FDI and Corruption: Panel Evidence from EU Member States

Abstract

This paper examines the relationship between corruption and foreign direct investment (FDI) inflows in European Union member states using a dynamic panel framework. Using an unbalanced EU panel from 2002 to 2022 and an Arellano–Bond difference-GMM specification, we model inward FDI inflows per capita as a function of institutional integrity (measured by Transparency International’s Corruption Perceptions Index), market size, development level, and trade integration. The results show a robust positive association between improvements in perceived integrity (higher CPI scores) and increases in inward FDI inflows per capita, conditional on macroeconomic controls and dynamic adjustment. Market size and trade variables have the expected signs, while GDP per capita is the empirically sensitive margin, consistent with the idea that higher development can indicate greater purchasing power but also higher costs and saturation effects in advanced economies. Robustness checks using the inverse hyperbolic sine transformation—suited to heavy tails, zeros, and negative net flows—confirm that the governance association is not an artifact of scaling. The findings highlight the importance of institutional quality and market openness as correlates of FDI attractiveness within the EU.

1. Introduction

Foreign direct investment (FDI) is one of the most visible outcomes of firms’ international location decisions. For host economies it can relax capital constraints, transmit technology and managerial practices, and generate spillovers through supplier linkages and competitive pressure (World Bank, 1993; Chirila-Donciu, 2014). In the European Union (EU), these channels matter not only for growth but also for the performance of the single market and long-term agendas associated with the Lisbon Strategy and the Sustainable Development Goals. Recent assessments note that the EU remains a major outward investor, yet its relative position as a destination has weakened, making inward attractiveness a renewed policy concern (European Council, 2025; Eurostat, 2025b).

Corruption can influence these location decisions by changing expected returns and the distribution of risk. It may raise transaction costs, create discretionary barriers, and substitute informal payments for transparent rules—the classic “grabbing-hand” mechanism (Krueger, 1974; Shleifer & Vishny, 1993; Mauro, 1995; Tanzi & Davoodi, 1997). A minority view argues that where regulation is excessively burdensome, bribes may occasionally speed up administrative processes (“greasing the wheels”) (Leff, 1964; Lui, 1985). In FDI applications, empirical findings mirror this tension: many studies report a deterrent effect of corruption (Wei, 2000; Abed & Davoodi, 2000; Al-Sadig, 2009), while some show weak or context-dependent relationships (Quazi et al., 2014; Navickas et al., 2016).

In the EU, where regulatory density is comparatively high but enforcement quality and rule-of-law credibility still vary across member states, both the “grabbing-hand” and “helping-hand” mechanisms remain plausible, so the sign and magnitude of the governance–FDI link are ultimately empirical.

For EU member states the issue is not merely academic. Although institutional standards are comparatively high, perceived corruption continues to affect public trust and is repeatedly cited as a drag on competitiveness and investment (European Commission, 2022; European Union, 2025; European Parliament, 2025). Existing cross-country evidence often pools EU and non-EU economies or relies on broad governance composites; fewer studies focus on the CPI, which remains the indicator most visible in policy debates (Rose & Heywood, 2013; Torsello & Venard, 2015; Hamilton & Hammer, 2018). As a consequence, estimates for advanced-economy settings remain unsettled, both in sign and in magnitude (Zangina et al., 2020; Blanco-Alcántara et al., 2022; Farinha & López-de-Foronda, 2024; Dincă & Strempel, 2025).

This paper therefore concentrates on a parsimonious set of determinants that are repeatedly identified as first-order drivers of inward FDI: perceived corruption (CPI), aggregate market size (GDP), income level (GDP per capita), and trade intensity, proxied by exports and imports per capita. GDP captures demand and scale effects relevant for market-seeking investors (Kosteletou & Liargovas, 2000; Moudatsou & Kyrkilis, 2011; Jones et al., 2020). GDP per capita reflects purchasing power and development, but in mature economies it can also proxy higher costs that discourage efficiency-seeking projects (Veeramacheneni et al., 2007; Sahiti et al., 2018; Sabir et al., 2019). Trade variables may complement FDI in vertical and export-platform strategies, yet substitution is also possible, which helps explain mixed findings in the literature (Dunning, 1980; Aizenman & Noy, 2006; Albulescu & Goyeau, 2019; Majeed & Ahmad, 2007; Coconoiu, 2014).

Overall, three issues motivate a fresh EU-wide assessment: (i) corruption remains theoretically central but empirically fragile; (ii) GDP and GDP per capita are often used interchangeably, despite reflecting distinct mechanisms; and (iii) trade and FDI may be jointly determined. Methodologically, many earlier contributions employ cross-sectional regressions or static panels, which struggle with persistence in FDI and with reverse causality (Leamer, 1983; Leamer & Leonard, 1983).

To address these points, we estimate a dynamic panel model for 27 EU member states over 2002–2022, using FDI inflows per capita as the dependent variable. The specification allows for inertia in inflows and uses difference-GMM (Arellano & Bond, 1991) to handle unobserved country effects and endogeneity, following standard guidance for dynamic panels (Baltagi, 2008; Wooldridge, 2010; Roodman, 2009).

The empirical analysis focuses on four questions: (1) whether changes in perceived corruption (CPI) are associated with FDI per capita; (2) whether larger market size (GDP) attracts more inward investment; (3) whether higher income levels (GDP per capita) raise or reduce marginal inflows in advanced economies; and (4) how exports and imports per capita relate to inward FDI across EU members.

This paper contributes in three ways. First, it provides EU-wide evidence on the FDI–corruption relationship using the CPI over a long post-2002 horizon that spans enlargement waves, the global financial crisis, and recent geopolitical disruptions. Second, it estimates corruption, market size, income level, and disaggregated trade measures jointly within a dynamic panel framework that accounts for persistence and endogeneity. Third, by restricting attention to the EU, it speaks directly to policy debates in advanced economies that aim to pair high institutional standards with sustained inward investment.

This paper proceeds as follows. Section 2 summarizes the relevant theoretical mechanisms and the empirical evidence. Section 3 describes the data, variable construction, and the econometric approach. Section 4 reports the estimation results, Section 5 interprets them in relation to prior findings, and Section 6 concludes with policy implications and directions for further research.

2. Literature Review

This section summarizes the main theoretical arguments and empirical findings linking inward FDI to corruption and to the macro controls used here (GDP, GDP per capita, exports per capita, and imports per capita).

Statistical manuals treat FDI as cross-border investment that conveys a lasting managerial interest; empirically, this is typically operationalized by an ownership stake of around 10% or more of voting power. Applied studies use both flows and stocks, reported in levels, per capita, or relative to GDP, and sometimes decomposed into equity, reinvested earnings, and intra-company debt (Pekarskiene & Susniene, 2015). Using a large cross-country panel, Kerner (2014) shows that both flow and stock measures respond to demand and growth conditions and tend to be lower where rule of law is weak and corruption is more prevalent, while capital-account liberalization is associated with higher inflows.

Through ownership and control links, FDI can affect host economies well beyond the initial capital transfer—via technology diffusion, managerial know-how, access to export networks, and competitive pressure on incumbents (Chirila-Donciu, 2014). For EU entrants and smaller economies, these channels feature prominently in debates about convergence and productivity. Modi (2004) documented the rapid acceleration of global FDI since the 1970s and its increasing importance relative to trade flows. The net effect is not mechanically positive: inward projects can enlarge the tax base and diversify production, but they also imply profit repatriation and may intensify competition faced by domestic firms. In line with European Central Bank (2018) reporting, the EU is both a major outward investor and a large destination, and current policy strategies emphasize improving the climate for high-quality inward investment.

In this paper, corruption is understood as the misuse of public or entrusted authority for private benefit, encompassing bribery, kickbacks, clientelism and related practices that distort allocation and weaken trust (European Commission, 2022). The concept spans petty and grand forms and may be systemic or episodic (Rose & Heywood, 2013). Despite measurement debates, the Corruption Perceptions Index remains the most visible cross-country proxy in applied work. Evidence for EU members suggests that stronger property-rights enforcement precedes improved “freedom from corruption”, pointing to an institutional channel relevant for investors (Mance & Pečarić, 2016). Torsello and Venard (2015) stress that any empirical proxy is necessarily partial and that historical and cultural context can shape how corruption operates.

A useful starting point is Becker’s (1968) expected-utility framework: a corrupt act is chosen when the expected payoff exceeds the expected penalty, which is governed by detection probability and sanctions. Macro evidence suggests that corruption reallocates public spending towards large, discretion-heavy projects and away from maintenance and human capital, with adverse growth consequences (Mauro, 1995; Tanzi & Davoodi, 1997). Lambsdorff (1999) emphasizes the role of legal institutions and inequality. Dynamic-panel studies also point to non-linearities, suggesting that reductions from high baseline levels produce the largest growth payoffs (Ahmad et al., 2012). Against this backdrop, any short-run administrative “speed-up” effects are typically dominated by the longer-run distortionary costs.

The FDI implications are usually framed between two mechanisms. Under the “grabbing-hand” view, corruption behaves like an unpredictable tax that raises transaction costs and discourages entry (Krueger, 1974; Shleifer & Vishny, 1993; Tanzi & Davoodi, 1997). Under a “helping-hand” view, bribes may substitute for red tape in heavily regulated settings (Leff, 1964; Beck & Maher, 1986; Lui, 1985). For multinationals, Tanzi (1998) notes that corruption can even be used to secure preferential treatment, which could attract certain projects. Kaufmann (1997) cautions that such “advantages” depend on discretion and can differ across institutional contexts. Zhou (2007) proposes a useful distinction: non-discriminatory payments that merely obtain routine services tend to deter investment, while discriminatory payments that deliver privileged treatment may, in some settings, increase FDI. In the EU, where regulatory density is high but enforcement quality and rule-of-law credibility vary across member states, both mechanisms are plausible, so the sign and magnitude of the governance–FDI link remain an empirical question.

Empirical studies on corruption and FDI reflect this theoretical ambiguity. Early work by Wheeler and Mody (1992) analyzed US manufacturing multinationals in 42 countries and found that the corruption proxy was insignificant. In contrast, Wei (2000), using bilateral FDI stocks for 12 source and 45 host countries, reported a significant negative effect of corruption on FDI. Smarzynska and Wei (2002) combined firm-level data from Eastern Europe and the former Soviet Union with alternative corruption indices and found that corruption reduces the profitability and volume of FDI. Abed and Davoodi (2000) showed that transition economies with lower corruption attract higher FDI per capita. More recent panel studies have produced mixed results. Hossain (2016) found large negative effects of corruption on FDI for a broad set of countries, while Quazi et al. (2014) obtained results consistent with a “helping-hand” mechanism in some contexts. The diversity of findings is partly explained by differences in corruption measures (CPI, World Bank governance indicators, commercial risk ratings), sample composition (developed versus developing economies), model specification, and estimation method (cross-section, static panel, dynamic panel).

This sensitivity is closely related to the broader econometric critique by Leamer (1983) and Leamer and Leonard (1983), who argued that, in the presence of many plausible explanatory variables and flexible functional forms, results are vulnerable to “specification search”. FDI is influenced by a wide set of macroeconomic, institutional, and structural variables, and researchers can alter variable definitions (levels, logs, squared terms), sample periods, or subsets of regressors until they obtain “interesting” coefficients. Studies on corruption and FDI are no exception: some use log FDI per capita as the dependent variable and log GDP per capita as a regressor (Al-Sadig, 2009), while others employ different combinations of logs, levels, and non-linear terms (Castro & Nunes, 2013). This flexibility complicates comparison and raises concerns about robustness.

Beyond corruption, the literature identifies GDP, GDP per capita, and trade as central macroeconomic determinants of FDI, but their effects are not uniform. Sahiti et al. (2018) emphasize that the impacts of overall GDP (market size) and GDP per capita (living standard) on FDI remain debated: some studies find positive and significant effects, while others find weak, negative, or insignificant relationships. Veeramacheneni et al. (2007) note that research has focused more on the impact of FDI on growth than the reverse and call for more work on how income and market size attract FDI. For developing economies, Mottaleb (2007) finds that GDP and its growth stimulate FDI, while higher corruption reduces it; GDP per capita is positive in some cases, and trade integration generally supports inflows. Moosa and Cardak (2006), using a large sample including EU members, report mostly positive effects of GDP and GDP per capita and positive roles for exports and infrastructure, whereas poor business environment indicators deter FDI.

In the EU context, Kosteletou and Liargovas (2000) show that market size positively affects FDI, while exchange rate volatility is detrimental. Jones et al. (2020) also find that market size, its growth, and transport infrastructure promote FDI, whereas high taxation and extensive government intervention can discourage it. Simionescu (2016), using a Bayesian panel approach for EU members, finds that the GDP–FDI relationship can vary across countries and over time, although it is predominantly positive. Moudatsou and Kyrkilis (2011) report that GDP per capita has significant positive effects on FDI in EU and ASEAN members and argue that FDI in the EU is largely growth-driven. Botrić and Škuflić (2006) show that, in Central and South-Eastern Europe, the impacts of market size and living standards can be positive or negative depending on the sample, although trade generally supports FDI. Sabir et al. (2019) distinguish between developed, transition and emerging economies and find that GDP per capita has a negative effect on FDI in developed economies but a positive effect in emerging and transition countries; institutional quality (including corruption) and trade integration tend to promote FDI, while inflation is detrimental.

Trade-related determinants have also attracted substantial attention. In the OLI paradigm, Dunning (1980) treats trade variables (exports, imports) as part of location advantages, and FDI can substitute for or complement trade depending on the motive (horizontal versus vertical FDI). Aizenman and Noy (2006), using a panel of low-, middle- and high-income economies, show that total trade and GDP per capita generally increase FDI, but the role of merchandise versus services trade is heterogeneous, and institutional variables, including corruption and democracy, have ambiguous effects. Albulescu and Goyeau (2019) argue that the joint dynamics of exports, imports and FDI in EU member states remain understudied despite their importance for EU trade and investment policy. They found bidirectional and positive relationships between inward FDI, exports and imports in Central and Eastern European countries. Other studies report similarly mixed evidence. Ekholm et al. (2007) define “exporting FDI” and distinguish vertical and horizontal forms, emphasizing that export-oriented zones and production networks make exports an important determinant of FDI. Erkekoglu and Kilicarslan (2016) find that exports, GDP, lower corruption and lower inflation are associated with higher FDI in a large cross-country panel, although institutional indicators differ in their impacts. Majeed and Ahmad (2007) find a positive effect of GDP growth and exports on FDI in low- and middle-income economies. Boubacar (2013), using US FDI in OECD economies, finds that exports and GDP positively affect FDI, whereas total population and some integration dummies have negative effects. Coconoiu (2014) shows that in EU countries the effects of exports and imports on FDI are usually positive, though their significance and relative strength vary. Root and Ahmed (1978) find that import capacity has a positive effect on FDI in low- and middle-income economies, whereas GDP and total trade can sometimes have negative coefficients. Chaisrisawatsuk and Chaisrisawatsuk (2007) report positive two-way relationships between trade and FDI for OECD and ASEAN economies, while Brainard (1997) documents mostly positive effects of exports, imports and GDP per capita on FDI, but with considerable cross-country variation.

In short, corruption and the macro controls employed here are repeatedly identified as core correlates of inward FDI, but results differ across samples and methods. Corruption is most often found to reduce investment, although conditional or insignificant effects appear in some settings. GDP typically enters with a positive sign, whereas GDP per capita can switch signs in high-income groups, consistent with a cost-versus-market-size trade-off. Trade indicators often complement FDI in global value chains, but substitution is also plausible. For the EU, there is still limited long-horizon evidence that considers CPI, GDP, GDP per capita, and disaggregated trade measures jointly within a dynamic framework that addresses both persistence and endogeneity. These considerations motivate our empirical specification and the hypotheses stated in Section 3.1.

3. Data and Methods

3.1. Hypotheses

The literature reviewed above implies clear sign predictions for governance quality and market size, while income-level effects in advanced economies can be ambiguous because higher GDP per capita proxies both purchasing power and higher cost levels. In a highly regulated setting such as the European Union, this ambiguity is empirically relevant. We therefore test the following hypotheses:

H1. 

Improvements in perceived integrity (a higher CPI score, i.e., lower perceived corruption) are associated with higher inward FDI inflows per capita.

H2. 

Increases in market size (GDP) are associated with higher inward FDI inflows per capita.

H3. 

Changes in GDP per capita have an ambiguous association with inward FDI inflows per capita in advanced EU economies, reflecting a purchasing-power channel versus a cost-saturation channel.

H4. 

Greater trade integration (exports and imports per capita) is associated with higher inward FDI inflows per capita.

3.2. Data, Sample, and Scope

The empirical analysis uses annual country-level data to study inward FDI in the EU. Building on recent discussions of EU investment attractiveness (OECD, 2025a, 2025b, 2025c; Amdam & Benito, 2022; Saurav & Kuo, 2020), we assembled a panel for the 27 member states covering 2002–2022.

Data availability varies across variables and countries, so the baseline panel is unbalanced. The estimations use all observed country–year pairs. Macroeconomic and trade series are taken from the World Development Indicators (World Bank), while perceived corruption is measured by the Corruption Perceptions Index published by Transparency International (Transparency International, 2024).

All transformations, estimations, and post-estimation diagnostics are carried out in EViews 14.

3.3. Variables

The dependent variable is FDI inflows per capita (FDIPC), measured as net FDI inflows divided by total population (current US dollars per capita).

The core explanatory variables are chosen to reflect market size, income level, institutional quality, and trade integration. Market size is proxied by gross domestic product in current US dollars (GDP). The standards of living are proxied by GDP per capita in current US dollars (GDPPC). Perceived corruption is proxied by the Corruption Perceptions Index (CPI), where higher values indicate lower corruption. Trade intensity is proxied by exports per capita (EXPPC) and imports per capita (IMPPC), both in current US dollars per capita. Table 1 summarizes the variables and data sources.

Table 1. Variables.

Variable	Meaning	Description	Database
FDI pc	FDI per capita	Investment activity	World Bank (WDI)
GDP	Gross domestic product	Market size	World Bank (WDI)
GDP pc	GDP per capita	Living standard	World Bank (WDI)
CPI	Corruption perception index	Corruption	Transparency international
EXPPC	Export per capita	Trade	World Bank (WDI)
IMPPC	Import per capita	Trade	World Bank (WDI)

FDI per capita is used because it scales flows by population and improves comparability across EU member states with different country sizes. The focus on FDI determinants is motivated by both policy relevance and remaining empirical gaps, as emphasized in the literature and policy documents (e.g., OECD, 2025a; European Council, 2025; Eurostat, 2025a, 2025b).

To represent market size, we use GDP (current USD), the standard demand proxy in location-choice studies (World Bank, 2025; Callen, 2025; Park et al., 2018). GDP per capita captures the level of development and purchasing power (Eurostat, 2025a; Grigoly & Ley, 2012). Including both variables help separate pure scale effects from income-level effects, which need not move together across EU members. Because GDP and GDP per capita are mechanically related, multicollinearity is possible. Therefore, we interpret their coefficients jointly and treat specifications that exclude one as robustness checks. It is important to recognize that total GDP reflects absolute market size, a primary driver of horizontal FDI, while GDP per capita indicates average consumer purchasing power, which shapes demand. Along with a country’s trade openness, these factors determine whether FDI and imports act as complements or substitutes, ultimately defining the coefficients and directional signs in trade-investment models.

CPI is included because corruption is a central institutional factor in the FDI literature and remains the subject of substantive debate regarding measurement and mechanisms (Shleifer & Vishny, 1993; Aidt, 2003; Kaufmann, 1997; Zhou, 2007). This is particularly relevant for the EU, where anti-corruption policy and institutional quality are frequently discussed as determinants of economic outcomes (e.g., European Commission, 2022; European Union, 2025; European Parliament, 2025; Blanco-Alcántara et al., 2022; Dincă & Strempel, 2025). The focus on corruption is also motivated by the fact that parts of the literature still treat the corruption–FDI nexus as under-researched or insufficiently resolved in comparable samples (Zangina et al., 2020).

Trade is measured by exports per capita and imports per capita because exports and imports are basic components of international integration and are commonly used in empirical work examining the trade–FDI linkage (Aizenman & Noy, 2006; Majeed & Ahmad, 2007; Albulescu & Goyeau, 2019). Using both exports and imports allows the analysis to capture trade exposure more fully than a single indicator.

The model is intentionally parsimonious. In short-to-moderate T panels estimated by GMM, adding many correlated controls can rapidly expand the instrument set, weaken over-identification diagnostics, and increase the risk of overfitting. We therefore focus on a small set of core determinants that are theoretically central and consistently available for all EU members: institutional quality (CPI), market size (GDP), development level (GDP per capita), and trade integration (exports/imports per capita or openness). Other determinants frequently used in the FDI literature (e.g., human capital, unit labor costs, corporate taxation, infrastructure quality, political stability indices, and sectoral composition) are relevant but are treated here as potential omitted factors and priorities for future extensions.

3.4. Econometric Model and Estimation

Country panels are appropriate here because they exploit both cross-country differences and within-country changes over time, while allowing for unobserved heterogeneity that would bias pooled regressions (Baltagi, 2008; Wooldridge, 2010).

To ensure that the subsequent dynamic panel estimates are not driven by spurious regression due to the persistence of the lagged variable or due to the unit roots, we first examine the time-series properties of all variables used in the baseline specification.

Table 2 reports Augmented Dickey–Fuller (ADF) unit root tests conducted for each series both in levels and in first differences.

Table 2. Unit root test results (ADF).

At Level		FDIPC	PIPC	CPI	GDPPC	OPEN	EXPPC	IMPPC
With Constant	t-Statistic	0.4659	0.0739	0.3013	0.9878	0.1294	0.9841	0.9928
	Prob.	0.1129	0.0935	0.2542	0.5412	0.9849	0.9931	0.9989
With Constant and Trend	t-Statistic	0.5543	0.1251	0.1777	0.9735	0.0475	0.1494	0.9647
	Prob.	0.9542	0.1577	0.4707	0.6464	0.6276	0.9578	0.9977
Without Constant and Trend	t-Statistic	0.1904	0.0271	0.7106	0.9989	0.8279	0.9980	0.9875
	Prob.	0.3385	0.0085	0.9332	0.9316	0.9547	0.9966	0.9754
At First Difference		d(FDIPC)	d(PIPC)	d(CPI)	d(GDPPC)	d(OPEN)	d(EXPPC)	d(IMPPC)
With Constant	t-Statistic	0.0297	0.0002	0.0005	0.1126	0.0066	0.0087	0.0140
	Prob.	0.0002	0.0006	0.0157	0.0139	0.1263	0.0038	0.1476
With Constant and Trend	t-Statistic	0.0060	0.0020	0.0029	0.2525	0.0459	0.0549	0.0584
	Prob.	0.0123	0.0026	0.0378	0.0582	0.1327	0.0175	0.0738
Without Constant and Trend	t-Statistic	0.0020	0.0000	0.0000	0.0706	0.0003	0.0438	0.0289
	Prob.	0.0000	0.0000	0.0023	0.0015	0.0470	0.0042	0.0331

Null Hypothesis: the variable has a unit root. Lag Length based on SIC. Probability based on one-sided p-values.

ADF unit root tests are reported only as a diagnostic; the Arellano–Bond framework does not require pretesting. The evidence of strong persistence, especially for FDI, supports specifying the model in first differences, so we work with I(0) changes while eliminating time-invariant country effects. The ADF results also indicate that trade openness (OPEN) remains non-stationary under several deterministic specifications; therefore, we exclude OPEN and instead use EXPPC and IMPPC. We use Difference GMM as the baseline estimator because, under the standard internal-instrument conditions, it delivers consistent inference and, crucially in panels with limited N, allows strict control of the instrument count through a disciplined lag depth.

System-GMM can improve efficiency when variables are highly persistent, but it relies on additional moment conditions and can generate very large instrument sets in small panels, potentially overfitting endogenous components and weakening Hansen tests. Given the EU’s relatively small cross-section and the need to keep instruments parsimonious, we take difference-GMM as the baseline and emphasize careful diagnostics (Hansen J and Arellano–Bond tests) and robustness checks via alternative specifications and transformations. System-GMM is therefore best viewed here as a potential extension rather than the baseline specification. System-GMM (Arellano–Bover/Blundell–Bond) can improve finite-sample performance when the series are persistent, but it can also lead to instrument proliferation. For this reason, it is best used with strict instrument reduction, such as collapsed instruments and limited lag depth.

We report the usual specification checks for difference-GMM: serial-correlation tests in the differenced residuals (AR(1) and AR(2)) and an overidentification test for the instrument set (Hansen J or Sargan, depending on the weighting matrix). A key requirement is failure to reject AR(2); otherwise, lagged levels may be invalid instruments. Roodman (2009) provides practical guidance on disciplined instrument design.

Because EU member states are exposed to common global shocks (e.g., the global financial crisis and pandemic-era disruptions), a full set of year effects is desirable; however, adding many time indicators in a small-N GMM setting can interact with instrument proliferation, so we treat time saturation as a priority robustness extension.

EU member states face common macroeconomic shocks, such as changes in global financial conditions, pandemic-related disruptions, and energy price shocks. In principle, year fixed effects appropriately control for such time-common factors. However, in dynamic panel GMM, including many time dummies expands the instrument set and increases instrument proliferation. Therefore, we keep the baseline specification parsimonious and treat time-effect saturation as a robustness check, applying conservative instrument restrictions when including time dummies.

3.5. Model Construction

Equation (1) shows the model with the single time-shifted (lagged) variable:

$$y_{it}=\beta y_{it-1}+u_i+{\epsilon }_{it},\hspace{1em}\left|\beta \right|<1$$

(1)

Symbol $y_{it}$ represents the dependent variable in the period $t$; $y_{it-1}$ is the dependent variable with one period lag from $t$; and $u_i$ represents time-invariant effects, while ${\epsilon }_{it}$ represents the error term.

Individual effects are taken as stochastic. Furthermore, necessary assumption for the invariance of the model are errors ${\epsilon }_{it}$ that are serially uncorrelated. Individual time-invariant effects are mostly related to past effects of the dependent variable in our model, which indicates an endogeneity issue.

The econometric model can be expressed as follows in Equation (2):

$$FDI{PC}_{it}=\alpha +{\beta }_{1}FDI{PC}_{it-1}+{\beta }_2{GDP}_{it}+{\beta }_{3}GDP{PC}_{it}+{\beta }_4{CPI}_{it}+{\beta }_{5}EXPORTS{PC}_{it}+{\beta }_{6}IMPORTS{PC}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(2)

With the coefficient variables ${\beta }_1$ to ${\beta }_6$ already explained in Table 1, we provide explanations of extra symbols: ${\mu }_i$ captures country specific effects; ${\lambda }_t$ denotes time effects; and ${\epsilon }_{it}$ is the idiosyncratic error term.

4. Results

This section presents the empirical results of the dynamic panel model examining the relationship between corruption and foreign direct investment (FDI) in the 27 European Union (EU) member states over the period 2002–2022. The analysis was conducted using the Generalized Method of Moments (GMM) framework, specifically the Arellano–Bond estimator, to address potential endogeneity and unobserved heterogeneity across countries. Following Wooldridge (2010), dynamic panel estimators such as the Arellano–Bond GMM are particularly suitable when the dependent variable exhibits persistence over time and when regressors may be correlated with past shocks. This approach helps mitigate simultaneity bias and omitted-variable problems commonly found in macro-panel settings. This approach allows for consistent estimation of dynamic relationships by using lagged values of endogenous variables as instruments.

4.1. Baseline Results

Table 3 presents the estimation results for the baseline model, where the dependent variable is foreign direct investment per capita (FDI p.c.), expressed in US dollars per capita. The explanatory variables are gross domestic product (GDP), GDP per capita (GDP p.c.), the Corruption Perceptions Index (CPI), and trade indicators (exports and imports per capita). All variables enter the baseline in first differences, as required by difference-GMM. Table 2 shows substantial persistence in levels for several series, while first differences are generally stationary under standard deterministic specifications.

Table 3. GMM estimation results for FDI per capita (dependent variable: FDIPC).

Variable	Coefficient		Std. Error		t-Statistic	Prob.
FDIPC(−1)	0.071995		0.003384		21.27484	0.0000
GDP	3.83 × 10−8		3.42 × 10−9		11.18457	0.0000
GDPPC	−2.251329		0.065371		−34.43910	0.0000
CPI	857.1569		34.35116		24.95278	0.0000
EXPPC	0.319296		0.027950		11.42396	0.0000
IMPPC	1.189130		0.008391		141.7114	0.0000
Mean dependent var		−85.80906		S.D.dependent var		30,003.78
S.E. of regression		33,195.89		Sum squared resid		5.97 × 1011
J-statistic		20.17061		Instrument rank		27
Prob(J-statistic)		0.510479

Method: Panel-Generalized Method of Moments. Transformation: First differences. Sample: 2002–2022. Periods included: 21. Cross-sections included: 27. Total panel (unbalanced) observations: 548. White period (period correlation) instrument weighting matrix. Instrument specification: @DYN(FDIPC,−2).

Because the model is estimated in first differences, the coefficients should be interpreted as associations between year-to-year changes (e.g., a one-point improvement in CPI and the corresponding change in FDI per capita), not as long-run level effects.

Table 3 presents the dynamic panel estimates obtained with the Arellano–Bond difference-GMM approach, where first differencing eliminates time-invariant country effects and lagged values provide internal instruments. The lag of FDI per capita is positive and highly significant, indicating meaningful persistence in FDI dynamics. The corruption indicator (CPI) enters with a positive and strongly significant coefficient, consistent with the view that improvements in perceived institutional integrity are associated with higher FDI per capita. Trade intensity, captured by exports and imports per capita, also shows positive and statistically robust associations with FDI. In line with Wooldridge (2010), diagnostic tests such as the Hansen J-statistic and the Arellano–Bond serial correlation tests were conducted to verify the validity of the overidentifying restrictions and the absence of serial correlation in the residuals. The Hansen J-statistic (p = 0.510) fails to reject the null hypothesis of instrument exogeneity, indicating that the chosen instruments are valid and not correlated with the error term. The decisive serial-correlation diagnostic for difference-GMM is the AR(2) test on the differenced residuals: failure to reject AR(2) supports the maintained assumption that the idiosyncratic error in levels is not serially correlated, which is required for lagged-level instruments to be valid. In our baseline, AR(2) is not rejected (p = 0.355), so we do not find evidence of second-order serial correlation. The AR(1) result is not used as an instrument-validity check because first differencing often induces negative first-order correlation mechanically. Thus, the GMM estimates can be considered consistent and unbiased.

The results of the econometric model can be expressed as follows in Equation (3):

$$FDI{PC}_{it}=0.07\times FDI{PC}_{it-1}+3.83\times {10}^{-8}\times {GDP}_{it}-2.25\times GDP{PC}_{it}+857.16\times {CPI}_{it}+0.32\times EXP{PC}_{it}+1.19\times IMP{PC}_{it}+{\epsilon }_{it}$$

(3)

4.2. Robustness Checks

Because net FDI flows can be extremely volatile and may be negative in years of disinvestment, the robustness analysis replaces FDI per capita in USD with an inverse hyperbolic sine (IHS/asinh) transform, which is defined for negative and zero values while remaining log-like for large magnitudes. Formally,

$$asinh\left(X_{i,t}\right) = ln(X_{i,t}+\sqrt{X_{i,t}^2+1})$$

(4)

Thus, on the example of FDIPC,

$$asinh\left({FDIPC}_{i,t}\right) = ln({FDIPC}_{i,t}+\sqrt{{FDIPC}_{i,t}^2+1})$$

(5)

This transformation is defined for all real numbers (including negatives) and behaves similarly to the logarithm when values are large, while remaining approximately linear around zero. The aim is not to change the economic question but to verify that the institutional effect (CPI) is not driven by the scale of FDI per capita or by a few extreme observations. As a robustness control for cross-border financial conditions, PI denotes portfolio investment per capita and is transformed analogously as follows:

$$asinh\left({PIPC}_{i,t}\right) = ln({PIPC}_{i,t}+\sqrt{{PIPC}_{i,t}^2+1})$$

(6)

The baseline specification is reported in original units; the asinh/IHS specifications are reported strictly as robustness checks to reduce sensitivity to extreme flows and to retain observations with zero or negative net FDI.

We retest for robustness purposes by adding the portfolio investment per capita (PI) variable and by exchanging the EXPPC and IMPPC variables with the trade openness variable (OPEN):

$$Trade openness (OPEN)= {\displaystyle \frac{{EXPPC}_{i,t}+{IMPPC}_{i,t}}{{GDPPC}_{i,t}}}$$

(7)

Table 4 and Table 5 use a balanced 2007–2022 panel and confirm a positive CPI effect.

Table 4. GMM estimation results for FDI per capita (dependent variable: FDIASINH).

Variable	Coefficient		Std. Error		t-Statistic	Prob.
FDIASINH(−1)	−0.031895		0.012598		−2.531697	0.0180
PIASINH	−0.131906		0.018570		−7.103129	0.0000
CPI	0.229916		0.036751		6.256010	0.0000
GDPPC	0.000229		7.48 × 10−5		3.068377	0.0051
OPEN	−8.232452		2.051649		−4.012602	0.0005
Mean dependent var		−0.059447		S.D.dependent var		6.828432
S.E. of regression		6.613759		Sum squared resid		17,977.88
J-statistic		22.74184		Instrument rank		26
Prob(J-statistic)		0.357896

Method: Panel Generalized Method of Moments. Transformation: First differences. Sample: 2007–2022. Periods included: 16. Cross-sections included: 26. Total panel (balanced) observations: 416. White period (cross-section cluster) standard errors and covariance (d.f. corrected). Instrument specification: @DYN(FDIASINH,−2) @DYN(CPI,−2).

Table 5. GMM estimation results for FDI per capita (dependent variable: FDIASINH).

Variable	Coefficient		Std. Error		t-Statistic	Prob.
FDIASINH(−1)	−0.077194		0.005285		−14.60741	0.0000
PIASINH	−0.154775		0.009762		−15.85465	0.0000
GDPASINH	3.304156		1.121887		2.945178	0.0069
CPI	0.245868		0.033367		7.368612	0.0000
Mean dependent var		−0.059447		S.D.dependent var		6.828432
S.E. of regression		6.550041		Sum squared resid		17,676.05
J-statistic		24.55995		Instrument rank		26
Prob(J-statistic)		0.318578

Method: Panel-Generalized Method of Moments. Transformation: First differences. Sample: 2007–2022. Periods included: 16. Cross-sections included: 26. Total panel (balanced) observations: 416. White period (cross-section cluster) standard errors and covariance (d.f. corrected). Instrument specification: @DYN(FDIASINH,−1) @DYN(PIASINH,−1).

Arellano–Bond tests indicate significant first-order serial correlation in the differenced residuals (AR(1) p = 0.0000), as expected under first differencing, while the absence of second-order correlation (AR(2) p = 0.7410) supports the validity of the lagged instruments.

The CPI coefficient from the asinh specification should not be compared directly to the CPI coefficient from the level specification because the dependent variable is measured on a different scale. A one-unit change in CPI changes FDIPC by an amount that depends on the level of FDIPC:

$$∆FDIPC\approx \beta CPI\times \sqrt{{FDIPC}^2+1}$$

(8)

in USD per capita. For large positive FDIPC, the effect becomes approximately proportional (log-like):

$$∆FDIPC\approx \beta CPI\times FDIPC$$

(9)

This explains why a “small” asinh coefficient can correspond to a “large” USD coefficient. Numerically, the two reported estimates align at typical high-flow magnitudes: 857.1569/0.229916 ≈ 3728.13 and 0.229916 × 3728.13 ≈ 857.1569. The coefficients therefore do not contradict each other; they reflect the same relationship expressed in levels versus a log-like transformation.

As an additional robustness check, we re-estimate the dynamic model using inverse hyperbolic sine transformations of the capital-flow variables. This keeps observations with zero or negative net flows and reduces the influence of extreme values, while preserving the main interpretation of changes in FDI over time. The model is again estimated with Arellano–Bond difference-GMM in first differences.

The robustness results are consistent with the baseline conclusions. CPI remains positive and highly significant, indicating that improvements in perceived integrity are associated with stronger FDI dynamics. Portfolio investment enters with a negative and significant coefficient, suggesting a substitution or compositional effect between different types of capital inflows. GDP remains positive and significant, in line with the market-size channel. The negative lagged dependent variable indicates mean reversion in annual FDI changes.

Diagnostic tests are consistent with a well-specified difference-GMM model: AR(1) is significant in the differenced residuals (p = 0.0005), as expected, whereas AR(2) is clearly insignificant (p = 0.9161), indicating no evidence of second-order serial correlation.

Table 6 presents the final robustness specification estimated by Arellano–Bond difference-GMM (2007–2022) with clustered standard errors, combining CPI, GDP per capita, trade openness, and portfolio investment as controls.

Table 6. GMM estimation results for FDI per capita (dependent variable: FDIASINH).

Variable	Coefficient		Std. Error		t-Statistic	Prob.
FDIASINH(−1)	−0.026707		0.007985		−3.344650	0.0009
PIPCASINH	−0.125366		0.015474		−8.101467	0.0000
CPI	0.288031		0.025500		11.29542	0.0000
GDPPC	0.000429		2.63 × 10−5		16.32636	0.0000
OPEN	−11.76018		0.817759		−14.38098	0.0000
Mean dependent var		−0.059447		S.D. dependent var		6.828432
S.E. of regression		6.679016		Sum squared resid		18,334.40
J-statistic		19.01255		Instrument rank		26
Prob(J-statistic)		0.584335

Method: Panel-Generalized Method of Moments. Transformation: First differences. Sample: 2007–2022. Periods included: 16. Cross-sections included: 26. Total panel (balanced) observations: 416. White period (cross-section cluster) standard errors and covariance (d.f. corrected). Instrument specification: @DYN(FDIASINH,−2).

CPI remains positive and highly significant, confirming a robust institutional effect on FDI. GDP per capita is positive, while trade openness and portfolio investment are negative and significant, consistent with compositional or substitution patterns across cross-border activity. The negative lag term indicates mean reversion in annual FDI changes. Diagnostics support the robustness specification: AR(1) is significant as expected (p = 0.0003), AR(2) is not significant (p = 0.6995), and instrument validity is not rejected by the Hansen J-test. Additional robustness checks using alternative scaling (e.g., FDI as a share of GDP) and time-effect saturation are natural extensions but are beyond the current revision’s scope.

5. Discussion

The baseline difference-GMM estimates reveal a clear pattern consistent with this paper’s hypotheses. The coefficient on the lagged dependent variable is positive and highly significant, indicating persistence in FDI dynamics: changes in inflows show partial adjustment rather than one-period mean reversion. Because the model is estimated in first differences, coefficients are best interpreted as associations between year-to-year changes in the regressors and the corresponding change in inward FDI inflows per capita, conditional on the dynamic term and included controls.

Governance quality, proxied by the CPI, has a positive and highly significant coefficient, supporting H1: improvements in perceived integrity (higher CPI values) are associated with increases in inward FDI inflows per capita. This supports the standard “deterrence” mechanism, where corruption raises effective transaction costs, amplifies regulatory uncertainty, and increases perceived risk of contract non-enforcement or discretionary expropriation, discouraging long-term investment. However, the sign of the corruption–FDI relationship is not settled a priori in the broader literature. A competing “grease the wheels” argument suggests that in heavily regulated environments, informal payments can reduce delay costs and thereby increase investment incentives. The EU setting is informative because it includes both high-quality institutional regimes and heterogeneous administrative burdens across member states. The findings align more strongly with the deterrence view than with the grease view, suggesting that within this sample and specification, the predictability and credibility of rules outweigh any short-run facilitation effects.

Market size (GDP) is positive and statistically significant, consistent with H2 and market-seeking motives: larger markets provide scale, demand depth, and scope for local production. The GDP coefficient is numerically small, which is unsurprising because GDP is measured at the aggregate level while the dependent variable is per capita flows and the estimation is in differences. Thus, the contribution of market size is better understood as directional support for the location-choice channel rather than as a literal marginal-dollar effect.

GDP per capita is the empirically sensitive margin in the EU context (H3). In advanced samples, GDP per capita proxies both purchasing power and higher cost levels. The baseline negative coefficient is compatible with a cost-saturation interpretation: higher-income member states may attract skill-intensive or headquarters-type projects yet deter cost-sensitive investment or shift marginal projects toward lower-cost EU locations. Importantly, the robustness exercises indicate that the GDP per capita association is sensitive to sample composition and scaling, cautioning against a single-mechanism interpretation. In this framework, GDP per capita should be treated primarily as a development-level control capturing multiple channels rather than as a single structural driver.

Trade integration variables (exports and imports per capita) are positive and significant, supporting H4 and reinforcing the relevance of openness and cross-border integration for FDI location decisions. The stronger association for imports per capita than for exports per capita is economically plausible in the EU environment: inward FDI is often embedded in production networks with intensive cross-border sourcing of intermediate inputs, while exports may reflect either market access or substitution between exporting and local production. This asymmetry highlights the value-chain dimension of EU integration rather than contradicting conventional trade–FDI complementarities.

The robustness checks using the inverse hyperbolic sine transformation strengthen the institutional conclusion. The CPI effect remains positive and statistically strong when the dependent variable is transformed to reduce the influence of extreme observations and to accommodate zero or negative net flow values without dropping data. This reduces the likelihood that the institutional association is a mechanical by-product of heavy tails or a particular scaling choice.

Interpreting these results requires consideration of the design’s limitations. The CPI, produced by Transparency International, is based on perceptions and aggregates multiple institutional dimensions into a single index; it cannot identify which specific enforcement, procurement, judicial, or regulatory mechanisms drive the observed association. More generally, GMM inference relies on maintained moment conditions and careful instrument selection; therefore, the coefficients are best interpreted as robust conditional associations rather than fully structural causal effects. Future research could improve mechanism identification by combining sectoral FDI data with more granular governance indicators and by applying designs that exploit quasi-exogenous institutional shocks or policy reforms within the EU.

6. Conclusions

Using dynamic panel (difference-GMM) estimates for EU member states over 2002–2022, we find that better perceived integrity (a higher CPI score, i.e., lower perceived corruption) is robustly associated with higher inward FDI inflows per capita. Market size (GDP) enters positively, while the income effect is specification-sensitive: GDP per capita is negative in the baseline trade per capita specification but turns positive once the model is re-estimated on a balanced 2007–2022 sample using asinh-transformed flows and an alternative set of controls (portfolio investment and trade openness). Trade variables are positively related to FDI when exports and imports are entered separately, whereas openness carries a negative coefficient in the extended specification, which is consistent with partial import-substitution motives for FDI once overall tradability is controlled for.

The main empirical message is therefore narrow but policy-relevant: within the EU, improvements in governance quality captured by the CPI are consistently linked with stronger FDI performance, even after accounting for dynamics, endogeneity, and standard post-estimation diagnostics (instrument validity and lack of second-order serial correlation). At the same time, CPI is a perception-based aggregate indicator and cannot by itself identify the underlying institutional mechanisms; likewise, GMM results depend on the maintained moment conditions and disciplined instrument design. Future research should triangulate corruption with rule-of-law and property-rights enforcement indicators, test nonlinearities and distributional heterogeneity (e.g., threshold or quantile effects), and strengthen identification with external sources of institutional variation or alternative estimators (e.g., System-GMM and bias-corrected LSDV) where appropriate.

In conclusion, this study reaffirms that corruption remains one of the most significant barriers to investment activity and sustainable economic development in the EU. Reducing corruption through transparent governance and strong institutions not only enhances foreign investment inflows but also promotes broader economic resilience and social trust: key foundations for long-term prosperity.