The Effects of Financial Stability, Economic Growth, and Industrial Production on Environmental Performance: Evidence from Türkiye

Abstract

Environmental performance constitutes a core dimension of the environmental, social, and governance (ESG) framework; however, empirical evidence for developing economies such as Türkiye remains limited, particularly regarding regime-dependent effects. Accordingly, this study examines the effects of financial stability, economic growth, and industrial production on environmental performance for Türkiye within a Markov-switching error correction and time-varying parameter state–space framework using annual data for 1990–2023. The findings show that financial stability is insignificant in the low error-variance regime but is negatively associated with environmental performance in the high error-variance regime. Economic growth is negatively associated with environmental performance in the low error-variance regime and becomes insignificant in the high error-variance regime. In contrast, industrial production has a positive and statistically significant effect on environmental performance in both regimes. The error correction mechanism indicates that short-run disequilibria are gradually corrected, supporting the existence of long-run convergence. Furthermore, the time-varying parameter state–space estimates indicate that these effects change over time. Overall, the findings indicate that the effects of financial stability, economic growth, and industrial production on environmental performance differ depending on regime periods and changing economic conditions.

1. Introduction

The environmental, social and governance (ESG) framework offers a widely recognized conceptual basis for assessing sustainable development performance at the global level. The environmental dimension of the ESG framework is closely aligned with the United Nations Sustainable Development Goals (SDGs), which aim to build a more sustainable future by 2030. It is also grounded in the Paris Climate Agreement (COP21), which seeks to mitigate climate risk on a global scale (United Nations Framework Convention on Climate Change, 2015). Within this context, the energy and environmental economics literature predominantly addresses climate risk within the scope of the SDGs, with environmental performance commonly assessed using indicators such as carbon dioxide (CO₂) emissions (Wen et al., 2021; Ali et al., 2024; Dridi et al., 2025; Khalid et al., 2025; Shaheen et al., 2025), ecological footprint measures (Abbas et al., 2025; Çabaş et al., 2025; Kelly & Nembot Ndeffo, 2025; Tunio et al., 2025) and the load capacity factor (Adebayo et al., 2025; Aybudak et al., 2025; Degirmenci et al., 2025; Javed et al., 2025).

One factor that has become increasingly important in the assessment of environmental performance is financial stability. It influences environmental performance through channels such as investment and capital allocation decisions, the environmental orientation of foreign direct investment, technological advancement processes, and risk management dynamics (Cole & Elliott, 2005; Weber, 2017; Krueger et al., 2020; Berg et al., 2022). These effects, however, are not homogeneous; they vary according to macro-financial conditions through differences in production structures and patterns of resource use. While periods of financial stability may constrain long-term environmental pressures by fostering environmentally sustainable investments (Cole & Elliott, 2005; Lan et al., 2012; Owen et al., 2018; Ren et al., 2022), they may also encourage the expansion of pollution-intensive activities by lowering perceived risk (Li et al., 2019; Ji et al., 2021; Uche et al., 2023). Likewise, technological progress can enhance environmental efficiency, yet it may simultaneously result in increased resource consumption through rebound effects (Özsoy, 2024; Iqbal et al., 2025). Consequently, the effect of financial stability on environmental performance is not fixed or unidirectional over time but is instead conditioned by economic regimes and evolving macro-financial conditions.

Within the SDG framework, Goal 8 (decent work and economic growth) represents one of the key objectives that directly captures the link between environmental performance and economic growth (Wursthorn et al., 2011; Chang & Hao, 2017). Although economic growth raises income levels, it simultaneously generates negative externalities affecting land, air, water, and biodiversity, thereby exerting pressure on environmental performance and potentially intensifying climate risks (Erdogan & Solarin, 2025). In this regard, SDG Goal 9 (industry, innovation and infrastructure), which seeks to expand the role of industrial production in economic growth, adopts an approach that places environmental performance at the forefront through the integration of environmentally sensitive production processes and the management of resource efficiency (Faizi et al., 2025). The composition of energy demand across different sources within countries’ production processes therefore emerges as a crucial factor in managing the environmental risks associated with economic growth and industrial activity (Rehman et al., 2019; Gani, 2021; Ma & Wang, 2025). Nevertheless, as energy mixes, production structures, and macroeconomic conditions evolve over time, the roles played by these factors in shaping environmental performance and climate risk also display a dynamic structure.

Regime changes, structural breaks, and macro-financial conditions, which are particularly common in developing countries, require that the determinants of environmental performance be analyzed within a dynamic framework. In that context, this study focuses on Türkiye’s economy, as a developing country that reflects these characteristics. More specifically, Türkiye is a rapidly growing economy with a substantial role in production and trade, and it frequently experiences structural breaks, regime changes, and internal economic fluctuations. Despite this, its production structure, especially in electricity generation and industrial activities, remains heavily dependent on fossil fuels. Moreover, as a party to COP21, Türkiye has committed to reducing its reliance on fossil fuels and to promoting the construction of a sustainable environment by 2030. In the empirical literature on Türkiye, environmental performance has been predominantly examined through indicators such as CO₂ emissions (Dar & Asif, 2018; Pata, 2018), ecological footprint (Destek & Sarkodie, 2019; Tutgun, 2025), and the load capacity factor (Yurtkuran & Güneysu, 2023; Yardımcı & Oskay, 2025). Based on the literature, this study constructs a composite environmental performance indicator using a Markov-switching error correction model and a time-varying parameter state–space approach, and it examines the effects of financial stability, economic growth, and industrial production on environmental performance in Türkiye over the period of 1990–2023.

This study seeks to contribute to the empirical literature in three main respects. First, it analyzes the short-term effects of financial stability, economic growth, and industrial production on environmental performance within a regime-dependent framework. By doing so, it highlights that these relationships may differ according to regime dynamics and macroeconomic conditions rather than reflecting average effects. Second, it employs multiple indicators for both environmental performance and financial stability in order to assess the macro finance–environment nexus across different regime periods and under varying variance conditions. Lastly, it examines changes in financial stability, economic growth, industrial production, and the speed of adjustment to equilibrium over time, thereby complementing the existing linear and non-linear time-series literature.

The remainder of this study is structured as follows. Section 2 reviews the theoretical framework and empirical literature and identifies literature gaps. Section 3 describes the data sources and outlines the methodological frameworks. Section 4 presents and discusses the empirical results. Finally, Section 5 concludes this study.

2. Theoretical Framework and Empirical Literature

2.1. Theoretical Framework

The theoretical framework of this study is based on the role of financial stability, which has gained increasing attention in the environmental performance and climate risk literature, in shaping exposure to climate risk. A substantial strand of this literature emphasizes that environmental performance indicators should not be treated solely as outcome variables reflecting environmental conditions, but also as indicators capturing the exposure of firms and economies to environmental and climate-related risks (Azzone et al., 1996; Ilinitch et al., 1998; Hoffmann & Busch, 2008). From this perspective, the role of financial stability is reflected in how environmental risks are accumulated, distributed, and managed, rather than in direct quantitative improvements or deteriorations in environmental performance (Landi et al., 2022; Sun et al., 2022). Accordingly, financial conditions influence the environmental dimension primarily through risk-related channels associated with investment capacity, financing conditions, and risk management behavior. This perspective is in line with prior studies that conceptualize environmental performance within a broader financial risk framework (Donaldson, 2001; Dafermos et al., 2018; Shahrour et al., 2023).

Financial stability shapes exposure to environmental risks through several transmission mechanisms operating within the broader framework of climate risk. In this context, the literature commonly identifies three main channels through which financial stability affects climate risk exposure: the scale channel, the composition channel, and the technology channel. The scale channel is associated with increased production volumes and higher energy consumption, which tend to raise carbon intensity and thereby increase potential exposure to climate risk (Grossman & Krueger, 1995; Sadorsky, 2010; Jalil & Feridun, 2011; Zhang, 2011; Shahbaz et al., 2013a, 2013b, 2013c; Kim et al., 2015; Acheampong, 2019). The composition channel, particularly in the context of trade liberalization, alters the sectoral and geographical distribution of environmental risks by encouraging the relocation of pollution-intensive activities toward countries with relatively weaker environmental regulations (Antweiler et al., 2001; Tamazian & Rao, 2010; Sadorsky, 2011; Krueger et al., 2020). The technology channel captures the dual role of technological change: while the adoption of cleaner production processes can mitigate environmental risks, efficiency gains may also generate rebound effects that increase total resource use, thereby limiting or even offsetting reductions in exposure to climate risk (Tamazian & Rao, 2010; Zhang, 2011; Acheampong, 2019; Kihombo et al., 2021; Chung et al., 2025). Importantly, the role of financial stability in shaping climate risk through these channels is not homogeneous across different regimes. The literature suggests that environmental risks are relatively more manageable under conditions of low financial volatility, whereas heightened uncertainty during periods of high volatility may weaken the risk management capacity and exacerbate the exposure to climate-related risks (Benkraiem et al., 2025; Ghani & Qin, 2025). Accordingly, financial conditions influence the environmental dimension primarily through risk-related channels associated with investment capacity, financing conditions, and risk management behavior.

2.2. Empirical Literature

Empirical studies on Türkiye’s economy have examined the relationship between financial development and environmental quality using a variety of indicators to capture environmental degradation. The existing literature predominantly relies on single environmental measures, which has resulted in heterogeneous findings depending on the indicator employed, the sample period, and the econometric methodology. For instance, Ozturk and Acaravci (2013), among the earliest studies to proxy environmental quality with CO₂ emissions, examined the period of 1960–2007 using an autoregressive distributed lag (ARDL) approach. Their results indicated that financial development had no statistically significant effect on environmental quality in the long run, while it contributed to environmental degradation in the short run, suggesting that the environmental implications of financial development vary between the short and long run. Similarly, Pata (2018) examined the long-run relationship between financial development and environmental quality for the period of 1974–2014 using ARDL, fully modified ordinary least squares (FMOLS), canonical cointegration regression (CCR), and regime cointegration tests. The results indicated that financial development significantly deteriorated environmental quality in the long run and that this relationship evolved over time, exhibiting sensitivity to structural breaks. Furthermore, Dar and Asif (2018) analyzed the period of 1960–2013 by employing ARDL and threshold cointegration techniques and found that financial development contributed to improvements in environmental quality through emission reductions; however, the strength and direction of this relationship varied in the presence of internal structural breaks. In a related study, Rjoub et al. (2021) applied structural break unit root tests along with FMOLS, DOLS, and CCR estimators and reported that financial development plays a moderating role in enhancing environmental quality, highlighting that the environmental impacts of financial factors are neither linear nor constant over time.

Empirical studies that assess environmental quality using the ecological footprint indicator likewise report non-linear empirical relationships. For instance, Destek and Sarkodie (2019) employed the augmented mean group (AMG) estimator for the period of 1977–2013 and found that financial development initially exerted a negative effect on environmental quality, but contributed to environmental improvement once a certain threshold was surpassed. This evidence suggests the presence of threshold behavior in the relationship between financial development and environmental outcomes. In contrast, Tutgun (2025) analyzed the period of 1992–2021 using Fourier ARDL, FMOLS, DOLS, and CCR techniques and revealed that financial development increased the ecological footprint, thereby worsening environmental quality. These findings indicated that when non-linear modeling approaches are applied, the environmental effects of financial development may remain adverse.

Environmental quality has also been evaluated using the load capacity factor as an alternative indicator. In this context, Yurtkuran and Güneysu (2023) analyzed the period of 1980–2018 using Fourier ADF and augmented ARDL (AARDL) techniques and found that financial development reduced the load capacity factor, thereby deteriorating environmental quality. Yardımcı and Oskay (2025), employing ARDL, FMOLS, DOLS, and CCR estimators, reported that green finance enhanced environmental quality, whereas financial globalization generated differing effects in the short and long run. More recently, Yurtkuran and Güneysu (2026) examine the period spanning 2008Q2 to 2020Q4 in the Türkiye economy using Fourier ARDL, FMOLS, and CCR methodologies. Their findings indicated that financial activities contribute to environmental degradation, whereas financial technologies play a mitigating role by reducing pollution levels. Notably, as the sample period encompasses crisis periods in Türkiye, the results found that different components of the financial system may exert heterogeneous environmental effects in terms of both direction and magnitude.

Existing empirical findings for Türkiye’s economy suggest that the relationship between financial development and environmental quality varies depending on the indicators, time period, methodology, and structural breaks considered. However, prior empirical studies do not systematically examine how this heterogeneity is shaped under different economic and financial regimes, and environmental quality is predominantly assessed using single indicators. Within that context, this study aims to contribute to the literature on Türkiye’s economy in several respects. First, it employs multiple indicators to capture both environmental performance and financial stability. Second, it adopts a regime-dependent and time-varying framework to examine the effects of financial stability, economic growth, and industrial production on environmental performance. Accordingly, this study formulates three hypotheses, which are presented below:

Hypothesis 1: 

The effect of financial stability on environmental performance differs across regimes.

Hypothesis 2: 

The effect of economic growth on environmental performance differs across regimes.

Hypothesis 3: 

The effect of industrial production on environmental performance differs across regimes.

3. Econometric Methodology

3.1. Data Description

This study examines the impact of financial stability (FS), economic growth (GDP), and the industrial production index (IP) on environmental performance (EP) in Türkiye’s economy. The baseline econometric model is specified as follows:

$${EP}_t={\beta }_0+{\beta }_1{FS}_t{ + \beta }_2{GDP}_t+{\beta }_3{IP}_t+e_t$$

(1)

In Equation (1), ${EP}_t$ denotes environmental performance, while ${FS}_t$, ${GDP}_t$, and ${IP}_t$ denote financial stability, economic growth, and the industrial production index, respectively. The parameter ${\beta }_0$ denotes the constant term, and ${\beta }_1$, ${\beta }_2$, and ${\beta }_3$ capture the estimated coefficients. The subscript $t$ indicates the time dimension, and the sample covers Türkiye’s economy over the period of 1990–2023. Moreover, $e_t$ represents the error term.

The environmental performance (EP) indicator reflects the environmental dimension of the ESG framework within the scope of the UN SDGs (6, 7, and 13). Following Diaye et al. (2022), the EP index is derived using principal component analysis (PCA) based on multiple environmental indicators, including PM2.5 air pollution (mean annual exposure, µg/m³), access to at least basic sanitation services (% of population), forest area (% of land area), combustible renewables and waste (% of total energy), renewable electricity output (% of total electricity output), and renewable energy consumption (% of total final energy consumption). Data are sourced from the World Development Indicators (World Bank, 2025b) database.

We construct the financial stability (FS) variable using PCA based on broad money (% of GDP), private credit by deposit money banks to GDP (%), and domestic credit to the private sector by banks (% of GDP). The WDI database provides the broad money and domestic credit variables, while the Global Financial Development Indicators (World Bank, 2025a) database provides private credit by deposit money banks to GDP. We include economic growth (GDP) as a control variable measured by GDP per capita (constant 2015 US$), obtained from the WDI database. We also include the industrial production (IP) as an additional control variable to capture industrial activity, measured as the industrial production index (2015 = 100) and obtained from the International Financial Statistics (International Monetary Fund, 2025) database.

Table 1. Data descriptions, measurements, and sources.

Variable	Measurement	Data Source
Environmental performance (EP)	Composite index constructed from PM2.5 air pollution (µg/m3), basic sanitation access (% of population), forest area (% of land area), combustible renewables and waste (% of total energy), renewable electricity output (% of total electricity output), and renewable energy consumption (% of total final energy consumption)	Author’s calculations based on WDI
Financial stability (FS)	Composite index constructed from broad money (% of GDP), private credit by deposit money banks (% of GDP), and domestic credit to private sector by banks (% of GDP)	Author’s calculations based on WDI
Economic growth (GDP)	GDP per capita (constant 2015 US$)	WDI
Industrial production (IP)	Industrial production index (2015 = 100)	IFS

presents the data descriptions, measurements, and data sources.

3.2. Markov-Switching ECM Methodology

This study employs a Markov-switching error correction model (MS–ECM) to examine the effects of FS, GDP, and IP on EP in Türkiye’s economy, in which the error correction mechanism is specified within a Markov-switching framework. The MS approach, originally introduced by Hamilton (1989), allows economic relationships to vary across unobservable regimes and captures regime changes such as expansion and contraction within a single framework. Unlike threshold-based models such as self-exciting threshold autoregressive (SETAR), threshold autoregressive (TAR), and smooth transition autoregressive (STAR), the MS framework does not rely on predetermined threshold values to characterized regime changes (Deschamps, 2008; Dueker et al., 2013). Instead, the MS framework defines regime dynamics through probabilistic transition rules. This structure enables regime-dependent behavior to be modeled in a more flexible and adaptive manner, particularly under conditions of structural uncertainty and abrupt economic changes (Hamilton, 1989; Maitland-Smith & Brooks, 1999; Wu & Chen, 2007).

In this study, we integrate the Markov transition structure with the residual-based two-stage error correction approach proposed by Engle and Granger (1987). In the first stage, we estimate the long-run relationship using ordinary least squares (OLS); once the cointegration condition is satisfied, we incorporate the resulting residuals into the model as an error correction term (ECT). This study, following Psaradakis et al. (2004) and Kapetanios et al. (2006), assumes that the cointegration vector and the ECT are common across unobservable regimes governed by a first-order Markov process. Accordingly, the MS–ECM framework adopted in this study allows for regime-specific error variances while maintaining a common long-run equilibrium relationship and a common ECT across regimes. The MS–ECM model employed in this study is specified as follows:

$${\mathsf{\Delta }EP}_t={{\beta }_0}^{\left(S_t\right)}+{\sum }_{k=1}^p{{\beta }_{1k}}^{\left(S_t\right)}{\mathsf{\Delta }FS}_{t-k}+ {\sum }_{k=1}^q{{\beta }_{2k}}^{\left(S_t\right)}{\mathsf{\Delta }GDP}_{t-k}+{\sum }_{k=1}^r{{\beta }_{3k}}^{\left(S_t\right)}{\mathsf{\Delta }IP}_{t-k}+{\lambda }^{\left(S_t\right)} {ECT}_{t-1}{+ e}_t$$

(2)

In Equation (2), the terms ${\mathsf{\Delta }EP}_t$, ${\mathsf{\Delta }FS}_{t-k}$, ${\mathsf{\Delta }GDP}_{t-k}$, and ${\mathsf{\Delta }IP}_{t-k}$ denote first-differenced variables, whereas $s_t$ captures coefficients that vary across regimes. Under this framework, the Markov transition probabilities are specified as follows (Burke & Rosenblatt, 1958; Billingsley, 1961):

$$s_t=\{\begin{array}{c}{p}_{11}\\ p_{12}\\ p_{21}\\ p_{22}\end{array} \begin{array}{c}{s}_{t }\\ s_t \\ s_t \\ s_t \end{array}\begin{array}{c}=\\ =\\ =\\ =\end{array} \begin{array}{c}1\\ 2\\ 1\\ 2\end{array} \begin{array}{c}{s}_{t-1 }\\ s_{t-1} \\ s_{t-1} \\ s_{t-1} \end{array}\begin{array}{c}=\\ =\\ =\\ =\end{array} \begin{array}{c}1\\ 1\\ 2\\ 2\end{array}$$

(3)

In Equation (3), $p_{11}$ denotes the probability that the process remains in the first regime in the subsequent period given that it is currently in the first regime, while $p_{12}$ represents the probability of transitioning from the first to the second regime. Similarly, $p_{21}$ refers to the probability of switching from the second to the first regime, and $p_{22}$ denotes the probability that the process stays in the second regime in the next period when it is initially in the second regime. Accordingly, the Markov transition matrix can be expressed as follows (Burke & Rosenblatt, 1958):

$$P\left(s_t=j |{ s}_{t-1}=i\right)=p_{ij}\hspace{1em}\hspace{1em}\hspace{1em}{\sum }_{j=1}^K{p}_{ij}=1$$

(4)

In Equation (4), $s_t$ denotes the unobserved regime, while $s_{t-1}$ represents the regime in the previous period. The index $j$ refers to the current regime, $i$ denotes the past regime, and $p_{ij}$ indicates the corresponding transition probability. Moreover, the condition ${\sum }_{j=1}^K{p}_{ij}=1$ implies that each row of the transition matrix in Equation (3) sums to unity.

This study allows the error variance to differ across regimes within the MS–ECM framework, as specified below (Frömmel et al., 2005):

$$s_t=1\to |{ e}_t\sim N(0,{\sigma }_1^2)\hspace{1em}\hspace{1em}\hspace{1em}{s}_t=2\to |{ e}_t\sim N(0,{\sigma }_2^2)$$

(5)

In Equation (5), ${\sigma }_1^2$ denotes the variance of the error term in regime 1, while ${\sigma }_2^2$ represents the variance in regime 2.

3.3. Time-Varying Parameter State–Space Methodology

In this study, we employ the MS–ECM framework to capture regime-dependent short-run dynamics under different variance structures. Nevertheless, from an economic theory perspective, the possibility that the influence of explanatory variables on the dependent variable may evolve gradually over time points to the need for a more flexible modeling framework. Accordingly, the existing literature frequently adopts complementary approaches that allow for time-varying parameters, most notably the rolling window method and state–space models estimated using the Kalman filter. More specifically, the rolling window approach relies on the OLS estimator and involves repeatedly estimating parameters over moving subsamples of the data (Ozcan et al., 2018; Balcilar et al., 2019). A key limitation of this method is that the window length is pre-specified (Zivot & Wang, 2003). If an appropriate window length cannot be determined, parameter estimates may become biased, and coefficients may exhibit excessive sensitivity to the choice of window (Pesaran & Timmermann, 2002; Bentz, 2003). Moreover, the use of a rolling window entails a loss of observations and prevents the full sample information from being utilized simultaneously (Swanson & White, 1997; Zivot & Wang, 2003). Therefore, despite its ability to estimate dynamic forecasts, this approach remains subject to several limitations.

On the other hand, the Kalman filter, first proposed by Kalman (1960), constitutes a widely used framework for the dynamic estimation of time-series models. In contrast to the rolling window approach, it accommodates both stationary and non-stationary processes within a unified framework and permits estimation that accounts for stochastically evolving parameters and unobservable (latent) states (Kalman, 1960; Kalman & Bucy, 1961; Broto & Ruiz, 2004). In addition, it avoids observation loss during the estimation process and facilitates a smoother representation of parameter evolution over time by recursively updating prediction errors and their associated conditional variances (Kalman, 1960; Woods, 2005). Drawing on these features, this study employs a time-varying parameter state–space model estimated using the Kalman filter to capture the time-varying effects of FS, GDP, and IP on EP, as well as the dynamic adjustment of the ECT. Within this framework, the model is specified in terms of a transition equation and a measurement equation. The transition equation in the Kalman filter framework can be expressed as follows (Kalman, 1960; Durbin & Koopman, 2012; Hamilton, 2020):

$${EP}_t={\phi }_{0,t}+{\gamma }_{FS,t}{FS}_t+{\delta }_{GDP,t}{GDP}_t+{\vartheta }_{IP,t}{IP}_t+{\lambda }_{ECT,t}{ECT}_t+v_t$$

(6)

In Equation (6), ${\gamma }_{FS,t}$, ${\delta }_{GDP,t}$ and ${\vartheta }_{IP,t}$ denotes the time-varying coefficients associated with FS, GDP, and IP, respectively, while ${\lambda }_{ECT,t}$ captures temporal variations in the speed of adjustment toward the long-run equilibrium. In addition, $v_t$ represents the observation error term.

The observation equation within the Kalman filter framework is specified as follows (Meinhold & Singpurwalla, 1983; Frühwirth-Schnatter, 1995):

$${EP}_t={h^{\prime }}_t{w}_t+v_t v_t\sim N(0,R)$$

(7)

where h_t and w_t are defined as:

$$h_t=\left[\begin{array}{c}1\\ {FS}_t\\ {GDP}_t\\ {IP}_t\\ {ECT}_t\end{array}\right] w_t=\left[\begin{array}{c}{\phi }_{0,t}\\ {\gamma }_{FS,t}\\ {\delta }_{GDP, t}\\ {\vartheta }_{IP, t}\\ {\lambda }_{ECT, t}\end{array}\right]$$

In Equation (7), ${EP}_t$ denotes the observed dependent variable, while, $h_t$ is the vector of observed explanatory variables, and $w_t$ contains the time-varying coefficients. The measurement error term $v_t$ is assumed to be normally distributed with zero mean and variance $R$, i.e., $v_t\sim N(0,R)$.

The state equation, which describes the evolution of the underlying structural relationship over time, is specified as follows (Frühwirth-Schnatter, 1995; Kandepu et al., 2008):

$$w_t=w_{t-1}+{\omega }_t {\omega }_t\sim N(0,Q)$$

(8)

In Equation (8), $w_t$ denotes the unobserved state vector of time-varying coefficients at time $t$. The disturbance term ${\omega }_t$ captures stochastic variations in the coefficients and is assumed to follow a normal distribution with zero mean and covariance matrix $Q$, i.e., ${\omega }_t\sim N(0,Q)$.

4. Empirical Results and Discussion

4.1. Unit Root Tests Results

The Augmented Dickey–Fuller (ADF) (Dickey & Fuller, 1979) and Phillips–Perron (PP) (Phillips & Perron, 1988) unit root test results are presented in

Table 2. Unit root tests results.

Variables	ADF Test				PP Test				ID
	Level		First (Δ)		Level		First (Δ)
	I	T & I	I	T & I	I	T & I	I	T & I
EP	−0.79	−1.92	−4.77	−4.69	−0.76	−1.92	−4.70	−4.61	[I(1)]
	(0.81)	(0.62)	(0.00) *	(0.00) *	(0.82)	(0.62)	(0.00) *	(0.00) *
FS	−0.90	−1.67	−5.01	−4.94	−0.90	−1.69	−4.97	−4.90	[I(1)]
	(0.77)	(0.74)	(0.00) *	(0.00) *	(0.78)	(0.73)	(0.00) *	(0.00) *
GDP	0.67	−2.58	−5.74	−5.78	1.97	−2.47	−6.50	−7.64	[I(1)]
	(0.99)	(0.29)	(0.00) *	(0.00) *	(0.99)	(0.34)	(0.00) *	(0.00) *
IP	1.95	−1.18	−4.63	−5.51	6.74	−0.32	−4.55	−11.25	[I(1)]
	(0.99)	(0.90)	(0.00) *	(0.00) *	(0.99)	(0.99)	(0.00) *	(0.00) *

Notes: * denotes significance at the 1% level. I = intercept model; T & I = intercept-and-trend model. ID denotes integration of degree. Source: author’s calculations.

. The ADF test indicates that the EP, FS, GDP, and IP variables are non-stationary at their levels under both the intercept (I) and intercept–trend (I & T) specifications. However, all variables become stationary and statistically significant at the 1% level after first differencing. Similarly, the PP test results reveal that the variables are non-stationary in levels across both model specifications, while their first differences exhibit stationarity at the 1% significance level. Overall, these findings confirm that all variables are integrated of order one, I(1), thereby satisfying the stationarity requirement for the MS–ECM framework employed in this study.

4.2. Cointegration Test Results

The results of the Johansen (1991) cointegration test, which examines whether variables that are stationary at the I(1) level move together in the long-run, are reported in

Table 3. Johansen cointegration test results.

Statistics	r = 0	r ≤ 1	r ≤ 2	r ≤ 3
Trace statistics	58.68 (0.00) *	36.99 (0.00) *	21.03 (0.00) *	8.45 (0.00) *
Max. eigenvalue statistics	21.68 (0.24)	15.96 (0.23)	12.59 (0.09)	8.45 (0.03)

Notes: * denotes significance at the 1% level. Source: author’s calculations.

. The trace test statistics indicate the existence of a cointegration relationship among the variables at the 1% significance level up to r ≤ 3. In contrast, the maximum eigenvalue test does not provide evidence supporting the same conclusion. However, since the trace test assesses the joint number of cointegrating vectors and is generally regarded as offering more robust inference, it is adopted as the primary criterion for evaluation. Taken together, the results indicate the existence of a long-run relationship of at least three periods among the variables, although the maximum eigenvalue test does not provide supporting evidence.

4.3. Markov-Switching ECM Estimates

The estimation results from the MS–ECM model are presented in

Table 4. Estimated coefficients from the MS–ECM model.

Panel A: Regimes
	Regime 1				Regime 2
Variables	Coef.	S.E	z-stat	prob	Coef.	S.E	z-stat	prob
ΔFS	0.185	0.157	1.174	0.240	−0.493	0.223	−2.209	0.027 **
ΔGDP	−4.584	1.897	−2.416	0.016 **	2.036	1.558	1.307	0.191
ΔIP	0.043	0.013	3.244	0.001 *	0.031	0.013	2.377	0.018 **
constant	0.174	0.067	2.594	0.009 *	−19.417	14.699	−1.321	0.187
logsigma (1–2)	−2.184	0.414	−5.277	0.000 *	−1.365	0.179	−7.623	0.000 *
Panel B: Common
		Coef.		S.E		z-stat		prob
ECT(−1)		−0.290		0.091		−3.176		0.002 *
Panel C: Diagnostic Tests
Tests		Statistics
Jarque–Bera (J–B)		2.24 (0.33)
ARCH (1–1)		3.21 (0.09)
Portmanteau (1)		2.68 (0.10)
Portmanteau (1–2)		2.83 (0.24)
Portmanteau (1–3)		5.11 (0.16)
AIC		0.88
SIC		1.43

Notes: * and ** denotes significance at the 1% and 5% levels, respectively. Source: author’s calculations.

. According to the findings, the FS does not exhibit a statistically significant effect on EP in regime 1, whereas in regime 2, increases in FS are associated with a reduction in the EP at the 5% significance level. The empirical results provide evidence in support of [$H_1$], positing that the effect of FS on EP differs across regimes. With respect to GDP, the results indicate a statistically significant negative effect on EP in regime 1 at the 5% level, while the corresponding coefficient in regime 2 is statistically insignificant. The findings partially support [$H_2$], as the effect of GDP is regime-dependent in low error-variance periods but becomes statistically insignificant in high error-variance regimes. Furthermore, the IP displays a positive and statistically significant relationship with EP in both regimes. Specifically, increases in the IP are associated with an increase in EP of approximately 1% in regime 1 and 5% in regime 2, suggesting that the magnitude of the IP effect differs across regimes. The results support $[H_3$], positing that IP exerts statistically significant effects across regimes, although the magnitude of these effects differs.

On the other hand, the parameters logsigma 1 and logsigma 2 represent the logarithmic forms of the estimated error variances for regime 1 and regime 2, respectively. Both estimated variance parameters are negative and less than unity. Nevertheless, the estimated logsigma 2 value exceeds logsigma 1, implying that the error variance in regime 2 is comparatively higher than that observed in regime 1. The ECT is specified as common across both regimes. More specifically, the ECT is negative (−0.29) and statistically significant at the 1% level, indicating that approximately 29% of deviations from the long-run equilibrium are corrected in the subsequent period.

Finally, the adequacy of the MS–ECM specification is assessed using a set of diagnostic tests, including the Jarque–Bera (J-B) normality test, ARCH (1–1), the Portmanteau (1–3) test, and the Akaike (AIC) and Schwarz (SIC) information criteria. The J-B test results indicate that the residuals are normally distributed, while the ARCH test suggests no evidence of conditional heteroskedasticity at the 10% significance level. In addition, the Portmanteau test results do not reveal autocorrelation up to three lags. Taken together with the AIC and SIC values, these findings support the adequacy of the MS–ECM specification.

The estimated regime classification indicates that regime 1 spans a total of 25 years, covering the periods of 1991–2014 and 2015, whereas regime 2 comprises 8 years over the period of 2016–2023, as reported in

Table 5. Estimated regime periods and transition possibility results.

Estimated Regime Periods
Regimes	Period (s)	Durations	AP	Regime types
Regime 1	1991–2014	24	0.99	Low error variance
	2015	1	0.60	Transition year
Regime 2	2016–2023	8	0.99	High error variance
Regime Transition Possibilities
Regime (t)	Regime (t + 1)	POT	AD (years)
$P\left(1\to 1\right)$	Regime 1	0.63	2.68
$P\left(1\to 2\right)$	$\mathrm{Regime} 1 \to$2	0.16	–
$P(2\to 1)$	$\mathrm{Regime} 2 \to$1	0.88	–
$P(2\to 2)$	Regime 2	0.12	1.14

Notes: AP, POT, and AD denotes average probabilities, transition probability, and average duration, respectively. Source: author’s calculations.

. The transition probabilities show that when Türkiye’s economy enters regime 1, which is characterized by low error variance, the probability of remaining in this regime is 63%. In contrast, upon entering regime 2, associated with high error variance, the probability of persistence in this regime is estimated at 16%. Moreover, the probability of transitioning from regime 2 back to regime 1 is relatively high, at 88%, while the probability of remaining in regime 2 is limited to 12%.

4.4. Time-Varying State–Space Estimates

The time-varying coefficient estimates for FS, GDP, IP, and the ECT in Türkiye’s economy over the period of 1992–2021 are obtained using the Kalman filter and reported in

Table 6. Time-varying coefficient estimates.

Years	($\mathit{\gamma }$) FSI	($\mathit{\delta }$) GDP	($\mathit{\vartheta }$) IP	($\mathit{\lambda }$) ECT
1992	0.1676	0.1575	0.2207	−0.0454
1993	−0.2306	0.7816	0.7565	−0.3559
1994	−0.0388	0.3933	0.4166	−0.4235
1995	−0.3311	−0.0034	−0.0561	−0.5204
1996	−0.5234	−0.2170	−0.2464	−0.4426
1997	−0.6363	−0.3880	−0.4509	−0.2322
1998	−1.5204	−0.2750	−0.3183	−0.9644
1999	−1.7312	0.0610	−0.0737	−0.3509
2000	−1.4917	−0.7598	−0.8680	−0.8170
2001	−1.5742	−0.6048	−0.6446	−0.8242
2002	−1.6412	−0.4171	−0.3444	−0.3996
2003	−1.6086	−0.4658	−0.3958	−0.4459
2004	−1.5175	0.1083	0.1360	−0.2485
2005	−1.6897	−0.2697	−0.3495	−0.0680
2006	−1.7355	−0.4212	−0.4774	0.0050
2007	−2.0599	−1.1822	−1.2776	0.5387
2008	−2.4398	−1.1213	−1.2163	0.5812
2009	−2.9179	−0.3268	−0.2071	0.1460
2010	−2.8978	−0.2613	−0.1454	0.1821
2011	−2.8695	0.0532	0.1747	0.4830
2012	−2.8128	0.3178	0.4006	0.8081
2013	−2.8601	0.3637	0.4122	0.8108
2014	−2.9810	−0.0151	−0.0020	1.0194
2015	−2.9217	0.2967	0.2642	1.1786
2016	−2.6693	0.5905	0.6158	1.2392
2017	−2.7105	2.0132	2.0393	0.9307
2018	−2.7461	2.0662	2.0574	0.9248
2019	−2.5635	1.9807	1.9933	0.4442
2020	−0.8007	2.2438	2.4769	−1.0255
2021	−0.6926	0.7985	0.7718	0.7080

Source: author’s calculations.

(see Appendix A). The estimates indicate that FS is associated with a negative effect on EP throughout the 1990s, a period marked by increasing economic integration. During the 1997–2015 period, the magnitude of this negative association becomes more pronounced, coinciding with major economic disruptions, including the 2001 banking crisis and the 2008 global financial crisis. From 2016 onward, the estimated coefficients suggest a gradual weakening of the negative effect of FS on EP. Nevertheless, as of 2021, the estimated coefficients indicate that FS continues to have a negative effect on EP. Overall, the time-varying estimates indicate that the effect of FS on EP evolves over time and exhibits sensitivity to different phases of economic activity in Türkiye.

Secondly, turning to the time-varying effect of GDP on EP, the estimated coefficients indicate a positive effect during the 1992–1994 period. Between 1995 and 2010, however, the effect of GDP on EP is predominantly negative, with the exceptions of 1999 and 2004. Following the global financial crisis of 2008, the estimated effect changes sign again after 2010, with GDP exhibiting a positive effect on EP from 2014 onward. After 2020, the magnitude of this positive effect gradually diminishes in the aftermath of the COVID-19 pandemic. Taken together, the time-varying estimates suggest that the effect of GDP on EP varies over time.

Thirdly, with regard to the time-varying coefficients capturing the effect of the IP on EP, the estimates indicate a positive effect during the 1992–1994 period. Between 1995 and 2010, the estimated coefficients predominantly point to a negative effect of IP on EP. After 2010, with the exception of 2004, the estimates suggest that the effect of IP on EP becomes positive again, although its magnitude varies over time. Following the COVID-19 pandemic, the positive effect of IP on EP gradually weakens. Collectively, the time-varying estimates indicate that the effect of IP on EP differs across periods and changes in both sign and size over time.

Finally, an examination of the time-varying ECT coefficients reveals notable changes over the sample period. The estimated ECT coefficient remains negative between 1992 and 2005, indicating that deviations from the long-run equilibrium were gradually adjusted during this period. Between 2006 and 2015, however, the coefficient exhibits a weakening adjustment pattern, suggesting a reduced speed of convergence toward the long-run equilibrium. Toward the end of the 2000s, the upward movement of the ECT coefficient implies a decline in the strength of the adjustment mechanism. Although the estimates point to a partial improvement in the adjustment process around 2020, this pattern does not persist into 2021, indicating that the speed of adjustment remained limited in the final period of the sample.

5. Conclusions

The progress of countries toward sustainable development objectives is commonly assessed within the framework of ESG performance, of which environmental performance constitutes a central component. Environmental performance remains closely linked to countries’ development trajectories, particularly in the context of production expansion and resource use. For developing economies, balancing economic activity with environmental performance remains a key challenge, with Türkiye providing a relevant case for analysis. Motivated by that context, this study examines the effects of financial stability, economic growth, and industrial production on environmental performance in Türkiye using a Markov-switching error correction framework and time-varying parameter state–space model covering the period of 1990–2023.

The empirical findings from the Markov-switching ECM indicate that the effects of financial stability and economic growth on environmental performance are regime-dependent and differ across regimes. Specifically, financial stability is statistically insignificant in the low error-variance regime, whereas in the high error-variance regime, it exhibits a negative and statistically significant association with environmental performance. In contrast, economic growth displays a statistically significant negative association with environmental performance in the low-variance regime, while this association becomes statistically insignificant in the high-variance regime. These findings suggest that the relationships between financial stability, economic growth, and environmental performance vary across regimes. Moreover, industrial production exhibits a positive and statistically significant association with environmental performance in both regimes, although the magnitude of this association differs across regimes. Consistent with these regime-dependent findings, results from the time-varying parameter state–space model indicate that the effects of financial stability, economic growth, and industrial production on environmental performance also vary over time. During periods of heightened economic stress, the estimated coefficients are associated with lower environmental performance, while in subsequent periods, these effects weaken or change sign. In addition, the time-varying estimates of the error correction mechanism suggest that the speed of adjustment toward the long-run equilibrium is not constant over time.

The findings indicate that the impact of economic activity on environmental performance in Türkiye is neither constant across regimes nor stable over time, highlighting the importance of accounting for regime-dependent and time-varying dynamics in environmental assessments. The results further suggest that financial stability, economic growth, and industrial production do not exert homogeneous effects, implying that the environmental implications of economic activity should be evaluated within prevailing macro-financial conditions. This study provides additional evidence on the macro finance–environment nexus in Türkiye’s economy by employing a framework that incorporates regime-dependent structures and time-varying coefficients, offering an alternative to approaches based on average effect assumptions. Despite these contributions, the analysis remains subject to certain limitations, as it focuses exclusively on the environmental dimension of ESG performance and does not incorporate variables such as technological intensity, research and development activities, or foreign direct investment. Future studies may extend this framework by integrating these factors to enable a more comprehensive and multidimensional assessment.