Assessing the Impact of Environmental Technology on CO₂ Emissions in Saudi Arabia: A Quantile-Based NARDL Approach

Abstract

Climatic change and environmental degradation have become a worldwide discourse. Green innovation is commonly viewed as a means of lowering environmental pollution in the era of climate change. Considering this, the primary purpose of this study is to investigate the effects of environmental technology (ET) on CO₂ emissions by controlling Saudi Arabia’s ICT use, energy use, energy intensity, and financial development. It uses a quantile-based multiple-threshold “nonlinear autoregressive distributed lag (NARDL)” estimation utilizing data from 1990 to 2020. It also conducts the ARDL and NARDL estimation techniques simultaneously for comparative outcomes. The Toda–Yamamoto (T-Y) causality assessment also crosschecks the primary multiple-threshold NARDL estimates. The outcomes reveal that ET promotes environmental pollution due to its low scale compared to the Kingdom’s technological base. ICT improves environmental quality, and energy consumption deteriorates it. All three estimation techniques confirm these findings. The multiple-threshold NARDL estimation appears robust and reveals damaging impacts of energy intensity and financial development on emissions. The T-Y causality assessment also authenticates the primary estimation outcomes. The outcomes have important implications for policymakers to focus on enhancing patents for ET, raising ICT diffusion, reducing energy intensity through generating more renewable energies, expanding financial support for ICT and green investments, and ensuring a sustainable environment.

1. Introduction

Global warming-related climate change and pollution of the environment have become recognized phenomena worldwide. The main contributor to climate change and environmental degradation has been human activity, particularly the use of fossil fuels. Therefore, policymakers worldwide have been concerned about contrasting climate occurrences, especially environmental contamination [1]. Against this backdrop, economists and decision-makers acknowledge green innovation as a potent instrument for lowering environmental pollution. Green innovation, in simple terms, refers to ecologically sensible inventions that improve the environment and ensure the efficient use of natural resources. It is a scientific development that lessens energy usage, reduces pollutant releases, and eventually enhances the environment. Many countries seek out green innovation and environmentally friendly industrial processes because they have a substantial capacity to handle environmental concerns. Research shows that green innovation can reduce CO₂ emissions and create a clean environment [2].

Environmental quality directly impacts both human life and the ecological system; environmental degradation damages people’s emotional and physical well-being, leading to a wide range of illnesses and disorders in addition to low-income levels. Climate change, rising temperatures, and animal species disappearance result from environmental degradation, a worldwide problem [3,4]. As a result, the United Nations targets lowering environmental pollution as one of the top priorities of the sustainable development goals. CO₂ emission is a widely used indicator of environmental pollution. An environment’s quality is evaluated based on whether pollutant emissions are present; the environment is of poor quality if there are significant CO₂ emissions and vice versa. Therefore, we measure the level of CO₂ released into the air to verify the environmental quality, and we work to understand how green innovation may lower the prevalence of CO₂ emissions. We focus on environmental technology (ET) as a stand-in for green innovation that might reduce CO₂ emissions and deliver a pristine environment for Saudi Arabia. Despite ET being widely available, if its reach is not sufficiently enlarged, it may not be capable of decreasing pollutant emissions [5]. As a result, this study aims to look into how Saudi Arabia’s CO₂ emissions have changed due to environmental technologies. Figure 1 illustrates ET and CO₂ emissions per capita in Saudi Arabia during 1990–2020. The percentage of ET in the Kingdom’s total technologies was consistently low and averaged 18.89% throughout the survey period [6], whereas the level of CO₂ emissions per capita has shown a rising trend.

Similarly, over the past three decades, “information and communication technology (ICT)” has made significant advancements, and empirical data suggests that using ICT to drive economic development is beneficial [8]. Additionally, due to globalization, many nations increasingly rely on ICT to boost work and energy efficiency across manufacturing and industry sectors [9]. ICT does affect the environment significantly, in addition to its excellent benefits for economic growth and human advancement. Many studies have investigated how ICT affects environmental pollution. Several studies examining how ICT affected the environment found that it affected environmental quality positively and negatively and even remained neutral. Thus, research on ICT’s impact on the environment has been inconclusive. Therefore, further research is needed to understand how ICT alters environmental quality.

Similarly, financial development affects CO₂ emissions in various ways. Financial growth results in easy credit and speeds up buying cars and home appliances for a convenient lifestyle; these products and goods release more CO₂ into the atmosphere through energy sources [10]. Easy credit in times of financial expansion brings possibilities for expanding business enterprises, which are energy-intensive and cause more CO₂ emissions [11]. However, financial growth may also hasten investments in R&D and green technology, reinforcing a favorable effect on environmental quality. Therefore, it may influence a country’s environmental quality in either a favorable or an unfavorable way, which is why this research explores financial expansion’s effect on Saudi Arabia’s CO₂ emissions.

Over the past 30 years, there have been substantial spikes in energy use due to rapid economic growth. Energy consumption is the main factor contributing to increased environmental pollution, which puts humankind and the environment at risk. The impact of energy use on environmental excellence has been a topic of many investigations, since energy use, specifically conventional energy, has contributed significantly to the degradation of the environment worldwide. As Saudi Arabia’s primary energy sources include petroleum and natural gas, it appears helpful to research how energy use affects Saudi Arabia’s air pollution. Of its total energy usage, 99.93% came from non-renewable energy sources in 2022 [12]). After implementing Vision 2030, Saudi Arabia has taken steps to further engage in significant initiatives to diversify its economy away from the oil sector. Thus, energy has become a crucial element of its vast development initiatives, which are often likely to compromise environmental quality. Hence, more investigation is required into how energy consumption influences environmental degradation in Saudi Arabia.

Energy intensity indicates how inefficiently an economy uses energy and is calculated using energy consumption per output unit. High energy intensity reveals high GDP costs or prices when utilizing energy and vice versa. However, high industrial output in terms of GDP results from substantial energy intensity, which is the source of increased environmental pollution, while low-energy-intensity countries have labor-intensive economies. The energy intensity of a product is determined by the amount of energy needed to make it; hence, utilizing less energy to produce it lowers the intensity. The impact of energy intensity on economic growth is widely studied, but its environmental effects must be well researched. This study examines how energy intensity will likely affect the Kingdom’s environment.

Therefore, this study considers ET’s symmetric and asymmetric impressions on environmental pollution, controlling for ICT, energy usage, energy intensity, and financial expansion employing the quantile-based multiple-threshold NARDL approach for the first time. This study makes several significant contributions that cover several key areas. Primarily, it is a pioneering effort to investigate the relationship between pollution and ET, focusing on Saudi Arabia, a critical energy-dependent nation. Secondly, this research contributes to the scholarly debate on pollution and ET by implementing a quantile-based decomposition of ET data. Finally, this study employs the multiple-threshold NARDL approach to unveil the nexus between decomposed ET data and pollution levels within the Saudi Arabian context for the first time. The expected results promise to help Saudi Arabia’s officials create workable plans to reduce environmental pollution in this energy-rich country. The following sections make up the remainder of the study. Section 2 appraises the literature, Section 3 explains the data and techniques, Section 4 offers the results and findings, and Section 5 concludes.

2. Literature Review

Linking ET to CO₂ emissions is a relatively recent phenomenon, and only a handful of studies exist in this field. In this section, we review the pertinent literature currently available and present our conclusion.

2.1. ET and CO₂ Emissions

Zhang et al. [13] used 30 provinces’ longitudinal data from 2000 to 2013 and the system GMM approach to underscore the consequence of environmental innovation for CO₂ releases in China. The results showed that EG increased carbon emissions, whereas environmental innovation typically decreased CO₂ discharges. Hashmi and Alam [14] investigated the impact of ET patents in 29 nations utilizing panel data from 1999 to 2014. The study determined that while a surge in ET patents by 1% reduced CO₂ emissions by 0.017%, an increase in income per person increased pollution levels.

Ahmad et al. [15] utilized panel data between 1990 and 2014 with an FMOLS technique and investigated, among other things, the impacts of innovation shocks on environmental degradation in 26 OECD countries. They revealed that positive innovation shocks enhanced environmental quality, and adverse innovation shocks aggravated it. Using data from 2000 to 2014, Ganda [16] employed the system GMM technique and examined the effect of invention and technology outlays on CO₂ emissions in 26 OECD nations. They uncovered that while R&D spending was negatively associated with CO₂ releases, economic growth, energy usage, and credit to the private sector had a positive connection.

Using data from the BRICS nations from 1980 to 2016, Khattak et al. [17] investigated the link between innovation, renewable energy, and CO₂ emissions. Employing the CCEMG method, they found that innovation could not reduce CO₂ emissions in four countries, except Brazil, while EG increased pollution among them. Ahmad et al. [18] explored the relationship between innovation—measured by R&D investment and the energy–environment–EG nexus—in 24 OECD nations using data from 1993 to 2014 and a simultaneous equation model. The study could not demonstrate any environmental effect of innovation and remarked that reducing pollutant emissions would not be successful if R&D spending was just directed at increasing output. It would have the desired environmental impact if directed only at reducing pollution. In a study of the asymmetries between ET and CO₂ emissions in highly decentralized countries, Lingyan et al. [19] used population, GDP, and fiscal decentralization as controls and the “Method of the Moments Quantile” estimate. They discovered that EG increased pollution while ET reduced CO₂ emissions in the middle to upper emissions quantiles.

Using quarterly data, Xin et al. [20] looked at the asymmetric consequence of green technology innovation on CO₂ releases in the USA. They showed that during periods of expansion, positive shocks to innovation in ET reduced CO₂ emissions, whereas adverse shocks to innovation in ET during periods of decline increased CO₂ emissions. Using annual data for the USA from 1980 to 2018 and the quantile ARDL technique, Sun et al. [21] evaluated the role of globalization and eco-innovation in cutting CO₂ emissions and controlling GDP. They noted that while GDP and globalization increased CO₂ emissions, eco-innovation decreased them.

Jiang et al. [22] used BRICS’s data for the 1985–2018 period, applied a “dynamic common correlated effect mean group (DCCEMG)” method, and analyzed the influence of ET on consumption-based CO₂ emissions. They reported that ET had an adverse long-term effect on emissions. Using data from the BRICS economies from 1990 to 2014 and the AMG estimate, Ali et al. [23] explored the link between green innovation and CO₂ emissions using FDI, energy usage, and EG as control factors. They found that FDI, energy usage, and EG increased emissions and green innovation considerably decreased them. Umar and Safi [24] investigated how green innovation and finance affected CO₂ emissions in the OECD nations, among other things. They showed that green innovation and financing significantly reduced emissions. Islam et al. [25] investigated the influence of remittance outflow on CO₂ emissions in Saudi Arabia, controlling ICT and environmental innovation, among others. Using the NARDL approach, the study revealed ICT’s positive impact on environmental quality and the negative effect of innovation’s adverse shocks. Islam [2] investigated the uneven impact of ET on CO₂ emissions controlling EG, energy usage, trade liberalization, and financial growth in Saudi Arabia using the NARDL estimate. The study reported that the small scale of the ET could not lower emissions.

2.2. Energy Use, Intensity, and CO₂ Emissions

Numerous studies have demonstrated the link between energy consumption and pollutant outputs. The impact of energy use on environmental excellence is a topic of many investigations since energy use, specifically conventional non-renewable energy, has contributed significantly to the degradation of the environment worldwide. Such studies, among others, include Islam [3] for South Asia, Ahmad et al. [18] for Pakistan, Ali et al. [23] for BRICS economies, Acheampong [26] for Sub-Saharan African (SSA) countries, Khan et al. [27] for BRICS countries, Zhao et al. [28] for China, and Adeleye et al. [29] for South Asia. These studies revealed a positive contribution of energy use to environmental degradation.

The impact of energy intensity on the environment has yet to be researched. Despite significant volumes of research available on the effects of energy use on the atmosphere, the consequence of energy intensity on the latter has received less attention, perhaps because energy usage and energy intensity are seen through the same lens. A few recent studies, such as Danish et al. [30] for the United States, Shokoohi et al. [31] for the Middle Eastern region, and Zhang et al. [32] for Morocco, examined the effects of energy intensity on environmental quality and reported its significant damaging impact on the environment. This study thus examines how energy and its intensity will likely affect the Kingdom’s environment.

2.3. ICT and CO₂ Emissions

Many studies have examined how ICT affects environmental pollution. Several studies examined ICT’s impact on the environment. The studies that found ICT’s positive influence on environmental quality include Islam and Rahaman [8] for GCC nations, Zhang and Liu [33] for China, Lu [34] for 12 Asian countries, Godil et al. [35] for Pakistan, Chien et al. [36] for BRICS countries, Ebaidalla and Abusin [37] for GGC nations, Shehzad et al. [38] for Pakistan, and Islam et al. [39] for the GCC region. These studies found ICT diffusion to be environmentally friendly.

There is another strand that highlights the devastating role of ICT. According to this strand, there is evidence that ICT deployments have resulted in environmental degradation. ICT utilization was shown, for instance, to harm environmental quality and to be the cause of carbon emissions into the atmosphere by Park et al. [40] for selected EU countries, Asongu et al. [41] for SSA economies, Arshad et al. [42] for South and Southeast Asian economies, Alatas [43] for a sample of 93 countries, and Appiah-Otoo et al. [44] for a sample of 110 economies. Appiah-Otoo et al. [44] revealed that ICT enhanced environmental sustainability in nations with high ICT quality but deteriorated the environment in countries with moderate and low ICT quality.

However, there is another strand: ICT has been neither environment-promoting nor environment-deteriorating; instead, it has been environmentally neutral. For example, Amri et al. [45], employing Tunisian data collected between 1975 and 2014, noted that ICT did not impact carbon dioxide emissions and remained environmentally neutral.

Anser et al. [46], using the ARDL bounds test and data from 1970 to 2018, investigated the impact of technological factors on CO₂ emissions in Saudi Arabia, considering GDP, industry value added, trade openness, energy use, and population density. They showed that computer communications and telephone and mobile subscriptions helped to reduce emissions, while technical cooperation grants enhanced them. Thus, research regarding the impact of ICT on CO₂ emissions is inadequate, and further research is needed to understand how ICT alters environmental quality.

2.4. Financial Development and CO₂ Emissions

The literature discusses various ways that financial development affects CO₂ emissions. The prominent viewpoint regarding FD’s contribution to CO₂ emissions is that financial expansion enhances environmental degradation. This strand of research highlights the easy availability of credit owing to financial development, which speeds up buying cars and home appliances and leads to CO₂ in the atmosphere. Several studies, including Ahmad et al. [18] for the Belt and Road Initiative countries, Acheampong [26] for SSA nations, Omri et al. [47] for the MENA region, Yang et al. [48] for the BICS nations, Islam [49,50] for South Asia and top remittance-receiving countries, and Habiba et al. [51] for emerging nations observed that FD increased emissions and thus degraded the environment.

However, financial growth may also hasten investments in R&D and green technology, which can reduce pollution levels and favor environmental quality. Several instances of such findings exist, including Abbasi and Riaz [52] for Pakistan, Tao et al. [53] for the OECD nations, and Shang et al. [54] for China. These studies observed how financial development helped environmental promotion in different countries.

However, there is evidence that FD is environmentally neutral; it does not influence the environment. For instance, Charfeddine and Kahia [55] witnessed FD’s neutral influence on MENA countries. Thus, financial expansion may affect a country’s environmental quality in either a favorable or an unfavorable way, or it may remain neutral.

The above-cited studies that linked ET to CO₂ emissions used several controlling variables. Most considered a panel setup, while only a few employed time-series country-level data. As a result, country-level studies are limited. Moreover, the results of previous research are mixed; most studies revealed the beneficial effects of ET on CO₂ emissions, and several reported insignificant negative impacts on environmental quality. Therefore, substantial research on the ET–CO₂ emissions nexus is mandatory to mitigate environmental pollution.

In the context of Saudi Arabia, the country has an apparent research gap in this field. Islam [2] and Islam et al. [25] are the only studies that have uncovered how ET asymmetrically contributed to environmental pollution in Saudi Arabia using NARDL modeling. The present study is a pioneering attempt to examine how ET may contribute to environmental pollution in a quantile-based multiple-threshold framework employing different control variables. Thus, our study differs from existing research, offers new insights into the ET–pollution nexus, and provides further implications for Saudi Arabian policymakers.

Saudi Arabia ranks among the top energy producers worldwide and is collaborating on environmentally relevant technologies. Despite this, there needs to be more research on how ET may affect the quality of the environment in the Kingdom, and there remains a research gap. Therefore, this study attempts to fill this research gap in the following ways. ➀ This study is the first to use a quantile-based multiple-threshold NARDL model to investigate how ET affects the Kingdom’s environmental quality. ➁ We also conduct the ARDL and NARDL estimation procedures to obtain comparative outcomes. Thus, our attempt closes an existing knowledge gap and researches the impact of ET on CO₂ discharge by controlling ICT, energy usage, energy intensity, and financial advancement in Saudi Arabia utilizing a quantile-based multiple-threshold NARDL approach.

3. Data and Methods

CO₂ emissions data are calculated using metric tons per person. Following Islam et al. [39] and Islam and Rahaman [8], ICT data are generated into an index using five variables (mobile cellular subscriptions, fixed broadband subscriptions, the population that uses the internet, fixed telephone subscriptions, all per 100 individuals, and the share of medium- and high-tech exports in all manufactured exports) based on principal component analysis (PCA).

The proportion of total technologies in the nation relevant to the environment is leveled as environmental technology (ET) and is utilized as a stand-in for green innovation. Financial expansion (FD) is quantified using monetary sector credit to the private sector as a fraction of GDP. Energy consumption (EC) is measured in quadrillion Btu. Energy intensity (EI) is calculated as energy consumption per unit of GDP based on purchasing power parities and expressed in thousand Btu per US dollar.

The data on CO₂, ICT, and FD are obtained from the “World Bank” [7]; data on ET are sourced from the OECD [6]; and data on EC and EI are taken from the EIA [12]. The variables are stated, and their statistics are included in

Table 1. Variables with their statistics.

Variable	Specification	Mean	Max	Min	SD.	Obs.
CO2	CO2 releases are expressed in metric tons per person.	13.35	17.26	10.71	2.09	31
ET	It is measured as a portion of the nation’s total technologies related to the environment.	18.84	62.50	4.66	9.89	31
ICT	ICT index is generated based on the PCA.	0.00	3.44	−2.29	1.95	31
EI	Energy consumption per unit of GDP is based on purchasing power parities and expressed in thousand Btu per US dollar.	5.72	7.42	4.10	1.00	31
EC	Total energy use is computed in quadrillion Btu	6.95	11.16	3.34	2.80	31
FD	Monetary sector credit to the private segment as a ratio of GDP is utilized as a proxy for financial expansion.	34.04	58.11	14.82	12.57	31

.

In Equation (1), a model states CO₂ as a function of ET, controlling for ICT, EC, EI, and FD.

$${CO}_2=f\left(ET,ICT,EI,EC,FD\right)\dots \dots \dots$$

(1)

We have used 31 years of annual data from the above sources [6,7,12] during 1990–2020, suitable for running a time-series analysis. The period of data is determined by the availability of OECD data on ET, which are available (ET data for Saudi Arabia are available up to the year 2020. We have used the 1990–2020 period; the 1994 and 1996 data are missing. We generated the missing data by averaging the preceding and following years’ data). Some standard unit root tests are used to assess the stationarity of the variables, which is necessary for conducting a time-series analysis. In addition, a breakpoint unit root test is also performed.

3.1. ARDL Model

First, we utilize the ARDL approach by Pesaran and Shin [56] and Pesaran et al. [57]. The long-run and short-run versions of the ARDL model of Equation (1) are demonstrated in Equations (2) and (3).

$$\begin{array}{ll}\hfill {{CO}_2}_t={\alpha }_0& +{\displaystyle \sum _1^l}{{\alpha }_1{CO}_2}_{t-1}+{\displaystyle \sum _0^m}{\alpha }_2{ET}_{t-1}+{\displaystyle \sum _0^n}{\alpha }_3{ICT}_{t-1}+{\displaystyle \sum _0^o}{\alpha }_4{EC}_{t-1}+{\displaystyle \sum _0^p}{{\alpha }_{5}EI}_{t-1}\\ & +{\displaystyle \sum _0^q}{{\alpha }_{6}FD}_{t-1}+e_{1t}\dots \dots .\end{array}$$

(2)

$$\begin{array}{ll}\hfill {{∆CO}_2}_t={\alpha }_0& +{\displaystyle \sum _1^l}{{\alpha }_1∆{CO}_2}_{t-1}+{\displaystyle \sum _0^m}{\alpha }_2{∆ET}_{t-1}+{\displaystyle \sum _0^n}{\alpha }_3{∆ICT}_{t-1}+{\displaystyle \sum _0^o}{\alpha }_4{∆EC}_{t-1}+{\displaystyle \sum _0^p}{{\alpha }_5∆EI}_{t-1}\\ & +{\displaystyle \sum _0^q}{{\alpha }_6∆FD}_{t-1}+{\beta ECT}_{t-1}+e_{2t}\dots \dots .\end{array}$$

(3)

With the F test, a null hypothesis of no association (H₀: ${\alpha }_1$ = ${\alpha }_2$ = ${\alpha }_3$ =${\alpha }_4$ = ${\alpha }_5$ = ${\alpha }_6$ = 0) is evaluated. A long-term association between the variables is established, and the null hypothesis is turned down if the F-statistic goes beyond the critical upper bound value.

3.2. Nonlinear ARDL Model

In the second step, we employ the nonlinear ARDL (NARDL) model suggested by Shin et al. [58] to capture the asymmetric impacts of the explanatory variable(s) on the dependent variable. This approach divides the explanatory variable ET into negative and positive partial sum series, as explained in Equations (4) and (5).

$${ET\_N}_t={\displaystyle \sum _{i=1}^t}{∆ET\_N}_t={\displaystyle \sum _i^t}min\left({∆ET}_i,0\right)\dots \dots \dots$$

(4)

$${ET\_P}_t={\displaystyle \sum _{i=1}^t}{∆ET\_P}_t={\displaystyle \sum _i^t}max\left({∆ET}_i,0\right)\dots \dots \dots$$

(5)

The NARDL model is an extended version of the simple ARDL model by Pesaran and Shin [56] and Pesaran et al. [57]. The long-run and short-run versions of the NARDL model in Equation (1) are demonstrated in Equations (6) and (7).

$$\begin{array}{ll}\hfill {{CO}_2}_t={\alpha }_0& +{\displaystyle \sum _1^l}{{\alpha }_1{CO}_2}_{t-1}+{\displaystyle \sum _0^m}{\alpha }_2{ET\_N}_{t-1}+{\displaystyle \sum _0^n}{\alpha }_3{ET\_P}_{t-1}+{\displaystyle \sum _0^o}{\alpha }_4{ICT}_{t-1}+{\displaystyle \sum _0^p}{{\alpha }_{5}EC}_{t-1}\\ & +{\displaystyle \sum _0^q}{{\alpha }_{6}EI}_{t-1}+{\displaystyle \sum _0^r}{{\alpha }_{7}FD}_{t-1}+e_{1t}\dots \dots .\end{array}$$

(6)

$$\begin{array}{ll}\hfill {{∆CO}_2}_t={\alpha }_0& +{\displaystyle \sum _1^l}{{\alpha }_1{∆CO}_2}_{t-1}+{\displaystyle \sum _0^m}{\alpha }_2∆{ET\_N}_{t-1}+{\displaystyle \sum _0^n}{\alpha }_3{∆ET\_P}_{t-1}+{\displaystyle \sum _0^o}{\alpha }_4{ICT}_{t-1}\\ & +{\displaystyle \sum _0^p}{{\alpha }_5∆EC}_{t-1}+{\displaystyle \sum _0^q}{{\alpha }_6∆EI}_{t-1}+{\displaystyle \sum _0^r}{{\alpha }_7∆FD}_{t-1}+{\beta ECT}_{t-1}+e_{2t}\dots \dots .\end{array}$$

(7)

3.3. Multiple-Threshold NARDL Model

Verheyen [59] first utilized the multiple-threshold NARDL model, which Pal and Mitra [60] used in a recent study. We have followed Verheyen [59] and Pal and Mitra [60] and decomposed the explanatory variable ET into three parts based on the data series’ quantiles. The quantile-based NARDL model strategy provides more in-depth insights than the ARDL and NARDL approaches.

To run the bi-threshold NADRL method, three partial sum series, ET_25, ET_25$~$75, and ET_75, are generated (ET_25 is the first quantile, ET_75 is the fourth quantile, and ET_25$~$75 combines the second and third quantiles) by establishing two thresholds at the 25th (referred to as ω25) and 75th (referred to as ω75) quantiles, which are designated based on the following Equations (8)–(10).

$${ET\_25}_t={\displaystyle \sum _{i=1}^t}∆{ET}_i={\displaystyle \sum _i^t}∆{ET\_25}_i\left({∆ET}_i\le \mathsf{\omega }25\right)\dots \dots$$

(8)

$${ET\_25~75}_t={\displaystyle \sum _{i=1}^t}∆{ET}_i={\displaystyle \sum _i^t}∆{ET\_25~75}_i\left(\mathsf{\omega }25<{∆ET}_i\le \mathsf{\omega }75\right)\dots$$

(9)

$${ET\_75}_t={\displaystyle \sum _{i=1}^t}∆{ET}_i={\displaystyle \sum _i^t}∆{ET\_75}_i\left({∆ET}_i>\mathsf{\omega }75\right)\dots \dots$$

(10)

Following the bi-threshold decompositions, the NARDL model of Equation (1) is shown in its long- and short-term variants in Equations (11) and (12).

$$\begin{array}{ll}\hfill {{CO}_2}_t={\alpha }_0& +{\displaystyle \sum _1^l}{{\alpha }_1{CO}_2}_{t-1}+{\displaystyle \sum _0^m}{\alpha }_2{ET\_25}_{t-1}+{\displaystyle \sum _0^n}{\alpha }_3{ET\_25~75}_{t-1}+{\displaystyle \sum _0^n}{\alpha }_4{ET\_75}_{t-1}{\displaystyle \sum _0^o}{\alpha }_5{ICT}_{t-1}\\ & +{\displaystyle \sum _0^p}{{\alpha }_{6}EC}_{t-1}+{\displaystyle \sum _0^q}{{\alpha }_{7}EI}_{t-1}+{\displaystyle \sum _0^r}{{\alpha }_{8}FD}_{t-1}+e_{1t}\dots \dots .\end{array}$$

(11)

$$\begin{array}{ll}\hfill ∆{{CO}_2}_t={\alpha }_0& +{\displaystyle \sum _1^l}{{\alpha }_1{∆CO}_2}_{t-1}+{\displaystyle \sum _0^m}{\alpha }_2{∆ET\_25}_{t-1}+{\displaystyle \sum _0^n}{\alpha }_3∆{ET\_25~75}_{t-1}\\ & +{\displaystyle \sum _0^n}{\alpha }_4{∆ET\_75}_{t-1}{\displaystyle \sum _0^o}{\alpha }_5{ICT}_{t-1}+{\displaystyle \sum _0^p}{{\alpha }_6∆EC}_{t-1}+{\displaystyle \sum _0^q}{{\alpha }_7∆EI}_{t-1}+{\displaystyle \sum _0^r}{{\alpha }_8∆FD}_{t-1}\\ & +{\beta ECT}_{t-1}+e_{2t}\dots \dots .\end{array}$$

(12)

where l, m, …. r each represent the optimum lag length of the respective variables using the Akaike information criteria (AIC). Figure 2 shows the plots of variable ET under three different modeling scenarios.

We check the unit root properties and ensure that the variables are stationary to implement the ARDL, NARDL, and multiple-threshold NARDL model estimations. All three approaches work well for any integrating order of the variables: I(0), I(1), or a mix of them. The dependent variable (CO₂) must be I(1) to run the above-mentioned techniques.

The models apply equally to large and small samples [61]. We use the bounds test version of the models to validate the long-run connection among relevant variables. Due to the small size of our sample—only 31 observations—we use the reference values of Narayan [62] for the F-statistic value instead of the critical values of Pesaran et al. [57]. A long-run correlation is confirmed if the anticipated F-statistic is higher than Narayan’s [62] crucial values. The long-term link between variables is then presented at level forms. An “error correction model (ECM)” is incorporated to show the short-term dynamics across the three approaches. ECT refers to “error correction term”; it seamlessly integrates all short-term data into long-term characteristics. If $\beta$ < 0 and is significant, a long-run association is verified, and the significant coefficient value for each regressor recognizes its short-run connection.

To detect the direction of causality among variables and check the robustness of three ARDL approaches, we apply the Granger causality assessment based on VAR developed by the “Toda–Yamamoto (T-Y)” method [63], which is thought to be better than the usual Granger causality assessment.

$$\begin{array}{ll}\hfill {{CO}_2}_t={\alpha }_{01}& +{\displaystyle \sum _1^{l+d_{max}}}{{\alpha }_1{CO}_2}_{t-1}+{\displaystyle \sum _0^{m+d_{max}}}{\beta }_1{ET\_25}_{t-1}+{\displaystyle \sum _0^{n+d_{max}}}{\gamma }_1{ET\_25~75}_{t-1}\\ & +{\displaystyle \sum _0^{0+d_{max}}}{\delta }_1{ET\_75}_{t-1}+{\displaystyle \sum _0^{p+d_{max}}}{\epsilon }_1{ICT}_{t-1}+{\displaystyle \sum _0^{q+d_{max}}}{{ϵ}_{1}EC}_{t-1}+{\displaystyle \sum _0^{r+d_{max}}}{{\theta }_{1}EI}_{t-1}\\ & +{\displaystyle \sum _0^{s+d_{max}}}{{\vartheta }_{1}FD}_{t-1}+u_1\dots \dots \end{array}$$

(13)

$$\begin{array}{ll}\hfill {ET\_25}_t={\alpha }_{02}& +{\displaystyle \sum _1^{l+d_{max}}}{{\alpha }_1{CO}_2}_{t-1}+{\displaystyle \sum _0^{m+d_{max}}}{\beta }_1{ET\_25}_{t-1}+{\displaystyle \sum _0^{n+d_{max}}}{\gamma }_1{ET\_25~75}_{t-1}\\ & +{\displaystyle \sum _0^{0+d_{max}}}{\delta }_1{ET\_75}_{t-1}+{\displaystyle \sum _0^{p+d_{max}}}{\epsilon }_1{ICT}_{t-1}+{\displaystyle \sum _0^{q+d_{max}}}{{ϵ}_{1}EC}_{t-1}+{\displaystyle \sum _0^{r+d_{max}}}{{\theta }_{1}EI}_{t-1}\\ & +{\displaystyle \sum _0^{s+d_{max}}}{{\vartheta }_{1}FD}_{t-1}+u_2\dots \dots \end{array}$$

(14)

$$\begin{array}{ll}\hfill {ET\_25~75}_t={\alpha }_{03}& +{\displaystyle \sum _1^{l+d_{max}}}{{\alpha }_1{CO}_2}_{t-1}+{\displaystyle \sum _0^{m+d_{max}}}{\beta }_1{ET\_25}_{t-1}+{\displaystyle \sum _0^{n+d_{max}}}{\gamma }_1{ET\_25~75}_{t-1}\\ & +{\displaystyle \sum _0^{0+d_{max}}}{\delta }_1{ET\_75}_{t-1}+{\displaystyle \sum _0^{p+d_{max}}}{\epsilon }_1{ICT}_{t-1}+{\displaystyle \sum _0^{q+d_{max}}}{{ϵ}_{1}EC}_{t-1}+{\displaystyle \sum _0^{r+d_{max}}}{{\theta }_{1}EI}_{t-1}\\ & +{\displaystyle \sum _0^{s+d_{max}}}{{\vartheta }_{1}FD}_{t-1}+u_3\dots \end{array}$$

(15)

$$\begin{array}{ll}\hfill {ET\_75}_t={\alpha }_{04}& +{\displaystyle \sum _1^{l+d_{max}}}{{\alpha }_1{CO}_2}_{t-1}+{\displaystyle \sum _0^{m+d_{max}}}{\beta }_1{ET\_25}_{t-1}+{\displaystyle \sum _0^{n+d_{max}}}{\gamma }_1{ET\_25~75}_{t-1}+{\displaystyle \sum _0^{0+d_{max}}}{\delta }_1{ET\_75}_{t-1}\\ & +{\displaystyle \sum _0^{p+d_{max}}}{\epsilon }_1{ICT}_{t-1}+{\displaystyle \sum _0^{q+d_{max}}}{{ϵ}_{1}EC}_{t-1}+{\displaystyle \sum _0^{r+d_{max}}}{{\theta }_{1}EI}_{t-1}\\ & +{\displaystyle \sum _0^{s+d_{max}}}{{\vartheta }_{1}FD}_{t-1}+u_4\dots \end{array}$$

(16)

$$\begin{array}{ll}\hfill {ICT}_t={\alpha }_{05}& +{\displaystyle \sum _1^{l+d_{max}}}{{\alpha }_1{CO}_2}_{t-1}+{\displaystyle \sum _0^{m+d_{max}}}{\beta }_1{ET\_25}_{t-1}+{\displaystyle \sum _0^{n+d_{max}}}{\gamma }_1{ET\_25~75}_{t-1}\\ & +{\displaystyle \sum _0^{0+d_{max}}}{\delta }_1{ET\_75}_{t-1}+{\displaystyle \sum _0^{p+d_{max}}}{\epsilon }_1{ICT}_{t-1}+{\displaystyle \sum _0^{q+d_{max}}}{{ϵ}_{1}EC}_{t-1}+{\displaystyle \sum _0^{r+d_{max}}}{{\theta }_{1}EI}_{t-1}\\ & +{\displaystyle \sum _0^{s+d_{max}}}{{\vartheta }_{1}FD}_{t-1}+u_5\dots \dots \end{array}$$

(17)

$$\begin{array}{ll}\hfill {EC}_t={\alpha }_{06}& +{\displaystyle \sum _1^{l+d_{max}}}{{\alpha }_1{CO}_2}_{t-1}+{\displaystyle \sum _0^{m+d_{max}}}{\beta }_1{ET\_25}_{t-1}+{\displaystyle \sum _0^{n+d_{max}}}{\gamma }_1{ET\_25~75}_{t-1}+{\displaystyle \sum _0^{0+d_{max}}}{\delta }_1{ET\_75}_{t-1}\\ & +{\displaystyle \sum _0^{p+d_{max}}}{\epsilon }_1{ICT}_{t-1}+{\displaystyle \sum _0^{q+d_{max}}}{{ϵ}_{1}EC}_{t-1}+{\displaystyle \sum _0^{r+d_{max}}}{{\theta }_{1}EI}_{t-1}+{\displaystyle \sum _0^{s+d_{max}}}{{\vartheta }_{1}FD}_{t-1}+u_6\dots \dots \end{array}$$

(18)

$$\begin{array}{ll}\hfill {EI}_t={\alpha }_{07}& +{\displaystyle \sum _1^{l+d_{max}}}{{\alpha }_1{CO}_2}_{t-1}+{\displaystyle \sum _0^{m+d_{max}}}{\beta }_1{ET\_25}_{t-1}+{\displaystyle \sum _0^{n+d_{max}}}{\gamma }_1{ET\_25~75}_{t-1}+{\displaystyle \sum _0^{0+d_{max}}}{\delta }_1{ET\_75}_{t-1}\\ & +{\displaystyle \sum _0^{p+d_{max}}}{\epsilon }_1{ICT}_{t-1}+{\displaystyle \sum _0^{q+d_{max}}}{{ϵ}_{1}EC}_{t-1}+{\displaystyle \sum _0^{r+d_{max}}}{{\theta }_{1}EI}_{t-1}+{\displaystyle \sum _0^{s+d_{max}}}{{\vartheta }_{1}FD}_{t-1}+u_7\dots \dots \end{array}$$

(19)

$$\begin{array}{ll}\hfill {FD}_t={\alpha }_{08}& +{\displaystyle \sum _1^{l+d_{max}}}{{\alpha }_1{CO}_2}_{t-1}+{\displaystyle \sum _0^{m+d_{max}}}{\beta }_1{ET\_25}_{t-1}+{\displaystyle \sum _0^{n+d_{max}}}{\gamma }_1{ET\_25~75}_{t-1}+{\displaystyle \sum _0^{0+d_{max}}}{\delta }_1{ET\_75}_{t-1}\\ & +{\displaystyle \sum _0^{p+d_{max}}}{\epsilon }_1{ICT}_{t-1}+{\displaystyle \sum _0^{q+d_{max}}}{{ϵ}_{1}EC}_{t-1}+{\displaystyle \sum _0^{r+d_{max}}}{{\theta }_{1}EI}_{t-1}+{\displaystyle \sum _0^{s+d_{max}}}{{\vartheta }_{1}FD}_{t-1}+u_8\dots \dots \end{array}$$

(20)

Equations (13)–(20) delineate the test procedure. The T-Y causality analysis accepts any order of integration among variables, including I(0), I(1), or I(2).

4. Results and Findings

4.1. Unit Root Assessment Results

The standard unit root tests are run to confirm the series’ attributes; the results are shown in

Table 2. Unit root test results.

Variable	ADF		PP		KPSS
	Level	∆	Level	∆	Level	∆
CO2	−1.272	−3.452 **	−1.151	−3.688 *	0.624	0.16 ***
ET	−3.329 **	-	−5.651 *	-	0.089 ***	-
ICT	0.655	−5.086 *	0.998	−5.083 *	0.704	0.252 ***
EC	−1.462	−2.222	−0.319	−3.646 **	0.696	0.166 ***
EI	−1.570	−6.154 *	−1.497	−6.180 *	0.601	0.252 ***
FD	−0.949	−4.452 *	−0.683	−8.513 *	0.706	0.450 **

Note: * p < 0.01, ** p < 0.05, and *** p < 0.10.

. The series are of multiple orders of integration because they are stationary at levels and first difference. The dependent variable (CO₂) is I(1), which is useful and allows us to perform ARDL, NARDL, and multiple-threshold NARDL estimations. The ADF and PP test results are used for the primary assessments, while the KPSS assessment results are referred to apply the “T-Y causality” check solely.

We also assess the breakpoint unit roots (UT) test employing an innovation outlier with a break and trend specification intercept. The outcomes are stated in

Table 3. Unit root with breakpoints.

Variable	ADF		Berak Date
	Level	∆
CO2	−3.641	−4.454 **	2017
ET	−6.698 *		1994
ICT	−1.476	−6.135 *	2017
EC	−1.462	−6.920 *	2017
EI	−7.192 *		2012
FD	−2.342	−6.039 *	2015

Note: * p < 0.01, ** p < 0.05.

. The UT test results illustrate that the variables are stationary and have no problems. Therefore, we proceed to the step-by-step estimation process outlined in the methodology section.

4.2. ARDL Estimates

The model is determined based on automatic lag specification, with a maximum of one lag for the explanatory factors and two for the explained variable. An ARDL (2, 0, 0, 0, 1, 0) model is selected, and the estimated output based on “Case 3: Unrestricted constant and no trend” is exhibited in

Table 4. ARDL bounds test, long-run and short-run outputs, and diagnostic results.

Test Statistic	Value	Significance	I(0)	I(1)
F-statistic	7.217 *		Finite sample: N = 30
k	5	5%	3.125	4.608
Actual N	29	1%	4.537	6.37
Variable	Coefficient	Std. Error	t-Statistic	p-Value
ET	0.028	0.007	3.931	0.001 *
ICT	−0.577	0.232	−2.485	0.022 **
EI	−0.064	0.141	−0.453	0.656
EC	1.206	0.173	6.988	0.000 *
FD	−0.019	0.016	−1.218	0.238
ECM Regression
C	3.461	0.496	6.979	0.000 *
∆(CO2(−1))	0.359	0.090	3.999	0.001 *
∆(EC)	1.684	0.167	10.086	0.000 *
ETC(−1)	−0.675	0.092	−7.357	0.000 *
Diagnostic Assessment	Statistic	Value	p-Value	Decision
“Breusch–Pagan–Godfrey”	Obs. R2	7.689	0.464	Homoscedastic
Normality	Jarque–Berra	1.156	0.561	Normal
Breusch–Godfrey	Obs. R2	3.605	0.165	No serial correlation
Ramsey RESET	F-statistic	0.947	0.343	Stable
CUSUM †				Stable
CUSUM of squares †				Stable

Note: * p < 0.01; ** p < 0.05. † See Figure 3.

.

It is hypothesized that there is no association between the variables of interest at level form under the null hypothesis. Nevertheless, a long-term connection between the variables is confirmed because the calculated F-statistic value [7.217] is substantially higher than the crucial values at various significance levels and rejects the null hypothesis. A long-run association among variables at a 1% level is established.

A significant positive coefficient of ET shows that the impact of environmental technology on air pollution is contributory, which is a stunning and unusual finding. Usually, ET is targeted to minimize the environmental impact, and thus, it must reduce environmental pollution. However, our findings are quite the opposite of what was expected and contradict all existing literature. While Khattak et al. [17] and Ahmad et al. [18] reported an insignificant impact of ET on the environment, Zhang et al. [13], Hashmi and Alam [14], Ahmad et al. [15], Ganda [16], Lingyan et al. [19], Xin et al. [20], Sun et al. [21], Jiang et al. [22], and Ali et al. [23] suggested that ET helped to reduce environmental pollution; our findings sharply contrast with them.

The coefficient of ICT is statistically significant; a rise in the value of the ICT index is likely to scale down CO₂ emissions and improve environmental quality. This outcome is consistent with Islam and Rahaman [8], Zhang and Liu [33], Lu [34], Godil et al. [35], Chien et al. [36], Ebaidalla and Abusin [37], Shehzad et al. [38], and Islam et al. [39], who reported an environmentally friendly role of ICT. This finding, however, opposes Park et al. [40], Asongu et al. [41], Arshad et al. [42], Alatas [43], and Appiah-Otoo et al. [44], who established its worsening effect on the environment. Our findings also contradict Amri et al. [45], who witnessed ICT to be ecologically neutral. The result has important implications for policymakers to raise the stake of ICT to mitigate environmental pollution and enhance its quality.

Energy use substantially impacts pollutant emissions, as evidenced by the positive and significant EC constant. The Kingdom primarily uses non-renewable energy sources for consumption, and the results are as expected, given the minimal use of renewable energy. Our finding supports the findings of previous studies in the literature, including Islam [3], Ahmad et al. [18], Ali et al. [23], Acheampong [26], Khan et al. [27], Zhao et al. [28], and Adeleye et al. [29], who demonstrated that energy use had a positive impact on CO₂ emissions.

When it comes to oil energy, the Kingdom has the upper hand. Moving from conventional to renewable energy sources to reduce CO₂ emissions may be difficult because Saudi Arabia is the world’s second-biggest oil producer and exporter. Given the country’s year-round access to ample solar and wind supplies, the Kingdom’s authorities may consider sourcing more energy from these sources. In addition to reducing environmental pollution, this will create new prospects for regional economic expansion.

The positive coefficient of EI is insignificant, showing its no impact on pollution. Usually, EI is expected to damage environmental properties and affect CO₂ emissions positively. However, our findings contradict the available literature and the recent results of Danish et al. [30], Shokoohi et al. [31], and Zhang et al. [32], who demonstrated a damaging effect of energy intensity on environmental quality.

FD delivers a negative yet insignificant coefficient, indicating its noncontribution to pollution. Thus, based on the ARDL assessment, FD remains environmentally neutral. Our finding aligns with Charfeddine and Kahia [55], who highlighted FD’s neutral influence on CO₂ emissions. However, this finding contradicts most literature, including Omri et al. [47], Acheampong [26], Ahmad et al. [18], Yang et al. [48], Islam [49,50], and Habiba et al. [51], who observed that FD accelerated emissions and worsened the environment. Similarly, our result also contrasted with Abbasi and Riaz [52], Tao et al. [53], and Shang et al. [54], who revealed that financial development helped environmental promotion. However, if responsible organizations and authorities take appropriate steps, financial expansion may lessen environmental damage. The findings might persuade policymakers to utilize financial institutions to introduce green financing and direct investments to increase the scope and stake of ET and achieve its considerable environmental impact.

The ECM estimation further shows that the one-period lag value of the dependent variable (CO₂) and EC affect emissions positively in the short run. The model is stable and fulfills all diagnostic criteria with a 67.5% speed of adjustment toward long-term stability.

4.3. One-Threshold NARDL Estimate

The single-threshold NARDL model is determined based on automatic lag specification, with a maximum of one lag for both the explained and explanatory variables. The chosen model is ARDL (1, 0, 0, 1, 1, 1, 0). The estimated outcomes based on “Case 3: Unrestricted constant and no trend” are exhibited in

Table 5. NARDL bounds test and long-run, short-run, and diagnostic outputs.

Test Statistic	Value	Significance	I(0)	I(1)
F-statistic	5.383	Finite sample: N = 30
k	6	10%	2.457	3.797
Actual N	29	5%	2.97	4.499
Variable	Coefficient	Std. Error	t-Statistic	p-Value
ET_N	0.052	0.013	3.882	0.001 *
ET_P	0.024	0.011	2.244	0.037 **
ICT	−0.894	0.299	−2.988	0.008 *
EI	0.886	0.428	2.071	0.052 ***
EC	1.131	0.279	4.052	0.001 *
FD	0.038	0.035	1.105	0.283
Long-run asymmetric test outcomes
Test statistic	Value	df	p-value
t-statistic	3.0967	19	0.006 *
F-statistic	9.589	(1, 19)	0.006 *
Chi-square	9.589	1	0.002 *
H0: C(1) = C(2); Ha: C(1) ≠ C(2)
ECM regression
C	0.861	0.173	4.989	0.000 *
∆(ICT)	−0.855	0.174	−4.903	0.000 *
∆(EI)	0.003	0.095	0.032	0.975
∆(EC)	1.498	0.163	9.165	0.000 *
ETC(−1)	−0.545	0.090	−6.032	0.000 *
Diagnostic Assessment	Statistic	Value	p-Value	Decision
“Breusch–Pagan–Godfrey”	Obs. R2	13.781	0.183	Homoscedastic
Normality	Jarque–Berra	0.721	0.697	Normal
Breusch–Godfrey	Obs. R2	3.69	0.158	No serial correlation
Ramsey RESET	F-statistic	3.162	0.093	Stable
CUSUM †				Stable
CUSUM of squares †				Stable

Note: * p < 0.01, ** p < 0.05, and *** p < 0.10; † See Figure 3.

.

ET, ICT, and EC’s impacts on environmental pollution are the same as those of the simple ARDL estimation. Both the positive and negative components of ET positively contribute to pollution levels. Thus, the analysis in Section 4.2 regarding the impact of the target variable (ET), including two control variables, ICT and EC, holds. The impact of FD on the latter remains insignificant, as earlier.

However, this ET outcome partly aligns with and contradicts Islam et al. [25] and Islam (2024). According to these two studies, the positive component of ET was statistically insignificant, and the negative components were pollution-enhancing. However, this study’s findings show that both positive and negative components of ET promote pollution. Similarly, the current findings also partly differ and partly contrast with Xin et al. [20], who revealed that ET’s positive shocks reduced CO₂ emissions, and its negative shocks increased CO₂ emissions.

In addition, ET’s negative and positive components generate positive coefficients that look alike. The Wald test has been used to check for the potential asymmetry between two coefficients of decomposed positive and negative elements. The results of the long-run asymmetric test, presented in Table 5, reject the hypothesis of no asymmetry. The coefficients of the positive and negative components are asymmetric over the long run.

The positive coefficient of EI indicates its assertive influence on emissions, as opposed to the ARDL estimation finding. Our NARDL finding asserts that an increase in energy intensity exerts pressure on air pollution and damages the environmental quality. This finding conforms with existing literature, such as Danish et al. [30], Shokoohi et al. [31], and Zhang et al. [32], who showed that energy intensity hurts the quality of the environment. Hence, any rise in energy intensity will adversely affect the environment in the long run. The outcome suggests that policymakers must look for alternative, efficient energy sources to reduce their energy intensity and mitigate environmental pollution.

Further evidence is that ICT asserts the same negative impact on emissions in the short run as it does in the long run. EC and EI have short-term positive impacts on emissions, similar to those in the long run. No short-run asymmetry is noticed between ET’s decomposed positive and negative components, as their coefficients are insignificant in the short run. With a 54.5% adjustment rate toward long-term stability, the model is stable and satisfies all diagnostic requirements.

4.4. Multiple-Threshold NARDL Estimates

The bi-threshold NARDL model is determined based on automatic lag specification, with a maximum of one lag for both the explained and explanatory variables. The chosen model is ARDL (1, 0, 0, 1, 0, 1, 1, 0). The estimated results are exhibited in

Table 6. Bi-threshold NARDL bounds test and long-run, short-run, and diagnostic outcomes.

Test Statistic	Value	Significance	I(0)	I(1)
F-statistic	6.022	Finite sample: N = 30
k	7	10%	2.384	3.728
Actual N	29	5%	2.875	4.445
Variable	Coefficient	Std. Error	t-Statistic	p-Value
ET_25	0.049	0.011	4.520	0.000 *
ET_25~75	0.132	0.019	6.989	0.000 *
ET_75	0.043	0.011	4.048	0.001*
ICT	−0.854	0.151	−5.674	0.000 *
EI	0.760	0.195	3.904	0.001 *
EC	1.086	0.111	9.750	0.000 *
FD	0.027	0.015	1.782	0.093 ***
Long-run asymmetric test outcome
Test statistic	Value	df	p-value
F-statistic	5.796	(2, 17)	0.012 **
Chi-square	11.592	2	0.003 *
H0: C(1) = C(2) = C(3); Ha: C(1) ≠ C(2) ≠ C(3)
ECM Regression
C	1.982	0.273	7.256	0.000 *
∆(ET_75)	0.020	0.003	6.534	0.000 *
∆(EI)	0.278	0.087	3.184	0.005 *
∆(EC)	1.541	0.124	12.441	0.000 *
ECT(−1)	−0.959	0.116	−8.247	0.000 *
Diagnostic Assessment	Statistic	Value	p-Value	Decision
“Breusch–Pagan–Godfrey”	Obs. R2	18.025	0.081	Homoscedastic
Normality	Jarque–Berra	1.082	0.582	Normal
Breusch–Godfrey	Obs. R2	0.614	0.736	No serial correlation
Ramsey RESET	F-statistic	0.001	0.980	Stable
CUSUM †				Stable
CUSUM of squares †				Stable

Note: * p < 0.01, ** p < 0.05, *** p < 0.10; † Figure 3.

.

The impacts of decomposed ET, ICT, EC, and EI on environmental pollution are very close to the single-threshold NARDL estimation outcomes. All three decomposed components of ET positively influence air pollution. Despite their positive impacts, the magnitudes of their impacts on the environment are not identical. Therefore, we have conducted the Wald test to verify any possible asymmetry among the coefficients of three decomposed variables. The outcomes of the long-run asymmetric test reported in Table 6 show that the hypothesis of no asymmetry is rejected. A long-run asymmetry exists among the coefficients of quantile-based decomposed variables.

The limited application of ET compared to total technologies in the country is one potential cause of its damaging environmental impact. Numerous studies have supported similar conclusions, showing that expenditures in innovation intended to boost output rather than improve the environment worsen the environment due to their uneven consequences (Mensah et al., [64]; Santra, [65]). Environment-related technologies comprise a relatively small portion of the Kingdom’s comprehensive technology, as mentioned in Section 1.

The result has significant implications for the Kingdom’s policymakers, who should increase ET’s share in the overall mix of technologies to improve the environment. To achieve this, enticements such as easy financing for environmentally friendly innovation may be promoted, increasing ET’s stake and lowering environmental pollutants.

Thus, the analysis in Section 4.3 regarding the impact of the target variable (ET) and three control variables, i.e., ICT, EC, and EI, holds. The influence of FD on CO₂ emissions becomes significant and positive. This finding is consistent with Omri et al. [47], Acheampong [26], Ahmad et al. [18], Yang et al. [48], Islam [49,50], and Habiba et al. [51], who revealed that FD increased emissions and deteriorated the environment. However, our findings differ from Abbasi and Riaz [52], Tao et al. [53], and Shang et al. [54], who stated a favorable contribution of financial development to the environment. Thus, following the bi-threshold NARDL estimation, the impacts of all control variables match the expected outcomes. This outcome will likely support policymakers in directing investments to green finance and increasing the scope and share of ET to achieve its considerable environmental impact.

Additionally, it proves that the short-term favorable effects of ET_75, EC, and EI on emissions are comparable to their long-term effects. There is no short-run asymmetry among the decomposed variables, as two of the three factors are insignificant. Assuming a 95.9% adjustment rate toward long-term stability, the model meets all diagnostic criteria and is stable.

Based on the above analyses, the bi-threshold NARDL estimates appear more robust than the ARDL and single-threshold NARDL estimates.

4.5. Causality Assessment

The optimum lag order is 2 using several criteria, including “LR, FPE, AIC, SC, and HQ”. The T-Y assessment generates several bidirectional and one-way causations; we have reported the relevant ones in

Table 7. T-Y causality outcomes.

Sl.	Causality	Chi-sq	df	p-Value	Direction
1	ET_25 to CO2	7.327	2	0.026 **	ET_25→CO2 One-way
	CO2 to ET_25	0.336	2	0.845
2	ICT to CO2	7.815	2	0.020 **	ICT↔ CO2 Two-way
	CO2 to ICT	7.267	2	0.026 **
3	EC to CO2	7.387	2	0.025 **	EC↔ CO2 Two-way
	CO2 to ENC	11.761	2	0.003 *
4	FD to CO2	11.365	2	0.003 *	FD→CO2 One-way
	CO2 to FD	1.073	2	0.585
5	EI to CO2	3.866	2	0.145	CO2→EI One-way
	CO2 to EI	9.649	2	0.008 *
6	ET_75 to ET_25	7.320	2	0.026 **	ET_75↔ ET_25 Two-way
	ET_25 to ET_75	8.904	2	0.012 **
7	ET_25~75 to EI	10.250	2	0.006 *	ET_25~75→EI One-way
	EI to ET_25~75	0.516	2	0.772
8	ET_75 to CO2	3.679	2	0.159	CO2→ET_75 One-way
	CO2 to ET_75	6.311	2	0.043 **
9	EI to FD	7.411	2	0.025	EI→FD One-way
	FD to EI	1.279	2	0.528

Note: * p < 0.01, ** p < 0.05.

.

The first causality runs from ET’s first quartile to pollution (ET_25→CO₂); it shows that ET at its first quartile unidirectionally causes pollution, and the reverse does not hold. The second two-way causation occurs between ICT and carbon emissions (ICT↔ CO₂). This implies that ICT significantly causes air pollution and vice versa. Both ET and ICT cause carbon emissions.

The third bidirectional causality occurs between energy consumption and pollutant emissions (EC↔ CO₂). There is a feedback association between two variables—energy consumption and environmental pollution—that cause each other. The fourth causation runs from FD to pollution (FD→CO₂), revealing their unidirectional association. It means financial expansion causes environmental degradation, and the opposite is untrue.

The fifth causation appears to be from emissions to energy intensity (CO₂→EI), which is supposed to be the opposite. However, the causality analysis is a transient phenomenon, and a stable long-run direction of causation will likely be from energy intensity to emissions. Moreover, EI can cause CO₂ emissions through the FD channel, as reported in the ninth one-way causality.

The sixth occurs between the fourth and first quartiles of ET (ET_75↔ ET_25), exhibiting a feedback relationship. It follows that the fourth quartile (ET_75) does not directly cause CO₂ emissions, but it invariably does so through the first quartile (ET_25).

Finally, causation exists from the second–third quartile to EI (ET_25$~$75→EI), signifying that the second–third quartile causes energy intensity. Thus, the second–third quartile, though it does not cause emissions directly, causes the latter through the energy intensity channel. Hence, the “T-Y causality assessment” authenticates the bi-threshold NARDL findings.

5. Conclusions and Implications

This study has examined how environmental technologies affect CO₂ emissions, controlling ICT, energy use, intensity, and financial advancement. We have used data spanning 31 years and applied the bi-threshold NARDL approach. We have also operated ARDL and single-threshold NARDL estimates to have comparable outcomes of the different estimates and check their robustness. The T-Y causality assessment is implemented to perform a further robustness check.

The bi-threshold NARDL bounds test supports the long-run cointegrating affiliation among variables. The outcomes of the bi-threshold NARDL indicate that the effects of ET on CO₂ emissions across the thresholds are positive. Two other assessment methods (ARDL and NARDL) also yield similar findings. Long-run asymmetry exists among decomposed components of ET based on both bi-threshold and single-threshold NARDL approaches. Usually, ET is likely to influence CO₂ emissions negatively, but all three estimation methods provide a different and stunning outcome that ET affects the latter positively. The limited use of ET compared to the country’s total technologies may be a reason for this unusual finding. This rare finding in the literature may alert Saudi Arabian policymakers to take immediate policy actions to address the low level of ET compared to overall technology in the Kingdom.

In all three estimation methods, ICT significantly reduces carbon emissions in the long run. Thus, ICT diffusion remains a good solution for environmental management in the country. Energy consumption and intensity are responsible for air pollution, as the Kingdom extensively utilizes fossil fuels to meet its growing energy demands. The impact of financial progress on pollution emissions remains positive and significant. The bi-threshold NARDL method remains much more robust than the other two estimation techniques.

The T-Y causality assessment further assesses the outcome of the main estimation procedure and endorses the robustness of the threshold NARDL findings. The implemented causality assessment produces several bidirectional and one-way causations, which support the strength of the bi-threshold NARDL results.

The findings have significant ramifications for the Kingdom’s government and policymakers to shape workable policies to reduce environmental pollution in this energy-rich country. Immediate action is needed to raise the scale of environmental technology compared to the overall mix of technologies in the country. It may enhance the scale of ICT diffusion for better environmental management. Additionally, since the nation has year-round access to abundant solar and wind energy supplies, steps must be taken to choose a more significant share of renewable energy sources for consumption. This will ease environmental pollution and offer new prospects for further economic development.

Moreover, policymakers must look for alternative, efficient energy sources to reduce their energy intensity and mitigate environmental pollution. Financial expansion can facilitate easy credit for green innovation, ICT, and renewable energy generation, likely decreasing environmental pollution.

Limitations and Further Study

The study is country-specific and confined to Saudi Arabia. It looks at the impact of ET on CO₂ emissions, controlling ICT, energy use, its intensity, and financial development. Future research may consider different control variables, including trade openness and foreign direct investment. Similar studies may be implemented for other economies and a panel of nations. They may investigate the influence of ET on environmental quality by incorporating the same or different set of controlled variables. Moreover, future research may consider different proxies for environmental technology, such as research and development expenditure and number of patents.