Investigation of the role of data conversion on prediction results: Evidence from USA's energy-related emissions and source-based energy use data

Abstract

Considering data conversion practices in empirical research, this research investigates the role of data conversion on prediction results in the United States (USA), where yearly and monthly data on energy-related carbon dioxide (CO₂) emissions and source-based energy consumption is available, which makes the USA an appropriate case for empirical analysis. In this context, this study considers CO₂ emissions as the dependent variable, uses source-based energy use indicators as the explanatory drivers, and performs cointegration regression (CR) approaches on monthly datasets between 1989/1 and 2023/12, which consist of monthly original series (MOS), monthly converted series by quadratic-average-approach (MCSQA), and monthly converted series by quadratic-average-sum (MCSQS). The empirical results reveal that (i) data conversion increases R² values and improves the goodness of fit criteria of the prediction models, where training and testing results are above 96%; (ii) data conversion causes a change in the coefficients of the explanatory variables. While the direction of the variables changes from MOS to MCSQA and MCSQS, it is the same across between MCSQA and MCSQS, but coefficients and p-values slightly differentiate; (iii) dynamic OLS approach has the highest prediction performance among approaches applied; (iv) the importance of source-based energy use indicators on CO₂ emissions differentiate. Overall, the study empirically demonstrates the increasing but varying impact of data conversion on prediction results. Accordingly, the study discusses to benefit of the use of converted data series in empirical predictions, where policymakers can benefit from increasing the impact of data conversion on prediction capacity and prevent incorrect prediction results.

Keywords

source-based energy use data conversion time series approaches the USA

Introduction

Climate change, which is a comprehensive term to represent various environmental problems including air pollution and temperature increases, has become one of the most important issues that the world has been facing in recent years.^1,2 In line with this negative progressing trend, countries and societies have become much more concerned about environmental degradation in recent years. Accordingly, solo efforts of the countries as well as collaborative efforts among countries and international initiatives have been developing day by day. As a result of these developing efforts, research on environmental degradation from different perspectives has been increasing as well.

Following the leading study of Grossman and Krueger,³ previous studies focused on mainly economy-related factors in examining environmental progress (e.g. Pao et al.,⁴ Magazzino et al.,⁵ and Rahman et al.⁶). On the other hand, compatible with the study of Kraft and Kraft,⁷ later studies have dealt much more with energy use for the impact on the environment (e.g. Sharif et al.⁸ and Kartal⁹). Thus, today's research has been mainly considering the role of energy use on environmental degradation by considering combinations of energy sub-types in the research.

The current literature is highly rich in terms of the investigation of the impact of energy use on the environment. There are countless research on this aspect. While some of these have used original datasets at different frequencies (e.g. monthly dataset by Sharif et al.⁸ and Ulussever et al.¹⁰; yearly dataset by Kartal et al.¹¹), some others have preferred to work with converted data series originated from the original (e.g. yearly) data into a selected one (monthly converted dataset by Sharif et al.¹²; quarterly converted dataset by Adebayo and Kirikkaleli,¹³ Meng et al.,¹⁴ and Xu et al.¹⁵). While either original or converted datasets have been used in these studies, based on the best knowledge, there has not been any study that investigates the role of data conversion on the prediction results. This point issues a literature gap and this gap can be researched empirically to determine the impact of data conversion on the prediction results.

Considering the literature gap defined, this study aims to search for the answers to the research question of “determining whether data conversion has an impact on the prediction results or not.” For this purpose, a country, that must have both low-frequency (i.e. original) and high-frequency (i.e. converted) datasets at different time intervals, can be an appropriate sample. From this point of view, the USA has come to the fore because it has both yearly and monthly datasets.¹⁶ For this reason, the USA is evaluated as one of the most appropriate cases to investigate answers to the research question. Thus, it is possible to uncover the impact of data conversion on the prediction results or not by focusing on the USA case.

The USA is the leading economy in the world in terms of its economic size.¹⁷ While the USA has sustained its pioneering position among the countries, it has been causing a high amount of CO₂ emissions, which takes place among the top five CO₂-emitting countries and consumes a high amount of energy.¹⁸ Accordingly, the USA has been facing environmental problems resulting from high energy use. Figure 1 demonstrates the progress of both CO₂ emissions and energy use at disaggregated levels across the years.

Figure 1.

The progress of energy-related CO₂ emissions and energy use.

As demonstrated, from the point of CO₂ emissions, the USA has a relatively horizontal trend between 1989 and 2023. Although there was an increasing trend until 2005, there has been a slightly decreasing trend since then. Also, from the point of their share in the total energy mix, there has been a constant increase in BEC, GEC, NEC, NGC, SEC, and WEC across the recent years, whereas there has been a decreasing trend in CEC and OEC as well as a horizontal trend in HEC. In summary, there has been a huge amount of CO₂ emissions and energy use in the USA case.

In line with the research gap as well as the appropriate position of the USA case for empirical analysis, the study investigates the role of data conversion on the prediction results. In doing so, the study benefits from the USA case by considering the USA's energy-related CO₂ emissions and source-based energy use data, using both original and converted datasets at the same time, and applying CR approaches as performing much better. In this context, the study uses totally three different datasets (i.e. MOS, MCSQA, and MCSQS) for the same environmental and energy indicators and applies a total of three CR approaches (i.e. CCR, DOLS, and FMOLS). As a result, the study defines that data conversion increases R² and goodness of fit criteria of the prediction models. Also, data conversion causes a change in the coefficients of the explanatory variables. Besides, the DOLS approach has the highest prediction performance among CR approaches. Moreover, the importance of source-based energy use on CO₂ emissions varies. Thus, the study empirically reveals the increasing impact on the prediction capacity of the models, whereas the impact differentiates. In this way, the study obtains clear answers to the research question.

Following up a comprehensive approach, this research provides various novelties to the literature. First, according to the best knowledge, this research is the first kind of its type, which investigates the impact of data conversion in terms of its impacts on prediction results. Second, the study uses a total of three different unique datasets to uncover answers to the research question by focusing on the USA case as an appropriate case. Third, the study applies three CR approaches as types of econometric approaches because such approaches perform better than others due to being derived from economic theories. With these points, the study comprehensively and empirically searches for the answers to the research question.

The study goes on the theoretical framework in the “Theoretical framework and literature review” section; methods in the “Methods” section; empirical results in the “Empirical results” section; conclusion, discussion, policies, and future research in the “Conclusion, discussion, policy endeavors, and future research” section.

Theoretical framework and literature review

Theoretical underpinning and mechanism

The relationship between energy use and environmental degradation is mainly based on both energy-led growth and energy growth hypotheses, proposed by Kraft and Kraft⁷ and Nachane et al.,¹⁹ in order. The relationship is also related to the fundamental environmental economic hypothesis as the environmental Kuznets curve (EKC) proposed by Grossman and Krueger.³ These hypotheses mainly imply some critical points about the relationship between energy use and the environment as (i) energy use is highly required for ensuring sustainable economic growth; (ii) the amount of energy used increases as economies grow; and (iii) the environmental problems increase as economies grow and further energy is used unless they are handled in an eco-friendly manner. Hence, by considering these hypotheses (i.e. energy-led growth hypotheses; energy growth hypotheses; EKC hypotheses), it can be stated that energy use, economic growth structure, and environmental problems are highly interrelated with each other. For this reason, dealing with energy use is a critical issue in terms of both economic growth structure and environmental problems as well.

Since energy use is a critical factor for economic growth, there is a significant impact of energy use on environmental degradation. As economic growth is achieved and further energy is used for this purpose, there will be an increase in environmental degradation, which results from much more fossil fuel-based energy use. Hence, a direct impact mechanism between energy use and environmental degradation exists.

Taking theoretical underpinning structured by Kraft and Kraft,⁷ Nachane et al.,¹⁹ and Grossman and Krueger³ into account as well as the direct impact mechanism of energy use on environmental degradation, a large body of research (e.g. Bilgili et al.,²⁰ Li et al.,²¹ and Wang et al.^22–24) has been prepared to uncover the relationship between energy use and environmental degradation and different results have been obtained in these studies as well. The literature is quite from this point of view.

Review of the empirical literature

In the contemporary literature, a large set of research has focused on the relationship between energy use and the environment from different perspectives. On the one hand, the literature can be classified based on energy types. In this context, a part of them have focused on fossil fuel energy use (e.g. Ulussever et al.²), whereas much newer ones have dealt with clean energy use (e.g. Sharif et al.⁸). On the other hand, the literature can be classified also according to the data type used. Some studies have used total energy use (i.e. aggregated level) data (e.g. Kartal²⁵), whereas others have preferred much more detailed (i.e. disaggregated level) data (e.g. Bilgili,²⁶ Kuşkaya and Bilgili,²⁷ Magazzino et al.,²⁸ Zhen et al.,²⁹ and Kuşkaya et al.³⁰). Moreover, recent studies have begun to apply different data conversion approaches to original datasets by relying on various purposes, such as increasing the number of observations, making data appropriate for certain econometric approaches, and ensuring consistency between various data intervals into a single one. Thus, some studies have worked with original datasets at different frequencies, whereas others have transformed data into different frequencies. Hence, there has been a growing literature on the converted data use and Table 1 presents a summary of the recent research.

Table 1.

A summary of data conversion approaches applied, data types used, and econometric approaches performed.

Conversion Approaches	Authors	Scope	Frequency	Econometric approaches
Panel A: QA approach	Adebayo and Kirikkaleli¹³	Japan	Quarterly	WC
Panel A: QA approach	Adebayo and Ullah⁴²	Sweden	Quarterly	WC
Panel B: QS approach	Sharif et al.¹²	Top-10 polluted countries	Monthly	QQR, GCQ, QR
	Hu et al.⁴³	Canada	Quarterly	WLMC
	Liu et al.⁴⁴	Japan	Quarterly	QQR, NCQ
	Pata et al.⁴⁵	China	Quarterly	NCQ, GCQ
	Ulussever et al.²	GCC countries	Quarterly	QQR, GCQ, QR
	Wang et al.⁴⁶	Finland	Quarterly	QQR, WQR
	Adebayo et al.⁴⁷	Thailand	Quarterly	Fourier GC
	Adebayo et al.⁴⁸	Germany	Quarterly	WLMC
	Adebayo et al.⁴⁰	USA	Quarterly	QQR, WQR
	Adebayo et al.⁴¹	USA	Quarterly	QQR-based KRLS
	Kartal et al.⁴⁹	Selected eight countries	Quarterly	QQR, GCQ, QR
	Meng et al.¹⁴	USA	Quarterly	QQR-based KRLS
	Xu et al.¹⁵	USA	Quarterly	QQR, GCQ, WQR

Note: ARDL: autoregressive distributed lag; GC: Granger causality; GCC: Gulf cooperation council; GCQ: Granger causality in quantiles; KRLS: kernel-based regularized least squares; NCQ; nonparametric causality in quantiles; QQR: quantile-on-quantile regression; QR: quantile regression; WC: wavelet coherence; WLMC: wavelet local multiple correlation; WQR: wavelet quantile regression; QA: quadratic average; QS: quadratic sum.

According to Table 1, the literature includes various studies that have applied data conversion approaches to convert data into different frequencies in examining the relationship between energy use and environmental degradation. Mainly both QS and QA approaches have been used by scholars for data conversion. On the other hand, a variety of econometric approaches have been applied to converted datasets to predict the relationship between energy use and environmental degradation.

Evaluation of the literature

In the literature, there are various studies that have dealt with the relationship between energy use and environmental degradation. The studies have mainly shown that energy use has an important impact on environmental degradation and the impact of energy use on the environment has differentiated based on energy types considered, scope of the studies, time intervals used, and econometric approaches applied for empirical investigation.

On the other hand, among all, some of these studies have used converted data, which is summarized in Table 1, for empirical analyses. When they are considered all together, some common points can be seen. First, the studies have generally used either the QA approach or the QS approach for data conversion. Such a data conversion has various purposes, such as increasing the number of observations, making data appropriate for certain econometric approaches, and ensuring consistency between various data intervals into a single one. Second, the studies have mainly used a data conversion approach to convert yearly data into quarterly data (there is an example of monthly conversion as well). Third, the studies have mostly applied time series econometric models on the converted datasets for empirical analysis. Hence, they have stated various conclusions depending on the scope.

The present studies have applied data conversion approaches and performed various econometric approaches (mainly time series ones) on converted data for empirical analysis. The use of converted data enables researchers to investigate different sides of energy economics and environmental sustainability by increasing the number of observations (required by various econometric models), making data appropriate for certain econometric approaches (ensuring normality, nonlinearity, and stationarity), and ensuring consistency between various data intervals into a single one to provide use of various indicators at the same data frequency. However, based on the best knowledge, no study has been analyzed to determine the impact of data conversion on prediction results, which is also important in terms of energy economics and environmental degradation. So, this point presents a literature gap.

In line with the literature gap defined, the study focuses on the USA case, which has both yearly and monthly datasets. Thus, it is possible to make an empirical examination of the impact of data conversion on the prediction results by first converting the yearly dataset into a monthly dataset and then using both original monthly and converted monthly datasets in the empirical analyses. Within the context of empirical analysis, the study uses three datasets (i.e. MOS, MCSQA, and MCSQS), which include both original monthly and converted monthly datasets, applies CR approaches as consistent with the current literature because CR approaches are types of econometric approaches and such approaches perform better than others due to being derived from economic theories. Hence, the study fills in this gap by presenting empirical answers to the research question (i.e. determining whether data conversion has an impact on the prediction results or not).

Methods

Data

To empirically investigate the role of data conversion on prediction results, the study focuses on the USA case. That is why the USA publishes the data on energy-related CO₂ emissions and source-based energy use for both annual and monthly frequencies.¹⁶ This data availability enables researchers to uncover the impact of data conversion by using both original data series and transformed data series comparatively.

Although data for some energy use types goes further back, the study aims to include all energy types in the analysis because the use of source-based energy consumption data is critical to make a robust prediction due to related CO₂ emissions with the energy use as approximated in Annex 1. According to basic approximation, by considering source-based energy consumption indicators, the study predicts 99.69% variations in CO₂ emissions.

Wind energy use data starts from 1982, solar energy use data starts from 1984, and some data on solar and wind energy use are lacking across various times between 1985 and 1988. Based on these causes, the study uses a dataset between 1989/1 and 2023/12, which is obtained from EIA¹⁶ as the most recent available data. Hence, the overall dataset includes 420 monthly observations, which constitutes the original data series (i.e. MOS).

Also, to uncover the impact of data conversion, the study considers yearly data between 1989 and 2023 and transforms this yearly data series into a monthly data series. In this context, the study applies both QA and QS approaches, which is consistent with the current literature as shown in Table 1, where most of the studies have applied these approaches for data conversion. Also, the use of quadratic approaches in data conversion provides the best-converted series than other data conversion approaches.^31–33 These are the causes why the study considers QA and QS approaches. Thus, the study obtains two different monthly converted data series (i.e. MCSQA and MCSQS) by converting yearly datasets into monthly datasets through the performance of QA and QS approaches in EViews software.

Following up the above-explained process, the study obtains and uses three data series (i.e. MOS, MCSQA, and MCSQS) in the analysis, which is demonstrated in Annexes 2 to 4.

Furthermore, the data series are divided into both training and testing samples. In this context, 80% of the data series (i.e. the first 336 observations) is used as a training dataset, which is between 1989/1 and 2016/12. Also, the remaining 20% of the data series (i.e. the last 84 observations) is used as a testing dataset, which is between 2017/1 and 2023/12. Thus, through using not only original and converted monthly datasets, but also using both training and testing samples, this study obtains reliable results in the investigation of the impact of data conversion on the prediction results by making a robust analysis.

Variables

By relying on the explained theoretical framework and examined empirical literature, this study comprehensively examines the data conversion impact on the relationship between energy-related CO₂ emissions and source-based energy use. In this context, the study considers all available energy use types in the prediction process because they are complete reasons for energy-related CO₂ emissions.

Table 2 shows the details of the variables.

Table 2.

Details of the variables.

Symbol	Variable description	Measure unit	Data source
CO₂	Carbon dioxide^a	Million Tons	EIA¹⁶
BEC	Biomass EC	Quadrillion British Thermal Unit	EIA¹⁶
CEC	Coal EC
GEC	Geothermal EC
NEC	Nuclear EC
NGC	Natural gas EC
OEC	Oil EC
SEC	Solar EC
WEC	Wind EC

Note: BEC, CEG, GEC, NEC, NGC, OEC, SEC, and WEC denote energy use from biomass, coal, geothermal, nuclear electric, natural gas, oil (petroleum), solar, and wind sources.

^a Shows the dependent variable.

Methodological process

In line with the aim of the study, a comprehensive methodological process is used in the prediction, which Figure 2 shows in detail.

Figure 2.

Methodological process.

In the first step, original data series are collected from the source of EIA.¹⁶ In the second step, monthly converted series are obtained by performing QA and QS approaches. In the third step, all collected data series (i.e. MOS, MCSQA, and MCSQS) are split into both training and testing samples.

Following the obtaining of datasets, in Steps 4–6, preliminary statistics of the variables including descriptive statistics, correlation matrix, and stationarity test are analyzed.

Across Steps 7–10, to predict the energy-related CO₂ emissions by using source-based energy use indicators under the investigation of data conversion impact, the CR approaches are applied (i.e. CCR, DOLS, and FMOLS), goodness of fit statistics are examined, the CR approaches are compared, and the results of the best approach are interpreted in detail. Further details about the methods applied can be obtained from the original studies (for CCR³⁴; for DOLS³⁵; and for FMOLS³⁶).

For empirical prediction, CR approaches, which are types of econometric approaches, are used because such approaches perform better than others due to being derived from economic theories. It is clear that these methods have some limitations, such as working stationary variables at I(1). Hence, these methods are selected based on data characteristics. Accordingly, this point has been put in the limitations subsection of the study, where future research can consider this point in designing new studies.

Prediction model

FMOLS technique, which is a residual-based test, provides efficient results for the variables.³⁶ Moreover, FMOLS is considered a reliable estimate when the sample size is small and eliminates the problems of endogeneity and serial correlation among the variables. DOLS provides better results than FMOLS and eliminates correlation among regressors.³⁵ By using source-based energy use data as well as performing CR approaches as the fundamental prediction approaches (i.e. CCR by Park³⁴; DOLS by Stock and Watson³⁵; FMOLS by Hansen and Phillips³⁶), this investigation considers equation (1) to predict CO₂ emissions:

\begin{aligned} C O_{2} = & β_{1} (B E C_{t}) + β_{2} (C E C_{t}) + β_{3} (G E C_{t}) \\ + β_{4} (N E C_{t}) + β_{5} (N G C_{t}) + β_{6} (O E C_{t}) \\ + β_{7} (S E C_{t}) + β_{8} (W E C_{t}) + ϵ^{θ} \end{aligned}

(1)

where

β_{1, \dots, 8}

denotes the prediction coefficients of the explanatory variables and

ϵ

denotes the error term.

Goodness of fit criteria

In examination of the data conversion impact on the prediction results, the study considers a set of goodness of fit criteria that includes R², root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The higher (lower) values of R² (RMSE, MAE, and MAPE) imply that the prediction approaches have a higher success in predicting dependent variables by using independent variables as well as the difference between actual and predicted values of the dependent variables is quite low. Accordingly, it can be stated that an approach, that has higher (lower) values of R² (RMSE, MAE, and MAPE), is better than others.

R² is used as an indicator to measure the changes in the dependent variable that are explained by independent variables in the approach. Also, RMSE, MAE, and MAPE are the diagnostic indicators that reflect the accuracy of the approach. All these goodness of fit criteria are computed by using the following equations^37,38:

RMSE = \sqrt{\frac{1}{n} \sum_{i = 1}^{n} {(y_{i} - {\hat{y}}_{i})}^{2}}

(2)

MAE = \frac{1}{n} \sum_{i = 1}^{n} | y_{i} - {\hat{y}}_{i} |

(3)

MAPE = \frac{100}{n} \sum_{i = 1}^{n} | \frac{(y_{i} - {\hat{y}}_{i})}{y_{i}} |

(4)

R^{2} = 1 - \frac{\sum {(y_{i} - {\hat{y}}_{i})}^{2}}{\sum {(y_{i} - \bar{y})}^{2}}

(5)

where y_i is the actual value,

{\hat{y}}_{i}

is the predicted value from the approach,

\bar{y}

is the mean of actual values, and n is the number of observations.

Empirical results

Descriptive statistics

After the collection and preparation of the original monthly dataset and converted monthly data sets, the empirical process goes on with the examination of fundamental statistics. So, the study first examines the descriptive statistics of the variables in the datasets to understand their basic characteristics. In this context, Table 3 presents the descriptive statistics.

Table 3.

Descriptive statistics.

Data type	Variable	Mean	Maximum	Minimum	SD	Skewness	Kurtosis	JB	JB Probability
MOS	CO₂	450.58	560.78	305.60	42.70	0.14	2.91	1.57	0.4564
	BEC	0.32	0.46	0.18	0.08	0.22	1.48	43.55	0.0000
	CEC	1.53	2.13	0.50	0.39	−0.85	2.83	50.54	0.0000
	GEC	0.01	0.01	0.00	0.00	−0.02	1.44	42.51	0.0000
	NEC	0.08	0.12	0.05	0.01	0.37	2.54	13.35	0.0013
	NGC	2.09	3.70	1.24	0.51	0.67	2.91	31.19	0.0000
	OEC	3.00	3.56	2.12	0.21	0.03	3.28	1.38	0.5016
	SEC	0.01	0.10	0.00	0.02	2.24	7.60	723.31	0.0000
	WEC	0.03	0.16	0.00	0.04	1.22	3.44	107.19	0.0000
MCSQA	CO₂	5406.99	6032.24	4538.86	375.64	−0.01	2.14	12.87	0.0016
	BEC	3.79	5.15	2.58	0.92	0.27	1.33	53.96	0.0000
	CEC	18.37	22.94	7.09	4.31	−1.03	2.83	74.45	0.0000
	GEC	0.09	0.12	0.06	0.02	0.05	1.25	54.02	0.0000
	NEC	7.74	8.49	5.32	0.81	−1.12	2.94	87.62	0.0000
	NGC	25.07	33.73	19.53	3.97	0.79	2.52	47.57	0.0000
	OEC	35.96	40.40	31.98	2.10	0.45	2.34	22.15	0.0000
	SEC	0.17	0.92	0.05	0.22	1.96	5.85	412.58	0.0000
	WEC	0.37	1.51	0.01	0.46	1.09	2.90	83.71	0.0000
MCSQS	CO₂	450.58	502.69	378.24	31.30	−0.01	2.14	12.87	0.0016
	BEC	0.32	0.43	0.21	0.08	0.27	1.33	53.96	0.0000
	CEC	1.53	1.91	0.59	0.36	−1.03	2.83	74.45	0.0000
	GEC	0.01	0.01	0.00	0.00	0.05	1.25	54.02	0.0000
	NEC	0.65	0.71	0.44	0.07	−1.12	2.94	87.62	0.0000
	NGC	2.09	2.81	1.63	0.33	0.79	2.52	47.57	0.0000
	OEC	3.00	3.37	2.66	0.18	0.45	2.34	22.15	0.0000
	SEC	0.01	0.08	0.00	0.02	1.96	5.85	412.58	0.0000
	WEC	0.03	0.13	0.00	0.04	1.09	2.90	83.71	0.0000

Note: JB and SD denote the Jarque-Bera and standard deviation, in order. BEC, CEG, GEC, NEC, NGC, OEC, SEC, and WEC denote energy use from biomass, coal, geothermal, nuclear electric, natural gas, oil (petroleum), solar, and wind sources. MOS: monthly original series; MCSQA: monthly converted series by quadratic average approach; MCSQS: monthly converted series by quadratic sum approach.

In MOS, all variables except for CO₂ and OEC have a nonnormal distribution. Similarly, all variables in both MOSQA and MOSQS are not normally distributed. In all three datasets, CO₂ has the highest value followed by OEC, NGC, and CEC. Also, CO₂ and NGC have the highest variations among all variables, in order. While the minimum and maximum values of variables are near to each other in MOS and MOSQS datasets, the values in MOSQA differ highly from those. Hence, it can be stated that there are significant variations in the variables and most of the variables are not normally distributed.

Correlation matrix

Table 4 shows the correlation coefficients between the variables.

Table 4.

Correlation matrix.

Data type	Variable	CO₂	BEC	CEC	GEC	NEC	NGC	OEC	SEC	WEC
MOS	CO₂	1.00
	BEC	−0.28	1.00
	CEC	0.73	−0.65	1.00
	GEC	−0.21	0.93	−0.60	1.00
	NEC	0.04	−0.14	0.09	−0.17	1.00
	NGC	0.19	0.55	−0.49	0.55	−0.05	1.00
	OEC	0.65	−0.12	0.43	−0.05	−0.03	−0.08	1.00
	SEC	−0.50	0.62	−0.80	0.59	−0.10	0.44	−0.11	1.00
	WEC	−0.46	0.82	−0.89	0.81	−0.13	0.64	−0.20	0.85	1.00
MCSQA	CO₂	1.00
	BEC	−0.47	1.00
	CEC	0.83	−0.75	1.00
	GEC	−0.35	0.95	−0.68	1.00
	NEC	0.25	0.61	−0.24	0.76	1.00
	NGC	−0.47	0.85	−0.87	0.84	0.62	1.00
	OEC	0.83	−0.25	0.43	−0.13	0.43	−0.08	1.00
	SEC	−0.63	0.65	−0.92	0.63	0.32	0.89	−0.17	1.00
	WEC	−0.62	0.86	−0.94	0.84	0.50	0.96	−0.23	0.94	1.00
MCSQS	CO₂	1.00
	BEC	−0.47	1.00
	CEC	0.83	−0.75	1.00
	GEC	−0.35	0.95	−0.68	1.00
	NEC	0.25	0.61	−0.24	0.76	1.00
	NGC	−0.47	0.85	−0.87	0.84	0.62	1.00
	OEC	0.83	−0.25	0.43	−0.13	0.43	−0.08	1.00
	SEC	−0.63	0.65	−0.92	0.63	0.32	0.89	−0.17	1.00
	WEC	−0.62	0.86	−0.94	0.84	0.50	0.96	−0.23	0.94	1.00

Note: Values denote correlation coefficients. BEC, CEG, GEC, NEC, NGC, OEC, SEC, and WEC denote energy use from biomass, coal, geothermal, nuclear electric, natural gas, oil (petroleum), solar, and wind sources. MOS: monthly original series; MCSQA: monthly converted series by quadratic average approach; MCSQS: monthly converted series by quadratic sum approach.

In MOS, CO₂ is negatively correlated with BEC, GEC, SEC, and WEC, whereas other variables are positively correlated. Among all, some variables have a higher correlation (e.g. CEC: 0.73; OEC: 0.65) and some others have a lower correlation (NEC: 0.04; NGC: 0.19). In MCSQA and MCSQS, CO₂ is negatively correlated with BEC, GEC, NGC, SEC, and WEC, whereas other variables are positively correlated. Among all, some variables have a higher correlation (e.g. CEC: 0.83; OEC: 0.83) and some others have a lower correlation (e.g. NEC: 0.25; GEC: 0.25). Hence, it can be stated that the variables have a mixed condition, where they have both higher and lower correlations with CO₂.

Stationarity test

Table 5 demonstrates the stationarity test results.

Table 5.

Stationarity test results.

Data type	Variable	ADF		PP		Decision
Data type	Variable	I(0)	I(1)	I(0)	I(1)	Decision
MOS	CO₂	0.5859	0.0000	0.4859	0.0001	S at I(1)
	BEC	0.9714	0.0000	0.8616	0.0001	S at I(1)
	CEC	0.2485	0.0000	0.2932	0.0000	S at I(1)
	GEC	0.9968	0.0000	0.8234	0.0001	S at I(1)
	NEC	0.4127	0.0000	0.3871	0.0000	S at I(1)
	NGC	0.9978	0.0000	0.5197	0.0000	S at I(1)
	OEC	0.7083	0.0000	0.6501	0.0001	S at I(1)
	SEC	0.9999	0.0000	0.7611	0.0000	S at I(1)
	WEC	0.9991	0.0004	0.9559	0.0000	S at I(1)
MCSQA	CO₂	0.9049	0.0000	0.7985	0.0000	S at I(1)
	BEC	0.9038	0.0000	0.9451	0.0000	S at I(1)
	CEC	0.9999	0.0000	0.9997	0.0000	S at I(1)
	GEC	0.8801	0.0014	0.8318	0.0000	S at I(1)
	NEC	0.9590	0.0000	0.9747	0.0000	S at I(1)
	NGC	0.9957	0.0000	0.9862	0.0000	S at I(1)
	OEC	0.3692	0.0000	0.3692	0.0000	S at I(1)
	SEC	0.9644	0.0001	0.9999	0.0001	S at I(1)
	WEC	0.3087	0.0435	0.9999	0.0000	S at I(1)
MCSQS	CO₂	0.4921	0.0000	0.5195	0.0000	S at I(1)
	BEC	0.9933	0.0000	0.9295	0.0000	S at I(1)
	CEC	0.1167	0.0000	0.1531	0.0000	S at I(1)
	GEC	0.9964	0.0007	0.9996	0.0000	S at I(1)
	NEC	0.9781	0.0000	0.9747	0.0000	S at I(1)
	NGC	0.9999	0.0000	0.9993	0.0000	S at I(1)
	OEC	0.7778	0.0000	0.6999	0.0000	S at I(1)
	SEC	0.9644	0.0000	0.9999	0.0000	S at I(1)
	WEC	0.3087	0.0435	0.9999	0.0000	S at I(1)

Note: Optimal lag is determined through automatic selection by using the Schwarz Info Criterion in the ADF test and Newey-West automatic using Bartlett kernel in the PP test. S, I(0), and I(1) denote the stationarity, stationarity in level, and stationarity in first difference. MOS: monthly original series; MCSQA: monthly converted series by quadratic average approach; MCSQS: monthly converted series by quadratic sum approach.

Based on both ADF and PP test results, all variables have a unit root that they are stationary at the I(1) level.

In the case of the overall evaluation of the fundamental statistics (i.e. descriptive statistics, correlations, and stationarity condition of the variables), it is much more appropriate to apply econometric approaches that consider nonstationary at I(0). Hence, considering data characteristics, the study performs CR approaches (i.e. CCR, DOLS, and FMOLS) on the datasets for applying a comprehensive empirical analysis.

Comparison of the prediction results by CR approaches

Followingly, the study performs the CR approaches. Table 6 reports RMSE, MAE, MAPE, and R² statistics of CR results for the datasets in a comparative way.

Table 6.

Comparison of the prediction approaches.

Data type	Prediction approach	Training dataset					Testing dataset
Data type	Prediction approach	RMSE	MAE	MAPE	R ²	Adj. R²	RMSE	MAE	MAPE	R ²	Adj. R²
MOS	CCR	12.86	11.33	2.74	89.279%	89.050%	17.83	14.55	3.50	75.217%	72.935%
	DOLS	2.25	1.78	0.39	99.714%	99.708%	1.63	1.30	0.32	99.793%	99.774%
	FMOLS	2.16	1.73	0.38	99.700%	99.694%	1.69	1.36	0.33	99.776%	99.756%
MCSQA	CCR	21.44	16.28	0.29	99.556%	99.546%	41.59	32.05	0.64	96.571%	96.255%
	DOLS	9.77	7.49	0.13	99.908%	99.906%	0.00	0.00	0.00	99.990%	99.990%
	FMOLS	10.07	7.30	0.13	99.921%	99.919%	0.00	0.00	0.00	99.990%	99.990%
MCSQS	CCR	1.79	1.36	0.29	99.556%	99.546%	3.47	2.68	0.64	96.557%	96.239%
	DOLS	0.81	0.62	0.13	99.908%	99.906%	0.00	0.00	0.00	99.990%	99.990%
	FMOLS	0.82	0.60	0.13	99.907%	99.905%	0.00	0.00	0.00	99.990%	99.990%

Note: MOS: monthly original series; MCSQA: monthly converted series by quadratic average approach; MCSQS: monthly converted series by quadratic sum approach; RMSE: root mean squared error; MAE: mean absolute error; MAPE: mean absolute percentage error; CCR: canonical cointegration regression; DOLS: dynamic ordinary least squares; FMOLS: fully modified ordinary least squares. Bold lines denote the best prediction approach among the alternatives. The value of “0.00” denotes the rounded values for too-small values.

According to Table 6, the R² values of the prediction models across the training dataset and testing dataset are quite similar and near each other. Also, there is a high level of consistency between training dataset results and testing dataset results. Besides, the results between the training dataset and testing dataset are above 70%, which is higher than the acceptable limit. Hence, it can be stated that CR approaches are successful in predicting CO₂ emissions by using source-based energy use indicators.

From the point of view of data conversion impact, data conversion improves goodness of fit criteria (i.e. RMSE, MAE, and MAPE) because the values on these parameters are generally smaller in MCSQA and MCSQS with regard to MOS. Similarly, data conversion increases the R² values of the prediction models. Besides, with regard to the QA approach, the QS approach has a much more powerful and beneficial impact in terms of gathering better prediction results.

Results of the best prediction approach

Among all CR approaches applied, the DOLS approach has the highest prediction performance across MOS, MCSQA, and MCSQS. Through defining that the DOLS approach is the best approach among the alternatives applied, Table 7 presents the prediction results for three datasets.

Table 7.

Prediction results of the DOLS approach.

Data type	Variable	Training set		Testing set
Data type	Variable	Coefficient	p-value	Coefficient	p-value
MOS	BEC	−26.74	0.0025	51.73	0.1119
	CEC	87.05	0.0000	94.01	0.0000
	GEC	1368.06	0.0005	−1785.58	0.0034
	NEC	−13.65	0.3192	30.17	0.1627
	NGC	57.35	0.0000	53.10	0.0000
	OEC	67.62	0.0000	62.57	0.0000
	SEC	−348.84	0.0004	−86.06	0.0000
	WEC	−59.26	0.1914	19.53	0.3329
MCSQA	BEC	4.73	0.5706	153.44	0.0000
	CEC	95.24	0.0000	96.02	0.0000
	GEC	171.98	0.6724	−4815.12	0.0000
	NEC	17.63	0.0006	35.75	0.0000
	NGC	38.73	0.0000	46.99	0.0000
	OEC	69.55	0.0000	53.58	0.0000
	SEC	−102.18	0.4204	−44.87	0.0000
	WEC	81.93	0.1077	69.89	0.0000
MCSQS	BEC	4.73	0.5706	153.44	0.0000
	CEC	95.24	0.0000	96.02	0.0000
	GEC	171.98	0.6724	−4815.12	0.0000
	NEC	17.63	0.0006	35.75	0.0000
	NGC	38.73	0.0000	46.99	0.0000
	OEC	69.55	0.0000	53.58	0.0000
	SEC	−102.18	0.4204	−44.87	0.0000
	WEC	81.93	0.1077	69.89	0.0000

DOLS: dynamic ordinary least squares; MOS: monthly original series; MCSQA: monthly converted series by quadratic average approach; MCSQS: monthly converted series by quadratic sum approach. BEC, CEG, GEC, NEC, NGC, OEC, SEC, and WEC denote energy use from biomass, coal, geothermal, nuclear electric, natural gas, oil (petroleum), solar, and wind sources.

In MOS, all variables except for BEC, NEC, and WEC have a statistically significant impact on CO₂ emissions. Also, GEC and SEC have a decreasing impact, whereas the remaining variables have an increasing one. Moreover, the GEC (SEC) has the highest increasing (decreasing) impact on CO₂ emissions. In MOS, the constructed prediction model can predict 99.77% variations in CO₂ emissions through consideration of source-based energy use indicators.

In MCSQA and MCSQS, all variables have a statistically significant impact on CO₂ emissions. Also, GEC and SEC have a decreasing impact, whereas the remaining variables have an increasing one. Moreover, the GEC (SEC) has the highest increasing (decreasing) impact on CO₂ emissions. Using either MCSQA or MCSQS, the constructed prediction model can predict 99.99% variations in CO₂ emissions through consideration of source-based energy use indicators.

Among all datasets, the prediction performance of converted datasets (i.e. MCSQA and MCSQS) is higher than the original dataset (i.e. MOS). While the difference between the original and converted datasets is lower in the testing dataset, it increases a bit in the testing dataset.

Overall, the results reveal that data conversion increases R² values and improves the goodness of fit criteria of the prediction models, while data conversion causes a change in the coefficients of the explanatory variables. Hence, the study empirically reveals the increasing impact on the prediction capacity of the models, but the varying impact of the data conversion on the prediction results.

The DOLS results show that both GEC and SEC have a certain declining impact on CO₂ emissions, whereas other source-based energy use indicators do not help combat emissions in the USA. This is an important point that policymakers should consider in trying to achieve the sustainable development goal (SDG) SDG-13 by benefitting from SDG-7 because the empirical results demonstrate that all types of clean energy do not help ensure carbon neutrality, instead only some of them are beneficial.

Conclusion, discussion, policy endeavors, and future research

Conclusion

In light of SDGs, which aim to make economies sustainable, as well as the increasing importance of climate change at the global level and developing awareness concerning environmental degradation, this study uncovers the role of data conversion, which is a recent and common practice, on prediction results. Accordingly, the study focuses on an important point in the empirical analysis, which has not been the case until now based on the best knowledge, although various conversion practices have been applied.

Compatible with the literature gap defined, this study focuses on the USA case for the empirical analysis because it is an appropriate case for such analysis because the USA publishes both monthly and yearly data for energy-related emissions and source-based energy use data. Hence, it is possible to use the original (i.e. monthly) dataset as well as the converted dataset from the yearly to monthly one by applying conversion approaches. Accordingly, within the empirical examination, the study considers CO₂ emissions as the most commonly used environmental degradation indicator, includes the most possible source-based energy use indicators, uses originally monthly and converted monthly datasets between 1989/1 and 2023/12, and applies CR approaches.

Through a comprehensive empirical process, the study defines that data conversion increases R² values and improves the goodness of fit criteria of the prediction models; data conversion causes a change in the coefficients of the explanatory variables; the DOLS approach has the highest prediction performance among CR approaches; the importance of source-based energy use on CO₂ emissions varies.

Based on empirical results, the study specifically determines that some renewable energy (i.e. solar and geothermal) types have a certainly declining impact on CO₂ emissions in the USA, whereas other types of clean energy are not beneficial in this respect.³⁹ While the results obtained are compatible with the literature (e.g. Kuşkaya,¹ Magazzino et al.,⁵ Adebayo et al.,⁴⁰ and Adebayo et al.⁴¹), the results of this research extend the current knowledge by investigating the impact of data conversion on the prediction results.

Discussion and policy endeavors

By benefitting from the USA case, the study defines the increasing impact of data conversion on prediction results. Accordingly, this research argues various policy endeavors.

Firstly, the study argues that converted data series should be included in empirical predictions for the achievement of SDG 7 and 13 to benefit from increasing impact on prediction capacity so that incorrect prediction can be prevented. Such an approach enables policymakers to benefit from the increasing impact of data conversion on prediction capacity. In this way, policymakers can have the opportunity to benefit from this specialty of data conversion in developing further environmental policies based on energy use indicators.

Secondly, compatible with the results of the study, it is logical to be helpful in data conversion for policymakers across various countries, where there is no high-frequency data in these countries, and data conversion is a necessity to apply certain types of econometric analysis to make predictions. Thus, from this point of view, the results of the study are highly generalizable for other countries in light of the current literature, where data conversion approaches have been frequently applied in various studies.

Thirdly, the empirical results demonstrate that geothermal and solar energy uses are certainly beneficial in declining CO₂ emissions in the USA. This determination implies that the USA should rely further on these types of energy use in the energy mix because other types of clean energy do not contribute to the decarbonization of the energy mix in the USA. The most probable cause of these findings may be that other clean energy sources do not have economies of scale, which prevents the efficiency of these sources.

Fourthly, some other clean energy types including solar and nuclear are not effective in curbing CO₂ emissions. The potential cause of this may be a displacing impact, where the allocation of sources to specific clean energy sources may cause a displacement of other clean energy sources. Hence, the inefficiency of such clean energy sources may be seen. In this context, policymakers should enhance the generation and use of effective clean energy sources (i.e. geothermal and solar energy) in the total energy mix, while they should make further analysis about the source allocation among clean energy sources to prevent possible displacement. Accordingly, policymakers should re-evaluate these types of clean energy use by making them either efficient or re-optimized so that they can help decline CO₂ emissions.

By considering the policy endeavors discussed, the USA can benefit from further optimization of the energy mix under the impact of data conversion on the prediction results. Hence, the USA can make further policy development based on better prediction results. Overall, the USA can benefit from these points in enhancing various SDGs, especially SDG-7 (clean energy) and SDG-13 (climate action).

The results of the study are not only important for the USA, but also they can be beneficial for other countries under the leading position of the USA in the world because many countries have followed the path of the USA. Accordingly, because all countries have been aiming to ensure SDG-13, which is related to slowing down climate change by limiting global warming, the data conversion practices can help develop countries, especially those, who are low-income countries, in obtaining high-frequency data. In this way, while they do not have to bear to cost of producing high-frequency data, they can have also the opportunity to benefit from using high-frequency data in empirical modeling. Hence, they can make further analysis and have much better options to put their environmental policies based on energy use indicators. On the other hand, such data conversion approaches and their benefits can be used in other environmental and energy-related areas so that both SDG-7 and SDG-13 can be achieved through various actions that are consistent with the sub-targets of these SDGs. Hence, the results of this study may have large application areas in participating in the achievement of SDGs for the USA as well as other countries across the world. Thus, this study may have a much broader societal impact on decarbonization, carbon neutrality, energy, and environmental points in ensuring climate change through succeeding SDGs.

Limitations and future research

In the investigation of data conversion impact on the prediction results, the study has tried to make a very detailed empirical analysis. Although such an approach is followed up, nevertheless, the study has some limitations.

First, due to the data availability, the study examines only the USA case. For this reason, new studies should test the validity of the findings of this research on other countries including both developed and developing, if the appropriate data can be available for them.

Second, the study considers mainly energy-related CO₂ emissions and source-based energy use indicators in investigating the impact of data conversion on the prediction results. However, this approach does not consider other potentially effective factors, such as economic growth. So, new studies can include such influential factors in further empirical analyses.

Third, the study makes the empirical investigation by relying on main econometric (i.e. CR) approaches. Therefore, new studies can either perform other time series approaches that are not included in this study or prefer to apply machine learning algorithms as well. In addition, new studies can consider applying Monte Carlo simulation to analyze the issue across various scenarios. Even, new research can compare time series approaches and machine learning approaches.

Fourth, some other points, such as unit root tests with structural breaks and cointegration analyses with structural breaks or regime shifts, can be considered in new studies.

Last, since the study considers only QA and QS approaches in data conversion, new studies can consider other data conversion approaches (e.g. constant, linear, cubic, etc.). In this way, the validity of this research can be checked for other data conversion approaches. By embedding the points mentioned above into new research, the literature on the issue can be enriched further.

Footnotes

Abbreviations

Acknowledgments

Not applicable.

Authors’ contributions

There is one author, who prepared this work. He read and approved the final manuscript.

Author’s note

The authors certify that they have no affiliations with or involvement in any organization or entity with any financial interest (such as honoraria; educational grants; participation in speakers’ bureaus; membership, employment, consultancies, stock ownership, or other equity interest; and expert testimony or patent-licensing arrangements), or non-financial interest (such as personal or professional relationships, affiliations, knowledge, or beliefs) in the subject matter or materials discussed in this article.

Availability of data and materials

Data will be made available on request.

Consent for publication

The authors are willing to permit the Journal to publish the article.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Ethics approval and consent to participate

Not applicable.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Mustafa Tevfik Kartal

Mustafa Tevfik Kartal is a capital market professional. He received a PhD in banking from Marmara University, İstanbul/Türkiye in 2017 and had an Associate Professor degree in banking from Inter-Universities Board, Ankara/Türkiye in 2020. His research interests mainly focus on finance, economics, energy, and environmental areas. He has authored 2 books, 2 book reviews, 23 book chapters, 202 articles, and 13 proceedings, most of which are indexed in SSCI, SCI-E, and SCOPUS indices. Also, 22 articles have been under review. Besides, he has taken a guest editorship role in some prestigious journals of Elsevier and Springer. Moreover, he has been working on various articles in finance, economics, energy, and environment areas. Furthermore, he has a reviewer role in 131 journals indexed in SSCI, SCI-E, and SCOPUS indices.

Annexes

Annex 1.

OLS approximation for the prediction of CO₂ emissions.

Variable	Coefficient	Std. Error	t-Statistic	Prob.
BEC	−28.48	4.31	−6.61	0.0000
CEC	89.62	0.96	93.32	0.0000
GEC	1360.08	192.49	7.07	0.0000
NEC	−12.25	7.13	−1.72	0.0865
NGC	56.65	0.32	174.58	0.0000
OEC	66.33	0.49	134.09	0.0000
SEC	−111.29	12.50	−8.90	0.0000
WEC	−80.47	16.13	−4.99	0.0000
R²	99.69	Adj. R²	99.68

OLS: ordinary least squares; CO₂: carbon dioxide. BEC, CEG, GEC, NEC, NGC, OEC, SEC, and WEC denote energy use from biomass, coal, geothermal, nuclear electric, natural gas, oil (petroleum), solar, and wind sources.

Annex 2. Graphical representation of the variables in monthly original series (MOS) dataset.

Annex 3. Graphical representation of the variables in monthly converted series by quadratic average approach (MCSQA) dataset.

Annex 4. Graphical representation of the variables in monthly converted series by quadratic sum approach (MCSQS) dataset.

References

Kuşkaya

. Residential solar energy consumption and greenhouse gas nexus: evidence from Morlet wavelet transforms. Renew Energy 2022; 192: 793–804.

Ulussever

Kartal

Kılıç Depren

. Effect of income, energy consumption, energy prices, political stability, and geopolitical risk on the environment: evidence from GCC countries by novel quantile-based methods. Energ Environ 2023. DOI: 10.1177/0958305X231190351.

Grossman

Krueger

. Environmental impacts of a North American Free Trade Agreement. National Bureau of Economic Research , No. W3914, 1991.

Pao

Yang

. Modeling the CO₂ emissions, energy use, and economic growth in Russia. Energy 2011; 36: 5094–5100.

Magazzino

Mele

Schneider

. A machine learning approach on the relationship among solar and wind energy production, coal consumption, GDP, and CO₂ emissions. Renew Energy 2021; 167: 99–115.

Rahman

Nepal

Alam

. Impacts of human capital, exports, economic growth and energy consumption on CO₂ emissions of a cross-sectionally dependent panel: evidence from the newly industrialized countries (NICs). Environ Sci Policy 2021; 121: 24–36.

Kraft

. On the relationship between energy and GNP. J Energy Dev 1978; 3: 401–403.

Sharif

Bhattacharya

Afshan

, et al. Disaggregated renewable energy sources in mitigating CO₂ emissions: new evidence from the USA using quantile regressions. Environ Sci Pollut Res 2021; 28: 57582–57601.

Kartal

. Production-based disaggregated analysis of energy consumption and CO₂ emission nexus: evidence from the USA by novel dynamic ARDL simulation approach. Environ Sci Pollut Res 2023; 30: 6864–6874.

10.

Ulussever

Kılıç Depren

Kartal

, et al. Estimation performance comparison of machine learning approaches and time series econometric models: evidence from the effect of sector-based energy consumption on CO₂ emissions in the USA. Environ Sci Pollut Res 2023; 30: 52576–52592.

11.

Kartal

Magazzino

Pata

. Marginal effect of electricity generation on CO₂ emissions: disaggregated level evidence from China by KRLS method and high-frequency daily data. Energy Strat Rev 2024; 53: 101382.

12.

Sharif

Mishra

Sinha

, et al. The renewable energy consumption-environmental degradation nexus in top-10 polluted countries: fresh insights from quantile-on-quantile regression approach. Renew Energy 2020; 150: 670–690.

13.

Adebayo

Kirikkaleli

. Impact of renewable energy consumption, globalization, and technological innovation on environmental degradation in Japan: application of wavelet tools. Environ Dev Sustain 2021; 23: 16057–16082.

14.

Meng

Khan

Adebayo

, et al. How do oil price uncertainty and economic policy uncertainty influence achievement of net-zero emissions targets by 2050? A quantile-based analysis. Int J Sust Dev World Ecol 2024; 31: 892–911.

15.

Adebayo

Khan

, et al. United States’ 2050 carbon neutrality: myth or reality? Evaluating the impact of high-tech industries and green electricity. J Clean Prod 2024; 440: 140855.

16.

EIA. Data of energy-related CO₂ emissions and energy use, https://www.eia.gov/totalenergy/data/monthly (March 27, 2024).

17.

World Bank. (2024). Data of GDP. https://data.worldbank.org/indicator, Accessed on 15 July 2024.

18.

Energy Institute. Statistical review of world energy data, https://www.energyinst.org/statistical-review/resources-and-data-downloads (March 27, 2024).

19.

Nachane

Nadkarni

Karnik

. Co-integration and causality testing of the energy–GDP relationship: a cross-country study. Appl Econ 1988; 20: 1511–1531.

20.

Bilgili

Nathaniel

Kuşkaya

, et al. Environmental pollution and energy research and development: an environmental Kuznets curve model through quantile simulation approach. Environ Sci Pollut Res 2021; 28: 53712–53727.

21.

Wang

Guo

. Revisiting the environmental Kuznets curve (EKC) hypothesis of carbon emissions: exploring the impact of geopolitical risks, natural resource rents, corrupt governance, and energy intensity. J Environ Manage 2024; 351: 119663.

22.

Wang

. Rethinking the environmental Kuznets curve hypothesis across 214 countries: the impacts of 12 economic, institutional, technological, resource, and social factors. Hum Soc Sci Commun 2024; 11: 1–19.

23.

Wang

, et al. Reinvestigating the environmental Kuznets curve (EKC) of carbon emissions and ecological footprint in 147 countries: a matter of trade protectionism. Hum Soc Sci Commun 2024; 11: 1–17.

24.

Wang

Zhang

. Does emerging country's urbanization redefine the environmental Kuznets hypothesis for carbon emissions? A novel perspective on national and subnational differences. Energy Environ 2024. DOI: 10.1177/0958305X24123062.

25.

Kartal

. The role of consumption of energy, fossil sources, nuclear energy, and renewable energy on environmental degradation in top-five carbon producing countries. Renew Energy 2022; 184: 871–880.

26.

Bilgili

. The impact of biomass consumption on CO₂ emissions: cointegration analyses with regime shifts. Renew Sust Energy Rev 2012; 16: 5349–5354.

27.

Kuşkaya

Bilgili

. The wind energy-greenhouse gas nexus: the wavelet-partial wavelet coherence model approach. J Clean Prod 2020; 245: 118872.

28.

Magazzino

Toma

Fusco

, et al.

Renewable energy consumption, environmental degradation and economic growth: the greener the richer?

Ecol Indic 2022; 139: 108912.

29.

Zhen

Ullah

Shaowen

, et al.

How do renewable energy consumption, financial development, and technical efficiency change cause ecological sustainability in European Union countries?

Energy Environ 2022 ; 34(7): 2478–2496.

30.

Kuşkaya

Bilgili

Muğaloğlu

, et al. The role of solar energy usage in environmental sustainability: fresh evidence through time-frequency analyses. Renew Energy 2023; 206: 858–871.

31.

Balcılar

Gupta

Pierdzioch

. Does uncertainty move the gold price? New evidence from a nonparametric causality-in-quantiles test. Resour Policy 2016; 49: 74–80.

32.

Bhattacharya

Paramati

Öztürk

, et al. The effect of renewable energy consumption on economic growth: evidence from top 38 countries. Appl Energy 2016; 162: 733–741.

33.

Shahbaz

Balcılar

Mahalik

, et al.

Is causality between globalization and energy consumption bidirectional or unidirectional in top and bottom globalized economies?

Int J Financ Econ 2023; 28: 1939–1964.

34.

Park

. Canonical cointegrating regression. Econometrica 1992; 60: 119–143.

35.

Stock

Watson

. A simple estimator of cointegrating vectors in higher order integrated system. Econometrica 1993; 61: 783–820.

36.

Hansen

Phillips

. Estimation and inference in models of cointegration: a simulation study. Adv Econ 1988; 8: 225–248.

37.

Willmott

Ackleson

Davis

, et al. Statistics for the evaluation and comparison of models. J Geophys Res-Oceans 1985; 90: 8995–9005.

38.

Kılıç Depren

Kartal

Ertuğrul

, et al. The role of data frequency and method selection in electricity price estimation: comparative evidence from Turkey in pre-pandemic and pandemic periods. Renew Energy 2022; 186: 217–225.

39.

Pata

. Renewable energy consumption, urbanization, financial development, income and CO₂ emissions in Turkey: testing EKC hypothesis with structural breaks. J Clean Prod 2018; 187: 770–779.

40.

Adebayo

Meo

Eweade

, et al. Analyzing the effects of solar energy innovations, digitalization, and economic globalization on environmental quality in the United States. Clean Technol Environ Policy 2024. DOI: 10.1007/s10098-024-02831-0.

41.

Adebayo

Meo

Eweade

, et al. Examining the effects of solar energy innovations, information and communication technology and financial globalization on environmental quality in the United States via Quantile-on-Quantile KRLS analysis. Sol Energy 2024; 272: 112450.

42.

Adebayo

Ullah

. Towards a sustainable future: the role of energy efficiency, renewable energy, and urbanization in limiting CO₂ emissions in Sweden. Sustain Dev 2024; 32: 244–259.

43.

Alola

Tauni

, et al.

Pathway to cleaner environment: how effective are renewable electricity and financial development approaches?

Struct Change Econ Dyn 2023; 67: 277–292.

44.

Liu

Irfan

, et al. A safe path towards carbon neutrality by 2050: assessing the impact of oil and gas efficiency using advanced quantile-based approaches. J Clean Prod 2023; 425: 138844.

45.

Pata

Luo

Kartal

, et al. Do technological innovations and clean energies ensure CO₂ reduction in China? A novel nonparametric causality-in-quantiles. Energ Environ 2023. DOI: 10.1177/0958305X231210993.

46.

Wang

Adebayo

, et al. Can Finland serve as a model for other developed countries? Assessing the significance of energy efficiency, renewable energy, and country risk. J Clean Prod 2023; 428: 139306.

47.

Adebayo

Pata

Akadiri

. A comparison of CO₂ emissions, load capacity factor, and ecological footprint for Thailand’s environmental sustainability. Environ Dev Sustain 2024; 26: 2203–2223.

48.

Adebayo

Alola

Ullah

. Probing the carbon neutrality drive of environmental-related technologies and energy transition in France and Germany: A novel time–frequency technique. Clean Technol Environ Policy 2024. DOI: 10.1007/s10098-024-02816-z.

49.

Kartal

Kılıç Depren

Ayhan

, et al. Quantile-based heterogeneous effects of nuclear energy and political stability on the environment in highly nuclear energy-consuming and politically stable countries. Appl Energy 2024; 365: 123237.