The Impact of Strategic Global Integration on Sustainable Human Development in Ethiopia: Disentangling the Roles of Trade and FDI

Abstract

Ethiopia presents a compelling paradox in sustainable development: despite decades of rapid economic growth, improvements in human well-being have not been commensurate. This study examines the role of global economic integration in resolving this paradox by analyzing the impact of trade openness (TOP) and foreign direct investment (FDI) on human development in Ethiopia from 1991 to 2021. We hypothesize that this paradox arises because the benefits of trade and FDI operate primarily through an income-growth channel, with a weaker direct effect on health and education capabilities. Moving beyond the standard Human Development Index (HDI), we construct a modified index (HDI*) that excludes the income component, allowing us to disentangle direct effects on health and education from indirect effects mediated through economic growth. Using the ARDL bounds testing approach, we find that TOP and FDI have significantly stronger long-run effects on standard HDI (0.343 and 0.214, respectively) than on the non-income HDI* (0.235 and 0.136). This indicates that approximately one-third (31.5%) of TOP’s and over one-third (36.4%) of FDI’s total benefit is income-mediated, while the remainder reflects direct capability enhancement. The analysis further reveals that institutional quality significantly amplifies these benefits, whereas inflation specifically undermines non-income dimensions, highlighting the acute vulnerability of social sectors to macroeconomic instability. We conclude that the Ethiopian paradox stems not from a failure of growth but from its weak translation into direct, sustainable gains in health and education. We recommend policies to strengthen institutional governance, attract FDI into health and education sectors, lower trade barriers for agricultural exports, and use trade agreements to address structural trade deficits and promote sustainable human development.

1. Introduction

The concept of human development shifts development economics from income accumulation to expanding opportunities, freedoms, and abilities to lead meaningful lives [1]. While Economic growth and capital accumulation are necessary, they are not sufficient; true development also requires qualitative gains in social and economic well-being [2].

To operationalize this human-centered approach, the Human Development Index (HDI) was created. It measures a country’s average achievement in three dimensions: health (a long and healthy life), education (knowledge) and income (a decent standard of living). By combining these, the HDI offers a more holistic view of development than income-based metrics alone [3]. Globally, human development metrics have improved in recent decades [4]. However, this progress remains uneven. Many African nations have seen significant economic growth without similar advances in human development, highlighting a disconnect between macroeconomic performance and citizens’ lived experiences [5].

This phenomenon is not unique to Africa. Globally, several countries have experienced similar disparities—a pattern often termed “growth without development”. For instance, Equatorial Guinea witnessed immense GDP per capita growth following oil discoveries in the 1990s, yet its HDI progress stagnated due to weak governance and poor social spending [6]. Similarly, rapid growth in resource-rich states like Angola and Azerbaijan did not translate into proportional gains in life expectancy or education quality [7,8]. Even in high-growth Asian economies like Cambodia and India, rapid export-led expansion has at times outpaced improvements in non-income dimensions of well-being [9,10]. These cross-national examples underscore that the Ethiopian paradox reflects a broader developmental challenge in which the mechanisms linking economic expansion to human capability expansion are often inefficient or broken.

Ethiopia exemplifies this paradox. It has been among the world’s fastest-growing economies for two decades, yet it is still classified as having low human development, with an HDI of 0.485 in 2021—a figure that is below global and regional averages [11]. Understanding why rapid economic growth has not translated into improved human development is a central challenge. The country has pursued reforms to increase trade openness and attract foreign direct investment (FDI), based on the premise that global integration drives development. However, trade openness remains at only 26.5% of GDP, significantly lower than the Sub-Saharan African average of 56%, and trade deficits persist at 10.1% of GDP [12]. These outcomes raise a critical question: have these integration policies genuinely fostered sustainable human development?

Conventional policy strongly favors trade openness and foreign direct investment, assuming they foster development through technology transfer and competition [13,14]. Yet, empirical evidence, globally and in Ethiopia, is inconsistent [15]. For instance, some studies report positive effects of trade openness and FDI on Ethiopia’s economic growth [16], while others find negative or negligible impacts [17]. More importantly, a significant gap exists in understanding their direct link to broader human development outcomes in Ethiopia. The existing limited studies often rely on cross-country panel data methods [18], which can mask critical nation-specific dynamics. This ongoing ambiguity shifts the key research question from whether trade and investment impact well-being to how and under what conditions they do.

We argue that this ambiguity fundamentally stems from a failure to disentangle income-mediated effects from direct capability effects. To address this, we employ a novel dual-model framework, estimating effects on both the standard HDI and a modified index (HDI*) that excludes the income component. This approach directly tests whether globalization’s benefits operate primarily through economic growth or directly enhance health and education, thereby offering a new explanation for Ethiopia’s sustainable development paradox.

This study investigates how trade openness and FDI have affected Ethiopia’s sustainable human development from 1991 to 2021 using time-series analysis. It is guided by two main research questions: What are the short- and long-term effects of trade openness and FDI on human development? To what extent do these factors directly affect non-income dimensions of human development, independent of their effects on economic growth?

Our three contributions are as follows: First, this study provides a novel methodological contribution by developing and applying a formal dual-model framework that disentangles income-mediated from direct capability effects. By estimating parallel models for the standard HDI and a modified HDI* (excluding income), we move beyond the limitations of composite indices and offer a clearer diagnostic tool for understanding development pathways.

Second, we provide critical country-specific evidence for Ethiopia using time-series analysis. Most existing studies on this topic employ cross-country panels, which may obscure nation-specific dynamics and historical path dependencies. Our longitudinal analysis captures Ethiopia’s unique post-1991 reform trajectory, offering insights directly relevant to its policy landscape.

Third, the study makes a substantive theoretical contribution by empirically testing the capability-based critique of growth-centric development models in a specific national context. Our findings validate the theoretical proposition that economic growth and human development are linked through distinct, quantifiable channels whose relative strength depends on structural and policy conditions.

2. Literature Review

2.1. Theoretical Framework and Hypothesis Development

The relationship between globalization and development is theorized through two dominants, yet distinct, paradigms. This raises a critical question: Which pathway, income-mediated or direct capability, dominates in influencing human development? To answer this, we consider two perspectives.

The first paradigm, based on neoclassical and endogenous growth theories, proposes an income-mediated pathway. In this framework, TOP and FDI boost GDP per capita, mainly through efficiency gains, technology spillovers, and capital accumulation [19]. Higher national income then enables governments to allocate more resources to health and education, while also raising household disposable income. Thus, greater resource availability becomes the central mechanism translating economic growth into improved human development outcomes. Accordingly, under this paradigm, TOP and FDI are expected to exert a strong, positive effect on the standard HDI—which includes income components—and a weaker or delayed impact on HDI dimensions not directly linked to income.

The second paradigm, championed by [3], argues for the direct capability pathway. This human-centered perspective suggests that trade openness and foreign direct investment can directly improve health and education, regardless of prior income growth. The mechanisms include: importing medical supplies and educational technology, which increases access to modern treatments and learning tools; knowledge transfers from foreign enterprises, which enhance service efficiency through new practices and training; and the construction of clinics and schools by socially responsible investors, which provides immediate infrastructure for health and education. For instance, consider a rural clinic in Ethiopia that receives an imported ultrasound machine. This investment not only enables midwives to diagnose and monitor pregnancies more effectively but also reduces maternal and infant mortality rates significantly, thereby making the capability expansion feel immediate and tangible. This pathway centers on directly expanding human capabilities.

These theoretical pathways are critically tested in the Ethiopian context, where specific structural characteristics determine their relative dominance. The country’s economy relies heavily on primary commodity exports and has attracted FDI concentrated in capital-intensive, enclave sectors (e.g., construction, large-scale agriculture). This pattern suggests that the benefits of trade and FDI are likely channeled more through fiscal revenues and GDP growth than through direct knowledge spillovers or improvements in social service delivery. Consequently, the persistent ambiguity in the empirical literature regarding globalization’s impact stems from a methodological failure to separate these two effects. We contend that in Ethiopia, structural factors strongly justify the thesis that the income-mediated channel will dominate. Therefore, testing the hypothesis that globalization’s sustainable benefits are channeled more strongly through economic growth than through direct capability expansion provides a targeted, structural explanation for Ethiopia’s core development paradox.

To empirically test which pathway dominates, we formulate the following hypotheses based on our dual-model framework. As illustrated in Figure 1, our conceptual model posits that trade openness and foreign direct investment influence human development through two primary channels: an income-mediated channel and a direct capabilities channel. This yields the following central hypotheses:

H1: 

Trade openness and foreign direct investment have a significant positive long-run effect on the standard Human Development Index (HDI).

H2: 

The direct effect of trade openness and foreign direct investment on the non-income HDI* will be statistically weaker than their total effect on the standard HDI, supporting the primacy of the income-mediated channel in Ethiopia.

The diagram illustrates the dual-model approach for testing H1 and H2. H1 is tested via Model A, estimating the total effect of TOP and FDI on the standard HDI. H2 is tested via Model B, estimating their direct effect on the non-income HDI* (controlling for income). The income-mediated effect is derived as the difference between these estimated effects.

2.2. Synthesizing Empirical Evidence: Patterns and Contradictions

Empirical research examining the relationship between trade openness (TOP) and foreign direct investment (FDI) and human development (HD) reveals considerable inconsistency in findings. These contradictory findings stem from variations in both the transmission channels and contextual factors across studies. Differences emerge between analyses focusing on direct effects on health [20,21] and education versus indirect, growth-mediated effects [22,23], as well as broader contextual factors [24]. A significant source of variation is the choice of developmental metric [25,26]; studies employing the standard HDI often support the income-mediated perspective [27], whereas those focusing on non-income dimensions may support the direct capabilities pathway [18,28]. Our dual-index framework (HDI and HDI*) directly addresses this discrepancy, demonstrating that the choice of metric is a key reason for the empirical disparity and resolving the apparent contradiction in the literature. Additionally, the inconsistency in the effects of trade openness may result from its varying impact depending on the types of goods traded, such as extractive resources compared to human-capital-intensive goods [29,30] vs. [31].

In addition to these transmission mechanisms, the literature highlights that the benefits of TOP and FDI are not automatic but are mediated by a country’s specific conditions, including financial inflows, macroeconomic stability, and institutional quality. Official Development Assistance (ODA) represents a critical source of external financing, yet its effectiveness remains debated. Proponents argue it stimulates growth by addressing capital scarcity [32], while critics contend it can foster dependency and displace domestic savings [33]. This ambiguity is reflected in Ethiopia, where findings report both positive [34,35] and negative effects [36]. Remittances constitute another significant financial inflow, often contributing to poverty reduction and consumption smoothing [37,38], while serving a critical lifeline in nations facing economic challenges [39].

Macroeconomic stability, particularly inflation control, is a critical precondition, as it erodes household purchasing power and public sector budgets, thereby directly curtailing resources for health and education [40,41]. Moreover, institutional quality fundamentally determines developmental outcomes. Strong institutional frameworks are vital for channeling the gains from globalization into productive areas and are essential for sustained growth [42], whereas political instability can disrupt economic activity and deter investment [43].

In the Ethiopian context, most existing studies have focused on the effects of foreign direct investment and trade on economic growth rather than on broader human development outcomes, and their findings are often contradictory. For instance, ref. [44] identified a positive effect of FDI on economic growth, while ref. [45] reported negative impacts. Similarly, ref. [46] found a positive relationship between trade openness and growth, whereas refs. [47,48] observed that trade openness worsened the balance of payments and constrained growth. This ambiguity highlights the necessity of examining broader development metrics in relation to the stated hypotheses [49], especially as some scholars contend that FDI may negatively affect development under certain conditions [50].

To our knowledge, no empirical study has systematically investigated the dual-path effects of both trade and FDI on human development outcomes in Ethiopia, especially using time-series methodologies capable of capturing temporal dynamics and long-term relationships. This gap is particularly significant given Ethiopia’s unique economic characteristics, including its substantial agricultural dependence, structural trade deficits, and ongoing institutional reforms. These features may determine whether the income-mediated or direct capability pathway dominates, making the Ethiopian case both theoretically and empirically critical for understanding how globalization impacts well-being in agrarian, structurally disadvantaged economies.

3. Methodology

3.1. Data Source and Variables

This study investigates the impact of trade openness and FDI on Ethiopian sustainable human development. It uses annual time-series data spanning the period 1991 to 2021. The year 1991 is significant because it marks Ethiopia’s economic and political transition after the fall of the Derg regime [51]. This period begins the nation’s market-oriented reforms and trade liberalization policies [52]. It also marks Ethiopia’s gradual integration into the global economy. Thus, this timeframe is optimal for analyzing the long-term impact.

Data were collected methodically from reputable international institutions for reliability and transparency. Primary sources include the World Bank’s World Development Indicators (WDI) for macroeconomic and sectoral data, the International Monetary Fund (IMF) databases for financial statistics, the United Nations Development Programme (UNDP) for human development indices, and the Worldwide Governance Indicators (WGI) [53] for institutional quality metrics. Additional national data were sourced from the National Bank of Ethiopia (NBE) to complement international sources. To minimize data variability and address heteroscedasticity, all variables, except one variable, undergo a logarithmic transformation prior to analysis. This enables the transformation of data toward enhanced linearity, thereby facilitating the interpretation of the regression results. Due to data availability constraints, interpolation was used to supplement missing values for some variables.

3.1.1. Measures of Variables

All variables represent Ethiopian national data. To provide a nuanced analysis, the research employs two distinct dependent variables and a set of control variables to isolate the effects of the primary variables of interest. The specific measurement and data sources for all variables are summarized in

Table 1. Variable Description and source.

Variable	Notation	Description	Source
Human Development Index	lnHDI	Log of standard HDI	UNDP
Modified HDI	lnHDI*	Log of HDI (health + education only)	Author’s Calculation
Trade Openness	lnTOP	Log of (Exports Imports)/% GDP	IMF, WDI
Foreign Direct Invest.	lnFDI	Log of Net FDI inflows (% of GDP)	WB, WDI
Economic Development	lnGDP	Log of real GDP per capita (constant 2015 $)	WB, WDI
Institutional Quality	IQ	Governance and Institutional Performance Index	WGI
Remittance	lnREM	Personal remittances received (% of GDP)	WB, WDI
Public spending	lnPS	Public spending (education + health sector) % of total budget	NBE, WB
Macroeconomic Stability	lnINF	Consumer Price Index annual %	IMF, WDI

.

Dependent Variable

The analysis uses two dependent variables, evaluated in separate models that share the same explanatory variables, with one control variable differing. The dependent variables are the standard Human Development Index (HDI) and a modified Human Development Index. Both will be measured as logarithms and denoted as lnHDI and lnHDI*, respectively.

• The natural logarithm of the standard Human Development Index (lnHDI): The HDI is a composite statistic of life expectancy, education, and per capita income indicators, serving as a comprehensive measure of socio-economic development as used in numerous prior studies [54,55]. It measures three main dimensions: longevity (life expectancy), educational achievement (mean and expected years of schooling), and a decent standard of living (GNI per capita).
• The natural logarithm of a modified Human Development Index (lnHDI*): This variable is constructed by recalculating the index using only the health (life expectancy) and education (mean and expected years of schooling) indices, while explicitly excluding the income component (Gross National Income, GNI). The recalculation follows the UNDP’s standard geometric mean formula. This formulation allows for the separation of direct income effects from health and education outcomes in human development.

Independent Variables

• The natural logarithm of Trade Openness (lnTOP): This is defined as lnTOP, where TOP equals the sum of exports and imports of goods and services as a percentage of GDP. This is a standard metric that reflects a nation’s degree of international trade integration [56].
• The natural logarithm of net Foreign Direct Investment (lnFDI): This is measured as net FDI inflows as a percentage of GDP. In previous studies, FDI has often served as an independent variable and is typically normalized by GDP to account for differences in country size, facilitating cross-country comparisons [28,57]. This measurement is particularly significant for developing countries, where FDI constitutes the primary source of external financing.

Control Variables

The model includes key control variables to consider other critical factors affecting human development.

• Personal Remittances (lnREM): Measured as personal remittances received as a percentage of GDP. Remittances are a crucial source of external finance that can help reduce poverty and promote socio-economic stability [58]. Inflation (lnINF): The annual inflation rate, measured by the annual percentage change in the Consumer Price Index. It serves as a proxy for macroeconomic instability [59,60]. Inflation is chosen over other indicators (e.g., fiscal balance, exchange rate volatility) for two reasons: its direct effect on purchasing power and social welfare, and consistent data availability. Institutional Quality Index (IQI): This is a composite measure of governance effectiveness. Reference [61] identified six key dimensions of governance. To avoid potential weighting biases and account for the intercorrelations among these dimensions, this study applied Principal Component Analysis (PCA) to create a single composite measure that reflects overall governance quality, rather than relying on a single indicator. Combined Public Spending (lnPS): This variable measures the combined public expenditure on education and health as a percentage of the total budget, capturing direct public investment in human capital. Lastly, GDP per capita (lnGDP). This variable is included only in the modified HDI (lnHDI*) model to control for the overall level of economic development, as the income component is excluded from the dependent variable [62].

3.2. Theoretical and Empirical Model Specification

3.2.1. Theoretical Dual Pathways Models

To provide a rigorous theoretical foundation for testing Hypotheses H1 and H2, we draw on two established traditions, endogenous growth theory and Sen’s capability approach, and translate them into a three-equation system that disentangles income-mediated and direct capability pathways.

Endogenous growth theory posits [63] that trade openness (TOP) and foreign direct investment (FDI) foster growth via market expansion, technology transfer, and productivity spillovers. In this view, globalization affects human development primarily through raising GDP per capita (the income-mediated channel). In contrast, Sen’s capability approach [64] emphasizes that TOP and FDI may also directly enhance health and education capabilities—for example, through imports of medicines or corporate training programs—independently of income growth.

1. Income equation (captures the income-mediated channel)

Yt = α₀ + α₁ TOPt + α₂FDIt + Zt′γ + εt

(1)

2. Direct non-income capability equation

HDI*t = β₀ + β₁TOPt + β₂FDIt + β₃Yt + Xt′δ + νt

(2)

3. Composite human development equation (standard HDI)

HDIt = f (HDI*t, Yt)

(3)

where Yt = real GDP per capita, HDI*t = modified Human Development Index excluding the income component (health + education only), Zt and Xt are vectors of control variables, εt and νt are error terms.

In Ethiopia’s structural context—characterized by heavy reliance on primary commodity exports and FDI concentrated in capital-intensive, enclave sectors—we expect the income-mediated channel to dominate: ∣α₁∣, ∣α₂∣ > ∣β₁∣, ∣β₂∣. This theoretical expectation forms the basis for H2, predicting that the total effect of globalization on standard HDI will be larger than its direct effect on non-income capabilities.

Empirical Strategy for Decomposition: To test these hypotheses, we implement a two-model estimation strategy. First, we estimate a reduced-form model for the composite HDI (Model A, corresponding to Equation (3)), which captures the total effect of TOP and FDI. Second, we estimate the structural model for the non-income HDI* (Model B, Equation (2)), which, by controlling for Yt, isolates the direct capability effect (β₁, β₂). The income-mediated effect for each globalization variable is then calculated as the difference between the total and direct effects. This decomposition provides a transparent, theory-consistent method to quantify the relative strength of the two pathways and directly test H1 and H2.

3.2.2. Model Specification and Estimation Strategy

This study employs the Autoregressive Distributed Lag (ARDL) bounds testing approach to cointegration [65]. The ARDL framework is preferred for several reasons pertinent to our analysis: it is applicable to variables with different orders of integration (I (0) or I (1)); it is robust in studies with relatively small sample sizes; and it provides efficient estimates of long-run relationships and short-run dynamics simultaneously, even in the presence of endogenous regressors [66].

To test our hypotheses, we estimate two distinct ARDL models derived from the conceptual framework in Section 3.2.1. This dual-model strategy is designed to disentangle the total effect of globalization from its direct effect on health and education capabilities.

Model A (Total Effect): This reduced-form model estimates the total long-run effect of trade openness and foreign direct investment on the standard Human Development Index (HDI), capturing all pathways, including those mediated by income.

HDI = f (TOP, FDI, REM, IQ, PS, INF)

Model B (Direct Effect): This model isolates the direct effect on non-income capabilities by employing a modified dependent variable (HDI*) and controlling for the level of economic development (GDP per capita).

HDI* = f (TOP, FDI, GDP, REM, IQ, PS, INF)

The general bounds testing form of the ARDL model for both specifications can be expressed as follows [65,67]:

Model A:

$$\begin{gathered} \Delta \mathrm{l}\mathrm{n}{\mathrm{H}\mathrm{D}\mathrm{I}}_t={\mathsf{\alpha }}_0+{\sum }_{\mathrm{i}=1}^{\mathrm{p}}{\mathsf{\delta }}_{\mathrm{i}}\Delta \mathrm{l}\mathrm{n}{\mathrm{H}\mathrm{D}\mathrm{I}}_{\mathrm{t}-1}+{\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_2}{\mathsf{\beta }}_{1,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}{\mathrm{T}\mathrm{O}\mathrm{P}}_{\mathrm{t}-1}+{\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_2}{\mathsf{\beta }}_{2,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}\mathrm{F}\mathrm{D}{\mathrm{I}}_{\mathrm{t}-1}+{\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_3}{\mathsf{\beta }}_{3,\mathrm{i}}\Delta {\mathrm{I}\mathrm{Q}}_{\mathrm{t}-1}+ \\ {\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_4}{\mathsf{\beta }}_{4,\mathrm{i}}\Delta {\mathrm{l}\mathrm{n}\mathrm{R}\mathrm{e}\mathrm{m}}_{\mathrm{t}-1}+{{\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_5}{\mathsf{\beta }}_{5,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}{\mathrm{P}\mathrm{S}}_{\mathrm{t}-1}+{\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_6}{\mathsf{\beta }}_{6,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}\mathrm{I}\mathrm{N}{\mathrm{F}}_{\mathrm{t}-1}+\mathsf{\lambda }}_1{\mathrm{l}\mathrm{n}\mathrm{H}\mathrm{D}\mathrm{I}}_{\mathrm{t}-1}+{\mathsf{\lambda }}_2\mathrm{l}\mathrm{n}\mathrm{T}\mathrm{O}{\mathrm{P}}_{\mathrm{t}-1} \\ +{\mathsf{\lambda }}_3\mathrm{l}\mathrm{n}\mathrm{F}\mathrm{D}{\mathrm{I}}_{\mathrm{t}-1}+{\mathsf{\lambda }}_4{\mathrm{I}\mathrm{Q}}_{\mathrm{t}-1}+{\mathsf{\lambda }}_5\mathrm{l}\mathrm{n}\mathrm{I}\mathrm{N}{\mathrm{F}}_{\mathrm{t}-1}+{\mathsf{\lambda }}_{6}ln{\mathrm{R}\mathrm{E}\mathrm{M}}_{\mathrm{t}-1}+{\mathsf{\lambda }}_7{\mathrm{l}\mathrm{n}\mathrm{P}\mathrm{S}}_{\mathrm{t}-1}+{\mathrm{U}}_{\mathrm{t}} \end{gathered}$$

(4)

Model B:

$$\begin{gathered} \Delta \mathrm{l}\mathrm{n}\mathrm{H}\mathrm{D}{\mathrm{I}}_{\mathrm{t}}^{*}={\mathsf{\alpha }}_0+{\sum }_{\mathrm{i}=1}^{\mathrm{p}}{\mathsf{\delta }}_{\mathrm{i}}\Delta \mathrm{l}\mathrm{n}\mathrm{H}\mathrm{D}{\mathrm{I}}_{t-1}^{*}+{\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_2}{\mathsf{\beta }}_{1,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}{\mathrm{T}\mathrm{O}\mathrm{P}}_{\mathrm{t}-1}+{\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_2}{\mathsf{\beta }}_{2,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}\mathrm{F}\mathrm{D}{\mathrm{I}}_{\mathrm{t}-1}+{\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_3}{\mathsf{\beta }}_{3,\mathrm{i}}\Delta {\mathrm{l}\mathrm{n}\mathrm{G}\mathrm{D}\mathrm{P}}_{\mathrm{t}-1}+ \\ {\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_4}{\mathsf{\beta }}_{4,\mathrm{i}}\Delta {\mathrm{I}\mathrm{Q}}_{\mathrm{t}-1}+{\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_5}{\mathsf{\beta }}_{5,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}{\mathrm{R}\mathrm{E}\mathrm{M}}_{\mathrm{t}-1}+{{\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_6}{\mathsf{\beta }}_{6,\mathrm{i}}\Delta {\mathrm{l}\mathrm{n}\mathrm{P}\mathrm{S}}_{\mathrm{t}-1}+{\sum }_{\mathrm{i}=0}^{{\mathrm{q}}_7}{\mathsf{\beta }}_{7,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}\mathrm{I}\mathrm{N}{\mathrm{F}}_{\mathrm{t}-1}+ \\ \mathsf{\lambda }}_1\mathrm{l}\mathrm{n}\mathrm{H}\mathrm{D}{\mathrm{I}}_{t-1}^{*}+{\mathsf{\lambda }}_2\mathrm{l}\mathrm{n}\mathrm{T}\mathrm{O}{\mathrm{P}}_{\mathrm{t}-1}+{\mathsf{\lambda }}_3\mathrm{l}\mathrm{n}\mathrm{F}\mathrm{D}{\mathrm{I}}_{\mathrm{t}-1}+{\mathsf{\lambda }}_4{\mathrm{l}\mathrm{n}\mathrm{G}\mathrm{D}\mathrm{P}}_{\mathrm{t}-1}+{\mathsf{\lambda }}_5{\mathrm{I}\mathrm{Q}}_{\mathrm{t}-1}+{\mathsf{\lambda }}_6{\mathrm{l}\mathrm{n}\mathrm{R}\mathrm{E}\mathrm{M}}_{\mathrm{t}-1} \\ +{\mathsf{\lambda }}_7{\mathrm{l}\mathrm{n}\mathrm{P}\mathrm{S}}_{\mathrm{t}-1}+{\mathsf{\lambda }}_8\mathrm{l}\mathrm{n}\mathrm{I}\mathrm{N}{\mathrm{F}}_{\mathrm{t}-1}+{\mathrm{U}}_{\mathrm{t}} \end{gathered}$$

(5)

The dependent variable is HDI for Model A and HDI* for Model B. α₀ is the constant, and Δ terms capture short-run dynamics. δ₁ and β₁, i denote short-run coefficients, while λ shows long-run multipliers. U_t is the error term, and the optimal lag (p, q_1, q₂ …) is set using the Akaike Information Criterion AIC.

Once cointegration is established (i.e., a long-run relationship exists), the long-run model is derived from the estimated coefficients in Equations (4) and (5), respectively:

$${\mathrm{lnHDI}}_{\mathrm{t}}={\mathsf{\theta }}_0+{\mathsf{\theta }}_1\mathrm{lnTOP}+{\mathsf{\theta }}_2\mathrm{lnFDI}+{\mathsf{\theta }}_3\mathrm{lnREM}+{\mathsf{\theta }}_4\mathrm{IQ}+{\mathsf{\theta }}_5\mathrm{lnPS}+{\mathsf{\theta }}_6\mathrm{lnINF}+{\mathsf{\epsilon }}_{\mathrm{t}}$$

(6)

$$ln\mathrm{H}\mathrm{D}{\mathrm{I}}_{\mathrm{t}}^{*}={\mathsf{\theta }}_0+{\mathsf{\theta }}_1\mathrm{lnTOP}+{\mathsf{\theta }}_2\mathrm{lnFDI}+{\mathsf{\theta }}_3\mathrm{lnGDP}+{\mathsf{\theta }}_4\mathrm{lnREM}+{\mathsf{\theta }}_5\mathrm{IQ}+{\mathsf{\theta }}_6\mathrm{lnPS}+{\mathsf{\theta }}_7\mathrm{lnINF}+{\mathsf{\epsilon }}_{\mathrm{t}}$$

(7)

Note: The term θ, long-run coefficients (θ), are calculated from the estimated λ coefficients.

The short-run dynamics for both models are estimated using their Error Correction Model (ECM) representations, specified in Equations (8) and (9) as:

Model A ECM:

$$\begin{array}{ll}\Delta {\mathrm{l}\mathrm{n}\mathrm{H}\mathrm{D}\mathrm{I}}_t=& {\mathsf{\alpha }}_0+\sum _{\mathrm{i}=1}^{\mathrm{p}}{\mathsf{\delta }}_{\mathrm{i}}\Delta {\mathrm{l}\mathrm{n}\mathrm{H}\mathrm{D}\mathrm{I}}_{\mathrm{t}-1}+\sum _{\mathrm{i}=0}^{{\mathrm{q}}_2}{\mathsf{\beta }}_{1,\mathrm{i}}\Delta {\mathrm{l}\mathrm{n}\mathrm{T}\mathrm{O}\mathrm{P}}_{\mathrm{t}-1}+\sum _{\mathrm{i}=0}^{{\mathrm{q}}_2}{\mathsf{\beta }}_{2,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}\mathrm{F}\mathrm{D}{\mathrm{I}}_{\mathrm{t}-1}+\sum _{\mathrm{i}=0}^{{\mathrm{q}}_3}{\mathsf{\beta }}_{3,\mathrm{i}}\Delta {\mathrm{I}\mathrm{Q}}_{\mathrm{t}-1}+\\ & \sum _{\mathrm{i}=0}^{{\mathrm{q}}_4}{\mathsf{\beta }}_{4,\mathrm{i}}\Delta {\mathrm{l}\mathrm{n}\mathrm{R}\mathrm{E}\mathrm{M}}_{\mathrm{t}-1}+\sum _{\mathrm{i}=0}^{{\mathrm{q}}_5}{\mathsf{\beta }}_{5,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}{\mathrm{P}\mathrm{S}}_{\mathrm{t}-1}+\sum _{\mathrm{i}=0}^{{\mathrm{q}}_6}{\mathsf{\beta }}_{6,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}\mathrm{I}\mathrm{N}{\mathrm{F}}_{\mathrm{t}-1}+\mathsf{\varphi }{\mathrm{E}\mathrm{C}\mathrm{M}}_{\mathrm{t}-1}+{\mathrm{U}}_{\mathrm{t}}\end{array}$$

(8)

Model B ECM:

$$\begin{array}{ll}\Delta \mathrm{l}\mathrm{n}\mathrm{H}\mathrm{D}{\mathrm{I}}_{\mathrm{t}}^{*}={\mathsf{\alpha }}_0+& \sum _{\mathrm{i}=1}^{\mathrm{p}}{\mathsf{\delta }}_{\mathrm{i}}\Delta \mathrm{l}\mathrm{n}\mathrm{H}\mathrm{D}{\mathrm{I}}_{t-1}^{*}+\sum _{\mathrm{i}=0}^{{\mathrm{q}}_2}{\mathsf{\beta }}_{1,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}{\mathrm{T}\mathrm{O}\mathrm{P}}_{\mathrm{t}-1}+\sum _{\mathrm{i}=0}^{{\mathrm{q}}_2}{\mathsf{\beta }}_{2,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}\mathrm{F}\mathrm{D}{\mathrm{I}}_{\mathrm{t}-1}+\sum _{\mathrm{i}=0}^{{\mathrm{q}}_3}{\mathsf{\beta }}_{3,\mathrm{i}}\Delta {\mathrm{l}\mathrm{n}\mathrm{G}\mathrm{D}\mathrm{P}}_{\mathrm{t}-1}+\\ & \sum _{\mathrm{i}=0}^{{\mathrm{q}}_4}{\mathsf{\beta }}_{4,\mathrm{i}}\Delta {\mathrm{I}\mathrm{Q}}_{\mathrm{t}-1}+\sum _{\mathrm{i}=0}^{{\mathrm{q}}_5}{\mathsf{\beta }}_{5,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}{\mathrm{R}\mathrm{E}\mathrm{M}}_{\mathrm{t}-1}+\sum _{\mathrm{i}=0}^{{\mathrm{q}}_6}{\mathsf{\beta }}_{6,\mathrm{i}}\Delta {\mathrm{l}\mathrm{n}\mathrm{P}\mathrm{S}}_{\mathrm{t}-1}+\sum _{\mathrm{i}=0}^{{\mathrm{q}}_7}{\mathsf{\beta }}_{7,\mathrm{i}}\Delta \mathrm{l}\mathrm{n}\mathrm{I}\mathrm{N}{\mathrm{F}}_{\mathrm{t}-1}\\ & +{\varphi \mathrm{E}\mathrm{C}\mathrm{M}}_{t-1}+{\mathrm{U}}_{\mathrm{t}}\end{array}$$

(9)

The term $\mathsf{\varphi }$ECM_t−1 is the lagged error correction term, which is derived from the residuals of the estimated long-run equation. ϕ is the speed of adjustment coefficient. It measures the speed of adjustment back to long-run equilibrium following a short-run shock and must be negative, statistically significant, and between 0 and −1 for the model to be valid.

3.3. Econometric Procedures and Diagnostic Testing

The empirical analysis followed a structured sequence of econometric procedures to ensure the robustness and validity of the results. First, the stationarity properties of the time series were ascertained using the Augmented Dickey–Fuller (ADF) unit root test. This test is crucial for selecting appropriate models, performing cointegration analysis, ensuring model validity, avoiding spurious regressions, and improving forecast accuracy in econometrics. The ADF test handles more complicated models compared to the Dickey–Fuller test, and it is more robust [68].

Secondly, determining the optimal lag length of models is crucial before running more complex econometric tools. This ensures that the models are effective and well-defined. The process involves selecting the number of past observations to include in the models, which affects model complexity, computational efficiency, overfitting prevention, and forecasting accuracy. There are various lag selection criteria available, but the most frequently used are AIC, SBC, and HQ. Previous studies, including [69,70], have primarily utilized AIC as their lag selection criterion. The optimal lag length for the ARDL model is selected using the Akaike Information Criterion (AIC) to adequately capture the data-generating process without overfitting.

Following this, the presence of a long-run cointegrating relationship was tested using the bounds test for the joint significance of the lagged level variables. Upon establishing cointegration, the long-run coefficients and the associated Error Correction Mode (ECM) were estimated. The ECM’s error correction term (ECT) was scrutinized; a negative and statistically significant coefficient is required to confirm a stable convergence towards the long-run equilibrium following short-run deviations.

Finally, to validate the reliability of the chosen model, a battery of diagnostic checks was conducted. This included testing for serial correlation using the Breusch-Godfrey LM test, heteroskedasticity using the Breusch-Pagan-Godfrey test, and functional form misspecification using the Ramsey RESET test. Furthermore, the stability of the model’s parameters over the sample period was verified recursively using the CUSUM and CUSUMSQ tests.

4. Results

This section presents the empirical findings on how trade openness and foreign direct investment affect human development in Ethiopia. The analysis begins with diagnostic tests to validate the model’s assumptions, followed by the presentation and interpretation of both long-run and short-run results.

4.1. Pre-Estimation Diagnostics

Before estimating the Autoregressive Distributed Lag (ARDL) model, several diagnostic tests were conducted to ensure the robustness and validity of later analyses. This preparatory step lays the foundation for accurate model estimation.

4.1.1. Unit Root Test

Prior to estimation, verifying the stationarity of the variables is essential to avoid spurious regression results. The Augmented Dickey–Fuller (ADF) test results (

Table 2. Results of the Augmented Dickey–Fuller (ADF) Unit Root Tests.

Variable	At Level, I (0)	At First Difference, I (1)	Order of
	ADF Statistic (p-Value)	ADF Statistic (p-Value)	Integration
lnHDI	−4.920 (0.023) **	-	1(0)
lnHDI*	−1.009 (0.736)	−2.834 (0.05) **	I (1)
lnTOP	−0.954 (0.767)	−4.874 (0.04) **	I (1)
lnFDI	−2.611 (0.020) **	-	I (0)
lnGDP	−1.312 (0.998)	−2.963 (0.048) **	I (1)
IQ	−2.002(0.027) **	-	I (0)
lnREM	−1.802 (0.704)	−2.027 (0.026) **	I (1)
lnPS	−1.579 (0.063)	−3.232 (0.002) **	I (1)
lnINF	−1.598 (0.481)	−7.253 (0.035) **	I (1)

Note: ** denotes significance at the 5% level. Tests include an intercept.

) indicate a mix of integration orders: lnHDI, IQ, and lnFDI are stationary at level I (0), while lnHDI*, lnTOP, lnGDP, lnREM, lnPS, and lnINF are stationary at first difference I (1). This mix of I (0) and I (1) variables justifies the use of the Autoregressive Distributed Lag (ARDL) bounds testing approach for cointegration analysis.

4.1.2. Lag Length Selection and Cointegration Test

The optimal lag length for the ARDL model was selected using the Akaike Information Criterion (AIC), which indicated a maximum lag of 4 for both models (see Appendix A and Appendix B for full lag selection criteria results). The presence of a long-run cointegrating relationship was then tested using the ARDL bounds testing procedure.

As presented in

Table 3. ARDL Bounds Test for Cointegration.

Models	F-Statics	Lower Bound 5%	Upper Bound (5%)	Conclusion
HDI	10.3	3.99	5.06	Long-run relation exists
HDI*	11.19	3.99	5.06	Long-run relation exists

Note: Case III (Unrestricted intercept and no trend).

, the computed F-statistics for both the standard HDI model (10.3) and the non-income HDI* model (11.19) exceed the upper-critical bounds at the 5% significance level. This provides conclusive evidence to reject the null hypothesis of no cointegration, confirming a stable long-run relationship among the variables in both model specifications.

4.2. Long-Run Estimation and Interpretation

The confirmed cointegration allows for the estimation of long-run coefficients (

Table 4. Estimated Long-Run Coefficients using the ARDL Approach.

Variables	Model A: lnHDI	Model B: lnHDI*
	Coefficient (p-Value)	Coefficient (p-Value)
lnTOP	0.343 (0.040) **	0.235 (0.007) ***
lnFDI	0.214 (0.008) ***	0.136 (0.001) ***
lnGDP	--------------	0.129 (0.000) ***
IQ	0.157 (0.002) **	0.014 (0.335)
lnREM	0.102 (0.027) **	0.012 (0.032) **
lnPS	0.026 (0.025) **	0.047 (0.295)
lnINF	−0.002 (0.426)	−0.018 (0.070) *
Constant	0.051 (0.000) ***	0.010 (0.034) **
R-squared	0.802	0.832
Adjusted R-squared	0.723	0.758

Note: *, **, *** denote significance at the 10%, 5%, and 1% levels, respectively.

). The results strongly support H1 and H2.

Confirming our first hypothesis, both trade openness (TOP) and foreign direct investment (FDI) exhibit positive and statistically significant long-run effects on the standard Human Development Index (HDI). Specifically, a 1% increase in TOP leads to a 0.343% increase in HDI—a result that aligns with conventional growth theories and empirical literature [69], affirming that Ethiopia’s trade policies have contributed positively to aggregate outcomes through economic growth. Similarly, a 1% increase in FDI leads to a 0.214% increase, corroborating studies that identify FDI as a driver of aggregate development [70,71]. In particular, FDI-driven knowledge spillovers and capital formation are tangible drivers of human development in the Ethiopian context [72].

The central contribution of our study is the direct test of Hypothesis 2 through our dual-model framework. The key finding is that the coefficients for both TOP and FDI are significantly larger in the full HDI model (Model A) than in the non-income HDI model (Model B). For instance, TOP’s effect decreases from 0.343% on the full HDI to 0.235% on HDI*. Similarly, FDI’s effect drops from 0.214% to 0.136%.

This consistent differential allows for a precise decomposition of total effects into two distinct pathways (see

Table 5. Decomposition of Total Effects into Direct and Income-Mediated.

Variables	HDI	HDI*	Income	% of Total
	(Model A)	(Model B)	Mediated	Effect via Income
lnTOP	0.343	0.235	0.108	31.50%
lnFDI	0.214	0.136	0.078	36.40%

Note: The Income-Mediated Effect is calculated as (Coefficient in Model A) − (Coefficient in Model B). The percentage is calculated as (Income-Mediated Effect/Total Effect) * 100.

). We calculate that approximately 31.5% of TOP’s total impact and 36.4% of FDI’s total impact on human development are mediated through income growth. This quantification empirically validates our theoretical proposition that the income-mediated channel is the dominant, though not exclusive, transmission mechanism in Ethiopia. The significant, yet smaller, coefficients in the HDI* model simultaneously confirm the existence of a secondary direct capability channel, consistent with studies highlighting this mechanism [73,74] and FDI-related knowledge spillovers [75].

Control variables show that institutional quality (IQ) positively and significantly affects the full HDI (0.157) but is not significant for the non-income HDI*. This means that better institutions foster economic growth and overall human development, but may not directly improve health and education. Remittances (REM) have positive and significant effects in both models, but their impact is larger on the full HDI (0.102 vs. 0.012). This suggests remittances mainly contribute to economic development. Public spending (PS) is significant in the full HDI model (0.026), but not in the non-income HDI* model, indicating its impact works mostly through income. As expected, real GDP per capita has a positive and highly significant coefficient in Model B. This shows that economic growth is key to generating resources for health, education, and welfare. This matches the findings of [62,76]. Inflation (INF) has a nuanced effect. It is negative but statistically insignificant for the full HDI. For the non-income HDI*, it is negative and significant at the 10% level. This means that, while rising incomes may offset inflation in the long run, inflation directly harms health and education quality. This offers evidence for a key concern in development economics—macroeconomic instability erodes health and education.

4.3. Short-Run Dynamics and Error Correction

Table 6. Short-Run Error Correction Model (ECM) Results.

Variables	Model A: lnHDI	Model B: lnHDI*
	Coefficient (p-Value)	Coefficient (p-Value)
ΔlnHDI (−1)	0.232 (0.047) **	-
ΔlnHDI* (−1)	-	0.157 (0.002) ***
ΔlnTOP	0.116 (0.000) ***	0.015 (0.004) ***
ΔlnTOP (−1)	0.05 (0.242)	0.013 (0.195)
ΔlnFDI	0.084 (0.044) **	0.035 (0.057) *
ΔlnGDP	-	0.128 (0.032) **
ΔIQ	0.058 (0.046) **	0.014 (0.335)
ΔlnREM	0.073 (0.409)	0.026 (0.186)
ΔlnREM (−1)	0.013 (0.642)	0.003 (0.158)
ΔPS	0.018 (0.028) **	0.012 (0.032) **
ΔINF	−0.042 (0.022) **	−0.011 (0.027) **
ΔINF (−1)	−0.0012 (0.720)	−0.042 (0.022) **
EC (−1)	−0.284 (0.005) ***	−0.178 (0.015) ***

Note: *, **, *** denote significance at the 10%, 5%, and 1% levels, respectively.

presents short-run dynamics and error correction results. The error correction term (ECT), which shows how quickly variables return to equilibrium after a shock, is negative and significant at the 1% level in both models. This confirms the long-run cointegrating relationship and shows that deviations from equilibrium are corrected over time, restoring balance.

Error correction terms of −0.284 and −0.178 show that 28% and 18% of disequilibrium are corrected annually for Models A and B. This indicates that shocks or imbalances in these models are partially self-corrected each year, reflecting how quickly the system moves back toward equilibrium after a disturbance. The faster adjustment in the full HDI model suggests income components in this framework respond more rapidly to such shocks, helping to restore long-run equilibrium sooner. Furthermore, Significant coefficients on lagged dependent variables (ΔlnHDI (−1) and ΔlnHDI* (−1)) show persistence in human development, indicating improvements in one period drive positive momentum in the next.

In the short run, both trade openness and foreign direct investment (FDI) are key catalysts for human development in Ethiopia, but their impacts differ. The effect of trade (ΔlnTOP = 0.116) is much greater than FDI (ΔlnFDI = 0.084) on the standard Human Development Index (HDI), providing a stronger immediate, mainly income-driven boost. In comparison, FDI’s impact is smaller but more balanced across non-income factors.

Among the control variables, changes in institutional quality (ΔIQ) have a positive and significant short-run effect on the Human Development Index (HDI), but not on the adjusted Human Development Index (HDI*). In contrast, remittances (lnREM) do not show significant short-run effects in either model. Regarding public spending, changes in (ΔlnPS) have a positive and significant short-run effect in both models, with coefficients of 0.018 for HDI and 0.012 for HDI*. Thus, higher public spending can rapidly improve human development, especially in non-income areas. However, some short-run dynamics are not visible in the long run. Changes in inflation (ΔINF), for instance, have a significant negative impact on both models. Rising prices immediately threaten human welfare by limiting access to essentials, although this effect may be offset in the long run. Maintaining macroeconomic stability is therefore critical for short-term well-being. Finally, changes in economic growth (ΔlnGDP) in Model B have a positive short-run impact, as immediate income gains directly boost human development before the long-run equilibrium returns.

5. Model Validation and Robustness Checks

A suite of diagnostic tests was employed to validate the models’ specifications. The results, summarized in

Table 7. Diagnostic Test Results.

Test	HDI Model	HDI* Model
	Statistic (p-Value)	Statistic (p-Value)	Null Hypothesis	Conclusion
Breusch Godfrey LM	4.879 (0.163)	1.930 (0.207)	No serial correlation	Not rejected
Breusch–Pagan	3.970 (0.137)	1.088 (0.580)	Homoscedasticity	Not rejected
Jarque–Bera	1.695 (0.429)	1.228 (0.541)	Normal distribution	Not rejected
Ramsey RESET	0.470 (0.708)	0.142 (0.931)	Correct functional form	Not rejected
DW-statistic	2.342	2.184	No serial correlation	Not rejected

Note: calculated using Stata version 17.

, indicate that the models are well-specified. The Breusch-Godfrey test fails to reject the null hypothesis of no serial correlation. The Durbin-Watson d-statistic further supports this result. The Breusch-Pagan test indicates homoscedasticity in the residuals. Furthermore, the Jarque–Bera test does not reject the null hypothesis of normally distributed residuals.

Model Stability Tests

The parameters remain stable across time, indicating that our models capture consistent relationships. This stability, verified by the CUSUM and CUSUM of Squares (CUSUMSQ) tests shown in Figure 2 for both models, is crucial for reliable policy inference. The CUSUM test detects systematic changes in regression coefficients, while the CUSUMSQ test identifies abrupt, large shifts in the variance of the regression errors, reflecting sudden changes in model parameters.

The graphical results confirm that all test plots remain consistently within the 5% significance boundaries (dashed red lines). For Model A, both the CUSUM statistic (Figure 2a) and the CUSUMSQ statistic (Figure 2b) demonstrate stability, indicating no progressive, lasting shocks affected the long-run relationships between variables. Similarly, for Model B, the CUSUM (Figure 2c) and CUSUMSQ (Figure 2d) plots staying within their bounds indicate the absence of any major, short-term shocks. Therefore, the collective evidence from both tests strongly suggests that the parameters of both models are constant over the sample period.

6. Concluding Remarks

This study examined the impact of trade openness (TOP) and foreign direct investment (FDI) on sustainable human development in Ethiopia from 1991 to 2021. Using an autoregressive distributed lag (ARDL) dual-model framework, we distinguished total effects on the standard Human Development Index (HDI) from direct effects on a modified non-income HDI* (comprising only health and education components). The results confirm stable long-run cointegration and positive, significant effects of both TOP and FDI on the standard HDI, supporting Hypothesis 1. Crucially, the coefficients are consistently larger for the standard HDI than for HDI*, with the income-mediated channel accounting for approximately one-third of the total effect (31.5% for TOP and 36.4% for FDI; Table 5). This decomposition validates Hypothesis 2 and confirms the income-mediated pathway as the dominant, though not exclusive, transmission mechanism in Ethiopia.

These findings, interpreted through our dual-pathway conceptual framework (Figure 1), provide quantitative insight into Ethiopia’s “growth without development” paradox. Globalization has served as a potent driver of income generation but a weaker instrument for direct capability expansion in health and education. This asymmetry reflects the structural characteristics of Ethiopia’s global integration: trade concentrated in primary commodities (e.g., coffee, oilseeds) with limited linkages to social sectors [77], and FDI focused on capital-intensive enclave sectors (e.g., construction and large-scale agriculture) that boost GDP but generate few direct knowledge spillovers or social infrastructure improvements [78].

The differential impacts of control variables across models further illuminate these structural constraints. Institutional quality significantly enhances the standard HDI but not HDI*, indicating that governance improvements have primarily fostered growth-enabling environments (e.g., attracting investment and enforcing contracts) rather than directly improving the efficiency of health and education service delivery [79]. This pattern aligns with Ethiopia’s low economic freedom scores—particularly in judicial effectiveness, property rights, financial freedom, and investment freedom [80]—and its state-led development model, which has historically prioritized industrial policy and infrastructure over social sector reform [73].

Similarly, remittances and public spending show stronger associations with the income-inclusive HDI, reflecting their primary roles in consumption smoothing, small-scale investment, and growth-oriented projects. Inflation exerts a marginally significant negative effect on HDI*, highlighting the vulnerability of non-income capabilities to macroeconomic instability, as it erodes household purchasing power and constrains public social budgets even when nominal incomes rise.

Although the direct capability channel is secondary, it remains positive and statistically significant, indicating that meaningful spillovers from trade and FDI do occur. Collectively, these results resolve the Ethiopian paradox: the shortfall in human development is not due to insufficient growth from globalization but to the asymmetric strength of transmission pathways, conditioned by domestic structures and policies. The framework also reconciles contradictory findings in the literature, aligning with studies reporting positive globalization effects on composite HDI [70,81], while explaining weaker or null impacts on isolated social outcomes [82,83] or in high-inflation contexts [31,84].

We conclude that the Ethiopian development paradox—rapid growth without commensurate improvements in well-being—stems from the nature of its global integration. The gains from trade and FDI have been channeled predominantly through economic growth but have not robustly translated into direct enhancements of health and education capabilities. Therefore, policy must evolve from a framework of generic openness to one of strategic integration that specifically fosters direct capability expansion.

6.1. Policy Recommendations

To maximize the positive impact of economic integration on sustainable human development, Ethiopian policymakers should prioritize the following actions:

Firstly, actively pursue trade and investment policies that are specifically designed to generate social benefits. This includes incentivizing FDI in sectors like healthcare technology, pharmaceutical manufacturing, and educational services, and moving beyond simple trade liberalization to negotiate trade agreements that facilitate the import of essential medical and educational technology while supporting value-added exports that create higher-skill jobs.

Secondly, institutional reforms should be strategically targeted to translate economic growth into direct human development gains. Building on our finding that institutional quality boosts aggregate growth more than direct social outcomes, policy must focus on specific regulatory improvements, including enhancing labor market flexibility through measured deregulation to facilitate hiring in FDI-linked sectors and improving monetary policy frameworks to ensure stability. These targeted reforms in economic freedom will help channel the income gains from globalization into more direct job creation and human capital development. Concurrently, continued efforts in strengthening overall governance, reducing corruption, and improving public service efficiency remain fundamental to enhancing the developmental impact of globalization.

Thirdly, remittances should be leveraged as a source of development finance. Policies that reduce the cost of remittance transfers and channel them into productive investments in health and education can amplify their positive effects.

Furthermore, maintaining macroeconomic stability is paramount. The significant negative effect of inflation on non-income sustainable human development underscores the critical need for sound monetary and fiscal policies to control price volatility, thereby protecting household purchasing power and public sector investment in health and education.

Finally, enhancing data collection and conducting further research are essential. The government should prioritize the regular compilation of disaggregated social data at regional levels.

6.2. Limitations and Avenues for Future Research

This study offers important insights into Ethiopia’s development paradox, yet its findings must be considered within the context of several inherent limitations, each of which opens a clear pathway for further investigation. A primary constraint arises from the use of aggregated national-level data, which, while effective for identifying country-wide trends, may conceal significant regional disparities in how trade and foreign investment affect local development. Consequently, future research employing sub-national or provincial data could powerfully illuminate these spatial heterogeneities.

Furthermore, although the ARDL model establishes strong long-run associations, the analysis cannot definitively rule out issues of reverse causality. To strengthen causal claims, subsequent studies would benefit from applying specific causality tests, such as the Granger causality test within a Vector Error Correction Model framework.

Another set of limitations stems from the variables included in the model. While institutional quality was controlled for, broader dimensions of economic freedom—such as labor market flexibility or monetary policy—were not incorporated; these omissions may lead to an understatement of the barriers that weaken the direct translation of economic gains into social benefits. Extending the model to include such moderators could test the sensitivity of our core findings. Equally, the study’s focus on socio-economic development metrics means environmental sustainability indicators are absent, which limits a full assessment of sustainable development within Ethiopia’s climate-vulnerable context. Future work that integrates metrics related to the environmental impact of investment and trade would provide a more holistic view aligned with the Sustainable Development Goals.

Finally, the analysis treats trade openness and foreign direct investment as aggregate variables. A more granular, sectoral analysis examining, for instance, differences between primary commodity exports and manufactured goods or between capital-intensive and knowledge-intensive investment could reveal the structural drivers behind the dominant income-mediated pathway.

In summary, while this research disentangles key channels of impact for Ethiopia, exploring these avenues—regional disaggregation, enhanced causal inference, incorporation of omitted economic and environmental variables, and sectoral analysis—would substantially deepen and broaden our understanding of how global integration shapes sustainable human development.