Motivating Green Transition: Analyzing Fuel Demands in Turkiye Amidst the Climate Crisis and Economic Impact

Abstract

Decarbonizing the transportation sector is critical for sustainable development, particularly in rapidly urbanizing countries like Turkiye. This study analyzes fuel demand elasticities for diesel, gasoline, and LPG across 12 NUTS-1 regions of Turkiye in 2022, using a panel random effects SUR approach. The model accounts for regional variation and fuel interactions, producing robust estimates that uncover significant spatial and temporal differences in consumption patterns. Uniquely, diesel demand displays a significantly positive price elasticity, challenging the conventional assumption of inelasticity. Gasoline demand is moderately price-sensitive, while LPG appears relatively unresponsive. Strong cross-price elasticities—especially between diesel and gasoline—point to substitution effects that can inform more adaptive policy frameworks. Seasonal fluctuations and Istanbul’s outsized impact also shape national trends. These findings underscore the need for differentiated region- and fuel-specific strategies. While higher gasoline taxes may effectively reduce demand, lowering diesel and LPG use will require complementary measures such as infrastructure upgrades, behavioral incentives, and accelerated adoption of alternative fuels. The study advocates for regionally adjusted carbon pricing, removal of implicit subsidies, and targeted support for electric and hybrid vehicles. Aligning fiscal tools with actual demand behavior can enhance both the efficiency and equity of the transition to a low-carbon transportation system.

1. Introduction

Air pollution continues to represent a pervasive challenge in European urban areas, a point that has been repeatedly emphasized by the World Health Organization (WHO). In 2014, exposure to fine particulate matter resulted in approximately 428,000 premature deaths across 41 European nations [1]. While more recent data from the EU-28 indicates a slight decline, the mortality toll remains substantial, with an estimated 399,000 fatalities [1]. Furthermore, the efforts to combat climate change have, at times, produced unintended consequences, exacerbating pollution levels. For instance, policies designed to promote the use of diesel vehicles have inadvertently resulted in increased air pollutants, while the introduction of renewable wood burning in specific regions has led to elevated local particulate matter concentrations. These cases highlight the critical need for rigorous evaluation and holistic policymaking [1]. In the case of Turkiye, the adoption of Decree-Law 4856 in 2003 signified a pivotal moment in environmental governance, leading to the establishment of the Ministry of Environment and Forestry following Turkiye’s accession to the European Environment Agency [2]. Legislation was passed that mandated the development of an environmental inventory and the publication of a comprehensive state of the environment report, thereby underscoring a commitment to transparency and accountability [2]. Furthermore, the establishment of the National Sustainable Development Commission (NSDC) signifies a deliberate endeavor to orchestrate inter-institutional actions towards the realization of the Sustainable Development Goals [2]. The role of Turkiye’s Statistical Institute (TSI) is also crucial in ensuring the alignment of national data collection with international standards, thus ensuring consistency in global sustainability indicators.

Nearly 68% of the world’s population is expected to live in urban areas by the middle of the century, according to United Nations (UN) estimates. Given the projected global size of urban growth, it is anticipated that urban centers will house an additional 2.5 billion people by 2050 [3]. Analogous trends are also evident in Turkiye, where most residents are concentrated in provincial and district centers. Currently, 22 out of the country’s 81 provinces have populations exceeding one million [4]. As of 2023, the total population of the country stood at 85,816,199, with approximately 77% of people residing in urban areas [5]. The rapid urban expansion that has taken place has intensified environmental challenges, contributing to health risks and ecological threats responsible for 40% of severe diseases and premature deaths [6,7]. The leading environmental cause of premature mortality in the world is air pollution. Global healthcare costs associated with air pollution are predicted by the Organization for Economic Co-operation and Development (OECD) to increase significantly from USD 21 billion in 2015 to USD 176 billion by 2060. Improving air quality has the potential to buck this trend and yield significant economic gains [8]. A significant contributing factor to the deterioration of urban air quality is large-scale infrastructure development, which disrupts natural land surfaces and diminishes the ability of green spaces to regulate energy balance [9]. The increasing demands for agriculture, mining, residential settlements, and transportation have precipitated a marked increase in carbon dioxide (CO₂) emissions [10]. Methane (CH₄) and nitrous oxide (N₂O) are the next most important greenhouse gases (GHG), accounting for almost 74% of global emissions, according to the OECD Climate Action Monitor [11]. Because of their large sources from human activities and their large heat-trapping capacity, these three gases are major contributors to climate change [12]. According to recent estimates by the International Energy Agency [13], the transport sector will be responsible for about 8 gigatonnes (Gt) of CO₂ emissions in 2022, or about 21% of all energy-related emissions worldwide. About two-thirds of all GHG emissions came from industrial activities and the burning of fossil fuels, which were the main drivers of this overall increase [14]. Similarly, the OECD [15] and Statista [16] highlight that despite increasing efforts towards electrification, the sector continues to rely on fossil fuels, contributing to a growing carbon footprint. These findings underscore the critical role of the transport sector in climate change mitigation strategies. To address the issue of urban air pollution, it is imperative that a concerted effort is made to reduce GHG emissions. Projections indicate that by 2050, the global economy is expected to expand fourfold, resulting in an estimated 80% surge in energy demand compared to current levels [17]. As the world’s 17th largest economy and Europe’s sixth largest [18], it is incumbent upon Turkiye to align its economic trajectory with sustainable environmental policies. While international agreements aim to curb atmospheric pollutants, these initiatives must be reinforced by localized measures, including public awareness campaigns, the removal of high-emission vehicles from urban centers, and strategic urban zoning policies [1].

The latest data show that fossil fuels continued to dominate the global energy mix in 2022, accounting for 82% of total consumption [19]. Despite record growth in renewable energy, energy-related CO₂ emissions reached a new peak, increasing by 0.9%. The transport sector was a major contributor to this trend, with global oil demand rising to 97.3 million barrels per day, driven mainly by diesel and gasoline consumption. While gasoline demand returned to pre-pandemic levels, the use of diesel and other oil products exceeded these levels, highlighting the sector’s continued reliance on fossil fuels in the midst of a sluggish post-COVID recovery. Oil prices increased by around 40%, with Brent crude averaging around $101 per barrel by the end of 2022, the highest level since 2013 [19]. A substantial proportion of vehicles continue to operate using internal combustion engines, thereby releasing considerable quantities of GHGs into the atmosphere [20,21]. The combustion of a liter of gasoline, diesel, and liquefied petroleum gas (LPG) results in emissions of 2.3, 2.7, and 1.6 kg of GHGs, respectively [22]. The substantial reliance on fossil fuels in the transportation sector has resulted in a marked increase in global GHG emissions [21], with the level of CO₂ projected to rise further, thereby posing a grave threat to both the global climate system and human well-being [10]. Turkiye’s carbon footprint has expanded dramatically over the past few decades. Specifically, the per capita CO₂-equivalent emissions in 1990 were recorded at 3.8 metric tons, which increased to 6.3 metric tons by 2016 [23]. By 2018, Turkiye had attained the 15th position among the world’s top 20 CO₂-emitting nations, with total emissions reaching approximately 6 million metric tons per capita [24,25]. Within the purview of the OECD, the country is positioned among the top 10 countries contributing to GHG emissions [26]. Between 1990 and 2019, the country’s total GHG emissions surged by an alarming 130%, reaching 506 million metric tons [27]. Concurrently, global emissions escalated to over 50 billion metric tons of CO₂-equivalent by 2022, reflecting a persistent upward trend despite temporary declines during the COVID-19 pandemic [14,28]. The transportation sector continues to represent a substantial proportion of Turkiye’s total GHG emissions, accounting for 23.2% of the country’s total emissions—a figure that exceeds the global average of 14% [29]. The rapid expansion of transportation infrastructure in the country has been identified as a key factor in the exacerbation of environmental pressures, particularly with regard to energy consumption and CO₂ emissions. Projections indicate that by 2050, energy demand and CO₂ emissions from the transportation sector could increase by a staggering 3.4 times [21]. This precipitous rise in emissions has given rise to urgent concerns within the government, with discussions now underway on sustainable policies and mitigation strategies [10,30,31,32]. A preceding study emphasized the transportation sector’s significant role in Turkiye’s GHG emissions and outlined the country’s total emissions and commitments under the Paris Agreement [33]. In the context of the country’s ambitious economic growth objectives, a comprehensive analysis of emission sources is imperative for the development of alternative policy measures [34]. As a signatory to the UN Framework Convention on Climate Change (UNFCCC) and the Kyoto Protocol, Turkiye is obligated to take decisive action against global warming [35]. As an Annex I party, the country is required to submit annual inventory reports on GHG emissions and removals, in accordance with the guidelines established by the Intergovernmental Panel on Climate Change (IPCC) [27]. An analysis of the country’s CO₂ emissions reveals a worrisome upward trend, with the country accounting for 69% of total GHG emissions in 1990, and this figure rising to 80.5% by 2018 [36]. While numerous studies have examined GHG emissions in the transportation sector across different countries, research focusing on Turkiye remains limited, particularly those utilizing transportation datasets [21,24,29]. As Şahin [24] emphasized and Güzel and Alp [29] elaborated, there is an urgent necessity for detailed regional studies on the transportation sector to facilitate the development of more effective policy interventions.

Despite the growing body of literature on GHG emissions and the transport sector, there remains a significant gap in empirical research on regional fuel demand in Turkiye, particularly using disaggregated data across provinces and fuel types. Most existing studies [37,38] focus on national aggregates and fail to adequately capture spatial and seasonal heterogeneity between fuel types consumed. This study addresses such a gap by estimating province- and month-specific demand elasticities for diesel, gasoline, and LPG within a robust panel data framework using a seemingly uncorrelated regression (SUR) demand system. At the core of this research lies the critical problem of the insufficient understanding of how key economic variables—especially prices and income (e.g., gross domestic product (GDP))—shape fuel consumption patterns in a country that is both highly urbanized and heavily reliant on energy imports. Without such insight, it is difficult to design effective fiscal and environmental policies aimed at reducing emissions. Therefore, understanding the price, cross-price, and GDP elasticities of transport fuels is essential for policymakers to design effective instruments, such as taxation, and for sectors where such knowledge has important operational implications [38,39]. As highlighted by Hasanov [38], accurate forecasting and management of energy demand, especially for transport fuels, is a central policy concern for energy-importing countries such as Turkiye. This is further driven by growing public concerns about energy security and GHG emissions related to climate change [38,40,41]. For example, if fuel demand is found to be inelastic, taxation may have limited effectiveness in curbing both consumption and emissions. To this end, the study employs the SUR method within a random effects panel framework, which allows for the simultaneous estimation of multiple demand equations (e.g., diesel, gasoline, and LPG) while accounting for correlations between fuels and unobserved heterogeneity across cities in the country. Using data from Turkiye’s 81 provinces over a 12-month period in 2022, the study provides detailed elasticity estimates that are crucial for informing targeted climate policies. To our knowledge, this is the first study in Turkiye to integrate province-level disaggregation with the panel data SUR approach, providing novel empirical insights not only for national policy but also for countries with similar economic and environmental profiles.

2. Data and Methods

2.1. Data

This study uses a comprehensive 2022 panel dataset covering all 81 provinces of Turkiye, combining spatial (across provinces) and temporal (monthly) dimensions. Monthly data on diesel, gasoline, and LPG consumption (in metric tons), real GDP per capita, registered motor vehicles, and the number of neighborhoods and geographical areas of the provinces were obtained from the Turkish Statistical Institute [42]. Monthly fuel pump prices at the provincial level (in Turkish Lira [TL] per liter) were obtained from Turkish Petroleum [43], with nominal prices converted to real values using the Consumer Price Index (CPI) to ensure comparability over time.

The dependent variables—logarithmically transformed monthly consumption of diesel, gasoline, and LPG—are analyzed alongside key explanatory variables, including real fuel prices (TL per liter) and real GDP per capita (TL), both also expressed in logarithmic form. This transformation allows direct interpretation of the elasticity coefficients, accounts for potential non-linearity, and also improves model stability by reducing variance. Structural controls include the logarithmic values of both the number of residential settlements (e.g., villages, neighborhoods) and the geographical area of each province, while vehicle density (registered vehicles per capita) is included in its original form. These controls capture essential demographic and spatial heterogeneity. To improve spatial interpretability and to account for regional variation, the study uses the Nomenclature of Territorial Units for Statistics (NUTS), which is aligned with the European Union (EU) regional classification standards. As part of the EU harmonization process, the country has adopted a three-level NUTS structure based on population size, topographical characteristics, socio-economic development, and regional planning priorities. Accordingly, the country is divided into 12 regions at NUTS level 1, 26 at level 2, and 81 provinces at level 3 [44]. Regional dummy variables based on the 12 NUTS-1 (e.g., level 1) regions and seasonal dummies derived from monthly data are also included in the three-equation demand system to control for spatio-temporal variation. Estimation is carried out using SUR, which allows for a robust and integrated analysis of fuel demand elasticities across different fuel types. This methodological framework provides a detailed analysis of regional energy consumption dynamics in Turkiye. All estimations were performed using the R statistical software environment and the plm package for panel data modeling and SUR estimation. Confidence intervals for the estimated parameters in each equation were calculated in R, and custom scripts were developed to manage and process the monthly provincial dataset. Visualizations were generated using R’s base plotting system and the ggplot2 package. Together, these tools ensured robust model estimation and facilitated the reproducible creation of graphical results. To provide a clear and engaging context for the forthcoming model estimation,

Table 1. Variable definitions and sample descriptive statistics of variables.

Variables	Definition	Mean	Std. Dev.
Diesel quantity	Monthly diesel consumption in the province (metric tons)	25,274.70	40,862.69
Gasoline quantity	Monthly gasoline consumption in the province (metric tons)	3391.10	8014.215
LPG quantity	Monthly LPG consumption in the province (metric tons)	3177.17	4314.52
Real diesel price	Diesel pump liter price (Turkish Lira (TL)/per liter)	27.84	4.12
Real gasoline price	Gasoline pump liter price (TL/per liter)	25.40	4.85
Real LPG price	LPG pump liter price (TL/per liter)	13.52	2.25
Gross domestic product	Per capita annual gross domestic product in TL in each province	64,915.61	24,819.05
Vehicle numbers	Total number of vehicles registered on the road per month	319,276.08	620,751.39
Village numbers	Total number of villages/neighbors in each province	429.57	246.01
Area	Total area of the city in square kilometers	9598.21	6275.71
Population	Total population	1,053,979.96	1,890,874.44
Regions:
Istanbul	1 if Istanbul sub-region, 0 otherwise	0.012	0.110
Western Marmara	1 for provinces in the Western Marmara sub-region (including Tekirdağ, Edirne, Kırklareli, Balıkesir, and Çanakkale); 0 otherwise	0.062	0.241
Aegean	1 for provinces in the Aegean sub-region (Izmir, Aydın, Denizli, Muğla, Manisa, Afyonkarahisar, Kütahya, and Uşak); 0 otherwise.	0.099	0.299
Eastern Marmara	1 if the province belongs to the Eastern Marmara sub-region (Bursa, Eskişehir, Bilecik, Kocaeli, Sakarya, Düzce, Bolu, and Yalova); 0 otherwise	0.099	0.299
Western Anatolia	1 if the provinces are in the Western Anatolia sub-region (Ankara, Konya, and Karaman); 0 otherwise.	0.037	0.189
Mediterranean	1 for provinces in the Mediterranean sub-region (Antalya, Isparta, Burdur, Adana, Mersin, Hatay, Kahramanmaraş, and Osmaniye); 0 otherwise.	0.099	0.299
Central Anatolia	1 for provinces in the Central Anatolia sub-region (Kırıkkale, Aksaray, Niğde, Nevşehir, Kırşehir, Kayseri, Sivas, and Yozgat); 0 otherwise	0.099	0.299
Western Black Sea	1 if located in the Western Black Sea sub-region (Zonguldak, Karabük, Bartın, Kastamonu, Çankırı, Sinop, Samsun, Tokat, Çorum, and Amasya); 0 otherwise.	0.123	0.329
Eastern Black Sea	1 for provinces in the Eastern Black Sea sub-region (Trabzon, Ordu, Giresun, Rize, Artvin, and Gümüşhane); 0 otherwise	0.074	0.262
Northeastern Anatolia	1 for provinces in the Northeastern Anatolia sub-region (Erzurum, Erzincan, Bayburt, Ağrı, Kars, Iğdır, and Ardahan); 0 otherwise	0.086	0.281
Middle Eastern Anatolia	1 for provinces in the Middle Eastern Anatolia sub-region (Malatya, Elazığ, Bingöl, Tunceli, Van, Muş, Bitlis, and Hakkari); 0 otherwise	0.099	0.299
Southeastern Anatolia	1 for provinces located in the Southeastern Anatolia sub-region (Gaziantep, Adıyaman, Kilis, Şanlıurfa, Diyarbakır, Mardin, Batman, Şırnak, and Siirt); 0 otherwise	0.111	0.314

presents a comprehensive descriptive analysis of the dataset, which is further elaborated in the results section.

2.2. Methods

SUR provides a powerful framework for estimating systems of multiple equations (e.g., diesel, gasoline, and LPG), especially when their error terms are potentially correlated. Unlike separate equation-by-equation estimation, the SUR approach exploits cross-equation correlations to produce more efficient and consistent parameter estimates. By jointly modeling the equations and explicitly accounting for interdependencies in the error structure, the SUR model improves both the interpretability and the predictive accuracy of the system. This methodological advantage makes SUR particularly valuable in contexts where several related outcomes—such as fuel consumption types—are simultaneously influenced by a common set of explanatory variables. In this study, we use the SUR approach to jointly estimate the demand equations for diesel, gasoline, and LPG, treating each as a dependent variable within a unified analytical framework. Readers seeking a deeper understanding are strongly recommended to consult specialist textbooks on panel data analysis, particularly Croissant and Millo (2019) [45], which offers a comprehensive chapter on seemingly unrelated regressions alongside R implementation code. These resources provide thorough guidance on SUR estimation techniques and error component considerations, making them invaluable for gaining both theoretical insight and practical experience. Fuel demand equations for each fuel type (e.g., diesel, gasoline, and LPG) are compactly expressed using the random effects specification, chosen for its analytical tractability and suitability to the scope of this research:

$$Ln(Q_{it}^m)={\delta }^m+{\displaystyle \sum _{m=1}^3{\beta }^m\ln (P_{it}^m)}+{\displaystyle \sum _{j=1}^4{\lambda }_j^m}{x}_{jt}+{\displaystyle \sum _{k=1}^{10}{\gamma }_k^m}{R}_k+{\displaystyle \sum _{j=1}^3{\delta }_j^m}{S}_j+{\alpha }_i^m+{\epsilon }_{it}^m$$

(1)

where the dependent variable, ln($Q_{it}^m$) represents the natural logarithms of fuel consumption by type (m = diesel, gasoline, and LPG) at the provincial level i in month t. Each demand equation includes the natural logarithm of real pump prices, ln($P_{it}^m$) (in TL per liter) for the three fuel types (m) in each province i in month t, enabling elasticity estimation and addressing potential non-linear relationships. Additional covariates, x_jt, comprise the logarithm of the number of villages, per capita GDP, and land area in each province, as well as the ratio of registered vehicles to population—capturing structural and economic heterogeneity. The model also incorporates observed regional (R_k) and seasonal effects (S_j) through dummy variables. Specifically, it includes ten NUTS-1 regional indicators, using Western Marmara—including Istanbul—as the reference category. Initially, the regional classification comprised 12 regions, allowing for comparisons across 11 regions relative to the reference group. However, Istanbul was originally designated as a separate region due to its unique characteristics. Given its exceptionally high and anomalous consumption levels of diesel, gasoline, and LPG, Istanbul posed a risk of skewing the analysis. To mitigate this disproportionate influence, Istanbul was merged with the Western Marmara region, reducing the number of regions from 12 to 11. Additionally, three seasonal dummy variables were included, with one season omitted to ensure model identification. By estimating this system using the SUR method, the model exploits potential correlations across equations, thereby enhancing efficiency and yielding more robust insights into province-level fuel demand behavior in Turkiye. Finally, ${\alpha }_i^m$ captures the province-specific random effects for each fuel type, while ${\epsilon }_{it}^m$ denotes the idiosyncratic error terms for the diesel, gasoline, and LPG equations across provinces and months.

As it takes more than 12 months to estimate the effects of time and individual (province) factors simultaneously, our analysis is limited to a one-way random effects model that focuses solely on variation at the provincial level. This model is implemented within the SUR framework, a well-established method of panel data analysis that is particularly suited to addressing unobserved heterogeneity across units (such as provinces) over time. In this context, we apply the joint model to three dependent variables—diesel, gasoline, and LPG consumption—each specified at the provincial level over twelve months. The mathematical formulation and estimation procedure for this model have been extensively documented in the literature (e.g., Croissant and Millo [45]). In the one-way random effects specification, each cross-sectional unit (i.e., province) is assumed to have an individual-specific random effect (α_i) that captures time-invariant, unobserved factors affecting fuel demand. The model for each dependent variable (e.g., $Ln{Q}^{diesel}$, $Ln{Q}^{gasoline}$, and $Ln{Q}^{LPG}$) is represented as follows:

$$\begin{array}{l}Ln{Q}_{it}^{diesel}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}=X_{it}{\beta }^{diesel}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}+{\alpha }_i^{diesel}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}+{\epsilon }_{it}^{diesel}\\ Ln{Q}_{it}^{gasoline}=X_{it}{\beta }^{gasoline}+{\alpha }_i^{gasoline}+{\epsilon }_{it}^{gasoline}\\ Ln{Q}_{it}^{LPG}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}=X_{it}{\beta }^{LPG}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}+{\alpha }_i^{LPG}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}+{\epsilon }_{it}^{LPG}\end{array}$$

(2)

where X and β denote the set of explanatory variables (including a constant term) and their corresponding coefficients in Equation (1), respectively. When we stack the equations for all three fuel types, we obtain a system of seemingly unrelated regressions, allowing us to jointly estimate the models while accounting for potential correlations among their error terms:

$$Y=X\beta +Z\alpha +\epsilon$$

(3)

where $Y=(Ln{Q}^{diesel},Ln{Q}^{gasoline},Ln{Q}^{LPG}{)}^{\prime }$ is the vector of dependent variables (log-transformed fuel demands), $\mathrm{X}=I_3\otimes X$ and $\beta ={\left({\beta }^{diesel},{\beta }^{gasoline},{\beta }^{LPG}\right)}^{\prime }$ represent the block-diagonal design matrix and corresponding slope coefficients, $Z={\left(Z_1,Z_2,Z_3\right)}^{\prime }=I_3\otimes Z$ and $\alpha ={\left({\alpha }^{diesel},{\alpha }^{gasoline},{\alpha }^{LPG}\right)}^{\prime }$ capture province-specific random effects, and $\epsilon ={\left({\epsilon }^{diesel},{\epsilon }^{gasoline},{\epsilon }^{LPG}\right)}^{\prime }$ denotes the error vector. Each column of Z corresponds to province-level intercepts via a unit vector 1_n, representing province-specific effects. The system can be rewritten more compactly using Kronecker product notation, (⊗), as:

$$Y=\left(I_3\otimes X\right)\beta +\left(I_3\otimes Z\right)\alpha +\epsilon =W\theta +\epsilon$$

(4)

where $W=\left[I_3\otimes X,I_3\otimes Z\right]$, $\theta ={\left(\beta ,\hspace{0.17em}\hspace{0.17em}\alpha \right)}^{\prime }$, and $\epsilon \sim N\left(0,\Omega \right)$ with $\Omega ={\sigma }_{\alpha }^{2}Z{Z}^{\prime }+{\sigma }_u^{2}I$, capturing both cross-equation error correlations and idiosyncratic shocks.

A key strength of the SUR approach is its ability to account for contemporaneous correlation between error terms, which greatly improves the efficiency of parameter estimation. Estimation methods typically include the General Method of Moments (GMM), Generalized Least Squares (GLS) (we use GLS by estimating the $\widehat{\theta }={\left(W^{\prime }{\Omega }^{-1}W\right)}^{-1}{W}^{\prime }{\Omega }^{-1}Y$ and then decompose β and α sets accordingly), or Maximum Likelihood Estimation (MLE), all of which are designed to account for these cross-equation dependencies [45].

This study applied three diagnostic tests to evaluate the reliability of the panel data estimations and the accuracy of the model assumptions: the Hausman specification test (H), the slope homogeneity test (SH) and the cross-sectional dependence test (CD). Although unit root tests are standard diagnostic tools in panel data analysis, they are unreliable over short time horizons. With only 12 monthly observations per unit (e.g., province), these tests lack the statistical power required to support robust conclusions. Therefore, any claims about stationarity are inherently tentative in this context. These tests assess estimator consistency, parameter stability, and spatial interdependence, respectively. First, the Hausman specification test (Hausman [46]) was used to determine whether the random effects (RE) estimator is consistent and efficient or whether fixed effects (FE) should be preferred due to the correlation between unobserved individual effects and regressors. The test statistics are given by the following:

$$H={\left({\widehat{\beta }}_{RE}-{\widehat{\beta }}_{RE}\right)}^{\prime }{\left[Var({\widehat{\beta }}_{RE})-Var({\widehat{\beta }}_{FE})\right]}^{-1}\left({\widehat{\beta }}_{RE}-{\widehat{\beta }}_{RE}\right)$$

(5)

where ${\widehat{\beta }}_{RE}$ and ${\widehat{\beta }}_{FE}$ are the estimated coefficients under random-effects (RE) and fixed-effects (FE), respectively, and Var(⋅) is the corresponding covariance matrix. Under the null hypothesis that there is no correlation between the unobserved individual effects and the regressors, the statistic (H) follows a chi-squared distribution (${\chi }_k^2$) with degrees of freedom equal to the number of regressors, k. Rejection of the null hypothesis in the Hausman specification test signals that the random effects (RE) estimator is inconsistent and favors the fixed effects (FE) model instead. For the test to yield valid results, however, it is crucial that both models are estimated using an identical set of explanatory variables. However, this poses a challenge: Regional dummy variables are time-invariant and are therefore automatically dropped from the FE model but are retained in the RE model, resulting in an asymmetry. To preserve comparability and adhere to the test’s assumptions, these dummies are excluded from the RE specification solely for the purpose of the Hausman test. This ensures that the test operates on a shared parameter space, enabling robust inference on the suitability of the RE estimator. We also tested the assumption of slope homogeneity using the test proposed by Pesaran and Yamagata [47]. The test statistics are given by the following:

$$\Delta =\sqrt{N}\left({\displaystyle \frac{\widehat{S}-k}{\sqrt{2k}}}\right),\hspace{0.17em}\hspace{0.17em}{\Delta }_{adj}=\sqrt{N}\left({\displaystyle \frac{N^{-1}\widehat{S}-{\rm E}(\widehat{S})}{\sqrt{Var(\widehat{S})}}}\right)$$

(6)

where N is the number of cross-sectional units and $\widehat{S}={\displaystyle {\sum }_{i=1}^N{\left({\widehat{\beta }}_i-\overline{\beta }\right)}^{\prime }{\widehat{V}}^{-1}}\left({\widehat{\beta }}_i-\overline{\beta }\right)$ is the sum of the squared deviations of individual slope estimates from their cross-sectional weighted mean, $\overline{\beta }={\left({\displaystyle \sum _{i=1}^N{\widehat{V}}_i^{-1}}\right)}^{-1}{\displaystyle \sum _{i=1}^N{\widehat{V}}_i^{-1}}{\widehat{\beta }}_i$, weighted by the variance-covariance of ${\widehat{\beta }}_i$ where ${\widehat{V}}_i={\widehat{\sigma }}_i^2{\left({X^{\prime }}_i{X}_i\right)}^{-1}$ (e.g., ${\widehat{\sigma }}_i^2$ is the variance estimate of the cross-section unit i, and X is the regressor matrix, while k is the number of regressors. $\widehat{S}$ is also known as Swamy’s test statistic and E($\widehat{S}$) is the expected value of Swamy’s test statistic with its corresponding estimated variance, $Var(\widehat{S})$. Under the null hypothesis of homogeneity of slopes, both Δ and Δ_adj are asymptotically distributed as standard normal (Δ, Δ_adj ∼ N(0,1)) [47]. Significant values indicate heterogeneity in the slope coefficients across provinces, indicating rejection of the pooled estimators. Finally, the presence of cross-sectional dependence (CD) was assessed using Pearson’s [48] CD test:

$$CD=\sqrt{{\displaystyle \frac{2T}{N(N-1)}}}{\displaystyle \sum _{i=1}^{N-1}{\displaystyle \sum _{j=i+1}^N{\widehat{\rho }}_{ij}}}$$

(7)

where ${\widehat{\rho }}_{ij}$ is the pairwise correlation coefficient of the residuals between cross-sectional units i and j, N is the number of cross-sectional units, and T is the number of time periods. Under the null hypothesis of cross-sectional independence, the CD statistic follows a standard normal distribution (CD ∼ N(0,1)). Significant test results imply the presence of cross-sectional dependence, which may require robust inference techniques to avoid biased standard errors and spurious significance levels [48].

3. Results

3.1. Preliminary Results

In 2022, Turkiye witnessed significant variations in the consumption of diesel, gasoline, and LPG (Table 1). Diesel consumption averaged around 25,000 metric tons per month, gasoline at 3400 metric tons, and LPG at 3200 metric tons. The standard deviations for diesel and gasoline were nearly double their averages, highlighting considerable variations. Interestingly, the price of diesel surpassed that of gasoline on a monthly basis, despite gasoline typically being the pricier option. LPG, on the other hand, was far more economical, costing nearly half as much as the other two fuels. Meanwhile, the number of total vehicles on the road per province per month hovered around 320,000. In 2022, the annual per capita gross domestic product (GDP) across all provinces (e.g., cities) amounted to approximately 65,000 TL, equivalent to around $4642 based on the exchange rate at that time. Table 1 also presents the statistical data for the other variables. The analysis also uses data from the 12 NUTS-1 sub-regions in Turkiye, alongside monthly dummy variables which account for temporal variation.

The regional distribution of consumption levels and pump prices for diesel, gasoline, and LPG across the country’s twelve sub-regions offers helpful advice for designing fiscally sound and environmentally effective policies—especially those aimed at promoting a low-carbon, greener future (

Table 2. Descriptive statistics of consumed quantities and prices of liquid fuel types by twelve sub-regions.

Regions	Quantity Consumed (Tons/Vehicles)			Prices (TL/L)
	Diesel Consumption	Gasoline Consumption	LPG Consumption	Diesel	Gasoline	LPG
	Mean (Std. Dev.)	Mean (Std. Dev.)	Mean (Std. Dev.)	Mean (Std. Dev.)	Mean (Std. Dev.)	Mean (Std. Dev.)
Istanbul	0.06284 (0.00480)	0.01358 (0.00144)	0.00421 (0.00040)	27.51292 (4.31761)	25.08066 (5.08108)	13.52003 (2.34507)
Western Marmara	0.13494 (0.09714)	0.01315 (0.00369)	0.00780 (0.00161)	27.69909 (4.14577)	25.26699 (4.88394)	13.46480 (2.26682)
Aegean	0.07686 (0.02110)	0.00890 (0.00319)	0.01118 (0.00417)	27.76580 (4.14393)	25.33262 (4.87969)	13.27715 (2.25104)
Eastern Marmara	0.10752 (0.03244)	0.01367 (0.00398)	0.01244 (0.00456)	27.62799 (4.13845)	25.13804 (4.80494)	13.22413 (2.24798)
Western Anatolia	0.07063 (0.01958)	0.00756 (0.00251)	0.01218 (0.00238)	27.80359 (4.18794)	25.37016 (4.92606)	13.51747 (2.25053)
Mediterranean	0.06215 (0.01445)	0.00762 (0.00182)	0.01185 (0.00476)	27.84924 (4.14644)	25.41907 (4.87621)	13.42201 (2.23920)
Central Anatolia	0.10883 (0.06019)	0.00830 (0.00216)	0.01689 (0.00673)	27.78909 (4.14489)	25.34911 (4.88185)	13.54063 (2.25671)
Western Black Sea	0.08312 (0.03216)	0.00836 (0.00243)	0.01380 (0.00363)	27.76830 (4.13028)	25.32191 (4.86880)	13.55280 (2.27175)
Eastern Black Sea	0.08530 (0.01998)	0.01042 (0.00331)	0.00967 (0.00410)	27.83641 (4.15083)	25.40253 (4.89399)	13.77044 (2.27196)
Northeastern Anatolia	0.09815 (0.04541)	0.01042 (0.00306)	0.01145 (0.00689)	27.95189 (4.14148)	25.51205 (4.88882)	13.83540 (2.26299)
Middle Eastern Anatolia	0.13390 (0.06752)	0.01294 (0.00487)	0.01546 (0.00559)	28.10659 (4.11240)	25.65394 (4.85382)	13.63614 (2.24527)
Southeastern Anatolia	0.15462 (0.08809)	0.01072 (0.00342)	0.01434 (0.00631)	28.01193 (4.12124)	25.57491 (4.86152)	13.53791 (2.24340)

and Figure 1). In response to a suggestion from one of our reviewers, for which we are grateful, we also calculated the fuel consumption of each vehicle type, providing a more detailed analysis. This table and figure analyze the average consumption levels, price structures, and distribution of each fuel type. The results highlight regional differences and assess statistical variability to inform more targeted energy policy evaluations. Diesel is the most widely consumed fuel type across all regions. The regions with the highest usage are Southeastern Anatolia (0.1546 metric tons per vehicle), Middle Eastern Anatolia (0.1339 metric tons per vehicle), and Western Marmara (0.1349 metric tons per vehicle). This pattern likely reflects the prevalence of commercial vehicles and agricultural activities, as well as a strong reliance on diesel-powered transport in urban and rural areas within these regions. In contrast, Istanbul (0.0628 metric tons per vehicle) and the Mediterranean region (0.0622 metric tons per vehicle) report the lowest levels of diesel consumption. This can be attributed to widespread public transport availability, shorter travel distances, and greater environmental awareness influencing vehicle choice in metropolitan areas. However, the highest gasoline consumption figures are found in Eastern Marmara (0.0137 metric tons/vehicle), Istanbul (0.0136 metric tons/vehicle), and Western Marmara (0.0132 metric tons/vehicle). By contrast, the lowest consumption is found in Western Anatolia and the Mediterranean regions at 0.0076 tons/vehicle. This distribution suggests a greater preference for gasoline-powered vehicles in more industrialized and urbanized areas, reflecting the higher prevalence of private car ownership and usage in metropolitan regions. Meanwhile, rural or less densely populated regions seem to prefer alternative, less expensive fuels, such as diesel or LPG. Finally, LPG consumption reveals a markedly different regional dynamic. The highest levels are observed in Central Anatolia (0.0169 metric tons per vehicle), Middle Eastern Anatolia (0.0155 metric tons per vehicle), and Southeastern Anatolia (0.0143 metric tons per vehicle). This highlights the widespread adoption of LPG as a viable, economical option in the country’s interior regions. This tendency likely reflects lower income levels and greater sensitivity to fuel costs in these regions. Notably, Istanbul reports the lowest LPG consumption (0.0042 metric tons per vehicle), potentially due to limited conversion infrastructure and heightened safety concerns in such a densely populated urban environment.

Beyond average consumption, the standard deviations reveal interesting insights into regional variability (see Figure 1). The highest volatility in diesel use is observed in Western Marmara (standard deviation: σ = 0.0971), where many data points reach the upper limit. The next highest volatility is found in Southeastern Anatolia (σ = 0.0881), followed by Middle Eastern Anatolia (σ = 0.0675). Gasoline consumption fluctuates most significantly in the Middle Eastern Anatolia region (σ = 0.0049), followed by Eastern Marmara (σ = 0.0040). LPG consumption fluctuates most in the northeastern Anatolia region (σ = 0.0069), followed by the central Anatolia region (σ = 0.0067). This variability may be driven by seasonal usage patterns, the composition of the vehicle fleet, or inconsistent user behavior over time.

Fuel pricing data highlights significant regional disparities. In Middle Eastern Anatolia, diesel and gasoline prices peak at 28.11 TL/L and 25.65 TL/L, respectively, while Southeastern Anatolia sees prices of 28.01 TL/L for diesel and 25.57 TL/L for gasoline (refer to Figure 2). These high prices can be linked to logistical issues, including long distances from refineries, less competitive market conditions, and increased distribution expenses. The challenging mountainous terrain and severe, prolonged winters in these areas further complicate fuel transportation and storage, raising overall supply chain costs. Conversely, Istanbul consistently has the lowest fuel prices for all three types, thanks to its efficient distribution network, strong competition, and strategic location near supply centers. For LPG, Northeastern Anatolia has the highest average price at 13.84 TL/L, while Eastern Marmara has the lowest at 13.52 TL/L. The Marmara region, particularly Istanbul, enjoys the lowest fuel prices in the country (see Figure 2). This pricing advantage stems from its closeness to major refineries and distribution centers, a well-developed transportation network, and a highly competitive fuel market driven by strong demand and market size. These structural and logistical improvements reduce transportation and distribution costs, thereby lowering prices for passenger vehicle users and companies involved in logistics, including passenger and freight transport. Notably, Istanbul also experiences the highest price volatility across all fuel types, especially for gasoline (σ = 5.08 TL/L), indicating a dynamic pricing landscape shaped by fierce market competition, regulatory changes, and fluctuating storage or supply conditions typical of large urban areas.

On the other hand, the monthly consumption patterns and pump price data of diesel, gasoline, and LPG in Turkiye (

Table 3. Descriptive statistics of consumed quantities and prices of liquid fuel types by months.

Variables/Months	Quantity Consumed (Metric Tons/Vehicles)			Prices (TL/L)
	Diesel Consumption	Gasoline Consumption	LPG consumption	Diesel	Gasoline	LPG
	Mean (Std. Dev.)	Mean (Std. Dev.)	Mean (Std. Dev.)	Mean (Std. Dev.)	Mean (Std. Dev.)	Mean (Std. Dev.)
January	0.08435 (0.05985)	0.00839 (0.00280)	0.00882 (0.00322)	21.80764 (0.15707)	21.45138 (0.15941)	14.09576 (0.19197)
February	0.07825 (0.04317)	0.00856 (0.00313)	0.00966 (0.00528)	25.99200 (0.16433)	25.63824 (0.16783)	14.75147 (0.19431)
March	0.08486 (0.05318)	0.00733 (0.00247)	0.01005 (0.00392)	32.15994 (0.16580)	29.98689 (0.17402)	16.82953 (0.20598)
April	0.09440 (0.05057)	0.00980 (0.00276)	0.01094 (0.00410)	30.52167 (0.15785)	28.18682 (0.15912)	16.32753 (0.19649)
May	0.10313 (0.05417)	0.01015 (0.00329)	0.01358 (0.00539)	33.54428 (0.15716)	34.27952 (0.50521)	16.01195 (0.21250)
June	0.09780 (0.05685)	0.00833 (0.00272)	0.01238 (0.00457)	32.02989 (0.14635)	32.72875 (0.13937)	14.69852 (0.19893)
July	0.11366 (0.05663)	0.01342 (0.00406)	0.01715 (0.00688)	29.43973 (0.15820)	26.73922 (0.15346)	13.39646 (0.24053)
August	0.12391 (0.06443)	0.01338 (0.00428)	0.01557 (0.00596)	31.99610 (0.15815)	24.03686 (0.17409)	13.27029 (0.21123)
September	0.11443 (0.06448)	0.01153 (0.00360)	0.01430 (0.00543)	23.89369 (0.17019)	19.90233 (0.15460)	11.08422 (0.18276)
October	0.11573 (0.06586)	0.01145 (0.00372)	0.01400 (0.00551)	27.13726 (0.17146)	21.30280 (0.15332)	10.33676 (0.20264)
November	0.10436 (0.06126)	0.00914 (0.00325)	0.01313 (0.00530)	23.06214 (0.17771)	20.66761 (0.15566)	10.75490 (0.19772)
December	0.11256 (0.06582)	0.01133 (0.00371)	0.01263 (0.00480)	22.48988 (0.17636)	19.82938 (0.16363)	10.67025 (0.19616)

and Figure 3) provide a valuable framework for understanding not only seasonal demand fluctuations but also the broader economic, social, and logistical dynamics that influence energy use throughout the year. In this context, analyses based on monthly average consumption per vehicle and price levels for each fuel type highlight not only cyclical variations but also the necessity of strategic energy management. According to the data (Table 3 and Figure 3), diesel consumption peaks in August (0.1239 metric tons/vehicle), followed by October (0.1157 metric tons/vehicle) and September (0.1144 metric tons/vehicle). This surge is closely associated with the intensification of agricultural activities during the summer months, along with increased long-distance travel for both individual (e.g., mostly for vacation) and commercial purposes. Conversely, the lowest diesel consumption is recorded in February (0.0783 metric tons/vehicle), likely due to limited transportation activities during harsh winter conditions and a heightened focus on energy efficiency. Gasoline consumption also exhibits a noticeable rise during the summer period (June–August), reaching its annual peak in August at 0.0134 metric tons per vehicle. This pattern is primarily driven by the increase in leisure travel and the prevalence of private vehicle use during the holiday season. The lowest consumption levels occur in March (0.0073 metric tons/vehicle) and January (0.0084 metric tons/vehicle), reflecting both climatic constraints and the influence of economic factors on fuel use behavior. LPG consumption, meanwhile, follows a somewhat different trend. The highest usage is observed in July–September (0.0172 metric tons/vehicle, 0.0156 metric tons/vehicle, and 0.0143 metric tons/vehicle, respectively), likely due to increased long-distance travel combined with a shift toward more economical fuel alternatives during vacation season. The lowest LPG consumption is recorded in January (0.0088 metric tons/vehicle), which may be attributed to limited technical accessibility caused by cold weather and a decline in user comfort associated with LPG-powered vehicles. To also deepen the analysis, December was designated as the reference month to assess whether monthly mean fuel consumption significantly deviated from that of December. Pairwise confidence intervals for all three fuel types are presented in Figure 4. In line with the suggestion of the reviewer whose input we gratefully acknowledge, our comparative analyses of fuel consumption are conducted on a per-vehicle basis. This allows for a more accurate assessment of regional consumption patterns. The analysis highlights significant seasonal trends in fuel consumption across the country’s provinces. For diesel, the first quarter of the year (January, February, and March) shows significant statistical differences in per-vehicle consumption compared to December, indicating a clear winter-related change. Gasoline consumption, on the other hand, shows significant variations in almost every month except May, September, and October, suggesting a more erratic consumption pattern. LPG follows a seasonal trend similar to diesel, reflecting comparable behavioral and climatic factors for different fuel types in the country. These variations, clearly shown in the figure, not only indicate a strong seasonal trend but also suggest deeper structural patterns in regional mobility and energy use. The consistency of these monthly changes across provinces supports the use of a random effects modeling approach. This method captures the inherent differences between provinces and accounts for region-specific baselines, providing a robust statistical framework for identifying both broad and localized trends. This is particularly important in energy policy research, where ignoring spatially concentrated variation could lead to inaccurate generalizations. These findings also have important implications for carbon mitigation strategies. Understanding the temporal and regional aspects of fuel consumption allows for more targeted decarbonization efforts. By strategically targeting policies to peak demand periods and high consumption areas, we can greatly improve the effectiveness and equity of climate policies and move toward a data-driven, regionally sensitive, and seasonally adaptive plan for a low-carbon transportation future in the country.

On the other hand, monthly variations in consumption across all three fuel types—diesel, gasoline, and LPG—are shaped by seasonal effects, economic preferences, and driving habits. Diesel consumption demonstrates the greatest variability in August–October and December (σ = 0.0659, σ = 0.0645, σ = 0.0644, and σ = 0.0658, respectively), underscoring the influence of peak agricultural activity and heightened transportation demand. In contrast, the lowest variability is seen in February (σ = 0.0432), a period marked by subdued mobility due to severe winter conditions and economic stagnation in Turkiye. For gasoline, the highest standard deviation occurs in August (σ = 0.0043), possibly reflecting behavioral diversification following the middle of the vacation season and regional differences in vehicle preferences. The lowest variability is found in March (σ = 0.0025), a month when the onset of spring travel renders consumption patterns more predictable in the country. LPG consumption, on the other hand, exhibits its highest fluctuation in June (σ = 0.0069), suggesting a greater susceptibility to seasonal demand shifts, particularly in inland regions where LPG is often chosen for its cost-effectiveness. The lowest variability for LPG is noted in January (σ = 0.0032), indicating relative stability in consumer behavior at the start of the year.

Looking at fuel prices (Table 3 and Figure 5), diesel reached its peak monthly average in May at 33.54 TL/L, followed by March at 32.16 TL/L and June at 32.03 TL/L. These increases during spring are likely influenced by refinery production cycles, rising global crude oil prices, or changes in tax policies. The lowest diesel prices were seen in January at 21.81 TL/L and December at 22.49 TL/L. December stands out as the lowest price point for all fuel types, likely due to year-end promotions, reduced demand, and temporary inflation control measures. Gasoline prices exhibited a similar pattern, peaking in May at 34.28 TL/L, reflecting a seasonal demand surge before summer. The lowest gasoline prices were recorded in September and December, around 19.90 TL/L and 19.83 TL/L, respectively, due to off-season demand drops and heightened market competition. LPG prices, generally more stable, peaked in March at 16.83 TL/L, coinciding with a shift in demand from industrial to automotive use. The lowest LPG price was noted in October at 10.34 TL/L.

In contrast to consumption trends, changes in pump prices are more closely linked to macroeconomic, geopolitical, and supply chain factors (see Figure 5). Diesel prices were the most volatile in November (σ = 0.1777 TL/L), reflecting instability in exchange rates and uncertainties in the global oil market. The lowest volatility was recorded in June (σ = 0.1464 TL/L), indicating a temporary impact of stable pricing policies. Gasoline prices exhibited a similar pattern, with the highest standard deviation also occurring in May (σ = 0.5052 TL/L), highlighting the vulnerability of both diesel and gasoline markets to international price fluctuations. The lowest volatility for gasoline was noted in June (σ = 0.1394 TL/L), suggesting a more cautious pricing strategy as the vacation season approached. LPG prices were most unstable in July (σ = 0.2405 TL/L), indicating that even alternative fuels are significantly affected by global petroleum product markets. The month with the least volatility for LPG prices was September (σ = 0.1828 TL/L). Lastly, as presented in Figure 6, the correlation matrix highlights substantial associations among fuel type prices, with particularly strong positive correlations observed between diesel and gasoline prices (correlation coefficient, r = 0.843, p < 0.001), and between gasoline and LPG prices (r = 0.829, p < 0.001). These high inter-fuel correlations reflect the interconnected nature of fossil fuel markets, which are shaped by common cost structures, regulatory environments, and demand-side dynamics. Moreover, a strong and statistically significant correlation is observed between the logarithm of provincial area and the number of villages (r = 0.741, p < 0.001), suggesting that as the territorial size of a province increases, the number of settlements tends to increase proportionally—reflecting a spatial pattern of rural dispersion associated with larger administrative regions. The variance inflation factors (VIFs) calculated for the unit pump prices of diesel, gasoline, and LPG were 4.47, 8.83, and 7.38, respectively. In addition, the variable representing the size of the provinces had the third highest VIF of 4.50. Of all the predictor variables, these had the highest levels of multicollinearity, which is consistent with the pairwise correlation coefficients discussed earlier. This pattern underscores the presence of notable interdependencies among price-related and spatial variables that warrant careful consideration in model specification and interpretation. Despite these associations, the variance inflation factor (VIF) for all the explanatory variables indicates that multicollinearity is not a concern, with all VIF values remaining well below the conventional threshold of 10. These diagnostics were performed for both versions of the analysis, including and excluding Istanbul. While the detailed results have been omitted for brevity, they are available from the authors upon request. This suggests that multicollinearity, while present to some extent, does not pose a significant threat to the validity of regression estimates. Therefore, the model can be considered sufficiently robust for inferential analysis, without necessitating immediate dimensionality reduction or variable exclusion strategies.

3.2. SUR Estimations

Before presenting the estimation results and model specifications in

Table 4. Estimates from the SUR system with the panel structure for liquid fuels considering NUTS-1 sub-regions in Turkiye, including Istanbul.

Variables	Diesel Quantity Demanded				Gasoline Quantity Demanded				LPG Quantity Demanded
	Estimate	Std. Err.	95% Confidence Interval		Estimate	Std. Err.	95% Confidence Interval		Estimate	Std. Err.	95% Confidence Interval
			Lower	Upper			Lower	Upper			Lower	Upper
Constant	−9.857 ***	3.600	−16.912	−2.802	−13.574 ***	3.552	−20.535	−6. 613	−6.761 *	3.515	−13.650	0.129
Log diesel price	0.744 ***	0.083	0.582	0.907	0.936 ***	0.073	0.794	1.078	0.862 ***	0.076	0.712	1.011
Log gasoline price	−0.532 ***	0.091	−0.710	−0.355	−1.184 ***	0.079	−1.339	−1.029	−0.215 ***	0.083	−0.378	−0.052
Log LPG price	−0.291 **	0.092	−0.470	−0.111	0.274 ***	0.080	0.117	0.431	−0.600 ***	0.084	−0.765	−0.435
Log GDP	1.186 ***	0.313	0.572	1.800	1.313 ***	0.309	0.707	1.918	0.609 **	0.306	0.009	1.208
Log village number	0.494 **	0.232	0.039	0.949	0.733 ***	0.229	0.283	1.182	0.867 ***	0.227	0.423	1.312
Log area	0.478 **	0.238	0.011	0.945	0.275	0.235	−0.186	0.736	0.214	0.232	−0.242	0.670
Vehicles per population	−1.983 **	0.845	−3.640	−0.326	−0.132	0.830	−1.759	1.495	0.926	0.826	−0.693	2.545
Regions:
Aegean	0.325	0.305	−0.273	0.922	0.145	0.300	−0.445	0.735	0.117	0.298	−0.467	0.700
Eastern Marmara	−1.618 ***	0.289	−2.184	−1.051	−1.325 ***	0.285	−1.885	−0.766	−1.627 ***	0.282	−2.180	−1.074
Western Anatolia	−0.197	0.287	−0.759	0.365	−0.373	0.283	−0.928	0.182	0.081	0.280	−0.468	0.630
Mediterranean	−0.563 *	0.290	−1.131	0.004	−0.757 ***	0.286	−1.317	−0.196	−0.601 **	0.283	−1.155	−0.047
Central Anatolia	0.058	0.454	−0.832	0.949	0.010	0.449	−0.870	0.889	0.456	0.444	−0.414	1.325
Western Black Sea	−0.880 ***	0.313	−1.494	−0.267	−0.741 **	0.309	−1.347	−0.135	−1.110 ***	0.306	−1.710	−0.511
Eastern Black Sea	0.161	0.359	−0.543	0.866	0.009	0.355	−0.687	0.705	−0.546	0.351	−1.234	0.142
Northeastern Anatolia	0.029	0.341	−0.639	0.698	0.062	0.337	−0.599	0.722	0.004	0.333	−0.650	0.657
Middle Eastern Anatolia	−1.458 ***	0.293	−2.033	−0.882	−1.113 ***	0.290	−1.681	−0.545	−1.209 ***	0.287	−1.771	−0.648
Southeastern Anatolia	2.458 ***	0.713	1.060	3.855	3.100 ***	0.704	1.720	4.480	2.359 ***	0.696	0.994	3.724
Seasons:
Winter	−0.213 ***	0.021	−0.254	−0.172	−0.276 ***	0.018	−0.312	−0.241	−0.227 ***	0.019	−0.264	−0.190
Some useful statistics:
σα	0.620				0.613				0.606
σu	0.175				0.153				0.161
R-squared	0.692				0.732				0.733
Overall R-squared	0.720
Huasman specification test	12.588 **  (p = 0.013)				0.071   (p = 0.999)				3.401   (p = 0.493)
Cross-dependence (CD) test	60.685 *** (p < 0.000)				135.900 *** (p < 0.000)				118.100 *** (p < 0.000)
Slope homogeneity (Δadj) test	−17.734 *** (p < 0.000)				−17.103 *** (p < 0.000)				−15.921 *** (p < 0.000)

Note: *, **, and *** indicate statistical significance at the 5, 10, and 1% levels.

and

Table 5. Estimates from the SUR system with the panel structure for liquid fuels considering NUTS-1 sub-regions in Turkiye, excluding Istanbul.

Variables	Diesel Quantity Demanded				Gasoline Quantity Demanded				LPG Quantity Demanded
	Estimate	Std. Err.	95% Confidence Interval		Estimate	Std. Err.	95% Confidence Interval		Estimate	Std. Err.	95% Confidence Interval
			Lower	Upper			Lower	Upper			Lower	Upper
Constant	−6.025	3.951	−13.768	1.718	−9.611 **	3.866	−17.187	−2.034	−1.914	3.682	−9.129	5.302
Log diesel price	0.757 ***	0.083	0.594	0.921	0.947 ***	0.073	0.804	1.091	0.872 ***	0.077	0.722	1.023
Log gasoline price	−0.532 ***	0.091	−0.710	−0.355	−1.193 ***	0.079	−1.349	−1.036	−0.215 **	0.084	−0.379	−0.051
Log LPG price	−0.320 ***	0.092	−0.500	−0.140	0.260 ***	0.081	0.101	0.419	−0.630 ***	0.085	−0.797	−0.464
Log GDP	0.977 ***	0.321	0.349	1.605	1.089 ***	0.314	0.475	1.704	0.312	0.299	−0.273	0.898
Log village number	0.697 ***	0.244	0.220	1.174	0.953 ***	0.238	0.485	1.420	1.173 ***	0.227	0.729	1.618
Log area	0.250	0.252	−0.245	0.744	0.028	0.247	−0.456	0.511	−0.129	0.235	−0.590	0.332
Vehicle per population	−3.022 ***	0.973	−4.930	−1.114	−1.296	0.946	−3.151	0.559	−0.858	0.910	−2.642	0.926
Regions:
Aegean	0.176	0.365	−0.540	0.892	0.152	0.357	−0.547	0.853	0.693 **	0.340	0.026	1.359
Eastern Marmara	−0.274	0.278	−1.016	0.467	−0.104	0.370	−0.829	0.621	0.317	0.353	−0.374	1.008
Western Anatolia	0.054	0.502	−0.930	1.038	0.171	0.491	−0.792	1.134	1.237 ***	0.468	0.320	2.154
Mediterranean	0.148	0.377	−0.591	0.886	0.327	0.369	−0.396	1.050	1.015 ***	0.351	0.326	1.703
Central Anatolia	−0.442	0.396	−1.218	0.334	−0.474	0.387	−1.233	0.285	0.491	0.369	−0.232	1.214
Western Black Sea	−1.022 **	0.400	−1.806	−0.237	−1.089 ***	0.392	−1.857	−0.322	−0.510	0.373	−1.241	0.221
Eastern Black Sea	−1.394 ***	0.437	−2.251	−0.537	−1.137 ***	0.428	−1.975	−0.298	−1.120 ***	0.408	−1.919	−0.321
Northeastern Anatolia	−2.132 ***	0.447	−3.008	−1.257	−1.724 ***	0.437	−2.579	−0.868	−1.648 ***	0.416	−2.464	−0.832
Middle Eastern Anatolia	−2.018 ***	0.469	−2.938	−1.098	−1.563 ***	0.459	−2.461	−0.664	−1.310 ***	0.438	−2.167	−0.452
Southeastern Anatolia	−0.916 **	0.465	−1.826	−0.005	−0.834 *	0.454	−1.724	0.056	−0.627	0.433	−1.476	0.222
Seasons:
Winter	−0.216 ***	0.021	−0.257	−0.176	−0.280 ***	0.018	−0.316	−0.244	−0.233 ***	0.019	−0.270	−0.195
Some useful statistics:
σα	0.619				0.606				0.576
σu	0.175				0.153				0.161
R-squared	0.668				0.707				0.748
Overall R-squared	0.710
Huasman specification test	12.208 **  (p = 0.016)				0.064   (p = 0.999)				3.815   (p = 0.432)
Cross-dependence (CD) test	60.224 *** (p < 0.000)				135.890 *** (p < 0.000)				117.530 *** (p < 0.000)
Slope homogeneity (Δadj) test	−14.063 *** (p < 0.000)				−12.988 *** (p < 0.000)				−11.574 *** (p < 0.000)

Note: *, **, and *** indicate statistical significance at the 5, 10, and 1% levels.

, it is important to address the structural peculiarity of Istanbul, which—based on all prior diagnostic assessments—exhibits clear outlier behavior in terms of fuel consumption and pricing dynamics. Given Istanbul’s disproportionate weight in both economic and demographic terms, its inclusion in a national model risks introducing self-referential bias into parameter estimates, potentially distorting the relationships observed across other provinces. To mitigate this issue, we employed a two-pronged strategy. First, Istanbul was merged with the Western Marmara region to diffuse its dominant influence within a broader geographical cluster. Second, we conducted a separate estimation by entirely excluding Istanbul from the dataset. As a result, two distinct modeling approaches were implemented: one in which Istanbul is incorporated within Western Marmara, and another in which Istanbul is fully excluded. This dual approach allows us to both account for and isolate Istanbul’s influence, offering a more nuanced and robust understanding of regional fuel demand patterns. Furthermore, we aimed to fully capture the seasonal effects in our analyses; however, when the spring and summer seasons were included as additional explanatory variables along with the winter season, high multicollinearity between the three fuel types emerged. Consequently, we excluded the spring and summer seasons from the analysis, focusing solely on the impact of winter compared to the rest of the year. This decision was guided by multicollinearity diagnostics: including more than one seasonal dummy variable while designating one as the reference category results in VIF values exceeding the conventional threshold of 10, which indicates severe multicollinearity. To preserve model stability and interpretability, therefore, the fuel demand equation system was restricted to a single seasonal dummy variable.

Table 4 and Table 5 provide key diagnostic statistics from the panel SUR estimations conducted separately for diesel, gasoline, and LPG prices, both with and without the inclusion of Istanbul. The R-squared and overall R-squared values remain consistently high across all models—ranging between 0.668 and 0.748—suggesting that the chosen explanatory variables effectively account for a substantial share of the variation in fuel demands. The error component variances (σₐ and σᵤ) show minimal fluctuation across specifications, indicating model stability. These findings confirm the adequacy of the SUR approach in capturing fuel demand dynamics in a panel structure, regardless of Istanbul’s inclusion.

Model selection diagnostics, particularly the Hausman specification test, offer valuable insight into the appropriate estimation strategy. For diesel, the test statistics are statistically significant in both datasets (p = 0.013 with Istanbul; p = 0.016 without), indicating that the fixed effects model is preferable due to correlation between the individual-specific effects and the regressors. In contrast, the gasoline and LPG models—under both data configurations—yield non-significant Hausman test statistics (p-values well above 0.05), supporting the use of a random effects specification in these cases. Although the Hausman specification test suggested that a fixed effects model would be more appropriate only for the diesel demand equation—while indicating that random effects are suitable for both gasoline and LPG—we ultimately estimated the system of demand equations for all three fuel types (diesel, gasoline, and LPG) under a random-effects framework using the SUR methodology. This decision was grounded in the need to maintain consistency across the system of equations and to accommodate unobserved heterogeneity across cities in a flexible yet coherent manner. Accordingly, for both datasets—with and without Istanbul—the panel SUR was estimated using a random-effects specification, which allows for inter-equation correlation while accounting for regional diversity in fuel consumption behavior. Importantly, the cross-dependence test is highly significant across all models (p < 0.000), validating the choice of the SUR estimator, which explicitly accommodates contemporaneous correlation across units. Additionally, given our data’s relatively short time dimension of 12 months, the adjusted delta (Δ_adj) test for slope homogeneity is a suitable and well-supported choice [47]. The results of this test strongly reject the null hypothesis of homogeneous slopes (t-statistics between −11.574 and −17.734), underscoring the need for heterogeneous slope modeling in capturing region-specific or fuel-specific dynamics.

Table 4 and Table 5 present a random effects panel SUR analysis of diesel, gasoline, and LPG demand in the twelve regions of Turkiye, comparing two scenarios: with and without Istanbul. The model assesses how factors such as fuel prices, GDP, land area, vehicle density, number of villages, and seasonal variation shape fuel consumption patterns. Given the economic and demographic weight of Istanbul, its inclusion significantly changes the magnitude and importance of key drivers. These findings provide important guidance for the design of region-specific energy policies, particularly in the context of Turkiye’s climate commitments under the Paris Agreement. As the transport sector remains a major contributor to national CO₂ emissions, such empirical insights are essential for designing effective, equitable, and targeted decarbonization strategies.

When Istanbul is included in the demand system, diesel prices exhibit a positive and statistically significant relationship with their own demand, with an elasticity coefficient of 0.744 (p < 0.01). To enhance transparency and provide readers with a statistical basis for interpreting these estimates, 95% confidence intervals have been also calculated for all system parameters and are presented as additional columns in Table 4 and Table 5, following a reviewer’s suggestion. Excluding Istanbul slightly increases its own elasticity to 0.757 (p < 0.01), suggesting that diesel demand is even more responsive to price changes in non-metropolitan areas. While both effects remain inelastic and counterintuitive, this paradox highlights the essential nature of diesel in the rural and economically less developed areas of the country, where public transportation is limited and the adoption of electric vehicles (EVs) remains minimal. In contrast, the price elasticity of gasoline demand is negative and statistically significant in both scenarios, indicating a more price-responsive consumption pattern. When Istanbul is included, the own-price elasticity of gasoline is −1.184 (p < 0.01), and when it is excluded, the own-price elasticity rises slightly to −1.193 (p < 0.01)—both indicating elastic demand for gasoline. This implies that a 1% increase in the real price of gasoline leads to a reduction in gasoline consumption of about 1.18% to 1.19%, underscoring the responsiveness of consumer/logistics hubs to gasoline price fluctuations, especially outside the metropolitan core. On the other hand, LPG demand shows a similarly negative and statistically significant own-price elasticity of −0.600 and −0.630 under the two scenarios, respectively. However, unlike gasoline, LPG demand remains price inelastic, suggesting its continued role as a low-cost and often irreplaceable alternative in vehicle energy use.

An examination of the cross-price elasticities elicits a fascinating asymmetry in the substitution behavior between different fuel types. In both datasets analyzed (e.g., with and without Istanbul), rising prices for gasoline and LPG lead to a significant decrease in diesel consumption, with the effect of gasoline price increases being much stronger than that of LPG. This indicates a closer competitive relationship between diesel and gasoline, probably due to their extensive use in transportation. On the other hand, gasoline consumption tends to increase when diesel and LPG prices increase. Such a simultaneous response suggests potential substitution, especially by LPG, which is often used in dual-fuel vehicles. However, the substantial increase in gasoline demand in response to higher diesel prices—almost 4.5 times greater than the effect of LPG—may also indicate anticipatory consumer/logistics hub behavior. As diesel prices rise, consumers may anticipate a general increase in fuel prices, leading them to use more gasoline while its price remains relatively stable. A particularly interesting trend can be observed for LPG consumption. While increases in diesel prices lead to a notable increase in LPG demand—possibly due to stockpiling or fear of substitution—higher gasoline prices have the opposite effect, significantly reducing LPG consumption. This may indicate a perception of tightening energy affordability, with rising gasoline prices suggesting broader fuel cost pressures, leading LPG users to adopt more cautious consumption habits. These complex relationships between fuels highlight the importance of understanding not only direct price elasticities, but also psychological and behavioral responses to expected future fuel costs.

GDP influences fuel demand significantly. The elasticity for GDP increases from 0.312 (p = 2.99) to 1.313 (p < 0.01), indicating that as economic growth accelerates, fuel consumption—along with associated emissions—rises in both metropolitan and rural regions. However, the effect is much stronger when Istanbul is included, reflecting the region’s pronounced income advantage. Higher-income areas such as Istanbul tend to have significantly higher demand for liquid fuels, underscoring the direct link between economic prosperity and energy consumption. In both datasets, an increase in the number of residential areas (such as villages and neighborhoods) is consistently associated with increased consumption for all three types of fuel. This trend is particularly pronounced for LPG and becomes even more significant when Istanbul is excluded from the analysis. The more compact urban layout of Istanbul is likely to allow for shorter transport distances, leading to less reliance on fuel-intensive travel. Conversely, areas outside Istanbul, with more dispersed settlements and longer commuting distances, appear to be more dependent on liquid fuels—especially LPG—for daily transport, highlighting the spatial and infrastructural difficulties of less densely populated regions. Similarly, vehicle density also plays a role, with higher vehicle concentration correlating with reduced fuel consumption, especially when using diesel. This suggests that in densely populated urban areas like Istanbul, the expansion of shared mobility options or increased public transportation usage could contribute to lower individual fuel consumption, helping reduce emissions.

Regional dummy variables provide valuable insight into how fuel demand varies in different regions of the country. For example, in the Eastern Marmara region, fuel demand decreases logarithmically from 1.325 to 1.627 when Istanbul is included as a reference category. However, such an effect becomes statistically insignificant (p > 0.05) when Istanbul is excluded, underscoring the critical role of Istanbul in driving fuel consumption in the Marmara region. Similarly, in the Mediterranean region, fuel demand is lower when Istanbul is included in the Western Marmara region, but when Istanbul is excluded, the region consumes more fuel than the Western Marmara region. This suggests that the urban structure of Istanbul, with its extensive transportation network and high vehicle density, plays an important role in shaping the fuel consumption patterns of the surrounding areas. In comparison, the Western Black Sea region has consistently lower fuel demand in both scenarios, whether Istanbul is included or not. This suggests that regions further away from Istanbul, especially those with less industrialization and infrastructure development, tend to have lower fuel consumption. Conversely, in the Eastern Black Sea region, when Istanbul is excluded, the region shows a lower demand for fossil fuels compared to the Western Marmara region, likely reflecting less reliance on motorized transport and a higher share of energy-efficient practices. Interestingly, in the Northeast Anatolia region, fuel demand is higher than in the Western Marmara region when Istanbul is included in the dataset, although this result is statistically insignificant. When Istanbul is excluded, fuel demand in the Northeast Anatolia region drops below that of the reference region. This could be due to the relatively lower vehicle density and fewer alternatives to fossil fuel consumption in these rural areas. In the Central Eastern Anatolia region, however, fuel demand is significantly lower compared to the Western Marmara region in both scenarios, which could be attributed to a combination of lower economic activity and fewer vehicles in the region. The Southeastern Anatolia region, on the other hand, presents a notable contrast: fuel consumption is significantly higher when Istanbul is included in the dataset but lower when Istanbul is excluded. This pattern might reflect the region’s dependence on fuel-intensive industries, which are more prominent when Istanbul’s logistics and economic activities are taken into account, but less so when Istanbul is excluded from the analysis. On the other hand, in both datasets analyzed, fuel consumption for all three fuel types was consistently lower during the winter months compared to the other seasons. This result, while expected, is particularly noteworthy in the context of environmental sustainability. The long and harsh winter conditions experienced in much of the country—often lasting up to six months—appear to significantly alter mobility patterns. In major metropolitan areas, residents are more likely to use public transportation than private vehicles during this season. In addition, adverse weather conditions tend to limit long-distance travel, further reducing fuel demand.

4. Discussion

Policymakers need robust empirical evidence to design effective energy pricing, taxation, and environmental policies. One of the central tools in this endeavor is the estimation of demand elasticities for petroleum products, which reveal how consumers and industries adjust their consumption in response to changes in fuel prices, income levels, and the relative prices of substitute fuels [37,38]. This study, based on seemingly unrelated regression estimates for diesel, gasoline, and LPG demand in 12 regions of Turkiye (both with and without Istanbul), contributes to a growing literature that seeks to measure these elasticities in a way that accounts for spatial, seasonal, and cross-fuel dynamics.

A key finding of this study is the surprisingly positive price elasticity of diesel demand, estimated at 0.744 with Istanbul and 0.757 without. These results diverge from much of the established literature, which often portrays diesel demand as relatively price inelastic, especially in the short run [37,38,49,50,51,52]. However, the Turkish case seems to be shaped by unique contextual factors [37,38]. Unlike in many developed economies, where diesel is primarily used by heavy-duty vehicles and industrial users, in Turkiye diesel is widely used by private vehicle owners-particularly in rural and peri-urban areas-because of its relative affordability per kilometer and superior fuel efficiency. As such, price increases may be more likely to induce behavioral changes in this consumer/industry segment, a pattern also suggested by evidence from Canada (Barla et al. [49]), Germany (Alberini et al. [53]), and Ghana (Mensah et al. [54]), where price sensitivity varies according to the purpose and necessity of fuel use. However, pricing mechanisms alone will not suffice, as the various diesel elasticity profiles identified in this analysis of Turkiye demonstrate that the country’s current fuel taxation system is environmentally unsound and economically wasteful. Policies must also address behavioral incentives that encourage speculative consumption, reform subsidy structures that distort market signals, and remove infrastructure barriers that impede the transition to cleaner alternatives. First, the unexpected positive elasticity of diesel demand highlights the limitations of relying solely on price mechanisms as a policy tool. Effective policies must also directly address speculative behavior, sector-specific subsidies that distort market signals, and infrastructure bottlenecks. Improving the transparency and predictability of fuel price adjustments can help discourage hoarding and speculative purchases; diesel subsidies in critical sectors such as agriculture and transport should be explicitly linked to measurable efficiency improvements, thereby encouraging more responsible consumption; and infrastructure and tax reforms need to be accelerated to enable a realistic and scalable shift to cleaner fuel alternatives. In addition, internalizing environmental costs through carbon-based taxation would send a clear and direct signal about the environmental impact of diesel consumption, encouraging freight and passenger transport, as well as car users, to make more sustainable choices consistent with global environmental imperatives. This holistic and integrated policy architecture offers a way to rationally rebalance demand and contribute meaningfully to the country’s climate commitments.

Conversely, gasoline demand in Turkiye shows strong price responsiveness, with elasticity estimates ranging from −1.184 to −1.193, consistent with global estimates reported by many researchers [50,52,54,55,56,57,58,59,60,61]. These findings suggest that gasoline users are more flexible in reducing consumption or shifting to alternatives, possibly due to the greater availability of public transportation or the discretionary nature of many gasoline-related trips, especially in urban centers. This is supported by Hymel and Small [62], who observed asymmetric consumer responses to rising and falling fuel prices, suggesting that perceptions and expectations may influence consumption patterns as much as actual price changes. Therefore, the distinct elasticity profiles of different fuels require a nuanced and carefully calibrated approach to taxation in Turkiye. Given the high price responsiveness of gasoline, increasing its environmental tax burden could serve as an effective tool to discourage consumption and align its cost structure with its environmental externalities.

LPG, a fuel widely used by low- and middle-income households in Turkiye for both cooking and transportation, has inelastic but statistically significant price elasticity estimates (−0.600 and −0.630). This reflects its essential role in household energy consumption and corroborates the findings of Mensah et al. [54], Hu et al. [61], and Yii et al. [60], who also report inelastic LPG demand, especially among lower-income groups. In line with previous research, as LPG is also widely used by economically vulnerable groups in Turkiye, a more cautious and equitable fiscal approach is required to avoid regressive effects.

The study also provides valuable insights into cross-price elasticities, which capture substitution behavior between fuel types. In particular, diesel demand decreases in response to rising gasoline and LPG prices (whether or not Istanbul is included), while increases in diesel prices induce substitution to gasoline. These patterns illustrate the interdependence of fuel markets, a phenomenon also observed in India by Bhuvandas and Gundimeda [63], who found that electricity is a potential substitute for fossil fuels in transportation. This finding is particularly relevant for policy in Turkiye, where expanding access to electric mobility and clean public transport could further incentivize a shift away from petroleum-based fuels. In this context, the entrenched role of LPG in rural areas—despite its low price elasticity—requires strong government support for innovation in LPG-related fuel efficiency technologies. This could include the development and widespread adoption of dual-fuel systems, low-consumption engine designs, and smart energy management technologies to reduce fuel consumption and associated emissions. In addition, the cross-price elasticities identified in this study have profound policy implications: any reform of fuel taxes or subsidies must carefully consider the substitution dynamics between fuel types, as these interactions can either enhance or inadvertently undermine the intended policy effects.

Moreover, the analysis reveals the significant role of income effects in shaping fuel demand. Gross domestic product, used here as a proxy for regional income, shows a strong positive and statistically significant relationship with fuel consumption when Istanbul is included in the model. This suggests that rising income leads to increased fuel consumption, supporting previous evidence from Dahl [64], Lin and Zeng [57], Hu et al. [61], who report positive income elasticities above unity for transportation fuels, while Gasim et al. [52] report a value below zero for Saudi Arabia. However, the weaker income effects in the specification excluding Istanbul point to the outsized influence of the capital on national averages, further highlighting the need for regionally differentiated policy approaches. In line with previous studies, Turkiye’s per capita income continues to rise, so an increase in overall fuel demand is inevitable. Redirecting this growing demand towards electric or hybrid vehicles through well-designed incentives and infrastructure development is a promising strategy for strengthening fiscal stability and ensuring long-term environmental sustainability in an increasingly globalized world. The comprehensive policy recommendations articulated in this study offer a sound path for increasing fiscal revenues and enhancing economic resilience. They also provide a strong, evidence-based foundation for decisive energy transition and renewed national commitment to tackling the 21st-century climate crisis. Turkiye’s current motor vehicle tax (MTV) must be fundamentally transformed. Rather than relying on outdated criteria such as engine size, the tax should be based on more relevant metrics, such as annual fuel consumption and associated carbon emissions. Higher taxes on high-consumption vehicles, alongside incentives for low-emission alternatives through tax rebates and other mechanisms, would increase public revenues and shift consumer preferences towards cleaner mobility options consistent with global sustainability goals.

From a spatial perspective, regional heterogeneity in fuel demand is evident. Peripheral and economically less developed regions, such as Eastern Anatolia and the Eastern Black Sea, consistently show lower fuel consumption, likely due to factors such as lower income levels, weaker infrastructure, and lower private vehicle ownership. These disparities are consistent with the findings of Del Granado et al. [65] and Arzaghi and Squalli [58], who document how regional income disparities shape energy consumption patterns and exacerbate the distributional effects of fuel price reforms. We therefore argue for the introduction of a regionally differentiated carbon tax that recognizes the country’s diverse economic and consumption patterns in Turkiye. In high-consumption urban centers such as Istanbul, for example, a targeted carbon tax could discourage demand and generate important revenues to finance improvements to public transport systems and investments in clean energy infrastructure. This would promote a more sustainable urban mobility ecosystem. Accelerating the transition to electric vehicles (EVs) is also critical, and this should be driven by direct and compelling financial incentives, strategic tax exemptions to improve affordability, and the widespread deployment of reliable and accessible charging infrastructure. The sensitivity of demand to price signals in diesel- and gasoline-dependent regions demonstrates that well-designed fiscal instruments and parallel infrastructure investments could deliver transformative results in EV adoption. In parallel, the strategic expansion of public transport infrastructure, particularly in underserved rural and peri-urban areas, would deliver significant environmental benefits and promote social equity by providing inclusive, sustainable mobility options.

On the other hand, seasonal dynamics also play a non-negligible role. Demand for all fuel types tends to decline during the winter months, as captured by the statistically significant seasonal dummy variables. This could reflect a combination of factors, including adverse weather conditions that reduce vehicle use and possibly a greater reliance on alternative heating sources. The observed seasonal variation is consistent with the findings of Molloy and Shan [66], who highlight the lagged nature of behavioral responses to fuel price shocks, particularly in colder climates. Finally, the methodological choice to use a panel SUR model is another strength of this study. By allowing for contemporaneous correlation between error terms across fuel types, the SUR approach captures unobserved shocks or common economic influences (such as regulatory changes or macroeconomic volatility) that affect all fuel types simultaneously. This multivariate framework is well-supported in the literature (Labandeira et al. [59]; Gasim et al. [52]) and provides more efficient and nuanced elasticity estimates compared to single-equation models.

5. Conclusions

Turkiye’s transportation sector is facing a pivotal moment—caught between heavy reliance on energy imports and the growing urgency to mitigate greenhouse gas emissions. The combination of these dynamics, coupled with the country’s reliance on imported energy and mounting pressure to reduce greenhouse gas emissions, has given rise to substantial fiscal and environmental challenges. In this context, policy responses must be based on robust empirical evidence and adapted to regional differences. This study contributes to this effort by estimating the price elasticities of diesel, gasoline, and LPG using a regionally disaggregated random-effects panel SUR model. To capture Istanbul’s outsized influence, two model specifications were employed: one including the city and one excluding it. This revealed meaningful spatial heterogeneity in fuel demand. The model demonstrated consistent robustness across specification tests and confirmed significant variations in fuel demand across geography and season. However, using a single-year dataset has limitations, particularly with regard to performing comprehensive time-series diagnostics, such as unit root or autocorrelation tests. Therefore, the findings should be interpreted within a short-term analytical framework. Notably, the positive price elasticity estimated for diesel demand—an unusual result—may be influenced by potential endogeneity issues, such as reverse causality or omitted variable bias. These issues could include industrial activity or transport demand that could not be fully captured. Although control variables were included, data constraints limited the ability to formally address these issues. Despite these limitations, the study provides valuable insights. It confirms that fuel price responsiveness in Turkiye is far from uniform and is strongly influenced by socio-economic, geographical, and behavioral factors. These findings challenge standard assumptions about fuel demand in developing economies and emphasize the importance of differentiated fuel-specific policies. Looking ahead, future studies should utilize longer-term, high-frequency data to explore regional and seasonal variations.