4.1. Empirical Results
We begin by reporting the results of essential preliminary tests, such as cross-sectional dependency, homogeneity, and unit root tests. Cross-sectional dependence is critical when examining the relationships between variables in panel data models. Due to the prospect of countries becoming dependent on one another as a result of numerous economic ties and the effect of shared causes, ignoring spillover effects might result in incorrect inference and misspecification issues. Thus, we begin by examining cross-sectional dependence. Additionally, slope homogeneity tests are included. As shown in Table 5
, we report five cross-sectional dependence tests: the Lagrange multiplier (LM) test developed by Breusch and Pagan [80
], the adjusted Lagrange multiplier (LM-adj) test developed by Pesaran et al. [81
], the cross-sectional dependence (CD) test developed by Pesaran [82
], the LM cross-sectional dependence (CDLM
) test developed by Pesaran [82
], and the adjusted CDLM
-adj) test developed by Baltagi et al. [84
]. Another critical preliminary test in panel estimation is the model coefficients’ country-specific heterogeneity. We employ two alternative tests to determine slope homogeneity. The first test is Pesaran and Yamagata’s [85
] truncated slope homogeneity (
) test with Blomquist and Westerlund’s [86
] heteroskedasticity and autocorrelation consistent covariance (HAC) adjustment. The second test is an adjusted version of the
test for small samples, designated by
. Each test is constructed using a pooled ordinary least squares regression with six different model specifications. LETD is the dependent variable in each model. Each of the Models 1 to 5 include the variables LDA, LEDU, LFDIR, LSOC, and LTOP1, but each also includes one of the variables LCO2
, LEPI, LGDP, LGTD, and LTAI as independent variables, in the provided order. Model 6 augments the independent variables in Model 1 with the variables LCO2
, LEPI, LGDP, LGTD, and LTAI.
The cross-sectional dependence tests shown in Table 5
indicate that, with just three exceptions, the null hypothesis of no cross-sectional dependence is rejected for all countries in all models at all significant levels. Thus, the results reveal the existence of cross-sectional dependence among countries, implying that, as a result of the 58 countries’ high degree of integration, the shock that originated in one of these 58 countries appears to have propagated to the others. Another significant point is that the unit root tests should be chosen depending on the cross-sectional dependence test results. Due to the existence of cross-sectional dependence in the data we employ, the second-generation unit root test should be employed to ensure that the tests are efficient.
Additionally, Table 5
includes the results of slope homogeneity testing. At all levels of significance, these tests do not reject the null hypothesis of slope homogeneity across countries. As a result, the slope is unlikely to vary by country, and the influence of independent variables on green technology diffusion appears to be homogeneous across 58 countries.
The panel data method that should be utilized is determined by the data’s stationarity property. Thus, we conduct unit roots tests prior to estimating empirical relationships. We employ second-generation unit root tests due to the existence of cross-sections in the data. Among the second-generation unit root tests that allow for cross-sectional dependency, we use the cross-section augmented Im-Peseran-Shin (CIPS) test of Peseran [87
], the modified CIPS (M-CIPS) tests of Westerlund and Hosseinkouchack [88
], the panel analysis of nonstationarity in idiosyncratic and common components (PANIC) based panel unit root test (PANIC-
) of Westerlund and Larsson [89
], and the bias adjusted version of the PANIC-
). The results of the unit root tests are given in Table 6
. The results in Table 6
demonstrate that all tests substantially reject the unit root null hypothesis with constant and constant plus trends specifications at the 1% significance level, with the exception of LCO2
, for which the PANIC-based tests do not strongly reject the unit root null. Given that the CIPS and M-CIPS tests agree in all cases, we infer that all series are stationary and can be estimated using the stationary GMM.
Since all series are stationary, and the concerns are of endogeneity, multicollinearity, autocorrelation, the system GMM method is used. We provide the system GMM estimates in Table 7
, together with their standard errors, which are given in parentheses below the parameter estimates. Baltagi [90
] points out that the system GMM estimator has the most desirable properties in the presence of endogenous regressors for stationary dynamic panels with large cross-sectional (
) and short and fixed time (
) dimensions, which hold in our case with
. The green technology diffusion—the dependent variable in our study—is largely driven by general technology diffusion and innovation. Therefore, some of the regressors, such as the per capita income, CO2
emissions, environmental performance, technological achievement, and foreign direct investment, potentially depend on green technology diffusion. None of the other approaches based on the common correlated effects estimator of Pesaran [91
] are appropriate in our case since they do not account for variable endogeneity. These alternative approaches work better for long
and at best require weakly exogenous regressors (see e.g., Chudik and Pesaran [92
] and Baltagi [90
]). As Sarafidis [93
] and Sarafidis and Wansbeek [94
] point out, the GMM estimator can eliminate cross-sectional dependence and maintain its consistency. Moreover, a consistent GMM estimator is feasible using a subset of the instruments based on exogenous instruments (Sarafidis [93
] and Sarafidis and Wansbeek [94
]). Indeed, following Sarafidis et al. [95
], we perform cross-sectional dependence tests on the residuals of the system GMM estimates (denoted SYR-CD), which are reported in Table 7
. The SYR-CD test results show that the system GMM estimator accounts for all the cross-sectional dependence in the data, and our results do not suffer from cross-sectional dependence.
The estimates are obtained using a two-step system GMM since standard errors from a single-step system GMM are always asymptotically inefficient. In system GMM, the dependent variable’s second to third lags and the first differences of all explanatory variables except lagged dependent variables are employed as instruments for the first differenced equation. The level equation is estimated using the first differences of endogenous variables as instruments. Due to the concerns of multicollinearity, autocorrelation, and endogeneity, we do not include all possible variables in each model; rather, we include a subset, as described in Table 1
. The level and first-differenced models are assessed using the Lagrange multiplier Arellano-Bond autocorrelation tests for accurate dynamic specification and the Sargan-Hansen J
-test for instrumental variable validity.
For the dynamic system GMM estimation with one lag, we should reject the first order serial correlation by the Lagrange multiplier Arellano-Bond test—LM-AR(1)—and not reject the second order serial correlation AR(2) using the LM-AR(2) test. The LM-AR(1) tests’ results in Table 7
are all significant at the 1% significance, confirming the AR(1) specification. However, some of the LM-AR(2) tests are significant at the 5% level (Models 1–3, 6–8). For these cases, we estimate the models with two lags and find that both LM-AR(2) and LM-AR(3) tests are insignificant, invalidating the AR(2) specification. Therefore, all models are estimated with one lag of the dependent variable. Additionally, the Sargan-Hansen J
-test results for the validity of over-identification constraints in Table 7
do not reject the null hypothesis of valid over-identification restrictions at all commonly used significance levels for all models that we estimate. As a result, we conclude that the specified instrumental variables are valid and that an AR(1) dynamic specification is sufficient to capture autocorrelation. The SYR-CD cross-sectional dependence test of Sarafidis et al. [95
], reported in Table 7
, which is valid for GMM estimates, indicates that the system GMM estimator adequately accounts for cross-sectional dependence in the data. As a result, our findings are not subject to cross-sectional dependence.
We estimate the eleven distinct models specified in Table 1
to evaluate the effects of the variables examined in this study as potential determinants of green technology diffusion. Our approach is to estimate regression models with key variables from Table 1
, such as democratic accountability, foreign direct investment, socioeconomic conditions, and income distribution, while also developing instruments with the same subset. The estimates are summarized in Table 7
. The empirical data indicate that the democratic accountability variable has a positive sign and is statistically significant across all specifications at the commonly used 10%, 5%, and 1% significance levels. Thus, more democratic conditions facilitate the adoption of green technologies, a finding that corroborates Zecca and Nicolli’s [15
] empirical evidence. Similarly, except for models 1 and 4, the effect of foreign direct investment is large with a positive sign in all other models. This empirical conclusion corroborates Halkos and Skouloudis’s [38
] finding that FDI has a beneficial influence on green technology development in a static panel data model. As a result, this finding demonstrates the critical importance of foreign direct investment in a country’s spread and development of green technologies.
The education index is closely linked with a number of other factors, including the TAI, per capita GDP, environmental performance index, and CO2
emission. This is because education is one of the variables used to construct environmental performance and technological achievement indices. The association between education level and GDP is well established in the literature on economic growth (e.g., [96
]). Additionally, a few studies establish an empirical link between education and CO2
emission (e.g., [98
]). Taking this into consideration, we include education in Models 1,3,5, and 7, but not in others, in order to avoid the identification problem that emerges when education is a sub-component of another variable, or multicollinearity. With the exception of Model 5, where education has a positive effect on green technology dissemination but is not statistically significant, the education index is statistically significant at all usual significance levels and has a positive sign in all models where education is included. Thus, it is recognized that a country’s education level promotes the dissemination and development of green technology, most likely with a greater influence when combined with other encouraging elements. This finding is consistent with Bartel and Lichtenberg’s [25
] finding that highly educated individuals have a comparative advantage in adapting to and learning new technologies, which results in an increase in demand for new technologies.
According to Table 7
, the socioeconomic conditions variable has positive coefficients and is statistically significant in all models except for Model 4, where the parameter estimate is statistically insignificant. As the empirical evidence indicates, socioeconomic conditions in countries have a positive effect on green technology diffusion. To gain a better understanding of the diffusion of green technology, we take another underlying variable into account: income inequality. In this study, inequality is quantified by the income share of the top 1% of highest-income households, denoted by the variable TOP1. In all models in which it is included, the inequality variable has a significant estimate at the 1% level and a negative sign, with the exception of Model 2, where the coefficient estimate is statistically insignificant. This finding is consistent with Vona and Patriarca’s [13
] empirical findings. Thus, the findings imply that a more equitable income distribution fosters green innovation in societies, while increasing economic inequality retards the diffusion and development of green technologies.
emission variable is included in models 2, 7, 9, and 10, but not in others due to its strong association with the education index, per capita GDP, and the TAI. CO2
emission has a positive and statistically significant effect on the diffusion of green technologies in each specification at the 5% level of significance. As a result, in countries with larger CO2
emissions, the diffusion of green technologies is enhanced. According to Alataş [101
], environmental technologies have a statistically insignificant positive influence on CO2
emission in the EU15’s transport sector. In comparison to our findings, this is a reverse causality, albeit a statistically insignificant one. Due to the fact that we employ system GMM estimation, this reverse causality does not affect our estimates. However, Du et al. [14
] find contradictory evidence. They demonstrate that whereas green technology advances have a negligible influence on CO2
emissions in less developed economies, they have a considerable impact on CO2
emissions in industrialized ones. However, this conclusion contradicts the findings of Naseem and Guang Ji [102
], who find a statistically significant negative relationship between renewable energy usage and CO2
Due to the substantial association between education index and per capita GDP, the EPI is included in Models 3, 7–11, but not in others to avoid potential identification and multicollinearity. The results reveal that the EPI is statistically significant at the 1% level of significance in each model. Thus, our study establishes that in those countries where environmental performance is high the diffusion of green technologies is also stronger, indicating that prioritizing environmental issues causes faster adaptation of green technologies. GDP per capita is included in models 3, 9, and 11 because of its strong association with the education index, CO2
emission, environmental performance index, and socioeconomic conditions index. Except for Model 11, in which the effect of GDP is shown to be statistically insignificant, the GDP variable shows a positive sign and statistical significance at usual significance levels in relation to the diffusion of green technologies. It is inferred that countries with a higher per capita income will invest more in green technology and inventions, as well as file more patent applications. The conclusion is consistent with Vona and Patriarca [13
] and Zecca and Nicolli [15
], who demonstrate a positive and statistically significant association between GDP and environmental technology. Thus, adaptability and dissemination of green products and technologies are related to the per-capita income level and income distribution in all countries. Models 5, 10, and 11 include the general technology diffusion variable. The coefficient estimates for the general technology diffusion is positive and significant at the 1% level in all model specifications. Thus, the empirical evidence supports the notion that green technologies have a strong relationship and increase concurrently with the general diffusion of technology.
Lastly, the TAI, which is calculated by the authors, appears in Models 6 and 8, and is excluded from others due to its high correlation with the education, CO2
, GDP, and general technology diffusion variables. The estimates in Table 7
reveal that the technology achievement index has a positive effect on green technology diffusion, which is significant at the 1% level. Green technology adoption occurs more rapidly in countries with a high level of TAI because these countries have a track record of easily resolving environmental issues, minimizing environmental degradation, and enhancing environmental quality.
In summary, green technology diffusion is related to a society’s per capita income, education level and investment profile, socioeconomic conditions and democratic accountability, CO2 emissions, environmental performance, and general technology diffusion and innovation adaption. As a result, we use the system GMM technique to estimate parameters for various model specifications while taking into account the relationship between green technology diffusion and its determinants. As demonstrated in the table, adaption to green technology diffusion is positive and significant for all variables except inequality. The TOP1 variable’s coefficient is statistically significant (except in model 2) and has a negative sign. This finding indicates that the diffusion of green technologies is strongly tied to the income distribution among countries. Additionally, among the 11 models, the most statistically significant specifications of model 3, model 6, and model 8 might be picked. All parameter coefficients are expected and statistically significant at the 1% level of significance in these models. The technical achievement score and the environmental performance index, in particular, are significant and favorably connected with the dissemination of green technologies.
In summary, green technology diffusion is related to the per capita income, education level, and investment profile of a country. Socio-economic conditions and democratic accountability, CO2 emissions, environmental awareness, and adaptation of general innovations are also significant determinants of green technology diffusion. We estimate various model specifications using the system GMM method considering the interaction between green technology diffusion and its determinants, which may lead to endogeneity issues. Our estimates show that green technology diffusion relates positively and significant to all variables except for income inequality, for which the relationship is negative and significant. This conclusion demonstrates a substantial relationship between green technology dissemination and income distribution across countries, which is one of the study’s novel results. Two further novel findings concern the technical achievement index and the environmental performance index, both of which are examined for the first time in our study. We demonstrate that these variables are major predictors of green technology diffusion. With rare exceptions, the majority of parameter estimates in the 11 models we estimate are statistically significant at traditional significance levels.
The determinants of green technology diffusion were examined empirically in the context of the general technology diffusion trends, environmental performance, democratic accountability, income distribution, income level, socioeconomic conditions, and technological achievement of nations using a large panel of 58 nations. Our empirical results indicate that the relationships between green technology diffusion and income level, education level, investment profile, socioeconomic conditions, democratic accountability, CO2
emission, environmental performance, brow technology diffusion, and technological achievements are statistically significant and positive in selected countries. Thus, improvements in these factors help increase the adaptation and spread of green technologies. Firstly, the income level of countries is a predominant factor for enhancing green products and sustainable development. Fatima et al. [103
] find that the ratio of consumption of renewable energy to CO2
emission descends when income rising. At the same time, some studies [15
] conclude that higher GDP increases carbon emissions. Secondly, the investment profiles of countries play an important role in the adaption of green innovations. Khan et al. [105
] show significant causal relationships between policies of exports and imports, income level, and green innovation that have resulted in changes to consumption-based CO2
emission levels in G7 countries. Likewise, Shahzad et al. [106
] find that export diversification greatly decreases CO2
emissions in selected developed and developing countries. These outcomes are comparable to the conclusion of Andersson [107
] on Chinese exports. In this respect, our results are complimentary to the empirical evidence presented in these studies, but in a more extensive coverage of countries, which includes both developing and developed countries.
Thirdly, economic development, socio-economic conditions, democratic accountability, technological innovation, and environmental performance are particularly prominent factors shaping green technologies. Wang et al. [108
] indicate that political freedom and institutional quality help to reduce CO2
emission levels when GDP growth and financial development are enhancing environmental degradation. The result is in line with the findings of [109
]. Zuhair and Kurian [111
] recognize some socio-economic barriers which affect the diffusion of green products, such as the gender gap, lack of human and financial capacity, loss of community spirit, and lack of environmental awareness. Moreover, Chaudhry et al. [112
] find that technological innovations have a significantly positive relationship with environmental indicators in higher-income East-Asian and Pacific countries.
Energy-efficient and environmentally friendly technologies, as well as patents, promote environmental technology development. Paramati et al. [113
] show that green technology helps to decrease energy consumption and improve energy efficiency. These results are similar to the outcomes of [114
]. Another study by Arbolino et al. [116
] concludes that economic variables have an important role in the diffusion of green technology policies by using the EPI.
Overall, our empirical analysis demonstrates how the determinants of green technology diffusion make a positive contribution to attaining a sustainable environment as described in the relevant literature. However, our results go beyond the previous studies by including a larger set of factors that affect environmental technology diffusion. Our study obtains complimentary evidence to previous literature from a broader range of time periods and a larger number of countries that includes both developing and developed economies. More importantly, our study is the first to obtain evidence that green technology does not independently develop and diffuse from the general or brown technology trends and technological achievement of countries. The general technology diffusion and achievement trends in a country are significant drivers of the green technology trends.