Does technology shape the relationship between FDI and growth? A panel data analysis

ABSTRACT

The theoretical literature on the FDI-growth relationship suggests that FDI with high technological content might play a peculiar role. To investigate the existence and magnitude of this peculiar effect, we collected data for 28 countries over the period 1989 to 2019 and used GMM techniques for empirical estimation. We employed the U.N. International Standard Classification (ISIC) Revision 3 to classify FDI data and follow the criteria laid down by the Organization for Economic Cooperation and Development (OECD) to distinguish FDI data by technological content. The empirical findings of this article confirm that technology plays an important role in determining the FDI-Growth relationship. Interestingly, the empirical evidence supports a U-shaped relationship between FDI and economic growth when FDI is disaggregated by different technological contents. More specifically, we find strong evidence that in the manufacturing sector, FDI with a higher technological content exhibits a positive association with growth in the host country. We also find evidence pointing towards a positive relationship between FDI and growth in the host country at the other end of the technology spectrum (low-tech). Further investigation confirms the robustness of these findings across different estimation techniques as well as across different sampling strategies.

KEYWORDS:

• Panel data models
• foreign direct investment
• economic growth
• manufacturing and service industries
• technology

JEL CLASSIFICATION:

• C23
• F21
• F43
• O14
• O57

I. Introduction

The relationship between Foreign Direct Investment (henceforth FDI) and economic growth (henceforth EG) in the host country has been studied widely in the literature. Such keen interest was sparked by the surge in the flow of FDI to a number of host countries¹ which has taken place in recent decades as multinational firms looked for different ways of expanding their business beyond their national borders (Figure 1).

Figure 1. FDI inflows, global and by group of economies 1990–2020.

Source: Data from UNCTAD, World Investment Report 2021

Based on empirical evidence, there is no consensus on whether such a relationship exists or the direction of causality.² A broader consensus has been reached on the existence of a relationship between FDI and EG after taking into account threshold effects in a number of covariates, like, but not limited to, human capital, financial development, etc.³ However, very few studies to date have tried to probe whether the existence of a relationship between FDI and EG depended on the sector of industry the FDI was flowing into or the type of FDI involved.⁴

Until recently, lack of granularity in FDI data was a major constraint for anybody who wanted to explore these issues. As these data become more plentiful and more detailed and cover longer time span, this constraint is being relaxed. In particular, in the past data constraints forced scholars and policymakers to treat FDI as a whole, irrespective of the type and sector of the economy. However, more recently, a growing interest has started to appear in probing whether the relationship between FDI and EG changes depending, for example, on the sector concerned (agriculture, manufacturing or services), on the type of FDI involved (horizontal or vertical FDI are dimensions that have been investigated and so too are greenfield investment or mergers and acquisitions) or whether it changes depending on the country FDI originates from. More details on these attempts to distinguish FDI in some way will be given later in the literature review section.

The present study is an attempt at bridging the gap in the existing literature in this direction. More specifically, we focus on the role played by technology transfers to the host country as an important channel linking FDI to EG (two of the most seminal studies in this respect are De Mello 1997; Borensztein, Gregorio, and Lee 1998). To this end, this article aims at probing whether the effect of FDI on growth varies depending on its technological content. Our objective is to show that the effect of FDI on the EG and development of the recipient nation depends crucially on the technology embodied in it. More specifically, we disaggregate FDI into different sectors and technological contents to differentiate the role of each sector and content on EG. In other words, what we would like to show is that the heterogeneity in the technological content of diverse types of FDI matters.

Based on the evidence we gather, we find that, after differentiating FDI by technological content, a U-shaped relationship emerges relating FDI, technology and EG. At both ends of the technological spectrum, FDI flows with high and low technological content are highly conducive to EG, while the two intermediate types of FDI with medium technological content are not. This result holds across two different estimation techniques – generalized method of moments (GMM) and fixed effects (FE) and two sampling methods – first including countries for which the FDI value is non-zero and then using a full sample of all countries.⁵ (More details are given in the main body of the article below). We therefore conclude that this result is not sensitive to the econometric methodology used or to the inclusion or not of zero values for the relevant FDI type. This study contributes to the existing literature in several ways: as mentioned, we uncover important technology spillovers on the economic performance of host countries. But we add to this more nuance and additional findings. First, we find that the positive effect of technology is limited to FDI with high technological content. In contrast, as that technological content drops, the effect of FDI on the EG of the host country becomes indiscernible at best. We conjecture that this finding may be due to other, less benign effects of FDI that may offset wholly or partially the positive effects of technology. However, we also uncover an intriguing feature of FDI at the other end of the technology spectrum, namely low-technology FDI. This FDI type is found to have a positive impact on the economic growth of the recipient country. In this instance, this positive effect is probably not due to technology spillovers. Here, we conjecture that low-tech FDI may better fit with the economic texture of recipient countries⁶

The article is organized as follows: Section II briefly reviews the literature on FDI and EG, which is relevant for our stated goals; Section III lays out the econometric model to be estimated and sketches the econometric challenges being faced in the estimation; Section IV details data, data sources and descriptive statistics; Section V illustrates the results of the estimation exercise, while Section VII mentions possible extensions and future line of work and draws conclusions.

II. A brief look at the existing empirical literature

The literature on the relationship between FDI and EG is extensive, but its review is beyond the scope of this article. Since the focus of this article is to exclusively investigate how different types of FDI impact EG, we shall restrict our review to the literature that specifically covers this part.

Alfaro (2003) can be considered as the first attempt at investigating the above question. By using cross-country data for 47 countries over the period 1981–1999, she differentiated investment into three sectors (primary, manufacturing and services) and studied the effect of FDI inflows in these sectors on EG. She found that the effect of FDI on EG was sector-dependent. Her empirical findings showed that FDI into primary sector had a negative impact on growth, the impact of FDI in manufacturing was positive, while the effect of FDI in the services sector was ambiguous.Alfaro and Charlton (2007) in an extension to this work distinguished different qualities of FDI by adding specific industry characteristics. Using a panel of 29 countries and 15 years (1985–2000), they found a positive relation between FDI and EG in industries with higher skills requirements and in industries more reliant on external capital.

Beugelsdijk, Smeets, and Remco (2008) distinguished the growth effect of FDI into horizontal (market seeking) FDI⁷ and vertical (efficiency seeking) FDI. Growth effects were estimated using a sample of 44 host countries over the period 1983 to 2003. The article also splits the sample into developed and developing countries to probe whether differences depend on country-status. The empirical findings confirmed positive and significant growth effects from both horizontal and vertical FDI in developed countries, while no significant relationship was found for the sample of developing countries. The article also found that the effect of horizontal FDI to be about 50% larger than that of vertical FDI.

Chakraborty, Nunnenkamp, and Nunnenkamp (2008) used panel data of 15 industries for India segregating it into three sectors: manufacturing, primary, and services (1987–2000) and found a positive impact of manufacturing FDI on EG but no evidence of a similar relationship was found for primary sector FDI, while the services sector had only transitory effects. Massoud (2008) did a similar exercise using data from Egypt (1974–2005) and found that only manufacturing FDI promotes EG (when interacting with human capital). In another paper (Wang 2009) used a sample of 12 Asian economies over the period 1987–1997. He further split the FDI data into manufacturing and non-manufacturing FDI and investigated their impact on EG of host countries.. The empirical results showed that manufacturing FDI had a positive and significant impact on EG of the host country, whereas the effect of FDI inflows in non-manufacturing sectors was ambiguous.

Further studies on FDI attempted to split business activities (both in the manufacturing and services sectors) on the basis of their technological content. Hatzichronoglou (1997), based on R&D intensities in each sector, classified sectors by technological content. Mani (2000) used the same classification as Hatzichronoglou (1997) while studying high-tech exports for a sample of OECD countries. This classification was then finalized in OECD (2003) and referred to in subsequent OECD documents covering technology intensity (see, e.g. OECD 2011). We use this classification in our attempt to differentiate FDI by technological content.

2015) use data from the WB–IFC Enterprise Survey to determine backward linkages from FDI for developing countries. Part of the work is to determine the motivation of foreign-owned firms across sectors. The study finds that ‘some sectors (food, wood, auto and auto parts) are more prone than others (textiles and electronics) in developing backward linkages’. Gönel and Aksoy (2016) investigated the relationship between FDI and growth when FDI is differentiated by three sectors of activity. They found that while aggregate FDI has a significant and positive impact on EG, FDI belonging to the three sectors on their own has no effect on EG. They do find, however, that a certain absorptive capacity in the host country is conducive to FDI having a positive effect on EG.

In a recent paper, Shah et al. (2019) looked at the relationship between sectoral FDI and domestic investment in Pakistan. They used an autoregressive distributed lag (ARDL) model for data from Pakistan over the period 1980–2012. They tackle the specific question by splitting FDI into three sectors, primary, manufacturing and services. The results suggest that FDI in the manufacturing and services sectors has a positive impact on domestic investment, while primary sector FDI does not influence domestic investment. In another paper, Jana, Sahu, and Pandey (2019) employed a time-varying parameter model for data from India covering the period 1995–2016 to study the effects of sector-specific FDI inflows on economic growth. The findings of this article revealed that in the case of India, the agriculture sector FDI does not have any positive impact on agriculture production or growth. For manufacturing and services sectors, the study found a positive relationship between FDI and EG.

Ingham, Read, and Elkomy (2019) did a similar study using data from Egypt (1990–2007). They used a fixed effect model for unpublished data from the General Authority of Free Zones and Investment (GAFI). The findings show a positive growth effect in manufacturing and petroleum sectors but not in other sectors, such as finance and retail, tourism, construction and agriculture. Interestingly, the manufacturing sector also exerts positive spillovers on the latter sectors.

Doytch and Uctum (2019) used a sample of 14 Asia Pacific economies, with data spanning 1985–2012 to investigate the impact of services FDI on EG. They first disaggregated FDI by manufacturing and services and then further disaggregated services into financial services, trade services, and business services. Interestingly, this article found services FDI to have a positive impact on growth, while no effect on growth was found for manufacturing FDI.

The present article can be considered as an extension of the above literature. There are two major differences in this article. First, while the existing literature is limited to individual country or to a specific region, this article uses a large global sample of 48 developing and developed countries. Second, the classification of FDI flows by sector in the existing studies is limited to broad sectors, such as agriculture, manufacturing and services. In this article, we use a more detailed classification of FDI based on technological content and include more sectors (see section III below). Interestingly, the results of this article are contrary to Gönel and Aksoy (2016) and we did not find any empirical support for the effect of aggregate FDI on EG of the host country. However, our results suggest an impact on EG for FDI with high and low technological content, respectively.

III. Econometric model, methodology and estimation challenges

The main objective of this article is to investigate whether differentiating FDI by typein particular, distinguishing FDI on the basis of the technological content embedded into it, can help shed further light on the nature of the relationship between FDI and EG beyond what we already know about it.

As a first step, we differentiate FDI by sectors and then by technological content for each sector. The World Investment Directory (henceforth W.I.D.) classifies FDI data according to the U.N. International Standard Classification (ISIC) Revision 3 (UNSD 2021). This provides a very detailed and accurate classification of economic activities, which makes the task of grouping sectors by technological content far easier. We follow the OECD criteria for differentiating sectors by technological content (OECD 2003), and accordingly use four different groups in the manufacturing (secondary) sector. As for the services sector, we split the sector into two groups: high tech and low tech according to the use of incorporated technologies, R&D intensities and the qualification (skillness) of the workers. To do so, we follow Table 2 in Anghel (2006) which in turn draws from 2003). Next, we employ FDI/GDP ratios for specific sectors and sub-sectors. These details are provided in Tables 1 and 2.

Table 1. FDI decomposition by sector and by technology.

Secondary	High Tech	Aircraft and Spacecraft
		Pharmaceuticals
		Office, accounting and computing machinery
		Radio, TV and communication equipment
		Medical, precision and optical instruments
	Medium High Tech	Electrical machinery and apparatus, n.e.c.
		Motor vehicles, trailers and semi-trailers
		Chemicals excluding pharmaceuticals
		Railroad equipment and transport equipment, n.e.c.
		Machinery and equipment, n.e.c.
	Medium Low Tech	Building and repairing of ships and boats
		Rubber and plastics products
		Coke, refined petroleum products and nuclear fuel
		Other non-metallic mineral products
		Basic metals and fabricated metal products
	Low Tech	Manufacturing, n.e.c.; Recycling
		Wood, pulp, paper, paper products, printing and publishing
		Food products, beverages and tobacco
		Textiles, textile products, leather and footwear
Services	High Tech	Post and Telecommunications
		Financial Intermediation
		Renting and Business Activities
		Education, Health and Social Work
	Low Tech	Services n.e.c.

Table 2. FDI variables with definitions.

Variable Name	What it measures	Variable Name	What it measures	Variable Name	What it measures
FDI	Total FDI/GDP ratio	FDISEC	Secondary FDI/GDP ratio	HTFDISec	High Tech FDI Sec/GDP Ratio
				MHTFDISec	Medium High Tech FDI Sec/GDP Ratio
				MLTFDISec	Medium Low Tech FDI Sec/GDP Ratio
				LTFDISec	Low Tech FDI Sec/GDP Ratio
		FDISERV	Services FDI/GDP ratio	HTFDIServ	High Tech FDI Services/GDP ratio
				LTFDIServ	Low Tech FDI Services/GDP ratio
		FDIPRIM	Primary FDI/GDP ratio

For empirical estimation, we, first, run a benchmark panel regression (Equation 1(1) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{FD}{I}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(1) ) with the (annual) growth rate of GDP per capita as the dependent variable and centred on the FDI/GDP ratio as the main regressor as follows:

(1) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{FD}{I}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(1)

where $g_{\mathit{it}}$ is the (real) annual growth rate of GDP per capita for country i at time t and $\mathit{FD}{I}_{\mathit{it}}$ is the FDI/GDP ratio for country i at time t. We also include the lagged value of the dependent variable (${\mathrm{g}}_{\mathrm{it}-1}$) as first regressor and a set of standard control variables included in the vector ${\mathbf{A}}_{\mathbf{it}}$. These comprise domestic credits (d_credit), inflation, population growth (pop_growth), trade, government expenditures (gov_exp), gross fixed capital formation (gfcf) and our measure of human capital (hc), which is a human capital index taken from Penn World Tables 9.1⁸ (see Section V on data and sources for more detail). All the controls’ values are measures for country i at time t. It follows that $\mathit{\gamma }$ is a vector of coefficients corresponding to the vector of controls ${\mathbf{A}}_{\mathbf{it}}$ ⁹and ${\epsilon }_{\mathit{it}}$ is a standard error term. The need to account for any remaining country specific characteristics not already controlled for would suggest the use of the fixed effect model (in short: FE) in this regression. The Hausman test confirms that the fixed effects model is superior to the random effects model in our case.¹⁰ However, we can further reduce the risk of obtaining biased results by employing the system Generalized Method of Moments technique (system GMM). GMM estimation is an improvement on FE estimation in at least three ways.

First, the model, as specified in Equation (1)(1) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{FD}{I}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(1) is dynamic insofar as it includes the lagged value of the dependent variable as one of the regressors. GMM estimation enables us to account for the dynamic nature of the model.

Second, the length of the period under study (from 1989 to 2019) increases the likelihood of structural breaks somewhere along the line. Once again, GMM estimation is equipped to handle structural breaks in the data, while FE estimation is not.

Finally, and perhaps most decisively, GMM estimation is a lot better than FE estimation at addressing the endogeneity bias, frequently lurking in this type of panel data, where it is hard to know ex ante the direction of causality (if any) between the designated dependent variable (EG in this case) and independent regressors of interest (here: FDI).

For the sake of completeness of exposition and intellectual openness, it must be said that GMM estimation carries the risk of some bias of its own. For example, a blind use of this technique can lead to instrument proliferation and consequent so-called overfitting bias. A large time dimension T can also bias the results. Finally, the use of weak instruments (not correlated with the dependent variable but not orthogonal to the other regressors) can bias the results in the direction of the OLS estimator.

We test the weakness of the instruments by running a standard Sargan test. We address the overfitting bias caused by instrument proliferation in two ways. We suitably reduce the number of lags and apply the collapse option available in Stata. We report on the number of instruments for all GMM regressions in each table.

In this paper, we use the GMM model in regression (1) and all the other annual panel regressions used in this article. However, in order to allay any fears related to GMM estimation, we also report our results from FE estimation in appendix B, Tables B1, B2 and B3.

The second equation differentiates FDI by type. In Equation (2)(2) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_3\mathit{FDISE}{C}_{\mathit{it}}+{\beta }_4\mathit{FDISER}{V}_{\mathit{it}}+{\beta }_2\mathit{FDIPRI}{M}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(2) , we replace the total FDI/GDP ratio (in Equation 1(1) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{FD}{I}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(1) ) with the sum of sectoral FDI/GDP ratios from the Primary, Secondary and Services sectors. Everything else is the same as in Equation (1)(1) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{FD}{I}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(1) .

(2) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_3\mathit{FDISE}{C}_{\mathit{it}}+{\beta }_4\mathit{FDISER}{V}_{\mathit{it}}+{\beta }_2\mathit{FDIPRI}{M}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(2)

Next, in order to explore differences in FDI types with differing technological content associated with the growth rate of GDP per capita, in Equation (3)(3) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{HTFDISe}{c}_{\mathit{it}}+{\beta }_3\mathit{MHTFDISe}{c}_{\mathit{it}}+{\beta }_4\mathit{MLTFDISe}{c}_{\mathit{it}}+{\beta }_5\mathit{LTFDISe}{c}_{\mathit{it}}+{\beta }_6\mathit{HTFDISer}{v}_{\mathit{it}}+{\beta }_7\mathit{LTFDIServ}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(3) we replace the FDI/GDP ratio with the sum of the six sub-variables resulting from the full FDI decomposition according to technological content. These technology-related distinctions are referred as high-tech (HTFDI), medium high-tech (MHTFDI), medium low-tech (MLTFDI) and low-tech (LTFDI). Accordingly, we replace the variable FDI with the sum of the sub-variables HTFDISec, MHTFDISec, MLTFDISec, LTFDISec, HTFDIServ and LTFDIServ and their respective coefficients, then add the same set of control variables as in Equation (1)(1) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{FD}{I}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(1) . As a result, Equation (3)(3) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{HTFDISe}{c}_{\mathit{it}}+{\beta }_3\mathit{MHTFDISe}{c}_{\mathit{it}}+{\beta }_4\mathit{MLTFDISe}{c}_{\mathit{it}}+{\beta }_5\mathit{LTFDISe}{c}_{\mathit{it}}+{\beta }_6\mathit{HTFDISer}{v}_{\mathit{it}}+{\beta }_7\mathit{LTFDIServ}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(3) looks as follows:

(3) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{HTFDISe}{c}_{\mathit{it}}+{\beta }_3\mathit{MHTFDISe}{c}_{\mathit{it}}+{\beta }_4\mathit{MLTFDISe}{c}_{\mathit{it}}+{\beta }_5\mathit{LTFDISe}{c}_{\mathit{it}}+{\beta }_6\mathit{HTFDISer}{v}_{\mathit{it}}+{\beta }_7\mathit{LTFDIServ}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(3)

Equation (3)(3) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{HTFDISe}{c}_{\mathit{it}}+{\beta }_3\mathit{MHTFDISe}{c}_{\mathit{it}}+{\beta }_4\mathit{MLTFDISe}{c}_{\mathit{it}}+{\beta }_5\mathit{LTFDISe}{c}_{\mathit{it}}+{\beta }_6\mathit{HTFDISer}{v}_{\mathit{it}}+{\beta }_7\mathit{LTFDIServ}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(3) is the main equation we use in this article to empirically verify our research question.

As part of a robustness test, we run an additional equation (Equation 4(4) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{HTFDISe}{c}_{\mathit{it}}+{\beta }_3\mathit{MHTFDISe}{c}_{\mathit{it}}+{\beta }_4\mathit{MLTFDISe}{c}_{\mathit{it}}+{\beta }_5\mathit{LTFDISe}{c}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(4) ) which focuses on the secondary sector only by including its four different types of FDI as in Equation (3)(3) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{HTFDISe}{c}_{\mathit{it}}+{\beta }_3\mathit{MHTFDISe}{c}_{\mathit{it}}+{\beta }_4\mathit{MLTFDISe}{c}_{\mathit{it}}+{\beta }_5\mathit{LTFDISe}{c}_{\mathit{it}}+{\beta }_6\mathit{HTFDISer}{v}_{\mathit{it}}+{\beta }_7\mathit{LTFDIServ}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(3) but leaving out the two FDI types of the services sector. Hence, Equation (4)(4) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{HTFDISe}{c}_{\mathit{it}}+{\beta }_3\mathit{MHTFDISe}{c}_{\mathit{it}}+{\beta }_4\mathit{MLTFDISe}{c}_{\mathit{it}}+{\beta }_5\mathit{LTFDISe}{c}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(4) includes the sum of FDI/GDP ratios from the secondary sector only, differentiated by technological content:

(4) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{HTFDISe}{c}_{\mathit{it}}+{\beta }_3\mathit{MHTFDISe}{c}_{\mathit{it}}+{\beta }_4\mathit{MLTFDISe}{c}_{\mathit{it}}+{\beta }_5\mathit{LTFDISe}{c}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(4)

Since the empirical literature on FDI-EG relationship is inconclusive, in this article, we expect the sign to be positive or negative. Lagged GDP is also expected to have a positive sign. For sector-specific FDI, the literature (as stated in Section II above) has gathered mixed evidence. However, a majority of studies suggest a positive relationship between manufacturing sector FDI and EG, while the same is not supported for services and primary sectors. In this article, we have disaggregated industries into secondary, primary and services sectors and then further disaggregated the data for the secondary and services sectors by their technological content as discussed above. To the best of our knowledge, there is no research is available on this disaggregation.¹¹ In our view, technology will further strengthen the relationship.

As for control variables, theoretically, domestic credit (to the private sector) is expected to have a positive impact on growth. However, empirical evidence in the literature is mixed. Khalid and Marasco (2019) discussed this issue and found a negative relationship between EG and domestic credit.¹² Hence, we leave the impact of domestic credit on growth on empirical verification. Inflation is expected to have a negative impact on growth. Neoclassical growth theory suggests that population increase can impact growth negatively. Crowding-out literature suggests that increases in government expenditure should have a negative impact on growth, especially if the country is facing a persistent budget deficit. Both trade and gross investment are expected to have a positive impact on growth. Finally, human capital should help improve productivity and impact growth in a positive manner.

IV. Data, sources and descriptive statistics

This section discusses the data used in our study. First, we illustrate and discuss FDI data and sources, including a discussion of the criteria and the sources that were used for the FDI decomposition we adopted. Later, we list the data and sources concerning all the other variables used in the study.

As stated earlier, the main constraint we faced in the empirical analysis was the availability of FDI data classified by the sector of activity. Unfortunately, there is no single source where all data are available. To overcome this issue, we use different sources to collect data for the variables used in this study.¹³ Typically, data collection in this respect is the responsibility of individual countries’ central banks or statistical offices. Additionally, several international agencies (United Nations, OECD, World Bank, to name but a few) have at different times attempted to collate data concerning countries in different geographical areas. In this study, we collected data for 48 countries, spanning the period between 1989 and 2019. The number of countries after dropping to zero values for the relevant FDI variables goes down to 28. We also used the full sample to segregate data into the high-income countries’ (31 countries) and European region (26 countries).

It is important to note that the sample of 48 countries (both developed and developing) used in this study includes five main regions: 6 countries are from East Asia and the Pacific, 26 countries are from Europe and Central Asia, 13 are from North and Latin America, 1 (Tunisia) is from the Middle East and Africa and 2 (India and Pakistan) are from South Asia. The most notable absences from this sample, due to lack of detailed FDI data, are China and Russia. The full list of countries is available in Appendix A, Table A1.

We used three different sources to collect data. Data on human capital come from the Penn World Tables (version 9.1). This source gives a human capital index, which is based on the average years of schooling from Barro and Lee (BL, 2013) and an assumed rate of return to education, based on Mincer equation estimates around the world (Psacharopoulos, 1994).¹⁴ Data for the dependent variable, growth rate of GDP per capita, as well as data on all other regressors – domestic credit, government expenditure (both measured as ratios to GDP), population growth (annual percent), trade (defined as the sum of exports and imports of goods and services, measured as a share of GDP), gross fixed capital formation (percent of GDP) and inflation *GDP deflator, percent) are obtained from the World Bank World Development Indicators¹⁵ Individual countries’ central banks and statistical bodies are used to collect data on FDI-related variables (FDI and its classification by sectors and technology content).

We first report the summary and correlation statistics across all variables concerned (see Appendix Tables B1 and B2):

One apparent feature in the data is that the number of observations drops drastically when it comes to FDI variables, which confirms that FDI data constitute the bottleneck for this study. However, it is of note that we managed to keep the number of observations above the 1000 mark for all the FDI variables when classified by technological content. The very small magnitude of the ratios in the FDI decomposition based on technology reflects the fact that the FDI inflow (the numerator) is generally a tiny proportion of GDP (the denominator).

For correlation statistics, the correlation coefficients provide a first hint of the findings that will come with the regressions. For example, the FDI is negatively correlated with EG (value: −0.036), whereas HT FDI Sec is positively correlated with EG (value: 0.0893).

V. Results and robustness checks

The results of empirical estimation are presented and discussed below. We discuss here the results of the GMM regressions after dropping zero values for the FDI/GDP ratios.¹⁶ These results are presented in Table 3.

Table 3. Dynamic system GMM non-zero countries.

	(1)	(2)	(3)	(4)
${\mathrm{g}}_{\mathrm{it}-1}$	0.265***	0.243***	0.232***	0.242***
	(5.12)	(4.34)	(4.12)	(4.76)
FDI	0.0511
	(1.58)
FDISEC		0.297***
		(2.86)
FDISERV		0.000329
		(0.01)
FDIPRIM		−0.0473*
		(−1.87)
HTFDISec			1.680***	1.450***
			(3.13)	(2.96)
MHTFDISec			−0.0731	−0.0492
			(−0.33)	(−0.23)
MLTFDISec			−0.0600	0.00230
			(−0.14)	(0.01)
LTFDISec			1.155***	1.197***
			(4.22)	(3.70)
HTFDIServ			0.0602***
			(3.18)
LTFDIServ			−0.0574
			(−0.68)
d_credit	−0.0138***	−0.0146***	−0.0195***	−0.0226***
	(−3.33)	(−3.19)	(−4.77)	(−4.86)
inflation	−0.00397	−0.0258	−0.0276	−0.151**
	(−0.11)	(−0.67)	(−0.75)	(−2.39)
pop_growth	−1.088***	−1.389***	−1.578***	−1.536***
	(−4.25)	(−6.72)	(−7.40)	(−6.62)
trade	0.00816*	0.00470	−0.00297	−0.00497
	(2.02)	(0.89)	(−0.72)	(−1.13)
gov_exp	−0.252**	−0.217***	−0.226***	−0.248***
	(−3.57)	(−3.71)	(−4.25)	(−4.65)
gfcf	0.0783	0.168***	0.181***	0.167***
	(1.18)	(3.97)	(4.52)	(3.79)
hc	0.105	0.0507	0.552	0.170
	(0.21)	(0.09)	(1.02)	(0.28)
_cons	5.103**	3.134	2.363	5.090**
	(2.11)	(1.56)	(1.31)	(2.15)
N	618	544	523	531
Instruments	14	16	20	18
AR(1) p-value	0.000	0.000	0.000	0.000
AR(2) p-value	0.008	0.108	0.150	0.126
AR(3) p-value	0.468	0.215	0.384	0.258
Sargan	0.762	0.767	0.456	0.322
Hansen	0.581	0.597	0.121	0.143

Sargan and Hansen p-values are reported in the table. Both Sargan and Hansen test the null that instruments as a group are exogenous. In all our models, we fail to reject the null hypotheses, thereby indicating that the instruments are valid.

Notes: 1. Sargan and Hansen p-values are reported in the table. Both Sargan and Hansen test the null that instruments as a group are exogenous. In all our models, we fail to reject the null hypotheses, thereby indicating that the instruments are valid.2. The test for serial correlation in the residual shows presence of first order serial correlation (AR1) while absence of second order serial correlation (AR2) in all models except in the model in equation 1. In this model, serial correlation of order 2 is found. Consequently, we have used higher order lags as instruments in this model, as suggested in Arellano and Bond (1991).

We start with Equation 1.(1) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{FD}{I}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(1) In contrast to many earlier studies with respect to FDI regressors, we find that the association between FDI Total and EG, though positive, is not significant, at least in GMM regression. This unexpected result might be due to some heterogeneity in the effect of different FDI types across sectors on EG. To verify this, we run Equation (2)(2) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_3\mathit{FDISE}{C}_{\mathit{it}}+{\beta }_4\mathit{FDISER}{V}_{\mathit{it}}+{\beta }_2\mathit{FDIPRI}{M}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(2) through (4(4) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{HTFDISe}{c}_{\mathit{it}}+{\beta }_3\mathit{MHTFDISe}{c}_{\mathit{it}}+{\beta }_4\mathit{MLTFDISe}{c}_{\mathit{it}}+{\beta }_5\mathit{LTFDISe}{c}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(4) ).

Before we do that, we briefly comment on the control variables. Domestic credit is statistically significant and has a negative sign as expected. Consistent with theoretical expectations, inflation has a negative sign but is mostly statistically insignificant except for Equation 4.(4) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{HTFDISe}{c}_{\mathit{it}}+{\beta }_3\mathit{MHTFDISe}{c}_{\mathit{it}}+{\beta }_4\mathit{MLTFDISe}{c}_{\mathit{it}}+{\beta }_5\mathit{LTFDISe}{c}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(4) Similarly, a faster growth in population has a negative and significant impact on EG. However, the result concerning Trade is surprising. Theoretically and empirically, we anticipated trade to have a positive relationship with EG. Trade serves as an indicator of openness and is expected to enhance growth. The results in Table 4 confirm this positive relationship, but only for models 1 and 2 where the coefficient of trade is positive but only significant in model 1. In models 3 and 4, the trade coefficient is negative but statistically insignificant. Statistical insignificance may be interpreted as no relationship between trade and growth. As expected, government expenditure has a negative and statistically significant sign. Domestic investment (gfcf) has a positive impact on growth and is mostly significant (except model 1). Finally, human capital (hc) has a positive relationship with growth but is statistically insignificant. In general, our control variables are significant and have the expected sign.¹⁷

Table 4. Total FDI and growth, comparative view.

	GMM (non zero)	FE (non zero)	GMM (full sample)	Fixed Effects (full sample)
FDI (Total)/GDP	0.0511	0.0321*	−0.00489**	−0.00223

t statistics in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

Equation (2)(2) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_3\mathit{FDISE}{C}_{\mathit{it}}+{\beta }_4\mathit{FDISER}{V}_{\mathit{it}}+{\beta }_2\mathit{FDIPRI}{M}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(2) , which features FDI decomposition broadly by sector, indicates that there is a highly significant and positive association between FDI in the manufacturing sector and EG. This result is in line with a good part of the earlier literature on the subject. FDI in services is positive but not significant, while in primary sector, FDI is negative and slightly significant (at 10%) confirming the presence of heterogeneity across sectors.

As already discussed earlier, the main goal of this study is not to arrive at a final settlement in the debate on the growth-fostering role of FDI but rather to investigate the role of technology (if any) in the FDI-EG relationship. To this end, we decompose FDI by technological content as explained in Section III and run regressions (3) and (4). As already mentioned, we do this mindful of the argument that technology may be one of the main channels through which FDI may benefit recipient countries. Regression (3) features the full decomposition. Now, with respect to the secondary sector, the FDI/GDP ratios at both ends of the technology spectrum are significant and positive, indicating a positive association between FDI of these types and EG.

The coefficient attached to HT FDI Sec is 1.680, that is, an increase in HT FDI Sec by 100% points (by 1) increases the growth rate of GDP per capita by 168% points. Obviously, for a correct interpretation, we need to rescale to account for the small magnitude of the relevant variables. Hence, for example, a 10% increase in the mean value of HT FDI Sec for our sample, from 0.05% to 0.055%, will be associated with an increase of 0.084% percentage points in the growth rate of GDP per capita (obtained by multiplying the original coefficient by the actual increase in percentage points, 0.005). Similar interpretations can be given to other significant coefficients in the regression. Regression (3) points to the very important role that HT FDI Sec may have in the FDI-EG relationship. As we move down the technology spectrum, the coefficients lose significance, only to return significantly (at the 1% level) at the other end of the spectrum with a coefficient value of 1.155. This result points to LT FDI Sec as another important determinant of EG. The same regression also shows the importance of technology in the Services sector where the coefficient of HT FDI Serv, with a value of 0.0602, is highly significant and positive.

In regression (4) which focuses on the manufacturing sector only, the results are almost identical to those of regression (3). Hence, an increase in HT FDI Sec by 100% points would now have a slightly weaker impact on EG by raising the growth rate of GDP per capita by 145% points. In other words, a 10% increase in HT FDI Sec would increase EG by 0.0072%. As in Equation 3(3) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{HTFDISe}{c}_{\mathit{it}}+{\beta }_3\mathit{MHTFDISe}{c}_{\mathit{it}}+{\beta }_4\mathit{MLTFDISe}{c}_{\mathit{it}}+{\beta }_5\mathit{LTFDISe}{c}_{\mathit{it}}+{\beta }_6\mathit{HTFDISer}{v}_{\mathit{it}}+{\beta }_7\mathit{LTFDIServ}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(3) , neither MHT FDI Sec nor MLT FDI Sec would have a significant impact on EG, but LT FDI Sec would (at the 1% significance level).¹⁸ As a robustness check, we also ran all four equations using FE estimation. These results are reported in Appendix B, Table B5.¹⁹

Digging deeper: zero values, stratification by income and by geography

One important remark is in order with respect to the FDI/GDP ratios. When decomposing FDI by type, the data on FDI/GDP ratios include several zero values, particularly with respect to FDI with high technological content. One simple way of dealing with this issue is to drop the zero values from the sample for the variable of concern. In spite of its apparent simplicity and seeming crudeness, this strategy may not be clearly inferior to a number of alternative strategies that entail the use of alternative estimators [Kareem and Kareem 2019]. This is particularly true when the variable featuring the zero values is not the dependent variable but one (or more) of the independent regressors as in our case He et al. (2014). For this reason, in this study we chose to give baseline status to the results after dropping the zero values in the FDI/GDP ratios but also to retain and report results for the full dataset inclusive of the zero values. In this sub-section, instead of offering a detailed description of all the results for the full sample, we chose to only underline the main differences in results between the full sample and the baseline case with non-zero countries. Accordingly, results of GMM and FE estimates (for the full sample) are provided in Appendix Table B6 and B8, respectively.

In Equation (1)(1) $g_{\mathit{it}}={\beta }_0+{\beta }_1{g}_{\mathit{it}-1}+{\beta }_2\mathit{FD}{I}_{\mathit{it}}+\mathit{\gamma }{\mathit{A}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$(1) , FDI is now negative and significant in GMM regression (but not in FE regression). This is most likely due to the fact that, in contrast with the baseline case, the manufacturing sector FDI now does not have a significant impact on EG. It is remarkable, however, that after diversifying FDI by technology, there is no longer any major difference between the baseline results and the full-sample regressions: in all cases both High-Tech FDI and Low-Tech FDI have a significant and positive impact on EG, whereas both middle categories, Middle-High-Tech and Middle-Low-Tech FDI fail to exhibit any relationship with EG. As already pointed out in the introduction, this U-shaped relationship between different FDI types (by technology) and EG is remarkably robust throughout different estimation techniques (GMM or FE) as well as different sampling strategies (only non-zero countries versus full sample). Finally, to control for heterogeneity in our sample, we stratified countries by geography and by income. With the former stratification, the sample was shrunk to contain European countries only. Europe was the only geographical area that guaranteed a sufficient number of countries to keep the analysis meaningful. As for the stratification by income, we restricted the sample to high-income countries only. Once again, we could not give consideration to other income groups separately, as this would have shrunken the sample size too far. For the same reason in all these further stratifications, we did not drop countries with zero values. Both Europe and the high-income group returned results that are by and large similar to those already obtained when all countries were considered. Since there is an overlap between the European sample and the high-income sample, only GMM and FE estimation results for the high-income countries sample are reported in Appendix Table B7 and B9, respectively. Results concerning the European sample are available upon request.

In particular, we again find no evidence of any effect of total FDI on EG, but we do find that manufacturing FDI has a positive impact on EG, while the effect of primary FDI on EG is negative. The effect of FDI types on EG after classifying FDI by technology are also confirmed along the same lines as in the sample with all countries, though in these instances of Europe and the high-income group, the positive association of High-Tech FDI and Low-Tech FDI with EG is less significant than before. The fall in the significance level, which is now between 5% and 10%, is perhaps due to a weaker impact, technology spillovers might have on well-developed economies such as those in Europe and those classified as high income. Finally, the signs and significance levels for all control variables are the same as in the full sample including the whole world.

VI. Discussion

In view of the above remarks, our results are interesting and contain more than one element of surprise. First, we do not find any statistically significant relationship between aggregate FDI and growth in the host country. This is somewhat surprising already since this finding is not in line with many past studies on the same topic. There are, however, quite a number of studies that, like ours, fail to spot a relationship between aggregate FDI and growth, hence our result is not entirely unexpected.²⁰ We attribute the lack of a relationship to the heterogeneity of effects on EG across different FDI types that manifests itself in the classification by sector.

The evidence gathered after classifying FDI by technological content reinforces this argument. This evidence points to the existence of a U-shaped relationship between different FDI types and growth, which is robust across all different estimation techniques as well as all sampling strategies. To display this robustness most clearly, we decided to display the main results once again in comparative format across estimation techniques and sampling methods. This is done in Tables 4, 5 and Table 6.

Table 5. FDI by sectors and growth, comparative view.

	GMM (non zero)	FE (non zero)	GMM (full sample)	Fixed Effects (full sample)
FDI (Sec)/GDP	0.297***	0.220*	0.0591	0.0961***
FDI (Serv)/GDP	0.000329	−0.0487**	0.00109	0.00623
FDI (Prim)/GDP	−0.0473*	−0.0971***	−0.0246*	−0.0103

t statistics in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

Table 6. FDI by technology and growth, comparative view.

	GMM (non zero)	FE (non zero)	GMM (full sample)	Fixed Effects (full sample)
HT FDI (Sec)/GDP	1.680***	2.042***	1.193**	1.694***
MHT FDI (Sec)/GDP	−0.0731	−0.229	0.0115	−0.193
MLT FDI (Sec)/GDP	−0.0600	−0.114	0.0283	−0.162
LT FDI (Sec)/GDP	1.155***	0.607**	1.013***	0.650**
HT FDI (Serv)/GDP	0.0602***	0.0594***	0.0145	0.0101
LT FDI (Serv)/GDP	−0.0574	−0.0389	−0.0223***	−0.0208**

t statistics in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

Table 4 concerns total FDI, while Table 5 is about FDI by sector of activity, and finally Table 6 regards FDI by technological content. No further detailed comment is provided here as these results were amply illustrated earlier on in the text.

This evidence shows that FDI affects growth by playing its role as a vehicle for technology transfers to the host country. However, once the technological content drops, as for the intermediate categories of MHT FDI Sec and MLT FDI Sec, we cannot find a statistically significant positive relationship with growth. Clearly, with respect to these FDI types, the technology spillover involved is weaker and perhaps not sufficient to create that positive association between FDI and economic growth. Even more striking is that the positive relationship between FDI and growth resurfaces when a flow of LT FDI Sec is involved.

To explain the U-shaped relationship that emerges in our study between different FDI types and economic growth, we conjecture the following role for technology: different FDI types seem to be carriers of different effects. The positive effects probably include technology spillovers that can be interpreted broadly as inclusive of new managerial skills and novel production processes. These effects will work most strongly when FDI is characterized by high technological content.²¹ The spillover effects may become weaker as the technological content decreases, allowing for other, perhaps more negative influences to prevail. For medium-high and medium-low tech FDI, foreign investment carried out by large multinational enterprises may crowd out existing smaller local firms in at least two ways. First, the larger multinational firms, by exploiting their larger size, enjoy scale economies that can raise entry barriers, which in turn change the competitive structure of the market in which they enter. The new, higher barriers to entry may push pre-existing local firms out of business and may discourage new ones from entering the market.²² Second, the multinational firms may also be able to obtain privileged access to credit, particularly if local credit markets are less developed and more shallow. As a result, local firms may find it more difficult to finance their business.²³ These two factors, higher barriers to entry requiring more funding and rationed access to credit, which reduces funding, jointly have the potential to crowd local firms out of the market in which they operate after the entry of larger multinational enterprises. If a sufficient number of local firms succumb, the net effect of foreign direct investment on the growth rate of the recipient economy may become unclear or even negative.²⁴ However, when its technological content is lowest, we again find strong evidence of a positive impact of FDI on the EG of the host country. Often low-tech FDI is of the efficiency-seeking type, and it is likely to be labour-intensive: For this reason its goal may be to take advantage of the availability of cheaper labour in the recipient country. FDI in the textile sector is one such example. To give this conjecture some footing in economic theory, we invoke a standard Heckscher-Ohlin argument here: the arrival of labour-intensive low-tech FDI would help disproportionately expand the production possibilities of labour-abundant host countries thereby explaining its positive impact on the EG of these countries.²⁵ In order to further verify and strengthen our results, we also run a regression of low-tech FDI alone on growth (keeping all the other FDI types out) and found a positive and significant impact of low-tech FDI on growth in the GMM model for the non-zero baseline case.²⁶

The above discussion clarifies that FDI is not all equal and differentiating it by technological content is crucial. Both high tech and low tech FDI hold the highest promise of fostering economic growth in the host economy. Policy makers of countries interested in attracting FDI should strive to focus on the high technology and low technology sectors and give less encouragement to investment flowing into medium-high-tech and medium-low-tech sectors.

VII. Conclusions and suggestions for future work

This study exploits the increasing level of detail in FDI data that is slowly but steadily becoming available to bring to the fore the role played by technology in the FDI-EG relationship. Despite all the efforts and a very large body of literature, a certain amount of controversy still lingers on about the existence and direction of causality in this relationship. However, the contribution of this study is rather clear: FDI types are not all equal in affecting the economy of the host country. Indeed, a particularly effective way to reveal the heterogeneity of different FDI types in this respect is to classify them by technological content and track the different impact each type has on the EG of the host country. In this respect, our results do not leave room for much doubt: technology plays a crucial role in the growth-fostering function of FDI. This study shows that this result is robust across different estimation techniques as well as across different sampling strategies. It is shown that the conclusions reached on the matter did not change whether GMM estimation or FE estimation was used. Likewise, conclusions were the same whether countries with zero values were dropped (baseline case) or not. We also found that low-tech FDI fosters EG too though probably through different channels.

The results have important policy implications. However, though pursuing GDP growth is important, there is now a broad consensus that this is not the only measure of the well being of countries. In this regard, the United Nations has produced a list of sustainable development goals (SDG). Specifically, when devising policies geared towards attracting FDI, policymakers around the world should ensure that the effects of inward FDI on the recipient countries are in line with the United Nation’s Sustainable Development Goals 8 and 9. SDG 8 calls for inclusive and sustainable economic growth, while SDG 9 is to build resilient infrastructure, promote inclusive and sustainable industrialization and foster innovation United Nations (2015). Policies designed to achieve inclusive and sustainable growth would bring important societal impacts.

This study can be extended in a number of ways. As more data become available, the sample size can be broadened. This step allows us to focus on specific geographical areas or income groups besides Europe and the high-income countries as already done here.

Another possibility would be to compare and contrast the short run and the long-run effects of different FDI types on EG. Finally, it would be interesting to probe if there is a nexus between the various FDI technological categories and the country of origin of these FDI types.

Disclosure statement

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

Notes

1 Although flows of capital experienced a drop since 2016 for developed countries, transitional and developing economies continued to enjoy an upward trend in FDI flows until 2019.

2 If there is any causality between the two, it can run both ways and there are multiple channels through which this causality may manifest itself (see, e.g. Almfraji and Almsafir, 2014, for a literature review. For causality from EG to FDI, see, e.g. Lamsiraroj and Doucouliagos (2015).

3 Seminal papers on threshold effects in the FDI – EG relationship are Borenzstein et al. (1998) for human capital, Alfaro et al. (2008) for financial development and Kose et al. (2009) for both financial and institutional development in the broader context of the relationship between financial integration and EG. Khalid and Marasco (2019) investigate if the impact of FDI on growth is dependent on the channel of integration used to attract FDI.

4 Alfaro (2003) is the pioneer study in this respect. Other relevant references are given in Section 2 on the literature review.

5 While decomposing FDI by type, we noted that the data on FDI/GDP ratios include several zero values, particularly with respect to the FDI with high technological content. Hence, in the main analysis we used a sub-sample of countries by removing those with zero values for the variable of concern. We re-estimate the model by using a full sample for consistency check.

6 A more complete discussion of these arguments is given in Section 5.2.

7 Another major distinction in FDI is horizontal versus vertical. Horizontal FDI is where multi-plant firms duplicate roughly the same activities in multiple Countries. It is believed that the bulk of FDI is horizontal rather than vertical.

8 Penn World Table 9.1

9 Indeed, in all other equations throughout this article too, the vector of control variables is always preceded by a vector of coefficients.

10 The value of the chi-square test statistic (with 7 degrees of freedom) was 67 (with p-value = 0.000) for the annual regression of equation 1. Thus the null hypothesis that the random model is appropriate is rejected. The Hausman test was also carried out for the 5-year average regressions of equation 8 (see below). Test results (not reported but available upon request) confirmed the fixed effect model as the appropriate choice in that case too.

11 Marasco (2021) takes a preliminary look at the data using this disaggregation and as such constitutes a partial exception.

12 Law and Singh (2014) and Arcand et al. (2015) found similar empirical evidence.

13 This is because while FDI data worldwide have now been collected more or less systematically for a very broad cross section of countries for at least 30 years, and they are readily accessible from sources such as the UNCTAD, data collection on FDI differentiated by sector of industry has been much more scattered and hard to come by.

14 As explained in the document “Human Capital in PWT 9.0”.

15 Please see Section 4 for description and discussion of the model used and the control variables included in this study..

16 In general, it is true that small N can introduce a sample bias in GMM estimation. This will depend on whether N is big enough to accommodate all relevant instruments. In our case N = 28 and we have 14, 16, 20 and 18 instruments respectively in our regressions. To further address this concern, we also tested the robustness of our results by shortening T to less than N (from 31 to 25 by starting the sample from 1995 instead of 1989) and found no change in our results. We also run fixed effects regressions for all four specifications and the results were in line with GMM.

17 We also performed some additional tests to ensure that model estimates are valid. First, we conducted VIF tests for multicollinearity for all four models. These results are reported in Appendix table B3. These results do not suggest any multicollinearity. Second, we performed panel unit root tests to check any possible stationarity. Again, the results, in general, reject non-stationarity. These results are reported in Appendix table B4.

18 See footnote 21.

19 As a further robustness check, we also grouped data into 5-years intervals and then re-estimated the model using both GMM and fixed effects regressions for equations (1) through (4) on the corresponding averages. For the sake of brevity, we do not report these results in the article. However, these results are available from corresponding author upon request. It is important to note, though, that grouping the data into 5-years average significantly reduced the number of observations and the power of the tests. Nonetheless, the results confirm the results obtained from the annual regressions. In particular, after we disaggregate data by technological content (equation 3a), we have confirmation that, in the manufacturing sector, the positive association between FDI at both ends of the technology spectrum (high and low) and EG remains significant, at the 1% and 5% levels respectively, while the two intermediate categories of FDI (medium-high-tech and medium-low-tech) remain statistically insignificant.

20 Almfraji and Almsafir (2014). Also see Forte and Moura (2013).

21 Borenzstein et al. (1998) and De Mello (1997) are some of the early seminal articles on technology spillovers of FDI. For more recent contributions, see literature review by Shushu Feng (2020).

22 For example, Kokko and Thang (2014) find evidence of crowding-out while documenting the impact of FDI on the survival of domestic private firms in Vietnam.

23 Harrison and McMillan (2003) in a study centered on Ivory Coast, a country with market imperfections, find that local firms are more credit-constrained than foreign firms and that borrowing by foreign firms exacerbates the credit constraints of domestic firms

24 Marasco (2021) contains the same discussion of these effects .

25 Nunnenkamp and Spatz (2003) is an example of an earlier study advancing this kind of argument.

26 We run individual regressions for the other FDI types too which confirmed the results of Table 3. These results are reported in Appendix table B10.

Appendix A

Table A1. List of countries.

1. Argentina	17. France^(N, H)	33. Pakistan
2. Armenia^(N)	18. Germany^(N, H)	34. Paraguay
3. Australia^(H)	19. Greece^(N, H)	35. Peru
4. Austria^(N, H)	20. Hong Kong^(H)	36. Poland^(N, H)
5. Bolivia	21. Hungary^(N, H)	37. Slovak Republic^(N, H)
6. Brazil^(N)	22. India^(N)	38. Slovenia^(H)
7. Canada^(H)	23. Indonesia^(N)	39. Spain^(N, H)
8. Costa Rica	24. Ireland^(N, H)	40. Sweden^(H)
9. Croatia^(N, H)	25. Italy^(N, H)	41. Switzerland^(N, H)
10. Cyprus^(H)	26. Japan^(N, H)	42. Thailand
11. Czech Republic^(H)	27. Korea^(N)	43. Trinidad and Tobago^(H)
12. Denmark^(N, H)	28. Latvia^(H)	44. Tunisia
13. Dominican Republic	29. Lithuania^(N, H)	45. Turkey^(N)
14. El Salvador	30. Mexico^(N)	46. United Kingdom^(N, H)
15. Estonia^(H)	31. Netherlands^(N,H)	47. United States^(N, H)
16. Finland^(N, H)	32. Norway^(N, H)	48. Uruguay^(N, H)

This table represents the full sample of countries. The superscript ‘N’ stands for the non-zero subsample, while ‘H’ stands for high income.

Appendix B

Table B1. Summary statistics.

	Obs	Mean	Std. Dev.	Min	Max
$g_{\mathit{it}}$	1,442	2.337	3.672	−40.74	24.667
FDI	1,192	6.637	31.651	−29.762	349.476
FDISEC	1,149	0.657	3.614	−12.658	48.865
FDISERV	1,356	4.791	24.513	−52.856	254.17
FDIPRIM	1,063	1.142	8.911	−68.402	101.179
HTFDISec	1,103	0.065	0.449	−2.178	11.254
MHTFDISec	1,117	0.1268	0.547	−7.452	6.693
MLTFDISec	1,152	0.1586	1.719	−2.351	40.802
LTFDISec	1,164	0.1139	0.6018	−2.764	10.712
HTFDIServ	1,355	2.389	13.37	−53.596	246.055
LTFDIServ	1,375	2.370	14.201	−10.277	171.99
d_credit	1,401	72.864	51.307	0.1858	253.57
inflation	1,442	27.137	259.66	−27.63	6261.24
pop_growth	1,487	0.697	0.863	−5.814	4.439
trade	1,447	80.648	52.786	13.75	442.62
gov_exp.	1,445	16.480	4.838	2.976	27.935
gfcf	1,445	22.226	5.056	10.524	44.345
hc	1,384	2.867	0.552	1.351	3.794

Table B2. Correlation table (full correlation data – inclusive of control variables).

	${\mathrm{g}}_{\mathrm{it}}$	FDI	FDI SEC	FDI SERV	FDI PRIM	HT FDISec	MHT FDISec	MLTFDISec	LT FDISec	HT FDIServ	LT FDIServ	d_credit	inflation	pop_growth	trade	gov_exp	gfcf	hc
${\mathrm{g}}_{\mathrm{it}}$	1
FDI	−0.04	1
FDISEC	0.03	0.21	1
FDI SERV	−0.04	0.93	0.11	1
FDI PRIM	−0.03	0.83	0.04	0.68	1
HTFDI Sec	0.09	−0.02	0.81	−0.10	0.14	1
MHT FDISec	−0.01	0.02	0.57	−0.01	0.04	0.16	1
MLT FDISec	0.03	0.06	0.94	0.00	−0.03	0.10	0.05	1
LT FDISec	0.02	0.03	0.77	−0.05	0.06	0.13	0.18	0.81	1
HTFDIServ	−0.03	0.73	0.12	0.88	0.46	−0.15	0.00	−0.01	−0.09	1
LT FDIServ	−0.03	0.92	0.10	0.90	0.74	−0.04	−0.02	0.00	−0.01	0.57	1
d_credit	−0.22	−0.05	−0.01	0.00	−0.07	0.03	0.06	−0.02	0.01	0.07	−0.06	1
inflation	−0.13	−0.01	−0.01	−0.02	0.00	−0.01	−0.01	0.00	−0.01	−0.02	−0.01	−0.07	1
pop_growth	−0.10	−0.02	0.04	−0.05	0.04	0.07	−0.01	0.04	0.07	−0.05	−0.03	−0.05	0.03	1
Trade	0.10	0.01	0.02	0.03	0.01	0.10	0.01	0.00	0.01	0.04	0.01	0.22	−0.07	−0.21	1
gov_exp.	−0.19	−0.14	0.01	−0.10	−0.15	−0.02	0.03	0.00	0.06	−0.05	−0.13	0.25	−0.10	−0.46	0.00	1
gfcf	0.25	−0.17	−0.06	−0.17	−0.13	0.03	−0.03	−0.01	−0.03	−0.15	−0.16	0.18	−0.07	−0.06	0.13	−0.01	1
hc	−0.07	−0.19	−0.01	−0.14	−0.20	0.04	0.07	0.01	0.00	−0.07	−0.16	0.49	−0.10	−0.50	0.23	0.55	0.16	1

Table B3. Variance inflation factor.

Variable	VIF (1)	VIF (2)	VIF (3)	VIF (4)
${\mathrm{g}}_{\mathrm{it}-1}$	1.44	1.44	1.43	1.40
FDI	1.12
FDISEC		1.03
FDISERV		2.38
FDIPRIM		2.18
HTFDISec			1.17	1.16
MHTFDISec			1.15	1.15
MLTFDISec			1.04	1.04
LTFDISec			1.23	1.23
HTFDIServ			1.53
LTFDIServ			1.58
d_credit	1.56	1.59	1.62	1.56
İnflation	1.04	1.05	1.05	1.04
pop_growth	1.65	1.78	1.88	1.78
trade	1.15	1.17	1.18	1.18
gov_exp	1.69	1.72	1.79	1.79
gfcf	1.33	1.34	1.35	1.31
hc	2.19	2.28	2.26	2.23
Mean VIF	1.46	1.63	1.45	1.41

Table B4. Panel unit root test results.

Variable	Fisher	Fisher with trend	Pesaran	Pesaran with trend
${\mathrm{g}}_{\mathrm{it}-1}$	613.651***	482.569***	−13.271***	−10.230***
FDI	299.405***	322.218***	−8.729***	−5.897***
FDISEC	551.514***	526.926***	−9.243***	−6.770***
FDISERV	354.564***	371.125***	−9.487***	−8.387***
FDIPRIM	419.013***	314.774***	−10.616***	−7.262***
HTFDISec	336.895***	300.227***	3.481	3.528
MHTFDISec	396.772***	342.435***	−1.437*	−0.586
MLTFDISec	721.743***	478.738***	−8.894***	−7.058***
LTFDISec	649.283***	538.801***	−8.214***	−6.142***
HTFDIServ	483.568***	427.311***	−10.875***	−8.591***
LTFDIServ	570.807***	557.525***	−11.674***	−12.665***
d_credit	79.171	73.613	1.585	4.808
İnflation	547.238***	468.688***	−17.125***	−14.635***
pop_growth	277.567***	419.163***	−6.974***	−7.781***
trade	64.654	123.078**	−1.798**	0.714
gov_exp	174.078***	131.018***	−2.568***	1.591
gfcf	222.984***	227.223***	−1.902**	−0.717
hc	677.131***	145.507***	9.551	16.648

For first generation tests, we have employed the Fisher test as developed by Maddala and Wu (1999); while for the second generation tests, Pesaran (2007) panel unit root test has been used. Null for both tests is that the series have unit root.

Table B5. Fixed effects in non-zero countries.

	(1)	(2)	(3)	(4)
${\mathrm{g}}_{\mathrm{it}-1}$	0.142***	0.110	0.0818	0.114
	(2.62)	(1.62)	(1.27)	(1.61)
FDI	0.0321*
	(1.91)
FDISEC		0.220*
		(1.93)
FDISERV		−0.0487**
		(−2.32)
FDIPRIM		−0.0971***
		(−4.27)
HTFDISec			2.042***	1.698***
			(4.49)	(4.74)
MHTFDISec			−0.229	−0.214
			(−1.35)	(−1.48)
MLTFDISec			−0.114	−0.142
			(−0.32)	(−0.33)
LTFDISec			0.607**	0.566**
			(2.40)	(2.13)
HTFDIServ			0.0594***
			(4.18)
LTFDIServ			−0.0389
			(−0.42)
d_credit	−0.0236**	−0.0292***	−0.0298***	−0.0282***
	(−2.56)	(−3.31)	(−3.47)	(−3.06)
inflation	0.00803	−0.00949	−0.0117	−0.119
	(0.18)	(−0.20)	(−0.26)	(−1.60)
pop_growth	−0.924***	−1.656***	−1.686***	−1.751***
	(−3.75)	(−8.06)	(−6.73)	(−5.65)
trade	0.0102	0.0256	0.0247	0.0147
	(0.67)	(1.48)	(1.55)	(0.91)
gov_exp	−0.624***	−0.461***	−0.472***	−0.494***
	(−3.35)	(−3.54)	(−3.54)	(−3.58)
gfcf	0.0780	0.285***	0.277***	0.243***
	(0.88)	(4.81)	(5.11)	(3.74)
hc	0.0801	−0.328	−0.938	−1.638
	(0.06)	(−0.24)	(−0.80)	(−1.38)
_cons	12.62**	6.221	8.367**	12.66**
	(2.41)	(1.52)	(2.19)	(2.64)
N	618	544	523	531

t statistics in parentheses¸* p < 0.1, ** p < 0.05, *** p < 0.01.

Table B6. Dynamic system GMM full sample.

	(1)	(2)	(3)	(4)
${\mathrm{g}}_{\mathrm{it}-1}$	0.277***	0.255***	0.239***	0.242***
	(5.93)	(5.21)	(5.12)	(5.63)
FDI	−0.00489**
	(−2.46)
FDISEC		0.0591
		(1.07)
FDISERV		0.00109
		(0.23)
FDIPRIM		−0.0246*
		(−1.80)
HTFDISec			1.193**	1.131**
			(2.41)	(2.26)
MHTFDISec			0.0115	0.0270
			(0.05)	(0.11)
MLTFDISec			0.0283	0.0141
			(0.09)	(0.05)
LTFDISec			1.013***	0.973***
			(3.49)	(3.16)
HTFDIServ			0.0145
			(1.68)
LTFDIServ			−0.0223***
			(−4.35)
d_credit	−0.0137***	−0.0134***	−0.0155***	−0.0153***
	(−4.88)	(−4.42)	(−4.84)	(−4.71)
Inflation	−0.000830***	−0.000806***	−0.000874***	−0.000904***
	(−8.26)	(−7.32)	(−9.33)	(−9.44)
pop_growth	−0.998***	−1.171***	−1.392***	−1.283***
	(−5.79)	(−6.19)	(−7.19)	(−6.37)
Trade	0.00531**	0.00447*	0.00283	0.00284
	(2.03)	(1.93)	(1.63)	(1.49)
gov_exp	−0.227***	−0.227***	−0.253***	−0.228***
	(−6.01)	(−5.39)	(−6.00)	(−5.38)
Gfcf	0.114***	0.149***	0.140***	0.143***
	(3.13)	(5.05)	(4.82)	(4.97)
Hc	0.242	0.0646	0.112	0.172
	(0.74)	(0.19)	(0.30)	(0.45)
_cons	3.674***	3.567***	4.330***	3.567***
	(3.48)	(3.27)	(3.85)	(3.38)
Instruments	14	16	20	18
N	1062	954	906	931

t statistics in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

Table B7. Dynamic system GMM high-income countries.

	(1)	(2)	(3)	(4)
${\mathrm{g}}_{\mathrm{it}-1}$	0.333***	0.331***	0.294***	0.274***
	(6.78)	(6.22)	(5.91)	(6.02)
FDI	0.0201
	(0.96)
FDISEC		0.273***
		(2.83)
FDISERV		0.00633
		(1.55)
FDIPRIM		−0.0476***
		(−4.98)
HTFDISec			1.160**	1.045*
			(2.10)	(1.95)
MHTFDISec			−0.00164	0.00475
			(−0.01)	(0.02)
MLTFDISec			0.0757	0.0727
			(0.23)	(0.22)
LTFDISec			1.019***	0.970***
			(3.14)	(2.78)
HTFDIServ			0.0179
			(1.49)
LTFDIServ			−0.0127
			(−0.13)
d_credit	−0.0125***	−0.0107***	−0.0139***	−0.0131***
	(−3.61)	(−4.39)	(−4.87)	(−4.79)
inflation	0.0617**	0.0445	0.0454	0.0556
	(2.52)	(1.37)	(1.34)	(1.69)
pop_growth	−0.866***	−1.025***	−1.173***	−1.151***
	(−3.47)	(−5.10)	(−5.98)	(−5.80)
trade	0.00474**	0.00450**	0.00270	0.00316
	(2.00)	(2.14)	(1.48)	(1.57)
gov_exp	−0.202***	−0.186***	−0.217***	−0.211***
	(−3.66)	(−3.69)	(−4.25)	(−4.17)
gfcf	0.0440	0.0988***	0.0908***	0.0951***
	(0.89)	(3.57)	(3.12)	(3.37)
hc	0.267	−0.213	0.0193	−0.194
	(0.64)	(−0.48)	(0.04)	(−0.34)
_cons	4.249*	4.137**	4.615**	5.023**
	(1.93)	(2.35)	(2.25)	(2.44)
Instruments	14	16	20	18
N	689	624	596	603

t statistics in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

Table B8. Fixed effects (full sample).

	(1)	(2)	(3)	(4)
${\mathrm{g}}_{\mathrm{it}-1}$	0.134***	0.104***	0.0966**	0.102***
	(3.42)	(2.49)	(2.37)	(2.53)
FDI	−0.00223
	(−0.74)
FDISEC		0.0961***
		(3.30)
FDISERV		0.00623
		(1.57)
FDIPRIM		−0.0103
		(−0.79)
HTFDISec			1.694***	1.646***
			(5.03)	(4.80)
MHTFDISec			−0.193	−0.173
			(−1.04)	(−0.93)
MLTFDISec			−0.162	−0.154
			(−0.65)	(−0.63)
LTFDISec			0.650**	0.639**
			(2.31)	(2.31)
HTFDIServ			0.0101
			(1.65)
LTFDIServ			−0.0208** (−2.49)
d_credit	−0.0302***	−0.0350***	−0.0397***	−0.0384***
	(−4.55)	(−4.79)	(−5.21)	(−5.60)
inflation	−0.00126***	−0.00121***	−0.00130***	−0.00131***
	(−11.90)	(−11.54)	(−12.09)	(−11.06)
pop_growth	−1.471***	−1.938***	−1.944***	−1.888***
	(−6.41)	(−8.30)	(−6.47)	(−6.33)
trade	0.0142*	0.0216**	0.0221***	0.0182***
	(1.68)	(2.41)	(3.32)	(2.92)
gov_exp	−0.534***	−0.523***	−0.531***	−0.498***
	(−5.48)	(−5.89)	(−6.13)	(−5.82)
gfcf	0.185***	0.287***	0.267***	0.262***
	(2.84)	(6.71)	(6.13)	(6.09)
hc	1.068	1.041	0.780	0.724
	(0.90)	(0.84)	(0.63)	(0.66)
_cons	5.977*	3.665	5.140*	5.019*
	(1.86)	(1.26)	(1.78)	(1.89)
N	1062	954	906	931

t statistics in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

Table B9. Fixed effects high-income countries.

	(1)	(2)	(3)	(4)
${\mathrm{g}}_{\mathrm{it}-1}$	0.162***	0.159***	0.135***	0.126**
	(3.85)	(3.01)	(2.64)	(2.48)
FDI	0.00749
	(0.54)
FDISEC		0.205*
		(1.87)
FDISERV		0.00633
		(1.06)
FDIPRIM		−0.0376***
		(−3.45)
HTFDISec			1.596***	1.537***
			(4.33)	(4.38)
MHTFDISec			−0.197	−0.198
			(−1.12)	(−1.21)
MLTFDISec			−0.184	−0.200
			(−0.81)	(−0.85)
LTFDISec			0.558*	0.571**
			(2.00)	(2.11)
HTFDIServ			0.0173*
			(1.86)
LTFDIServ			−0.0175
			(−0.26)
d_credit	−0.0277***	−0.0300***	−0.0339***	−0.0322***
	(−4.02)	(−3.95)	(−4.12)	(−4.52)
inflation	0.0582	0.0341	0.0319	0.0350
	(1.99)	(0.96)	(0.99)	(1.10)
pop_growth	−1.455***	−1.673***	−1.676***	−1.807***
	(−5.86)	(−8.78)	(−6.57)	(−6.05)
trade	0.0197*	0.0249**	0.0275***	0.0251***
	(1.80)	(2.29)	(3.78)	(3.53)
gov_exp	−0.443***	−0.429***	−0.470***	−0.473***
	(−3.57)	(−4.01)	(−4.01)	(−4.27)
gfcf	0.204***	0.271***	0.256***	0.249***
	(2.71)	(5.54)	(4.88)	(4.72)
hc	0.0952	−0.627	−1.748	−1.428
	(0.06)	(−0.47)	(−1.21)	(−0.98)
_cons	6.495	6.808	11.43*	10.80*
	(1.09)	(1.37)	(2.05)	(1.97)
N	689	624	596	603

t statistics in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

Table B10. Secondary sector FDI classified by technology-dynamic system GMM (non-zero countries).

	(4.1)	(4.2)	(4.3)	(4.4)
${\mathrm{g}}_{\mathrm{it}-1}$	0.212***	0.278***	0.302***	0.271***
	(4.67)	(5.69)	(3.87)	(5.35)
HTFDISec	1.274**
	(2.24)
MHTFDISec		0.163
		(0.64)
MLTFDISec			0.188
			(0.32)
LTFDISec				0.439**
				(2.27)
d_credit	−0.0217***	−0.0175***	−0.0160***	−0.0172**
	(−4.35)	(−3.74)	(−2.99)	(−3.46)
inflation	−0.140**	−0.137**	−0.149**	−0.0708
	(−2.01)	(−1.93)	(−2.23)	(−1.19)
pop_growth	−1.228***	−1.157***	−1.135***	−1.024***
	(−3.71)	(−4.59)	(−4.38)	(−3.37)
trade	0.00181	0.00362	0.00333	0.00643
	(0.32)	(0.75)	(0.74)	(1.11)
gov_exp	−0.274***	−0.243***	−0.234***	−0.255***
	(−3.83)	(−4.57)	(−4.48)	(−3.66)
gfcf	0.110	0.112**	0.115**	0.0457
	(1.43)	(1.91)	(1.96)	(0.63)
hc	0.154	−0.172	−0.395	0.146
	(0.24)	(−0.29)	(−0.73)	(0.26)
_cons	6.362**	6.153**	6.516***	6.362**
	(2.41)	(2.46)	(2.82)	(2.23)
N	542	581	596	613
Instruments	15	15	15	15
AR(1) p-value	0.001	0.000	0.000	0.000
AR(2) p-value	0.029	0.046	0.055	0.027
AR(3) p-value	0.214	0.146	0.118	0.327
Sargan	0.813	0.399	0.408	0.933
Hansen	0.700	0.226	0.215	0.872

Sargan and Hansen p-values are reported in the table. Both Sargan and Hansen test the null that instruments as a group are exogenous. In all our models, we fail to reject the null hypotheses, thereby indicating that the instruments are valid.