Foreign Direct Investment, Institutions, and Domestic Investment in Develo** Countries: Is there a Crowding-Out Effect?

Abstract

Despite using a common database for a sample of 46 develo** countries to evaluate the impact of foreign direct investment (FDI) inflows on domestic investment (DI), two recent articles on the subject (Morrissey and Udomkerdmongkol in World Dev 40(3):437–445. 10.1016/j.worlddev.2011.07.004, 2012 and Farla et al. in World Dev 88:1–9, 2016. 10.1016/j.worlddev.2014.04.008), produced conflicting results. The current paper contributes to the debate by using a larger panel database of 105 develo** countries from 2002 to 2018 while controlling for financial development. We make use of the system generalized method of moments (S-GMM). Our findings do not support a crowding-in effect of FDI; instead, we found that FDI crowded out domestic investment. The findings underscore that institutions played no role in the FDI–DI nexus. Furthermore, there is no strong evidence that good institutions promoted investment in develo** countries from 2002 to 2018.

Introduction

Morrissey and Udomkerdmongkol (2012, henceforth MU) and Farla et al. (2016, henceforth FCV) both explored the same database of 46 develo** countries for the period 1996–2009 to assess the effect of foreign direct investment inflows (FDI) on domestic investment (DI). Both papers are remarkable in two respects. They expanded the research frontier. They are, to date, to the best of our knowledge, the only investigations to have simultaneously tested the crowding-in hypothesis with respect to FDI inflows in develo** countries and analyzed the role of institutions in the FDI–DI nexus. These two papers are further notable because they yielded contradictory results even though they drew on the same database. Using Blundell and Bond’s (1998) system Generalized Method of Moments (S-GMM), MU found that good institutions (low levels of corruption and political stability) encourage total investment (FDI and private investment). The authors further found that FDI inflows crowded out DI in develo** countries and the crowding-out effect increased with the quality of institutions (except where governance was measured by political stability¹). From the authors’ perspective when poor governance deters FDI, domestic investment is expected to be higher. Consequently, when institutions improve, FDI flows in, reducing the private investment opportunities so that total investment (FDI and private investment) increases by less than FDI. FCV in contrast found that FDI inflows crowded in DI, thus establishing the absence of evidence for the crowding-out effect. FCV further reported that institutions had a negative mediating effect on the FDI–DI nexus. The authors did not confirm that good institutions promote total investment².

The contradiction in their respective results led FCV to outline the limitations of the MU study. FCV argued that MU had four methodological and measurement limitations: (1) the use of a poor proxy for DI—total investment net of FDI inflows and public investment; (2) the failure to comply with the rule of thumb of generalized method of moments (GMM) that requires the number of groups to be greater than the number of instruments when the sample size is relatively small (data for only 46 develo** countries is used in this study)³; (3) the omission of dummy time variables in the econometric estimations which are relevant to exclude time-related shocks from the error terms; (4) the failure to apply the Windmeijer’s (2005) correction approach to get robust standard errors.

We argue that the study of FCV itself faces some limitations as well. Though the study is irreproachable from a methodological standpoint, the authors’ interpretation of their study’s findings can be misleading. For instance, concluding that FDI inflows crowd in DI when FDI has a positive and significant impact on total investment, measured by gross fixed capital formation (GFCF), is quite restrictive and goes against the commonly accepted interpretation in the literature (MU; Morrissey and Udomkerdmongkol 2016). FCV seems to have confused GFCF with private domestic investment. As private domestic investment is a component of total investment (the dependent variable), the marginal effect of FDI inflows should be greater than unity for one to infer a crowding-out effect. Although FCV found a positive and significant coefficient for FDI inflows, it consistently falls below unity, which implies a crowding-out effect⁴. Therefore, the major real point of contestation between MU and FCV boils down to the influence of institutions on domestic investment. FCV reported the negative mediating effect of institutions based on aggregated Worldwide Governance Indicators (WGIs), and, we suspect that had they considered the individual governance indicator (as MU did) their study may have arrived at different results, which may not have contradicted MU’s findings.

In this study, we re-evaluate MU and FCV’s findings concerning the importance of institutions to investment and regarding the surprising negative modulating effect of institutions on the FDI–DI nexus, by addressing the shortcomings and limitations of MU that were identified by FCV. Theoretically, institutions represent an important indicator for investors. North (1990) argued that capital-friendly political and economic institutions are crucial for both foreign and domestic investment because they reduce transaction and production costs and improve the business environment. Furthermore, growing evidence suggests that institutional weaknesses such as political instability, a biased judiciary system, and corruption deter FDI and DI (Adams 2009; Asiedu et al. 2009; Busse and Hefeker 2007; Ezeoha and Cattaneo 2012; Gugler and Peev 2010; Mauro 1996; Tag 2021; Uberti 2020). Good institutions that protect and enforce property rights provide incentives for multinationals who intend to operate in sectors with strong technological spillovers (i.e.,: manufacturing) (Adams 2009; Crespo and Fontoura 2007; Jude and Levieuge 2017). With good institutions, local firms may enjoy greater productivity and absorptive capacity that allow them to seize available investment opportunities. In contrast, in bad institutional frameworks, the elite’s rent-seeking behavior tends to raise production and transaction costs, discouraging domestic investors and encouraging multinationals to rather invest in the primary (and not secondary) sector where spillovers are rare (Shah et al. 2020; Jude and Levieuge 2017). Therefore, the assertions that the crowding-out effect of FDI inflows is greater in countries with good governance (as reported by MU and FCV) and that institutions have no significant effect on total investment (as highlighted by FCV), require further analysis.

We depart from MU and FCV in many ways. First, we used a more substantial database of panel data from 105 develo** countries over the period 2002–2018. The scope of our study, encompassing a broad sample of develo** countries and an extended timeframe, facilitated the execution of sub-period analyses while ensuring that the number of instruments did not exceed the number of groups (countries). In contrast to both FCV and MU, we adopted a more comprehensive approach that integrates additional determinants of investment. We incorporate financial development and investigate its interaction with FDI. Financial development is an essential determinant of investment, as a well-developed financial market facilitates the allocation of financial resources to profitable and viable investment projects (Levine 2005). In develo** countries, where the investment ability of local firms is often hindered by liquidity constraints and financial market imperfections (Islam et al. 2020; Harrison et al. 2004; Jongwanich and Kohpaiboon 2008), financial development is frequently perceived as a credible alternative to supporting domestic investment (Misati and Nyamongo 2011; Ndikumana 2000). In addition to being critical for domestic investment, the financial market is the second most important medium—after the real market—through which FDI inflows affect domestic investment (Agosin and Machado 2005; Jude 2019). It is well known that FDI inflows expand the availability of liquidity for local investors through currency appreciation and decreased interest rates (Jude 2019). As a result, a well-developed financial market can mitigate the crowding-out effect or enhance the crowding-in effect of FDI inflows. Given the significance of the financial market to the FDI–DI nexus, neglecting financial development in the investment equation (as done by MU and FCV) may introduce omission bias.

In addition to excluding financial development in their regression, MU arbitrarily ignored “government effectiveness,” which is one of the six WGIs, without any clear justification for why this indicator was dropped. This shortcoming was not properly addressed by FCV either. Though WGIs are highly correlated (Buchananet al. 2012; Kurul 2017), they measure and capture different aspects of institutions in such a way that one cannot discriminate between them without running into information loss (Kurul 2017). Government effectiveness (GE) captures the quality of bureaucracy and public service provision, the commitment of the government to the announced policies, and the independence of civil servants from political pressure. These parameters are important to investors.

Our findings support the notion that FDI has a positive impact on total investment. Given that the marginal effect of FDI inflows is less than unity, we conclude that FDI inflows crowd out domestic investment. Our findings do not support the idea that the crowding-out effect of FDI remains more pronounced in countries with good institutions, as suggested by FCV and MU. Instead, we observe a neutral mediating effect of institutions on the FDI-domestic investment relationship. Furthermore, both the direct and indirect effects of financial development were found to be statistically insignificant.

Beyond its academic contributions, the central message from our research is that FDI inflows tend to displace domestic investment in develo** countries. This discovery emphasizes the necessity for policymakers to meticulously weigh the potential downsides of FDI and develop strategies to maximize the advantages associated with FDI inflows. Suggestions for policy action encompass enhancing the business climate to facilitate the diffusion of technology, promoting collaborations between multinational corporations and local investors, and broadening investment prospects for local enterprises, particularly within the manufacturing domain.

The paper is structured as follows. “Relevant Literature” section gives a snapshot of the existing literature. “Stylized Facts on FDI and Total Investment in Develo** Countries” section presents some stylized facts on FDI and total investment in develo** countries. “Methodology” section outlines the methodology used whereas “Data and Sources” section presents the data and its source. The main findings are reported in “Empirical Results” section. “Discussion of the Main Findings” section discusses the main findings and “Concluding Remarks” section concludes the paper and proposes a direction for future research.

Relevant Literature

Inflows of FDI to develo** countries may either stimulate or deter domestic investment. The literature refers to the former as the crowding-in effect, and to the latter as the crowding-out effect of FDI inflows. FDI can stimulate domestic investment through linkages effects (Markusen and Venables 1999; Marcin 2008), the transfer of technologies or technological spill-overs, and by loosening financial constraints to domestic investors (Harrison et al. 2004; Harrison and McMillan 2003; Blalock et al. 2008).

Though the growing flows of FDI in develo** countries have aroused interest in their macroeconomic effects, few studies have explored their effects on domestic investment as highlighted in Morrissey and Udomkerdmongkol (2012). Moreover, the limited existing evidence is mixed and inconclusive

For instance, while one strand of empirical evidence supports the existence of a crowding-out effect (Adams 2009; Agosin and Machado 2005; Elheddad 2019; Harrison and McMillan 2003; Morrissey and Udomkerdmongkol 2012; Mutenyo et al. 2010), another thread of the literature argues that FDI crowds-in domestic investment instead (Al-Sadig 2013; Ashraf and Herzer 2014; Farla et al. 2016; Ndikumana and Verick 2008; Omri and Kahouli 2014).

Yet other empirical studies have found the results to be mixed (Adams 2009; Agosin and Machado 2005; Chen et al. 2017; Jude 2019; Diallo et al. 2021). For instance, Agosin and Machado (2005) report that FDI had no impact on domestic investment in Africa, Latin America, and Asia from 1971 to 2000, while it crowded out DI in Latin America in some sub-periods. Jude (2019), in a study of 10 Central and Eastern European countries over the period 1995–2015, finds that FDI crowds out in the short term, but crowded in domestic investment in the long run. Based on a panel database of 42 Sub-Saharan African countries from 1990 to 2003, Adams (2009) found that FDI had a negative contemporaneous effect on domestic investment and a positive effect in the latter period. Diallo et al. (2021), using a panel of 40 sub-Saharan African countries over 1980–2017, saw no effect in the short run but a crowding-in effect in the long run. Using an aggregate measure of FDI, a study by Chen et al. (2017) was not able to detect any significant effect of FDI on domestic investment in China from 1994 to 2014; However, by considering the entry mode, the study found that equity joint venture crowds in domestic investment, whereas wholly foreign-owned firms crowd it out.

FDI may crowd out domestic investment when local firms are less productive than foreign ones (Markusen and Venables 1999). The presence of multinational firms may exert no significant effect on domestic investment when foreign firms buy little from local ones (Rodriguez-Clare 1996). According to Agosin and Machado (2005), because foreign firms have technological superiority over domestic firms, they may take away investment opportunities that would otherwise be available to domestic investors. The authors further posit that the crowding-out effect would be even larger when foreign and domestic investors operate in the same sector. FDI may also crowd-out local enterprises by eroding their market shares (Adams 2009). Yet, the rise in the interest rate, when foreign investors finance their production costs through the local financial market is a source of the crowding-out effect (Harrison and McMillan 2003).

Though they advanced reasons to explain the mixed evidence, few studies offered a detailed explanation for FDI’s inconsistent delivery of the expected crowd-in effects. The existing evidence attributes FDI’s lack of effect and its crowding-out effects, to low levels of absorptive capacity in host countries (Al-Sadig 2013; Farla et al. 2016; Jude 2019; Morrissey and Udomkerdmongkol 2012). Indeed, host countries with lower absorptive capacities have low productivity which prevents them from making use of technology, knowledge, and other skills transferred by multinationals (Adams 2009). Host countries’ absorptive capacity is influenced by the quality of institutions, financial development, availability of human capital, and macroeconomic policies (Alfaro et al. 2008; Alguacil et al. 2011). Jude (2019) and Al-Sadig (2013) reported, respectively, that financial development and human capital positively strengthen the effects of FDI inflows on domestic investment.

Good institutions have the potential to affect the relationship between FDI and domestic investment. Good institutions strengthen host countries’ absorptive capacities because countries with better institutions are more likely to invest in human capital (Acemoglu et al. 2001; Klomp and De Haan 2013) and have a well-developed financial market (Beck and Levine 2003). From Cezar and Escobar’s (2015) theoretical perspective, the institutional distance between origin and host countries, the adaptation costs, and the productivity cut-off above which it is profitable to settle greenfield investment all increase when institutions are bad, which leads to lower FDI inflows.

Nonetheless, MU and FCV challenge the theoretical perspective that posits a positive relationship between good institutions and an increased crowding-in effect of FDI on domestic investment. Both authors identified a negative mediating effect of institutions on the FDI-domestic investment nexus and the relationship between FDI and domestic investment, except for political stability (as reported by MU). Distinct explanations are offered by the authors. According to FCV’s perspective, within the institutional framework of develo** countries, the rent-seeking behavior that leads elites to provide foreign investors with preferential access to the industry (local market) harms the industrial structure and domestic investment in the long run. The negative mediating effect of institutions on the FDI-domestic investment relationship becomes apparent when the detrimental impact of rent-seeking activities outweighs the expected spillover effects from multinationals. For MU, foreign and local investors react differently to the quality of institutions. They concede that poor institutions deter FDI inflows such that domestic investment must increase to compensate so as not to disrupt investment opportunities. In contrast, improved institutions stimulate the entry of MNCs, reducing investment opportunities for local firms so that total investment increases less in proportion to the increase in FDI flows.

Stylized Facts on FDI and Total Investment in Develo** Countries

Lucas (1990) remarked that foreign capital does not flow to develo** countries as expected. He attributed the reluctance and low flow of FDI to develo** countries to the lack of human capital. Subsequent studies explain Lucas’s paradox as a consequence of structural challenges including market imperfections, shallow financial markets, lax macroeconomic policies, and institutional poverty (Alfaro et al. 2008; Papaioannou 2009). However, FDI flows to the Global South have been rising since the 1990s, as depicted by Figure 1. Though net FDI inflows made up a larger proportion of GDP in high-income countries (HICs) than in develo** countries between the 1970s and the 2000s, low-income countries (LICs) had a greater share of net FDI inflows (as a percentage of GDP) between 2010 and 2019 (Figure 1). Since the 1970s, net FDI inflows as a percentage of GDP have grown steadily in low-income countries whereas they have declined in the last decade for HICs and middle-income countries (MICs). LICs also seem more dependent on FDI inflows since these made up an average of 4% of their GDP in the period 2010–2019, whereas FDI contributed only 2% of GDP in the average HIC. Nonetheless, these growing levels of FDI suggest that develo** countries have been financing their economies with foreign capital whose macroeconomic effects must not be ignored.

Fig. 1

Source World Bank database, 2023

Net FDI inflows (% of GDP) by income group.

A closer look at Fig. 2 indicates that total investment appears less volatile than net FDI. Importantly, while total investment as a percentage of GDP grew from an average of 23% in the 1970s to 31% in the decade 2010–2019 in develo** countries, the average HIC saw it fall, from almost 25 to 20%, in the same period. The concomitant growth of FDI and total investment in develo** countries highlighted in Figs. 1 and 2 would explain the positive correlation among both variables as indicated by Fig. 3. Such a figure presumes that FDI inflows are less likely to crowd in domestic investment. The negative correlation between domestic private investment and FDI inflow strengthens the assumption of the crowding-out effect. Further causal analyses are needed to infer a crowding-in (out) effect.

Fig. 2

Source World Bank database, 2023

Total investment, or GFCF (% of GDP) by income group.

Fig. 3

Source World Bank database, 2023

Correlation between FDI and total (and domestic) investment, 2002–2018.

Methodology

Empirical Model and Variable Selection Justification

The FCV and MU’s empirical model augmented with financial development and its interaction with institutions is adopted. For a country $i$ at time $t$, the model is specified as follows:

$$ \begin{aligned} {\text{GFCF}}_{i,t} = & \beta_{0} + \beta_{1} {\text{GFCF}}_{i,t - 1} + \beta_{2} {\text{FDI}}_{i,t} + \beta_{3} {\text{INST}}_{i,t} + \beta_{4} {\text{FIN}} + \beta_{5} {\text{INST}}_{i,t} *{\text{FDI}}_{i,t} \\ & \; + \beta_{6} {\text{FIN}}_{i,t} *{\text{FDI}}_{i,t} + \theta^{\prime } X_{i,t} + v_{i} + d_{t} + \varepsilon_{i,t} \\ \end{aligned} $$

(1)

where ${\upbeta }_{0}$, ${\upbeta }_{1}$, ${\upbeta }_{2}$, ${\upbeta }_{3}$, ${\upbeta }_{4}$, ${\upbeta }_{5}$, ${\upbeta }_{6}$ and $\theta^{\prime }$ are parameters to be estimated. ${\upbeta }_{1}$ is the speed of adjustment, ${\beta }_{2}$ the direct effect of FDI on total investment (GFCF), ${\beta }_{3}$ and ${\beta }_{4}$ measure the direct effect of institutions (INST) and financial development (FIN), respectively, ${\beta }_{5}$ and ${\beta }_{6}$ stands for the mediating role of institutions and financial development, $\theta^{\prime }$is a vector of parameters that capture the effects of control variables ($\text{X}$). ${v}_{i}$ stands for time-invariant unobservable fixed effects, ${\varepsilon }_{i,t}$ is an idiosyncratic error term; ${d}_{t}$ captures a set of dummy time variables integrated into the model to exclude time-related shocks from the error terms, that is contemporaneous correlation.

$\text{FDI}$ stands for net FDI inflows expressed as a percentage of gross domestic product ($\text{GDP}$). FDI is the net inflow of investment to acquire a long-term management interest in an enterprise located in the host countries, generally 10% or more of the voting rights. The World Bank measures FDI as the sum of equity capital, reinvestment of earnings, and other long-term and short-term capital as displayed in the balance of payments. Net FDI inflows are no less than new foreign investment inflows net of disinvestment. In develo** countries where new foreign investment outpaces outflows of FDI (Levasseur 2002), net FDI inflows are a good proxy for FDI and are widely used in the literature (Morrissey and Udomkerdmongkol 2012; Tag 2021).

The lack of data on domestic investment has led researchers to proxy domestic investment by the residual obtained when public investment (GFCFPU) and FDI are subtracted from total investment, that is GFCF (Adams 2009; Agosin and Machado 2005; Morrissey and Udomkerdmongkol 2012). However, this proxy was openly criticized in the literature (Ndikumana and Verick 2008), particularly by FCV for several reasons. First, FDI and GFCF measure different realities and use different conceptual frameworks. FDI, as a component of the balance of payments, integrates financial flows such as re-invested earnings, intra-company loans, and equity stake, whereas GFCF proceeds from the national accounts. Second, although both include greenfield investment, GFCF does not include mergers and acquisitions (M&As), which leads to a larger bias in the residual if the share of M&As in the FDI inflows is non-negligible. The bias could be larger and become a worrying issue in develo** countries, and African countries particularly, where M&As have been growing owing to financial liberalization and privatization (Levasseur 2002; Diallo et al. 2021). Third, the use of the residual (domestic private investment) as a dependent variable constrains the coefficients of FDI and GFCFPU to negative values as $\text{FDI}$ and GFCFPU stand as explanatory variables in the investment equation. Finally, the bias gets worse if net FDI inflows instead of the FDI inflows are deducted from GFCF because net FDI inflows include disinvestment (Farla et al. 2016).

Acknowledging these critiques, we opted to use GFCF as a proxy for DI as done by FCV. Although it does not directly capture domestic investment, GFCF stands as a superior proxy compared to generated private domestic investment, because it avoids, at least, the shortcomings underlined by FCV and therefore minimizes the biases in the estimates. Nonetheless, using GFCF as a dependent variable entails a consequential interpretation of the coefficient of FDI since GFCF includes both domestic and foreign investments. From Morrissey and Udomkerdmongkol’s (2016) perspective, even though a negative coefficient of FDI does not imply a crowding-out effect, a positive coefficient is insufficient to infer a crowding-in effect when one uses GFCF as the independent variable. We expect the coefficient of FDI (${\beta }_{2}$) to be positive and significant. Such a result does not entail a crowding-in effect as ${\beta }_{2}$ is neither the marginal effect of FDI inflows on total investment nor the direct effect of FDI inflows on DI. Rather, ${\beta }_{2}$ is the direct effect of FDI inflows on total investment ($\text{Domestic investment}+\text{FDI}$). For a robustness check, we employ private investment, specifically private gross fixed capital formation (GFCFPR), as the independent variable. Private domestic investment was also employed to replicate the approach of FCV and MU based on our dataset.

The mediating role of institutions, which is an attribute of the host countries’ absorptive capacity, is also addressed. This mediating role can be justified in several ways. First, foreign investors are more likely to initiate greenfield investments, which have been proven to crowd in domestic investment (Elheddad 2019; Jude 2019), in good institutional environments because of the low transaction and production costs embedded in inclusive institutions (North 1990). Second, growing evidence suggests that foreign investors are more likely to transfer technologies to local firms when governance infrastructure is capital-friendly and property rights are well protected (Flu et al. 2021). Third, multinational firms (MNCs) increase the amount of investment because adaptation costs are low in an optimal institutional setting (Cezar and Escobar 2015; Lederman et al. 2010; Busse and Hefeker 2007). Protected intellectual property, which is an attribute of capital-friendly and inclusive institutions (political and economic), facilitates greater FDI spillovers (Fu et al. 2011) and investment in R&D as well in the manufacturing sector (Farla et al. 2016). To capture the described mediating effect, we introduced an interaction term of institutions and FDI inflows into the empirical equation (INST*FDI). If these described effects are strong, institutions would have a strong positive mediating effect on the relation between FDI and domestic investment. Therefore, we expect ${\beta }_{5}$ to be positive and significant.

We operationalized institutions using binary variables derived from the six World Bank Worldwide Governance Indicators (WGIs) (Kaufmann et al. 1999): (1) control of corruption (CC); (2) governance effectiveness (GE); (3) political stability and absence of violence (PS); (4) regulatory quality (RQ); (5) rule of law (RL); and (6) voice and accountability (VA). Each WGI falls within the range of − 2.5 to 2.5. A value closer to the upper limit indicates good institutional quality, while a value closer to the lower limit suggests poor institutional quality. In the primary estimations and in accordance with FCV, institutions are gaged by the index of institutions computed through principal component analysis (PCA). The index of institutions is equated to the first principal component.

To replicate the approaches of MU and FCV (and in subsequent estimations), institutions were alternatively represented by binary variables based on the percentile rank of WGIs (CC, GE, PS, RQ, RL, and VA). The binary variables take a value of 1 for a given country at a given time if the WGIs’ ranks are above their 50th percentile and 0 otherwise. Following MU, an institutional binary variable equal to 1 implies that institutions are capital-friendly; otherwise, they are considered capital-unfriendly.

Moreover, to address multicollinearity concerns that may arise from high correlation among the institutional variables (Buchanan et al. 2012), we refrained from integrating them into the same regression. Consequently, we estimated separate equations for each institutional variable. The direct impact of each institutional variable on total investment is captured by β₃. We anticipate β₃ to be positive and significant, as institutions hold significance for domestic investors (Edgardo Campos et al. 1999; North 1990).

In addition to the institutions, we accounted for financial development (FIN) which is also an attribute of host countries’ absorptive capacities. Both foreign and domestic investors thrive when the financial sector is well-developed. A booming financial market contributes to an optimal allocation of financial resources, the monitoring of investments, and the diversification of risk (Levine 2005). Such a market promotes private investment through the easing of financial constraints (Braga Tadeu and Moreira Silva 2013; Misati and Nyamongo 2011). The accessibility to external funds through host countries’ financial markets allows foreign investors to easily cover their upfront fixed costs and finance their production, especially when foreign investors’ financial institutions are reluctant to cover the costs of FDI (Desbordes and Wei 2017). Empirical evidence suggests that the macroeconomic effect of FDI is greater when financial markets and institutions are well-developed since they enable technological spillovers (Osei and Kim 2020). As we did for institutions (as done by MU and FCV), we measured financial development (FIN) by a binary variable equal to 1 when the International Monetary Fund’s (IMF) index of financial development (IFD) is above its 50th percentile, and 0 otherwise. The IFD is a multidimensional metric that accounts for access and efficiency of financial markets and institutions. Therefore, the IFD is more advanced than other commonly used indicators of financial development, such as credit to the private sector as a percentage of GDP, monetary aggregates, and market capitalization. We expect ${\beta }_{4}$ and ${\beta }_{6}$ to be positive and significant.

To infer a crowding-in effect, the marginal effect of FDI inflows on total investment must be greater than unity. That is total investment must significantly expand more in proportion to the increase in FDI inflows (Morrissey and Udomkerdmongkol 2012; 2016). Equation 2 presents the marginal effect of FDI inflows on total investment which is still a function of the regimes of institutions and the level of financial market development. As countries with good institutions are more likely to have a well-developed financial market (Beck and Levine 2003), we maintain that $FD=0$ if $\text{INST}=0$ and $FD=1$ if $\text{INST}=1$. This assumption eases the interpretation of Eq. 2. The marginal effect of FDI inflows on total investment is shown by ${\beta }_{2}+{\beta }_{5}+{\beta }_{6}$ in a capital-friendly institutional framework and by ${\beta }_{2}$ in poor institutional regimes.

$$ \partial {\text{GFCF}}_{i,t} /\partial {\text{FDI}}_{i,t} = \beta_{2} + \beta_{5} {\text{INST}}_{i,t} + \beta_{6} {\text{FD}}_{i,t} $$

(2)

Several factors other than FDI, institutions, and a well-developed financial market can promote or discourage DI. We captured their effects by adding a set of two control variables $\text{X}$ as done in FCV and MU. These include economic growth ($\text{GROWTH}$) and public investment (GFCFPU). According to the well-known accelerator effect, the expansion of GDP increases investment at the firm level. Empirical studies confirm the accelerator effect by establishing a positive relationship between economic growth and private investment (Ashraf and Herzer 2014; Farla et al. 2016; Ndikumana and Verick 2008). In addition to inducing an accelerator effect, economic growth is often used as an indicator of current and future market potentials (Agosin and Machado 2005; Morrissey and Udomkerdmongkol 2012). It is also well established that bond-financed public investments raise the real interest rate, crowding out private investment (Aschauer 1989). Public debt, contracted to finance public investments, can reduce private investment in a situation where the available financial resources for the private sector are limited.

Identification Strategy

Our empirical model is dynamic since GFCF (as a percentage of GDP) lagged by one period and is considered an explanatory variable. Since estimating a dynamic model with an Ordinary Least Square (OLS) estimator would bias the estimates, we relied on Blundell and Bond’s (1998) two-step system GMM estimator to fit the empirical model as done by FCV. This estimator has proved advanced compared to Arellano and Bond’s (1991) and Arellano and Bover’s (1995) estimators because system GMM uses greater moment conditions, and therefore more instruments. In addition to the common Arellano-Bond-type orthogonality conditions, the system GMM estimator uses moment conditions on the level equation to produce efficient estimates. Roodman (2009b) makes the point that system GMM estimators can help handle fixed effects and endogenous regressors while avoiding dynamic panel bias. Furthermore, they can accommodate unbalanced panel data.

Another source of concern arises from the potential for simultaneous causality between FDI and DI. While FDI may either crowd in or crowd out DI, a higher level of DI could signal to foreign investors that the business environment is favorable and transaction costs are minimal (Agosin and Machado 2005; Shah et al. 2020). From the perspective of Ndikumana and Verick (2008), foreign investors interpret a higher level of DI in host countries as an indication of a high return on capital. However, it might also be the case that government incentives to domestic investors, aimed at stimulating their investment, attract foreign investors (Shah et al. 2020).

The application of System GMM has been instrumental in mitigating potential simultaneous causality biases arising from a reciprocal relationship between total investment and FDI. By utilizing all available internal instruments, we instrumented FDI using its past values, in contrast to FCV, who used second lags as instruments in the transformed equation and first differences in the level equation. The use of more instruments, while ensuring that the number of groups exceeds the number of instruments, results in more efficient estimates. These internal instruments are exogenous to GFCF, as the contemporaneous total investment is less likely to influence the past values of FDI. In line with MU and FCV, we treated institutions as exogenous variables. Similarly, we treated financial development as exogenous, given that both share similar characteristics in determining a country’s absorptive capacity. Therefore, the interactions proceed from exogenous variables (FDI and FIN) to an endogenous variable (FDI).

In contrast to FCV, who considered interactions as exogenous, we regarded interactions as predetermined, employing their lagged values as instruments only in the level equation (not in the transformed equation). This approach allows us to reduce the number of instruments compared to a case where interactions are assumed to be endogenous. Still, the proliferation of instruments remains the chief concern with system GMM. Roodman (2009b) offers two reasons why this could be an issue. First, too many instruments can overfit the instrumented variable, leaving it unable to remove the endogenous component and bias the estimates toward those from non-instrumenting estimators. Second, when the instrument count is high, the standard errors in the two-step GMM tend to be downward biased (Roodman 2009b; Windmeijer 2005). To reduce the number of instruments below the total number of groups, that is the number of countries, we used the command “collapse” proposed by Roodman (2009a). We further use the Windmeijer robust estimator for the two-step covariance matrix to nullify any bias that could stem from the two-step variance of system GMM. To ascertain the quality of our estimates, we performed two post-estimation tests including the Arellano-Bond test and the Hansen test. The former tests the null hypothesis of the absence of second-order autocorrelation of the error terms, whereas the latter tests the null hypothesis of the joint validity of instruments. In both cases, the null hypothesis is not rejected if the p-value associated with each test is greater than 5%. The Sargan test is an alternative overidentification test. However, we chose the Hansen test over the Sargan test because the former is robust to heteroskedasticity or autocorrelation (Roodman 2009a).

Data and Sources

Our dataset comprises yearly observations from 105 develo** countries across the period 2002–2018⁵. From the overall list of 134 develo** countries, we excluded eight countries that lack FDI data (Cuba, Eritrea, Korea, Kossovo, Micronesia, Somalia, South Soudan, and Syria), twenty countries lacking public investment data (Belaros, Jamaica, Kyrgyz, Kiribati, Marshall Island, Papua New Guinea, Palau, Russian Federation, Solomon Island, Lybia, Sudan, Suriname, Timor-Lest, Tonga, Turkiye, Tuvalu, Turkmenistan, Vietnam, Vanuatu, and Zimbabwe) and one country that does not have both FDI and public investment data (Samoa). The absence of annual observations for the World Bank’s governance indicators (WGIs) before 2002 justifies the commencement of the study period (2002–2018). Data pertaining to the index of financial development (FIN), public investment (GFCFPU), and private investment (GFCFPR) were sourced from the International Monetary Fund’s (IMF) database. Information on the growth rate of GDP (GROWTH), net foreign direct investment inflows (FDI), gross fixed capital formation as a percentage of GDP (GFCF), and institutions gaged by governance indicators (WGIs) (control of corruption (CC), government effectiveness (GE), political stability and absence of violence (PS), regulatory quality (RQ), rule of law (RL), and voice and accountability (VA)) were extracted from the World Bank database.

Domestic private investment data were computed as total investment (GFCF) minus FDI inflows (net), and public investment was determined based on the methodology proposed by MU. Utilizing principal component analysis (PCA), the index of institutions (WGI) is defined as the first principal component. Table 1 presents the descriptive statistics for the study variables. The coefficient of variation (CV), the ratio of standard deviation to the mean, –– indicates that FDI and domestic private investment (DPI) exhibit greater dispersion compared to financial development and investment (private, public, and total). The minimum value in the data on FDI inflows is negative because some countries⁶ had negative net FDI inflows at a point in time. This implies that new investment inflows are inferior to disinvestments such as the repatriation of capital and repayment of loans. The FDI negative values were excluded from regressions that involved the logarithm of net FDI inflows. Negative minimum values for GROWTH imply that some countries experienced negative growth rates at a point in the study time frame. Furthermore, the correlation coefficients presented in Table 2 indicate that ($1$) FDI inflows, economic growth rate, financial development index, investment (private, domestic private, and public), and WGIs are positively correlated with total investment (GFCF); ($2$) while FDI is positively correlated with total investment, it is negatively associated with domestic private investment, presuming more of crowding-out than the crowding-in effect of FDI inflows. We emphasize that these correlations do not imply causation.

Table 1 Descriptive statistics.

Table 2 Correlation matrix between the variables of the study.

Table 3 presents the difference in the mean of FDI/GDP and GFCF/GDP with respect to different institution regimes. Such descriptive statistics are relevant in assessing the relative importance of total investment and FDI inflows across host countries’ institutional regimes. High institutional quality (INST=1) denotes capital-friendly institutions, whereas lower institutional quality (INST=0) suggests capital-unfriendly institutions. Both FDI/GDP and GFCF/GDP display positive and significant mean differences for all institutional variables except GE and RL for FDI/GDP and VA for GFCF/GDP. However, the overall picture seems to denote that countries with capital-friendly institutions seem to attract a greater amount of FDI inflows and greater gross fixed capital formation. Both domestic and foreign investors are therefore sensitive to host countries’ institutional superstructure.

Table 3 The difference in FDI/GDP and GFCF/GDP by institutional regime.

However, domestic, and foreign investors seem to value the attributes of governance differently because the magnitudes of the mean differences vary across institutional regimes. The mean difference of GFCF/GDP is greater for CC, PS, GE and RL, and the mean difference of FDI/GDP is higher for PS and VA. Consequently, domestic investors are reluctant to invest when corruption is common, governments are not effective in delivering and implementing public policies, political instability, and the lack of rule of law prevails. In contrast, foreign investors settle down when host countries are politically stable and more democratic. It is worth stressing that the mean difference of FDI/GDP is insignificant whereas the mean difference of GFCF/GDP is significant for government effectiveness (GE) and rule of law (RL). Domestic investors seem to be more sensitive to the rule of law and governments being effective in delivering public policies and services than foreign investors are. As no causality can be inferred at this level, further analyses are required to establish causally that institutions matter for investment.

Empirical Results

Effects of FDI Inflows and Institutions on Total Investment

Replicating MU

We commenced our analysis by replicating MU. Utilizing system-GMM, we conducted a regression with domestic private investment as the dependent variable. The independent variables included FDI, institutions measured by the Worldwide Governance Indicators (WGIs), their interaction with FDI, growth rate, public investment, and lagged domestic private investment, using a sample comprising 46 develo** countries (formerly used by MU). Each WGI was integrated, with the exception of Government Effectiveness (GE), in line with MU’s approach. We did not attempt to collapse the instruments, nor did we correct standard errors. Institutions and the interaction term between institutions and FDI were assumed to be exogenous following MU’s approach. Table 4 presents the results without the interaction terms, while Table 5 includes the interaction terms between institutions and FDI.

Table 4 Replication of MU without interaction terms, dependent variable: DPI

Table 5 Replication of MU with interaction terms, dependent variable: DPI

Table 4 reveals a negative and significant coefficient for FDI, indicating a crowding-out effect of FDI on domestic private investment. Additionally, the coefficients of each institutional indicator are negative and significant, except for the coefficient of regulatory quality. In Table 5, the coefficients of FDI remain unchanged, staying negative and significant, reinforcing the conclusion that FDI crowds out domestic private investment. However, the coefficients of institutions turn positive and are mostly significant for control of corruption, political stability, and regulatory quality. The coefficients of the interaction term between institutions and FDI are negative and significant. In both tables, the coefficients of GROWTH are positive and significant, while the coefficients of GFCFPU are negative but significant.

It is important to note some slight differences between MU and our replicated findings. In MU’s original findings, the direct effects of institutions on domestic investment are positive and significant, except for control of corruption. The coefficient of the interaction of the rule of law with FDI is insignificant, while the coefficient of the interaction of FDI with political stability is positive. In our replication reported in Table 5, the coefficient of the interaction of FDI with political stability is negative, and the direct effects of rule of law and voice and accountability are insignificant. These discrepancies were anticipated, primarily because we constructed the dataset ourselves, unlike FCV, who requested the dataset from MU. The sources of these slight differences can be outlined in three main points: firstly, we did not fill in missing data for governance indicators as MU did using component models. Secondly, our dataset covers the period 2002–2018, while the replication of MU is based on the period 2002–2009. Lastly, since MU did not specify the instruments used, we utilized all lagged values of endogenous variables as instruments.

Despite these slight differences, our replication maintains the central message from MU, as evident in Table 5: (1) institutions promote domestic investment, (2) FDI crowds out domestic private investment, and (3) the crowding-out effect of FDI is more pronounced when prevailing institutions are capital-friendly.

Replicating FCV

We now shift our focus to replicating FCV using our dataset. In adherence to FCV’s methodology, domestic investment was proxied by GFCF, and institutions were captured by the index of institutions computed through principal component analysis (PCA). Precisely, the index of institutions was equalized to the first principal component. FDI and its interaction with the index of institutions were considered endogenous, while institutions were maintained as exogenous, aligning with FCV’s approach. The instruments were collapsed to adhere to the rule of thumb, which stipulates those instruments must be fewer than the number of groups, and Windmeijer’s (2005) approach was utilized to obtain robust standard errors. The estimation results are presented in Table 6. Columns 3 and 4 illustrate replications of FCV with and without collapsing the instruments. Columns 1 and 2 depict replications of MU (as the dependent variable is domestic private investment) with and without collapsing the instruments, using an alternative measure of institutions in compliance with FCV.

Table 6 Replication of FCV with interaction terms, dependent variable: GFCF

The coefficient of FDI remains consistent with MU in Columns 1 and 2, even with the varied measurement of institutions. Intriguingly, the coefficient of FDI turns positive and significant during the replication of FCV, as reported in Columns 3 and 4 of Table 6. FCV interpreted the positive effect of FDI on total investment as a sign of the crowding-in effect of FDI inflows. In Columns 1 and 2, institutions had an insignificant effect on total investment, although it became significant in Column 3, only to revert to insignificance in Column 4. The coefficients of GROWTH and GFCFPU turn insignificant compared to replications of MU reported in Table 5. The coefficients of the control variables do not exhibit robustness across alternative dependent variables or altered GMM systems. Despite slight differences, which find justification in the reasons mentioned earlier, the core message from FCV persists: (1) there is no strong evidence that good institutions promote investment, (2) FDI "crowds in" domestic investment, and (3) the interaction between foreign investment and governance has a negative mediating effect on investment.

As noted previously, FCV’s interpretation of the coefficients of FDI (Columns 3 and 4 of Table 6) is inaccurate, misleading, and deviates from the commonly acknowledged interpretation in the literature (Morrissey and Udomkerdmongkol 2016). Since total investment includes domestic private, public, and foreign investments, GFCF must increase in a greater proportion than the increase in FDI inflows for one to infer a crowding-in effect. However, this is not the case with FCV’s findings. Rather, the marginal effects of FDI are significantly positive and inferior to one. The appropriate interpretation of this finding would be that FDI stimulates total investment, and not that FDI crowds in domestic private investment, as concluded by FCV. The central contention between MU and FCV now narrows down to the direct effect of institutions on investment, warranting further investigation. The unexpected negative interacting effect of institutions and FDI inflows on investment (total investment for FCV and domestic private investment for MU) is a critical point deserving scrutiny. We remain committed to evaluating both points and have augmented MU and FCV’s empirical model with financial development and its interactions with institutions.

Results of the Augmented Model

We estimated the empirical models of MU and FCV using the entire sample of 105 develo** countries covering the period 2002–2009. In Columns 1 and 2 (Table 7), the dependent variable is domestic private investment (DPI), while it becomes total investment (GFCF) in the subsequent third and fourth columns. Our findings closely align with FCV: The coefficients of FDI are significant; while negative in Columns 1 and 2, they turn positive in columns 3 and 4. The crowding-out effect of FDI inflows on domestic investment remains robust with the expanded sample size. The impact of institutions on total investment becomes significant, and the interacting effect of institutions and FDI on total investment is negative and significant. The coefficients of the lagged dependent variables are significant, as expected, confirming the suitability of GMM in modeling the impact of FDI on total and domestic investments. As anticipated, there is evidence of first-order serial correlation (AR1) and no evidence of second-order serial correlation (AR2). The p values of Hansen statistics are not significant, affirming the validity of the instruments used.

Table 7 MU and FCV’s models based on the current study’ sample—105 develo** countries

Subsequently, we estimated our empirical model—the augmented MU and FCV’s empirical model—which includes time dummies. Column 3 of Table 8 presents the main results of our empirical model with interaction terms. For brevity, we did not report the coefficients of the time dummies. The estimation of parameters of our empirical model adheres to the system-GMM approach akin to FCV. Institutions were measured comprehensively by the index of institutions, and both variables of interest (FDI and GFCF) were expressed in logarithmic terms. Therefore, the coefficient of FDI can be directly interpreted as the direct elasticity of GFCF with respect to FDI in the remainder of the paper. We primarily reported the number of instruments, the number of groups, and the p-values of the Hansen J and Arellano-Bond tests. The statistics associated with the AR (2) test are significant, indicating no second-order autocorrelation in the idiosyncratic error term. The Hansen J test’s statistics are insignificant, aligning with instrument exogeneity. The number of groups exceeds the number of instruments, suggesting compliance with Roodman’s (2009a) rule of thumb, ensuring unbiased estimates. The positive and significant coefficient of GFCF (as a percentage of GDP) lagged by one period implies a positive correlation between current and past values of GFCF. The diagnostic results in Column 3 of Table reveal that the estimates are unbiased across all the reported equations.

Table 8 Results of the system-GMM model-105 develo** countries over 2002–2018

Column 1 of Table 8 presents the estimates of the empirical model without interaction terms. The coefficients of FDI, GROWTH, and GFCFPU are significant, while the coefficients of INST, FIN, and the interaction terms are insignificant. The same results hold after controlling for each worldwide governance indicator (see Panel A of Table A1). When the interaction of FDI and INST is included (Column 2), and even after controlling for each WGI (see Panel B of Table A1), the coefficient of FDI and GFCFPU remain significant, GROWTH turns insignificant, while INST, FIN, and the interaction terms remain insignificant. The results presented in Column 2 are akin to those in Column 3 of Table 8, which includes both interactions—FDI-institutions and FDI-financial development nexuses. Even when institutions are measured by control of corruption (CC), government effectiveness (GE), political stability and absence of violence (PS), regulatory quality (RQ), rule of law (RL), and voice and accountability (VA), the results in Columns 3 still hold (refer to Panel B of Table A2).

In summary, from columns 1 to 3, the coefficient of FDI is positive and significant but less than unity. The coefficient of GROWTH, significant in Column 1, becomes insignificant in columns 2 and 3, while the coefficient of GFCFPU is significant across all three columns. INST, FIN, and their interaction with FDI inflows show no significant impact on total investment. Overall, we find compelling evidence that FDI inflows deter domestic investment, in line with FCV and MU. Unlike MU and FCV, there is no robust evidence that the interaction between FDI inflows and institutions has a negative mediating effect on investment. Consistent with FCV, our findings seem to affirm that institutions do not exert a significant impact on total investment.

Robustness Checks

An Alternative Measure of Financial Development

In this paper, we have operationalized institutions and financial development through binary variables, distinguishing between two regimes: capital-friendly institutions and capital-unfriendly institutions (or well-developed financial markets and shallow financial markets, respectively). From the primary outcomes presented in Column 3 of Table 8, we ascertain that FDI displaces domestic investments, while public investment stimulates total investment. Additional findings emphasize that institutions, financial development, and their interactions with FDI inflows do not yield a significant impact on total investment.

In a preliminary robustness check, we sought to determine the resilience of these results to an alternative measure of financial development. Rather than utilizing a dummy variable, we represented financial development using the raw value of the IMF’s financial development index. The results, as disclosed in Column 4 of Table 8, mirror the previously reported findings: FDI displaces domestic investment, and institutions, financial development, and their interactions with FDI inflows demonstrate no substantial effect on total investment. These outcomes persisted when we conducted the empirical model estimation with an alternative measure of financial development, while institutions were gaged by binary WGIs (CC, GE, PS, RL, RQ, and VA), as detailed in Panel A of Table A3 in the annexures.

An Alternative Measure of Institutions

In the second robustness check, we gaged institutions using the raw indices of Worldwide Governance Indicators (WGIs), encompassing CC, GE, PS, RL, RQ, and VA, rather than binary variables as detailed in Table 4. The results, derived through system GMM and presented in Column 5 of Table 8, are as follows: the coefficient of FDI remains positive, significant, and below unity. However, the coefficients of INST, FIN, INST × FDI, and INST × FIN are deemed insignificant. These outcomes reaffirm that institutions, FDI, financial development (FD), and the associated interaction terms lack a significantly positive impact on GFCF.

Consistent with prior observations, the coefficients of CC and GE are positive and significant, with the interaction of CC and FDI demonstrating a negative correlation with GFCF. Once again, these results further negate the crowding-in hypothesis of FDI and uphold the crowding-out hypothesis. They underscore that public investment and institutions are consequential for private investment, influencing both domestic and foreign investors. Importantly, the essence of our findings remains unaltered when institutions are appraised through the raw estimated values of worldwide governance indicators (WGIs), including CC, GE, PS, RL, RQ, and VA (refer to Panel B of Table A3 in the annexes).

An Alternative Identification Strategy and Measure of Domestic Investment

So far, the transformed equation in the system GMM has been derived by taking the first difference of the empirical model. We now scrutinize the robustness of the main results under an alternative identification strategy. Pursuing this objective led us to employ an orthogonal transformation to obtain the transformed equation in the system GMM. Encouragingly, our findings remain robust to this transformation (results are available but not reported to conserve space).

In addition, we opted for an alternative measure of domestic investment. Rather than using GFCF, we employed private gross fixed capital formation as a proxy for domestic investment, capturing investments exclusively made by the private sector. Remarkably, the coefficients of FDI and GFCFPU persist in being positive and significant, while the impacts of institutions, financial development, and their interactions with FDI inflows remain consistently insignificant (results are available but not reported to save space).

Sub-period Analysis

We conducted an additional analysis by performing a sub-period examination. The coefficients of the empirical model were estimated for two distinct sub-periods, namely 2002–2009 and 2010–2018, using the sample of 105 develo** countries. Results for the sub-periods 2010–2018 and 2002–2009 are respectively presented in Columns 6 and 7 of Table 8. In Column 6, none of the coefficients for the study variables are statistically significant, whereas in Column 7, the coefficients of FDI, GROWTH, and GFCFPU attain significance. These outcomes persist even when institutions were measured using binary WGIs (refer to Panels A and B of Table A4).

Specifically, the coefficients of FDI exhibit weak significance in Columns 1 and 6, and insignificance in Columns 2 to 5 of Panel A. Conversely, the coefficients of institutions consistently maintain significance across Panel B of Table A3 (in the appendices). The sub-period analysis appears to affirm that institutions, financial development, and their interactions with FDI inflows do not propel total investment. While the hypothesis of a crowding-in effect appears improbable in both sub-periods, the crowding-out effect of FDI inflows seems more pronounced in the 2010–2018 sub-period than in the 2002–2009 sub-period.

Discussion of the Main Findings

Our study reveals that FDI inflows crowd out domestic investment in develo** countries. We robustly observe that FDI inflows have a positive impact on total investment, with a marginal effect less than unity. This establishes the absence of the crowding-in effect of FDI inflows on domestic investment. This finding aligns with FCV and MU, suggesting that FDI inflows diminish incentives for domestic investment. The market channel appears to be the mechanism at work, where foreign firms, perceived as more productive than local counterparts, capitalize on available investment opportunities, leading to the displacement of inefficient local firms. This lack of crowding-in effect may be attributed to the prevalence of foreign firms in primary sectors where technological spillover rarely occurs (Adams 2009; Dupasquier and Osakwe 2006). The dominance of greenfield investment in develo** countries (Levasseur 2002; Diallo et al. 2021) could explain the contemporaneous crowding-out effect identified in the current article. For instance, Chen et al. (2017) found that wholly foreign-funded enterprises crowd out domestic investment, while equity joint ventures exhibit a crowding-in effect in China. A recent investigation by Jude (2019) reported that greenfield investment crowds out domestic investment in the short run with a marginal crowding-in effect in the long run.

A significant finding pertains to the direct and mediating effects of institutions. Even after accounting for financial development and using a large sample of develo** countries, we found no evidence that good institutions encourage investment, contrasting with the findings of FCV. We discovered a nonsignificant mediating effect of institutions on total investment. This contradicts FCV and MU, who documented a negative mediating effect of institutions on the FDI–GFCF nexus. We suspect that the misspecification problem inherent in MU and FCV’s empirical model, stemming from their failure to control for financial development, may be the source of this divergent outcome. Although financial development has no significant impact on domestic investment, it might be correlated with FDI and institutions in a way that ignoring it would lead to omission bias problems. However, the insignificant effect of institutions on total investment found in our study is unexpected. This may occur if foreign and domestic investors react differently to institutions, as suggested by MU. Improving the quality of institutions could attract greater FDI inflows that may displace domestic investment opportunities (crowding-out effect), leaving total investment unchanged if both effects cancel each other out.

Financial development can either promote or discourage investment. A well-developed financial market has the potential to alleviate financial constraints on local investors, stimulating domestic investment (Ndikumana 2005; Braga Tadeu and Moreira Silva 2013; Misati and Nyamongo 2011). Conversely, financial development may discourage domestic investment when it is biased toward a prioritized sector, as seen in some develo** countries where the banking system favors the industrial sector over services when it comes to attributing credits. Ang (2009) demonstrated that directed credit programs favoring prioritized sectors can discourage private investment in countries like India and Malaysia. Lahcen (2004) suggested that financial development may hurt private investment when credit is allocated to households rather than local firms, reducing the availability of loans for the business sector. Unexpectedly, our study found no significant effect of financial development. Whether financial development has a neutral, positive, or negative effect on investment is an empirical question. The observed neutral effect in our study may result from a balance between the positive and negative effects of financial development on investment, considering the diverse financial policies implemented in develo** countries.

Two important methodological caveats must be highlighted. First, institutions and financial development were assumed to be exogenous. A growing literature suggests that institutions are endogenous. From Acemoglu et al.’s (2001) perspective, early European mortality rates affect the colonization strategy (the settlement) which determines the current institutions. Where the mortality rates were high, settlers were more likely to establish extractive institutions that persisted after independence. Institutions, in turn, determine the level of financial development (Acemoglu and Johnson 2005). It has been documented in the literature that countries’ legal origins determine both institutions and financial development (Porta et al. 1997). Despite our awareness of the endogenous nature of institutions and financial development, both variables were assumed to be exogenous to comply with MU and FCV. Such an assumption prevents the proliferation of the number of instruments. Another caveat rests on the fact that FDI has been instrumented with internal instruments solely, which may not entirely dissolve the confounding effects of time-varying unobserved heterogeneities. However, we retain confidence in the estimates for two reasons. First, the coefficient of FDI remains stable across all estimations. Second, GMM techniques are advanced techniques that handle partially—if not completely—unobserved heterogeneities. By including a transformed equation obtained through differentiation, system-GMM cancels out any time-invariant heterogeneities. Future studies may want to use either external instruments or an instrumental variable approach.

Concluding Remarks

The current study aims to reevaluate the findings of Morrissey and Udomkerdmongkol (2012, henceforth MU) and Farla et al. (2016, henceforth FCV) regarding the crowding-out effect of FDI inflows and the mediating role of institutions. Both MU and FCV utilized the same panel database of 46 develo** countries from 1996 to 2009. While MU observed a negative impact of FDI inflows on domestic investment (DI), FCV reported a positive effect of FDI inflows on total investment with a coefficient of FDI less than unity. Despite both studies seemingly supporting the crowding-out hypothesis, FCV misleadingly concluded a crowding-in effect of FDI inflows. However, the results regarding the impact of institutions were contradictory, with MU suggesting that institutions promote DI, while FCV found inconclusive evidence.

To address these discrepancies, our study employs a larger dataset covering 105 develo** countries from 2002 to 2018, ensuring external validity. We enhance MU and FCV’s empirical model by incorporating financial development and its interaction with institutions to mitigate omission variable bias. Utilizing the system-GMM and adhering to FCV’s methodological protocol, we collapse instruments and employ Windmeijer’s (2005) correction method to address the limitations highlighted by FCV in MU. Dummy time variables are introduced to account for contemporaneous correlation of error terms. Domestic investment (DI) and foreign investment are proxied using the logarithm of gross fixed capital formation and net FDI inflows, respectively. Worldwide governance indicators measure institutions, while the IMF’s Index of Financial Development gages financial development. These variables are transformed into binary form in the primary estimations, with alternative measures explored in robustness checks.

Our replication of FCV and MU’s findings with our dataset confirms the crowding-out effect of FDI on domestic private investment and the negative mediating effect of institutions on the FDI-DI nexus. In agreement with FCV, we find no significant effect of institutions on investment. Importantly, our study does not robustly support the notion that institutions mediate the FDI–DI relationship in develo** countries. Furthermore, financial development and its interaction with FDI inflows do not exhibit a significant impact on investment.

The core message of our study is that FDI inflows tend to crowd out domestic investment, with a heightened crowding-out effect in recent years. This finding underscores the need for policymakers to carefully consider the potential negative aspects of FDI and formulate policies that promote the benefits associated with FDI inflows. Policy recommendations include improving the business environment to facilitate technological spillovers, fostering partnerships between multinational and local investors, and widening investment opportunities for local firms, especially in the manufacturing sector.

Despite the robustness of our findings and in addition to the potential caveats underlined in the discussion section, further limitations should be acknowledged. First, the aggregate measure of FDI inflows may overlook nuances related to entry modes, sectoral distribution, and types of firms involved. Future research could delve into these aspects for a more comprehensive understanding. Second, this study relied on macroeconomic data to investigate the impact of FDI inflows on domestic investment while controlling for the quality of institutions. The exploration of firm-level data could offer detailed insights into the mediating role of institutions in the FDI-DI nexus.

Appendix

See Tables

Table 9 Results of the system GMM without interaction terms, controlling for financial development

9,

Table 10 Results of the system GMM, controlling for financial development and interaction terms

10,

Table 11 Results of the system GMM– alternative measures of institutions and financial development

11, and

Table 12 Results of the system GMM– sub-period analyses

12.