The study uses annual time series for Togo, a member country of the Economic Community of West African States (ECOWAS), over the period from 1980 to 2022. These data come from a variety of sources, such as the World Bank’s latest World Development Indicators database ( World Bank, 2023 ) and public data from the Global Ecological Footprint Network’s National Ecological Footprint and Biocapacity Accounts ( Global Footprint Network, 2023 ) ( Figure 1 , Table 1 ).

The variables used are:

Notes: * and ** indicate significance at the 5% and 10% level, respectively. Source: Authors, based on data from the World Bank ( World Bank, 2023 ) and the Global Ecological Footprint Network ( Global Footprint Network, 2023 ).

To investigate the long-term impact of the service sector on environmental quality, the study draws on the economic literature and specifies an empirical model linking environmental quality to the service sector, economic growth, physical capital and the active workforce in the Togolese economy. These variables have all been included.

${\text{CO}}_2{}_t=f\left({\text{SERVICE}}_t,{\text{PIBTETE}}_t,{\text{FBCF}}_t,{\text{TRAVAIL}}_t\right)$ (1)

where t is the time period.

To examine the relationship between environmental quality and the tertiarization of economic activities, a number of econometric steps need to be taken following the study of a time series. The first step is to test for the presence of unit roots in the series. In the economic literature, it has been shown that the presence of a unit root in a series is likely to create spurious regressions in the estimates, leading to poor results. In this work, we performed the unit root tests of Augmented Dickey-Fuller ( Dickey & Fuller, 1981 ; Philips & Perron, 1988 ; Kwiatkowski et al., 1992 ). These tests are well documented in the economic literature. When the dependent or explained variable is stationary in the first difference, an error-correction model (ECM) is estimated. On the other hand, if it is stationary in level, then we estimate the empirical model using the heteroscedasticity-robust ordinary least squares (OLS) method. In the second approach, we test for the presence of cointegration between the series of the empirical model using the Johansen test ( Johansen, 1995 ). Cointegration is the presence of a long-term relationship between several non-stationary variables. For a cointegrating relationship to exist between two variables, two conditions must be met. Firstly, the variables must be non-stationary and integrated of the same order. Secondly, there must be at least one linear combination of these variables that is stationary. In the presence of cointegration, we estimate a vector error correction model (VECM), also known as a cointegrated vector autoregression (cointegrated VAR). The Johansen cointegration test a multivariate generalization of the augmented Dickey-Fuller test. To determine the number of cointegrating vectors, Johansen proposed two alternatives: the Trace test and the maximum eigenvalue test.

The following general model is used:

$\begin{array}{c}\ln {\text{CO}}_2{}_t=\beta +{\alpha }_1\ln {\text{SERVICE}}_t+{\alpha }_2\ln {\text{PIBTETE}}_t\\ +{\alpha }_3\ln {\text{FBFC}}_t+{\alpha }_4\ln {\text{TRAVAIL}}_t+{\epsilon }_t\end{array}$ (2)

To perform the Johansen cointegration test, we begin by determining the optimal number of lags to incorporate into the autoregressive process. This number is determined according to the AIC (Akaike Information Criterion), FPE (final prediction error), SBIC (Schwarz’s Bayesian information criterion) and HQIC (Hannan and Quinn information criterion) criteria. VAR models characterized by their dynamic structures, modeling each endogenous variable in the system as a function of the lagged values of all other endogenous variables in the system.

The VAR(p) model in equation 2 is presented as follows:

$Y_t=A_0+A_1{Y}_{t-1}+A_2{Y}_{t-2}+\cdots +A_P{Y}_{t-p}+{\epsilon }_t$ (3)

Avec ${\epsilon }_t\sim N\left(0,\text{Σ}\right)$. $Y_t$ contains the endogenous variables

The VAR can be rewritten as VECM:

$\Delta Y_t=\Pi Y_{t-1}+{\Gamma }_i\Delta Y_{t-i}+A_0+{\epsilon }_t$ (4)

Cointegration relationships are determined by the $\Pi$ matrix, which describes the adjustment towards equilibrium following an exogenous shock.

If the matrix Π has reduced rank r < k (k denotes the number of variables), then there exist two matrices (k * r), α et β each of rank r such that Π = α𝛽r where:

Johansen’s cointegration test is based on estimating the rank r of the matrix Π, which designates the number of cointegration vectors.

Three cases can be distinguished:

Two statistics are defined for the Johansen test: the trace statistic and the maximum eigenvalue statistic.

We use a powerful modeling procedure called dynamic ordinary least squares (DOLS) proposed by Stock & Watson ( Stock & Watson, 1993 ) to check the robustness of the VECM results. The Stock Watson method is a robust approach that corrects for the endogeneity of the explanatory variables by incorporating initial values and first-difference lags of these variables, as well as serially correlated errors through a GLS procedure. The DOLS estimator improves on ordinary least squares (OLS) estimates by taking into account dynamic sources of bias and small sample sizes.

The application of unit root tests indicates that all variables are stationary in the first difference. To this end, all variables are integrated of order 1 (I(1)). The results are shown in Table 2 .

Note: Values in brackets are p-values for the ADF and PP tests. For the KPSS test, on the other hand, they are the Z-statistics of the decision thresholds (10%: 0.119; 5%: 0.146 2.5%: 0.176; 1%: 0.216). The optimal lag length for each variable was determined using the varsoc function. For the ADF and PP tests, *p < 0.10; **p < 0.05; ***p < 0.01 indicate rejection of the unit root null hypothesis at the 10%, 5% and 1% significant levels, respectively. For the KPSS test, *p < 0.10; **p < 0.05; ***p < 0.01 indicate acceptance of the null hypothesis of series stationarity at significant levels of 10%, 5% and 1%, respectively. I(d) denote the order of integration of the variable. Source: Authors, based on data from the World Bank ( World Bank, 2023 ) and the Global Ecological Footprint Network ( Global Footprint Network, 2023 ).

Note: AIC (Akaike Information Criterion), FPE (final prediction error), HQIC (Hannan and Quinn information criterion), and SBIC (Schwarz’s Bayesian information criterion). *Indicates the optimum VECM offset length. Source: Authors, based on data from the World Bank ( World Bank, 2023 ) and the Global Ecological Footprint Network ( Global Footprint Network, 2023 ).

The FPE, AIC and HQIC information criteria indicate an optimal VAR lag length of 4. This leads to the estimation of a VECM with three (3) lags, as for the cointegration test. ( Table 3 )

Johansen’s cointegration test is applied, taking the constant trend and three lags. At rank 1, the trace statistic is below the critical value at the 5% threshold, indicating the presence of a cointegrating relationship. The same is true of the maximum eigenvalue test. There is, therefore, a cointegrating relationship between tertiary activities and carbon dioxide emissions in Togo. The results are presented in Table 4 .

Note: *** and ** indicate the presence of a cointegrating relationship at rank i at the 1% and 5% significant levels, respectively. Null hypothesis (H0) is the absence of cointegration. Source: Authors, based on data from the World Bank ( World Bank, 2023 ) and the Global Ecological Footprint Network ( Global Footprint Network, 2023 ).

As Johansen’s cointegration test revealed a cointegrating relationship, we are therefore interested in the first equation, which contains our variable of interest, and we estimate the long run block of the chosen model by the maximum likelihood method as follows ( Table 5 ):

Note: *p < 0.10; **p < 0.05; ***p < 0.01 indicate significance of coefficients at the 10%, 5% and 1% significant levels, respectively. Values in brackets [] indicate Z-statistics associated with these coefficients. Source: Authors, based on data from the World Bank ( World Bank, 2023 ) and the Global Ecological Footprint Network ( Global Footprint Network, 2023 ).

Estimation of the cointegration relationship leads to the following long-term equation:

$\begin{array}{c}\ln {\text{CO}}_2{}_t=1.562-1.703\ln {\text{SERVICE}}_t-3.927\ln {\text{PIPTETE}}_t\\ -0.278\ln {\text{FBCF}}_t+3.058\ln {\text{TRAVAIL}}_t\end{array}$ (5)

A 1% increase in the value added of tertiary sector activities to GDP (lnSERVICE) generates a 0.587% reduction in carbon dioxide (CO₂). There is also a negative effect on gross domestic product per capita (lnPIBTETE) and gross fixed capital formation (lnFBCF). In fact, a 1% increase in gross domestic product per capita and gross fixed capital formation respectively generates a 0.255% and 3.597% drop in CO₂. On the other hand, the workforce in the Togolese economy, represented by the active population, has a positive impact on carbon dioxide emissions.

Robustness estimates using dynamic ordinary least squares reveal a negative and significant effect of the value added of tertiary activities to GDP on carbon dioxide emissions. Thus, in the long term, the value added of tertiary activities to GDP leads to a reduction in carbon dioxide emissions. This result confirms that of the VECM. ( Table 6 )

Note: *p < 0.10; **p < 0.05; ***p < 0.01 indicate significance of coefficients at the 10%, 5% and 1% significant levels, respectively. Null hypothesis (H0) is the absence of cointegration. Values in brackets [] indicate the Z-statistics associated with these coefficients. The kernel used is the quadratic function (qs). Source: Authors, based on data from the World Bank ( World Bank, 2023 ) and the Global Ecological Footprint Network ( Global Footprint Network, 2023 ).

Short-term results show that the effect of value added in the service sector is not significant on environmental quality. The results are presented in Table 7 .

Note: *p < 0.10; **p < 0.05; ***p < 0.01 indicate significance of coefficients at the 10%, 5% and 1% significant levels, respectively. Null hypothesis (H0) is the absence of cointegration. Values in brackets [] indicate Z-statistics associated with these coefficients. Source: Authors, based on data from the World Bank ( World Bank, 2023 ) and the Global Ecological Footprint Network ( Global Footprint Network, 2023 ).

Validation tests of the VECM model show no evidence of violation of the assumptions of the classical linear regression model ( Table 8 ). Figure 2 also shows that all eigenvalues are below unity, and the cointegrating relationship exists on the unity circle. The model is therefore stable.

Notes: *indicates acceptance of the null hypothesis. Source: Authors, based on data from the World Bank ( World Bank, 2023 ) and the Global Ecological Footprint Network ( Global Footprint Network, 2023 ).

The empirical results of this study show that increasing the value added of the tertiary sector relative to GDP leads to a reduction in carbon dioxide emissions. This is a major result that underlines the importance of the tertiarization of a country’s economy in wealth production, and remains consistent with the findings of Kouton et al. ( Kouton et al., 2024 ) and Senzele ( Senzele, 2022 ). Thus, the greater the importance of tertiary activities in a country’s wealth production, the fewer CO₂ emissions there are, and therefore the less air pollution there is. On the other hand, per capita gross domestic product has a negative effect on carbon dioxide emissions. There are two reasons for the negative effect of per capita income on CO₂. The first is that economic growth in developing countries is driven more by tertiary activities, which contribute more to growth. Empirical research by Awad ( Awad, 2021 ) has shown that the structural transformation of economies in underdeveloped countries is mainly due to the migration of labor from the agricultural sector to the service sector. The second reason is linked to the use of per capita income in developing countries. This income is used more by households for consumption expenditure on goods and services, housing expenditure, healthcare expenditure, etc. This result is in line with that of Keho ( Keho, 2015 ). The positive effect of the labor force (active population) used in the Togolese economy on CO₂ emissions could be explained by the dynamics of their fuel consumption. Indeed, with a fast-growing population and rapid urban development, workers generally use two or four-wheeled vehicles that consume fossil fuels to get to their place of work. What’s more, given their low incomes, Togolese workers generally opt for older vehicles when purchasing a means of transport, which are cheaper but consume more fuel. According to a diagnostic report ( SSATP, 2019 )³ on sustainable mobility and accessibility policies in Togolese cities, the number of private cars (excluding two-wheelers) in circulation in Togo in 2015 was estimated at 140,000 units, or 19 cars/1000 inhabitants. And this rate has grown exponentially since 2005 (+7% between 2005 and 2010, then +30% between 2010 and 2015). This rate is higher in Togo than in neighboring countries such as Burkina Faso and Mali, which have more than twice the population of Togo. Furthermore, these factors of the capital component of the workers’ share combined with the technological effect or the productivity rate improve the impact on CO₂. In this respect, the capital used is increasingly clean, as is technology, which justifies its negative effects outweighing the positive effect of labor. The effect of investment measured by gross fixed capital formation on CO₂ emissions is insignificant.

Comparative analysis of Togo’s results with those of studies such as Kouton et al. (2024) , who worked on the case of Ivory Coast, reveal the same similarity as that of Ivory Coast in terms of the service sector’s high share of value added in GDP and the improvement in the rise of tertiary activities in reducing CO₂ emissions. Thus, the relationship between the tertiarization of activities and environmental quality in Togo is the same as in Côte d’Ivoire. An analysis of the added value of the agricultural, industrial and service sectors in WAEMU countries reveals that these countries have the same economic structure, i.e., the service sector is the biggest contributor to growth, and the workforce migrates more to this sector, to the detriment of the others. This already gives us an intuition of the relationship between the tertiarization of activities and the quality of the environment that may exist in these countries. Empirical studies are needed in these other countries to confirm exactly the similar relationship found in Togo and Côte d’Ivoire.