Machine Learning the Future of Inclusive and Sustainable Growth: The Role of Entrepreneurial Ecosystems

Abstract

Entrepreneurial ecosystems are increasingly viewed as drivers of sustainable and inclusive development, yet their true capacity to translate entrepreneurship into inclusive outcomes remains insufficiently understood. This study addresses this gap by linking national-level entrepreneurship indicators to three dimensions of inclusive growth—decent work and economic growth (SDG 8), innovation and infrastructure (SDG 9), and inequality reduction (SDG 10)—using data from the Global Entrepreneurship Monitor’s National Expert Survey (NES) for 37 countries over the period 2020 to 2024. By integrating multistage machine learning techniques with institutional and ecosystem theories, the study captures the nonlinear and interdependent dynamics among entrepreneurial framework conditions. The results reveal that development outcomes depend not on isolated factors but on complementarities among finance, infrastructure, and R&D transfer. Entrepreneurial ecosystems foster innovation and productive employment, yet their inclusiveness hinges on institutional efficiency, governance quality, and redistributive capacity. The findings show that policy intent diverges from policy impact, as governmental support and cultural norms exert adverse effects when institutional coherence and absorptive capacity are weak. The research advances entrepreneurship theory by moving beyond linear, additive models toward a systemic understanding of ecosystem complementarity, and offers policy insights, emphasizing that innovation-led growth must be embedded in institutional coherence and social inclusion to achieve sustainable and equitable development.

1. Introduction

In an era defined by accelerating technological change and widening inequality, the question is no longer whether entrepreneurship drives growth, but what kind of growth it creates, and for whom. While entrepreneurship is often celebrated as a universal engine of progress, its developmental outcomes remain far from uniform. Some nations harness entrepreneurial dynamism to expand decent work and technological capacity, while others experience rising inequality and fragile employment despite high start-up activity. Understanding this divergence requires moving beyond aggregate growth narratives toward a systemic examination of how national entrepreneurial ecosystems shape the quality, inclusiveness, and sustainability of development.

Most existing studies linking entrepreneurship and sustainable development adopt either an aggregate perspective or a growth-centred focus. On the one hand, sustainability research often relies on broad composite indices, such as the SDG Index, which collapse multiple goals into a single score, obscuring how entrepreneurial ecosystems may affect specific outcomes like employment, innovation, or inequality differently [1,2]. On the other hand, much of the entrepreneurship literature equates development with economic growth alone, emphasizing GDP expansion or firm creation while sidelining crucial equity and innovation dimensions [3]. This reliance on macroeconomic growth metrics risks painting an incomplete picture; for example, entrepreneurship may drive GDP growth yet simultaneously exacerbate inequality or fail to foster decent and inclusive employment.

This dual tendency limits our understanding of how entrepreneurial ecosystem conditions may differentially shape the multi-faceted landscape of inclusive development, which spans decent work and economic growth (SDG 8), innovation and infrastructure (SDG 9), and inequality reduction (SDG 10). Addressing this gap requires a disaggregated approach that links specific ecosystem pillars to distinct developmental outcomes. By focusing on the Global Entrepreneurship Monitor’s National Expert Survey (NES), this study applies an ecosystem lens that captures the institutional, financial, educational, and infrastructural dimensions underpinning entrepreneurial activity. Such granularity allows for a more precise mapping of how systemic conditions translate into social and economic progress. In this study, inclusive growth is operationalized as a development process that combines productive employment creation, innovation-driven structural transformation, and equitable distribution of economic gains. Thus, SDG 8, SDG 9, and SDG 10 are selected as complementary and policy-relevant dimensions that capture opportunity, resilience, and equity most directly linked to entrepreneurial ecosystem dynamics.

Accordingly, we investigate three central questions:

RQ1. To what extent do entrepreneurial ecosystem conditions predict cross-country outcomes in inclusive growth (SDGs 8, 9, and 10)?

RQ2. How are inclusive growth outcomes projected to evolve under alternative entrepreneurial policy scenarios, and what insights can machine-learning–based forecasting provide for future growth pathways?

RQ3. Which entrepreneurial ecosystem conditions exert the strongest causal effects on inclusive growth, and how can causal machine-learning methods identify the most effective policy levers?

Traditional econometric approaches are limited in addressing these questions because entrepreneurial factors are inherently interdependent [4]. High levels of multicollinearity inflate standard errors and conceal the systemic nature of entrepreneurial ecosystems, where policy, finance, education, and culture co-evolve. Indeed, the 12 NES indicators exhibit strong pairwise correlations, often exceeding 0.7 (see Figure 1), confirming that entrepreneurial environments function as tightly coupled systems rather than isolated determinants. Machine learning (ML) offers a methodological solution to this complexity [5].Techniques such as the Elastic Net mitigate multicollinearity through regularization, while XGBoost models non-linearities, thresholds, and cross-pillar interactions [6]. Together, these models reveal how ecosystem configurations shape inclusive growth across the employment, innovation, and inequality dimensions. By complementing predictive modelling with causal machine-learning methods such as DoubleML and SHAP analysis, the study moves beyond traditional linear inference to identify policy-relevant, causally robust levers of inclusive growth.

In doing so, this research bridges a crucial gap in the literature on entrepreneurship and development. It reframes entrepreneurship not as a source of aggregate economic expansion but as a systemic and context-dependent driver of inclusive, innovation-led, and equitable development. Accordingly, machine learning is employed not merely as a predictive tool, but as a theory-consistent method for modeling the nonlinear and interdependent mechanisms emphasized in entrepreneurial ecosystem and endogenous growth theories. The structure of this study is as follows. Section 2 reviews the theoretical and empirical literature linking entrepreneurship, ecosystem dynamics, and inclusive development, highlighting methodological gaps that motivate ML approach. Section 3 describes the data, variables, and analytical design. Section 4 presents the empirical results, and Section 5 discusses their theoretical interpretation. Section 6 concludes the study, while Section 7 outlines the policy implications and directions for future research.

2. Literature Review

2.1. Entrepreneurship and Development

The intellectual foundation linking entrepreneurship to national economic development lies in the Schumpeterian concept of “creative destruction,” which positions innovation as the essential force driving productivity and long-term economic dynamism. This idea was formalized within Endogenous Growth Theory in the 1980s, primarily through the work of [7,8]. These models demonstrated that sustained growth is driven by endogenous factors, specifically, investment in human capital, knowledge accumulation, and R&D, rather than exogenous variables. Theorists of this movement argued that R&D and human capital investments serve as sustained growth engines, thereby positioning the innovative firm as the central agent converting knowledge into economic value. Crucially, endogenous growth theory establishes that long-run economic expansion is highly dependent on policy measures that influence the incentive structure for innovation. This theoretical mandate validates the focus on policy-quality metrics as the appropriate set of independent variables for evaluating national development outcomes.

Following the endogenous growth models, subsequent theoretical development, rooted in institutional theory [9,10], focused on the critical role of context in determining whether entrepreneurial action is productive or unproductive. This shift culminated in the entrepreneurial ecosystem framework. An entrepreneurial ecosystem is defined as “a set of interdependent actors and factors coordinated in such a way that they enable productive entrepreneurship within a particular territory” [11]. The lineage of this ecosystem thinking can be traced through successive generations of entrepreneurship scholarship [12], reflecting its evolution from individual-centred theories to system-based models of entrepreneurial dynamics. The Entrepreneurial Ecosystem is characterized by its relational dimension, meaning entrepreneurship takes place within a community where actors interact to access resources, knowledge, and infrastructure [13]. The system encompasses several core entrepreneurial framework conditions (EFCs), including finance, policy, education, R&D, infrastructure, and culture. Since the effect of any single component is contingent upon the quality of the system’s other pillars, the entrepreneurial ecosystem framework requires an analytical approach capable of modelling systemic interdependence and complex interactions.

Traditional entrepreneurship literature has primarily adopted a classical growth paradigm, equating development solely with macroeconomic metrics like GDP expansion. However, this approach is critically limited, as rapid aggregate economic growth often exacerbates inequality, concentrating benefits and relying on ex-post redistribution rather than structural inclusion [14]. The dominance of GDP metrics is insufficient to provide sufficient guidance for building equitable societies. This recognition has necessitated a paradigm shift toward inclusive development, which recognizes that development must be defined by equitable distribution, resilience, and the broad provision of opportunities for all populations.

The conceptual integration requires viewing entrepreneurship as a direct instrument for social and financial inclusion. Inclusive entrepreneurship aims to ensure that all citizens, regardless of gender, age, or socioeconomic status, have an equal opportunity to start and scale a successful business [15]. This mandate is particularly relevant for marginalized groups, such as immigrant women, who face multifaceted challenges requiring strategic policy shifts toward inclusive strategies. By adopting a disaggregated approach focusing on specific SDG outcomes, the research explicitly examines the tension within the development literature, whether ecosystem factors that maximize innovation (SDG 9) can simultaneously mitigate inequality (SDG 10). This underscores the necessary conceptual shift from “entrepreneurship for growth” to “entrepreneurship for inclusion.”

2.2. Ecosystem Dynamics and Inclusive Growth

The entrepreneurial ecosystem framework provides the analytical foundation for understanding how the institutional and contextual environments shape entrepreneurial outcomes and, in turn, national development trajectories. It conceives entrepreneurship as an embedded process within a system of interdependent institutions, networks, and resources that collectively determine the conditions for business creation and growth [11,16,17]. Within this framework, the EFCs represent the critical enablers of entrepreneurial performance, including finance, government policy, education, infrastructure, R&D transfer, and cultural support. In addition, national culture exerts a significant influence on entrepreneurial behaviour and development outcomes, shaping risk tolerance, innovation, and opportunity perception in systematic ways [18]. The interaction among these pillars forms a complex, adaptive system in which policy interventions often yield non-linear and conditional effects, depending on the balance and maturity of other components.

The NES operationalizes this framework across countries by systematically capturing expert evaluations of national entrepreneurial environments. It offers a harmonized dataset on the institutional, financial, and infrastructural factors that enhance or constrain entrepreneurial activity. However, the NES’s richness comes with analytical challenges. High multicollinearity among ecosystem pillars (indicators) inflates standard errors, making it difficult to disentangle their distinct contributions [19,20,21]. Policymakers and entrepreneurs, for instance, often diverge markedly in their evaluations of ecosystem quality. Such dependencies and biases reduce the reliability of traditional regression-based inference, reinforcing the need for more robust machine-learning and causal inference techniques that can model interaction effects, handle complex data structures, and uncover high-impact levers of inclusive growth [22].

Anchored in this framework, the present study links entrepreneurial ecosystem conditions to the multi-dimensional concept of inclusive growth, as reflected in three interrelated Sustainable Development Goals (SDGs 8, 9, and 10). Together, these goals capture the economic, technological, and social dimensions of sustainable progress. Entrepreneurial ecosystems advance SDG 8 by fostering productive employment and business dynamism, yet the quality of work generated depends on institutional safeguards and human-capital alignment. Moreover, innovation-driven entrepreneurship may boost output but also displace labour if labour policies lag behind technological change [14]. The same ecosystems promote SDG 9 by enabling R&D transfer, infrastructure development, and technological upgrading, where finance, knowledge diffusion, and absorptive capacity collectively underpin industrial resilience and innovation capability. However, these advances do not automatically translate into the equitable outcomes envisioned in SDG 10. Without fair access to resources, inclusive financing, and institutional trust, entrepreneurship can reinforce structural inequalities and spatial divides rather than alleviate them [23,24]. Hence, while entrepreneurship can act as a bridge between innovation and social progress, the net inclusiveness of its impact depends on whether ecosystem mechanisms operate as open systems or instead privilege established actors.

In sum, the entrepreneurial ecosystem framework offers a comprehensive lens for understanding how interdependent national conditions influence distinct dimensions of inclusive development. Its implementation through NES data highlights both the promise and the complexity of entrepreneurship as a vehicle for sustainable progress, thereby justifying the adoption of advanced empirical approaches capable of capturing systemic interdependence, non-linearity, and causal complexity.

2.3. Theoretical Alignment Between Entrepreneurial Ecosystems, Endogenous Growth, and Machine Learning

Entrepreneurial ecosystem theory conceptualizes entrepreneurship as a systemic and interdependent process, where outcomes emerge from the interaction of multiple institutional, financial, and knowledge-based conditions rather than from isolated inputs [11,17]. This systemic logic closely aligns with endogenous growth theory, which emphasizes that sustained development arises from complementarities between human capital, R&D, finance, and institutional incentives [7,8]. Both traditions therefore imply nonlinear and conditional relationships, where the marginal effect of any single factor depends on the configuration of the wider system.

Machine learning methods are theoretically well suited to this ecosystem-based view of development. Unlike linear econometric models, nonlinear algorithms such as XGBoost explicitly accommodate interaction effects, thresholds, and complementarities among predictors, thereby operationalizing the core propositions of ecosystem and endogenous growth theories. In this study, machine learning is not used as a purely technical forecasting tool, but as a theory-consistent analytical framework that allows development outcomes to emerge endogenously from complex ecosystem configurations.

From a theoretical standpoint, the key entrepreneurial ecosystem pillars examined in this study map directly onto endogenous growth mechanisms. Financing for entrepreneurs relaxes credit constraints and enables risk-taking and innovation, translating knowledge into productive investment. R&D transfer and infrastructure enhance knowledge diffusion and absorptive capacity, accelerating innovation and structural transformation (SDG 9). Physical and services infrastructure lowers transaction costs and facilitates firm scaling, supporting productivity and employment (SDG 8). By contrast, governmental support and cultural norms operate primarily as institutional moderators rather than direct growth inputs; their effectiveness depends on governance quality, implementation capacity, and the equitable distribution of opportunity [9,18,23].

The study’s multi-stage machine learning design mirrors this theoretical structure. Predictive models capture the joint explanatory power of ecosystem configurations (RQ1), forecasting models simulate dynamic trajectories under alternative policy scenarios (RQ2), and causal machine learning isolates the net structural effects of individual pillars within the system (RQ3). In doing so, the empirical strategy directly reflects the theoretical premise that inclusive development is a system-level outcome shaped by interacting entrepreneurial conditions rather than additive policy levers. See Figure 2 that iilustrates the conceptual model of this study.

2.4. Methodological Gaps and the Turn to Machine Learning

Entrepreneurial ecosystem theory conceptualizes entrepreneurship as an emergent outcome of interacting institutional, financial, and knowledge-based conditions rather than as the result of isolated inputs. Central to this framework are systemic interdependence, complementarities, and nonlinear effects across ecosystem pillars [11,17]. However, the methodological inability of prior entrepreneurial ecosystem research to isolate these effects can be traced to its reliance on traditional linear approaches, which are ill-equipped to handle the high-dimensional complexity, nonlinearity, and multicollinearity inherent in ecosystem analysis [25]. As a result, much empirical work has remained misaligned with the theoretical premises of ecosystem research, limiting its ability to capture joint and conditional dynamics across ecosystem dimensions.

Machine-learning methods provide a theory-consistent empirical response to these challenges. Tree-based ensemble models, such as eXtreme Gradient Boosting (XGBoost) and Random Forest, are particularly suited to modelling threshold effects, interactions, and nonlinear relationships among complex systems, yielding superior predictive performance relative to additive models [4,26,27]. Model-agnostic interpretability tools, such as SHAP, further translate complex model structures into theoretically meaningful contributions by quantifying how individual ecosystem indicators shape specific SDG outcomes under different configurations. In the causal domain, machine-learning techniques grounded in the Neyman orthogonality principle—such as Double Machine Learning—enable the identification of structural effects by flexibly adjusting for high-dimensional confounding while preserving valid inference [28]. Together, these methods operationalize entrepreneurial ecosystems as complex, interdependent systems, aligning empirical estimation with endogenous growth and institutional theories that emphasize policy effectiveness, implementation capacity, and systemic efficiency.

2.5. Integrative Framework and Research Gap

The review demonstrates a clear conceptual pathway from entrepreneurial ecosystems to multi-dimensional development. However, empirical literature has failed to provide the necessary granular details due to reliance on aggregated outcomes and methodologically inadequate tools. The crucial justification for this study’s advanced methodology lies in overcoming the empirical limitations of prior studies, which could not identify specific policy levers due to pervasive multicollinearity and subjective data [21].

The central research gap is the absence of empirical evidence that rigorously tests the differential causal impact of specific, collinear entrepreneurial ecosystem components on the disaggregated, multi-dimensional outcomes of inclusive development (SDGs 8, 9, and 10). Previous work has struggled to provide actionable policy insights because the aggregation of outcomes obscures the specific pathways leading to growth versus equity, and methodological instability prevents the confident isolation of independent policy levers.

This study’s methodology is designed to overcome these constraints. By providing a comprehensive pipeline to both forecast system behaviour under hypothetical policy changes (RQ2) and identifying the strongest, debiased causal drivers (RQ3), the research offers a theoretically grounded and empirically robust analysis that can specify which ecosystem pillars must be adjusted to optimize particular, often conflicting, dimensions of inclusive development.

3. Methodology

3.1. Data Preprocessing

The entrepreneurial ecosystem conditions provided by NES capture expert assessments of twelve Entrepreneurial Framework Conditions (EFCs). Each NES indicator is originally measured on a standardized 1–9 Likert scale, where higher values indicate more favourable entrepreneurial conditions. To enhance interpretability and ensure consistency with the SDG outcome measures, all NES indicators are linearly rescaled to a 10–100 index prior to analysis, preserving relative cross-country differences while maintaining the original ordering of scores. In addition, the inclusive and sustainable development outcomes are measured using country-level SDG indices for SDG 8, SDG, and SDG 10. Each SDG outcome is defined as the composite index score reported on a 0–100 scale. These indices are designed for cross-country comparability and are used directly in both predictive and forecasting analyses.

The dataset contains a limited number of missing observations, amounting to ten data points (less than 0.4% of the full sample). All missing values occur in SDG 10 for Oman and Saudi Arabia across the study period. To address these gaps, a peer-country proxy approach is employed, using Qatar as the reference country due to its geographic proximity and comparable economic structure. Qatar exhibits a stable SDG 10 score of 73 throughout the observation window; this value is therefore used to impute the missing SDG 10 observations for Oman and Saudi Arabia.

Prior to model estimation, SDG outcome variables are retained on their original 0–100 scale, while NES indicators are standardized when required by linear models and left unscaled for tree-based algorithms. This preprocessing strategy ensures numerical stability, preserves cross-country comparability, and supports reproducibility across all empirical stages of the analysis.

3.2. RQ1—Prediction

To assess the predictive power of entrepreneurial ecosystem conditions for the dimensions of inclusive development (SDG 8, SDG 9, and SDG 10), we estimate separate models for each SDG, applying a two-tiered modelling strategy using the NES data from the Global Entrepreneurship Monitor covering the years 2020–2024. First, the Elastic Net regression serves as the linear baseline, balancing interpretability with regularization to address multicollinearity among NES indicators, Equation (1). Second, we employ XGBoost as the nonlinear benchmark, capable of capturing complex interactions and threshold effects that may characterize the link between entrepreneurial ecosystems and development outcomes, Equation (2). To avoid temporal or geographic leakage, we implement grouped cross-validation by country, ensuring that models are tested on unseen economies. Model performance is evaluated using the coefficient of determination (R²) and the normalized root mean squared error (NRMSE%) (Equations (3) and (4) respectively), which together capture explanatory power and scale-independent predictive accuracy. Condition importance is further interpreted through SHapley Additive exPlanations (SHAP), see Equation (5), providing insights into the contribution and direction of influence of each NES indicator on the respective SDG outcomes.

$${\widehat{Y}}_{i_{new},k}={\widehat{\beta }}_{0,k}+{\sum }_{j=1}^{12}{X}_{i_{new},j}{\widehat{\beta }}_{j,k}$$

(1)

where:

${\widehat{Y}}_{i_{new},k}$: The Predicted Score for SDG $k$ for the new observation $i_{new}$.

${\widehat{\beta }}_{0,k}$: The Estimated Intercept (Bias) term for SDG $k$, derived from the Elastic Net optimization.

$X_{i_{new},j}$: The observed score for the j-th NES indicator (input feature) for the new observation.

${\widehat{\beta }}_{j,k}$: The Estimated Coefficient (Weight) for the j-th NES indicator, derived from the Elastic Net optimization.

$${\widehat{Y}}_{i_{new},k}=\Phi \left(X_{i_{new}}\right)={\sum }_{t=1}^{500}{f}_t\left(X_{i_{new}}\right)$$

(2)

where:

${\widehat{Y}}_{i_{new},k}$: The Predicted Score for SDG $k$ for the new observation $i_{new}$.

$\Phi (X_{i_{new}})$: The overall prediction function of the XGBoost model.

$t:$ The defined total number of decision trees.

$f_t\left(X_{i_{new}}\right)$: The prediction generated by the t-th individual decision tree.

$X_{i_{new}}$: The vector of the 12 input NES indicators for the new observation.

$$R_k^2=1-\left[{\displaystyle \frac{\sum _{i=1}^{185}{\left(Y_{\mathrm{i},k}-{\widehat{Y}}_{i,k}\right)}^2}{{\sum _{i=1}^{185}{(Y}_{i,k}-{\overline{Y}}_k)}^2}}\right]$$

(3)

$${\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}}_{\mathrm{k}}={\displaystyle \frac{\sqrt{\left(\frac{1}{n}\sum _{i=1}^{185}{({\mathrm{Y}}_{i,k}-{\widehat{Y}}_{i,k})}^2\right)}}{Y_{k,max}-Y_{k,min}}}$$

(4)

where:

$Y_{i,k}$: The Actual Observed Score for SDG $k$ for observation $i$.

${\widehat{Y}}_{i,k}$: The Predicted Score for SDG $k$ (from the T = 500 XGBoost model) for observation $i.$

$n$: The total number of observations (37 countries × 5 years = 185).

$Y_{k,max}-Y_{k,min}$: The Range of the actual observed SDG scores.

$$\Phi \left(X_{i_{new}}\right)={\Phi }_0+\sum _{j=1}^{12}{\Phi }_j\left(X_{i_{new}}\right)$$

(5)

where:

$\Phi \left(X_{i_{new}}\right):$ The Predicted Score for the specific SDG outcome $k$ for the new observation ($i_{new}$)

${\Phi }_0:$ The Base Value. This is the average of the predicted SDG score across the entire training dataset.

${\Phi }_j\left(X_{i_{new}}\right)$: The SHAP Value for the j-th NES indicator. This is the calculated contribution of that specific NES indicator’s value.

$j$: Index from 1 to 12 the (NES inputs).

Before model estimation, we undertake three key data preparation steps. First, to ensure fair penalization of predictors, the twelve NES indicators are standardized to have a mean of 0 and a variance of 1 when used in the Elastic Net models, while no scaling is applied for XGBoost since tree-based methods are scale-invariant. Second, the outcome variables are defined separately, with three prediction tasks corresponding to SDG 8, SDG 9, and SDG 10, each modelled independently using the NES indicators as predictors. Finally, to avoid information leakage across time within the same country, we employ grouped cross-validation by country, ensuring that all yearly observations of a given country remain within either the training or the testing folds. This design thereby enhances the robustness and external validity of the predictive results.

3.3. RQ2—Forecasting and Scenario Analysis

Building on the predictive evidence obtained in RQ1, the analysis for RQ2 focuses on forecasting the evolution of SDG 8 and SDG 9, while SDG 10 is excluded due to the limited predictive relevance of entrepreneurial ecosystem conditions for inequality outcomes. XGBoost is adopted as the primary forecasting technique, given its superior performance in RQ1 (see Section 4.1). Moreover, compared to other nonlinear models such as Random Forest, XGBoost provides superior computational efficiency and regularization, producing more stable and parsimonious estimates in small cross-country panel datasets. In particular, when coupled with SHAP analysis, it also enables consistent and interpretable feature attribution [29].

The forecasting models are trained on NES indicators from 2020 to 2024 and validated using grouped cross-validation by country to ensure generalizability. Forecasts extend to 2030, consistent with the SDG timeline. To explore alternative policy pathways, three scenario simulations are constructed: (1) a Finance-led scenario emphasizing access to finance, government support, and taxation, (2) an Innovation-led scenario prioritizing R&D transfer, infrastructure, and professional support systems, and (3) a balanced scenario involving moderate improvements across all conditions. Forecast accuracy is assessed using R² and NRMSE% from historical back-tests, and results are presented both numerically and graphically to illustrate projected deviations from the trend baseline under each scenario.

3.4. RQ3—Causality and Condition Importance

To identify which entrepreneurial ecosystem conditions most strongly influence inclusive growth outcomes, this study applies a causal machine learning framework, using the same NES dataset with the ecosystem indicators as explanatory variables and the SDG 8, 9, and 10 as outcomes. The analysis combines Double Machine Learning (DoubleML) for causal identification with SHAP-based feature importance for interpretability. Double ML estimates the debiased causal effects of each factor while controlling for confounding through cross-fitted ML models, whereas SHAP values (derived in RQ1) rank variables by their overall contribution, see Figure 3. Together, these complementary approaches allow us to distinguish correlation from causation. Feature importance (SHAP) provides a ranking of predictors, while causal ML estimates the magnitude and direction of impacts. In this way, the framework bridges the gap between prediction accuracy and policy relevance by revealing what actually drives change.

4. Results

4.1. Prediction Power

The predictive analysis reveals that entrepreneurial ecosystem conditions are strong determinants of inclusive growth outcomes across 37 countries during the period 2020–2024. The results in

Table 1. Predictive performance of Elastic Net and XGBoost models in estimating SDGs. The higher R² and lower NRMSE% values indicate more accurate predictions.

SDGs	Model	R2	NRMSE%	Top 3 NES Levers (XGBoost Importance)
SDG 8	Elastic Net	0.41	6.2	Financing for entrepreneurs, governmental support & policies, and taxes and bureaucracy
	XGBoost	0.63	5
SDG 9	Elastic Net	0.48	6.7	R&D transfer, physical & services infrastructure, and commercial & professional infrastructure
	XGBoost	0.70	5.3
SDG 10	Elastic Net	0.19	9.4	Governmental support & policies, cultural & social norms, and taxes & bureaucracy
	XGBoost	0.39	7.5

show that the NES indicators predict SDG 8 (Decent Work) and SDG 9 (Innovation and Infrastructure) with high accuracy, as the XGBoost model explains 63% and 70% of the cross-country variance, respectively, while maintaining low prediction errors (NRMSE = 5.0% and 5.3%). In contrast, predictive performance for SDG 10 (Reduced Inequality) is weaker (R² = 0.39; NRMSE = 7.5%), consistent with the understanding that inequality is largely shaped by redistributive and institutional mechanisms beyond the entrepreneurial ecosystem framework.

Across all three goals, XGBoost consistently outperforms the linear Elastic Net model, underscoring the importance of nonlinearities and complementarities among ecosystem pillars. These findings suggest that the impact of individual conditions—such as finance or education—depends on their interaction with others; for example, financial depth contributes to stronger innovation only when supported by effective education systems and adequate infrastructure. Such interdependencies are better captured by nonlinear machine-learning models than by traditional linear approaches.

In contrast, the weakest predictors across all SDGs reveal areas where entrepreneurial ecosystems exert limited influence. As shown in Figure 3, government programmes and basic school entrepreneurial education consistently exhibit the lowest SHAP values, indicating minimal explanatory power for the outcomes of SDG 8, SDG 9, or SDG 10. This pattern suggests that while targeted initiatives and early-stage entrepreneurial curricula may contribute to awareness, they are insufficient drivers of systemic progress in employment, innovation, or inequality. Their weak and uniform influence across SDGs implies that these factors operate more as supportive background conditions rather than active levers of inclusive development. Strengthening these domains may therefore require deeper structural reforms rather than isolated expansion of training or government initiatives.

Together, these results suggest that progress in employment and innovation depends on the strength and interaction of multiple ecosystem pillars, while inequality is shaped more by broader institutional and redistributive forces. Notably, elements such as government programmes and basic entrepreneurial education appear too weak to drive meaningful change on their own, emphasizing the need for more integrated, system-wide approaches to achieve truly inclusive development.

4.2. Forecasting and Scenario Analysis

Under the linear trend extrapolation specified in Equation (6), both SDG 8, and SDG 9 are projected to improve steadily through to 2030, albeit with notable variation across countries, as reported in

Table 2. Forecast summary of SDG 8 and 9 under Trend (2020–2024 trajectory extended) baselines, showing country-level averages for 2025–2029 and projections for 2030.

Country	SDG 8		SDG 9		Country	SDG 8		SDG 9
	Mean	2030	Mean	2030		Mean	2030	Mean	2030
	2025–2029		2025–2029			2025–2029		2025–2029
Brazil	75.6	77.2	67.9	69.6	Mexico	73.9	72.4	59.9	51.5
Canada	83.4	85	76.4	78.2	Morocco	73.7	77.2	53.2	52.1
Chile	76.3	77.9	68.1	69.7	Norway	79.2	79.2	90.8	88.2
China	77.3	78.9	70	71.7	Oman	55.4	54.1	77.7	80
Colombia	74.7	76.1	67	68.5	Poland	81.5	83.5	69.5	78.8
Croatia	78.5	80.1	71.4	73	Qatar	68.5	68.8	75.2	67.4
Cyprus	69.9	68.1	54.7	51.5	Saudi Arabia	74.5	74.9	81.5	80.9
France	80.3	80.3	86.5	79.6	Slovenia	81.4	80.9	85.9	83.4
Germany	79.5	77.7	71.9	56.9	South Africa	70.7	70.1	59.8	57.6
Greece	76.4	76.2	73.5	71	South Korea	82.9	81.8	92.1	90.5
Guatemala	65.2	64.1	35.6	38.6	Spain	77.5	75.9	83.9	81.2
Hungary	74.9	67.7	54.4	45.6	Sweden	81.8	81.6	92.7	91
India	73.5	69.8	78.6	92.8	Switzerland	83.2	82.8	94.7	93
Israel	74.2	74.3	61.1	60.1	Arab Emirates	78.9	80.7	87.3	88.6
Italy	77.1	69.9	78.9	72	United Kingdom	79.9	79.2	88.6	86.9
Japan	69.5	60.9	62.4	57.8	United States	82.3	82.5	90.3	88.7
Latvia	77.8	78.3	73.1	72.6	Uruguay	82.9	80.6	85.8	90.5
Lithuania	60.2	54.1	78.2	80	Venezuela	73.4	71.4	38.8	51.8
Luxembourg	80.5	76.3	69.2	67

. These projections reinforce the earlier findings, indicating that innovation outcomes (SDG 9) are considerably more responsive to incremental enhancements in entrepreneurial ecosystem conditions than employment outcomes (SDG 8). Countries such as Canada, Austria, and South Korea display pronounced gains in SDG 9, reflecting the compounding effects of strong R&D transfer and infrastructural progress. In contrast, economies such as Brazil and Morocco show only modest improvements, suggesting persistent structural bottlenecks that limit the translation of ecosystem dynamics into employment and innovation gains.

$$X_{c,j,t,k}=X_{c,j,2024}+\left(t-2024\right)\left({\displaystyle \frac{X_{c,j,2024}-X_{c,j,2020}}{2}}\right)$$

(6)

where:

$k$: $k$ ∈ {SDG 8, SDG 9}

$X_{c,j,t,k}$: The Extrapolated Score for NES indicator $j$ in country $c$ for the prediction year $t$ (2025 ≤ t ≤ 2030), used as input for the SDG $k$ prediction model.

$X_{c,j,2020}$: The Observed Score for NES indicator $j$ in country $c$ in the first observation year (2024).

$X_{c,j,2024}$: The Observed Score for NES indicator $j$ in country $c$ in the last observation year (2024).

The comparison of the 2025–2029 averages with the 2030 baseline forecasts reveals clear patterns of uneven progress across countries, with innovation-related outcomes showing stronger momentum than employment gains. As shown in Figure 4, the slope chart effectively captures the magnitude and direction of these changes, illustrating that SDG 9 generally follows a steeper upward trajectory than SDG 8. For clarity, only a representative subset of countries is displayed, since including all economies would obscure cross-country trends. Taken together, Table 2 and Figure 4 indicate that while both decent work and innovation outcomes advance over time, the most pronounced gains occur in SDG 9, thereby emphasizing the need for sustained investment in entrepreneurial ecosystems to achieve the 2030 Agenda targets.

To deepen this analysis, we designed a set of policy shock scenarios to test how targeted interventions in specific entrepreneurial ecosystem conditions could alter the baseline forecasts using the scenario shock adjustment in Equation (7). The scenarios were defined in line with theoretical expectations and prior literature:

• Scenario A (Finance-led): +5 points to Financing for Entrepreneurs, Governmental Support & Policies, and Taxes and bureaucracy. This represents policies emphasizing access to capital, regulatory support, and advanced skills.
• Scenario B (Innovation-led): +5 points to R&D Transfer, Commercial & Professional Infrastructure, and Physical Infrastructure & Services. This reflects innovation-oriented strategies focused on knowledge transfer and infrastructure capacity.
• Scenario C (Balanced): +3 points applied equally to all 12 NES indicators, simulating a broad-based but less intensive reform package.

$$\begin{gathered} {\widehat{Y}}_{c,k,t}^{forcast}={\Phi }_k\left(X_{c,j,t,}^{forcast} |j=1\dots 12\right) \\ {X}_{c,j,t}^{forcast}=\left\{\begin{array}{c}{X}_{c,j,t}^{Baseline}+\delta ,ifj=j*\\ X_{c,j,t}^{Baseline} ,ifj\ne j*\end{array}\right. \end{gathered}$$

(7)

where:

${\widehat{Y}}_{c,k,t}^{forcast}$: The Predicted SDG Score for country $c$, for outcome $k$ (SDG 8 or SDG 9), in year $t$, resulting from the policy shock.

${\Phi }_k$: The specific trained XGBoost model for SDG $k$.

$X_{c,j,t}^{forcast}$: The new, modified score for NES indicator $j$ in country $c$ for year $t$ under the policy intervention. This input is fed into the XGBoost model.

$X_{c,j,t}^{Baseline}$: The Baseline Score calculated using the Linear Trend Extrapolation we defined. This serves as the starting point for the shock.

$j*$: The specific NES indicator being targeted by policy intervention.

$\delta$: The magnitude of the policy shock.

Before examining the policy shock scenarios, it is important to establish the reliability of the forecasting models.

Table 3. Model validation results for SDG 8 (Decent Work) and SDG 9 (Innovation & Infrastructure), showing predictive accuracy under grouped cross-validation and back-testing (2020–23 → 2024).

Target	Evaluation	R2	NRMSE%
SDG 8	Grouped CV (3-fold)	0.51	6
SDG 9	Grouped CV (3-fold)	0.63	5.2
SDG 8	Back-test (2020–23 → 2024)	0.47	6.1
SDG 9	Back-test (2020–23 → 2024)	0.66	5

reports validation results for SDG 8 and SDG 9 using grouped cross-validation and a back-testing strategy (train 2020–23 → test 2024). The models achieve strong predictive accuracy for SDG 9 (R² around 0.63–0.66, NRMSE ≈ 5%), while performance for SDG 8 is weaker (R² ≈ 0.47–0.51, NRMSE ≈ 6%). This confirms the robustness of the forecasting approach.

The logic behind these experiments was to compare targeted versus comprehensive policy interventions, and to identify which levers generate the strongest improvements in dimensions of inclusive development. The models were re-estimated using the same XGBoost forecasting framework as in the baseline to ensure comparability.

Table 4. Global average impacts of three policy shock scenarios on SDG 8 and 9, reported as changes (Δ) relative to the trend baseline for 2025–2029 and 2030.

Scenario	ΔSDG 8	ΔSDG 8 2030	ΔSDG 9	ΔSDG 9 2030
	Avg 25–29		Avg 25–29
(A) Finance-led (+5)	−12.92	−11.06	−1.15	1.48
(B) Innovation-led (+5)	−0.16	0.14	8.21	8.05
(C) Balanced (+3) all conditions	−13.32	−10.88	13.2	15.47

shows a clear divergence across strategies. Scenario B (Innovation-led) delivers the largest and most consistent gains for SDG 9 (≈+8 points by 2030 on average), confirming that innovation outcomes are particularly sensitive to R&D and infrastructure levers. Scenario A (Finance-led) produces modest improvements in SDG 9 (+1.5 by 2030) but mixed effects on SDG 8, with some countries experiencing declines due to nonlinear interactions between finance, regulation, and other pillars. Finally, Scenario C (Balanced) produces the largest overall increases in SDG 9 (+15.5 points by 2030). However, it is associated with substantial average decline in SDG 8 (−11 points), reflecting potential trade-offs when all ecosystem pillars are moved simultaneously.

At the country-level Figure 5 underscores these patterns and reveals important heterogeneity. Under Scenario B, countries such as Croatia, Poland, and South Africa emerge as strong beneficiaries, registering double-digit gains in SDG 9 by 2030. By contrast, Scenario A shows that finance-oriented reforms do not yield broad employment gains; in fact, the United States, China, and Brazil experience declines in SDG 8 despite small boosts in SDG 9, highlighting nonlinear trade-offs in the ecosystem. Scenario C raises SDG 9 sharply in many European economies (e.g., Sweden and Germany, with projected gains of over +40 points by 2030), but SDG 8 simultaneously falls, illustrating that balanced reforms may shift benefits toward innovation capacity at the expense of employment inclusiveness.

Some individual cases stand out. Guatemala and Venezuela, for example, achieve remarkable improvements in SDG 9 across scenarios (often exceeding +30 points), suggesting that even relatively weaker ecosystems can achieve outsized innovation gains if the right levers are targeted. Conversely, Uruguay shows modest improvements across scenarios, reinforcing that well-performing ecosystems may yield only incremental benefits from additional shocks.

Together, the baseline and scenario results provide an answer to the second research question, while current trajectories already point to steady progress, targeted innovation-driven reforms (Scenario B) are the most effective in accelerating SDG 9, whereas employment outcomes (SDG 8) remain harder to shift and may require complementary labour market or redistributive policies beyond the NES framework. These findings highlight both the potential and the limits of entrepreneurial ecosystems in shaping inclusive development, reinforcing the central insight from our first research question that innovation is more directly ecosystem-driven than employment.

To assess whether the scenario findings are sensitive to the choice of forecasting algorithm, we re-estimated the models using Random Forest regression as a robustness check. Validation results are reported in

Table 5. Validation results for Random Forest models of SDG 8 and 9, using grouped cross-validation and back-testing.

Target	Evaluation	R2	NRMSE
SDG 8	Grouped CV (3-fold)	−0.11	29.3
SDG 8	Back-test (2020–23 → 2024)	0.66	12.5
SDG 9	Grouped CV (3-fold)	0.01	26.7
SDG 9	Back-test (2020–23 → 2024)	0.77	11.5

. While Random Forests performed less consistently across countries under grouped cross-validation (R² values close to zero), back-testing on the temporal split yielded stronger results, particularly for SDG 9 (R² = 0.77, NRMSE = 11.5). Importantly, the direction of effects remained consistent with the XGBoost analysis: innovation outcomes (SDG 9) are more predictable and responsive to ecosystem conditions than employment outcomes (SDG 8). These findings confirm that the main scenario conclusions do not depend on a single modelling approach.

4.3. Causality Analysis

The final analytical step bridges predictive insight with causal rigour, specifically addressing RQ3: identifying which entrepreneurial ecosystem conditions exert the strongest causal influence on inclusive development outcomes (SDGs 8, 9, and 10). To estimate the debiased causal effects of each NES indicator, the study applies the Double ML framework, leveraging its ability to overcome high-dimensional confounding and multicollinearity, see Equation (8). Subsequently, SHAP values derived from the predictive analysis are incorporated to assess the robustness and interpretability of these causal findings. This integration provides a comprehensive view, identifying both the magnitude of the causal effects and the practical relevance of each policy lever.

As shown in

Table 6. Combined DoubleML–SHAP Summary of Causal Effects and Feature Importance across SDG 8, SDG 9, and SDG 10.

Ecosystem Factor	SDG 8		SDG 9		SDG 10		Overall Interpretation
	DML Effect 1	SHAP Mean	DML Effect	SHAP Mean	DML Effect	SHAP Mean
Financing for entrepreneurs	1.89 ↑	0.137	9.19 ↑	0.098	8.52 ↑	0.09	Strong positive lever across all SDGs; financing drives inclusive innovation and employment.
Governmental support & policies	−0.86 ↓	0.126	−1.26 ↓	0.11	−3.94 ↓	0.121	Consistently negative causal effect despite moderate importance → policy design inefficiency or rigidity.
Taxes & bureaucracy	−1.73 ↓	0.119	−1.01 ↓	0.049	0.17	0.108	Regulatory burden hinders SDG 8 and 9; neutral for SDG 10.
Governmental programmes	2.85 ↑	0.039	2.76 ↑	0.043	1.26 ↑	0.037	Positive causal impact with low SHAP → targeted programmes effective but not yet systemic.
Basic school entrepreneurial education	−1.88 ↓	0.045	−1.11 ↓	0.053	−0.26 ↓	0.044	Limited short-term causal gains; foundation not translating yet into SDG outcomes.
Post-school entrepreneurial education	−1.62 ↓	0.073	−1.94 ↓	0.073	−8.79 ↓	0.083	Negative across all; possible skills–market mismatch or lagged returns.
R&D transfer	1.08 ↑	0.058	8.73 ↑	0.155	8.30 ↑	0.059	One of the strongest dual levers—high causal + high SHAP; critical for innovation-driven inclusivity.
Commercial & professional infrastructure	−3.32 ↓	0.078	−2.19 ↓	0.128	2.58 ↑	0.056	Negative for productivity but slightly positive for inequality; fragmented business ecosystem.
Internal market dynamics	0.04	0.083	−0.53	0.061	0.85 ↑	0.05	Weak causal effects overall; modest role in SDG 10.
Internal market openness	0.85 ↑	0.052	−0.63	0.084	5.52 ↑	0.078	Strong for SDG 10 → openness mitigates inequality; mixed for SDG 9.
Physical & services infrastructure	1.08 ↑	0.066	2.92 ↑	0.141	2.32 ↑	0.067	Positive across all → enabling backbone for inclusive growth.
Cultural & social norms	−1.89 ↓	0.094	−1.18 ↓	0.066	−7.85 ↓	0.115	Consistently negative causal effects—entrenched norms still constrain inclusivity.

¹ The isolated causal contribution of an ecosystem factor to an SDG outcome, separated from correlated ecosystem conditions.

, the combined Double ML–SHAP analysis highlights three consistently high-impact levers (financing for entrepreneurs, R&D transfer, and physical and services infrastructure), each demonstrating strong causal effects reinforced by high predictive importance across all SDGs. These represent the structural pillars of inclusive, innovation-driven growth. In contrast, governmental programmes show positive causal effects but only modest predictive relevance, implying effective but limited reach, while governmental support and policies and cultural and social norms exhibit high SHAP values but negative causal impacts, signalling inefficiencies and institutional rigidity.

$${\widehat{\alpha }}_{j,k}={\displaystyle \frac{\sum _{i=1}^{37}\sum _{t=1}^5\left(Y_{i,t}^{\left(k\right)}-{\widehat{g}}_k\left(W_{i,t}^{\left(j\right)}\right)\right)\left(T_{i,t,j}-{\widehat{m}}_j\left(W_{i,t}^{\left(j\right)}\right)\right)}{\sum _{i=1}^{37}\sum _{t=1}^5{\left(T_{i,t,j}-{\widehat{m}}_j\left(W_{i,t}^{\left(j\right)}\right)\right)}^2}}$$

(8)

where:

${\widehat{\alpha }}_{j,k}$: Debiased causal effect of condition $j$ on SDG $k$.

$Y_{i,t}^{\left(k\right)}:$ SDG $k$ outcome in country $i$, year $t$.

$T_{i,t,j}$: value of NES ecosystem indicator $j$, in year $t$ for country $i$.

${\widehat{g}}_k\left(W_{i,t}^{\left(j\right)}\right)$: Denotes the predicted SDG $k$ outcome based on all NES indicators and country–year fixed effects.

${\widehat{m}}_j\left(W_{i,t}^{\left(j\right)}\right)$: Denotes the predicted value of ecosystem condition $j$ given all control conditions.

Further validation is presented in

Table 7. Rank-Correlation and Robustness Matrix between Double ML Causal Effects and SHAP Importance across SDGs.

SDG	Correlation	Alignment Strength	Interpretation
SDG 8	0.67	Strong positive alignment	The ranking of causal and predictive importance aligns well. Most high-effect variables (e.g., Financing for entrepreneurs, Governmental programmes, R&D transfer) are also top SHAP contributors, validating their robustness as employment and productivity drivers.
SDG 9	0.82	Very strong alignment	Strongest convergence among all SDGs, both methods highlight R&D transfer, Financing, and Infrastructure as the most powerful innovation levers, confirming structural causality rather than mere correlation.
SDG 10	0.58	Moderate alignment	Partial convergence. Financing and Market openness are causally positive and SHAP-salient, but Cultural norms and post-school education show high SHAP yet negative DML effects, indicating policy friction or delayed returns.

, where rank correlations between Double ML and SHAP rankings are consistently positive (0.67 for SDG 8, 0.82 for SDG 9, and 0.58 for SDG 10) indicating strong alignment between causal influence and predictive importance. The robustness matrix confirms that financing, R&D, and infrastructure show consistent convergence across all SDGs, whereas institutional and cultural domains remain misaligned, limiting their effective causal influence.

The consistent positive rank correlations confirm that the key drivers identified by Double ML are not artefacts of the model structure but genuinely correspond to high-importance features in the predictive space. Financing, R&D transfer, and infrastructure emerge as robust, causally validated policy levers across the inclusive development spectrum, while institutional and cultural domains remain the main barriers limiting causal effectiveness despite their high SHAP relevance.

5. Discussion

The empirical results indicate that entrepreneurial ecosystems are critical yet context-dependent mechanisms linking entrepreneurship to inclusive development. Across analyses, finance, infrastructure, and R&D transfer emerge as the most decisive ecosystem pillars shaping employment (SDG 8) and innovation (SDG 9), while inequality outcomes (SDG 10) respond only weakly to the same drivers. This asymmetric responsiveness suggests that while entrepreneurial ecosystems are effective in fostering productivity and innovation, their capacity to deliver equitable outcomes is limited in the absence of complementary institutional and redistributive mechanisms.

The predictive analysis further shows that ecosystem conditions interact rather than operate independently. The superior performance of XGBoost relative to Elastic Net demonstrates that development outcomes are shaped by nonlinearities and complementarities among ecosystem pillars. SHAP-based diagnostics reveal that financial access and supportive regulation are most strongly associated with job creation, whereas R&D transfer and infrastructure are more closely linked to innovation outcomes. These effects materialize through co-evolution across pillars rather than through isolated improvements.

The forecasting analysis extends these findings from static associations to dynamic trajectories. Scenario simulations show that innovation outcomes respond elastically to ecosystem strengthening, particularly in middle-income economies where R&D and infrastructure investments accelerate technological diffusion and catch-up growth. In contrast, employment gains remain modest across scenarios, highlighting structural rigidities that limit the translation of innovation into broad-based job creation. Notably, the Balanced scenario generates strong innovation gains but is associated with a decline in SDG 8, indicating a trade-off between rapid ecosystem upgrading and short-run employment absorption.

Causal inference deepens this empirical picture. Double Machine Learning results confirm that finance, R&D transfer, and infrastructure exert robust structural effects on inclusive development outcomes. In contrast, Governmental Support & Policies and Cultural & Social Norms exhibit weak or negative causal effects. When compared with the predictive results, this divergence reveals a clear distinction between perceived ecosystem support and structurally effective policy levers.

These empirical patterns carry important theoretical implications for entrepreneurship, innovation, and development research. The strong role of finance, infrastructure, and R&D transfer in driving employment and innovation aligns with Schumpeterian and endogenous growth theories, which locate knowledge accumulation, innovation, and capital formation at the core of long-term productivity and structural transformation [7,8]. At the same time, the weak responsiveness of inequality outcomes supports institutional theory’s argument that inclusiveness depends primarily on governance quality and redistributive capacity rather than on market dynamism alone [9,10].

More broadly, the results shift the understanding of entrepreneurial ecosystems from linear, additive systems toward nonlinear and complementary networks. The observed interaction effects validate the systemic perspective of ecosystem theory, in which the marginal contribution of any single pillar depends on the configuration of the wider system [11,17]. This perspective is further reinforced by the scenario results, which reveal that innovation-led ecosystem upgrading can outpace labour-market adjustment, echoing long-standing insights that technological change may initially widen productivity and income gaps before redistribution mechanisms respond [30].

The divergence between predictive and causal effects for Governmental Support & Policies and Cultural & Social Norms further refines institutional interpretations of entrepreneurship-led development. While these factors appear as strong predictors in perception-based models, their causal effects weaken or turn negative once core structural drivers are isolated. This finding highlights a gap between policy intent and policy impact: visible policy commitment and pro-entrepreneurial norms may signal an enabling environment without necessarily producing inclusive outcomes. This interpretation is consistent with longitudinal evidence showing that legitimacy and perception-based ecosystem dimensions often evolve asynchronously from structural drivers of entrepreneurial performance, particularly during periods of systemic stress and recovery [31]. Consistent with [23], entrepreneurship policies may emphasize expansion over effectiveness, while cultural support can both enable and constrain entrepreneurship depending on access to opportunity and institutional fairness [18]. Together, these results position policy and cultural conditions as context-dependent moderators whose developmental impact hinges on institutional coherence and absorptive capacity, in line with the institutional-efficiency logic emphasized by [32].

6. Conclusions

This research demonstrates that entrepreneurial ecosystem conditions are strong predictors of inclusive growth, particularly in shaping employment (SDG 8) and innovation (SDG 9). Finance, R&D transfer, and infrastructure emerge as the most consistent levers, translating entrepreneurial dynamism into sustained productivity and technological advancement. However, inequality (SDG 10) remains weakly responsive, underscoring that entrepreneurship alone cannot ensure equity without complementary institutional and redistributive mechanisms.

The results also reveal marked asymmetries across contexts. Innovation is highly elastic to R&D and infrastructural investment (especially in middle-income economies where absorptive capacity is expanding) whereas employment gains depend more on institutional adaptability and labour-market flexibility. These dynamics confirm that entrepreneurship-led development is nonlinear and contingent, requiring policy coherence and institutional depth to balance innovation with inclusion.

Causal evidence reinforces that growth becomes inclusive only when financial, knowledge, and institutional systems evolve synergistically. Conversely, governmental support and policies and cultural norms exert limited or adverse effects when not supported by effective governance and absorptive capacity.

In sum, entrepreneurial ecosystems appear to be important but not sufficient for inclusive growth. While they support innovation and employment, equitable outcomes remain contingent on institutional quality, policy coordination, and social inclusion, underscoring the need to situate entrepreneurship within broader development frameworks.

7. Theoretical and Policy Implications

This research advances entrepreneurship and development theory by empirically demonstrating that entrepreneurial ecosystems function as nonlinear, interdependent systems rather than additive collections of isolated conditions. The superior performance of nonlinear models (XGBoost, Double ML) confirms that development outcomes emerge from complementarities among finance, infrastructure, and R&D transfer, validating the systemic logic proposed by ecosystem theorists, [17]. The results also refine institutional and cultural perspectives: government support and social norms are not universal enablers but context-dependent moderators whose effects depend on governance quality and absorptive capacity. By integrating the principles of complexity and institutional efficiency into the entrepreneurship–development nexus, this research extends Schumpeterian and endogenous-growth frameworks toward a more dynamic and systemic model of inclusive development.

The findings underscore the need for coordinated, ecosystem-oriented policymaking that reinforces the interdependence of finance, R&D transfer, and infrastructure while enhancing institutional capacity to translate policy intent into tangible outcomes. Governments should embed inclusiveness within innovation policy by promoting equitable access to finance, education reform, and labour-market protections that sustain participation in knowledge-driven economies. Achieving inclusive growth therefore requires aligning entrepreneurial opportunity with redistributive balance and institutional coherence. Future research should extend this multi-method framework to longitudinal and micro-level analyses, integrating measures of governance, trust, and cultural adaptation to capture how entrepreneurship shapes sustainable and equitable development [33].