Assessing the Long-Run Role of Renewable Electricity Generation in Reducing CO₂ Emissions and Supporting Industrial Labour Absorption in Morocco

The Paris Agreement has increased pressure on countries to align economic development with low-greenhouse-gas and climate-resilient pathways [1]. In Morocco, this commitment gained greater visibility through the hosting of COP22 in Marrakech in 2016, which highlighted the country’s climate-policy position and renewable-energy ambitions [2]. Morocco’s energy decarbonisation strategy has progressively become an important component of its long-term sustainable-development agenda, linking climate mitigation, electricity-system transformation, renewable-energy deployment and industrial-sector involvement [3]. Electricity occupies a central position in this strategy because Morocco’s power system remains strongly dependent on fossil fuels. In 2018, the electricity sector accounted for approximately 40% of energy-related CO₂ emissions, while fossil fuels supplied about 81% of electricity generation, including 59% from coal, compared with around 19% from renewable sources [3].

Morocco aims to increase the renewable share of installed electricity capacity to 52% by 2030, while its long-term strategy considers renewable shares of 70% by 2040 and 80% by 2050, together with carbon neutrality by 2050. Large-scale projects such as the Noor Ouarzazate Solar Complex illustrate the progressive implementation of these objectives. Electricity accounted for 19% of Morocco’s total final energy consumption in 2023 [4]. Consequently, electricity-sector decarbonisation cannot be fully achieved without progressively substituting coal and other fossil-fuel-based electricity generation with renewable sources. Renewable electricity generation, denoted as renewable electricity generation (REN), is therefore measured as the share of renewables in total electricity generation and used as a realised production-based indicator of Morocco’s electricity decarbonisation pathway.

Beyond its environmental dimension, Morocco’s decarbonisation process raises an industrial-employment question. Industrial employment represented approximately 23% of total employment in 2023, confirming the importance of industry in the national employment structure [5]. In this study, industrial employment is used as a broad indicator of industrial labour absorption, activity expansion and sector participation in decarbonisation. It does not measure direct green-job creation, but helps assess whether renewable-electricity expansion supports industrial labour absorption and industrial-sector involvement.

Figure 1 presents the evolution of electricity generation by source in Morocco over 1990–2023. Renewable electricity generation has expanded progressively, reflecting the growing contribution of hydro, wind and solar power to the national electricity system. However, coal remains a major source of electricity and continues to occupy an important position in the electricity mix. This coexistence between renewable expansion and persistent fossil-fuel dependence shows that Morocco’s electricity-decarbonisation pathway should be understood in relation to changes in both renewable and non-REN. It also raises the question of whether Morocco’s dependence on coal-fired electricity acts as a predictive driver of the future evolution of REN. Coal-based electricity generation is therefore included as a control variable in the CO₂-emissions model and examined in the complementary Granger causality analysis.

tu_pian_7.png

Despite growing research on renewable energy, CO₂ emissions and employment, three gaps remain. First, many studies rely on renewable-energy consumption, installed capacity or individual technologies rather than REN as a realised share of total generation.

Second, environmental and employment outcomes are generally examined separately, limiting understanding of their joint relationship with electricity decarbonisation.

Third, limited analysis examines whether renewable-electricity expansion is associated with industrial labour absorption, used as a broad indicator of the expansion of industrial activity and its capacity to generate employment. This study addresses these gaps through two complementary models centred on REN. Model 1 examines the long-run association between REN and CO₂ emissions within an Environmental Kuznets Curve (EKC)-based specification, in which income and its squared term capture income nonlinearity, while coal-based electricity generation is included as a control variable. Model 2 examines the long-run association between REN and industrial employment, considering productive and digital conditions, while education expenditure and broad money are included as control variables. The Autoregressive Distributed Lag (ARDL) bounds-testing approach estimates long-run relationships and short-run adjustments, while the Granger causality analysis examines whether past changes in Morocco’s dependence on coal-fired electricity provide predictive information about the future evolution of REN. The following hypotheses are tested:

H1: Renewable electricity generation is negatively associated with CO₂ emissions per capita in the long run.

H2: Renewable electricity generation is positively associated with industrial-employment share in the long run.

H3: Long-run cointegration exists among the dynamic variables of both models.

H4: Past changes in Morocco’s dependence on coal-fired electricity provide predictive information about the future evolution of REN.

The remainder of the article is organised as follows: Section 2 reviews the relevant literature, Section 3 presents the methodology, Section 4 reports and discusses the empirical results, Section 5 provides the conclusions, and Section 6 outlines the study’s limitations and areas for future research.

3.1 Model specification

The empirical analysis uses two complementary models to examine whether REN is associated with lower CO₂ emissions and greater industrial labour absorption in Morocco. This structure addresses the environmental and industrial-employment dimensions of electricity decarbonisation within the empirical framework.

Model 1 adopts an EKC-based control specification following Grossman and Krueger [6] and Panayotou [7]. Following Danish et al. [25], LNCO₂ is used as the dependent variable and represents the natural logarithm of CO₂ emissions per capita from OWID [26]. LNREN is the main explanatory variable, and its coefficient is expected to be negative because renewable electricity may displace carbon-intensive generation [8-12, 27, 28]. LNGDPPC represents the effect of per capita income on emissions, while LNGDPPC2, calculated as the square of LNGDPPC, captures nonlinearity in the income–emissions relationship [25]. LNCOAL is included as a control variable because coal-based electricity generation is an important source of CO₂ emissions and its inclusion helps reduce omitted-variable bias [28, 29]. The Model 1 specification is adapted from Danish et al. [25].

$\text{LNC}{{\text{O}}_{{{2}_{t}}}}={{\beta }_{0}}+{{\beta }_{1}}\text{LN}RE{{N}_{t}}+{{\beta }_{2}}\text{LNG}DPP{{C}_{t}}+$${{\beta }_{3}}\text{LN}GDPPC{{2}_{t}}+{{\beta }_{4}}\text{LNCOA}{{\text{L}}_{t}}+{{\varepsilon }_{t}}$         (1)

Model 2 examines the long-run association between REN and industrial employment. LNEMP is the dependent variable and reflects industrial labour absorption rather than direct green-job creation [18]. LNREN is the main explanatory variable, and its coefficient is expected to be positive because renewable-energy expansion may support employment through construction, installation, operation, maintenance and related industrial activities [13, 18]. LNPROD represents industrial labour productivity [30], while LNCEL measures mobile-cellular subscriptions as an indirect proxy for digital connectivity and technology adoption [17]. LNEDUC captures human-capital development [31], while LNM2 captures financial conditions [16]; both are included as fixed control regressors [29]. The coefficient on LNCEL may be negative if technology adoption reduces industrial labour absorption by substituting for certain tasks. The Model 2 specification is adapted from Bajja et al. [18].

$\text{LNEM}{{\text{P}}_{t}}={{\alpha }_{0}}+{{\alpha }_{1}}\text{LN}RE{{N}_{t}}+{{\alpha }_{2}}\text{LN}PRO{{D}_{t}}$$+{{\alpha }_{3}}\text{LN}CE{{L}_{t}}+{{\alpha }_{4}}\text{LNEDU}{{\text{C}}_{t}}+{{\alpha }_{5}}\text{LNM}{{2}_{t}}+{{\varepsilon }_{t}}$          (2)

All variables are transformed into natural logarithms, while LNGDPPC2 is constructed as (LNGDPPC)². Annual data cover the period 1990–2023, providing a maximum of 34 observations. Table 1 presents the variable definitions, measurement units and data sources.

Table 1. Description of the study variables, measurement units, and data sources

Note: In the unit column, t denotes metric tonnes. LNGDPPC2 is constructed as (LNGDPPC)² to capture the potential nonlinear effect of economic growth on CO₂ emissions. OWID [26] denotes Our World in Data, WDI denotes World Development Indicators, and IEA denotes the International Energy Agency.

3.2 Econometric strategy

The analysis begins with descriptive statistics. The Augmented Dickey–Fuller (ADF) and Phillips–Perron (PP) unit-root tests are then applied to determine whether the variables are stationary at level, I(0), or after first differencing, I(1) [32-34]. An ADF breakpoint unit-root test is then used to account for a structural break in each series [35, 36].

The breakpoint test includes an intercept and trend, allows for a break in the intercept, selects the break date using the minimum Dickey–Fuller t-statistic and determines the lag length using the Schwarz Information Criterion [37]. These tests also confirm that none of the variables is integrated of order two, I(2). The ARDL bounds-testing approach developed by Pesaran et al. [38] is then used because it is suitable for small samples and accommodates variables integrated at I(0) and I(1).

Given the annual dataset of 34 observations, one lag is retained to capture the maximum available information while maintaining an effective sample of 33 observations. Within this restriction, the final lag structure of each model is selected using the Akaike Information Criterion [39]. Cointegration is assessed by comparing the calculated F-statistic with the corresponding asymptotic critical values reported by Pesaran et al. [38]. A long-run relationship is confirmed when the F-statistic exceeds the upper-bound critical value [38]. When cointegration is confirmed, the conditional ARDL error-correction model (ECM) is estimated to distinguish short-run changes from long-run relationships [38].

The conditional ARDL specifications of the CO₂ emissions and industrial-employment models are formulated following Pesaran et al. [38], expressed in natural logarithmic form following Wooldridge [29], and presented in Eqs. (3) and (4), respectively:

$\begin{gathered}\Delta \mathrm{LNCO}_{2_{\mathrm{t}}}=\beta_0+\sum_{\mathrm{i}=1}^{\mathrm{p}-1} \beta_{1i} \Delta \mathrm{LNCO}_{2_{\mathrm{t}-\mathrm{i}}} +\sum_{\mathrm{j}=0}^{\mathrm{q}-1} \beta_{2 \mathrm{j}} \Delta \mathrm{LNREN}_{\mathrm{t}-\mathrm{j}}+\sum_{\mathrm{j}=0}^{\mathrm{q}-1} \beta_{3 \mathrm{j}} \Delta \mathrm{LNGDPPC}_{\mathrm{t}-\mathrm{j}} +\sum_{\mathrm{j}=0}^{\mathrm{q}-1} \beta_{4 \mathrm{j}} \Delta \mathrm{LNGDPPC2}_{\mathrm{t}-\mathrm{j}}+\gamma_1 \mathrm{LNCO}_{2 \mathrm{t}-1} \\ +\gamma_2 \mathrm{LNREN}_{\mathrm{t}-1}+\gamma_3 \mathrm{LNGDPPC}_{\mathrm{t}-1} +\gamma_4 \mathrm{LNGDPPC2}_{\mathrm{t}-1}+\theta_1 \mathrm{LNCOAL}_{\mathrm{t}}+\varepsilon_{\mathrm{t}}\end{gathered}$         (3)

$\begin{gathered}\Delta \text {LNEMP}_{\mathrm{t}}=\alpha_0+\sum_{\mathrm{i}=1}^{\mathrm{p}-1} \alpha_{1 \mathrm{i}} \Delta \text {LNEMP}_{\mathrm{t}-\mathrm{i}} +\sum_{\mathrm{j}=0}^{\mathrm{q}-1} \alpha_{2 \mathrm{j}} \Delta \text { LNREN }_{\mathrm{t}-\mathrm{j}}+\sum_{\mathrm{j}=0}^{\mathrm{q}-1} \alpha_{3 \mathrm{j}} \Delta \text {LNPROD}_{\mathrm{t}-\mathrm{j}} +\sum_{\mathrm{j}=0}^{\mathrm{q}-1} \alpha_{4 \mathrm{j}} \text { △LNCEL }_{\mathrm{t}-\mathrm{j}}+\delta_1 \text { LNEMP }_{\mathrm{t}-1}+ \\ \delta_2 \text { LNREN }_{\mathrm{t}-1}+\delta_3 \text { LNPROD }_{\mathrm{t}-1}+\delta_4 \text { LNCEL }_{\mathrm{t}-1} +\mathrm{Z}_1 \text { LNEDUC }_{\mathrm{t}}+\mathrm{Z}_2 \text { LNM }_{\mathrm{t}}+\varepsilon_{\mathrm{t}}\end{gathered}$     (4)

In Eqs. (3) and (4), Δ denotes the first-difference operator, while ${{\beta }_{0}}$ and ${{\alpha }_{0}}$ are the constant terms. The coefficients ${{\text{ }\!\!\beta\!\!\text{ }}_{1\text{i}}}$ and ${{\text{ }\!\!\alpha\!\!\text{ }}_{1\text{i}}}$ capture the short-run persistence of the dependent variables, whereas ${{\text{ }\!\!\beta\!\!\text{ }}_{2\text{j}}}-{{\text{ }\!\!\beta\!\!\text{ }}_{4\text{j}}}$ and ${{\text{ }\!\!\alpha\!\!\text{ }}_{2\text{j}}}-{{\text{ }\!\!\alpha\!\!\text{ }}_{4\text{j}}}$ measure the short-run effects of changes in the explanatory variables. The coefficients $\gamma_1-\gamma_4$ and ${{\delta }_{1}}-{{\delta }_{4}}$ are associated with the lagged-level variables and are jointly tested to determine the existence of long-run relationships.

The coefficient ${{\theta }_{1}}$ is associated with the fixed control regressor LNCOALₜ, while $Z_1$ and $Z_2$ are associated with the fixed control regressors LNEDUCₜ and LNM2ₜ, respectively. Cointegration is confirmed when the bounds-test F-statistic exceeds the relevant upper critical bound. In the corresponding ECM, a negative and statistically significant error-correction coefficient indicates convergence toward long-run equilibrium and measures the speed of adjustment. A pairwise Granger causality test is subsequently conducted between coal-based and REN using one lag. Following Granger [40], the general specification is expressed as follows:

$Y_t=\alpha_0+\sum_{i=1}^p \alpha_i Y_{t-i}+\sum_{j=1}^q \beta_j X_{t-j}+\varepsilon_t$           (5)

where, $Y_t$ denotes the dependent variable, ${{\text{X}}_{\text{t}-\text{j}}}$ denotes the explanatory variable, $\text{p}$ and $\text{q}$ represent their respective lag lengths, and εₜ is the error term. The hypotheses are specified as follows: H₀: ${{\text{ }\!\!\beta\!\!\text{ }}_{1}}$ = ${{\text{ }\!\!\beta\!\!\text{ }}_{2}}$ = ··· = ${{\text{ }\!\!\beta\!\!\text{ }}_{\text{q}}}$ = 0; H₁: at least one ${{\text{ }\!\!\beta\!\!\text{ }}_{\text{j}}}$ ≠ 0, for j = 1, …, q. Rejecting H₀ when the p-value is below 5% indicates that past values of Xₜ provide useful information for predicting Yₜ [40]. To satisfy the stationarity requirement of the conventional Granger causality test, variables integrated of order one, I(1), were expressed in first differences, while I(0) variables were retained in levels [40, 41].

Finally, model validity and robustness are assessed using diagnostic and stability tests. The Breusch–Godfrey LM test examines serial correlation [42, 43], while the Breusch–Pagan–Godfrey test assesses heteroskedasticity [44].

Residual normality is evaluated using the Jarque–Bera test [45], and parameter stability is examined through the CUSUM and CUSUM of Squares tests [46]. Together, these tests assess the robustness of the estimated models by determining whether they satisfy the principal econometric assumptions and whether their coefficients remain stable throughout the study period.

The results show that REN is associated with lower CO₂ emissions and greater industrial labour absorption in Morocco. The bounds tests confirm cointegration in both models, supporting H3.

First, REN is negatively and significantly associated with CO₂ emissions in the long run, with an elasticity of − 0.068928, supporting H1. Renewable-electricity expansion is therefore associated with lower CO₂ emissions. However, the small coefficient indicates that this effect remains limited while fossil fuels, particularly coal, continue to occupy an important position in the electricity mix.

Second, REN is positively and significantly associated with industrial employment, with an elasticity of 0.053054, supporting H2. This relationship reflects the industrial sector’s engagement in electricity decarbonisation through labour absorption rather than automatic green-job creation. Local employment benefits may arise through construction, equipment supply, cabling, metal structures, installation, grid connection, technical services, operation and maintenance. However, dependence on imported technologies limits domestic production of high-technology renewable-energy equipment and reduces employment in these activities.

Third, the Granger result shows that past changes in coal-based electricity generation provide predictive information about the future evolution of REN, supporting H4. Economically, Morocco’s dependence on coal-fired electricity emerges as a significant predictor of the future evolution of REN. This finding reinforces the fundamental rationale for decarbonisation: progressively reducing reliance on carbon-intensive electricity through the development of REN, limiting potential exposure to pollution-related taxes and carbon costs, strengthening Morocco’s contribution to global environmental protection, and supporting the fulfilment of its international climate commitments. When considered together with the long-run results, the finding shows that Morocco’s renewable-electricity strategy responds to persistent coal dependence while contributing to lower CO₂ emissions and greater industrial labour absorption. The error-correction, diagnostic and stability results support the robustness of both models.

Overall, the findings underline the importance of reducing fossil-fuel dependence, strengthening domestic renewable-equipment production, developing workforce skills and aligning renewable expansion with employment-oriented industrial policies.

We are grateful to Mohammed V University in Rabat, the Faculty of Legal, Economic and Social Sciences-Agdal (FSJES-Agdal), and the Laboratory of Applied Economics for their institutional support.