Greener standards, cleaner production? Empirical evidence from ISO 14001 adoption

ISO 14001 is a globally recognized certifiable standard for environmental management systems (EMS), developed by the International Organization for Standardization (ISO), a worldwide federation of national standards bodies (ISO member bodies). The standard is designed to improve firms’ environmental performance by promoting systematic environmental management, continuous improvement, legal compliance, risk management, stakeholder engagement, and resource efficiency. The ultimate goal is to mitigate the environmental impacts of organizational activities, such as excessive resource consumption, pollution, waste generation, greenhouse gas emissions, and other harmful effects.

Despite widespread adoption, the actual effectiveness and impact of ISO 14001 certification remain matters of debate (Treacy et al., 2019). Most empirical research on ISO 14001 has focused on individual countries and collected cross-sectional data by distributing questionnaires to company managers (Boiral et al., 2018). Longitudinal studies based on broader international samples and grounded in measurable quantitative data are very scarce (Arocena et al., 2021, 2023). Moreover, although ISO 14001 is fundamentally oriented toward improving environmental outcomes, standard firm-level studies have predominantly focused on analyzing the impact of the certification on corporate financial performance, while its effects on environmental performance have received considerably less attention.

One particularly underexplored area is the effect of ISO 14001 adoption on corporate energy performance. As highlighted in our literature review, few empirical studies have directly examined this relationship, despite growing concerns among researchers about whether EMS standards genuinely encourage improvements in energy efficiency and the adoption of renewable energy (e.g., Laskurain et al., 2015). In fact, to our knowledge, no previous study has analyzed the effect of ISO 14001 adoption on renewable energy. Furthermore, most existing empirical studies rely on static ordinary least squares (OLS) and panel data models, which fail to account for the influence of past performance on current outcomes and thus omit the dynamics of energy performance.

This study aims to fill these gaps. Specifically, we assess the relationship between increased firm adoption of ISO 14001 on four dimensions of corporate energy and environmental performance: (i) energy intensity, defined as total direct energy consumption divided by output; (ii) total carbon emissions rate, measured as firm total greenhouse gases emissions per unit of energy used; (iii) carbon intensity from non-renewable energy sources, measured as firm greenhouse gases emissions per unit of non-renewable energy used; and (iv) utilization of renewable energy sources, measured as the proportion of renewable energy used relative to total energy consumption.

We use a system dynamic generalized method of moments (GMM) estimator on a panel of 512 multinational, non-service sector companies over the period 2013–2022 to examine whether broader adoption of ISO 14001is associated with reductions in energy intensity and carbon emissions, as well as increased reliance on renewable energy sources. The dynamic panel approach is particularly well-suited to this analysis, as it accounts for persistence (or state dependence) in the dependent variables—our energy performance metrics—while controlling for unobserved firm-level heterogeneity and potential endogeneity. This method estimates the contemporaneous association between ISO 14001 certification levels and performance metrics, conditioning on their past values, under the maintained system-GMM moment conditions. Furthermore, it facilitates the distinction between the short-term and long-term effects of ISO 14001 adoption on a firm’s energy and environmental performance. To our knowledge, this is the first study to comprehensively assess the role of ISO 14001 adoption in enhancing energy efficiency, lowering emissions, and encouraging the use of renewable energy sources. Our findings offer new insights into the benefits of EMS standards and contribute to a more nuanced understanding of ISO 14001’s effectiveness in practice.

The remainder of the paper is structured as follows. Sect. "The ISO 14001 framework and hypotheses development" outlines the theoretical framework and the development of hypotheses. Sect. "Review of the empirical literature on the impact of ISO 14001 adoption on environmental and energy performance" provides a comprehensive and up-to-date review of the empirical literature. Sect. "Methods" describes the methodology employed in the empirical analysis. Sect. "Data and variables" presents the data and defines the variables employed. Sect. "Results" reports the results. Finally, Sect. "Conclusion" discusses the main conclusions and their implications.

A significant body of research has analyzed the impact of ISO 14001 adoption on diverse financial performance metrics, including profitability, productivity, operating costs, stock price, Tobin’s Q, and fixed asset turnover.² Studies of the impact of ISO 14001 adoption on environmental performance are less abundant and follow two empirical approaches. The first bases the measurement of environmental performance on the perceptions of firm managers interviewed via questionnaires. A typical study using this approach asks managers of ISO 14001-certified plants/firms about the extent to which they believe their plant’s environmental performance has increased, decreased or remained unchanged following the adoption of the standard. While practitioners' opinions can provide valuable insights into their experiences in implementing the standard in their facilities, they do not rely on quantifiable measures of environmental impact and therefore do not allow a rigorous evaluation of the differential impact of ISO 14001 adoption across firms and time. As highlighted by Boiral et al. (2018), respondents’ opinions may be influenced by social desirability bias or self-report bias, but studies following this approach rarely address the risk of these biases.³ We therefore do not consider this stream of the literature. Instead, our interest is in studies following the second empirical approach: using quantitative measures of environmental impacts as dependent variables. Table 1 provides an updated review of these studies.

Several insights can be drawn from Table 1. First, with the exception of Arocena et al. (2021), all studies are country-specific. Second, some studies focus on specific industries, such as paper, electronics, and iron and steel. Third, some of the analyses are very basic statistical analyses that are limited to mean tests without additional controls. Only five studies deal with panel data models. Fourth, a brief examination of the last column of Table 1 reveals mixed and inconclusive results regarding the impact of ISO 14001 adoption on environmental performance. While some studies find a positive effect of ISO 14001-certified facilities on environmental performance compared to their non-certified counterparts, others find no statistically significant impact. Finally, more than half of the papers define a toxic chemical emissions index as the dependent variable to measure environmental impact (Aravind & Christman, 2011; Baek, 2018; Gómez & Rodríguez, 2011; King et al., 2005; Nishitani et al., 2012; Potoski & Prakash, 2005; Russo, 2009; Szymanski & Tiwari, 2004). Four studies use diverse indicators of CO₂ emissions (Arocena et al., 2021; Prasad & Mishra, 2017; Sam & Song, 2022; Testa et al., 2014), while one focuses on water pollutant emissions (Barla, 2007). Only three papers consider energy-related performance variables (Haider & Mishra, 2021; Jeong & Lee, 2022; Zobel, 2016).

Among the studies of energy-related performance, Zobel (2016) examined the impact of ISO 14001 adoption on the rates of change of different environmental performance dimensions in a sample of 114 Swedish manufacturing firms. The change in energy use was defined as the mean value of the percentage changes in the firm's use of electricity, fossil fuel, and total energy. The results of a t-test indicated that firms with ISO 14001 certification did not have superior performance compared to their non-certified counterparts. Haider and Mishra (2021) estimated the energy expenditure efficiency of 82 Indian iron and steel firms using stochastic frontier analysis. Instead of physical units, they measured energy in monetary terms, namely, firm energy expenditure on power and fuel. In a subsequent regression stage, the efficiency score was regressed on several explanatory variables, including a dummy variable indicating whether the firm had ISO 14001 certification. The results showed no statistically significant effect, suggesting that ISO 14001-certified companies did not have higher energy performance than non-certified firms.

Jeong and Lee (2022) provided the most comprehensive empirical analysis of the impact of ISO 14001 adoption EMS on energy efficiency. Specifically, they examined the effects of ISO 14001 adoption on the reciprocal of energy intensity, or energy productivity, within a panel of manufacturing plants in South Korea. They also explored the impact on emissions of certain air pollutants (NOx, SOx, and particulate PM-10), water pollutants (suspended solids, phosphorus, and nitrogen), and tons of solid waste. Interestingly, they found that ISO 14001 adoption had a negative impact on energy efficiency (productivity) but a positive impact on environmental outcomes. Furthermore, they discovered that ISO 14001 adoption led to increased use of environmental equipment by plants, which in turn reduced energy productivity. Thus, their research suggests the existence of a trade-off between energy efficiency and environmental outcomes.

We formulate the following dynamic panel data model:

Equation (1) is a p-order dynamic panel model, as it includes p lags of the dependent variable (\(\sum_{s}{\delta }_{s} {y}_{i,t-s})\) as a regressor on the right-hand side. In our analysis, the dependent variable \({(y}_{\text{i},\text{t}})\) represents one of four metrics for firm i at time t: energy intensity, greenhouse gas emission rate, renewable energy ratio, or carbon intensity from non-renewable energy sources. The primary explanatory variable of interest is ISO14001, which indicates firm i’s level of ISO 14001 certification at time t, \({X}_{\text{i},\text{t}}\) are other explanatory variables, and \({D}_{t}\) are year dummies. The error term in parentheses in Eq. (1) consists of two components, \({\eta }_{i} \sim i.i.d N\left(0, {\sigma }_{\eta }^{2}\right)\) and \({\epsilon }_{it} \sim i.i.d N\left(0, {\sigma }_{\epsilon }^{2}\right)\), and it is assumed to be serially uncorrelated i.e. \(E\left({\epsilon }_{is}{\epsilon }_{it}\right)=0\; \text{for }{\text{t}} \ne s\). The first term \({(\eta }_{i})\) is a firm-specific effect that accounts for unobserved influences on the dependent variable that are unique to each firm and remain constant over time. The second component \(({\epsilon }_{it})\) is the idiosyncratic error capturing time-varying, unit-specific random disturbances, and reflects the effects of factors that are not explicitly accounted for by the model and that vary across both units and time.

The rationale for including the lagged dependent variable among the regressors is to capture short-term dynamics. For instance, if we consider a firm’s energy intensity, the amount of energy consumed each year depends on factors such as available production technology and resources, which largely depend on investments and decisions made in the past. These factors may remain constant or change very slowly over time and thus have minimal short-term impact on energy intensity fluctuations. Consequently, the energy intensity for a given year is partially influenced by the previous year’s level. This rationale also applies to CO₂ emission rates and renewable energy usage. Therefore, incorporating the lagged dependent variable \({y}_{t-p}\) as a regressor for \({y}_{t}\) suggests that the past value of the dependent variable affects its current value. Thus, the coefficient \(\delta\) on the lagged term reflects a persistence or inertia effect, where the process evolves over time, with the current state partially depending on previous states. A high positive value of \(\delta\) in Equation [1] suggests that the dependent variable tends to stay close to its past values. Thus, if \(\delta\) is close to 1, the process shows high persistence or slow adjustment to shocks, indicating that inertia strongly influences the evolution of the dependent variable and the system adjusts slowly to change takes a while to adjust to changes. Conversely, a low value of \(\delta\) indicates that the system adjusts quickly to shocks or changes. Therefore, including at least one lag of the dependent variable in the estimation provides a measure of persistence in the dependent variable. Furthermore, this approach facilitates the comparison of the short-run (or contemporaneous) coefficients of the independent variables with their corresponding long-run effects, which can be calculated after estimation.

Some studies include the lagged dependent variable among the regressors in OLS and conventional fixed-effect (FE) or random-effect (RE) models to tackle the endogeneity issue that arises from reverse causality. However, this approach is not appropriate because \({y}_{it-1}\) is not a strictly exogenous regressor; by construction, the lag of the dependent variable \({y}_{it-1}\) is correlated with the fixed effect \({\eta }_{i}\). This makes estimators biased and inconsistent. Even when the within estimator wipes out \({\eta }_{i}\), the bias is determined by the correlation between \({y}_{i,t-1}\) and \({\overline{\epsilon }}_{i}\) (Nickell, 1981).

We employ the two-step system GMM proposed by Blundell and Bond (1998). Applying the dynamic panel data GMM estimator provides consistent and unbiased parameter estimates, assuming that unobserved heterogeneity remains constant or time-invariant. Arellano and Bover (1995) and Blundell and Bond (1998) introduced an approach that integrates equations in levels into the estimation process to form a stacked system comprising both first-differenced and level equations. This is known as system GMM. To be valid, the system GMM must fulfill the assumptions of first-difference GMM models (Arellano & Bond, 1991) as well as the additional assumption of stationarity:

The idea behind this assumption is that, instead of transforming the regressors to eliminate the unobserved heterogeneity (\({\eta }_{i}\)), the instruments are transformed to be exogenous to \({\eta }_{i}\). This exogeneity is achieved under the assumption that the correlation between the unobserved effect and the explanatory variables remains constant throughout the considered time period, which implies additional linear moment conditions for the equation in levels:

When the additional initial conditions assumptions hold true, adding these extra moment conditions for the equations in levels can significantly enhance efficiency and reduce finite sample bias, particularly in scenarios where \(\updelta\) approaches 1 or when the \({y}_{it}\) series displays increased persistence (Blundell & Bond, 2023).

Furthermore, unlike the instrumental variable (IV) methodology, which requires identifying external variables as instruments to address endogeneity issues, the system GMM method uses lagged values of the explanatory variables as instruments. These lags, by construction, are exogenous to the error term.

The primary data source employed in our empirical analysis is the LSEG (formerly Refinitiv) Eikon database, a widely used commercial platform providing standardized financial, environmental, social, and governance (ESG) information on publicly listed companies worldwide.⁴ LSEG compiles firm-level data from public reports, sustainability disclosures, and regulatory filings. In line with standard practice for dynamic GMM models, the panel length must remain relatively short, typically not exceeding ten years (Kiviet, 2020) to avoid over-identification, as the number of instruments tends to explode with the time dimension (Rodman 2009, p. 128). At the same time, a minimum of four years is required to test for the absence of second-order serial correlation. Accordingly, our sample includes only firms for which Eikon provides data on key variables—such as ISO 14001 certification—for at least four consecutive years. To address potential data anomalies, we removed outliers defined as observations falling more than three standard deviations from the industry mean, after grouping data by industry. The resulting dataset is an unbalanced panel of 512 companies operating in non-service sectors over the period 2013–2022, yielding a total of 3,128 firm-year observations. Table 2 presents a summary of the sample composition by industry and region.

We analyze the impact of ISO 14001 on four indicators: energy intensity (EI), greenhouse gases emission rate (GGR), renewable energy ratio (RER) and carbon intensity from non-renewable energy sources (CINRE). EI is defined as the total energy used by the firm, measured in gigajoules, divided by real output. Real output is the firm’s sales expressed in international dollars using purchasing power parity (PPP) rates, at constant 2017 prices. GGR encompasses the firm’s emissions of the seven greenhouse gases specified in the Kyoto Protocol: carbon dioxide (CO₂), methane (CH₄), nitrous oxide (N₂O), hydrofluorocarbons (HFCs), perfluorocarbons (PFCs), sulfur hexafluoride (SF₆) and nitrogen trifluoride (NF₃). The quantity of each gas is measured in tons of carbon dioxide equivalent (CO₂eq). GGR is computed by dividing the aggregate quantity of greenhouse gases emitted by the firm by its total energy consumption. RER is the ratio of energy from renewable energy sources to total energy used. Finally, CINRE is computed by dividing the total quantity of greenhouse gases emitted by the firm by the firm’s consumption of non-renewable energy.

At this point, we would like to emphasize that our analysis considers only the effect of ISO 14001 on a company’s direct energy consumption and emissions. In other words, we do not account for potential spillover effects that ISO 14001 adoption may have on the company’s suppliers. In this regard, Arimura et al. (2011) found that firms adopting ISO 14001 certification are more likely to implement advanced green supply chain management practices, such as assessing their suppliers’ environmental performance and requiring them to adopt specific environmental measures.

Our primary independent variable, ISO14001, measures the proportion of an MNE's facilities that have implemented the ISO 14001 certification, following Arocena et al., (2021, 2023). Consequently, this variable ranges from 0 for companies with no certified facilities to 1 for those with all facilities certified under the environmental standard. While the share of a firm’s facilities certified under ISO 14001 is a useful proxy for the intensity of adoption, there is no a priori reason to assume that this proportion translates linearly into performance improvements. For instance, a company might begin by certifying its largest, most strategic, or most visible plants and later extend certification to smaller or less critical facilities. In such a scenario, stronger impacts would likely occur at lower levels of certification intensity, followed by diminishing marginal effects as certification expands. Conversely, a firm might adopt a more cautious strategy, initially certifying smaller or easier-to-implement sites before targeting larger or more complex operations. In that case, performance impacts could increase as certification intensity grows. It is also reasonable to expect that early certifications might focus more on improving energy efficiency and altering consumption patterns, while investments in cleaner technologies and renewable energy, which often require a deeper level of commitment, might be undertaken only once higher levels of environmental certification are achieved.

To account for these possibilities, we include the squared term of the ISO 14001 certification variable in our empirical models. This allows us to test for the presence of non-linear effects and to assess whether marginal increases in certification intensity exhibit diminishing, increasing, or threshold patterns.

In Table 2, the last row of the sixth column indicates that 34% of the firms in the sample do not have any ISO 14001-certified plants. By contrast, 24% of companies have all their plants certified, as shown at the bottom of the seventh column. The last column of Table 2 shows that on average, firms in the sample have implemented the certification in 51% of their facilities.

Control variables include several firm- and country-specific factors that may influence corporate energy and environmental performance independently of ISO 14001 adoption. We account for the capital intensity of the firm (K/L), defined as the ratio of fixed assets to the total number of employees. Capital-intensive production processes are generally expected to consume more energy and produce higher carbon emissions compared to labor-intensive ones. However, a higher capital-labor ratio may also reflect the use of modern, energy-efficient equipment, resulting in reduced energy intensity (Arocena et al., 2024). Firm size (SIZE), measured by the number of employees, is included because larger firms typically have greater financial and organizational resources to invest in energy and environmental improvements and often face greater external pressure to demonstrate environmental responsibility. Additionally, we account for the degree of implementation of certified quality management systems (QMS), such as ISO 9001 certification. On the one hand, QMS can facilitate knowledge transfer whereby quality management practices contribute to more effective EMS (Albuquerque et al., 2007; Darnall et al., 2008; Vastag, 2004). On the other hand, a strong emphasis on product quality and customer satisfaction may lead to conflicting priorities, potentially diverting attention and resources away from energy saving initiatives and efforts and investments to reduce emissions. Similar to ISO14001, QMS ranges from 0 for companies with no certified facilities to 1 for those where all facilities are certified under the quality management standard.

In addition, the variable LAW accounts for the effectiveness of a country's legal system and its institutions in enforcing laws, as defined in the Rule of Law indicator developed by the World Bank.⁵ The variable is given in units of a standard normal distribution and ranges from −2.5 to 2.5. Stricter rule of law can be expected to contribute to the avoidance of violations and poor energy and environmental performance. All variables are logged, except for LAW and those that represent proportions, namely ISO14001, QMS and RER.

Note that with GMM estimation, time-invariant variables such as country and sector dummies are not included directly, as the first-differencing transformation eliminates them and their inclusion in the levels equation would create redundancy with the firm-specific effects. Instead, firm-level heterogeneity is addressed through differencing, and common time shocks are controlled by including year dummies.

Finally, to correctly implement the dynamic panel GMM methodology, we classify explanatory variables based on their expected correlation with the model's idiosyncratic error term. Thus, variables are categorized as endogenous, predetermined, or strictly exogenous with respect to the contemporaneous error term (Kiviet, 2020). Specifically, ISO14001, KL, SIZE, and QMS are treated as endogenous in all models. ISOCOUNTRY and LAW are treated as predetermined in the EI and GGR models, but as exogenous in the RER and CINRE models. This latter distinction is primarily driven by the smaller number of observations in the RER and CINRE models and the corresponding need to limit the number of instruments.

Table 3 provides a statistical summary, presenting the means, standard deviations, and correlation coefficients of the variables. To assess the potential for multicollinearity, we examine both the bivariate correlations and the variance inflation factor (VIF) values. The correlation coefficients across all variables are consistently very low, indicating a minimal risk of collinearity issues or redundancies. This conclusion is further supported by the VIF analysis, where the maximum VIF value is 1.30, well below the commonly accepted threshold of 10 (Kutner et al., 2005). These results indicate that multicollinearity is not a concern in our estimated models.

Table 4 presents the estimates of Eq. (1) for each dependent variable. Before delving into the estimates, let us first discuss the results of the standard tests related to the specification and validity of the dynamic system GMM model. All standard tests (Roodman, 2009) are shown at the bottom of Table 4.

The Arellano-Bond AR(1) and AR(2) tests assess first-order and second-order serial correlation in the first-differenced residuals under the null hypothesis of no serial correlation. In dynamic models, a significant consideration is the adequacy of the lags included to account for the dynamic components of the empirical relationship. If the incorporated lags are sufficient, any historical value beyond the lags can serve as a potentially valid instrument because it remains exogenous to current shocks. If the assumptions underlying our specification are correct, then the residuals in first differences, AR(1), are expected to exhibit correlation, whereas there should be no serial correlation in second differences AR(2). The result of these tests for the four models confirm the rejection of the null hypothesis of first-order serial correlation (p-values < 0.1), but we cannot reject the null hypothesis of second-order serial correlation (p-value > 0.1). This indicates that the use of a dynamic GMM specification is correct, whereas the OLS estimator is inconsistent.

The Hansen test of overidentification is under the null hypothesis that all instruments are valid. This test assesses the validity of the instruments employed and detects any absence of correlation between the instruments and the error term. All p-values are higher than 0.1, indicating that we cannot reject the null hypothesis. Note also that the number of instruments is well below the number of observations. Therefore, we can conclude that the established instruments for the three models are valid.

As mentioned above, the system GMM estimator assumes that any correlations between the endogenous variables and unobserved effects remain constant over time. This assumption is critical for incorporating the level equations and using lagged differences as instruments. To test this assumption, we employ the difference-in-Hansen test of exogeneity, which operates under the null hypothesis that the instruments used in the level equations are indeed exogenous. The validity of the system GMM estimator in our analysis hinges on failing to reject this null hypothesis. Table 4 shows the results of this test, with all p-values exceeding 0.1. This finding indicates that we cannot reject the null hypothesis, which supports the premise that the additional instruments used in the system GMM estimations are exogenous and validates the use of this estimator in our analysis.

Furthermore, Roodman (2009) pointed out that the estimated coefficient of the lagged dependent variable should lie within a specific range: the upper bound derived from OLS estimation and the lower bound obtained through the least squares dummy variable (LSDV) approach. Moreover, a credible estimate is expected to be below one, as values exceeding unity would suggest unstable dynamics, leading to accelerating divergence from equilibrium values. The results presented in the final rows of Table 4 confirm that all calculated values fall within this specified range.

Finally, it is important to note that all standard errors in our analysis are robust two-step standard errors adjusted using Windmeijer's (2005) correction. This adjustment is essential for ensuring the accuracy and reliability of the estimates, as it prevents the downward bias of the standard errors (Roodman, 2009).

Table 4 shows that the estimated coefficients for the lagged dependent variable are notably high across all four models. The first and third columns of Table 4 reveal that the coefficients in the EI and RER models are particularly high, at 0.966 and 0.937, respectively. These values confirm the significant persistence in energy intensity and the renewable energy ratio, suggesting that changes in energy use driven by factors such as ISO14001 will take considerable time to fully materialize. This long-run impact reflects the presence of strong inertia in both energy consumption patterns and the adoption of renewable energy sources. The lagged dependent variable in the CINRE model, while still significantly positive, is noticeably lower, as shown in column four. This suggests a lower degree of persistence in clean investment behavior, indicating that changes in this area may respond more quickly to influencing factors.

Unlike the other models, in the GGR model, a second lag of the dependent variable is included to capture the richer dynamic structure of the data and to adequately control for autocorrelation. Specifically, when only the first lag was included, the model failed the Arellano–Bond AR(2) test, indicating the presence of second-order serial correlation in the differenced residuals. As shown in Table 4, both lag coefficients are positive, indicating that past values of the dependent variable have a cumulative positive impact on the present value. This dynamic dependence is captured by the sum of the two lag coefficients \({\widehat{\delta }}_{1}+{\widehat{\delta }}_{2}=0.873\). This result suggests a strong persistence of the dependent variable, although it is slightly lower than those observed for EI and RER.

Regarding the interpretation of the coefficients of independent variables, Greene (2018) explained that adding dynamics to a model greatly alters the interpretation of the equation compared with static models. In a static model, where no lagged variable is included, the independent variables account for the full set of information that produces the observed outcome y_it. However, when a lagged dependent variable is added, “the equation now has the entire history of the right-hand-side variables, so that any measured influence is conditioned on this history; in this case, any impact of x_it represents the effect of new information” (Greene, 2018, pp. 523–524). As Piper (2023) highlights, the estimated coefficients reflect the short-term or contemporaneous impact of the independent variables conditional on the history of the model, which is captured by the lagged dependent variable.

At this point, it is important to note that the inclusion of exogenous and predetermined regressors helps mitigate, but does not fully eliminate, endogeneity concerns. The system GMM approach relies exclusively on internal instruments (lagged variables) and does not exploit any external exogenous variation in ISO 14001 adoption. Given the nature of our global panel, identifying comparable and consistent exogenous shocks across countries and industries (e.g., regulatory changes, policy interventions) is highly challenging, if not impossible. Nevertheless, the possibility that unobserved time-varying factors influence both certification decisions and environmental outcomes cannot be ruled out. Consequently, we interpret and present our findings as associations rather than definitive causal effects.

The first two columns of Table 4 present negative and statistically significant coefficients for ISO14001 in both EI and GGR models with values of −0.179 and −0.141, respectively. These results indicate that higher levels of ISO 14001 implementation are associated with more efficient energy consumption and lower greenhouse gas emissions per unit of energy used, thereby supporting hypotheses H1 and H2. Since ISO14001 is measured as a proportion (scaled from 0 to 1) and both EI and GGR are expressed in logarithmic form, the coefficients represent semi-elasticities. This means that a one percentage point increase in ISO 14001 adoption corresponds to a 0.179% reduction in EI and a 0.141% reduction in GGR in the short run.

Additionally, the third column of Table 4 reports a positive and statistically significant coefficient for ISO14001 in the RER model. This indicates that greater adoption of the standard is associated with increased use of renewable energy sources, supporting hypothesis H3. In this model, both the dependent variable RER and ISO14001 are proportions. Therefore, a one percentage point increase in ISO14001 leads to a 0.055 percentage point increase in RER.

Table 4 also shows that the coefficient for (ISO14001)^2 is statistically significant only in the RER model. This implies that, with the exception of RER, increases in the certification share exhibit no evidence of non-linear effects. By contrast, in the RER model, the positive sign of the squared term suggests the presence of increasing effects of ISO 14001 certification on renewable energy adoption. This is particularly interesting, as it indicates that renewable energy adoption is more strongly associated with firms exhibiting higher levels of ISO 14001 certification. Specifically, while the marginal effect of ISO 14001 is 0.098 at the median, it rises to 0.338 at the third quartile. As suggested above, this pattern likely reflects that a higher share of ISO 14001 certification is associated with a greater commitment to investing in more ambitious environmental initiatives.

Regarding the control variables, K/L is statistically significant only in the CINRE model, suggesting that more capital-intensive firms emit less CO₂ per unit of non-renewable energy consumed. QMS exhibits a positive and statistically significant coefficient in the GGR model, indicating that broader implementation of quality management systems is associated with a higher emissions rate. ISOCOUNTRY displays the expected negative sign, although it is statistically significant only in the EI model. Finally, LAW shows statistically significant negative coefficients in both the GGR and CINRE models, suggesting that more stringent environmental legislation is associated with a reduction in emissions intensity.

The long-run effect of a time-varying independent variable is obtained by dividing its short-run estimated coefficient \(\widehat{\beta }\) by (\(1-\widehat{\delta }\)), where \(\widehat{\delta }\) is the estimated coefficient of the lagged dependent variable. In other words, the short-run effect is scaled by the factor \(\frac{1}{\left(1-\widehat{\delta }\right)}\). For the EI and RER, the long-run effects of ISO14001 are −5.26 and 0.87, respectively. Thus, a one percentage point increase in ISO14001 is associated with a 5.26% in EI decrease in the long run, and 0.87 percentage point increase in RER after accounting for dynamic feedback effects.

In the case of the GGR model, which includes two lags of the dependent variable, the long-run effect of ISO14001 is calculated by dividing the estimated coefficient by \(\left(1-{\widehat{\delta }}_{1}-{\widehat{\delta }}_{2}\right)\), specifically \(-0.141/\left(1-0.715-0.158\right)=-1.11\). This implies that a one percentage point increase in ISO14001 results in a 1.11% long-run decrease in GGR.

As noted in our literature review, several previous studies employing static OLS and FE models have failed to find a significant effect of ISO 14001 adoption on energy-related outcomes. However, these approaches fail to model dynamics, which we have shown, play a critical role in this context. Ignoring such dynamics can lead to omitted variable bias and misattribution of effects. Static estimators treat all explanatory variables as strictly exogenous—an assumption that is unlikely to hold here, given plausible presence of reverse causality and omitted variable concerns. Variables such as ISO 14001 adoption, capital intensity, and firm size are likely influenced by past performance and may reflect unobserved firm characteristics. Failing to account for these dynamics introduces bias and inconsistency in the estimates.

For comparison, we estimate static OLS and FE models using our dataset. The results are presented in Table 5. As shown, ISO 14001 is not statistically significant in any of the models, with the exception of the CINRE model, where it appears with a positive and significant coefficient—suggesting that higher levels of ISO 14001 implementation are associated with increased emissions intensity from non-renewable energy sources. Additionally, Table 6 reports the results from OLS and FE models that include lagged dependent variables. ISO14001 remains mostly statistically insignificant, except for a marginally significant negative effect in the EI model. However, as Wintoki et al. (2012) note, while this specification improves explanatory power—as evidenced by higher R² values relative to those in Table 5—these models remain inconsistent due to the correlation between the lagged dependent variable and the unobserved fixed effects (Nickell, 1981).

As we have stated earlier, the system GMM estimator is specifically designed to address these concerns. It accounts for endogeneity, unobserved heterogeneity, and dynamic feedback effects. Based on both the diagnostic tests and the theoretical advantages of the estimator, the system GMM results provide more robust and reliable evidence, forming a sound basis for the conclusions and policy implications discussed in the next section.

In this paper, we analyze the effectiveness of implementing ISO 14001 in organizations with respect to energy utilization, a topic that has been largely overlooked in the literature. The scant empirical evidence available remains inconclusive regarding the actual impact of implementation of the standard on energy efficiency and emissions reduction. Our study distinguishes itself from previous research by employing dynamic modelling through a system GMM estimation approach. The results indicate that greater implementation of ISO 14001 is associated with lower firms’ energy intensity and an increased share of renewable energy use, both of which contribute to lowering greenhouse gas emission rates. However, we find no evidence that higher levels of ISO 14001 certification result in cleaner (i.e., lower carbon) consumption of non-renewable energy sources (i.e., fossil fuels). These results suggest that firms in our sample have not adopted ISO 14001 merely as a symbolic or reputational gesture, as might be expected under a greenwashing strategy.

Our analysis also confirms a strong persistence in all dependent variables, highlighting the significant cumulative effect of ISO 14001 over time. This is consistent with the notion that environmental certification should be seen not as a one-time goal, but as a continuous learning process, involving progressive stages, each marked by specific challenges and critical success factors (Boiral, 2011). Furthermore, the long-run influence of ISO 14001 adoption as captured by the lagged dependent variables, suggests that static modelling approaches would likely produce biased estimates and fail to capture the full impact of certification.

Our findings suggest several managerial and policy considerations. From a managerial perspective, the evidence that increased ISO 14001 implementation is associated with lower energy intensity and greater reliance on renewable energy sources suggests that firms can derive tangible operational and environmental benefits from adopting the standard. These outcomes support the view that ISO 14001 is more than a symbolic or reputational tool; rather, it can serve as a substantive mechanism for driving energy-related performance improvements. Managers should therefore approach ISO 14001 not as a compliance formality but as a strategic investment in long-term energy efficiency and sustainability. Moreover, the persistence observed in outcomes further underscores the importance of sustained commitment and continuous improvement. To be effective, ISO 14001 should be integrated into broader organizational routines and learning processes, with expectations for gradual but cumulative returns over time.

The data also reveal that 34% of the firms in our sample lack any certification, while only 24% have achieved full certification, with the remaining firms holding partial certifications. This indicates considerable potential for broader adoption of certified EMS. It is important to note that the firms analyzed are large, publicly listed international companies with substantial resources, making them more likely to adopt and implement the environmental standard in a meaningful way. The adoption of ISO 14001 can entail significant costs, which may be a barrier, especially for small and medium-sized enterprises (SMEs). As Bansal and Bogner (2002) suggest, the costs of implementation can sometimes outweigh the perceived economic benefits of the standard. Moreover, the internalization of ISO procedures can face resistance within organizations due to a lack of resources, personnel, training, and knowledge. This often results in a poor adaptation of ISO systems to the specific needs of an organization, limiting their effectiveness. Arocena et al. (2023) argue that SMEs are more likely to adopt the standard symbolically, whereas large firms are better positioned to undertake substantive implementation.

From a policy perspective, our findings point to an opportunity for public decision-makers to implement targeted interventions to encourage the adoption and effective implementation of voluntary EMS certifications. Policymakers could focus on reducing the costs and enhancing the benefits of EMS implementation. Examples of such policies could include offering training programs for ISO 14001 and providing fiscal incentives and financial support, particularly for SMEs and companies in specific industries. Additionally, more direct interventions could aim to create a more demanding economic environment regarding EMS certification adoption. For instance, policymakers might require EMS certification as a prerequisite for companies to qualify as suppliers in the public sector or to gain access to certain public programs, such as investment incentives for R&D. These initiatives should be designed to generate spillover effects throughout companies’ supply chains, encouraging broader environmental improvements (Arimura et al., 2011).

Nevertheless, the absence of effects on the carbon intensity of fossil fuel consumption points to a limitation of ISO 14001 in addressing certain dimensions of the energy transition, particularly in fossil fuel-dependent sectors. That is, ISO 14001 is likely to be a more useful tool for reducing emissions in sectors with greater potential for electrification, while its effectiveness appears limited when it comes to encouraging investments in cleaner fossil fuel processes or emissions abatement technologies. This is particularly relevant for firms where the structural challenges of electrifying industrial processes hinder the transition to renewable electricity—especially those that are heavily reliant on fossil energy, such as in the cement or chemical industries. This suggests that ISO 14001, while valuable, may need to be complemented by targeted policies—such as carbon pricing, sector-specific decarbonization mandates, or investment support for clean technologies—to fully address emissions from non-renewable energy use.

Finally, the dynamic nature of the observed effects in this study emphasize the need for policymakers and researchers to move beyond static assessments when evaluating the environmental impacts of voluntary standards. Time-lagged impacts and cumulative learning effects are essential components of the ISO 14001 trajectory, and policy evaluations should account for these dimensions when designing or assessing interventions aimed at promoting voluntary environmental certification schemes.

This study has several limitations. First, we lack data on the age of ISO 14001 certifications within each company. It is plausible that differences in energy and environmental performance may arise due to a learning curve associated with implementing the standard over time, which we do not account for. This limitation also extends to broader organizational and operational changes driven by ISO 14001 certification in the long term, which could also affect energy intensity. Second, we do not control for whether the companies in our sample adopted other certified EMSs, such as the European Eco-Management and Audit Scheme (EMAS), or energy management systems like ISO 50001, which could significantly influence energy performance. Third, as noted earlier, our sample consists of large companies, with an average size exceeding 14,000 employees. As such, caution is warranted when generalizing these results to SMEs. Four, dynamic system GMM is a data-intensive estimation technique that requires a panel data with a “small T, large N” structure. The number of cross-sectional units (N) in our database is not sufficiently large to allow for investigation of heterogeneity across subsamples, such as industries or regions. Future research using larger and more granular datasets could provide deeper insights into sectoral or geographical heterogeneity in greater depth. Addressing these limitations should be a priority for future studies.