Poverty Dynamics Under Changing Measurement Frameworks: The Role of Foreign Direct Investment in Vietnam

Abstract

Vietnam’s sustained poverty reduction has coincided with rising foreign direct investment (FDI) and a major shift from income-based to multidimensional poverty measurement, raising challenges for interpreting poverty dynamics and the role of FDI across regimes. This study examines the relationship between FDI and poverty reduction in Vietnam by accounting for poverty persistence, regional heterogeneity, and changes in poverty measurement. Using provincial panel data for 2002–2022 and a System GMM framework, three main findings emerge. First, poverty dynamics differ across measurement regimes: during the income-poverty period (2002–2016), poverty dynamics exhibited lower persistence and faster convergence, whereas under the multidimensional framework (2016–2022), poverty became more persistent and convergence slowed, reflecting the increasingly structural nature of remaining deprivation. Second, FDI is negatively associated with poverty under both measures, but its effects are conditional and uneven. Interaction effects indicate that the poverty-reducing impact of FDI depends on provincial income levels and initial deprivation, with weaker effects in provinces facing deeper multidimensional poverty. Third, higher FDI exposure is associated with greater poverty persistence, reflecting the spatial concentration of FDI in better-off regions rather than a poverty-increasing effect. The analysis is subject to limitations related to measurement regimes, and results are interpreted as conditional associations. Policy implications highlight that the poverty-reducing effects of FDI depend critically on investment quality, the strength of local production linkages, and complementary public spending, particularly in provinces facing persistent deprivation.

1. Introduction

Foreign direct investment (FDI) has long played an important role in the development trajectories of transitional economies such as Vietnam. In these economies, domestic savings are often insufficient to support the scale of investment required for sustained growth and economic modernization. As noted by Todaro and Smith (2020) and Hayami (2001), FDI helps narrow this gap by supplying long-term external capital that domestic financial systems are often unable to provide. By providing employment opportunities, supporting structural transformation, and easing financial constraints, FDI-driven growth creates channels through which poverty reduction may occur. This causal chain is particularly relevant for economies moving from central planning toward market-based systems. As a result, both theoretical arguments and empirical evidence continue to view the attraction and effective use of FDI as an important component of inclusive development strategies.

Over the past three decades or so, Vietnam has achieved substantial progress in reducing poverty. As measured against the World Bank’s poverty line, the income-based poverty rate declined from 58 percent in 1993 to 37.4 percent by 1998 (IMF—International Monetary Fund, 2001). This downward trend persisted into the early 2000s, with poverty falling to 28.9 percent in 2002 (GSO VHLSS, 2002–2010) and declining further to 5.8 percent by 2016 (GSO—General Statistics Office of Vietnam, 2002–2022). That same year marked an important shift in how poverty was measured, as Vietnam moved from a purely income-based standard to a multidimensional framework. Not surprisingly, the multidimensional poverty rate exceeded the monetary rate, as it captures not only income-based deprivation but also deprivations in access to basic social services: health, education, housing, clean water and sanitation and information accessibility. In 2016, multidimensional poverty was recorded at 9.2 percent and fell steadily to 4.2 percent by 2022 (GSO—General Statistics Office of Vietnam, 2002–2022). This sustained reduction in poverty should not be attributed to rapid economic growth alone. It also reflects the contribution of targeted rural development policies, the expansion of non-farm employment, and Vietnam’s deeper integration into global value chains. Viewed as a whole, the country’s early reform period represents a notable case in which poverty reduction proceeded in tandem with broadly shared growth among transitional economies.

Vietnam’s sustained reduction in poverty has coincided with a substantial increase in foreign direct investment. Since the early 1990s, FDI inflows have averaged approximately 6 percent of GDP, positioning Vietnam among the major FDI recipients in Southeast Asia (T. P. Nguyen, 2012; GSO—General Statistics Office of Vietnam, 2002–2022). These inflows have contributed to the expansion of the manufacturing sector, supported rapid export growth, and enabled technology transfer, particularly in provinces that have attracted multinational enterprises. The parallel evolution of rising FDI and declining poverty gives rise to important empirical and policy considerations. A central question is how, and to what extent, FDI has contributed to poverty reduction in Vietnam, whether through direct effects or indirect channels. Closely related is the question of transmission mechanisms, including employment generation, productivity spillovers, export-oriented industrialization, and increases in public revenue.

In the broader international literature, empirical research has examined how foreign direct investment affects poverty through different channels. Much of the empirical literature links FDI to faster economic growth, technological upgrading, and job creation, all of which shape living standards and poverty outcomes. However, relatively few studies provide direct, context-specific estimates of the FDI–poverty relationship. In Vietnam, empirical research is limited by the lack and inconsistency of provincial-level data, as well as by the lack of analyses built within a rigorous and systematic methodological framework. Most existing studies, based on data from the 1990s and early 2000s, only provide preliminary evidence of the role of FDI in supporting poverty reduction, while failing to fully clarify the dynamic mechanisms of impact, inter-local differences, and changes in poverty measures over time.

This study addresses several limitations in the existing literature by offering new empirical evidence on the relationship between foreign direct investment (FDI) and poverty reduction in Vietnam. First, the analysis employs two distinct poverty measures—income poverty for the period 2002–2016 and multidimensional poverty for 2016–2022—to examine whether the effects of FDI differ across monetary and non-monetary dimensions of deprivation. This dual-measure approach helps clarify how changes in Vietnam’s poverty measurement framework interact with the development impacts associated with FDI.

Second, the study investigates beta convergence in provincial poverty through a dynamic specification that incorporates lagged poverty. This framework allows for an assessment of whether provinces with higher initial poverty levels tend to experience faster subsequent poverty reduction in a transitional economy characterized by pronounced regional disparities. The results indicate partial beta convergence, suggesting that while poverty remains persistent, provinces continue to adjust toward a long-run equilibrium. In this context, FDI may affect not only average poverty levels but also the dynamics of convergence across regions.

Third, the analysis examines interaction effects between FDI and structural conditions in order to capture heterogeneity in the poverty-reducing impact of FDI. Interaction terms between FDI and per capita income are used to assess whether local absorptive capacity and development levels condition the effectiveness of FDI in reducing poverty. Furthermore, interactions between FDI and lagged poverty are introduced to evaluate whether FDI influences the speed of poverty convergence across provinces, thereby indicating whether its benefits are concentrated in poorer or more developed regions.

Fourth, the study constructs a harmonized poverty series by proportionally linking income-based and multidimensional poverty indicators in the transition year of 2016. This procedure yields a unified provincial panel covering the period 2002–2022 and provides a basis for testing the robustness of FDI effects under alternative poverty measurement regimes.

Finally, the empirical analysis applies the system GMM estimator to address endogeneity, the dynamic persistence of poverty, and unobserved provincial heterogeneity. This estimation strategy is well suited to capturing the joint dynamics between FDI, poverty, and convergence. Taken together, the empirical framework enables a comprehensive and credible assessment of the role of FDI in poverty reduction in Vietnam. The findings offer relevant implications for Vietnam’s development strategy and provide useful insights for other developing and transitional economies seeking to enhance the inclusive development impact of FDI.

The remainder of the paper is structured as follows. Section 2 reviews the related literature, beginning with the theoretical framework linking foreign direct investment to growth and poverty reduction, followed by a discussion of the relevant empirical evidence and the specific motivations for the present study. Section 3 outlines the empirical methodology and describes the data and sources used in the analysis. Section 4 presents the estimation results and provides a detailed discussion of the main findings. Section 5 concludes by summarizing the key results and drawing out their policy implications.

2. Literature Review

2.1. Theoretical Framework

Foreign direct investment (FDI) is widely regarded as an important driver of development in emerging and low-income economies, and a substantial theoretical literature links FDI to poverty reduction. Existing studies commonly distinguish two broad channels through which FDI may affect poverty: direct employment effects and indirect effects operating through economic growth. The strength and effectiveness of these channels depend on underlying mechanisms as well as on sectoral composition and institutional conditions. With respect to direct effects, FDI can contribute to poverty reduction through job creation. When multinational enterprises (MNEs) invest in labor-intensive manufacturing or service activities, they expand employment opportunities for local workers and frequently offer wages that exceed those paid by domestic firms. These wage differentials are often associated with higher productivity, greater capital intensity, and adherence to international labor standards (IFC—International Finance Corporation, 2000; Lipsey & Sjöholm, 2004; OECD—Organisation for Economic Co-Operation and Development, 2002). Beyond direct employment, FDI also generates indirect employment through backward and forward linkages, as foreign firms increase demand for local suppliers, distributors, and related support services (UNCTAD—United Nations Conference on Trade and Development, 1994; Javorcik, 2004). Essentially, these processes can raise household incomes and improve living standards, particularly in regions that attract export-oriented and labor-absorbing investment (Chudnovsky & López, 1999; Klein et al., 2001).

With respect to indirect effects, FDI may influence poverty reduction through its contribution to long-run economic growth. At the macroeconomic level, FDI helps bridge the gap between limited domestic savings and the level of investment required for sustained development, a central proposition of the dual-gap framework (Chenery & Strout, 1966). In line with endogenous growth theory, which highlights the roles of capital accumulation, knowledge formation, and productivity gains (Romer, 1990; Barro & Sala-i-Martin, 2004; Aghion & Howitt, 1998), FDI expands the capital base and improves the efficiency of investment. At the microeconomic level, multinational enterprises (MNEs) act as channels for technology transfer, managerial know-how, and organizational upgrading. Through mechanisms such as demonstration effects, labor mobility, supplier–buyer linkages, and competitive pressure, foreign firms generate spillovers that extend beyond the foreign sector (Blomström & Kokko, 1998; Borensztein et al., 1998; Javorcik, 2004; Görg & Greenaway, 2004). These firm-level spillovers enhance productivity, strengthen human capital, and raise the absorptive capacity of domestic enterprises, thereby reinforcing long-run growth dynamics.

FDI may also give rise to crowding-in effects, whereby the presence of foreign firms induces domestic enterprises to expand production, adopt new technologies, and upgrade capabilities, particularly in sectors characterized by strong foreign–domestic linkages (Agosin & Mayer, 2000). These effects are frequently reinforced by complementary public investment, as governments upgrade infrastructure in order to attract or retain foreign investors, including in rural and less-developed regions (T. P. H. Nguyen, 2003; UNCTAD—United Nations Conference on Trade and Development, 2005). In addition, FDI contributes to an expanded tax base through corporate taxation and export-related revenues, thereby increasing the fiscal resources available for pro-poor expenditures on health, education, infrastructure, and social protection. Through these combined channels, FDI-driven growth can support poverty reduction by expanding employment opportunities, raising wages, enhancing productivity, and enabling inclusive public investment, in line with theories of structural transformation, pro-poor growth, and fiscal redistribution (Lewis, 1954; Kuznets, 1973; Ravallion & Chen, 2003; Klasen, 2010).

Despite the mentioned potential benefits, the poverty-reducing effects of foreign direct investment are not automatic. Insights from labor-market theory, industrial organization, and the economics of technological change point to conditions under which FDI may yield limited—or even adverse—distributional outcomes. When foreign investment is concentrated in capital-intensive, extractive, or high-technology activities, employment generation tends to be modest and skewed toward skilled labor, which poorer households often cannot supply (Feenstra & Hanson, 1997; Blomström & Kokko, 1998). In economies characterized by abundant low-skilled labor but weak human-capital formation, such sectoral patterns are unlikely to translate the contributions of FDI into substantial gains in employment or income for the poor. Technological upgrading associated with multinational enterprises may further intensify skill-biased labor demand, displacing low-skilled workers or widening wage differentials. At the same time, competitive pressure from foreign firms can prompt domestic producers to automate production or rationalize operations, with potential reductions in low-wage employment (Aitken et al., 1996; Rodrik, 1997). These considerations highlight that the poverty impact of FDI depends critically on sectoral composition, local absorptive capacity, labor-market structure, and the degree of alignment between foreign investment and the host economy’s human-capital endowment (Jenkins & Thomas, 2002). Thus, if FDI is concentrated in enclave or capital-intensive sectors, local absorptive capacity is limited, labor markets are segmented, and foreign investment is poorly matched with domestic skill endowments, spillover and crowding-in effects remain constrained, thereby limiting the potential of FDI to generate broad-based welfare gains.

2.2. Empirical Studies

Early empirical research focused primarily on the growth effects of FDI, giving comparatively less attention to its direct implications for poverty outcomes. This emphasis reflected the prevailing view that economic growth constitutes a central mechanism for poverty reduction (Ravallion & Chen, 2003; Dollar et al., 2016). Accordingly, a large body of studies has examined the poverty implications of FDI indirectly by evaluating whether FDI stimulates growth through channels such as capital accumulation, productivity gains, and technology diffusion. To study these relationships, researchers have employed a broad set of econometric approaches, including Granger-causality tests, multivariate regression frameworks, vector autoregressive (VAR) and vector error-correction (VEC) models, dynamic panel estimators, and, more recently, system GMM and panel cointegration techniques (Carkovic & Levine, 2005; Hansen & Rand, 2006; Alfaro et al., 2010). While early Granger-causality analyses were useful in identifying short-run associations between FDI and growth, they are viewed as not explicitly considering the long-run dynamics that are critical for understanding development processes.

Empirical evidence on the relationship between FDI and economic growth remains mixed, reflecting cross-country differences in institutional quality, financial development, absorptive capacity, and the sectoral composition of foreign investment. A number of studies—including Hsiao and Hsiao (2006), Basu and Guariglia (2007), and Apergis et al. (2008)—find that FDI promotes growth by expanding capital formation and generating technology spillovers. In a related vein, Granger-causality analyses by Choe (2003) and Hansen and Rand (2006) identify bidirectional causality, indicating that FDI both contributes to economic growth and responds to it. Chowdhury and Mavrotas (2006) also find two-way causality between FDI and growth for Malaysia and Thailand using the Toda-Yamamoto test. Other studies, however, report weaker or more limited effects. Fry (1993) documents positive growth impacts only for Pacific Basin economies, while Duttaray (2001) finds that FDI enhances growth in fewer than half of the developing countries in his sample. Evidence at the sectoral level further suggests that FDI may stimulate growth in some industries while constraining it in others, depending on the structure of inter-industry linkages and the intensity of competitive pressures (Basu & Guariglia, 2007).

Several influential studies question whether FDI exerts a robust and independent effect on economic growth. Employing panel-data frameworks with stronger controls for endogeneity, Carkovic and Levine (2005) find no systematic association between FDI and long-run growth for 72 countries over the years 1960–1995. Mencinger (2003) reports that, in a number of Central and Eastern European transition economies, FDI inflows coincided with slower growth during the period 1994–2001, raising concerns related to crowding-out effects, enclave-type investment, and profit repatriation. More recent research emphasizes that the growth impact of FDI is highly conditional, materializing primarily in economies characterized by well-developed financial systems, sound institutions, macroeconomic stability, and sufficient human capital (Alfaro et al., 2010; Herzer, 2012; Iamsiraroj & Ulubaşoğlu, 2015). The analysis by Bruno and Campos (2013) further supports this perspective, documenting substantial heterogeneity in estimated FDI effects across regions, empirical methods, and time periods. Overall, the empirical evidence suggests that while FDI can promote growth, its benefits are neither automatic nor evenly distributed.

Relative to the extensive literature on foreign direct investment and economic growth, a growing number of studies explicitly examine the relationship between FDI and poverty. However, much of this literature assesses the poverty effects of FDI indirectly, typically by evaluating whether FDI promotes income growth, employment, or welfare-related outcomes rather than by modeling poverty measures directly. Early evidence provided by Klein et al. (2001) indicates that FDI contributed to poverty reduction indirectly through employment creation, multiplier effects, and increased female labor-force participation. Focusing on African economies over the period 1990–2007, Gohou and Soumaré (2012) report bidirectional causality between FDI and welfare indicators, including per capita income and the human development index, while also documenting substantial regional heterogeneity in these effects. Ucal (2014) reports a significant negative association between FDI and poverty across 26 developing countries, suggesting that higher FDI inflows are linked to lower levels of income poverty. More recent studies emphasize the conditional nature of these relationships, showing that FDI improves welfare outcomes primarily in countries with strong institutional frameworks, adequate absorptive capacity, and well-functioning financial systems (Reiter & Steensma, 2010; Gohou & Soumaré, 2012; C. P. Nguyen et al., 2021). In a comprehensive survey of the literature, Magombeyi and Odhiambo (2017) likewise conclude that FDI generally contributes to poverty reduction, although its impact depends critically on sectoral composition, institutional quality, and the choice of poverty measure.

In the case of Vietnam, empirical evidence on the FDI–poverty nexus remains relatively limited, whether assessed directly or indirectly, although several studies offer important insights. Using provincial panel data for the period 1990–2000, T. P. H. Nguyen (2003) finds that FDI contributed positively to economic growth but did not lead to a direct reduction in poverty, with its effects operating primarily through the growth channel. By contrast, Tran (2006), analyzing data from 12 provinces over 1992–2002, reports that FDI reduced poverty through both direct and indirect effects. More recent contributions point to similarly conditional and heterogeneous outcomes. Anwar and Nguyen (2011) show that FDI supported poverty reduction by increasing labor demand and facilitating industrial upgrading. Bui et al. (2019) find that FDI improved household welfare through the expansion of non-farm employment and enhanced human-capital accumulation. Do et al. (2021) employ a spatial econometric framework using provincial panel data for 2010–2016 and demonstrate that the poverty-reducing effects of FDI are not confined to individual provinces but also generate significant spillovers across regions. Notably, while the results by Do et al. (2021) indicate that FDI lowers poverty directly and through improvements in human capital, they also identify an indirect channel through international trade that may exacerbate poverty, highlighting a complex trade-off between deeper global integration and local welfare outcomes. Overall, the Vietnam-specific literature indicates that the poverty impacts of FDI are highly context-dependent and heterogeneous, shaped critically by provincial structural conditions and absorptive capacity.

2.3. Motivation

The existing empirical literature points to several important but unresolved limitations that motivate the present study. Although many contributions report that FDI supports economic growth and may be associated with poverty reduction, the available evidence remains fragmented, context-dependent, and frequently indirect. Most empirical analyses rely on single-dimensional poverty indicators, typically income-based measures, and thus do not examine whether FDI affects non-monetary forms of deprivation that have become central to Vietnam’s social policy framework since 2016. Moreover, relatively few studies investigate poverty convergence across regions or assess whether FDI influences the speed at which poorer provinces catch up, an issue of particular relevance in a transitional economy marked by substantial regional disparities. The literature also offers limited insight into how differences in provincial absorptive capacity—including income levels, institutional quality, human capital, and economic structure—condition the poverty-reducing effects of FDI. Furthermore, in the Vietnamese context, much of the empirical evidence is based on pre-2010 data or restricted provincial samples, despite significant economic restructuring, changes in the composition of FDI inflows, and the transition toward multidimensional poverty measurement. Methodological limitations, such as static model specifications, insufficient treatment of endogeneity, and the limited use of interaction terms, further restrict earlier studies’ ability to capture the joint dynamics of FDI, poverty persistence, and regional heterogeneity.

Building on these limitations, the present study offers several contributions. It employs two poverty measures—income poverty for the period 2002–2016 and multidimensional poverty for 2016–2022—to assess whether the effects of FDI differ between the monetary measure and the multidimensional measure of deprivation. To achieve the objective of testing whether the poverty impact of FDI varies with local absorptive capacity or initial disadvantage, the specification of the empirical model incorporates interaction terms, including the interaction between FDI and per capita income and the interaction between FDI and lagged poverty. While the inclusion of the lagged poverty in the empirical specification helps focus on the analysis of beta convergence of poverty in a dynamic framework, the inclusion of the interaction between FDI and lagged poverty allows an examination of whether FDI influences the speed of this convergence. Furthermore, to assess the stability and robustness of the estimated FDI effects and other control variables across poverty regimes, the study constructs a harmonized poverty series covering the full period from 2002 to 2022 solely as a robustness exercise, in order to assess whether the estimated relationships are sensitive to alternative poverty definitions rather than to establish strict intertemporal comparability across measurement regimes. Finally, estimation is conducted using the system GMM approach in order to address endogeneity, dynamic persistence, and unobserved provincial heterogeneity. Together, these components provide a more detailed and policy-relevant assessment of the role of FDI in poverty reduction in Vietnam, with implications that extend to other transitional economies.

3. Methodology and Data

3.1. Empirical Methodology

The empirical strategy is designed to capture the dynamic relationship between foreign direct investment and poverty across Vietnamese provinces. Vietnam provides an analytically relevant case due to the coexistence of rapid FDI inflows, pronounced regional disparities, and a formal shift from income-based to multidimensional poverty measurement. These features motivate the use of a dynamic panel approach that explicitly accounts for poverty persistence and heterogeneity across regions.

The empirical framework adopted in this study is grounded in modern growth theory, which emphasizes capital accumulation, technological progress, and labor supply as key determinants of long-run economic growth (Romer, 1990; Aghion & Howitt, 1998; Barro & Sala-i-Martin, 2004; Jones, 2019). By raising incomes and expanding employment opportunities, sustained growth creates channels through which factors such as investment, human capital, and foreign direct investment can influence poverty outcomes, both directly and indirectly (Dollar et al., 2016; Magombeyi & Odhiambo, 2017). Within this framework, the analysis employs a dynamic panel specification in which provincial poverty is modeled as a function of its own lagged value and a set of variables capturing economic performance, as summarized in Equation (1).

$$p_{i,t}=\alpha +{\beta }_1{p}_{i,t-1}+{\beta }_{2}in{c}_{i,t}+ {\beta }_{3}fd{i}_{i,t}+{\beta }_{4}ca{p}_{i,t}+{\beta }_{5}la{b}_{i,t}+ {\beta }_{6}go{v}_t+{\epsilon }_{i,t}$$

(1)

In Equation (1), the dependent variable $p_{i,t}$ denotes the provincial poverty ratio. The inclusion of the lagged poverty term $p_{i,t-1}$ captures state dependence in poverty dynamics, consistent with recent convergence research showing that poverty tends to exhibit strong inertia and that regional poverty rates commonly follow a process of beta-convergence (Hulme & Shepherd, 2003). The remaining regressors represent widely recognized determinants of growth and poverty reduction. Provincial real income per capita, $in{c}_{i,t}$, serves as a proxy for overall economic performance. The ratio of provincial production capital to income, $ca{p}_{i,t}$, reflects capital deepening. The ratio of total registered foreign direct investment to income, $fd{i}_{i,t}$, captures external capital inflows and associated spillover effects. National government expenditure as a share of national GDP, $go{v}_{i,t}$, represents the stance of fiscal policy. The labor–population ratio, $la{b}_{i,t}$, is included as an indicator of local labor supply. Finally, ${\epsilon }_{it}$ denotes the composite error term, which can be expressed as ${\mu }_i+{\vartheta }_{it}$, where ${\mu }_i$ represents unobserved time-invariant provincial effects and ${\vartheta }_{it}$ is an idiosyncratic disturbance varying over time. Thus, this set of variables is consistent with the FDI–growth and growth–poverty studies, which emphasize the role of capital accumulation, public spending, labor market conditions, and foreign investment in shaping welfare outcomes (Alfaro et al., 2004; Iamsiraroj & Ulubaşoğlu, 2015; Nkoro & Uko, 2022).

To observe the heterogeneity in the relationship between poverty and FDI across provinces, the model incorporates two interaction terms that reflect observed patterns of FDI allocation in Vietnam. In practice, FDI tends to be concentrated in provinces with better infrastructure, more developed urban economies, and a skilled labor force, which does not necessarily originate from poorer households. The first interaction term, $fd{i}_{i,t}\times in{c}_{i,t}$, is employed to test whether the poverty effect of FDI depends on provincial absorptive capacity, measured by income per capita. This consideration is particularly relevant in the Vietnamese context, where higher-income provinces are typically more successful in attracting FDI, yet the resulting gains may accrue primarily to firms and skilled workers rather than being transmitted to poorer households. The second interaction term, $fd{i}_{i,t}\times p_{i,t-1}$, is used to examine whether the marginal effect of FDI depends on initial poverty levels, capturing the possibility that the impact of FDI on poverty is different in provinces that start from higher levels of deprivation, where employment opportunities and access to basic services are more constrained. The interaction terms are specified linearly to capture average conditional relationships and should not be interpreted as ruling out nonlinear or threshold effects related to absorptive capacity.

The empirical analysis proceeds sequentially: the baseline specification is first augmented with the income-based interaction to capture heterogeneity related to absorptive capacity, after which the interaction with lagged poverty is introduced to assess heterogeneity associated with initial conditions. The resulting specification is presented as follows:

$$p_{i,t}=\alpha +{\beta }_1{p}_{i,t-1}+{\beta }_{2}in{c}_{i,t}+ {\beta }_{3}fd{i}_{i,t}+{\beta }_{4}ca{p}_{i,t}+{\beta }_{5}la{b}_{i,t}+ {\beta }_{6}go{v}_t+{\gamma }_{1}fd{i}_{i,t}\times in{c}_{i,t}+{\gamma }_{2}fd{i}_{i,t}\times p_{i,t-1}+{\epsilon }_{i,t}$$

(2)

Equation (2) helps assess how provincial economic capacity and baseline deprivation jointly shape the extent to which FDI translates into poverty reduction. First, within the framework of the dynamic panel specification, poverty convergence is evaluated by whether the coefficient on lagged poverty is less than unity, which would indicate that provinces with higher initial poverty tend to experience faster subsequent poverty reduction. This interpretation follows the standard convergence approach in the literature on regional growth and welfare dynamics (Barro & Sala-i-Martin, 2004). However, in the setting with the existence of the interaction term $fd{i}_{i,t}\times p_{i,t-1}$, the marginal effect of initial poverty on current poverty is modified by Equation (3):

$${\displaystyle \frac{\partial p_{i,t}}{\partial p_{i,t-1}}}={\beta }_1+{\gamma }_{2}fd{i}_{i,t}$$

(3)

Beta-convergence exists when this term lies between zero and one, implying that, conditional on FDI, provinces with higher initial poverty tend to reduce poverty faster over time (Barro & Sala-i-Martin, 2004). Subtracting $p_{i,t-1}$ from both sides of Equation (2) yields an expression that describes how changes in the poverty rate depend on initial poverty and a set of control variables, including the interaction terms. This transformation allows the model to be expressed in growth form. Accordingly, the conditional beta-convergence parameter can be derived using the following formula:

$$Conditional beta convergence={\beta }_1+{\gamma }_{2}fd{i}_{i,t}-1.$$

(4)

Equation (4) indicates that beta convergence of poverty depends on the level of FDI inflows. In this specification, ${\gamma }_2$ indicates how FDI modifies the extent to which past poverty affects current poverty across provinces. A negative value of ${\gamma }_2$ indicates that higher FDI reduces poverty persistence and slows beta convergence, whereas a positive value implies that FDI increases persistence and accelerates beta convergence. This specification allows FDI to enter the model in two roles: as a factor directly affecting current poverty and as a variable that influences the speed at which provincial poverty levels converge over time.

Second, the marginal effect of FDI on poverty does not only depend on ${\beta }_3$, but is also conditioned on the coefficients of the interaction terms, ${\gamma }_1$ and ${\gamma }_2$, as per Equation (5):

$$Marginal effect of FDI={\beta }_3+{\gamma }_{1}inc+{\gamma }_2{p}_{t-1}$$

(5)

Based on Equation (5), we can observe the extent to which the marginal effect of FDI on poverty differs across different income levels, given a certain level of initial poverty or across different poverty levels given a certain level of income.

Provincial poverty is measured using two indicators that reflect the evolution of Vietnam’s poverty assessment framework: an income-based poverty rate for the period 2002–2016 and a multidimensional poverty index for 2016–2022. The income-based measure identifies households as poor when per capita income falls below nationally defined thresholds, thus capturing monetary deprivation. By contrast, the multidimensional poverty index records deprivations across multiple dimensions, including education, health, living conditions and access to basic services, consistent with Vietnam’s post-2016 approach to welfare assessment (UNDP—United Nations Development Programme, 2022). The transition in 2016, therefore, represents a structural break in poverty measurement rather than a continuous scale. Accordingly, the empirical analysis is conducted separately for each poverty regime, and estimated coefficients are interpreted within-regime rather than as part of a single continuous poverty trajectory.

In addition, a harmonized poverty series is constructed solely for robustness purposes. This series rescales the multidimensional poverty index to an income-poverty–consistent metric using the rule-of-three method, with 2016 serving as the reference year in which both income-based and multidimensional poverty measures are available. The main estimations rely exclusively on the original poverty indicators to preserve their distinct conceptual interpretations, while the harmonized series provides a supplementary sensitivity check on the stability of the results over the period 2002–2022.

Because poverty follows a dynamic process and key covariates—including income, capital accumulation, and foreign direct investment (FDI)—are potentially endogenous, this study employs the two-step System Generalized Method of Moments (System GMM) estimator. This approach is particularly appropriate for dynamic panel data, as it addresses endogeneity and unobserved province-specific heterogeneity—issues that may arise when poverty, income, and FDI evolve jointly over time (Arellano & Bover, 1995; Blundell & Bond, 1998; Roodman, 2009). In the absence of credible external instruments or quasi-experimental variation at the provincial level, the empirical strategy adopts a triangulation approach, and the estimated coefficients are interpreted as conditional associations rather than strictly causal effects. To further assess potential reverse causality from a predictive (Granger) perspective, we complement the dynamic System GMM framework with a heterogeneous panel Granger non-causality test (Dumitrescu & Hurlin, 2012), reported in Section 4.1.

Modeling poverty convergence requires the inclusion of lagged poverty, which generates correlation between the regressors and unobserved provincial effects, thereby rendering ordinary least squares and fixed-effects estimators biased and inconsistent. System GMM overcomes this problem by exploiting internal instruments: lagged levels are used to instrument the first-differenced equation, while lagged differences serve as instruments for the level equation (Arellano & Bover, 1995). This combined system enhances efficiency relative to Difference GMM and mitigates weak-instrument concerns, particularly in empirical settings where poverty, income, and FDI exhibit gradual adjustment (Blundell & Bond, 1998; Roodman, 2009). The two-step implementation further improves efficiency by applying a heteroskedasticity-robust weighting matrix that fully exploits the available moment conditions (Arellano & Bover, 1995; Windmeijer, 2005). A growing body of empirical research in development economics and the FDI–poverty literature documents the suitability of System GMM for capturing dynamic interactions and simultaneity among key development variables (Carkovic & Levine, 2005; Saini & Singhania, 2018; Asongu & Odhiambo, 2020). The use of this estimator therefore provides a robust basis for credible inference on the effects of FDI, structural factors, and provincial growth dynamics on poverty outcomes in Vietnam.

A set of diagnostic tests is employed to assess the validity of the System GMM estimates, with particular emphasis on instrument relevance, serial correlation, and overall model specification. Autocorrelation in the residuals is examined using the Arellano–Bond test. In dynamic panel models, first-order serial correlation in the differenced residuals (AR(1)) is expected as a consequence of the transformation, whereas the absence of second-order serial correlation (AR(2)) is a necessary condition for model validity. A significant AR(2) statistic would indicate correlation between the lagged instruments and the error term, thereby violating the moment conditions. Instrument validity is further evaluated using the Hansen J-test of overidentifying restrictions, which tests the null hypothesis that the full set of instruments is exogenous. A Hansen p-value within an acceptable range supports the appropriateness of the chosen instrument set. The Sargan test provides a related diagnostic under the assumption of homoskedastic errors; however, because it is not robust to heteroskedasticity, it is reported only as supplementary evidence in the two-step System GMM framework. In addition, careful attention is paid to instrument proliferation, as an excessive number of instruments can weaken the Hansen test and lead to overfitting. Consistent with established guidance (Roodman, 2009; Labra & Torrecillas, 2018), the number of instruments is kept below or close to the number of cross-sectional units. Thus, the AR(1) and AR(2) tests, the Hansen and Sargan statistics, and explicit limits on instrument count provide a coherent basis for evaluating model adequacy. Consistency across these diagnostics supports the reliability of the System GMM estimates in capturing the dynamic relationships among FDI, growth-related factors, and provincial poverty.

While the System GMM framework is well suited for addressing dynamic persistence and endogeneity, it does not explicitly model spatial dependence across provinces. Spatial dynamic panel models would be conceptually appropriate for capturing inter-provincial spillovers; however, their implementation is not feasible in the present study due to the absence of robust and defensible province-level spatial weight matrices that consistently capture economic and logistical integration over time. This limitation is therefore explicitly acknowledged.

3.2. Data and Sources

Table 1 reports the variables used in the analysis, along with their definitions and data sources. All data are collected from the General Statistics Office (GSO), mainly from the annual Statistical Yearbooks (2002–2022) and the biennial Vietnam Household Living Standards Surveys (VHLSS) conducted between 2002 and 2010. The dependent variable, $p_{i,t}$, follows Vietnam’s official poverty measures: income poverty for the period 2002–2016 and the Multidimensional Poverty Index (MPI) for the period 2016–2022. Because the VHLSS is conducted every two years, income-poverty data are unavailable for 2003, 2005, and 2009. To obtain a balanced annual panel suitable for dynamic estimation, these missing observations are interpolated using the observed linear decline in poverty at both the national and provincial levels. For instance, official poverty rates reported for 2010–2016 (14.2%, 12.6%, 11.1%, 9.8%, 8.4%, 7.0%, and 5.8%) imply interpolated values of 12.65%, 9.75%, and 7.1% for 2011, 2013, and 2015, which closely align with subsequently released official figures. This correspondence supports the application of the same interpolation method to the provincial data, where only three observations are missing. This approach preserves long-run poverty trends but may smooth short-run fluctuations, implying that estimates of poverty persistence are likely conservative.

Table 1. Variables, Definitions, Measurements, and Data Sources.

Variable	Definition and Measurement	Data Source
$p_{i,t}$	Poverty incidence of province i in year t. For period 2002–2016: income poverty; For period 2016–2022: Multidimensional poverty index.	GSO Statistical Yearbooks; GSO-VHLSS (biennial surveys, 2002–2010)
$in{c}_{i,t}$	Average per capita income of province i in year t, adjusted by consumer price index	GSO Statistical Yearbooks
$fd{i}_{i,t}$	Ratio of total registered FDI capital to total provincial income.	GSO Statistical Yearbooks
$ca{p}_{i,t}$	Ratio of average production capital to total provincial income.	GSO Statistical Yearbooks
$la{b}_{i,t}$	Ratio of total labor to total population of province i in year t.	GSO Statistical Yearbooks
$go{v}_t$	Ratio of total government expenditure to national GDP in year t.	GSO Statistical Yearbooks
$fd{i}_{i,t}\times in{c}_{i,t}$	$\mathrm{Interaction}\mathrm{variable}\mathrm{between}fd{i}_{i,t}$$\mathrm{and}in{c}_{i,t}$.	Constructed from GSO data
$fd{i}_{i,t}\times p_{i,t-1}$	$\mathrm{Interaction}\mathrm{variable}\mathrm{between}fd{i}_{i,t}$$\mathrm{and}{p}_{i,t-1}$.	Constructed from GSO data

The explanatory variables reported in Table 1 are constructed directly from data published in the General Statistics Office (GSO) Statistical Yearbooks. These include real per capita income, the ratio of registered foreign direct investment (FDI) to provincial income, the ratio of production capital to income, the labor–population ratio, and the ratio of government expenditure to GDP. All the variables follow the definitions provided in Table 1. Income measures are expressed in real terms using the consumer price index (CPI), while ratio variables are calculated from officially reported aggregate values. Interaction terms are generated as simple multiplicative combinations of the relevant variables. Thus, the resulting dataset offers a consistent and transparent foundation for the empirical analysis.

4. Estimation Results and Discussion

4.1. Estimation Results

Table 2, Table 3 and Table 4 report the System GMM estimates of provincial poverty dynamics in Vietnam across different time horizons and corresponding poverty measures. Accordingly, Table 2 presents the results of the estimation of models that employ the income-based poverty measure over the period 2002–2016, while Table 3 reports those based on the use of the multidimensional poverty measure for the period 2016–2022. Table 4 provides estimation results based on a harmonized poverty measure throughout the whole period 2002–2022 for the purpose of assessing the robustness of models corresponding to the two distinct poverty measurement regimes. Overall, the diagnostic tests support the validity of the estimated models. The Arellano–Bond statistics indicate the presence of first-order serial correlation in the differenced residuals, as expected, and provide no evidence of second-order serial correlation. The Hansen and Sargan tests do not reject the validity of the instrument set; the number of instruments remains below the number of cross-sectional units. The F-statistics confirm the joint significance of the regressors.

Table 2. System GMM Estimation Results for Poverty Dynamics using the monetary poverty metric for the period 2002–2016.

Variables	Model 1	Model 2
$p_{i,t-1}$	0.2933 *** (0.1065)	0.2822 ** (0.1125)
$in{c}_{i,t}$	−0.1701 *** (0.0322)	−0.1571 *** (0.0313)
$fd{i}_{i,t}$	−0.0231 *** (0.0088)	−0.0486 *** (0.0174)
$ca{p}_{i,t}$	0.0158 ** (0.0074)	0.0213 ** (0.0078)
$la{b}_{i,t}$	−0.0622 (0.1159)	−0.1593 (0.1055)
$go{v}_t$	−0.8803 (0.1864)	−0.6540 *** (0.2246)
$fd{i}_{i,t}\times in{c}_{i,t}$	0.0025 * (0.0013)	0.0029 ** (0.0020)
$fd{i}_{i,t}\times p_{i,t-1}$	-	0.1585 ** (0.0743)
Constant	0.7737 *** (0.1548)	0.5205 *** (0.1471)
Observations	818	817
Number of groups	59	59
Instruments	10	13
F-statistic	640 (p = 0.000)	629 (p = 0.000)
AR(1) p-value	0.000	0.000
AR(2) p-value	0.488	0.346
Sargan p-value	0.394	0.681
Hansen p-value	0.550	0.489

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10.

Table 3. System GMM Estimation Results for Poverty Dynamic using Multidimensional Poverty metric for the period (2016–2022).

Variables	Model 3	Model 4
$p_{i,t-1}$	0.6128 *** (0.1654)	0.7390 *** (0.0884)
$in{c}_{i,t}$	−0.0858 * (0.0490)	−0.0454 * (0.0249)
$fd{i}_{i,t}$	−0.1529 ** (0.0717)	−0.1578 * (0.08830)
$ca{p}_{i,t}$	0.0016 (0.0014)	0.0038 *** (0.00085)
$la{b}_{i,t}$	−0.0258 (0.0362)	−0.0077 (0.0284)
$go{v}_t$	0.0409 ** (0.1626)	0.1865 * (0.0965)
$fd{i}_{i,t}\times in{c}_{i,t}$	0.0445 ** (0.0208)	0.0385 * (0.0202)
$fd{i}_{i,t}\times p_{i,t-1}$	-	0.1443 * (0.0846)
Constant	0.2363 (1832)	0.1234 (0.0806)
Observations	472	472
Number of groups	59	59
Instruments	10	12
F-statistic	1345 (p = 0.000)	2260 (p = 0.000)
AR(1) p-value	0.066	0.006
AR(2) p-value	0.587	0.178
Sargan p-value	0.118	0.204
Hansen p-value	0.165	0.133

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10.

Table 4. Estimation Results for Poverty Dynamic using harmonized metric for the period (2002–2022).

Variable	Model (5)	Model (6)
$p_{i,t-1}$	0.4083 *** (0.0828)	0.4527 *** (0.1561)
$in{c}_{i,t}$	−0.1235 *** (0.0239)	−0.0881 *** (0.0324)
$fd{i}_{i,t}$	−0.0647 ** (0.0252)	−0.2245 ** (0.1097)
$ca{p}_{i,t}$	0.0045 (0.0033)	0.0039 (0.0032)
$la{b}_{i,t}$	−0.0434 (0.0668)	0.04341 (0.1462)
$go{v}_t$	−0.2894 ** (0.994)	−0.0578 (0.1739)
$fd{i}_{i,t}\times in{c}_{i,t}$	0.0226 ** (0.0085)	0.0644 *** (0.0318)
$fd{i}_{i,t}\times p_{i,t-1}$	—	0.3578 ** (0.1740)
Constant	0.4854 *** (0.1067) ***	0.2893 ** (0.1246) **
Observations	1298	1273
Groups	59	59
Instruments	15	14
F-statistic	918 (p = 0.000)	520 (p = 0.000)
AR(1), p-value	0.001	0.050
AR(2), p-value	0.167	0.120
Hansen test, p-value	0.331	0.280
Sargan test, p-value	0.145	0.108

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05.

In general, the presented results of estimation (Table 2 and Table 3) indicate persistence of poverty across all the empirical models and across the two different time horizons. All the coefficients of lagged poverty are significantly positive, though their magnitudes are different across the two periods. Real capita income and FDI consistently exhibit to be the significant drivers of poverty reduction. The positive coefficients of the interaction terms between FDI and income and between FDI and lagged poverty consistently indicate that the poverty-reducing effect of FDI is conditionally linked to provincial absorptive capacity, weakening as income levels rise and as poverty becomes more persistent. Government expenditure exerts a poverty-reducing effect under the regime of income-based poverty measure (2002–2016), but this is not confirmed under the regime of multidimensional poverty measure (2016–2022). Capital accumulation and labor intensity have limited and unstable effects in both periods.

As presented in Table 4, the estimation results based on the harmonized poverty series indicate that the main coefficients retain their signs and conditional patterns. However, the pooled estimates yield just average effects as compared to those associated with each distinct time horizon. This indicates that the impact of FDI on poverty varies across poverty regimes rather than remaining constant over time. For this reason, Table 4 is used to assess coefficient stability, while the main findings and policy implications are drawn from the regime-specific results in Table 2 and Table 3. A more detailed discussion of the estimates follows in the next section.

Given concerns regarding potential reverse causality between FDI and poverty, we additionally implement the Dumitrescu and Hurlin (2012) heterogeneous panel Granger non-causality test. As reported in Table 5, for the 2002–2016 income-poverty period—where sufficient time observations are available—the test is conducted using first-differenced variables to ensure stationarity. The results indicate that short-run changes in FDI Granger-cause changes in poverty (Z-bar tilde = 2.81, p = 0.0049), while no significant reverse predictive relationship from poverty to FDI is detected (Z-bar tilde = 0.37, p = 0.7123). Because poverty measurement underwent a structural change in 2016, the test cannot be reliably implemented for the 2016–2022 multidimensional poverty period due to its short time dimension. To assess consistency across measurement regimes, the test is also applied to the harmonized poverty series over 2002–2022, yielding qualitatively similar results: short-run changes in FDI Granger-cause changes in poverty (Z-bar tilde = 3.70, p = 0.0002), again with no evidence of reverse causality from poverty to FDI (p = 0.9703). These findings provide supportive evidence of temporal directionality in a predictive sense rather than structural causal proof and are complementary to the System GMM framework.

Table 5. Dumitrescu and Hurlin (2012) Panel Granger Non-Causality Test Results.

Poverty Measure	Period	Null Hypothesis	Z-Bar Tilde	p-Value	Conclusion
Monetary Poverty	2002–2016	$\mathsf{\Delta }fdi$$\mathrm{does}\mathrm{not} \text{Granger-cause} \mathsf{\Delta }p$	2.81	0.0049	Reject H0
Monetary Poverty	2002–2016	$\mathsf{\Delta }p$$\mathrm{does}\mathrm{not} \text{Granger-cause} \mathsf{\Delta }fdi$	0.37	0.7123	Do not reject H0
Harmonized series of Poverty	2002–2022	$\mathsf{\Delta }fdi$$\mathrm{does}\mathrm{not} \text{Granger-cause} \mathsf{\Delta }p$	3.70	0.0002	Reject H0
Harmonized series of Poverty	2002–2022	$\mathsf{\Delta }p$$\mathrm{does}\mathrm{not} \text{Granger-cause} \mathsf{\Delta }fdi$	0.04	0.9703	Do not reject H0

Notes: The table reports the Dumitrescu and Hurlin (2012) heterogeneous panel Granger non-causality test using first-differenced variables to ensure stationarity. The Z-bar tilde statistic is reported due to moderate time dimension (small-sample correction). Rejection of the null hypothesis indicates Granger causality for at least one panel unit.

4.2. Discussion

This section interprets the empirical findings in light of the national and international literature, emphasizing their economic and theoretical implications for Vietnam’s development trajectory.

4.2.1. Poverty Persistence Across Periods

The presented results in Table 2 and Table 3 confirm the existence of poverty persistence across provinces, but the magnitudes of poverty persistence are different across the two examined periods. During 2002–2016, when poverty is measured using an income-based indicator, the coefficient on lagged poverty is positive but relatively small (around 0.28–0.29 reported in Table 1). This suggests limited persistence and relatively fast adjustment. The result is consistent with Vietnam’s development conditions at the time: poverty levels were high at the outset but declined rapidly as growth accelerated, employment expanded, and structural change intensified. In this setting, income gains translated quickly into poverty reduction, allowing provinces with higher initial poverty to record faster declines. Similar patterns have been observed in Vietnam and other rapidly growing Asian economies, where income poverty responds strongly to growth in early and middle stages of development (Ravallion & Chen, 2007; World Bank, 2018b).

By contrast, the period after 2016 shows much stronger persistence, with the coefficients on the lagged poverty rising to about 0.61–0.74 as reported in Table 3. This shift reflects two main factors. First, Vietnam adopted a multidimensional poverty measure in 2016, which includes deprivations in education, health, living conditions, and accessibility to information in addition to income. These dimensions tend to adjust slowly and are less sensitive to short-term economic growth. Second, by 2016, monetary poverty had already fallen to low levels, making further reductions more difficult and increasingly concentrated among structurally disadvantaged groups. As poverty declines, additional improvements become smaller, leading to greater inertia in measured outcomes.

Results based on the harmonized poverty series reported in Table 4 help reconcile these two regimes. The estimated coefficients on lagged poverty for the full 2002–2022 period fall between those reported in Table 2 and Table 3, indicating that pooled estimates combine two distinct dynamic processes rather than reflecting a single stable relationship. This intermediate level of persistence supports the view that Vietnam experienced a structural shift in poverty dynamics, driven by both declining poverty levels and the transition to a more demanding measurement framework.

This finding is consistent with the multidimensional poverty literature, which emphasizes that non-income deprivations—such as education, health, and living conditions—tend to adjust more slowly than monetary poverty and are therefore less responsive to short-run income growth (Alkire & Foster, 2011; UNDP—United Nations Development Programme, 2019). Empirical evidence from both country-specific and cross-country studies supports this interpretation. For example, Baulch and Masset (2003) show that non-monetary indicators of deprivation in Vietnam, including nutrition and education, exhibit substantially greater persistence over time than income-based poverty measures, even during periods of rapid growth. Similarly, longer-run analyses of multidimensional poverty document the prevalence of chronic deprivations that persist across multiple periods and dimensions, reflecting deep structural constraints rather than weak policy effectiveness (Alkire et al., 2017). Taken together, this body of evidence suggests that observed increases in poverty persistence and slower convergence following measurement transitions are primarily driven by the structural nature of remaining deprivations, rather than by a decline in the effectiveness of poverty-reduction policies.

4.2.2. Conditional Beta Convergence of Poverty

In the dynamic specification, beta convergence is assessed through the coefficient on lagged poverty. When the interaction term is included, the marginal effect of initial poverty on current poverty is expressed as ${\beta }_1+{\gamma }_{2}fd{i}_{i,t}$, as shown in Equation (3). Conditional beta convergence exists when this marginal effect lies between zero and one, conditional on the level of FDI. For the period using monetary-based poverty measurement (2002–2016), the coefficient on lagged poverty (${\beta }_1)$ is approximately 0.2822, indicating relatively low persistence. Because the coefficient on the interaction term between FDI and lagged poverty $\left({\gamma }_2\right)$ is positive (0.1585), the effective persistence parameter increases with FDI exposure. Evaluated at the sample mean FDI value of 0.655, the effective persistence parameter equals 0.386. The implied conditional beta convergence is −0.614, indicating strong convergence and rapid adjustment. This finding is consistent with the fast reduction in poverty observed during a period when income growth translated efficiently into welfare improvements.

For the period using multidimensional poverty measurement (2016–2022), the coefficient on lagged poverty is substantially higher $({\beta }_1=0.7390$), implying greater persistence. The coefficient on the interaction term between FDI and lagged poverty remains positive $\left({\gamma }_2=0.1443\right)$, further increasing effective persistence. Using the sample mean FDI value of 0.753, the effective persistence parameter rises to 0.848. The implied conditional beta convergence is −0.152, reflecting the greater inertia associated with non-income deprivations.

Overall, the results indicate that beta convergence continues to hold when conditioning on FDI in both poverty regimes. However, higher FDI exposure raises effective poverty persistence by strengthening the link between past and current poverty.

4.2.3. Marginal Effects of FDI Across Income and Poverty Levels

This subsection extends the analysis of poverty persistence and convergence by examining how the marginal effect of foreign FDI on poverty varies across income and poverty distributions. Because Vietnam’s poverty measurement changed after 2016, marginal effects are estimated separately for the income-based poverty period (2002–2016) and the multidimensional poverty period (2016–2022), using results from the System GMM models. Estimating effects by period allows the interactions among FDI, income, and initial poverty to be interpreted within the appropriate measurement framework and stage of development. It should also be noted that, in the absence of explicit spatial modeling, the estimated FDI effects may capture a combination of local and inter-provincial influences rather than purely localized impacts.

Table 6 reports the marginal effects of FDI on income poverty for the period 2002–2016. These effects are evaluated at selected points of the income and poverty distributions to capture heterogeneity during a phase of rapid growth and income-driven poverty reduction. Panel A presents marginal effects at the 25th, 50th, and 75th percentiles of income, with the poverty rate held at its sample mean (0.1520). FDI is associated with a negative and statistically meaningful reduction in poverty across all income levels. However, the magnitude of the effect declines modestly as income rises, from −0.018 at the 25th percentile to −0.016 at the 75th percentile. This pattern indicates that FDI was broadly pro-poor during 2002–2016 but had stronger effects in lower-income provinces, where income growth and job creation translated more directly into reductions in monetary poverty. At higher income levels, additional FDI yields smaller marginal gains, consistent with a stage in which many provinces had already benefited from earlier growth-led poverty reduction.

Table 6. Marginal effects of FDI on monetary poverty rate (2002–2016).

Marginal effect formula
Marginal effect of FDI = ${\beta }_3+{\gamma }_{1}inc+{\gamma }_2{p}_{t-1}$
Inputs (coefficients from estimation)
${\beta }_3$	−0.0486
${\gamma }_1$	0.0029
${\gamma }_2$	0.1585
Mean inc (ln of real per capita income)	2.5590
Mean p (monetary poverty rate)	0.1520
Panel A—Across income quantiles (holding monetary poverty rate at mean)
Income quantile	Real income per capita	Monetary poverty rate (held)	Marginal effect of FDI
P25	2.2475	0.1520	−0.0180
P50	2.5318	0.1520	−0.0172
P75	2.8547	0.1520	−0.0162
Panel B—Across monetary poverty rate quantiles (holding income at mean)
Monetary poverty rate quantile	Monetary poverty rate	Real income per capita (held)	Marginal effect of FDI
P25	0.0742	2.5590	−0.0294
P50	0.1284	2.5590	−0.0208
P75	0.2107	2.5590	−0.0078

Panel B examines marginal effects across the poverty distribution, with income fixed at its mean value (2.559). A clearer gradient appears in this case. The poverty-reducing effect of FDI is strongest in provinces with lower initial poverty (−0.031 at the 25th percentile) and weakens as poverty increases, falling to −0.011 at the 75th percentile. This pattern suggests that even during the income-poverty regime, FDI was less effective in provinces with higher initial poverty, suggesting the presence of structural constraints such as limited absorptive capacity, weaker labor skills, and fewer linkages between foreign firms and poor households.

These results are consistent with earlier findings of low poverty persistence and rapid beta convergence during 2002–2016. In a period marked by strong growth and industrial expansion, FDI contributed to poverty reduction mainly through employment creation, wage increases, and export-oriented manufacturing. These channels tend to operate more effectively in provinces with better infrastructure, skills, and market integration. Provinces with very high poverty levels appear to have benefited less from FDI, even under an income-based poverty framework, reflecting persistent spatial and structural disparities (World Bank, 2018a).

Table 7 presents the marginal effects of FDI on poverty during the period 2016–2022. Across income quantiles, FDI is associated with stronger poverty reduction at lower income levels, with the effect weakening at higher income levels. The marginal effect is most negative at the lower end of the income distribution (around −0.031 at the 25th percentile) and becomes smaller at higher income levels (around −0.015 at the 75th percentile, holding poverty at its mean). Although the average effect of FDI remains negative, heterogeneity across income groups is more pronounced. A similar pattern appears across poverty quantiles: the poverty-reducing effect of FDI is strongest at lower poverty levels (about −0.031 at the 25th percentile) and weakens as poverty becomes more severe (about −0.020 at the 75th percentile). These results indicate that FDI is less effective in provinces facing deeper and more persistent deprivation.

Table 7. Marginal effects of FDI on multidimensional poverty rate (2016–2022).

Marginal effect formula
Marginal effect of FDI = ${\beta }_3+{\gamma }_{1}inc+{\gamma }_2{p}_{t-1}$
Inputs (coefficients from System GMM)
${\beta }_3$	−0.1578
${\gamma }_1$	0.0385
${\gamma }_2$	0.1443
Mean inc (ln of real per capita income)	3.1982
Mean p (Multidimensional poverty rate)	0.0820
Panel A—Across income quantiles (holding multidimensional poverty rate at mean)
Income quantile	Real income per capita	Multidimensional poverty rate (held)	Marginal effect of FDI
P25	2.9900	0.0820	−0.0308
P50	3.2036	0.0820	−0.0226
P75	3.3987	0.0820	−0.0151
Panel B—Across multidimensional poverty rate quantiles (holding income at mean)
Multidimensional poverty rate quantile	Multidimensional poverty rate value	Real income per capita (held)	Marginal effect of FDI
P25	0.0226	3.1982	−0.0314
P50	0.0547	3.1982	−0.0268
P75	0.1029	3.1982	−0.0198

This increased heterogeneity is consistent with the transition from income-based to multidimensional poverty measurement, which explicitly includes access to basic social services. As shown by the persistence and convergence results, multidimensional poverty adjusts more slowly, suggesting that further reductions depend less on income growth and more on institutional capacity and service provision. In this setting, FDI—often concentrated in urban, industrial, or service activities—can raise income and employment without producing comparable improvements in non-income dimensions of welfare. Vietnam’s experience after 2016 supports this view: despite continued strong FDI inflows, poverty reduction slowed as remaining deprivations became more structural and less responsive to market-driven growth alone (World Bank, 2022).

The marginal-effects results are consistent with the dynamic findings on poverty persistence and convergence. During 2002–2016, rapid convergence and low persistence coincided with relatively uniform, though gradually diminishing, poverty-reducing effects of FDI. In this period, as featured by strong income growth and industrial expansion, FDI reduced poverty mainly through job creation, wage increases, and export-oriented manufacturing—channels that operate more effectively where absorptive capacity and market integration are stronger. Provinces with very high poverty levels benefited less, reflecting constraints related to skills, infrastructure, and local production linkages. This pattern aligns with evidence that Vietnam’s early poverty reduction was driven largely by growth and labor reallocation, while spatial and structural disparities remained (World Bank, 2018a, 2022). During 2016–2022, higher persistence and slower convergence were accompanied by sharper heterogeneity in marginal effects, with FDI providing limited benefits in provinces with more entrenched multidimensional poverty. This does not imply that FDI increases poverty; rather, it suggests that its poverty-reducing role increasingly depends on local conditions and complementary factors such as human capital and access to basic services.

The finding that foreign direct investment reduces poverty but with diminishing effectiveness in provinces facing deeper deprivation is consistent with a broader body of international evidence emphasizing the conditional nature of FDI’s welfare effects. Prior studies show that the poverty impact of FDI depends critically on local absorptive capacity, institutional quality, and complementary economic conditions (Reiter & Steensma, 2010; Gohou & Soumaré, 2012; Klein & Hadjimichael, 2003). In line with this perspective, Daud et al. (2025) find that while FDI is generally associated with poverty reduction, its effects are highly heterogeneous across income quantiles and are mediated by factors such as trade openness and income inequality. Conversely, where key threshold conditions are not met or supporting resources are misallocated, the poverty effects of FDI may be modest or even adverse, as documented by Sattar et al. (2022). Taken together, these studies suggest that FDI does not operate as an automatic engine of poverty reduction; rather, its effectiveness depends on the extent to which host regions possess the institutional and structural capacity to translate external investment into broad-based welfare gains.

These findings are consistent with Vietnam’s broader spatial and structural development patterns. Empirical evidence generally suggests that FDI can improve employment and household welfare outcomes. However, the magnitude of these benefits varies across regions and population groups. Consequently, uneven growth patterns associated with FDI may give rise to economic imbalances and distributional disparities, potentially widening inequality (Bui et al., 2019; Do et al., 2021; Le et al., 2021; McLaren & Yoo, 2016). Evidence also points to heterogeneous productivity and linkage effects of FDI, with spillovers often found to be weak or statistically insignificant in regions and sectors lacking sufficient absorptive capacity and domestic supplier networks (Ngoc & Ramstetter, 2004; Anwar & Nguyen, 2011; Newman et al., 2015; Görg & Greenaway, 2004; Javorcik, 2004; P. V. Nguyen et al., 2020). As Vietnam shifted toward a multidimensional poverty framework, these uneven effects became more apparent, helping to explain why higher FDI exposure does not necessarily lead to faster poverty convergence in provinces with persistent deprivation.

From a theoretical perspective, these results support models of conditional convergence in which capital inflows accelerate growth but do not automatically reduce spatial inequality when structural constraints persist (Barro & Sala-i-Martin, 2004; Ravallion & Chen, 2003). FDI therefore acts as a growth-enhancing but not necessarily convergence-enhancing force unless complemented by policies that strengthen local capabilities.

4.2.4. Impact of Control Variables on Poverty Dynamics

In addition to income and foreign direct investment (FDI), the estimation results point to different and uneven roles for other growth-related factors commonly discussed in the growth literature. These variables are conceptually linked to poverty reduction, but their effects vary across periods and poverty measures.

The coefficients on production capital (cap) are quite small across the two periods and models, suggesting weak and unstable links with poverty reduction. Under the income-poverty framework during the years 2002–2016, these coefficients are positive and statistically significant in both Model 1 and Model 2. Under the multidimensional poverty framework during the years 2016–2022, the coefficients on production capital (cap) turn out to be much smaller and statistically insignificant in one specification. This result implies that provinces with higher capital–income ratios did not experience faster declines in income poverty once income growth, FDI, and poverty persistence are taken into account. This finding is consistent with evidence from related studies, which suggests that capital accumulation alone may not guarantee inclusive growth, particularly when investment is concentrated in capital-intensive activities with limited employment creation and weak linkages to poor households (McCaig & Pavcnik, 2013). Furthermore, under the multidimensional poverty framework, when poverty measurement places greater emphasis on access to basic social services, increases in production capital may contribute to output but tend to have a weaker and more indirect effect on multidimensional poverty outcomes, as many deprivations are structural and not immediately responsive to income growth (Alkire & Foster, 2011; World Bank, 2022). The estimation results based on the harmonized poverty series for 2002–2022 support this interpretation, as the coefficients on production capital lose statistical significance when the two poverty regimes are pooled, indicating limited robustness across different poverty measurement frameworks. Taken together, production capital has no independent poverty-reducing effect, and any association with poverty outcomes is largely mediated through income growth rather than direct employment or welfare channels.

Government expenditure (gov) exhibits markedly different effects on poverty across periods. Under the income-based poverty measurement framework during 2002–2016, the coefficients on government expenditure are negative and statistically significant, suggesting that higher public spending is associated with lower income poverty. This finding is consistent with Vietnam’s experience in the 2000s, when large public investment programs and national target programs contributed to rapid poverty reduction through infrastructure development, rural investment, education expansion, and improved access to basic health services (World Bank, 2002, 2009). In contrast, under the multidimensional poverty framework during 2016–2022, the coefficients on government expenditure become positive and statistically significant in one model. This result should not be interpreted as evidence of a causal poverty-increasing effect of government spending. Rather, it is consistent with the fact that government resources are deliberately allocated toward provinces and population groups facing deeper and more persistent deprivation. In Vietnam’s post-2016 policy context, government expenditure is often directed toward the development of electricity, roads, schools, and health facilities in mountainous, remote, and structurally disadvantaged areas. These areas normally experience higher degrees of multidimensional poverty persistence, where improvements in non-income dimensions—such as education, health access, housing, and basic services—materialize only gradually. Consequently, higher government spending may coincide contemporaneously with higher rates of multidimensional poverty. Similar findings have been reported in the multidimensional poverty literature, where policy intensity and observed deprivation are positively correlated in the short run due to spatial targeting and persistence of non-monetary deprivations (World Bank, 2022; Alkire et al., 2017). The results of estimation using the harmonized series of poverty indicate negative or insignificant coefficients on government expenditure, thus consolidating the view that the poverty effect of government spending depends on the poverty measure and provincial economic context.

The coefficients on labor force (lab) are statistically insignificant across all specifications. This outcome does not mean a lack of relevance of employment for poverty reduction. Rather, it reflects structural features of the labor market in Vietnam. During earlier stages of economic transition and the process of monetarization, employment growth played an important role in reducing poverty. However, the workforce in Vietnam is essentially featured as having a large share of low-productivity, informal, or vulnerable jobs. Under such conditions, higher labor participation alone may not be clearly associated with improved household welfare or meaningful poverty reduction, as gains in poverty reduction depend critically on job quality and productivity (World Bank & MPI—Ministry of Planning and Investment of Vietnam, 2016; World Bank, 2014; McCaig & Pavcnik, 2013; Fields, 2011). Limited variation in labor force participation across provinces may also reduce the explanatory power of labor force once income, investment, and dynamic effects are controlled for (P. T. Nguyen, 2025). More importantly, the labor–population ratio captures only the quantity of labor, not its quality. It does not reflect differences in skills, wages, job security, formality, or sectoral allocation. As a result, labor quantity emerges as a weak predictor of poverty outcomes across both poverty regimes.

5. Concluding Remarks, Policy Implications, and Limitations

This study examines the role of FDI in poverty reduction in Vietnam by accounting for poverty dynamics, regional heterogeneity, and the shift from income-based poverty measurement to multidimensional poverty measurement. Using provincial panel data for 2002–2022 and a system GMM approach, the analysis leads to three main findings. First, poverty dynamics in Vietnam differ clearly across the two frameworks of poverty measurement. During the income-poverty period (2002–2016), poverty persistence was relatively low and beta convergence was rapid, reflecting a stage of development in which income growth translated effectively into poverty reduction. After 2016, when Vietnam adopted a multidimensional poverty framework, poverty became more persistent and convergence slowed. This change does not signal policy failure. Instead, it reflects the structural nature of remaining deprivation once income poverty has largely been reduced. Multidimensional poverty depends on education, health, housing, and access to basic services, which improve more slowly than income. These results highlight the need to interpret poverty dynamics within their institutional and measurement context rather than assuming stable relationships over time.

Second, FDI is empirically proven to be an important driver of poverty reduction, but its effects are conditional and uneven across regions. In both frameworks of poverty measurement, FDI is negatively associated with poverty, confirming its pro-development role in the Vietnamese economy. However, the analysis of interaction effects indicates that the marginal impact of FDI on poverty depends on provincial income levels and initial poverty conditions. During 2002–2016, FDI was broadly pro-poor but more effective in provinces with lower poverty and stronger absorptive capacity. During 2016–2022, the conditional poverty impact of FDI became more pronounced. FDI is more effective for poverty reduction in provinces with lower income levels. In the meantime, FDI is less effective for poverty reduction in provinces facing deeper and more persistent deprivation.

Third, the empirical results suggest that FDI affects the speed of poverty convergence across provinces. The positive coefficients on the interaction term between FDI and lagged poverty imply that provinces with greater FDI inflows tend to experience higher effective poverty persistence, thus lower beta convergence. This finding does not suggest that FDI increases poverty. Rather, it reflects the spatial and structural pattern of FDI allocation in Vietnam. In practice, FDI has increasingly concentrated in manufacturing, services, and industrial clusters located in provinces with better infrastructure, skilled labor, and market access. These provinces typically exhibit already low poverty rates, such that additional FDI inflows generate relatively small marginal reductions in poverty. By contrast, in poorer or more remote provinces, FDI projects—often capital- or skill-intensive—may generate limited spillovers to disadvantaged households. As a result, while FDI may contribute to poverty reduction in better-positioned regions, it does so unevenly, thus slowing the relative pace at which poorer provinces converge toward lower poverty levels.

The three main findings indicate that FDI does not operate as a uniform driver of inclusive growth. Although FDI can contribute to poverty reduction, its effects appear to be shaped by local absorptive capacity, institutional conditions, and the nature of deprivation being addressed. This interpretation helps reconcile Vietnam’s continued success in attracting FDI with the observed slowdown in poverty reduction after 2016 and highlights why growth-oriented investment alone cannot resolve remaining structural poverty.

These findings point to several important policy implications. First, FDI policy may be geared to move beyond a focus on investment volume toward greater emphasis on quality and local linkages. In earlier stages of FDI attracting, broad FDI inflows supported income growth and monetary poverty reduction. Under a multidimensional poverty framework, priority should shift toward projects that create stable jobs, raise skills, and strengthen links with domestic firms. Without such linkages, additional FDI may reinforce spatial concentration rather than support convergence.

Second, government spending also plays a critical complementary role. The interaction between FDI and poverty persistence indicates that private investment alone cannot address structural poverty. Targeted spending on education, health, basic infrastructure, and social protection is needed to raise absorptive capacity in poorer provinces and enable them to benefit more fully from FDI, particularly when poverty is multidimensional.

Third, both FDI policy and poverty-reduction strategies should be formulated with explicit attention to regional differentiation. Uniform national policy frameworks, although grounded in the expectation that FDI supports poverty reduction, may be less effective in structurally disadvantaged regions. More targeted approaches—such as encouraging labor-intensive or service-oriented FDI in poorer provinces, combined with investments in skills development—are more likely to support inclusive outcomes.

Vietnam’s experience indicates that FDI remains relevant for poverty reduction, although its role has evolved over time. As poverty increasingly reflects structural and multidimensional constraints, the effectiveness of FDI becomes more closely tied to the presence of supportive institutions, human capital, and well-targeted public policy. These insights extend beyond Vietnam and are relevant for other developing and transitional economies that seek to deploy FDI as an instrument for inclusive and sustainable development.

Several limitations of this study reflect data and design constraints inherent in long-run provincial analyses and are acknowledged here to clarify interpretation and indicate directions for future research.

First, the analysis spans two distinct poverty measurement regimes—income-based poverty (2002–2016) and multidimensional poverty (2016–2022)—which introduces an unavoidable conceptual discontinuity. Although this issue is addressed through regime-specific estimations and the use of a harmonized poverty series for robustness checks, full comparability across regimes cannot be achieved with currently available data. Future research could explore alternative harmonization strategies or latent-variable approaches to poverty measurement when suitable indicator-level data become available.

Second, due to the biennial nature of household surveys during part of the study period, a small number of poverty observations are interpolated to construct a balanced panel for 2002–2016. While the largely monotonic decline in income-based poverty mitigates this concern, interpolation may smooth short-term fluctuations and thus yield conservative estimates of poverty persistence. Future studies could assess robustness using alternative missing-data strategies.

Third, although the System GMM estimator is employed to mitigate endogeneity, causal interpretation remains limited by the observational nature of the data. Accordingly, the estimated effects of FDI on poverty should be interpreted as conditional associations, derived from a triangulation strategy combining dynamic panel estimation, robustness across specifications, and consistency with the existing literature.

Fourth, spatial spillovers are not explicitly modeled due to the absence of robust provincial spatial weight matrices, and estimated FDI effects may therefore reflect a combination of local and inter-provincial influences. In addition, the analysis relies on registered FDI due to data availability, which reflects investment commitments rather than fully realized or quality-adjusted investment. As a result, implementation delays, project cancellations, and heterogeneity in investment quality across sectors and technologies cannot be fully captured. This measurement limitation may lead to an overstatement of effective investment and is, therefore, explicitly acknowledged.

Fifth, interaction terms are specified linearly to capture average conditional relationships, while potentially nonlinear or threshold effects related to absorptive capacity are not explicitly modeled. Future research could explore such nonlinearities using threshold or quantile-based approaches when data and identification strategies permit.

Overall, these limitations primarily reflect data constraints rather than shortcomings of the empirical framework. The findings should therefore be interpreted as evidence of conditional relationships shaped by structural and regional heterogeneity, providing a foundation for future work using richer data and complementary research designs.