Gas in Transition: An ARDL Analysis of Economic and Fuel Drivers in the European Union

Abstract

This study investigates the short- and long-run drivers of natural gas consumption in the European Union using an ARDL bounds testing approach. The analysis incorporates GDP per capita, liquid fuel use, and solid fuel use as explanatory variables. Augmented Dickey–Fuller tests confirm mixed integration orders, allowing valid ARDL estimation. The results reveal a statistically significant long-run relationship (cointegration) between gas consumption and the energy–economic system. In the short run, the use of liquid fuel exerts a strong positive influence on gas demand, while the effects of GDP materialise only after a two-year lag. Solid fuels show a delayed substitutive impact, reflecting the ongoing transition from coal. An error correction model confirms rapid convergence to equilibrium, with 77% of deviations corrected within one period. Recursive residual and CUSUM tests indicate structural stability over time. These findings highlight the responsiveness of EU gas demand to both economic and policy signals, offering valuable insights for energy modelling and strategic planning under the European Green Deal.

1. Introduction

The energy transition in Europe is unfolding in an environment of intensifying geopolitical risk, volatile energy prices, and increasingly stringent climate commitments. Against this backdrop, natural gas occupies a particularly ambivalent position. On the one hand, it is often seen as a “bridge fuel” in the transition toward decarbonisation due to its lower carbon intensity compared to coal and oil. On the other hand, the EU’s heavy dependence on gas imports—particularly from politically sensitive regions—has exposed serious vulnerabilities in terms of energy security, especially in the wake of the 2022 Russian invasion of Ukraine.

Understanding the structural drivers of natural gas consumption is critical for developing effective and resilient energy policies. While a vast body of research exists on the relationship between energy consumption and macroeconomic variables, much of it tends to focus either on aggregate energy use or electricity demand. Studies that explicitly analyse natural gas consumption as a distinct dependent variable—particularly within a substitution framework—remain relatively limited. Even fewer contributions adopt a pan-European perspective, despite the EU’s increasingly integrated energy market and harmonised climate legislation.

Recent contributions [1,2,3,4] have begun to explore fuel-switching dynamics under regulatory and market constraints. M. Busu [4,5] and others [6,7,8] have demonstrated the value of macro-level econometric models in estimating long-term substitution patterns. Still, many of these studies are either sector-specific or focus on price responses without accounting for broader structural shifts in fuel composition. Additionally, literature has often neglected to explicitly link empirical findings to current EU strategies, such as REPowerEU and Fit for 55, which are at the heart of the bloc’s decarbonisation and energy security agendas.

This study addresses these gaps by empirically modelling both the long-run and short-run determinants of natural gas consumption in the EU using an ARDL approach. We explicitly incorporate GDP per capita, liquid fuel use, and solid fuel use as explanatory variables, capturing both the effects of economic growth and the dynamics of fuel substitution. The ARDL model is particularly well suited to our context, as it handles small sample sizes, mixed integration orders (I(0)/I(1)), and provides both error correction and elasticity estimates. Moreover, its transparency and interpretability make it preferable to machine learning models such as LSTM or XGBoost for structural economic analysis.

Our empirical analysis spans the period from 1990 to 2021 and includes harmonised annual data for all 27 EU Member States, accounting for both early and later accession countries. By modelling the EU as a single macroeconomic bloc, we can trace region-wide substitution patterns and policy-relevant responses, rather than relying solely on fragmented national analyses. Particular attention is given to the responsiveness of gas demand to changes in liquid and solid fuel consumption, which is interpreted as structural shifts in the heating, transport, and industrial sectors.

The findings reveal statistically significant cointegration relationships, providing clear evidence of substitution between gas and other fuel types. These results align with the cross-fuel flexibility envisioned in the EU’s REPowerEU strategy and the demand modulation goals of the Fit for 55 package. More broadly, the study contributes to a growing literature on energy system transformation under uncertainty, providing a framework for evaluating the role of natural gas in a rapidly evolving energy landscape.

In doing so, this research aims to support both academic understanding and policymaking by offering empirically grounded insights into fuel substitution mechanisms, macroeconomic drivers, and the long-term implications of EU energy policy.

2. Literature Review

2.1. Energy–Economy Nexus and Fossil Fuel Demand

The relationship between energy consumption and economic growth has been a foundational theme in energy economics. Numerous empirical studies, including those by Apergis and Payne [2], Ribeiro et al. [5], Ozturk [9], Sadorsky [10], and Kalyoncu et al. [11], among others [1,2,3,4,12,13], have investigated the causal direction and strength of this relationship across diverse economies and energy types. Most analyses focus on aggregate energy consumption, with some evidence of bidirectional causality, particularly in industrialised countries. However, relatively few studies have disaggregated energy sources, such as natural gas, to evaluate their independent interactions with macroeconomic indicators.

In the context of Central and Eastern Europe, Myszczyszyn, J., and Suproń, B. [14] applied ARDL modelling to investigate the relationship between energy consumption and economic growth in Visegrad countries, identifying significant long-run elasticities between GDP and energy use. Their results suggest that fossil fuel demand is not only income elastic but also sensitive to structural changes in the economy, such as deindustrialisation or sectoral shifts.

These findings support the inclusion of GDP in dynamic models of gas demand. However, macroeconomic drivers alone may not fully explain fossil fuel behaviour, mainly when structural shifts in the energy mix are occurring concurrently.

2.2. ARDL and Dynamic Modelling Approaches

To handle time-series data with mixed integration orders (I(0) and I(1)), the Autoregressive Distributed Lag (ARDL) model has emerged as a preferred tool. Recent studies, such as those by Obadi and Korcek, demonstrate the model’s utility in capturing both environmental and economic dynamics in the EU context, showing significant long- and short-run interactions between renewable energy consumption, CO₂ emissions, and GDP [15]. Introduced by Pesaran et al. [16], the ARDL bounds testing framework enables simultaneous estimation of short- and long-run relationships without requiring all series to be stationary in the same order. Numerous energy economics studies have employed this method to investigate the causal relationships between energy use, income, and policy variables [8,14]. For example, a recent study applied a panel ARDL framework to assess the long-run determinants of renewable energy consumption in the EU, confirming GDP and oil prices as key structural drivers [5]. Shahbaz et al. [8] employed ARDL techniques to examine the effect of energy consumption and capital on economic growth in Pakistan, finding long-run cointegration among the variables. This reinforces the suitability of ARDL models in capturing both energy market dynamics and macroeconomic feedbacks in EU contexts.

Similarly, despite its widespread application, ARDL is rarely used to model gas demand for alternative fuels, such as liquid and solid sources. The present study addresses this gap by integrating cross-fuel interactions into an ARDL framework, capturing the dynamics of substitution or complementarity over time.

The analysis of energy consumption dynamics within the European Union (EU) using Autoregressive Distributed Lag (ARDL) modelling offers valuable insights into the relationship between energy consumption, economic growth, and sustainability initiatives. This literature review synthesises recent studies on the subject, highlighting how ARDL approaches have been employed to uncover both short-run and long-run causality in energy consumption patterns across EU member states.

One key study by Marinaș et al. [6] employs the ARDL methodology to investigate the correlation between renewable energy consumption and economic growth, specifically in Central and Eastern European countries from 1990 to 2014. Their findings suggest a significant relationship where an increased share of renewable energy was associated with enhanced economic performance, indicating that renewable energy consumption plays a critical role in achieving sustainable economic growth in these regions [6]. This research emphasises the utility of ARDL modelling in capturing both immediate and lagged effects of renewable energy integration on economic indicators.

While previous studies have successfully applied ARDL and cointegration methods to model energy consumption, they typically focus either on total energy use or single-country settings. Obadi and Korcek [15] further confirm the limitations of single-country ARDL approaches, suggesting that EU-wide modelling better captures structural interdependencies. Furthermore, the substitution dynamics between fuel types are often theorised but seldom integrated directly into empirical models. Recent work [4] supports this view by applying a panel ARDL framework to evaluate the macroeconomic impact of renewables across EU countries, confirming that disaggregated fuel types have differentiated effects on economic growth [4].

Rokicki et al. [16] examine how energy consumption and intensity have changed in response to the COVID-19 pandemic across various EU sectors. Their analysis reveals a complex interplay between economic activities and energy use, suggesting that short-term disruptions can lead to long-lasting changes in consumption trends. By utilising ARDL models to analyse these shifts, the study underscores the importance of understanding dynamic responses to external shocks, such as pandemics, which can significantly reshape energy landscapes [16].

Another relevant investigation by Bozkurt and Okumuş [3] employs panel data analysis to evaluate the Environmental Kuznets Curve hypothesis across selected EU countries, examining the interactions among economic growth, energy consumption, and CO₂ emissions. This study reinforces the applicability of ARDL approaches in determining how economic development influences energy consumption practices and environmental outcomes over time, further supporting the robustness of the model in empirical research [8].

Moreover, Simionescu et al. explore the implications of renewable energy on economic performance in the context of the European Green Deal, illustrating long-term projections of renewable energy integration within the energy consumption framework of EU nations. Their findings contribute to a deeper understanding of how economic policies aimed at environmental sustainability can influence energy usage patterns [17]. The ARDL model played a crucial role in elucidating these relationships by enabling the analysis of distinct temporal dynamics. Additionally, Tzeiranaki et al. [18] provide an analysis of residential energy consumption trends within the EU, utilising an econometric approach that reinforces findings from the aforementioned studies regarding consumption patterns and efficiency goals set by EU energy policies. This research reveals that understanding these dynamics is crucial for developing effective future energy policies [18].

In conclusion, the application of ARDL modelling techniques in studying the dynamics of energy consumption across EU countries reveals a multifaceted and dynamic relationship between energy use, economic growth, and external influences. Each of the reviewed studies illustrates the critical role of ARDL approaches in capturing these dynamics, effectively supporting the identification of paths toward sustainable energy consumption and efficient policy formulations in the EU.

2.3. Fuel Substitution and Energy Mix Dynamics

The concept of fuel substitution is central to energy transition modelling. Changes in the relative prices, policy incentives, or infrastructure availability can shift consumption patterns across fuel types [19]. For instance, natural gas has historically been considered a “transition fuel”—a cleaner alternative to coal, especially in heating, industrial, and electricity generation sectors.

The dynamics of fuel substitution and energy mix in the European Union (EU) are complex and shaped by various factors, including policy, technological advancements, and market conditions. This review synthesises recent literature to summarise the state of fuel substitution and changes in energy mix within the context of the EU’s energy transition.

Recent studies indicate that the EU’s energy mix is undergoing a significant transformation, primarily driven by an increase in renewable energy sources (RES) and an ongoing debate over the country’s dependency on fossil fuels. Ślosarski argues that while the transition towards renewable energy commenced decades ago, statistical evidence suggests that this shift does not significantly reduce greenhouse gas emissions associated with fossil fuel consumption [20]. The findings indicate that the impact of renewable energy on fossil fuel reliance might be more gradual and nuanced than simplistic interpretations would suggest.

The comparative analysis conducted by Daroń and Wilk provides insights into how varying degrees of RES implementation across EU countries influence their energy production sector [13]. Their assessment, based on Eurostat data, highlights that while renewable energies, such as solar and wind, are assuming a more prominent role, a substantial reliance on traditional energy sources persists. This highlights the slow pace of transition even in the face of EU policies designed to foster energy diversification and sustainability.

A household-level perspective is offered by Piekut, who discusses the increasing adoption of RES among EU households between 2004 and 2019 [21]. The data reveal developmental trends suggesting varying degrees of success across EU nations, influenced by socio-economic status and local policies. This underscores that while some regions have successfully integrated renewables at the household level, challenges remain, particularly in low-income areas.

In another vein, Tzeiranaki et al. examine the determinants and trends of residential energy consumption, contributing to our understanding of how consumer behaviour interacts with broader energy policies [18]. Their analysis highlights the significant role of energy efficiency measures and consumption patterns in shaping energy demands. The EU’s targets for energy consumption reductions by 2020 and 2030 are pivotal in this context, suggesting that achieving these benchmarks necessitates comprehensive policy implementation and consumer engagement.

Moreover, the geopolitics of energy supply has come into sharper focus following recent external pressures, as highlighted in the work by Liu et al. [22]. The authors discuss the EU’s response to the Russian–Ukrainian war, emphasising a strategic pivot towards domestic renewable energy sources as a means to enhance energy security. Furthermore, Müller and Teixidó-Figueras critically assess the role of the European Union Emissions Trading System (EU ETS), describing how effective carbon pricing can incentivise fuel switching in coal-reliant nations [23]. Their research underscores the importance of stable and high carbon pricing as a driver for energy diversification toward cleaner alternatives.

Simultaneously, challenges to implementing these changes are elaborated in studies examining barriers to RES installations [6] and the economic implications of fuel tax policies. Regulatory constraints, market design issues, and infrastructural challenges hinder the widespread adoption of Renewable Energy Sources (RES). Additionally, Proedrou [24] reflects on how geopolitical scenarios can reshape energy policies, suggesting that external pressures can facilitate the energy transition but may also create new dependencies on various fuel sources, potentially undermining long-term sustainability goals [20].

In conclusion, the literature indicates that while the EU is making strides towards diversifying its energy mix and increasing the share of renewable sources, substantial challenges remain. These include harmonising diverse national policies into a coherent framework, addressing socio-economic disparities in energy access, and ensuring that geopolitical uncertainties do not derail the push towards sustainability. The interplay of these dynamics will undoubtedly shape the EU’s energy landscape in the coming years.

3. Methodology

This study applies an Autoregressive Distributed Lag (ARDL) modelling framework to analyse the dynamic relationship between natural gas consumption and its structural determinants in the European Union. The model includes GDP per capita, liquid fuel consumption, and solid fuel consumption as key explanatory variables. These variables are chosen to reflect macroeconomic activity and sectoral energy shifts that influence gas demand across the EU’s integrated energy system.

GDP per capita serves as a proxy for economic growth and structural energy demand, in line with established energy economics literature [4,9]. Liquid and solid fuel consumption are included to capture fuel substitution dynamics within the final energy mix, particularly in the transport, heating, and power generation sectors. The use of these variables does not imply strict causality but rather reflects structural substitution effects that unfold over time. The ARDL model is particularly well suited for capturing both short-run and long-run interdependencies without imposing strong assumptions regarding exogeneity.

The empirical analysis is based on annual data spanning the period from 1990 to 2022. This time window encompasses both pre-accession and post-accession periods for Member States that joined the EU in later enlargement rounds, enabling a harmonised examination of structural energy transitions across the EU-27. Energy consumption data (natural gas, liquid fuels, and solid fuels, in metric tonnes) were obtained from the International Energy Agency (IEA) World Energy Balances [19,25]. GDP per capita figures (in constant 2015 USD) were sourced from the World Bank’s World Development Indicators.

The classical ARDL model was selected due to its flexibility in handling small samples and variables with mixed integration orders. (I(0) and I(1)), as confirmed by unit root testing. Alternative models, such as PMG-ARDL and CS-ARDL are designed explicitly for panel datasets and are therefore less applicable to this macro-level single-unit framework. Similarly, while machine learning techniques such as LSTM or XGBoost offer high predictive power, they lack transparency in coefficient interpretation and are therefore less appropriate for policy-relevant structural modelling.

Although asymmetric ARDL variants such as NARDL can model nonlinear dynamics, their application typically requires longer time series than those available here. Given the study’s objective—to identify structural substitution elasticities in gas consumption—the ARDL model offers an optimal balance of interpretability, empirical rigour, and methodological relevance.

3.1. Model Specification

The general form of the ARDL (p, q₁, q₂, q₃) model is specified as follows:

$${\mathit{Energy}\_\mathit{Gas}}_t={\alpha }_0+\sum _{i=1}^p{\varphi }_i{\mathit{Energy}\_\mathit{Gas}}_{t-i}+\sum _{j=0}^{q_1}{\beta }_j{\mathit{GDP}}_{t-j}+\sum _{k=0}^{q_2}{\gamma }_k{\mathit{Liquid}}_{t-k}+\sum _{l=0}^{q_3}{\delta }_l{\mathit{Solid}}_{t-l}+{\epsilon }_t$$

(1)

where:

-

${\mathit{Energy}\_\mathit{Gas}}_t$ is the consumption of natural gas in year t,

-

${\mathit{GDP}}_t$ is GDP per capita,

-

${\mathit{Liquid}}_t$ is liquid and fuel consumption.

-

${\mathit{Solid}}_t$ is solid fuel consumption.

-

${\epsilon }_t$ is the white noise error term.

Lag lengths were selected using the Akaike Information Criterion (AIC), balancing parsimony and model fit.

3.2. Unit Rot and Integration Testing

To verify the stationarity properties of the variables, the Augmented Dickey–Fuller (ADF) test was applied to each series in levels and first differences. The ARDL bounds testing procedure requires that none of the series be integrated of order 2 or higher.

3.3. Bounds Test for Cointegration

Following model estimation, we applied the ARDL bounds test for cointegration [19] to examine whether a long-run equilibrium relationship exists among the variables. In the absence of a functional bounds test interface in our environment, we supplemented this with the Engle–Granger residual-based test, which examines whether the residuals from the long-run ARDL model are stationary.

3.4. Error Correction Model (ECM)

If cointegration was found, the short-run dynamics were modelled using an Error Correction Model (ECM) derived from the ARDL specification. The ECM includes (1) differenced terms to capture short-run adjustments, and (2) a lagged residual term to reflect deviations from the long-run equilibrium. The coefficient on the error correction term reflects the speed of adjustment, indicating how quickly the system returns to equilibrium after a shock.

3.5. Robustness Checks

To assess model stability and sensitivity, we conducted the following tests: (1) alternative lag structures (e.g., ARDL(1,1,1), ARDL(2,2,2)); (2) inclusion/exclusion of explanatory variables (e.g., Solid fuel); (3) log–log model estimation for elasticity interpretation; and (4) structural stability tests using recursive residuals and Cumulative Sum of Recursive Residuals (CUSUM). All estimations were conducted using the statsmodels econometrics library in Python 3.10, ensuring replicability and transparency.

4. Results

Initial time-series plots (Figure 1) of gas consumption, GDP per capita, and alternative fuel use (liquid and solid) in the European Union revealed pronounced long-term trends. Gas usage exhibits cyclical volatility, with a moderate upward trend, while GDP per capita continues to grow steadily.

Notably, solid fuel use declines persistently, reflecting decarbonisation, whereas liquid fuel use exhibits relative stability.

Table 1. ADF test results for raw data.

Variable	Test Statistic	p-Value	Critical Value (1%)	Critical Value (5%)	Critical Value (10%)	Stationary
Energy_Gas_EU	−1.73246	0.414468	−3.66143	−2.96053	−2.61932	False
GDP_per_EU	−0.86658	0.798779	−3.66143	−2.96053	−2.61932	False
Energy_Liquid_EU	0.102067	0.966202	−3.66143	−2.96053	−2.61932	False
Energy_Solid_EU	−0.40646	0.909014	−3.67906	−2.96788	−2.62316	False

Note: Energy_Gas_EU—natural gas consumption in the EU; GDP_per_EU—GDP per capita in the EU; Energy_Liquid_EU—liquid fuel consumption; Energy_Solid_EU = solid fuel consumption. ADF test includes an intercept (no trend). Critical values correspond to MacKinnon’s one-sided p-values. “Stationary = False” indicates failure to reject the null hypothesis of unit root at all conventional levels.

presents the Augmented Dickey–Fuller (ADF) test results for each variable in its level form. None of the series is stationary at the 5% significance level. Upon first differencing, stationarity is confirmed for gas, GDP per capita, and solid fuel use (

Table 2. ADF test results for first differences.

Variable	Test Statistic	p-Value	Critical Value (1%)	Critical Value (5%)	Critical Value (10%)	Stationary
d_Energy_Gas_EU	−6.09467	1.02 × 10−7	−3.66992	−2.96407	−2.62117	True
d_GDP_per_EU	−6.1062	9.57 × 10−8	−3.66992	−2.96407	−2.62117	True
d_Energy_Liquid_EU	−2.38484	0.14606	−3.67906	−2.96788	−2.62316	False
d_Energy_Solid_EU	−5.23693	7.39 × 10−6	−3.67906	−2.96788	−2.62316	True

Note: d_Energy_Gas_EU = first difference of natural gas consumption in the EU; d_GDP_per_EU = first difference of GDP per capita; d_Energy_Liquid_EU = first difference of liquid fuel consumption; d_Energy_Solid_EU = first difference of solid fuel consumption. All variables are in raw (non-logged) form. The Augmented Dickey–Fuller (ADF) test includes an intercept only. Critical values are based on MacKinnon’s one-sided p-values. “Stationary = True” indicates rejection of the null hypothesis of a unit root at the 10% significance level or lower.

), validating the ARDL framework’s requirement for mixed I(0)/I(1) integration orders.

Using the AIC, the optimal specification was ARDL(1, 2, 0), which models gas consumption as a function of GDP per capita and liquid fuel use (Figure 2).

Table 3. ARDL(1,2,0) model summary.

Variable	Lag	Coefficient	p-Value
Energy_Gas_EU (L1)	1	0.259	0.038
GDP_per_EU (L2)	2	11.47	0.002
Energy_Liquid_EU (L0)	0	0.517	<0.001

Note: The table presents estimated coefficients and p-values from the ARDL(1,2,0) model, where the dependent variable is natural gas consumption in the EU. Lag refers to the number of periods by which the independent variable is lagged. Abbreviations: Energy_Gas_EU = natural gas consumption; GDP_per_EU = gross domestic product per capita; Energy_Liquid_EU = liquid fuel consumption. All variables are in levels (non-differenced), with real GDP per capita in constant 2015 EUR and energy consumption in tonnes.

summarises the model outputs.

-

Persistence is evident, with a significant lagged gas term (coef. = 0.259, p = 0.038).

-

GDP per capita exerts a delayed influence, with only the second lag being significant (coef. = 11.47, p = 0.002).

-

Liquid fuel use is immediately impactful (coef. = 0.517, p < 0.001), suggesting complementarities or shared macroeconomic drivers.

The model captures medium-term dynamics and reveals a high level of in-sample fit, particularly during periods of sustained growth and post-crisis recovery.

The Engle–Granger test on ARDL residuals confirms cointegration (ADF = −5.35, p < 0.001), implying a long-run equilibrium exists among the variables. This supports the derivation of an Error Correction Model (ECM).

The ECM results (

Table 4. ECM estimation results (ARDL-based model).

Variable	Coefficient	Std. Error	t-Statistic
ECT (Error Correction Term)	–0.770 **	0.144	–5.135
Δ(GDP_per_EU, lag 2)	12.473 **	3.112	4.008
Δ(Energy_Liquid_EU)	0.607 **	0.111	5.470
Δ(Energy_Gas_EU, lag 1)	0.310 *	0.134	2.309
Constant	101.8 *	43.52	2.339

Note: The table shows results from the short-run dynamics of the ARDL-based Error Correction Model (ECM), where the dependent variable is the first difference of natural gas consumption in the EU. ECT indicates the speed of adjustment back to long-run equilibrium after a shock. Δ denotes the first difference of the variable. Abbreviations: GDP_per_EU—gross domestic product per capita (real, 2015 EUR); Energy_Liquid_EU—liquid fuel consumption (mtoe); Energy_Gas_EU—natural gas consumption (tonnes). Dependent variable: Δ(Energy_Gas_EU), Estimation period: 1990–2021, Significance levels: * p < 0.1; ** p < 0.05.

) reveal rapid correction of disequilibrium, with an error correction term of −0.770 (p = 0.005).

Approximately 77% of deviations from equilibrium are corrected within one year. Short-run effects include a significant positive impact from liquid fuel consumption (coef. = 0.597, p = 0.007). The scatterplot (Figure 3) shows a negative relationship between the lagged residuals and the short-run change in gas consumption, indicating convergence toward long-run equilibrium.

The scatterplot in Figure 3 illustrates the short-run adjustment mechanism captured by the ECM term. Observations lie predominantly in the upper-left and lower-right quadrants, indicating a corrective relationship; when gas consumption deviates from its long-run equilibrium (positive or negative residuals), subsequent changes tend to move in the opposite direction. This negative correlation supports the validity of the error correction structure, confirming that deviations from equilibrium are self-correcting over time. The magnitude and speed of this adjustment, as reflected by the ECM coefficient (–0.741), indicate a relatively fast convergence, with over 70% of disequilibrium corrected within one year.

Lag structure sensitivity confirms the stability of findings across ARDL (1,1,1) and ARDL (2,2,2). The inclusion of solid fuel exhibits weak substitutive signals.

The log–log specification (Table 5) offers the following elasticity interpretations:

-

A 1% increase in liquid fuel use leads to ~0.86% increase in gas use (p < 0.001).

-

Solid fuel exhibits delayed adverse effects, consistent with trends in fuel substitution.

-

GDP remains economically relevant, although statistically weak in the short term.

Table 5. Log–log specification: estimated coefficients.

Variable	Coefficient	Std. Error	t-Statistic
ln(GDP_per_EU)	1.749 ***	0.388	4.51
ln(Energy_Liquid_EU)	0.412 ***	0.097	4.25
ln(Energy_Solid_EU)	–0.108 *	0.053	–2.04
Constant	–3.126 *	1.421	–2.20

Note: The table presents OLS estimates from a log–log regression, where the dependent variable is the natural logarithm of natural gas consumption in the EU. All variables are in natural logarithms; therefore, the coefficients can be interpreted as elasticities. Abbreviations: ln(GDP_per_EU) = log of real GDP per capita (constant 2015 EUR); ln(Energy_Liquid_EU) = log of liquid fuel consumption (tonnes); ln(Energy_Solid_EU) = log of solid fuel consumption (tonnes); ln(Energy_Gas_EU) = log of gas consumption (tonnes). Estimation method: Ordinary Least Squares (OLS). Period: 1990–2021. Significance levels: *** p < 0.01; * p < 0.1.

Table 5. Log–log specification: estimated coefficients.

Variable	Coefficient	Std. Error	t-Statistic
ln(GDP_per_EU)	1.749 ***	0.388	4.51
ln(Energy_Liquid_EU)	0.412 ***	0.097	4.25
ln(Energy_Solid_EU)	–0.108 *	0.053	–2.04
Constant	–3.126 *	1.421	–2.20

Note: The table presents OLS estimates from a log–log regression, where the dependent variable is the natural logarithm of natural gas consumption in the EU. All variables are in natural logarithms; therefore, the coefficients can be interpreted as elasticities. Abbreviations: ln(GDP_per_EU) = log of real GDP per capita (constant 2015 EUR); ln(Energy_Liquid_EU) = log of liquid fuel consumption (tonnes); ln(Energy_Solid_EU) = log of solid fuel consumption (tonnes); ln(Energy_Gas_EU) = log of gas consumption (tonnes). Estimation method: Ordinary Least Squares (OLS). Period: 1990–2021. Significance levels: *** p < 0.01; * p < 0.1.

The estimated long-run elasticity of gas consumption to GDP per capita is 1.75, indicating that gas demand responds more than proportionally to changes in economic activity. This suggests that gas remains a key energy input across sectors, and its use scales strongly with output levels. The elasticity of liquid fuel consumption (0.41) reflects partial substitution toward oil-based fuels, while the negative elasticity for solid fuels (−0.11) indicates a gradual shift away from coal. These patterns are consistent with an ongoing structural transformation in the EU’s energy system, where gas is increasingly serving as a transitional fuel. The results reinforce the relevance of EU strategies, such as REPowerEU and Fit for 55, which aim to promote flexibility and decarbonization in final energy use.

The close alignment between the actual and fitted values in Figure 4 further supports the robustness of the model specification. The ARDL model accurately captures the long-run equilibrium path of gas demand, indicating that the selected explanatory variables—GDP per capita, liquid fuels, and solid fuels—are suitable structural drivers of gas consumption in the EU context. The model’s predictive accuracy reinforces the suitability of the log–log specification for analysing elasticity-based responses in energy systems undergoing transition.

The visual inspection (Figure 5) of the recursive residuals indicates that the residuals fluctuate around the zero line without significant deviation, supporting the absence of structural instability in the estimated model. Combined with the CUSUM test result (p = 0.979), this confirms that the model’s parameters remain stable throughout the sample period. This stability is essential for drawing consistent policy inferences from both short-run and long-run coefficients.

This comprehensive ARDL framework reveals that EU gas consumption is jointly shaped by lagged economic performance and short-run interactions with liquid fuels. The model’s stability and robustness across specifications make it suitable for long-term forecasting and policy analysis in energy transition contexts.

5. Discussion

This study aimed to investigate the determinants of natural gas consumption across the European Union using a dynamic autoregressive distributed lag (ARDL) framework. The empirical results yield several noteworthy insights into the short-run adjustments and long-run relationships between gas usage, economic performance, and the composition of the energy mix.

Our findings highlight that GDP per capita exerts a lagged, positive influence on gas consumption, with statistical significance emerging only after a two-period delay. This delayed responsiveness is also evident in studies on renewable energy, such as Obadi and Korcek [15], who report similarly time-lagged effects of economic activity on clean energy consumption and emissions. This highlights the temporal disconnect between macroeconomic expansion and the energy system’s response, likely due to the inertia inherent in adapting industrial, residential, and infrastructural systems. This result aligns with earlier studies emphasising the non-instantaneous nature of energy demand elasticity in mature economies. Similar GDP-related lag effects were reported by Ribeiro et al. [5] in the context of renewables, suggesting a consistent temporal disconnect between economic growth and energy system adjustments across fuel types.

Notably, the absence of significant short-run GDP effects suggests that natural gas consumption is not highly sensitive to transitory economic shocks, a relevant finding for energy planners concerned with macroeconomic volatility.

A robust and consistently significant short-run association is observed between liquid fuel consumption and gas use. The positive elasticity (~0.86 in the log–log specification) suggests that these two energy sources are co-driven by similar sectoral demands, particularly in the heating, transport, and industrial sectors. This complementarity effect indicates the presence of joint consumption patterns, possibly moderated by seasonal variations and everyday exposure to climate-driven demand fluctuations.

The implication is that any transition away from liquid hydrocarbons (e.g., petroleum) must be carefully coordinated with gas supply planning to avoid mismatches or unintended rebound effects.

While often omitted in standard models, our robustness analysis suggests that solid fuel use, although generally declining, has a delayed and statistically significant adverse impact on gas demand. This supports the hypothesis that natural gas has functioned as a transitional substitute for coal in the EU’s decarbonisation trajectory. The presence of a delayed substitution effect strengthens calls for modelling energy systems with explicit fuel-switching mechanisms [15].

However, the effect is neither significant nor immediate, suggesting that technological and regulatory constraints mediate this substitution process.

The estimated error correction term of −0.77 indicates rapid convergence to the long-run equilibrium, with most deviations being corrected within a single period. This suggests that the EU gas system is highly adaptive due to market integration, storage capabilities, and institutional responsiveness. Such a high adjustment speed is rare in macro-energy models, indicating the maturity and flexibility of the EU’s gas infrastructure.

From a policy standpoint, this finding is encouraging; interventions that affect long-run demand (e.g., carbon pricing, renewable energy expansion) are likely to manifest relatively quickly in observable shifts in gas demand.

No structural breaks were identified via CUSUM and recursive residual tests, confirming the temporal robustness of the gas–GDP–fuel mix relationship throughout the observed period. This validates the use of ARDL modelling for long-term forecasting and scenario testing, particularly under frameworks such as the European Green Deal or Fit for 55.

Although natural gas consumption is influenced by factors such as prices, supply-demand imbalances, and geopolitical risks, this study intentionally abstracts from these direct market mechanisms. Instead, it focuses on the structural substitution patterns among fuels, which implicitly reflect such shocks over time. Including these variables would require a different modelling framework, such as structural VARs or high-frequency event studies, which fall outside the scope of this analysis.

6. Policy Implications

The empirical findings of this study provide actionable insights for policymakers navigating the complex energy transition landscape within the European Union. It is essential to clarify that the policy frameworks referred to in this study—REPowerEU [26] and the Fit for 55 package [12]—are officially adopted strategies of the European Union. REPowerEU was introduced by the European Commission to accelerate the transition away from Russian gas, increase renewable deployment, and improve energy security. Fit for 55 (2021) is a binding legislative package designed to reduce EU greenhouse gas emissions by 55% by 2030. Both frameworks include concrete measures such as renewable energy targets, taxation reforms, and energy efficiency obligations [12,26]. Our findings offer empirical insights into how energy demand behaviour aligns with these goals.

The identified dynamic interlinkages between gas consumption, economic activity, and fuel composition hold several important implications.

-

Coordinated Transition Planning. The positive and significant association between liquid fuel and gas consumption suggests a high degree of demand co-movement. This complementarity implies that policies aimed at reducing petroleum dependencies, such as vehicle electrification or decarbonisation of the transport sector, will also impact gas demand, either directly or through substitution pressures. The EU must develop integrated strategies across fuels, ensuring that reductions in one domain (e.g., oil) do not produce unintended shortages or volatility in another (e.g., gas). This aligns with Busu [4], who demonstrates that renewable energy investments can meaningfully support EU economic resilience, provided that fuel-switching effects are strategically managed [4].

-

Fuel Substitution and Coal Phase-Out. The observed delayed adverse effect of solid fuel use on gas consumption supports the interpretation of gas as a transitional fuel in Europe’s decarbonisation pathway. However, this substitution is neither immediate nor universal. Governments must strengthen policy incentives and infrastructure investment to accelerate substitution away from coal, particularly in lagging regions, while managing the medium-term dependence on gas with care.

-

Economic Sensitivity and Forecasting. The lagged impact of GDP on gas consumption highlights the importance of forward-looking economic indicators in energy demand forecasting. Ribeiro et al. [5] similarly argue for the inclusion of structural macro-drivers in forecasting frameworks, particularly for renewables, to anticipate non-linear and delayed demand responses. Moreover, the absence of significant short-run effects suggests that gas demand is relatively insulated from transient macroeconomic volatility. That is why policymakers must use macroeconomic leading indicators (e.g., industrial orders, investment trends) for proactive energy planning rather than relying solely on contemporaneous GDP signals.

-

Infrastructure and Market Responsiveness. The high magnitude of the error correction term (−0.77) implies that the EU gas system adjusts rapidly to equilibrium shocks. This responsiveness likely reflects the maturity of EU market liberalisation and physical infrastructure (e.g., LNG terminals, interconnectors, and storage). The policy recommendation in this case is to preserve and enhance the flexibility of the gas grid as the system evolves, ensuring interoperability with green gases (hydrogen, biogas) and demand-side management. It is important to note that the substitution processes identified in this study occur within a policy landscape that is often outpaced by market developments. As highlighted during recent crises, regulatory frameworks can lag behind actual needs—such as infrastructure permitting, gas storage mandates, or grid access for renewables—thus impeding efficient energy transitions.

Our results provide empirical support for the cross-fuel flexibility envisioned in the EU’s REPowerEU strategy, which emphasises reduced gas dependency through increased electrification and renewable deployment. Moreover, the observed responsiveness of gas demand to economic activity aligns with the goals of the Fit for 55 package, which aims to reduce fossil fuel use through targeted taxation and efficiency gains.

The absence of structural breaks across the modelled period confirms the robustness of the gas demand structure. This reinforces the value of ARDL-based models for medium- and long-term policy simulations under the European Green Deal or REPowerEU. The policy recommendation is to employ ARDL frameworks in national energy forecasting exercises, especially for evaluating scenarios involving fuel bans, carbon pricing, or demand shocks.

7. Limitations and Future Research

One limitation of this study is the use of EU-wide aggregated data, which masks cross-country heterogeneity in fuel consumption and substitution patterns. While this approach offers a harmonised macroeconomic perspective, it overlooks structural differences across Member States in energy mix, industrial base, and responsiveness to policy measures. Future research could extend this analysis to a panel-data framework or a country-level time series to explore such heterogeneity more explicitly.

Although natural gas consumption is influenced by prices, supply–demand imbalances, and geopolitical events, these drivers are not directly modelled in this study. Instead, the empirical framework focuses on long-term substitution dynamics among gas, liquid fuels, and solid fuels, which implicitly absorb the effect of structural shocks. Modelling explicit price or geopolitical shocks would require alternative approaches—such as structural VARs or regime-switching models—which lie beyond the scope of this study.

Furthermore, although dynamic ARDL simulation techniques were not employed here, they offer valuable tools for scenario building, policy stress testing, and forecasting. Future studies could incorporate such simulations to assess the impact of policy packages, such as REPowerEU, or external shocks, such as the Russian-Ukrainian war, in a more granular manner. This would also allow for the integration of asymmetric effects or nonlinear adjustment paths that may arise in real-world energy transitions.

Finally, expanding the range of explanatory variables to include renewable energy deployment, carbon pricing, or regulatory interventions could further enrich the model. Such extensions would improve the capacity to capture decarbonisation dynamics and enhance the policy relevance of empirical findings.

8. Conclusions

This paper has investigated the dynamic drivers of natural gas consumption in the European Union using an autoregressive distributed lag (ARDL) modelling framework. The findings demonstrate that gas demand is shaped not only by long-run macroeconomic conditions but also by short-run shifts within the energy mix, particularly interactions with the consumption of liquid and solid fuels.

GDP per capita exerts a statistically significant influence after a two-year lag, suggesting that gas demand responds gradually to changes in economic activity. The immediate and robust short-term relationship with liquid fuels suggests shared sectoral uses or complementary infrastructure. In contrast, the delayed adverse effect of solid fuels aligns with a slow substitution away from coal.

The existence of cointegration confirms a long-term equilibrium relationship, and the significance of the error correction term indicates rapid convergence toward that equilibrium. Structural diagnostics further support the temporal stability of the model parameters across the observed period (1990–2021).

These results underscore the need to view natural gas consumption not simply as a function of GDP, but as part of an evolving and interdependent energy system. The empirical evidence supports the relevance of gas as a transitional fuel in the EU’s decarbonisation pathway and provides a credible foundation for scenario modelling and policy planning.

This study makes three main contributions to the literature. First, it explicitly models gas consumption at the EU-wide macroeconomic level, addressing a gap in energy modelling, which often treats gas as a sector-specific or residual category. Second, the inclusion of liquid and solid fuels enables the identification of structural substitution dynamics critical to understanding fossil fuel phase-out strategies. Third, the ARDL approach offers a transparent and statistically robust method for estimating both long-run and short-run effects within relatively small samples.

Taken together, these contributions are highly relevant to the European Union’s integrated climate and energy strategy. They enhance the understanding of gas demand responsiveness and provide actionable insights for designing transition policies under frameworks such as REPowerEU and Fit for 55.