When Does Technological Capacity Reduce Carbon Emissions? The Conditioning Role of Trade Openness

Climate change and the resulting increase in carbon emissions have become one of the most important global challenges to sustainable economic growth [1, 2]. In this regard, in line with both the Paris Agreement and the 2050 net-zero emission targets, countries are trying to develop policy instruments that can reduce carbon intensity while maintaining economic growth. In this process, technological development and international trade play a critical role in supporting low-carbon development. However, the conditions under which technological progress and trade integration reduce or increase carbon emissions have not yet been fully clarified [3, 4].

Economies with high technological capacity can support the development of environmentally friendly technologies by increasing productivity and energy efficiency in production processes [5, 6]. Similarly, an increase in trade openness facilitates countries' access to advanced technologies, accelerates knowledge diffusion, and can encourage the more rapid adoption of clean production technologies [7, 8]. Nevertheless, empirical studies show that the effects of technological development and trade openness on carbon emissions differ significantly across countries [9, 10]. Therefore, theoretically, both technological capacity and trade openness are expected to contribute to the reduction of carbon emissions. However, this theoretical expectation is not always confirmed in the empirical literature.

Within this framework, trade openness emerges as an important factor shaping the environmental effect of technological capacity. In economies with high trade openness, technological innovations spread more rapidly, access to environmentally friendly production technologies becomes easier, and firms can adopt more efficient production processes [11, 12]. In contrast, in economies with limited trade integration, the same technological capacity may not provide similar environmental gains. Therefore, trade openness accelerates technology diffusion, and the interactions between technological innovation and trade openness may be decisive for carbon emissions. The findings [7, 13] indicate that the effect of technological capacity on carbon emissions may vary according to the level of trade openness across countries.

A significant part of the existing literature examines separately the relationship between technological capacity and carbon emissions or the relationship between trade openness and carbon emissions [14, 15]. In contrast, the question of whether the environmental effect of technological capacity varies depending on the level of trade openness has largely been neglected. Therefore, examining the interaction mechanism between technological capacity and trade openness contributes to filling an important gap in the literature.

This study aims to fill this gap in the literature. Within this scope, a balanced panel dataset consisting of 20 OECD countries covering the period 2007–2023 is used. It is tested whether the effect of technological capacity on carbon emissions varies according to the level of trade openness. Technological capacity is represented by a composite index Technological Capacity Index (TCI) constructed from indicators of high-technology exports, research and development expenditures, and computer technology patents using Principal Component Analysis (PCA). In the analyses, the System Generalized Method of Moments (System GMM) is used in order to take into account the dynamic structure and potential endogeneity problems [16, 17].

This study contributes to the literature in three respects. First, technological capacity is measured through a multidimensional index rather than a single indicator. Second, the interaction between technological capacity and trade openness is directly included in the model, and the conditional nature of the environmental effect of technological development is analyzed. Third and finally, by using marginal effects analysis, it is revealed how the effect of technological capacity on carbon emissions changes at different levels of trade openness. In addition, the interpretation of this relationship is supported by graphical presentation.

The remainder of the paper is organized as follows. The second section reviews the relevant literature. The third section describes the dataset, variables, and methodology. The fourth section presents the preliminary analysis. The fifth section reports and discusses the empirical findings, including the marginal effects analyses. The final section presents the conclusions and policy recommendations.

3.1 Data and variables

In this study, annual data covering the period 2007–2023 are used for 20 OECD countries (Austria, Belgium, Canada, the Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Japan, Latvia, the Netherlands, Norway, Poland, Portugal, Spain, Sweden, the United Kingdom, and the United States) in the baseline analysis. The sample composition is primarily determined by the availability of the WIPO patent data required to construct the technological capacity indicator. Specifically, the patent component of the TCI is calculated as the share of computer technology patents in total patents. Therefore, the sample includes OECD countries for which the WIPO data required to calculate this indicator are consistently available throughout the 2007–2023 period. Other OECD countries were excluded from the sample because the WIPO patent data required to consistently calculate the share of computer technology patents in total patents over the entire 2007–2023 period were not available. The resulting baseline sample is a balanced panel consisting of 20 countries observed over 17 years (N = 20, T = 17), with a total of 340 observations. Additional variables with different data coverage are used in the robustness analyses, as explained in Section 3. The definitions and data sources of the variables used in the study are presented in Table 1.

Table 1. Variable definitions and data sources

The sample period covers a period in which technological transformation accelerated, global trade integration increased, and low-carbon development policies gained importance. The data on carbon emissions were obtained from the OECD database, while the indicators constituting technological capacity were compiled from the World Bank (World Development Indicators) and the World Intellectual Property Organization (WIPO) databases.

The dependent variable is defined as the natural logarithm of per capita CO₂ emissions (lnCO₂). The main explanatory variable of the study is the TCI. This index is a composite indicator consisting of high-technology exports, the share of research and development expenditures in GDP, and computer technology patents in total patents. The index was constructed by using PCA in order to combine the common information into a single variable. Before applying PCA, all component variables were standardized, and the first principal component was used as the TCI.

Trade openness (lnopen) is represented by the natural logarithm of the ratio of total exports and imports to GDP. The real effective exchange rate (lnreer) and real income per capita (lnpergdp) are included in the model as control variables. The real effective exchange rate is considered a control variable because of its potential effects on external competitiveness, trade structure, and production activities. In addition, trade openness is mean-centered around its sample mean before constructing the TCI × Trade Openness interaction term, which is used to test whether the effect of technological capacity on carbon emissions varies according to the level of trade openness.

3.2 Construction of the Technological Capacity Index

Technological capacity is a multidimensional concept that refers to the ability of countries to develop new technologies, use existing technologies, and transform innovation activities into economic production [33]. Therefore, in measuring technological capacity, it is a more appropriate approach to use indicators reflecting innovation production and technological development in different dimensions rather than a single indicator [34].

In this study, a TCI representing the level of technological capacity was constructed. In constructing this index, three main indicators were used to jointly reflect the technological production, innovation activities, and knowledge production capacity of countries. These indicators are the share of high-technology exports (htx), the share of research and development expenditures in GDP (rd), and the share of computer-technology patents in total patents (comp_pat). These indicators represent the technological production capacity, innovation activities, and knowledge production ability of countries, respectively. In constructing the index, first, all variables were standardized using the z-score method in order to eliminate scale differences among the indicators. Then, PCA was applied in order to capture the common variation among the indicators and obtain a single composite index [35].

The preliminary diagnostic tests presented in Table 2 show that the dataset is suitable for PCA. The overall Kaiser–Meyer–Olkin (KMO) coefficient of 0.613 indicates that the dataset is acceptable for PCA according to the criteria of Kaiser [36]. In addition, the Variance Inflation Factor (VIF) values, calculated at approximately 1.30 for all variables, indicate that there is no serious multicollinearity problem among the indicators.

Table 2. Principal Component Analysis (PCA) results and construction of the Technological Capacity Index (TCI)

As shown in Table 2, the PCA results reveal the presence of a dominant first principal component. The eigenvalue of 1.758 for the first principal component (PC1) is above the threshold value of 1 suggested by Kaiser [37], and this component alone explains 58.6% of the total variance in the dataset. This finding shows that the common variance among the technological indicators can be adequately represented by a single composite dimension. Therefore, PC1 was used as the composite indicator representing technological capacity, and the TCI variable was constructed based on this component.

In addition, the component loadings of all indicators on the first principal component are positive and above 0.50. This result supports that PC1 successfully reflects the common technological capacity of countries. Accordingly, the TCI was constructed by using the coefficients of the first principal component of the standardized variables through the following linear combination:

$T C I=0.5181 z_{\text {htx}}+0.5799 z_{\text {rd}}+0.6287 z_{\text {com_pat}}$     (1)

3.3 Empirical strategy: System Generalized Method of Moments

In this study, the dynamic relationship between carbon emissions and technological capacity is estimated by using the System GMM. This method reduces the biases arising from potential endogeneity and unobserved country-specific effects while taking into account the dynamic nature of carbon emissions. Therefore, a dynamic panel data model including the lagged dependent variable is preferred in the analysis.

The System GMM estimator was developed by Arellano and Bover [16] and Blundell and Bond [17]. This method can provide more efficient and consistent estimates by using the first-difference and level equations together. In addition, it reduces the biases arising from endogeneity, fixed effects, and reverse causality by using the appropriate lags of the explanatory variables as instrumental variables.

For the baseline estimations, both one-step and two-step System GMM estimators were implemented using the lag (2–3) instrument structure. Before constructing the interaction term, trade openness was mean-centered around its sample mean. Accordingly, the baseline models include mean-centered trade openness and the interaction term between TCI and mean-centered trade openness. In the two-step estimations, the Windmeijer [38] correction was used in order to correct for small-sample bias. In order to evaluate the robustness of the findings, additional estimations were performed by using the alternative lag (2–4) instrument structure.

The dynamic panel data model used in the analysis is defined as follows:

$\ln C O_{2 i t}=\alpha \ln C O_{2 i, t-1}+\beta_1 T C I_{i t}+\beta_2 c_{-} \operatorname{lnopen}_{i t}+\beta_3\left(T C I c_{-} \operatorname{lnopen}\right)_{i t}+\beta_4 \ln \operatorname{reer}_{i t}+\beta_5 \operatorname{lnpergdp_{it}}+\eta_i+\epsilon_{i t}$     (2)

Here, c_lnopen denotes trade openness mean-centered around its sample mean, and the interaction term is constructed as TCI × c_lnopen. i represents the countries and t represents the time dimension. ηᵢ denotes the unobserved country-specific fixed effects, and $\epsilon_{i t}$ denotes the error term.

The validity of the model was evaluated by using the Arellano–Bond autocorrelation tests (AR(1) and AR(2)) and the Hansen overidentification test. In addition, in order to prevent the instrument proliferation problem, care was taken to keep the number of instruments lower than the number of cross-sectional units.

Table 5 presents the results of the baseline model and robustness analyses obtained using the one-step and two-step System GMM estimators. In these specifications, trade openness is mean-centered around its sample mean, and the interaction term is constructed using mean-centered trade openness.

Table 5. Baseline and robustness checks: System Generalized Method of Moments (System GMM) estimates

The diagnostic test results indicate that the estimated models satisfy the main validity conditions. The Arellano–Bond AR(2) tests do not reject the null hypothesis of no second-order autocorrelation (p-values = 0.171–0.206). Similarly, the insignificant results of the Hansen overidentification tests (p-values = 0.128–0.162) provide no evidence against the validity of the instruments used. In addition, keeping the number of instruments lower than the number of cross-sectional units in all model specifications indicates that the risk of instrument proliferation is limited.

The coefficient of the lagged dependent variable is positive and statistically significant at the 1% level in all specifications, and the estimated coefficients range from 1.065 to 1.110. However, coefficients above one may raise concerns regarding potentially unstable dynamics. Therefore, in order to examine the issue in more detail, Wald tests were conducted for the null hypothesis that the coefficient of the lagged dependent variable is equal to one. The results are presented in Table 6.

Table 6. Wald tests for the lagged dependent variable coefficient

The findings show that the null hypothesis cannot be rejected either in the one-step specification (p = 0.319) or in the two-step specifications with lags 2–3 (p = 0.311) and lags 2–4 (p = 0.504). In addition, the 95% confidence intervals for the coefficient of the lagged dependent variable include one in all three specifications. Therefore, although the estimated coefficients are above one, there is insufficient evidence to conclude that the coefficients are statistically different from one.

On the other hand, panel unit root tests were also conducted for lnCO₂ in order to examine the time-series properties of CO₂ emissions in more detail. The results are reported in Table 7.

Table 7. Panel unit root test results for lnCO₂

According to the results obtained, the Levin–Lin–Chu test rejects the null hypothesis of a unit root (p < 0.001), while the Im–Pesaran–Shin test also rejects the unit root hypothesis at the 5% significance level (p = 0.034). In addition, the t-bar statistic in the Pesaran CADF/CIPS test is −2.684, and the null hypothesis of a unit root is rejected at the 10% significance level according to the finite-sample critical values. The p-value of the standardized Z[t-bar] statistic is 0.037.

When the Wald tests and panel unit root tests are evaluated together, there is no consistent empirical evidence that CO₂ emissions follow an unstable dynamic process. However, since the estimated coefficients for the lagged dependent variable are above one, this finding is not disregarded, and the results regarding dynamic stability are interpreted cautiously.

Trade openness was mean-centered around its sample mean, and the mean-centered variable was used both as the main effect and in constructing the interaction term. Therefore, the zero value of the mean-centered trade openness variable corresponds to the average level of trade openness in the sample. Accordingly, the TCI coefficient shows the conditional effect of technological capacity on carbon emissions at the average level of trade openness. The TCI coefficient is negative and statistically significant at the 1% level in the one-step System GMM estimation (β = −0.058, p < 0.01). In the baseline two-step System GMM model (lag 2–3), the coefficient is negative but not statistically significant (β = −0.048, p > 0.10). Similarly, in the two-step model with the alternative lag structure (lag 2–4), the coefficient is negative but statistically insignificant (β = −0.036, p > 0.10). These results show that, at the average level of trade openness, the TCI coefficient is negative in all specifications, but this conditional effect is statistically significant only in the one-step estimation. However, since the model includes an interaction term, the effect of technological capacity on carbon emissions cannot be evaluated solely through the TCI coefficient; this effect varies depending on the level of trade openness [39].

The most important finding of the study is that the interaction coefficient between TCI and mean-centered trade openness is negative and statistically significant in all model specifications. Although the interaction coefficient ranges from −0.067 to −0.087, it remains stable across different estimation methods and alternative lag structures. This result indicates that the effect of technological capacity on carbon emissions may vary depending on the level of trade openness. Although technological capacity may initially increase production activities and energy use, increasing trade integration may strengthen knowledge diffusion, access to clean technologies, and environmental competitive pressure. Thus, as trade openness increases, the effect of technological capacity on carbon emissions becomes more negative, suggesting that technological capacity may contribute more strongly to emission reduction in more open economies. This finding is also consistent with theoretical approaches arguing that the interaction between technological progress and international trade openness shapes environmental outcomes [3, 7, 27].

The interaction coefficient maintaining similar signs and magnitudes across different estimation methods and lag structures supports the robustness of the main findings. However, to assess whether this coefficient is sensitive to the way TCI is constructed, the baseline model was re-estimated using two alternative technological capacity measures. First, an equal-weighted index was constructed using the three standardized indicators. Second, considering the relatively strong loading of high-technology exports on PC2, an alternative measure based only on R&D and computer-technology patents, excluding high-technology exports, was used. The interaction coefficient between technological capacity and trade openness remained negative in both alternative specifications and yielded results consistent with the finding in the baseline model (β = −0.110, p = 0.059 and β = −0.160, p = 0.001, respectively). The full estimation results are reported in Appendix Table A1. These findings indicate that the main interaction result is robust to alternative constructions of technological capacity.

When the control variables are examined, it is seen that the direct effect of trade openness on carbon emissions is positive. However, the statistical significance of this effect varies across specifications, reaching the 5% level in the baseline two-step model and the 10% level in the other specifications. The real effective exchange rate and income per capita variables are not statistically significant in the two-step estimations.

Beyond these baseline controls, the literature on the determinants of CO₂ emissions shows that emission dynamics are associated not only with economic and technological factors but also with demographic structure, energy mix, and environmental policy conditions. For example, urbanization is considered an important demographic factor associated with CO₂ emission dynamics through channels such as the expansion of urban areas, population density, and changes in energy use [40, 41]. Renewable energy use is important for emission dynamics because it changes the carbon intensity of the energy mix. Indeed, empirical findings for OECD countries show that renewable energy production is significantly associated with CO₂ emissions [42]. In addition, environmental policy stringency (EPS) is an institutional factor reflecting countries’ regulatory frameworks for emission reduction. Recent empirical findings show that the effect of EPS on CO₂ emissions may vary across countries and time-frequency dimensions and that, in some cases, more stringent environmental policies are associated with lower CO₂ emissions [43].

In this context, additional robustness analyses were conducted to assess the sensitivity of the baseline model to these factors. Urbanization, renewable energy use, and EPS included in the additional robustness analyses are not randomly selected additional controls but are considered potential omitted factors representing the demographic, energy, and institutional/policy dimensions of emission dynamics, respectively. The purpose of these analyses is not to re-examine the direct effects of these variables on CO₂ emissions, but rather to assess whether the main finding regarding the interaction between technological capacity and trade openness is preserved when these alternative factors are controlled for.

The data period used in the additional robustness analyses was determined by taking into account the common data availability of the additional control variables. Although the main analysis covers the 2007–2023 period, the data coverage of the additional controls differs. While urbanization data are available for the entire period of the main analysis, the data coverage of the renewable energy consumption indicator extends through 2022. In contrast, the current series of the OECD EPS index covers the 1990–2020 period [44]. Therefore, the common data period of 2007–2020 was used for the robustness analyses in order to allow the additional control variables to be evaluated jointly and the alternative specifications to be compared using the same sample. Since EPS data are not available for Latvia among the countries included in the study, the common sample was restricted to 19 countries. As a result of using a one-period lag of the dependent variable in the dynamic System GMM model, the effective estimation sample covers the 2008–2020 period and consists of 247 observations. Estimating all alternative specifications using the same country and period sample makes it possible to distinguish changes resulting from the inclusion of additional control variables from changes in sample composition. The results of the additional robustness analyses estimated using this common sample are presented in Table 8.

Table 8. Robustness checks with additional controls: two-step System Generalized Method of Moments (System GMM) estimates

Notes: 1. Column (1) reports the baseline model estimated using the common sample (19 countries; 2008–2020; N = 247). 2. Columns (2)–(4) sequentially add EPS, urbanization, and renewable energy use, respectively. 3. All specifications use the same country-year sample. 4. Trade openness is mean-centered around its sample mean, and the interaction term is constructed using mean-centered trade openness. 5. ***, ** and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 6. Robust standard errors are reported in parentheses. 7. TCI = Technological Capacity Index.

The results show that the inclusion of additional control variables in the analysis does not substantially change the direction and magnitude of the interaction coefficient between technological capacity and trade openness. In the baseline model estimated using the same sample, the interaction coefficient is found to be −0.076 (p = 0.033). When EPS is controlled for, the coefficient is estimated at −0.072 (p = 0.019), and when urbanization is also included in the model, it is estimated at −0.070 (p = 0.024), with the negative sign and statistical significance being preserved. In the most comprehensive specification, in which renewable energy consumption is also included in the control set, the direction and magnitude of the interaction coefficient are largely preserved (β = −0.072), but statistical significance is lost (p = 0.123). Despite the absence of a notable change in the magnitude of the coefficient, the increase in the standard error of the interaction coefficient from 0.029 to 0.045 indicates that this weakening is more consistent with a reduction in estimation precision than with a substantial change in the estimated relationship. In addition, the null hypothesis of the AR(2) test cannot be rejected in all specifications, and the Hansen test does not reject the joint validity of the instrumental variables.

Finally, the marginal effects analysis presented in Table 9 and visualized in Figure 1 supports the main findings obtained from the interaction coefficient.

Figure 1 shows that the conditional marginal effect of Technological Capacity Index (TCI) on CO₂ emissions varies depending on the level of trade openness. The figure reveals that the marginal effect decreases steadily as trade openness increases and becomes negative and statistically significant at high levels of trade openness. These results are consistent with the marginal effects estimates reported in Table 9 and support that trade openness plays a conditioning role in the relationship between technological capacity and carbon emissions.

Table 9. Conditional marginal effects of Technological Capacity Index (TCI) across different levels of trade openness

Table 9 presents the conditional marginal effect of technological capacity (TCI) on CO₂ emissions at different levels of trade openness. The findings show that the marginal effect of TCI on CO₂ emissions is not constant. They reveal that this effect varies depending on the level of trade openness. At the lowest level of trade openness, the marginal effect of technological capacity on carbon emissions is positive but not statistically significant (β = 0.036, p = 0.499). At the sample mean, although this effect becomes negative, it still does not become statistically significant (β = −0.048, p = 0.105). In contrast, at the highest level of trade openness, the marginal effect of technological capacity becomes negative and statistically significant (β = −0.104, p <0.001). These findings indicate that the environmental benefits of technological capacity become more evident, particularly at high levels of international trade integration. In other words, trade openness emerges as an important conditioning factor shaping the effect of technological capacity on carbon emissions. These results provide empirical evidence that the complementary relationship between technological progress and trade integration is important for environmental sustainability.