The Determinants and Growth Effects of Foreign Direct Investment: A Comparative Study

1. Introduction

Economic openness attracts foreign direct investment (FDI), which not only brings in capital but also transfers knowledge and technology to host countries that would not otherwise have access to it. As investment from other countries links FDI host countries to the global market, global value chains provide opportunities for the countries to diversify their exports and expedite their integration into the global economy, as noted by Dollar (2017) in his Brookings feature article.¹ Through global value chains, host countries are exposed to broader economic activities in widely separated locations that are interconnected. Local firms may benefit from substantial economies of scale, strengthening their competitive global market positions. The World Bank (2021) underscores the crucial role of policies that attract and facilitate FDI entry in enhancing investment promotion capacity. These policies lead to spillover effects through supply chain linkages, which bring new capital, technology, and knowledge into business practices, enabling local firms to better withstand the onslaught of global, scalable firms. Serpa and Krishnan (2018) indicate that firms interacting within a supply chain affect each other’s productivity, and productivity spillovers are more pronounced when the supply chain is more concentrated. Thus, countries that gain from productivity improvements accrued from FDI inflow are more likely to pursue policies favoring economic openness.

Productivity spillovers from multinational activities to local firms come with knowledge and technology to an extent that is non-excludable and non-rival in production. Blomstrom and Kokko (1998) identify three main channels where productivity spillovers occur from FDI inflows to host countries, including demonstration effects through imitation of newer technologies, competition effects through competitive pressure forcing local firms to improve their production efficiency, and knowledge spillover effects through skilled labor mobility. Developing countries tend to attract a sizeable share of FDI through comparative advantage mechanisms and more substantial capabilities to pursue economic growth. For these countries to achieve productivity spillover gains, FDI inflows need to be stable and complementary to the labor force structure within the local economy, where FDI inflows would bring the capital formation to an extent sufficient to absorb the growth in the labor force, adjusting labor and capital to the equilibrium where productivity spillover effects can be prevalent. Consequently, countries that better capture the effect of demonstration or contagion are more likely to benefit from FDI inflows.

One aspect of low wages in developing countries can be attributed to the lack of productive capital, which is the primary factor driving structural unemployment. For wages in these countries to rise, capital investments need to be sufficient to reach the steady state growth rate of labor supply to boost labor productivity, and FDI can be a source for capital investment to expand to the level where productivity will grow, rendering in terms of skill development and productivity spillover. Consequently, many developing countries turn to FDI attraction seeking foreign capital to mitigate capital constraints and fill the resource gap in their quest for economic development. In their empirical studies, Lipsey and Sjoholm (2004) and Hijzen et al. (2013) put forth a view that multinational corporations (MNCs) tend to offer wage premia compared to their domestic counterparts with which the gap between wages in local firms and MNCs forces domestic firms to face competitive pressure in the labor markets for higher wages. MNCs share knowledge or technology with their suppliers, which they then implement to increase productivity. In return, MNCs would demand higher-quality goods and resources from suppliers, forcing local firms to be more productive. Thus, productivity spillovers occur as FDI creates benefits that local suppliers and other firms share in the local markets.

This study investigates how FDI inflows can foster economic growth among countries at distinct stages of development, specifically tracing the productivity spillover effects of FDI inflows with different degrees of economic freedom and growth rates. We emphasize that FDI’s effects on productivity spillovers depend positively on labor market concentration in the host country and attempt to measure the effects of productivity spillovers by exploring the ratio of capital to the stock of labor efficiency rather than focusing on horizontal or vertical productivity spillovers as is commonly done. Our approach to productivity spillovers from a macroeconomic perspective allows us to examine the overall effects of FDI inflows for an economy rather than within a specific sector or the level of the economy where the viewpoints would be relatively segmented.

We apply the system generalized method of moments (GMM) approach proposed by Arellano and Bover (1995) and Blundell and Bond (1998) to estimate the impact of FDI inflows on economic growth through productivity spillover effects. Based on a panel of 84 countries over the period 2010-2019, the results suggest that productivity spillovers occur as FDI inflows increase labor productivity, leading to growth in real output. Economic freedom plays a pivotal role in promoting FDI inflows and productivity. However, economic freedom alone does not guarantee strong economic growth in OECD countries; it is more pronounced in enhancing economic growth in non-OECD countries. Population growth positively contributes to economic growth in OECD countries but could impede economic growth in non-OECD countries. Our results show that only economies with specific infrastructure and development would benefit more significantly from FDI inflows.

The remaining sections of the paper are organized as follows. Section 2 provides an overview of relevant literature. Section 3 describes the data and econometric methodology applied in the study. Section 4 presents a preliminary finding explaining why system GMM estimates are superior to difference GMM estimates in this study, followed by our empirical findings of the GMM estimates with interaction terms and data segmented into transition and OECD economies. The main findings are summarized with a concluding remark in the final section.

2. Literature Review

Despite extensive research on the connection between FDI inflows and economic growth, the empirical literature provides conflicting evidence regarding the existence of the productivity spillover effect of FDI inflows. Demena and van Bergeijk (2017) investigate 1450 spillover estimates from 69 empirical studies published in 1986-2013 and note the inconsistency of reported spillover findings caused by publication bias, characteristics of data, and model misspecification. This can be traced back to Blomström, Kokko, and Mucchielli (2003), who also address the empirical interpretation of the evidence that fails to explain the occurrence and essence of spillovers.

To disentangle the FDI-growth nexus, the varying results of studies have called forth a series of meta-analyses, attempting to discern the effect of productivity spillovers. A meta-analysis study by Gorg and Strobl (2001) shows evidence of publication bias and that the type of data used (industrial- or plant-level data, developing or developed country data sets, or whether the data is recent) is irrelevant. Meyer and Sinani (2009) suggest that the relationship between a country’s stage of economic development and spillovers is curvilinear. Countries with low and high levels of GDP per capita experience spillovers from FDI, but transition economies in between see a declining effect. Wooster and Diebel (2010) also present weak evidence of horizontal spillovers in developing countries, contrasting prior studies. However, the research is comparatively limited in examining spillovers to domestic firms in the same sector. Iršová and Havránek (2013) find similar results to Wooster and Diebel (2010) with no evidence of horizontal spillovers. They argued that past levels of FDI do not affect current spillovers, implying a linear effect, wherein a country saturated with FDI may still reap spillover benefits.

The extant body of literature mainly supports the idea that FDI positively impacts economic growth, with productivity spillovers primarily seen as beneficial to domestic firms through the transfer of technology and knowledge from MNCs. Amann and Virmani (2014) analyze the feedback effect of FDI on productivity and find that the impact of FDI enhancing productivity growth is greater for R&D-intensive developed countries to invest in emerging economies. Gorodnichenko, Svejnar, and Terrell (2014) present evidence to partially support Blomstrom and Kokko (1998), suggesting that domestic firms experience a faster increase in efficiency after supplying to MNCs. However, domestic firms purchasing from or competing with MNCs in the same industry do not appear to gain positive spillovers. This could be due to the cost of implementing new technology to compete with MNCs in the short run. Based on a sample of 17,625 Chinese manufacturing firms, Liu (2008) shows that FDI spillovers lower domestic firms’ productivity levels in the short run and that backward linkages are the most statistically significant channel for technology spillovers to occur, potentially due to the demand for high-quality goods supplied to MNCs by domestic firms. The result suggests that productivity spillovers from FDI are still positive and substantial overall.

Building on the channels by Blomstrom and Kokko (1998), Azman-Saini et al. (2010) examine 85 countries and generate evidence of productivity spillovers from FDI conditioning on high levels of economic freedom. They highlight how economic freedom and FDI jointly positively and significantly affect growth. However, FDI alone does not impact growth, concluding that this occurs because economic freedom facilitates FDI spillovers. Naanwaab and Diarrassouba (2016) also contend that economic freedom is a positive determinant of FDI but also note that this occurs only when economic freedom is sufficiently high, and the most critical components of economic freedom are legal structure and property rights, regulation, and freedom to trade. This is similar to Iamsiraroj and Ulubaşoğlu (2015) who find that financial development and trade openness are essential for a country to have the absorptive capacity for FDI to impact growth.² However, the effect subsides once the economic development and trade openness reach certain levels. Sambharya and Rasheed (2015) posit that economic and political freedom affects countries’ institutional environments, which can influence a firm’s decisions to enter markets. Higher economic freedom overall leads to increased FDI inflows. However, only four out of five of the Heritage Foundation’s economic freedom index components influence FDI inflows. The exception being trade and investment freedom contrasts Azman-Saini et al. (2010), who find that freedom to trade is essential.

Naanwaab and Diarrassouba (2016) show that economic freedom alone cannot attract FDI inflows in developing countries; human capital and economic freedom are jointly substantial. Human capital is critical in enhancing absorptive capacity so domestic firms can maximize productivity spillovers by utilizing foreign knowledge and technology (Blomström et al., 2003). Balasubramanyam, Sapsford, and Salisu (1999) also find that for human capital to act as a conduit for FDI to affect growth, a threshold of human capital is necessary. Similarly, Bengoa and Sanchez-Robles (2003) argue that FDI positively correlates with economic growth conditioned upon adequate levels of human capital. Fukase (2010) infers that FDI affects growth through human capital differently depending on the channel, industry, and institutions. Wooster and Diebel (2010) note that labor quality is inversely associated with spillover absorption, where higher labor quality leads to decreased productivity spillover benefits from FDI. Kristjánsdóttir (2010) implies that MNCs are less attracted by human capital and labor education levels as they originate from highly skilled labor countries. The labor cost difference between countries was a more significant determinant of FDI, as MNCs are more likely to take advantage of cost-saving labor.

Restrictive labor laws could inhibit knowledge transfer between MNCs and domestic firms through the labor movement. Using data from Vietnam and China, Vu, Gangnes, and Noy (2008) analyze the impact of FDI inflows on growth through labor in different economic sectors. They find that FDI inflows significantly and positively affect growth through indirect interaction with labor in both countries. However, the effect is not equally distributed among sectors. Higher levels of regulation could also impede the efficient allocation of resources, giving less opportunity for domestic firms to acquire funding to adopt new technologies; strong protection of property rights would entice more technology-oriented FDI to develop their R&D domestically; and freedom to trade would allow MNCs to export easily while producing goods at a lower cost in developing countries, creating competition and relationships with domestic firms in the process.

3. Data and Methodology

This study focuses on countries in distinct stages of economic development. In parts of the analysis, the distinction is made between OECD and non-OECD countries. A panel of 84 countries includes 34 OECD countries and 50 non-OECD countries over the 2010-2019 period. The variables included in the analysis are real GDP per capita, FDI inflows, economic freedom index, gross capital formation, population growth rate, labor force participation rate, and labor productivity. FDI inflows are the net inflow as a percentage of real GDP. Real GDP and gross capital formation are in constant 2015 US dollars. Labor productivity is output per employed person in constant 2015 US dollars. The population growth rate is the ratio between the annual increase in the population size and the total population. The labor force participation rate is the percentage of a country’s working-age population actively engaged in the labor market. Economic freedom is an index measuring the degree of freedom in a country comprised of five major components: 1) size of government, 2) legal structure and security of property rights, 3) access to sound money, 4) exchange with foreigners, and 5) regulation of capital, labor, and business.

Real GDP and gross capital formation were drawn from the World Bank World Development Indicator. The data on FDI inflows was obtained from the United Nations Conference on Trade and Development data center (UNCTAD). Economic freedom is the index compiled by the Fraser Institute. Population growth rate is compiled by the World Health Organization (WHO). Labor productivity and labor force participation rates were obtained from the International Labor Organization Statistics Database (ILOSTAT). Table 1 reports summary statistics for GDP growth rate, FDI inflows, and domestic capital formation in percentage of GDP, population growth rate, labor force participation rate, and labor productivity changes over the sample period based on the respective full sample, OECD, and non-OECD countries.³ Non-OECD countries appear to have higher GDP and population growth rates than OECD countries, while average labor productivity in the former is notably lower than in the latter. In terms of GDP size, transition economies also receive approximately twice the FDI inflows as a percentage of GDP than OECD economies.

Table 1. Summary Statistics for The Period 2010-2019.

Variable	Mean	Std. dev.	Minimum	Maximum
Full Sample
GDP Growth (%)	3.14	2.66	-12.09	20.11
FDI (% of GDP)	5.46	18.70	-40.29	280.13
Economic Freedom	7.16	0.81	4.45	8.62
Capital Cumulation (% of GDP)	18.32	20.30	-263.56	71.16
Population Growth (%)	1.18	1.37	-1.319	12.25
Labor Force Participation Rate (%)	61.99	8.912	39.25	89.05
Labor Productivity	38557.87	39987.14	928.9	181776
Non-OECD Countries
GDP Growth (%)	3.75	2.72	-12.09	10.19
FDI (% of GDP)	6.64	22.87	-24.90	280.13
Economic Freedom	6.78	0.80	4.45	8.25
Capital Cumulation (% of GDP)	18.32	24.87	-263.56	71.16
Population Growth	1.65	1.50	-0.92	12.25
Labor Force Participation Rate (%)	62.66	10.71	39.25	89.05
Labor Productivity	14119.7	13082.98	928.9	68993
OECD Countries
GDP Growth (%)	2.22	2.29	-11.30	20.11
FDI (% of GDP)	3.73	9.52	-40.29	81.33
Economic Freedom	7.73	0.40	6.56	8.62
Capital Cumulation (% of GDP)	18.34	10.46	-55.55	66.08
Population Growth	0.50	0.74	-1.32	2.94
Labor Force Participation Rate (%)	61.00	5.12	47.97	77.75
Labor Productivity	74496.36	39103.87	14338	181776

Table 1. Summary Statistics for The Period 2010-2019.

Variable	Mean	Std. dev.	Minimum	Maximum
Full Sample
GDP Growth (%)	3.14	2.66	-12.09	20.11
FDI (% of GDP)	5.46	18.70	-40.29	280.13
Economic Freedom	7.16	0.81	4.45	8.62
Capital Cumulation (% of GDP)	18.32	20.30	-263.56	71.16
Population Growth (%)	1.18	1.37	-1.319	12.25
Labor Force Participation Rate (%)	61.99	8.912	39.25	89.05
Labor Productivity	38557.87	39987.14	928.9	181776
Non-OECD Countries
GDP Growth (%)	3.75	2.72	-12.09	10.19
FDI (% of GDP)	6.64	22.87	-24.90	280.13
Economic Freedom	6.78	0.80	4.45	8.25
Capital Cumulation (% of GDP)	18.32	24.87	-263.56	71.16
Population Growth	1.65	1.50	-0.92	12.25
Labor Force Participation Rate (%)	62.66	10.71	39.25	89.05
Labor Productivity	14119.7	13082.98	928.9	68993
OECD Countries
GDP Growth (%)	2.22	2.29	-11.30	20.11
FDI (% of GDP)	3.73	9.52	-40.29	81.33
Economic Freedom	7.73	0.40	6.56	8.62
Capital Cumulation (% of GDP)	18.34	10.46	-55.55	66.08
Population Growth	0.50	0.74	-1.32	2.94
Labor Force Participation Rate (%)	61.00	5.12	47.97	77.75
Labor Productivity	74496.36	39103.87	14338	181776

Figure 1 displays the average annual real GDP growth rate, FDI inflows, and economic freedom for the sampled countries over the entire period (2010-2019). The fitted line in Figure 1(a) shows a weak positive relationship between FDI and economic growth (R² = 0.038). Meanwhile, the fitted line in Figure 1(b) illustrates a negative correlation between output growth and economic freedom (R² = 0.092). The economic freedom of high-growth transition economies appears to cluster around the lower end of the index, while FDI inflows contribute positively to economic growth. However, the correlations do not imply the causation of FDI to economic growth. If economic freedom is vital in fostering economic growth, one would expect that FDI inflows and economic freedom are complements.

The model specification is broadly consistent with other panel data approaches that use FDI to estimate economic growth (for instance, Caselli et al., 1996 and Alfaro et al., 2004). Our model also includes labor productivity and economic freedom index to measure the effect of productivity spillover and economic openness. Three conditional variables, gross capital formation, population growth rate, and labor force participation rate, are vital factors of production. The former measures the investment levels of countries, and both latter variables are used to gauge labor force growth in the countries. The economic freedom index will likely capture most other pertinent variables, such as trade openness and government policies.

We first specify a growth level equation as

$$y_{it}-y_{it-1}={{\beta }_1{y}_{it-1}+\beta }_2{FDI}_{it}+{\beta }_3{EF}_{it}+{\beta }_4{CF}_{it}\mathrm{}+{{\beta }_5{PG}_{it}+\beta }_6{LF}_{it}+{\beta }_7{LP}_{it}+{\eta }_i+{\gamma }_t+{\epsilon }_{it}\mathrm{}$$

(1)

where y_it denotes the logarithm of real GDP in country i in year t; FDI_it is foreign direct investment, EF_it is economic freedom, CF_it is gross capital formation, PG_it is the population growth rate, LF_it is the labor force participation rate, and LP_it is labor productivity. η_i is an unobserved country-specific effect reflecting differences in the initial level of efficiency. The period-specific intercept γ_t captures real income changes that are common to all countries, and ε_it is the error term. Equation (1) can be rewritten as a dynamic model in the level of GDP by adding y_it-1 to both sides as follows.

$$y_{it}={{\mathsf{\alpha }}_{1}y}_{it-1}+{\beta }_2{FDI}_{it}+{{\beta }_{3}EF}_{it}+{\beta }_4{CF}_{it}+{{{\beta }_5{PG}_{it}+\beta }_{6}LF}_{it}+{{\beta }_{7}LP}_{it}+{\eta }_i{+\mathrm{}\gamma }_t+{\epsilon }_{it}$$

(2)

where α = (1+β₁). Equation (2) explicates the dynamic structure of the growth equation. However, the OLS estimate of the coefficient on the lagged dependent variable could bias upward as $y_{it-1}$ is positively correlated with the country-specific effects η_i. As proposed by Arellano and Bond (1991), one approach to address this issue is to employ lagged levels of the regressors as instruments, given that the error term is not serially correlated, and the lagged explanatory variables are weakly exogenous. This alternative approach addresses the presence of the lagged endogenous variable by first differencing equation (2) to eliminate country-specific effects as

$$∆y_{it}={{\mathsf{\alpha }}_1∆y}_{it-1}+{\beta }_2{∆FDI}_{it}+{{\beta }_3∆EF}_{it}+{\beta }_4{∆CF}_{it}+{{{\beta }_5{PG}_{it}+\beta }_6∆LF}_{it}+{{\beta }_7∆LP}_{it}+{∆\epsilon }_{it}$$

(3)

where the first difference equation removes omitted-variable bias and country-specific effects. Then, following Arellano and Bond (1991), we set the following moment conditions:

$$\begin{array}{c}\hfill E\left[y_{it-s}∆{\epsilon }_{it}\right]=0\mathrm{for}\mathrm{}s\ge 2;t=3,\dots .T\hfill \\ \hfill E\left[{FDI}_{it-s}∆{\epsilon }_{it}\right]=0\mathrm{}\mathrm{for}\mathrm{}s\ge 2;t=3,\dots .T\hfill \\ \hfill E\left[{EF}_{it-s}∆{\epsilon }_{it}\right]=0\mathrm{}\mathrm{for}\mathrm{}s\ge 2;t=3,\dots .T\hfill \\ \hfill E\left[{CF}_{it-s}∆{\epsilon }_{it}\right]=0\mathrm{}\mathrm{for}\mathrm{}s\ge 2;t=3,\dots .T\hfill \\ \hfill E\left[{PG}_{it-s}∆{\epsilon }_{it}\right]=0\mathrm{}\mathrm{for}\mathrm{}s\ge 2;t=3,\dots .T\hfill \\ \hfill E\left[{LF}_{it-s}∆{\epsilon }_{it}\right]=0\mathrm{}\mathrm{for}\mathrm{}s\ge 2;t=3,\dots .T\hfill \\ \hfill E\left[{LP}_{it-s}∆{\epsilon }_{it}\right]=0\mathrm{}\mathrm{for}\mathrm{}s\ge 2;t=3,\dots .T\hfill \end{array}$$

Despite being able to control country-specific effects and simultaneity bias, difference GMM estimates potentially suffer from instrumental problems with small samples. If the dependent variable was persistent and close to being a random walk, applying difference GMM estimation would yield both a biased and inefficient estimate of α₁ in finite samples, particularly acute with a relatively shorter estimation period.

Moreover, economic growth is dynamic and dependent on past levels, and the independent variables are not strictly exogenous, given that FDI and domestic investment may be correlated with past and current realizations of the error. Empirical research in international economics acknowledges that FDI, a dynamic process strongly dependent on past observations, could also potentially lead to disturbances that are likely heteroskedastic and serially correlated. One source of endogeneity arises from the possibility that current values of FDI are a function of past economic performance. Blundell and Bond (1998) attribute the poor precision and significant finite sample bias of the difference GMM estimator to lagged levels of the independent variables being weak instruments and propose using system GMM to address the simultaneity bias in the independent variables. The system dynamic panel GMM estimator effectively avoids the deviation caused by this endogeneity problem by utilizing equations (1) and (3) to provide more consistent estimates as a second condition for utilizing system GMM. The approach reduces the biases and imprecision by using additional moment conditions wherein difference equations use level instruments, and level equations use difference instruments. The estimated data, which includes a short period and a large number of countries, meets the requirements of a small period and a large number of countries to be estimated by GMM empirically.

Following Arellano and Bover (1995), we include the additional moment conditions for the second part of the system, the level equation.

$$\begin{array}{c}\hfill E\left[(y_{it-s}-y_{it-s-1})({\eta }_i+{\epsilon }_{it})\right]=0\mathrm{for}\mathrm{}s=1\hfill \\ \hfill E\left[({FDI}_{it-s}-{FDI}_{it-s-1})({\eta }_i+{\epsilon }_{it})\right]=0\mathrm{for}\mathrm{}s=1\hfill \\ \hfill E\left[({EF}_{it-s}-{EF}_{it-s-1})({\eta }_i+{\epsilon }_{it})\right]=0\mathrm{for}\mathrm{}s=1\hfill \\ \hfill E\left[({CF}_{it-s}-{CF}_{it-s-1})({\eta }_i+{\epsilon }_{it})\right]=0\mathrm{for}\mathrm{}s=1\hfill \\ \hfill E\left[(PG-{PG}_{it-s-1})({\eta }_i+{\epsilon }_{it})\right]=0\mathrm{for}\mathrm{}s=1\hfill \\ \hfill E\left[({LF}_{it-s}-{LF}_{it-s-1})({\eta }_i+{\epsilon }_{it})\right]=0\mathrm{for}\mathrm{}s=1\hfill \\ \hfill E\left[({LP}_{it-s}-{LP}_{it-s-1})({\eta }_i+{\epsilon }_{it})\right]=0\mathrm{for}\mathrm{}s=1\hfill \end{array}$$

The GMM estimators are applied with two-step variants using the optimal weighting matrices, in which the moment conditions are weighted by a consistent estimate of their covariance matrix. This would be asymptotically more efficient, as explained in Arellano and Bond (1991). However, using the two-step estimator in small samples would potentially result in the proliferation of instruments, leading to biased standard errors and parameter estimates. To solve this problem, Windmeijer (2005) proposed a finite sample bias-corrected standard error formula for the two-step linear GMM. Following Roodman (2009), we use the above moment conditions and employ the two-step estimator.

4. Empirical Results

Before conducting the GMM tests, we apply four panel unit root tests to examine the stationary properties of the data series: Levin, Lin, and Chu (2002, henceforth LLC), Im, Pesaran, and Shin (2003, henceforth IPS), and the Fisher-ADF and Fisher-PP tests defined by Maddala and Wu (1999) and Choi (2001). Table 2 presents the results of these tests. At the 5 percent significance level, the LLC, IPS, and both Fisher tests provide strong evidence that the six series, GDP, FDI, EF, CF, LF, and LP, are stationary, while only two statistics LLC and IPS show that population growth is of the I (1) process.

Table 2. Panel Data Unit Root Test Result.

			Chi-Squared Test
Variable	Levin, Lin & Chu	Im, Pesaran & Shin	ADF-Fisher	PP-Fisher
GDP	-20.6888*	-3.1327*	764.4010*	32.5363*
FDI	-51.2237*	-24.3951*	808.5611*	34.9455*
Economic Freedom (EF)	-31.0783*	-15.7595*	940.1037*	42.1217*
Capital Cumulation (CF)	-10.7141*	-17.6577*	716.8125*	29.9402*
Population Growth	-0.00042	-0.0083	310.1695*	7.7560*
Labor Force Participation Rate (LF)	-77.5358*	-41.1428*	1246.3047*	56.2808*
Labor Productivity (LP)	-36.9462*	-4.9398*	720.6632*	30.1502*

Notes: * rejects the null hypothesis of unit root; lags in each test were defined by the Akaike information criterion.

Table 2. Panel Data Unit Root Test Result.

			Chi-Squared Test
Variable	Levin, Lin & Chu	Im, Pesaran & Shin	ADF-Fisher	PP-Fisher
GDP	-20.6888*	-3.1327*	764.4010*	32.5363*
FDI	-51.2237*	-24.3951*	808.5611*	34.9455*
Economic Freedom (EF)	-31.0783*	-15.7595*	940.1037*	42.1217*
Capital Cumulation (CF)	-10.7141*	-17.6577*	716.8125*	29.9402*
Population Growth	-0.00042	-0.0083	310.1695*	7.7560*
Labor Force Participation Rate (LF)	-77.5358*	-41.1428*	1246.3047*	56.2808*
Labor Productivity (LP)	-36.9462*	-4.9398*	720.6632*	30.1502*

Notes: * rejects the null hypothesis of unit root; lags in each test were defined by the Akaike information criterion.

Growth regressions are outlined with a series of models: pooled OLS, fixed effects, difference GMM, and system GMM. To apply dynamic panel data models with robust properties, we utilize both two-step difference GMM and system GMM to deal with heteroskedasticity and simultaneity biases likely to exist among the variables.⁴ According to Nickell (1981), consistent regression coefficients resulting from difference GMM and system GMM should lie within the coefficients stemming from pooled OLS (upper bound; upward bias) and fixed effects estimators (lower bound; downward bias).

Table 3 demonstrates regression results using pooled OLS, fixed effects, difference GMM, and system GMM estimators. We use Hansen’s (1982) J-test to verify GMM consistency that the instruments are valid. The null hypothesis assumes that the instrumental variables satisfy the conditions, and the alternative hypothesis means that the instruments are not entirely exogenous or have been incorrectly excluded from the model (Baum et al., 2003). The Hansen test results suggest that the null hypothesis is not rejected, thus allowing the use of the GMM models. We also consider the test of second-order serial correlation of the error term suggested by Arellano and Bond (1991), in which the null hypothesis of no serial correlation of the error term is not rejected, implying the GMM estimator is valid.

The coefficients on the lagged dependent variables in all four models have a value of less than or close to one, providing strong evidence of conditional convergence. The difference GMM estimate coefficient for the lagged dependent variable is close to the fixed effects estimator, which implies that the difference GMM estimate is somewhat downward biased. The system GMM estimator with the coefficient of the lagged dependent variable falling between pooled OLS and difference GMM estimators is much closer to the upper bound of the pooled OLS estimate, provides consistent and efficient estimates, overcomes the endogeneity problem, and is a better fit for the selected panel data.

The coefficient generated by the system GMM estimator shows that economic freedom estimates are positive and significant, confirming the positive relationship between economic freedom and growth. Capital investment from FDI inflows and domestic accumulation also drive economic growth, and both sources of capital formation are equally effective in explaining economic growth. While the evidence shows that FDI leads to economic growth, a multitude of factors can still affect this relationship. Hall and Jones (1999) find that social infrastructure drives productivity and growth success through high physical and human capital investment. Differences in social infrastructure across countries can cause differences in capital accumulation, educational attainment, and productivity, which leads to differences in income between countries, thus affecting long-run economic performance concerning productivity and investment.⁵

The coefficients of population growth and labor force participation rate are negative but not statistically significant, while the coefficient on labor productivity is negative and statistically significant with system GMM estimates. This could be partially explained by Meyer and Sinani (2009) who find a curvilinear relationship between productivity spillovers and the level of economic development. The sample comprises developed, developing, and less developed countries, so the link between productivity and growth could thus be negative or positive.

We then examine the role of FDI on growth through productivity spillover and economic freedom. Following the model developed in section 3, we interact FDI with respective labor productivity and economic freedom and use each interaction term as a regressor. To ensure that the interaction term does not proxy for FDI or the level of labor productivity and economic freedom, both latter variables were included in the regression independently. Thus, we run the following regressions:

$$y_{it}={{\mathsf{\alpha }}_{1}y}_{it-1}+{\beta }_2{FDI}_{it}+{{\beta }_{3}EF}_{it}+{\beta }_4{CF}_{it}+{{{\beta }_5{PG}_{it}+\beta }_{6}LF}_{it}+{{\beta }_{7}LP}_{it}+{{\beta }_8{FDI}_{it}\mathrm{*}LP}_{it}+{\epsilon }_{it}$$

(4)

$$y_{it}={{\mathsf{\alpha }}_{1}y}_{it-1}+{\beta }_2{FDI}_{it}+{{\beta }_{3}EF}_{it}+{\beta }_4{CF}_{it}+{{{\beta }_5{PG}_{it}+\beta }_{6}LF}_{it}+{{\beta }_{7}LP}_{it}+{{\beta }_8{FDI}_{it}\mathrm{*}EF}_{it}+{\epsilon }_{it}$$

(5)

As shown in Table 4, each regression uses a different indicator for the FDI effect: column (1) uses labor productivity, and column (2) uses economic freedom. The main result is that the coefficients of both the interaction terms are statistically significant with or without including FDI inflows separately as a regressor.⁶ This implies that the marginal effect of FDI on economic growth depends on the level of productivity and economic freedom. FDI inflows alone appear to be significant but negative in general. This may be because, without progress in economic freedom, FDI inflows do not affect growth.⁷ Without productivity improvement, it is also possible that FDI inflows are not channeled efficiently and can even negatively affect growth. The results suggest a productivity spillover effect and economic freedom play an enhancing role in driving FDI inflow to growth. A caveat regarding this inference is that the interaction terms in our main specification force the impact of FDI on growth to increase monotonically with the level of LP and EF.⁸ However, a certain level of LP and EF may be required before the effect of FDI prevails in host countries. Only countries with an extent of absorptive capacity and adopting more progressive and transparent economic policies will likely benefit from the presence of FDI inflows.

We further evaluate the data by the subsets of OECD and non-OECD countries. Tables 5(a) and 5(b) give the results of models estimated by system GMM. When the regressions exclude interaction terms, the coefficients on FDI are positive and statistically significant, and the FDI coefficient turns out to be approximately twice as great for OECD than non-OECD countries. The coefficients on domestic capital formation are also significant and approximately equal for both groups.

Table 5. (a). Regression Results with Alternative Specifications: non-OECD Countries. (b). Regression Results with Alternative Specifications: OECD Countries.

	(1)		(2)		(3)
Lagged GDP (log)	0.9975***		0.9926***		0.9981***
	(0.0059)		(0.0064)		(0.0039)
FDI as % of GDP	0.0015*		-0.0156*		-0.0111*
	(0.0009)		(0.0090)		(0.0067)
Economic Freedom (log)	0.1392***		0.1575*		0.0697
	(0.0480)		(0.0879)		(0.1066
Capital Cumulation as % of GDP	0.0016*		0.0020*		0.0034**
	(0.0008)		(0.0011)		(0.0014)
Population Growth (%)	-0.0003		-0.0004		-0.0004
	(0.0040)		(0.0029)		(0.0032)
Labor Force Participation Rate (log)	-0.0590		-0.0402		-0.0040
	(0.0439)		(0.0397)		(0.0547)
Labor Productivity (log)	-0.0275**		-0.0350***		-0.0168
	(0.0121)		(0.0122)		(0.0110)
FDI x LP			0.0017**
			(0.0008)
FDI x EF					0.0071*
					(0.0037)
AR(1) Test (p-value)	0.017		0.039		0.049
AR(2) Test (p-value)	0.147		0.120		0.101
Hansen Test (p-value)	0.251		0.349		0.167
Number of Observations	500		500		500
Number of Countries	50		50		50
Number of Instruments	47		49		38
Notes: The dependent variable is the log of GDP. Robust standard errors are in parentheses. Time dummies are included in all the regressions (not reported). *, **, and *** indicate that the coefficients are significant at respective 10, 5, and 1 percent levels. Both columns report two-step Blundell and Bond (1998) system-GMM estimator with Windmeijer finite-sample correction. Labor productivity, labor force participation rate, and population growth are predetermined variables; foreign direct investment and capital accumulation are treated as endogenous in the estimation and are instrumented with their lagged values. The number of instruments is collapsed by dropping deeper lags as instruments to avoid the problem of instrument proliferation and overfitting of the endogenous explanatory variables (Roodman, 2009)
		(1)		(2)		(3)
Lagged GDP (log)		0.9879***		0.9885***		0.9928***
		(0.0063)		(0.0082)		(0.0055)
FDI as % of GDP		0.0028**		-0.0143**		-0.0343***
		(0.0011)		(0.0067)		(0.0124)
Economic Freedom (log)		0.2943		0.0831		-0.0538
		(0.3695)		(0.1487)		(0.1928)
Capital Cumulation as % of GDP		0.0015*		0.0012*		0.0014*
		(0.0008)		(0.0007)		(0.0007)
Population Growth		0.0117*		0.0118**		0.0073*
		(0.0063)		(0.0062)		(0.0042)
Labor Force Participation Rate (log)		-0.1018		-0.0574		-0.0072
		(0.0913)		(0.0555)		(0.0384)
Labor Productivity (log)		-0.0066		-0.0063		-0.0080
		(0.0131)		(0.0091)		(0.0097)
FDI x LP				0.0014**
				(0.0006)
FDI x EF						0.0176***
						(0.0060)
AR(1) Test (p-value)		0.038		0.031		0.064
AR(2) Test (p-value)		0.141		0.174		0.102
Hansen Test (p-value)		0.291		0.240		0.123
Number of Observations		340		340		340
Number of Countries		34		34		34
Number of Instruments		34		27		29
Notes: The dependent variable is the log of GDP. Robust standard errors are in parentheses. Time dummies are included in all the regressions (not reported). *, **, and *** indicate that the coefficients are significant at respective 10, 5, and 1 percent levels. Both columns report two-step Blundell and Bond (1998) system-GMM estimator with Windmeijer finite-sample correction. Labor productivity, labor force participation rate, and population growth are predetermined variables; foreign direct investment and capital accumulation are treated as endogenous in the estimation and are instrumented with their lagged values. The number of instruments is collapsed by dropping deeper lags as instruments to avoid the problem of instrument proliferation and overfitting of the endogenous explanatory variables (Roodman, 2009)

Table 5. (a). Regression Results with Alternative Specifications: non-OECD Countries. (b). Regression Results with Alternative Specifications: OECD Countries.

	(1)		(2)		(3)
Lagged GDP (log)	0.9975***		0.9926***		0.9981***
	(0.0059)		(0.0064)		(0.0039)
FDI as % of GDP	0.0015*		-0.0156*		-0.0111*
	(0.0009)		(0.0090)		(0.0067)
Economic Freedom (log)	0.1392***		0.1575*		0.0697
	(0.0480)		(0.0879)		(0.1066
Capital Cumulation as % of GDP	0.0016*		0.0020*		0.0034**
	(0.0008)		(0.0011)		(0.0014)
Population Growth (%)	-0.0003		-0.0004		-0.0004
	(0.0040)		(0.0029)		(0.0032)
Labor Force Participation Rate (log)	-0.0590		-0.0402		-0.0040
	(0.0439)		(0.0397)		(0.0547)
Labor Productivity (log)	-0.0275**		-0.0350***		-0.0168
	(0.0121)		(0.0122)		(0.0110)
FDI x LP			0.0017**
			(0.0008)
FDI x EF					0.0071*
					(0.0037)
AR(1) Test (p-value)	0.017		0.039		0.049
AR(2) Test (p-value)	0.147		0.120		0.101
Hansen Test (p-value)	0.251		0.349		0.167
Number of Observations	500		500		500
Number of Countries	50		50		50
Number of Instruments	47		49		38
Notes: The dependent variable is the log of GDP. Robust standard errors are in parentheses. Time dummies are included in all the regressions (not reported). *, **, and *** indicate that the coefficients are significant at respective 10, 5, and 1 percent levels. Both columns report two-step Blundell and Bond (1998) system-GMM estimator with Windmeijer finite-sample correction. Labor productivity, labor force participation rate, and population growth are predetermined variables; foreign direct investment and capital accumulation are treated as endogenous in the estimation and are instrumented with their lagged values. The number of instruments is collapsed by dropping deeper lags as instruments to avoid the problem of instrument proliferation and overfitting of the endogenous explanatory variables (Roodman, 2009)
		(1)		(2)		(3)
Lagged GDP (log)		0.9879***		0.9885***		0.9928***
		(0.0063)		(0.0082)		(0.0055)
FDI as % of GDP		0.0028**		-0.0143**		-0.0343***
		(0.0011)		(0.0067)		(0.0124)
Economic Freedom (log)		0.2943		0.0831		-0.0538
		(0.3695)		(0.1487)		(0.1928)
Capital Cumulation as % of GDP		0.0015*		0.0012*		0.0014*
		(0.0008)		(0.0007)		(0.0007)
Population Growth		0.0117*		0.0118**		0.0073*
		(0.0063)		(0.0062)		(0.0042)
Labor Force Participation Rate (log)		-0.1018		-0.0574		-0.0072
		(0.0913)		(0.0555)		(0.0384)
Labor Productivity (log)		-0.0066		-0.0063		-0.0080
		(0.0131)		(0.0091)		(0.0097)
FDI x LP				0.0014**
				(0.0006)
FDI x EF						0.0176***
						(0.0060)
AR(1) Test (p-value)		0.038		0.031		0.064
AR(2) Test (p-value)		0.141		0.174		0.102
Hansen Test (p-value)		0.291		0.240		0.123
Number of Observations		340		340		340
Number of Countries		34		34		34
Number of Instruments		34		27		29
Notes: The dependent variable is the log of GDP. Robust standard errors are in parentheses. Time dummies are included in all the regressions (not reported). *, **, and *** indicate that the coefficients are significant at respective 10, 5, and 1 percent levels. Both columns report two-step Blundell and Bond (1998) system-GMM estimator with Windmeijer finite-sample correction. Labor productivity, labor force participation rate, and population growth are predetermined variables; foreign direct investment and capital accumulation are treated as endogenous in the estimation and are instrumented with their lagged values. The number of instruments is collapsed by dropping deeper lags as instruments to avoid the problem of instrument proliferation and overfitting of the endogenous explanatory variables (Roodman, 2009)

The interaction terms reveal a positive, robust link between FDI inflows and productivity and between FDI inflows and economic freedom for both groups of transition and OECD economies. While the coefficient estimates are similar for the variables of the FDI-productivity interaction term, the interaction of FDI inflows and economic freedom is more substantial for OECD than for non-OECD countries. A potential explanation is that the interaction term may have captured an overall tendency for FDI to contribute to transition and OECD countries with a minimum threshold of economic infrastructure. The absorptive capacity for OECD countries with greater economic freedom appears to be the critical improvement for economic growth ascribing to FDI inflows.

Economic freedom alone drives economic growth only in non-OECD countries, and FDI inflows affect growth negatively when interaction terms are included in the regression. FDI inflows are valuable in terms of productivity spillover benefits in general, and the spillover benefits for OECD and transition economies appear to be similar. OECD economies tend to explore a greater absorptive capacity in FDI spillovers through improved economic freedom. However, economic freedom alone does not guarantee strong economic growth in OECD countries yet is more pronounced in enhancing economic growth in non-OECD countries.

The average population growth rate in non-OECD countries is three times greater than the growth rate in OECD countries. A larger population could drive economic growth in OECD countries but could be a substantial economic burden in non-OECD countries. The effects of the labor force participation rate on growth are adverse but statistically insignificant for OECD and non-OECD countries. The estimated labor productivity results are also mixed for both groups but seem to play a more antagonistic role in the economic growth of non-OECD countries.

To investigate how economic growth and FDI inflows are correlated in separate groups of OECD and transition economies, Figure 2 plots the relationship between these variables for the observations from 2010 to 2019. The fitted lines show a significantly more positive relationship between the FDI and economic growth for OECD countries than non-OECD countries. While the economies of most non-OECD countries are in transition with relatively higher economic growth, OECD countries stand to gain more from FDI inflows. Figure 3 plots the correlation between economic growth and economic freedom separated into OECD and transition economies. The fitted lines depict a weak positive correlation between output growth and economic freedom for non-OECD countries but a negative correlation for OECD countries. OECD countries are substantially more conducive to higher economic freedom despite output growth responding to economic freedom slightly inversely.

Note: economic growth (in percentage) was computed using real GDP per capita downloaded from the World Bank World Development Indicators, Foreign direct investments are the net inflows of investment to acquire a lasting management interest and the economic freedom index was taken from the Fraser Institute.

5. Conclusion

There is a shared view among policymakers across the globe that FDI inflows are a significant source of external financing for developing countries, emerging economies, and countries in transition (UNCTAD, 2019). FDI inflows are essential for transferring capital and technology, improving productivity, and growing economies. FDI inflows provide the capital needed for economic development, a substantive impetus for employment gains, especially in economies transitioning towards higher development. For economies transitioning from one level of development to the next, economic openness provides fertile ground for the effective allocation of capital, leading to higher productivity and economic growth. As productivity spillovers can occur with FDI inflows bringing along capital and technology, many factors could explain why FDI inflows contribute to higher productivity.

Applying the GMM system estimator to a panel of 84 countries over the period 2010-2019, we examine the productivity spillover effect of FDI inflows and the extent to which economic freedom attracts foreign investment. By separating the data into respective groups of transition and OECD economies, we focus on the effect of FDI inflows on economic growth and the channels through which the effect takes place for countries in different stages of development. The empirical results suggest that FDI inflows are valuable in terms of productivity spillover benefits in general, and the spillover benefits are relatively comparable for OECD and transition economies. Economic freedom is a substantial element for FDI inflows, and OECD economies with more economic freedom tend to explore greater absorptive capacity in FDI spillovers. However, economic freedom alone does not guarantee strong economic growth in OECD countries yet is more pronounced in enhancing economic growth in non-OECD countries. Our results confirm from previous studies that only economies with a certain level of economic infrastructure and economic development would benefit more from FDI inflows.

In addition to examining FDI spillover effects in general, this paper contributes to the existing literature on the effect of FDI presence along with productivity and economic freedom, concentrating explicitly in separate groups of transition and OECD economies. Despite policies in effect that explicitly aim to attract foreign investment, empirical evidence on the growth effect of FDI inflows is still subject to the specific factors underlying countries. When multiple definitions of the spillover variables are used, the results could vary across countries. This is a potential area for further study to discern how the level of economic freedom and country development affects this overall relationship and its progression over time.