA contract and by this means may influence its impact on economic performance.

Hence, the purpose of this study is to examine the relationship between FDI and economic performance in Mexico over the years 1970-2018 when including human capital into the framework. This is analysed through the following research questions:

(1) Do FDI inflows and economic performance have a positive significant relationship in Mexico over the years 1970-2018 after including human capital?

(2) Do FDI inflows and human capital have a positive significant relationship in Mexico over the time 1970-2018?

(3) Are there Granger causal relationships between FDI, human capital and economic performance in Mexico over the years 1970-2018?

By answering these questions this paper complements the already existing literature through three main contributions: (i) investigates a long time period both pre and post the major trade reforms, (ii) applies the ARDL model which possesses several advantages over the cointegration models commonly used in former studies and (iii) most importantly, extends the nexus among FDI and economic development by including human capital into the

(8)

3

multivariate framework. However, to discern accurate relationships among variables covering approximately 50 years of data of an economy which have experienced large reforms and restructurings is unlikely (if not impossible) and thus, this paper instead intends to focus on the overall consequences of incorporating human capital into the context. Nonetheless, the findings of this paper may be of interest, not only to Mexican policymakers but to decision-makers in other emerging countries open to FDI.

By running an ARDL model and thereafter applying the Granger causality technique derived by Toda and Yamamoto (1995) and Dolado and Lütkepohl (1996) this paper investigates the relationship among FDI and economic performance in Mexico over the years 1970-2018 after incorporating human capital into the framework. When including human capital, FDI inflows and real GDP per capita have a statistically insignificant long-run relationship. However, this paper finds a Granger causal relationship running from FDI inflows to human capital. Human capital, on the other hand, precedes real GDP per capita and the main implication is thus that FDI may not spur economic performance directly, but indirectly through its significant impact on human capital.

The remaining of this paper is structured as follows: Section 2, Background – briefly informs about the trade reforms undergone by Mexico and presents the Mexican FDI inflows in numbers and figures. Section 3, Theoretical Framework - introduces theories regarding FDI, human capital and economic growth in the host country as well as the home country. Additionally, empirical research on the topic is summarised. Section 4, Data & Methodology - presents the data and methodology used to answer the research questions. Moreover, diagnostic tests are touched upon. Section 5, Empirical Results & Analysis – presents and discusses the empirical results from the ARDL models and the Granger causality tests. It furthermore, interlinks the findings to previous research. Section 6, Conclusion – concludes, gives recommendations to further research and discusses limitations in this paper.

(9)

4

2 Background

_____________________________________________________________________________________

This section briefly informs about the trade reforms undergone by Mexico and presents the Mexican FDI inflows in numbers and figures. Additionally, it touches upon the division of FDI among sectors.

______________________________________________________________________ In the 1940s the import substituting industrialisation (ISI) became the official development strategy in Mexico and was not relaxed until the late 1970s and early 1980s. Its central objective was to lessen foreign reliance and instead create internal markets in an attempt to increase domestic capability and economic development (Jordaan, 2008). In 1973 the government implemented the directive called the “Law to promote Mexican investment and regulate foreign investment” and thereby the government controlled large industries like the railroad constructions and oil drillings (The World Bank, 1996). Initially, the decree was successful with improved infrastructure, enlarged industries, large oil discoveries and high gross domestic product (GDP) growth rates, however in the beginning of 1980 the conditions worsened. Oil prices fell and interest rates rose, causing the financial situation to become deficient and disputed (Lal, 2017). Thus, the development strategy was changed and in 1986 Mexico joined the General Agreements of Tariffs and Trade (GATT), the NAFTA in 1994 and the World Trade Organisation (WTO) in 1995 (WTO, 2020). Today Mexico is one of the countries with the most free trade agreements and consequently its FDI inflows have increased1_{substantially since the 1990s as seen in Figure 1. Also, from this figure one}

can discern three major falls in the FDI inflows, in which the first one is the drop during 2001 and 2003. Decreased economic growth and privatisation, as well as an increased interest

1_{It is not evident that FDI inflows increase with trade agreements, since such contracts may excise one of the}

original benefits with direct investments - trade barriers.

0 5 10 15 20 25 30 35 40 45 50 Billio n US $

Net inflows Net outflows

Source: Computed by the author using data from the WDI (2020a)

(10)

5

in low-cost countries (e.g Asian economies), caused FDI inflows in Latin American countries to decrease (UN, 2004). Second, the global financial crisis in 2008 influenced the FDI inflows as well. However, the countries with the largest relative withdrawals during this time was transacted by the nations with the smallest absolute investments (i.e. Sweden, Italy and Germany). Since FDI is a long-term investment it commonly requires loads of physical capital and other fixed assets making withdrawals for large investors (e.g. the United States) during slowdowns more inefficient than to sustain investments (GASME, 2016). Lastly, in 2014 the value dropped to around 32 billion from 47 billion US$ a year earlier. However, the record value in 2013 is mainly a result of a Belgium acquisition of the Mexican beer company Grupo Modelo counted to over 13 billion US$ (UN, 2015) and thus the sharp decline in 2014 is rather a return to a somewhat more regular level. Besides, from Figure 1 it is also clear that Mexico is more inward-oriented in terms of direct investments than outward-oriented.

Moreover, in 2018 Mexico´s total stock of FDI counted to approximately 480 billion US$ which translates to almost 40% of its GDP (UNCTAD, 2019). Appendix A illustrates that Mexico´s annual FDI inflows as a percentage of GDP was at the lowest in 1987 with 0.80% and had its peak in 2001 with almost 4%. After 2001 Mexico has a mean of approximately 2.70% of GDP, which is slightly lower than the world average of 2.90% and the Latin American and Caribbean mean of 3.11% during the same time span.

Table 1, Mexico´s FDI net inflows by industry in million US$

Year Mining Agriculture Manufacturing Electricity Water Construction Finance Services

2005 275 10 13594 358 5.6 369 2069 11400 2006 430 -2 11185 -57 -6.5 597 4240 8977 2007 1832 79 13816 293 0.2 2371 5963 14095 2008 4516 61 9135 491 14.7 895 7158 14375 2009 1385 22 6835 62 15.3 688 3076 8959 2010 1541 100 14353 642 -3.2 408 2387 10117 2011 915 127 11718 -83 52.1 1678 2644 11163 2012 3013 145 9708 1217 25 1660 -2522 6007 2013 5745 211 31354 984 29.2 1073 -426 8819 2014 2679 178 18758 695 6 886 4823 7060 2015 1582 169 17704 1081 -10.8 2028 3466 13095 2016 919 89 17862 1294 7.7 1013 3499 9459 2017 1364 138 15353 2153 10.4 3020 2412 11559 2018 1506 50 16318 4957 -6.9 1531 2122 9261

(11)

6

Table 1 depicts that the manufacturing industry has received the largest amount of FDI with a peak in 2013 due to the Grupo Modelo deal. Latin America, and especially Mexico, is known for their maquiladoras – almost entirely duty- and tariff free factories using raw materials to process and thereafter export products. The maquiladoras receive the predominant part of the FDI inflows in the manufacturing sector (Jordaan & Rodriguez-Oreggia, 2012). The service sector is the second largest recipient and was the largest receiver during the financial crisis. Moreover, the finance and insurance sector, as well as the mining and quarrying industry, have large variations, whereas the remaining industries depict less deviations2_. _{Due to the still strong nationalism in Mexico (mainly a result of the law}

implemented in 1973) sectors such as the oil industry, infrastructure (especially the railroad construction) and the energy production receive little FDI (GASME, 2016).

Mexico is nowadays an export-oriented country which is enabled due to its many free trade agreements. Most important is the NAFTA membership which has contributed to that the U.S. is, by far, Mexico´s largest FDI investor and also trading partner. The U.S. investments primarily go to the manufacturing industry which, notably, also is the industry with the most exports to the U.S. (GASME, 2016). Actually, after Mexico pursued the NAFTA contract the characteristics of the FDI inflows has changed to more vertically integrated and accordingly, the investors seem to utilise the comparative advantages in Mexico (Waldkirch, 2010). However, the second largest FDI investor has for many years been Spain in which much of its investments are going to the finance and insurance sector due to the presence of several Spanish banks in Mexico. Other large continual investors are Canada, Germany and Japan in which the former mainly invests in the electricity and gas sector and the two latter in the manufacturing industry (GASME, 2016).

Concisely, the FDI inflows have experienced a rapid increase after the NAFTA contract in 1994, with the major part of the investments going to the manufacturing maquiladoras. Today, Mexico is one of the most open nations to FDI inflows when considering emerging countries, which is a large transformation from the ISI strategy in the 1970s. This suggests that the period under study covers large political reforms and consequently needs to be analysed with caution.

2_{The data is presented on a net basis and therefore the net inflows can be negative due to, for instance, reverse}

investment (a direct investment firm has acquired a financial claim on its direct investor) or disinvestment (liquidating or selling assets) (The World Bank, 2020).

(12)

7

3 Theoretical Framework

_____________________________________________________________________________________

This chapter introduces theories regarding FDI, human capital and economic growth in the host country as well as the home country. Additionally, more recent empirical research on the topic is summarised.

______________________________________________________________________

3.1 Definition of FDI

FDI is “the category of international investment that reflects the objective of a resident entity in one economy to obtain a lasting interest in an enterprise resident in another economy” (OECD, 2001, p.1). This definition puts no minimum level of ownership, whereas the World Development Indicators (WDI) (2020a) and the United Nations Conference on Trade and Development (UNCTAD) (2000) state that the foreign owner must possess 10% or more of the ordinary shares with voting rights to be classified as FDI. Additionally, FDI can take a number of forms and be conducted due to several reasons. A horizontal investment implies that the investor establishes the same type of business in the foreign country as it has operating in the domestic country and can therefore be seen as a substitute for exports. This type of investment is also known as market-seeking FDI because it generally aims to exploit new markets (Markusen & Venables, 2000). Instead, vertical FDI refers to establishments related to a company´s main business but are generally invested in firms operating in other stages in the production chain. This investment is also known as efficiency-seeking FDI since it seeks to reduce production costs (Helpman & Krugman, 1985). An agent can also invest in a business entirely unrelated to its current operations - conglomerate FDI, but since this often requires the entering of a new industry it usually takes the form of a joint venture with a firm already operating in the country and industry of interest (Herger & McCorriston, 2014). Some authors also distinguish mergers and acquisitions (M&As) from greenfield investment (GI), in which the former means a purchase of an already existing business and the latter usually implies that the parent company builds a subsidiary in the foreign country (UNCTAD, 2000).

However, since the central mission of this paper is not to address the effects of different types of direct investments, FDI will generally not be separated, neither in the theoretical framework nor in the econometric regressions.

(13)

8

3.2 FDI, human capital and economic growth in the host country

3.2.1 Linking FDI to economic growth

The neoclassical growth model by Solow (1956) is one of the first to mathematically explain the determinants of economic growth, highlighting the importance of labour and capital accumulation. In Solow´s model, the savings rate together with the rate of capital depreciation (i.e. the capital accumulation) is the short-run driver of growth, in which the capital stock will grow if the savings rate is above the required investment. Thus, FDI can help the host country boost its capital stock through new equipment and machinery and thereby economic performance. However, since the Solow model assumes diminishing returns to capital, the economy will always converge to the same steady-state growth rate and therefore FDI can, at best, spur economic growth in the short-run under this setting.

Nevertheless, a year later Solow (1957) asserts how technological progress is the key determinant of economic long-run growth and a similar conclusion is derived in the work by Romer (1990). However, Romer internalises the underlying mechanisms to the technological improvements which, in Solow´s model, are exogenous and is therefore referred to as the endogenous growth model. Romer´s model combines externalities from technology improvements with imperfect markets and collectively they give rise to increasing returns to scale. Hence, Romer argues that technology advancement (i.e. new types of capital3_{) is the}

main fuel to economic performance. This does not only bring motivation for the home country to exploit their business across borders, but simultaneously FDI from a developed home country can spur the economic progress in the host country through spillovers4_{. This}

can regard spillovers related to new physical technologies and innovations or improved human capital. In other words, FDI can enhance economic performance in the host country through for instance R&D activities, new technology, productivity, know-how and spillovers related to these (Nair-Reichert & Weinhold, 2001), meaning that both direct and indirect effects can be derived from the endogenous growth model. Moreover, some argue that FDI can increase competition and thereby the search for technological improvements, which consequently spur economic growth (Grossman & Helpman, 1994).

3_{While Romer (1990) stresses the importance of new types of capital as a source to economic growth, Aghion}

and Howitt (1992) highlight that improvements in already existing products are the main drivers.

4_{Despite that many of the researchers find positive outcomes from FDI inflows, some find negative effects in}

terms of crowding out (Agosin & Machado, 2005), negative wage spillovers (Harrison & Aitken, 1999) and income inequality (Choi, 2006).

(14)

9

3.2.2 Linking human capital to economic growth

Continuing with the endogenous growth model by Romer (1990), it also highlights that human capital is the key determinant to the growth of technological progress and hence economic development. His model derives how sustained economic growth is a function of the productivity of research, the level of human capital and the fraction of human capital that is devoted to research; the mark-up incentivises individuals to create innovations whereas wages in production cause people to produce consumption goods.

Concisely, the endogenous growth model explains how investments in primarily human capital, innovation and knowledge generate economic growth. Thus, policy measures related to human capital and innovations can enhance the economic performance within a nation. Despite that the model by Romer (1990) generally serves as a basis for the endogenous growth theory, the idea that human capital is a key determinant of growth is also emphasised in the work by Lucas (1988) and Rebelo (1991).

3.2.3 Absorptive capacity

Based on findings by Nelson and Phelps (1966) and Banhabib and Spiegel (1994), Borensztein et al. (1998) bring forward a theoretical model explaining how FDI is a main contributor to successful technology transfers, but only if a certain minimum level of human capital exists in the host country. Put differently, the stock of human capital can limit the technical diffusion and simultaneously also the effect of FDI. Since direct investments, ideally, transfer improved technology and knowledge spillovers from the home country to the host country it also requires the receiving country to capture and assimilate these positive effects. Borensztein et al. (1998) regard this assimilation as the absorptive capacity of the economy and emphasise how it is a function of the human capital stock. The basic intuition is that (developed) home countries transfer advanced technologies and skills, and without a certain level of education, the (emerging) receiving nation will not be able to preserve and utilise these innovations due to poor experience and know-how.

The theoretical link among FDI, human capital and economic growth in the host country can be summarised through three main points. First, FDI can enhance economic development through improved technology and knowledge transferred from the home country (Solow, 1957; Romer, 1990). Second, human capital is a main determinant to economic performance not only directly (Lucas, 1988), but is also a key element to

(15)

10

technology advancements and innovations (Romer, 1990). Lastly, Borensztein et al. (1998) argue that human capital does not only influence economic growth individually but is also a requirement for FDI to be effective. In other words, a certain threshold level of absorptive capacity seems to be required for FDI inflows to benefit the host country.

3.3 FDI, human capital and economic growth in the home country

The study by Hymer (1970) discerns portfolio investments from investments associated with some type of control over the business, i.e. FDI. Previously, FDI was merged with financial assets and thus simply being considered capital investments. His main hypothesis asserts that a market setting with perfect competition is insufficient for engagement in FDI to be beneficial. Instead, he proposes a version with market imperfections in which firms can gain from industry-specific advantages in foreign countries; engagement in FDI means having some control over the production process and consequently also the possibility to utilise inputs such as cheap labour, inexpensive material (efficiency-seeking FDI), less competition, human capital, new brands (asset-seeking), new markets (market-seeking) and natural resources (resource-seeking). Likewise, the research by Buckley and Casson (1976) stresses the importance of industry-related or even firm-specific investment characteristics as determinants of FDI engagements, rather than aggregated country-level aspects as commonly reviewed previously.

The work by Hymer (1970) and Buckley and Casson (1976) was later embraced by Dunning (1979, 1980) in the theory referred to as the OLI (Ownership, Location and Internalisation) framework, also known as the eclectic paradigm. This theory relies on three main points, in which the first regards the ownership advantage. The greater the (specific) competitive advantage in the home country, the more likely the firm is to invest abroad, i.e. engage in FDI. Second, Dunning emphasises that localisation can have a significant influence on the level of engagement in FDI, considering both economic, social, cultural and political factors. Lastly, he also highlights the importance of internalisation advantages for firms to engage in FDI. This means that firms must believe that their ownership advantage is best utilised internally rather than through subcontracts like licensing or joint ventures. To conclude, since engagement in FDI appears to depend on the firm-specific utilisation of various inputs, the level of human capital or the economic performance in a host country can be significant determinants of FDI inflows (Hymer, 1970; Buckley & Casson, 1976; Dunning, 1979, 1980).

(16)

11 3.4 Empirical research

The theoretical idea that FDI can fuel economic growth (Romer, 1990) has received inconclusive empirical results. On the one hand, some argue that the effect of FDI depends on the development level of a nation and confirm that the higher the income level, the larger the economic gains from FDI inflows (Blomström et al., 1992; Jyun-Yi & Chih-Chiang, 2008; Alvarado et al., 2017). On the other hand, other scholars find no clear evidence that the income level influences the nexus among FDI and growth, but rather the degree of (trade) openness (de Mello, 1997; Nair‐Reichert & Weinhold, 2001; Iamsiraroj & Ulubaşoğlu, 2015), human capital (Borensztein et al., 1998; Blomström & Kokko, 2002; Ramzan, Sheng, Fatima & Jiao, 2019), infrastructure (Nourzad, Greenwold & Yang, 2014) and financial markets (Alfaro, Chanda, Kalemli-Ozcan & Sayek, 2004; Villegas-Sanchez, 2009).

Concerning Mexico, Ramírez (2000) examines the relationship between FDI and the aggregate labour productivity during 1960-1995 and concludes that the stock of FDI has a positive significant effect on labour productivity growth. Similarly, Adames (2000) concludes that there exists a positive long-run relationship between FDI inflows and real GDP in Mexico during 1971-1995. However, Zhang (2001) emphasises how former literature mostly overlook the direction of the effects among FDI and economic growth and thus ignore potential issues associated with endogeneity. Hence, his study estimates the Granger causality between the stock of FDI and economic growth in 11 developing economies and concludes bi-directional Granger causality in the case of Mexico. Likewise, Alguacil et al. (2002) examine the direction among exports, FDI inflows and economic performance during 1980-1999 and argue that FDI likely is an important contributor to Mexico´s export-led growth. In a later study, they also highlight how FDI precedes national savings and thus influences the savings-growth nexus in Mexico (Alguacil, Cuadros & Orts, 2004). Another, more recent, trivariate relationship is assessed by Lal (2017) who estimates the linkages between FDI, trade openness and economic growth in Mexico, India and China during 1980-2011. He finds that the FDI stock Granger causes both trade openness and GDP in Mexico. Romero (2019) studies the relation among foreign capital and labour productivity during 1983-2018 and find positive long-run relationships but no short-run interactions and thus, at least partially, agrees with the conclusion by Ramírez (2000). Hence, despite being classified as an emerging economy, researchers tend to agree that the Mexican performance gains from FDI inflows at the aggregate level.

(17)

12

However, heretofore the mentioned studies do not consider the effect of human capital, not on economic performance nor on FDI and might therefore suffer from omitted variable bias; researchers assert that there exists a relationship among FDI and human capital. Some argue that FDI inflows may not be economically beneficial for the host country if not a certain threshold level of education is fulfilled (Li & Liu, 2005; Ramzan et al., 2019) and thus confirms the theory presented by Borensztein et al. (1998). Other suggest how it exists an inverse U-shaped nexus among the attraction of FDI and the level of human capital in the host country and thus argue that economies with poor human capital standards receive less FDI inflows than nations with more well-developed educational systems (Zhang & Markusen, 1999; Akin & Vlad, 2011). Thus, since human capital is a key determinant to economic growth and simultaneously seems to be correlated with FDI inflows, the exclusion of such a variable in a regression analysis match the conditions for omitted variable bias.

Nonetheless, there are studies investigating the linkages between FDI and human capital in Mexico. For instance, Jordaan (2008) concludes that average years of schooling is a key determinant to regional FDI inflows during 1989-2006. Similarly, Osuna (2016) estimates the effect of various educational degrees on FDI inflows and conclude that there exists a threshold level of tertiary education that needs to be fulfilled to attract FDI inflows. Hence, these studies consider the connection between FDI inflows and human capital, but does not reflect on their relation to economic performance. Their nexus to the Mexican economic growth is, however, analysed by Oladipo (2007) at the aggregate level and Jordaan and Rodriguez-Oreggia (2012) at the regional level. The former uses school enrolment in secondary education to proxy human capital and concludes that FDI and human capital are key determinants of the Mexican economic growth. The latter, on the other hand, use mean years of schooling but find negative insignificant results to economic growth; however, conclude that FDI is a positive significant contributor to the economic performance. Yet, none of these studies examines the simultaneous relationship among FDI, human capital and economic development and thereby ignore potential interrelationships between the variables. This trivariate framework has been studied on other emerging economies, e.g. Thailand (Tanna & Topaiboul, 2005) and Turkey (Ozyigit & Eminer, 2011) in which the former proxy human capital through the mean years of schooling and conclude that FDI precedes human capital but not GDP per capita. The latter, on the other hand, use the number of university graduates to represent human capital and find that human capital Granger causes FDI inflows. However, they find no evidence of FDI-led growth.

(18)

13 3.5 Hypotheses

As mentioned, previous studies examining the relationship between FDI and economic growth in Mexico tend to ignore the potential influence of human capital. Thus, the first research question aims to examine if FDI inflows and economic performance have a positive relationship in Mexico over the years 1970-2018 after including human capital. The corresponding hypothesis is presented in Hypothesis 1.

▪ FDI inflows have a positive relationship to real GDP per capita after controlling for the gross

school enrolment in tertiary education during 1970-2018

(Hypothesis 1) The second research question investigates if FDI inflows and human capital has a significant relationship in Mexico over the years 1970-2018, with the following hypothesis:

▪ The gross school enrolment in tertiary education has a positive relationship to FDI inflows during

1970-2018

(Hypothesis 2) These hypotheses are tested by the use of an ARDL cointegration model, presented in

section 4.3. However, as the causation among the variables is far from obvious, the direction of the relationships between FDI, human capital and economic performance is tested with the following hypotheses:

▪ FDI inflows Granger causes real GDP per capita during 1970-2018

(Hypothesis 3a)

▪ The gross school enrolment in tertiary education Granger causes real GDP per capita during

1970-2018

(Hypothesis 3b) ▪ The gross school enrolment in tertiary education Granger causes FDI inflows during 1970-2018

(Hypothesis 3c)

To examine these hypotheses, the technique developed by Toda and Yamamoto (1995) and Dolado and Lütkepohl (1996) is applied and described in detail in section 4.4. Figure 2 depicts a visual presentation of Hypothesis 3a, 3b and 3c.

GDP

FDI

_HC

(19)

14

4 Data & Methodology

_____________________________________________________________________________________

This section presents the data, econometric model and methodology used to answer the research questions. Moreover, diagnostic tests are touched upon.

______________________________________________________________________

4.1 Data

The data of all the variables are retrieved from the WDI by the World Bank and consists of annual data for the years 1970 to 2018. Furthermore, Table 2 summarises the abbreviations, short definitions, transformations and expected sign of the variables used in the regressions. Table 2, Brief variable description

Variable Abbreviation Definition Transformation Expected sign

Economic

performance GDP

Real GDP per capita (constant local currency)

Taking the natural

logarithm +

Foreign direct

investment FDI FDI, net inflows (% of GDP)

Taking the natural

logarithm

+/-Human capital HC Gross school enrolment in

tertiary education (%)

Taking the natural

logarithm +

Capital formation CF Gross capital formation (% of

GDP)

Taking the natural

logarithm +

Notes: Depending on the regression model estimated, FDI and economic performance are treated either as dependent or explanatory variables (see Model 1a and 2a for clarification) and thus their expected signs represent the expected outcome when being used as regressors.

Economic performance is measured by the GDP per capita in constant local5_currency.

GDP is calculated as the annual sum of gross value added by domestic producers, adding the product taxes and subtracting subsidies not considered in the value of production of the goods and services. The GDP is thereafter divided by the midyear population (WDI, 2020b). This the dependent variable in Hypothesis 1, 3a and 3b.

Foreign direct investment represents the direct investment equity inflows (equity capital,

earning reinvestments and other capital) in the host country as a share of GDP. To be classified as a direct investment, the foreign owner must possess at least 10 % or more of the ordinary shares with voting rights (WDI, 2020c). This is one of the main variables of interest

5_{I.e. Mexican pesos. The average closing price for the U.S. Dollar Peso exchange rate during 2018 was 19.22}

(20)

15

and both according to growth theories (Solow, 1957; Romer, 1990) and empirical literature (Li & Liu, 2005; Lal, 2017) FDI can influence the economic performance significantly. However, although previous researchers generally find positive interactions among FDI and economic growth in Mexico, the expected sign of the FDI variable is unknown due to two reasons: first, FDI may depend on the level of human capital (Borensztein et al., 1998) and second, some studies conclude insignificant or even negative effects from FDI in emerging countries (Hye, 2011; Alvarado et al., 2017). This is the dependent variable in Hypothesis 2 and 3c.

Human capital is measured as the gross6_{enrolment ratio in tertiary education as a share of}

the population age group that officially corresponds to tertiary education (WDI, 2020d). Other proxies for human capital could be literacy rates, mean years of schooling or investments in human capital. According to Borensztein et al. (1998) a minimum level of human capital is required for FDI inflows to be beneficial. This, in combination with poor data availability on other proxies, make the school enrolment in higher education suitable for the purpose. However, preferably, the graduation rates in tertiary education would be used, but due to data limitations the enrolment ratio is applied instead. The importance of human capital in stimulating economic growth is clear from the theories presented by Lucas (1988) and Romer (1990), however also highlighted by Barro (2001). Previous research, as presented earlier, frequently ignores the effect of human capital and thus, the inclusion of this variable is one of the main contributions of this paper.

Capital formation is represented by the gross capital formation, as a share of GDP. It

represents the additional expenditures to the fixed assets (e.g. land advancements, plant and equipment acquisitions) together with net changes in the firm inventory level (WDI, 2020e). Generally, a high capital formation can enable faster economic growth, indicating that capital formation is a main determinant of economic development. This variable is frequently used as a control variable in research with similar research questions (Belloumi, 2014; Kalai & Zghidi, 2019) and thus included as variable in this paper.

Lastly, using a similar approach as Pacheco-López (2005), a dummy variable named DNAFTA

is included to capture some of the large trade reforms and political changes experienced in the middle of the 1990s. It takes a value of zero (DNAFTA=0) for the years before the NAFTA

contract (1970-1993) and equals one (DNAFTA=1) the remaining years.

(21)

16 4.2 Descriptive statistics

As depicted in Table 3 the annual mean of real GDP per capita is approximately 116,953 pesos with a maximum value of 147,102 in 2018 and a minimum of 78,064 in 1970. This variable has the lowest (relative) deviation from its mean whereas the largest one belongs to FDI inflows with a mean of 1.872% of GDP and a standard deviation of 0.972%. The human capital variable had its peak in 2018 with a gross enrolment ratio in tertiary education of 42%. Table 3, Descriptive statistics over 1970-2018

Unit N Mean Median Maximum Minimum Std. Dev

GDP Real GDP per _capita 49 116,953 116,629 147,012 78,064 18,216

FDI % of GDP 49 1.872 1.785 3.972 0.395 0.972

HC Enrolment _ratio 49 19.016 15.798 42.094 5.288 8.910

CF % of GDP 49 22.425 22.793 25.947 17.834 1.746

Notes: The variables in this table are not transformed, i.e. no natural logs are taken on the data.

One of the major advantages of the ARDL technique is its usefulness when dealing with non-stationary data (see a thorough discussion in section 4.3). However, since such data commonly follows some type of trend it likely brings high correlation among the variables and thus issues with multicollinearity7_{. To avoid this, I have reduced the number of}

explanatory variables and includes only capital formation8_{as a control variable. On the other}

hand, as the intuition with control variables is to reduce problems with omitted variable bias, such as inconsistent and biased estimators, there is a trade-off between multicollinearity and ignored variables. However, as the structure of the ARDL model (the inclusion of lags) is a good remedy for omitted variable bias (Ghouse, Khan & Rehman, 2018) the issues associated with multicollinearity are regarded as more problematic in this specific context and thus the number of explanatory variables used is decreased. The correlation matrix in Table 4 informs that the variable associated with capital formation has the lowest correlation with all the remaining variables whereas the human capital depicts the highest correlation9_.

7_{Imprecise and unreliable estimators.}

8_{Other frequently used control variables are employment and trade openness. However, due to insufficient}

data on employment this variable cannot be included. Trade openness depicts a very high correlation to all the remaining variables and is therefore excluded from the models to ease problems with multicollinearity.

9_{The partial correlation, after controlling for time, is presented in Appendix B and depicts significantly lower}

correlation among the variables, signalling that the time trend likely is the underlying reason to the (remarkably) high correlation.

(22)

17

Table 4, Correlation matrix

GDP FDI HC CF

GDP 1

FDI 0.823*** 1

HC 0.971*** 0.821*** 1

CF 0.224* 0.001 0.132 1

Notes: (*), (**) and (***) indicate statistically significant at the 10%, 5% and 1% significance level respectively.

4.3 Autoregressive distributed lag model (ARDL), Bounds test

To test Hypothesis 1 and 2 this paper applies the ARDL technique, presented in Model 1a and 2a respectively. This method has become popular in the context of FDI and economic growth during recent years and has four major advantages over other cointegration techniques. First is its suitability regardless of whether the variables are stationary in their level form, I(0), after taking the first difference, I(1), or if the variables are partly cointegrated (Pesaran & Smith, 1998). Thus, the issue of different orders of integration frequently encountered in similar contexts can be disregarded. Second, the short-run coefficients, obtained from the unrestricted error correction model (ECM), can easily be extracted from the ARDL model. Third, if the sample size estimated is small, the technique is especially useful compared to other cointegration tests (Pesaran et al., 2001). Lastly, as seen in Model 1a, the ARDL does not employ a system of equations to determine the long-run relationships, but instead only uses a single equation set-up. The null hypothesis that λ1=λ2=λ3=λ4=0 (no cointegration) is tested through an F-test. However, since the

distribution of the test is non-standard and simultaneously depend on a nuisance parameter, no exact critical values are available. Instead, the model presents critical values for both the case in which all variables are assumed I(0) and a further one assuming all are I(1) and thence the name bounds test. If the F-statistic exceeds the upper critical value, I(1), the null hypothesis is rejected, whereas if it is lower than the critical value corresponding to I(0) it cannot be rejected. If the value falls between the two extremes the test is inconclusive (Pesaran et al., 2001).

∆𝐺𝐷𝑃𝑡= 𝛽0+ ∑𝑝𝑖=1𝛽1𝑖∆𝐺𝐷𝑃𝑡−𝑖+ ∑𝑝𝑖=1𝛽2𝑖∆𝐹𝐷𝐼𝑡−𝑖+ ∑𝑝𝑖=1𝛽3𝑖∆𝐻𝐶𝑡−𝑖+

∑𝑝𝑖=1𝛽4𝑖∆𝐶𝐹𝑡−𝑖+ 𝜆1𝐺𝐷𝑃𝑡−1+ 𝜆2𝐹𝐷𝐼𝑡−1+ 𝜆3𝐻𝐶𝑡−1+ 𝜆4𝐶𝐹𝑡−1+ 𝜆5𝐷𝑁𝐴𝐹𝑇𝐴𝑡+ 𝜆6𝐷1982𝑡+ 𝜆7𝐷1996𝑡+ ℇ𝑡

(23)

18

In which Σ represents the dynamics for the error correction, ∆ is the difference operator, λ signifies the long-run relationship and D1982 and D1996 are two dummy variables capturing

structural breaks in year 1982 and 1996 respectively (see discussion on page 20). However, to analyse how the relationship between FDI inflows and real GDP per capita changes when including human capital into the framework, a model dropping the human capital variable (see Model 1b) is run as well. Lastly, to examine to which extent the control variables influence the impact of FDI inflows on economic performance, a model including only FDI as an explanatory variable (see Model 1c) is examined.

∆𝐺𝐷𝑃𝑡= 𝛽𝑜+ ∑𝑝𝑖=1𝛽1𝑖∆𝐺𝐷𝑃𝑡−𝑖+ ∑𝑝𝑖=1𝛽2𝑖∆𝐹𝐷𝐼𝑡−𝑖+ ∑𝑝𝑖=1𝛽3𝑖∆𝐶𝐹𝑡−𝑖+ +𝜆1𝐺𝐷𝑃𝑡−1+ 𝜆2𝐹𝐷𝐼𝑡−1+ 𝜆3𝐶𝐹𝑡−1+ 𝜆4𝐷𝑁𝐴𝐹𝑇𝐴_𝑡+ 𝜆5𝐷1982_𝑡+ 𝜆6𝐷1996_𝑡+ ℇ𝑡 ∆𝐺𝐷𝑃𝑡= 𝛽𝑜+ ∑ 𝛽1𝑖 𝑝 𝑖=1 ∆𝐺𝐷𝑃𝑡−𝑖+ ∑ 𝛽2𝑖∆𝐹𝐷𝐼𝑡−𝑖+ 𝜆1𝐺𝐷𝑃𝑡−1+ 𝑝 𝑖=1 𝜆2𝐹𝐷𝐼𝑡−1+ 𝜆3𝐷𝑁𝐴𝐹𝑇𝐴𝑡+ 𝜆4𝐷1982𝑡+ 𝜆5𝐷1996𝑡+ ℇ𝑡

To test Hypothesis 2 this paper uses a similar approach as Kalai and Zghidi (2019) who treats

FDI as the dependent variable. The corresponding model is presented in Model 2a.

∆𝐹𝐷𝐼𝑡= 𝛼𝑜+ ∑𝑝𝑖=1𝛼1𝑖∆𝐹𝐷𝐼𝑡−𝑖+ ∑𝑝𝑖=1𝛼2𝑖∆𝐻𝐶𝑡−𝑖+ ∑𝑝𝑖=1𝛼3𝑖∆𝐺𝐷𝑃𝑡−𝑖+

𝜔1𝐹𝐷𝐼𝑡−1+ 𝜔2𝐻𝐶𝑡−1+ 𝜔3𝐺𝐷𝑃𝑡−1+ 𝜔4𝐷𝑁𝐴𝐹𝑇𝐴𝑡+ 𝜔5𝐷1982𝑡+ 𝜔6𝐷1996𝑡+ ℇ𝑡

Further, the error correction form of Model 1a and 2a is presented in Model 1aa and 2aa respectively. The error correction term (represented by ECT) indicates the time required for the variables to return to their long-run value after a shock, also known as the speed of adjustment. ∆𝐺𝐷𝑃𝑡 =𝜃𝑜+ ∑ 𝜃1𝑖∆𝐺𝐷𝑃𝑡−𝑖+ ∑𝑖=1𝑝 𝜃2𝑖∆𝐹𝐷𝐼𝑡−𝑖+ ∑𝑝𝑖=1𝜃3𝑖∆𝐻𝐶𝑡−𝑖+ 𝑝 𝑖=1 ∑𝑝𝑖=1𝜃4𝑖∆𝐶𝐹𝑡−𝑖+𝜃5𝐷𝑁𝐴𝐹𝑇𝐴+𝜃6𝐷1982𝑡+ 𝜃7𝐷1996𝑡+ 𝛾1𝐸𝐶𝑇𝑡−1+ ℇ𝑡 ∆𝐹𝐷𝐼𝑡=𝛿𝑜+ ∑ 𝛿1𝑖∆𝐹𝐷𝐼𝑡−𝑖+ ∑𝑖=1𝑝 𝛿2𝑖∆𝐻𝐶𝑡−𝑖+ ∑𝑝𝑖=1𝛿3𝑖∆𝐺𝐷𝑃𝑡−𝑖+ 𝑝 𝑖=1 +𝛿4𝐷𝑁𝐴𝐹𝑇𝐴+𝛿5𝐷1982𝑡+ 𝛿6𝐷1996𝑡+ 𝜇1𝐸𝐶𝑇𝑡−1+ ℇ𝑡

Nevertheless, the ARDL technique also has some drawbacks, with the largest one being that it cannot be used if any of the variables are I(2) since this leads to spurious regressions estimates (Pesaran & Shin, 1999). To consider this, unit root tests are conducted based on (1b)

(1c)

(2a)

(2aa) (1aa)

(24)

19

the approach presented by Elder and Kennedy (2001). Following their strategy, the variables estimated seem to follow either a random walk, random walk with drift or a trend stationary process since they are all expected to grow or remain constant over time. To test for the type of process, the Augmented Dickey-Fuller (ADF) and the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) tests are applied and presented in Table 5.

The ADF test reveals different results depending on whether only an intercept, or both a trend and intercept are included10_{. However, importantly, both tests show that all variables}

are I(0) or I(1) at the 5% significance level and thus not I(2), which is a requirement for the ARDL model to work properly. The variables associated with GDP, FDI and HC seem I(1) whereas capital formation appears to be stationary in its level form and thus I(0), which also is supported by the KPSS test. Since the variables have different orders of integration, the ARDL cointegration model is especially useful compared to other conventional cointegration techniques.

Table 5, Unit root tests

H ₀ : Unit rooot Augmented Dickey-Fuller

Intercept

Level ∆GDP Level ∆FDI Level ∆HC Level ∆CF

t-stat. -2.106 -3.564** -1.520 -8.683*** -1.447 -3.106** -3.713*** -4.100***

Intercept & Trend

t-stat. -3.396* -5.612*** -4.668*** -8.623*** -2.665 -3.543** -3.734** -4.054**

H₀: No unit root Kwiatkowski-Phillips-Schmidt-Shin

Intercept

LM-stat. 0.865*** 0.218 0.811*** 0.053 0.884*** 0.208 0.093 0.085

Intercept & Trend

LM-stat. 0.104 0.075 0.152** 0.054 0.147** 0.075 0.049 0.080

Notes: Observe that the null hypotheses are different for the two tests. Moreover (*), (**) and (***) indicate statistically significant at the 10%, 5% and 1% significance level respectively.

10_{From Appendix C one can discern that only the human capital variable appears to follow an increasing time}

trend, whereas the remaining variables seem to be rather constant over time. This suggests that the unit root test with both an intercept and trend is the most reliable regarding the human capital variable and the test using only the intercept for the remaining variables.

(25)

20

However, a further shortcoming of the ARDL model is its sensitivity to structural breaks. Therefore, Bai-Perron´s test for multiple breakpoints is applied and presented in Table 6. Since the F-statistics for the GDP variable are significant, the null hypothesis of no structural breaks can be rejected and thus indicate that the GDP variable has structural breaks in 1982 and 1996. To consider this, two dummy variables named D1982 and D1996 are included in the

regressions. Likewise, the human capital variable has a structural break present in 1982. Likely explanations are the Mexican debt crisis in 1982 (van Wijnbergen, 1990) and the Mexican peso crisis during 1994-1996 (Truman, 1996). Additionally, Appendix D presents the breakpoint unit root test, which allows for structural breaks in the trend process11_{. The}

remaining F-statistics are insignificant and hence it appears as if no statistical12_structural

breaks are present in these time series.

Table 6, Multiple breakpoint test

H ₀ : No structural breaks Bai-Perron multiple breakpoint test

F-stat. 7.262** 7.719** 1.824 8.964** 5.331

Year 1982-1982 1996-1996 1982-1982

Notes: (*), (**) and (***) indicate statistically significant at the 10%, 5% and 1% significance level respectively.

Additionally, the ARDL model is sensitive to autocorrelation since it causes inconsistent and thus inefficient estimators. This issue can, to some extent, be mitigated through the decision of the ideal number of lags used. The optimal lag selection is determined by the Akaike Information Criteria (AIC) since it does not underfit the model and thus reduce eventual problems with autocorrelation. However, to make sure that the models are well-specified diagnostic tests for autocorrelation, non-normality, heteroscedasticity and misspecification are conducted. To verify parameter stability the cumulative sum of recursive residuals (CUSUM) and CUSUM of squares (CUSUMSQ) are applied. Lastly, following the original approach by Pesaran et al. (2001) two versions of Model 1a and 2a are estimated: one including a trend variable and a second one excluding this variable. If the trend variable enters significantly and the model passes all the diagnostic tests, this version is preferred.

11_{Conventional unit roots tests are biased toward accepting a false null hypothesis, i.e. a false unit root null.} 12_{As touched upon previously, there might exist non-statistical structural breaks during the time period}

estimated. The Mexican economy has experienced large changes in terms of political structure, indicating that despite that the Bai-Perron test depict no sign of statistical structural breaks in the data, it is a strong assumption to assume no other types of structural changes.

(26)

21

4.4 Toda & Yamamoto and Dolado & Lütkepohl (TYDL)

The third research question examines the direction among FDI, human capital and economic performance in Mexico. A common technique to address the problem is the method developed by Granger (1969). However, since it requires the variables to be stationary and integrated of the same order, this paper instead applies the technique presented by Toda and Yamamoto (1995) and Dolado and Lütkepohl (1996) (henceforth TYDL) which solves the stationarity and integration issues present in the Granger approach. The TYDL method works regardless of whether the variables are integrated and non-cointegrated, stationary or cointegrated. This is a major advantage since the data used in economics usually are I(1), which means that the long-run interactions cannot be estimated. The test is, nevertheless, based on the Granger causality technique and uses a vector autoregressive (VAR) model, but has a higher statistical power compared to the test by Granger (1969). The TYDL models used to test Hypothesis 3a, 3b and 3c are specified in Model 3a, 3b and 3c respectively.

𝐺𝐷𝑃𝑡= 𝛼 + ∑ 𝛽𝑖𝐺𝐷𝑃𝑡−𝑖+ ∑ 𝛽𝑖𝐺𝐷𝑃𝑡−𝑖+ 𝑚+𝑑𝑚𝑎𝑥 𝑖=𝑚+1 𝑚 𝑖=1 ∑ 𝛾𝑗𝐹𝐷𝐼𝑡−𝑗+ ∑ 𝛾𝑗𝐹𝐷𝐼𝑡−𝑗+ 𝜆1𝐻𝐶𝑡 + ℇ𝑦𝑡 𝑚+𝑑𝑚𝑎𝑥 𝑗=𝑚+1 𝑚 𝑗=1 𝐹𝐷𝐼𝑡= 𝜔 + ∑ 𝛿𝑖𝐹𝐷𝐼𝑡−𝑖+ ∑ 𝛿𝑖𝐹𝐷𝐼𝑡−𝑖+ ∑ 𝜑𝑗𝐺𝐷𝑃𝑡−𝑗+ ∑ 𝜑𝑗𝐺𝐷𝑃𝑡−𝑗+ 𝜆1𝐻𝐶𝑡+ ℇ𝑋𝑡 𝑚+𝑑𝑚𝑎𝑥 𝑗=𝑚+1 𝑚 𝑗=𝑖 𝑚+𝑑𝑚𝑎𝑥 𝑖=𝑚+1 𝑚 𝑖=1 𝐺𝐷𝑃𝑡= 𝛼 + ∑ 𝛽𝑖𝐺𝐷𝑃𝑡−𝑖+ ∑ 𝛽𝑖𝐺𝐷𝑃𝑡−𝑖+ 𝑚+𝑑𝑚𝑎𝑥 𝑖=𝑚+1 𝑚 𝑖=1 ∑ 𝛾𝑗𝐻𝐶𝑡−𝑗+ ∑ 𝛾𝑗𝐻𝐶𝑡−𝑗+ 𝜆1𝐹𝐷𝐼𝑡 + ℇ𝑦𝑡 𝑚+𝑑𝑚𝑎𝑥 𝑗=𝑚+1 𝑚 𝑗=1 𝐻𝐶𝑡= 𝜔 + ∑ 𝛿𝑖𝐻𝐶𝑡−𝑖+ ∑ 𝛿𝑖𝐻𝐶𝑡−𝑖+ ∑ 𝜑𝑗𝐺𝐷𝑃𝑡−𝑗+ ∑ 𝜑𝑗𝐺𝐷𝑃𝑡−𝑗+ 𝜆1𝐹𝐷𝐼𝑡+ ℇ𝑋𝑡 𝑚+𝑑𝑚𝑎𝑥 𝑗=𝑚+1 𝑚 𝑗=𝑖 𝑚+𝑑𝑚𝑎𝑥 𝑖=𝑚+1 𝑚 𝑖=1 𝐹𝐷𝐼𝑡= 𝛼 + ∑ 𝛽𝑖𝐹𝐷𝐼𝑡−𝑖+ ∑ 𝛽𝑖𝐹𝐷𝐼𝑡−𝑖+ 𝑚+𝑑𝑚𝑎𝑥 𝑖=𝑚+1 𝑚 𝑖=1 ∑ 𝛾𝑗𝐻𝐶𝑡−𝑗+ ∑ 𝛾𝑗𝐻𝐶𝑡−𝑗+ 𝜆1𝐺𝐷𝑃𝑡+ ℇ𝑦𝑡 𝑚+𝑑𝑚𝑎𝑥 𝑗=𝑚+1 𝑚 𝑗=1 𝐻𝐶𝑡= 𝜔 + ∑ 𝛿𝑖𝐻𝐶𝑡−𝑖+ ∑ 𝛿𝑖𝐻𝐶𝑡−𝑖+ ∑ 𝜑𝑗𝐹𝐷𝐼𝑡−𝑗+ ∑ 𝜑𝑗𝐹𝐷𝐼𝑡−𝑗+ 𝜆1𝐺𝐷𝑃𝑡+ ℇ𝑋𝑡 𝑚+𝑑𝑚𝑎𝑥 𝑗=𝑚+1 𝑚 𝑗=𝑖 𝑚+𝑑𝑚𝑎𝑥 𝑖=𝑚+1 𝑚 𝑖=1

In which dmax represents the maximum order of integration (in this paper dmax = 1) and is included to test if the model is dynamically stable. m is the optimal lag length and is determined through the VAR lag order selection criteria13_{. ℇ}

t represents the residuals.

Moreover, the variables associated with human capital, FDI and economic growth are included as exogenous variables in Model 3a, 3b and 3c respectively. To test the null hypothesis of no Granger causality the modified Wald´s test is conducted.

13_{Consists of the criteria: Sequential modified LR statistic, Final prediction error, AIC, Schwarz-Bayesian}

criteria and Hannan-Quinn information criteria. If they provide different results, the AIC is used.

(3a)

(3b)

(27)

22

5 Empirical Results & Analysis

_____________________________________________________________________________________

This chapter presents the empirical results from the ARDL and TYDL regression models. The results are interpreted, discussed and interlinked to previous research.

______________________________________________________________________

5.1 Hypothesis 1

This study is mainly concerned with examining the relationship among economic performance and FDI inflows after including a variable measuring human capital, and the corresponding ARDL results are presented in Table 7 (page 24). When including only FDI as a determinant to economic performance (Model 1c) the coefficient associated with FDI inflows is negatively signed and insignificant. However, this model has problems with non-normally distributed residuals and additionally, the ECT(-1) variable is insignificant, implying that the model cannot be used. However, since this model primarily is used as a benchmark, no further adjustments are conducted to correct these constraints. When including the control variable capital formation (Model 1b), the FDI coefficient is still insignificant. Likewise, the variable associated with capital formation enters insignificantly.

However, when adding human capital to the regression model, one can discern obvious changes. As mentioned previously, this paper follows the approach by Pesaran et al. (2001) and estimate Model 1a both with and without a trend variable. By looking at Model 1a when including a time trend (first column in Table 7) all the explanatory variables and the trend variable are insignificant at the 5% significance level. In addition, it has problems with parameter instability. Consequently, this model will be disregarded and instead Model 1a without the trend variable is of primary interest (second column in Table 7). From this model one can discern that the FDI coefficient is now positive (0.020) but still insignificant. The result of the FDI variable thus disagrees both with theory (Romer, 1990) and previous research on the Mexican economy (Ramírez, 2000; Adames, 2000; Alguacil et al., 2002; Oladipo, 2007; Lal, 2017) saying that FDI is a significant determinant to economic growth. Insignificant or even negative relationships among FDI and economic performance are, on the other hand, commonly found when examining emerging economies as they often lack the know-how and technological skills needed to utilise the technical inflows associated with FDI (Blomström et al., 1992; Jyun-Yi & Chih-Chiang, 2008; Alvarado et al., 2017). In other words, this might suggest that Mexico does not possess the level of human capital required

(28)

23

to benefit from high FDI inflows and thereby supports the argumentation by Borensztein et

al. (1998) stating that emerging countries lack the sufficient amount of absorptive capacity.

Yet, the capital formation variable is now significant and suggests that a 1% increase in gross capital formation increases real GDP per capita by approximately 0.796%. However, it is worth reminding that the exact values of the coefficients are not the main focus of this paper. Instead, the sign, relative magnitude and how they alter when including human capital are of primary interest. As the inclusion of human capital changes the result, it may suggest that human capital is associated with the other explanatory variables and thus, the ignorance of such a variable may lead to biased results. The human capital coefficient is positive and highly significant and thus supports the main implications from Lucas (1988) and Romer (1990) and the empirical research on the Mexican economy by Oladipo (2007). As one might expect school enrolment14_{ratios to have lagged effects on economic performance, an alternative}

interpretation of the variable could be that the level of school enrolment captures the quality of schools; i.e. the positive human capital coefficient signals a good educational system. Additionally, the cointegration results for Model 1a indicate that there exists a long-run relationship among the explanatory variables during 1970 to 2018 in Mexico since the F-statistic (6.075) exceeds the upper critical value at the 1% significance level. Furthermore, Model 1a most likely does not suffer from autocorrelation, non-normality, heteroscedasticity or misspecification. Appendix E presents the CUSUM and CUSUMSQ for the model in which both plots fall within the bounds representing the critical values, implying that the model does not have problems with parameter instability. Lastly, the dummy variable measuring the influence after the NAFTA contract in 1994 is insignificant, suggesting that no major changes in the average economic performance can be found.

Continuing with the short term interactions (see Model 1aa in Table 7), all variables maintain the same sign as in the long-run model but have decreased in magnitude. Yet, the coefficient associated with ECT(-1) is negative and significant at the 5% level, and thus confirms the long-run relationship presented previously. Its coefficient of -0.295 indicates that 29.5% of any short-run disequilibrium from the long-run equilibrium is corrected within one year. In summary, Table 7 reveals that Hypothesis 1 can be rejected since FDI does not have a positive statistically significant relationship to GDP per capita after controlling for human capital in Mexico during 1970-2018.

(29)

24

Table 7, ARDL results for Hypothesis 1 and 2

Long run Short run

Model (1a) (1a) (1b) (1c) (2a) (2a) (1aa) (2aa)

Variables GDP GDP GDP GDP FDI FDI GDP FDI

c 1.894 8.586*** 4.517 12.016*** -4.892 -33.958* 2.534* -4.955*** FDI -0.230 0.020 -0.093 -0.690 - - 0.006 -HC 0.273* 0.214*** - - 2.314*** -0.321 0.063* 1.493 CF 0.791* 0.796** 2.338 - - - 0.318*** -GDP - - - - 0.115 3.017* - 0.095 DNAFTA 0.019 0.010 0.013 0.023 0.932*** 0.343** 0.010 0.932*** D1982 0.007 0.014 0.000 -0.039 -0.065 -0.523* 0.014 -0.065 D1996 0.025 0.005 0.030 0.031 -0.118 0.300 0.005 -0.118 TREND -0.030 - - - -0.076*** - - -0.063*** ECT(-1) -0.295** -0.822*** F-stat. 4.084** 6.075*** 7.232*** 4.595** 9.841*** 5.820*** Adj. R2 0.978 0.975 0.978 0.962 0.803 0.794 0.549 0.541 Lag structure (1,0,0,1) (1,0,0,2) (1,0,1) (1,0) (1,5,0) (4,0,1) Diagnostics test Breusch-Godfrey LM test 11.622 9.886 10.332 11.911 12.278 10.601 Jarque-Bera Normality test 0.293 3.589 25.001*** 20.334*** 0.490 2.173 Breusch-Pagan-Godfrey 1.666 2.956 0.863 0.642 8.092 9.652 Ramsey RESET 0.32 0.406 0.565 0.084 3.155 0.581

CUSUM Stable Stable Stable Stable Stable Unstable CUSUMSQ Unstable Stable Stable Stable Stable Stable

Notes: (*), (**) and (***) indicate statistically significant at the 10%, 5% and 1% significance level respectively. Additionally, the null hypotheses for the respective diagnostic tests are: no autocorrelation, normality, homoscedasticity and no misspecification.

(30)

25 5.2 Hypothesis 2

Consistently, also Model 2a is run both with and without a time trend as presented in Table 7. Since the model with the trend variable is significant with a value of -0.076 and simultaneously passes all diagnostic tests, this version is preferred over the model excluding the time trend. The F-statistics of 9.841 is highly significant and thus indicates a long-run relationship among all the variables estimated. Human capital appears to have a strong long-run relationship to FDI inflows through its coefficient of 2.314. The GDP variable is lower by a value of 0.115 but enters insignificantly. Yet, the short-run relations (Model 2aa in Table 7) depict a slightly different picture. Both explanatory variables have decreased in magnitude but are now statistically insignificant. The speed of adjustment coefficient is -0.822, suggesting that the correction rate to the long-run equilibrium in response to imbalances caused by short-run shocks in the previous period is 82.2%. Additionally, the dummy variable is highly significant and positive, suggesting that the share of FDI inflows has, on average, increased largely after NAFTA. Lastly, this model passes the CUSUM tests for parameter stability presented in Appendix F15_.

Thus, Hypothesis 2 cannot be rejected and the results of this paper suggest that human capital and FDI inflows have a positive statistically significant relationship in Mexico during 1970-2018. This finding questions the argumentation that Mexico does not have the absorptive capacity required to benefit from FDI inflows. However, as the causal link between the variables is far from conclusive, it is difficult to assess whether FDI helps spurring human capital or if it is the level of human capital that attracts FDI inflows into the Mexican economy.

5.3 Hypothesis 3a, 3b and 3c

Due to the inconclusiveness concerning the direction of causality among both GDP and FDI as well as human capital and FDI, this motivates a test for Granger causality which is presented in Table 8. The findings depict how GDP precedes FDI and not the other way around as concluded by Alguacil et al. (2002, 2004) and Lal (2017) and no bi-directional causality as suggested by Zhang (2001). The reason to the disagreement may be many, however, some explanations are likely the different time periods estimated, the methodology

15_{The CUSUM for Model 2a is a borderline case, however since the CUSUM-line does not seem to exceed the}

FDI, human capital and economic performance in Mexico : An ARDL cointegration and Granger causality approach