Introduction

Global environmental challenges are intensifying as the incline in global temperature and unexpected weather alterations become more and more evident. As a foremost GHG, CO2e is considered to be central in triggering climate change. In a new report from IPCC in 2023, CO2e was noted to have reached the highest levels over a two million years period and the 1.1 °C increase in global surface temperature in the last decades is alarming the nations to adopt cleaner technologies and to commit green innovation (IPCC 2023).

Conversely, the available carbon budgets aligned with meeting the temperature target outlined in the Paris Agreement were dwindling rapidly. As a result, the recent COP28 agreement in December 2023 announced “the end of the era of fossil fuel” (UNFCCC 2023). The COP28 agreement puts forth the fundamental targets: (i) engaging in strategies to cap the global temperature increases at 1.5 degrees Celsius, (ii) to support communities vulnerable to climate change impacts, (iii) they are attaining net zero emissions by 2050, in addition to, (iv) urging governments to accelerate the shift from fossil fuels to renewable energy sources such wind and solar, (v) providing financial and technological assistance to nations to achieve the upcoming climate commitments (UNFCCC 2023).

FinTech, an interdisciplinary field merging finance and technology, is increasingly relevant to environmental and energy concerns. As it becomes integral to daily life, there’s a pressing need for clean technology to reduce its ecological footprint. The expanding user base of FinTech amplifies energy demands and the necessity for cleaner energies. FinTech is characterized as technology-driven financial innovation, which has given rise to new business models, applications, processes, or products that hold the potential to significantly improve financial services (FSB 2023). Blockchain, cryptocurrencies, crowdfunding, and digital currency stand as the foundational pillars of FinTech tools. The FinTech industry is transforming various sectors, including online shopping, payment systems, insurance, cryptocurrency trading, and the credit market, particularly through the utilization of blockchain technology (Thakor 2020). Recent studies show that the founding pillars of FinTech are blockchain and cryptocurrencies, crowdfunding, and digital currencies, (Cai 2018; Muhammad et al. 2022). Almost all of the most popular FinTech means are generated through high levels of complex mathematical calculations done by computers that necessitate high levels of EC. Among these pillars, Bitcoin, the major cryptocurrency has a yearly EC predicted to be between 135 and 177 TWh due to PoW, the algorithm used in the cryptocurrency mining for Bitcoin (de Vries 2020) with complex calculations creating an EC equal to those from countries such as Thailand or Sweden (Kohli et al. 2023). Each cryptocurrency transaction as a FinTech demands 619 kWh of electricity, equal to the EC of an average US family for 21 days (Badea and Mungiu-Pupazan 2021). As the yearly EC for FinTech is more than many countries, the total energy needed for all FinTech leads to tremendous amounts. Further, the EC of total FinTech technologies is predicted to result in more carbon emissions than the annual carbon emission output of the Czech Republic and Qatar (Jiang et al. 2021). It can be accepted that FinTech is one of the many factors influencing the path that environmental pollution will follow.

It is of crucial importance to study the relationships between FinTech and energy in addition to their links to the environment. With this focus, this study centers on the FinTech markets by utilizing the FinTech stock market index. The FinTech stock index is designed to capture the performance of the FinTech companies in time and the inclines in the index are under the influence of firm performances and expectations for future firm performances. the fluctuations in the FinTech market also affect the energy markets. As the new technologies are fostered, energy demand will be positively affected, leading to upward shifts in energy prices. Further, the shifts in both energy and FinTech are expected to influence the environment through more CO2e due to the acceleration of EC in economies. However, it is natural to expect that the inclines or declines in FinTech would not have radical effects on normal to moderate levels measured with the quantiles of the data, but instead, the increases in FinTech and energy prices are expected to have influences at extreme levels, in the upper and lower directions. Such relations could be effectively captured with contagion models that focus on modeling the tail behavior between variables.

The data unavailability limit the empirical analyses in examining these complex dynamics. In this paper, we argue that the effects of FinTech development, coupled with the advancement of financial technologies, could influence the overall market capitalization and radical fluctuations of the stocks of FinTech companies and FinTech markets also affect the energy markets in various ways. Further, we argue that the extreme levels of FinTech markets lead to inclinations in the consumption of energy, which could only be sustainable as long as the incline in the EC would not lead to increased CO2e.Footnote 1

To overcome the data issues, we employ recent global FinTech market data that covers daily observations for over a decade in addition to EC and global CO2e to achieve a global perspective. The modeling of FinTech development-led extreme FinTech stock market fluctuations, in addition to oscillations in EC and CO2e is challenging due to its distributional characteristics of nonlinearity and heteroskedasticity, especially for the daily data employed in this study. The study employs two novel methods that help in measuring the tail behavior of FinTech development measured with FinTech stock market capitalization and size and their relations with EC and CO2e. At the first stage, the MS-GARCH-copula approach (Bildirici et al. 2022), allows capturing nonlinearity and heteroskedasticity simultaneously in marginal distributions of each series, and afterward, their joint distributions are effectively modeled to measure tail dependence and contagion relations at the upper and lower tails, or extreme levels of data. The second stage in this paper consists of the RSGCC tests, which incorporate the above-mentioned characteristics of data to measure nonlinear causal links under heteroskedasticity at the lower and upper tails at distinct regimes. The method is a generalization of a novel causality analysis proposed by Kim et al. (2020) for Granger causality testing under GARCH-copula. in the RSGCC method, the copula-based Granger causality testing is generalized to regime switching to examine causality dynamics under Markov-switching and heteroskedasticity (Bildirici et al. 2022).

The paper contributes both theoretical and empirical aspects. For the former, the paper is the first to examine FinTech, EC, and CO2e in terms of volatility contagion and causality dynamics. For the empirical aspect, the paper utilizes two novel techniques to control the nonlinear and volatile characteristics of the data and to produce insights into the relations at the tails of data. Further, with this modeling strategy, the paper also aims to overcome the data unavailability issue for FinTech and to provide the basis for future studies. In the last section, the findings of the novel methods will also be compared with the linear approach to causality modeling to point out the effects of omitting the nonlinearity and its possible translation into inefficient policy recommendations.

This study comprises five sections. Sect. "Literature" encompasses the literature review. Sect. "Methodology" introduces the econometric methodology. Sect. "Empirical results" provides the data and findings. The final section encompasses policy recommendations and conclusions.

Literature

The literature review will be conducted in three parts. The first focuses on the recent advancements in the literature in applying econometric models with a focus on daily data utilization for environmental implications of financial market fluctuations. Sharif et al. (2023) explore financial markets and their effects on environmental degradation and their findings underline the necessity of steps that should be taken towards greening of financial markets and towards achieving the environmental goals. A set of recent studies examine financial assets and ecological innovations. Among these, Sadiq et al. (2023) stress the need for greening of financial assets and the requirement of urgent eco-innovation. As a pillar of FinTech, Bitcoin has been central in terms of technology and innovation leading to EC and a high ecological footprint (de Vries et al. 2022). Bildirici et al. (2024) questioned the sustainability of metaverse by showing chaos, entropy, fractionality, and complexity in metaverse stock markets, EC from Bitcoin’s cryptocurrency mining and CO2e and they exposed contagion spillover and causality from metaverse to EC and environmental pollution. Gharbi et al. (2023) showed spillovers and risk connectedness amid the FinTech industry and macroeconomic dynamics that also inclined in the Covid-19 period.

Greening of FinTech technologies becomes an emerging factor to contribute to the environment. However, to our knowledge, there is no research on FinTech particularly with a focus on the nexus between FinTech stock markets, EC, and CO2e with emphasis on contagion and causality dynamics under nonlinearity and volatility. From a business perspective, the need for commitment to transforming FinTech companies into environmentally and socially responsible companies is highlighted (Carè et al. 2023). While FinTech markets are a relatively unexplored field in this respect, cryptocurrency markets are not due to their energy-hunger. Cryptocurrencies are considered among the three pillars of FinTech. They are criticized for their energy use and environmental effects, especially for those that adopt energy-hungry mining technologies such as the major cryptocurrency, Bitcoin. Bitcoin is predicted to demand yearly energy reaching those from countries, leading to a significant carbon footprint (de Vries et al. 2022). The technology behind plays a crucial role. The PoW algorithm and the Bitcoin transactions require high amounts of EC occurring due to complex calculations, and their yearly EC reaches the EC of countries such as Thailand (Kohli et al. 2023). In another study, each cryptocurrency transaction is predicted to demand 619 kWh of electricity, equal to the 3-week EC of a typical household in the USA (Badea and Mungiu-Pupazan 2021).

The EC of FinTech is also predicted to result in significant levels of annual carbon emissions (Jiang et al. 2021). These findings put FinTech at the center of research and FinTech should be considered to influence the path that environmental pollution will follow in the future. In addition to the challenges to achieve greener financial technology coupled with green energy and energy efficiency, the energy transition commitment of nations and relevant policies appear as a second factor in shaping the carbon footprint of FinTech and its environmental impact.

As a second strand of literature, the cryptocurrency markets have been examined for their energy demand and environmental friendliness and as noted in Sect. "Introduction", cryptocurrencies are among the main pillars of FinTech. The examination is given in Table 1. The connection between Bitcoin and environmental damage has been strengthened by certain research. O’Dwyert and Malone (2014) proposed a method to calculate the energy footprint of Bitcoin mining for the Bitcoin network in Ireland. Their estimated results emphasize that EC measured with the electricity demand of Bitcoin mining is unexpectedly high, on par with that of Ireland (O’Dwyert and Malone 2014). More recently, Wimbush (2018) showed that cryptocurrency mining and its technology could create more value than its costs; however, it has been uneconomic so far due to high level of energy and environmental costs, while Kamran et al. (2022) showed that Bitcoin as a FinTech could be used as a safe-haven with certain limits, in addition to its function to diversify the investment portfolios.

Table 1 Empirical literature review on cryptocurrencies as a pillar of FinTech
Full size table

A third strand of literature focuses on FinTech technologies, their innovation, and their effects on the environment. However, the focus is so far limited due to the availability of data. Further, the overlook suggests two different points, negative or positive, for the effects of FinTech on the energy and environment. Further, the studies are generally restricted to yearly datasets or questionnaire-based secondary data without time dimensions. Croutzet and Dabbous (2021) question whether FinTech encourages renewable energy in the total energy mix of 21 OECD countries with a balanced panel covering 2005–2018 and they conclude that the benefits of financial technology advancements in promoting renewable energy in the overall energy mix. Muganyi et al. (2022) examined how commerce, industrialization, green finance, and the FinTech sector affect pollution in the environment. Their empirical findings underscore the vital role of the FinTech industry and green finance in enhancing environmental quality and curbing pollution (Muganyi et al. 2022). Zhou et al. (2020) looked into how environmental pollution in China is affected by green funding. Elheddad et al. (2021) examined data from 29 selected OECD countries and demonstrated that electronic finance technologies contribute to a reduction in carbon emissions. Muhammad et al. (2022) investigated how 23 EU nations’ environmental efficiency was affected by the FinTech sector.

The overall investigation of the literature above suggests excessive ecological effects of cryptocurrencies as a pillar to FinTech and also to financial technologies through their effects on the EC to the environment. An overlook of policy recommendations is the need to focus on the greening of financial technologies and relevant innovation for environmental sustainability targets.

Methodology

MS-GARCH processes for marginal distributions

In an MS-GARCH model, a time series rt follows the following dynamics (Haas 2004),

$$r_{{\text{t}}} = \theta_{{0,\left( {st} \right)}} + \mathop \sum \limits_{i = 1}^{f} \theta_{{i,\left( {st} \right)}} r_{{{\text{t}} - i,\left( {st} \right)}} + \mathop \sum \limits_{n = 1}^{m} \lambda_{{n,\left( {st} \right)}} \varepsilon_{{t - n,\left( {st} \right)}} + \varepsilon_{{t,\left( {st} \right)}}$$
(1)
$$\sigma_{{t,\left( {st} \right)}}^{2} = \alpha_{{0,\left( {st} \right)}} + \mathop \sum \limits_{n = 1}^{q} \alpha_{{i,\left( {st} \right)}} \sigma_{{t - n,\left( {st} \right)}} + \mathop \sum \limits_{n = 1}^{p} \beta_{{i,\left( {st} \right)}} \varepsilon_{{t - n,\left( {st} \right)}}^{2}$$
(2)

here, Eq. (1) assumes that rt is a time series, which follows Markov-switching in its conditional mean with regime-dependent ARMA processes combined with Eq. (2), a regime-dependent conditional variance process (Bildirici and Ersin 2014). The model given in Eqs. (1) and (2) assumes ARMA terms in addition to the intercept (I) and the heteroskedasticity (H) to be governed with MS, therefore, the model above is annotated as an (MSIARMAH-GARCH) model. To ensure positivity of the variance process, the parameters \(\phi ,\,\,\alpha ,\,\,\kappa ,\,\,\beta \, > 0\) are restricted for non-negativity. Similarly, if no moving average terms are included, model reduces to a MSIARH-GARCH. A more tractable model is the MSIH-GARCH, in which the dynamics that rt follows assumes no autoregressive nor moving average terms in the mean. The MSIH-GARCH model,

$$r_{{\text{t}}} = \theta_{{0,\left( {st} \right)}} + \varepsilon_{{t,\left( {st} \right)}}$$
(3)
$$\sigma_{{t,\left( {st} \right)}}^{2} = \alpha_{{0,\left( {st} \right)}} + \mathop \sum \limits_{n = 1}^{q} \alpha_{{i,\left( {st} \right)}} \sigma_{{t - n,\left( {st} \right)}} + \mathop \sum \limits_{n = 1}^{p} \beta_{{i,\left( {st} \right)}} \varepsilon_{{t - n,\left( {st} \right)}}^{2}$$
(4)

Further, in case of insignificance of the intercept terms, the model reduces to a MS-GARCH process,

$$r_{{\text{t}}} = \varepsilon_{{t,\left( {st} \right)}}$$
(5)
$$\sigma_{{t,\left( {st} \right)}}^{2} = \alpha_{{0,\left( {st} \right)}} + \mathop \sum \limits_{n = 1}^{q} \alpha_{{i,\left( {st} \right)}} \sigma_{{t - n,\left( {st} \right)}} + \mathop \sum \limits_{n = 1}^{p} \beta_{{i,\left( {st} \right)}} \varepsilon_{{t - n,\left( {st} \right)}}^{2}$$
(6)

In all variants of the MS-GARCH models above, we assume that the switching standard error mechanism is conditional on the past,

$$\sigma_{{t - i - 1,\left( {st - i} \right)}} = E\left( {\varepsilon_{{t - i - 1,\left( {st - i - 1} \right)}} {|}s_{{{\text{t}} - i}} ,r_{{{\text{t}} - i - 1}} } \right)$$
(7)

where st is a Markovian process governing the switches among the regimes,

$$\pi_{nt} = \Pr \left[ {s_{t} = j\left| {r_{{{\text{t}} - 1}} } \right.} \right]\, = \sum\nolimits_{i = 0}^{1} {\Pr \left[ {s_{t} = j\left| {s_{t - 1} = i} \right.} \right]} \Pr \left[ {s_{t} = j\left| {r_{{{\text{t}} - 1}} } \right.} \right]\sum\nolimits_{i = 0}^{1} {\psi_{ni} } \pi_{it - 1}^{*}$$
(8)

with Log-likelihood (L) defined as,

$${\text{L}} = \mathop \prod \limits_{t = 1}^{T} f\left( {r_{{\text{t}}} {|}s_{t = i} , r_{{{\text{t}} - 1}} } \right)\Pr \left[ {\left. {s_{t = i} } \right|r_{{{\text{t}} - 1}} } \right].$$
(9)

There are different approaches to define \(\sigma_{{t,\left( {st} \right)}} .\) and \(\varepsilon_{{t - 1,\left( {st} \right)}}^{2}\) and the necessary stability conditions (Francq and Zakoïan 2001; Henneke et al. 2009) and the model in this paper assumes Francq and Zakoian’s approach similar to a set of studies (Bildirici et al. 2022; Bildirici and Sonustun 2021).

MS-GARCH-Copula for joint distributions

The joint distribution is measured with Clayton, Student’s t, and Gumble copula functions. Following Sklar’s extended theorem, the method assumes time-varying copulae, which leads to time-variation in the dependence structures (Bildirici et al. 2022). The following function defines copula as,

$$C_{\tau } (u_{1} \,,\,u_{2} \left| {\theta ,\,v} \right.) = \int\limits_{ - \infty }^{{{\tau_{u}{ - 1(u_{1} )}} }} {\int\limits_{ - \infty }^{{{\,\tau_{u}{ - 1(u_{2} )}} }} {\frac{1}{{2\pi \sqrt {(1 - \theta^{2} )} }}} } \exp \left( {1 + \frac{{s^{2} - 2\theta st + t^{2} }}{{v(1 - \theta^{2} )}}} \right)^{{ - \frac{ - v + 2}{2}}} \,{\text{dsdt}}$$
(10)

named as a Student’s t copula, which assumes symmetric tails,

$$\lambda_{{{\text{LT}}}} = \lambda_{{{\text{UT}}}} = - 2\tau_{v + 1} \left( {\sqrt {v + 1} \,\frac{{\sqrt {1 - \theta } }}{{\sqrt {1 + \theta } }}} \right) > 0$$
(11)

where \(\lambda_{{{\text{LT}}}}\) and \(\lambda_{{{\text{UT}}}}\) are lower and upper tail copulae. Asymmetric copula functions are suggested. For the following copula function,

$$C\left( {\hat{u}_{1t,} \hat{u}_{2t} {|} \theta } \right) = {\text{exp}}\left( { - \left[ {( - \ln \left( {u_{1} } \right))^{\theta } + ( - \ln \left( {u_{2} } \right))^{\theta } } \right]^{{\frac{1}{\theta }}} } \right)$$
(12)

Gumble copula is accomplished by,

$$\lambda_{{\text{U}}} = \,2 - 2^{{ - \frac{1}{\theta }}} ,^{{}} \lambda_{L} = 0\,$$
(13)

where the power coefficient \(\theta \, \in \left] {1,\,\, + \infty } \right[\).

The Clayton copula is given as,

$$C(u_{1} ,\,u_{2} \left| \theta \right.) = \,(u_{1}^{ - \theta } + \,\,u_{2}^{ - \theta } - 1)^{{ - \frac{1}{\theta }}} \,\,\,{\text{and}}\,\,\,\,\theta \, \in \left] {0,\, + \infty } \right[\,\,\,\lambda_{U} = 0\,$$
(14)

here, \(\lambda_{{\text{U}}} = 0\,,\lambda_{{\text{L}}} = \,2^{{ - \theta^{ - 1} }} \,\) are the upper and lower tail copula parameters. The estimation necessitates the maximization of the log-likelihood (LL),

$${\text{LL}} = \mathop \sum \limits_{t = 1}^{T} \ln \left[ {\mathop \sum \limits_{j = 1}^{n} \left( {\hat{u}_{1t,} \hat{u}_{2t} {|}s_{t = j} , \Upsilon_{t - 1} ,} \right)\Pr \left( {s_{t = j} {|}\Upsilon_{t - 1} } \right)} \right]$$
(15)

where the regime-switching probabilities are conditional on \(\Upsilon_{t - 1}\), noted as \({\text{P}} (s_{t} = j\left| {\Upsilon_{t - 1} } \right.)\) and \({\text{P}}\left( {s_{t = j} {|}\Upsilon_{t} } \right)\), the information set available at the period t-1 and t. By utilizing Kim’s filter (Kim 1994), the conditional probability is,

$${\text{P}} (s_{t} = j\left| {\Upsilon_{t - 1} } \right.) = \,\sum\limits_{i = 1}^{n} {{\text{P}}_{ij}^{t - 1} } \Pr (s_{t} = i\left| {\Upsilon_{t - 1} } \right.)$$
(16)

for the former. For the latter,

$${\text{P}}\left( {s_{t = j} {|}\Upsilon_{t} } \right) = \frac{{C\left( {\hat{u}_{1t,} \hat{u}_{2t} {|}s_{t = j} , \Upsilon_{t - 1} } \right){\text{P}}\left( {s_{t = j} {|}\Upsilon_{t - 1} } \right)}}{{\mathop \sum \nolimits_{j = 1}^{n} C\left( {\hat{u}_{1t,} \hat{u}_{2t} {|}s_{t = j} ,\Upsilon_{t - 1} } \right){\text{P}}\left( {s_{t = j} {|}\Upsilon_{t - 1} } \right)}}$$
(17)

Transition followed by smoothed transition probabilities are,

$${\text{P}}_{ij} = {\text{P}}\left( {s_{t = j} {|}s_{{t - 1 = {\text{i}}}} { },{ }\Upsilon_{t - 1} } \right) {\text{and}} {\text{P}}\left( {s_{t = j} {|}\Upsilon_{t} } \right) = {\text{ P}}\left( {s_{t = j} {|}\Upsilon_{t} } \right)\mathop \sum \limits_{i = 1}^{n} \frac{{{\text{P}}_{ij} {\text{ P}}\left( {s_{t + 1 = i} {|}\Upsilon_{t} } \right)}}{{\Pr \left( {s_{t + 1 = i} {|}\Upsilon_{t} } \right)}}$$
(18)

further, the dependence parameter follows a time-varying structure completed by,

$$\hat{\Upsilon }_{t} = \mathop {{\text{argmax}}}\limits_{\Upsilon } \sum\limits_{t = 1}^{T} {\ln (C(\hat{u}_{1t} ,} \,\hat{u}_{2t} ;\Upsilon_{t} ))$$
(19)

Granger causality (GC) testing with MS-GARCH-Copula

Kim et al. (2020) proposed a method to test GCs with the copula-based methods. The method generalizes GC (Granger 1969) to copulae. Relevant literature on the effectiveness of copulae in causality modeling exists (Nalatore et al. 2014; Zhu et al. 2016). The method is generalized to nonlinear and high-order causality modeling (Bildirici et al. 2022).

Let \(A = a_{t} , \;B = b_{t}\), \(A{ \nrightarrow }B\) stating no-causality from A to B; \(A \mapsto B\) denotes a causal link from A to B and the null hypothesis of no GC from A to B is in fact an equality of explanatory power of two functions,

$$f\left( {b_{t + 1} {|}b_{t}^{n} a_{t}^{m} } \right) = f\left( {b_{t + 1} {|}b_{t}^{n} } \right)$$
(20)

\(b_{t}^{n} = \left( {b_{t} , \ldots , b_{t - n + 1} } \right)\) and \(a_{t}^{m} = \left( {a_{t} , \ldots , a_{t - m + 1} } \right)\) are the available information for \(B\) and \(A\), n and m are orders, f defines conditional probability for density function. \(f\left( {b_{t + 1} {|}b_{t}^{n} a_{t}^{m} } \right)\) shows prediction of B at the next period t + 1, with available information from previous period t. The right-hand-side function, \(f\left( {b_{t + 1} {|}b_{t}^{n} } \right)\), excludes A at t + 1 in modeling and predicting B and utilizes B at t only for this purpose. H0: A↛B tests no GC against H1: A \(\mapsto\) B and the likelihood ratio test statistic is calculated as,

$$GC_{A \to B} = E\left[ {\log \frac{{f\left( {b_{t + 1} {|}b_{t}^{n} ,a_{t}^{m} } \right)}}{{f\left( {b_{t + 1} {|}b_{t}^{n} } \right)}}} \right]$$
(21)

Copulae at high dimensions could be recursively reduced toward low dimensions. The resulting representation ease implementation in addition to efficiency. GC representation becomes,

$$GC_{A \to B} = E\left[ {\log \frac{{f\left( {b_{t + 1} {|}b_{t}^{n} ,a_{t}^{m} } \right)}}{{f\left( {b_{t + 1} {|}b_{t}^{n} } \right)}}} \right] = E\left[ {\log \frac{{h\left( {b_{t + 1} ,{ }a_{t}^{m} {|}b_{t}^{n} } \right)}}{{f\left( {b_{t + 1} {|}b_{t}^{n} } \right) \times {\text{ g}}({\text{a}}_{{\text{t}}}^{{\text{m}}} |{\text{b}}_{{\text{t}}}^{{\text{n}}} )}}} \right]$$
(22)

where marginal density of A and B are given with g and f, conditional joint density of (A, B) with h. The conditional joint density becomes,

$$h\left( {b_{t + 1} ,{ }a_{t}^{m} {|}b_{t}^{n} } \right) = f\left( {b_{t + 1} {|}b_{t}^{n} } \right) \times g\left( {a_{t}^{m} {|}b_{t}^{n} } \right) \times c(u, v|b_{t}^{n} )$$
(23)

here \(u = F(b_{t + 1} |b_{t}^{n} )\) is the conditional marginal distribution of B, \(v = G(a_{t}^{m} |b_{t}^{n} )\) is for A, and c is the copula density function. After substitution of Eqs. 20 and 21, copula-based GC causality from A to B (\(CGC_{A \to B}\)) is calculated as,

$$CGC_{A \to B} = E\left[ {\log c\left( {F\left( {b_{t + 1} {|}b_{t}^{n} } \right), G\left( {a_{t}^{m} {|}b_{t}^{n} } \right){|}b_{t}^{n} } \right)} \right]$$
(24)

The approach allows testing GC under regime dependence with copulae. In cases that deviate from Gaussian distributions, copula-based GC leads to better properties without assuming normality (Lee & Yang 2014). The MS-GARCH-copula-causality method above provides regime-dependent testing of GC under different tails and regimes (Bildirici et al. 2022).

Empirical results

Data

The dataset in this study consists of three variables for FinTech, EC, and CO2e variables and our daily sample covers 02 Jan. 2012–28 December 2022. The data source for the FinTech and EC is the Refinitiv Eikon database. For the FinTech, the STOXX Global FinTech index is selected due to being the longest dataset available with the largest coverage. The index includes companies leveraging FinTech to revolutionize how financial services reach consumers and enhance the competitiveness of traditional providers. With ongoing FinTech development and increased support from governments and regulators, these companies are well-placed to capitalize on the persistent FinTech trend, potentially significantly impacting their future revenues. The FinTech index in the analysis is denoted as FIN. To achieve daily EC data, we utilized daily EC in TWh, which is annotated as EC, and the data are obtained from Datastream. Atmospheric CO2e is obtained from the CO2 Earth Database, and these data measure CO2e in ppm and are considered as a measure of CO2e concentrations in the global atmosphere. Throughout the analysis, the CO2e concentration series is annotated as CO2e.

Basic statistics for the distributional properties of the dataset are represented in Table 2. Series in levels are subject to natural logarithms to remove the effects of units and to reduce factors such as excess skewness. The dataset is further first-differenced. Due to logarithm rules, after first differencing, differenced series represent daily percentage changes. The first-differencing is a result of the unit roots in the series in levels as will be seen in Table 3 and is conducted to attain stationarity of data.

Table 2 Descriptive statistics
Full size table
Table 3 Tests for unit roots and homoskedasticity
Full size table

In Table 2, variables, especially those in first differences, are subject to excess kurtosis, combined with skewness, signs of heavy-tailed and leptokurtic distributions for FIN and EC, and CO2e series. These also translate into JB tests, which indicate non-normality for all series.

Table 3 displays the outcomes of unit root tests and two tests were conducted: the conventional linear augmented Dickey–Fuller (ADF) test and the KSS nonlinear unit root test (Kapetanios et al. 2003). Unlike ADF, KSS is a unit root test under nonlinearity since it tests the unit root against a smooth transition autoregressive (STAR) alternative. Both linear and nonlinear test results indicate I (1) processes for all level series and in the next section, analyses are maintained with the first differenced variables.

The second part of Table 3 also reports ARCH–LM and White homoskedasticity tests. The former test points at ARCH-type heteroskedasticity and the latter confirms heteroskedasticity with nonlinear forms. Both tests point to the requirement of modeling FIN, EC, and CO2e series with models considering heteroskedasticity.

Table 4 BDS test
Full size table

Further, we tested these series with BDS nonlinearity tests in Table 4, which indicate dependence and nonlinearity. Overall results are given in Tables 3 and 4 lead to the modeling of FIN, EC, and CO2e with models that utilize nonlinear and heteroskedastic characteristics of the dataset, and the results provide the basis for modeling with MS-GARCH-copula modeling.

Model estimation results

The models in this section benefit from regime-switching copula for contagion and tail dependence followed by regime-switching GARCH-copula-based GC tests. In the first stage, the model estimations are presented and the results are given in Table 5, consisting of two sections, parameter estimates for the models with regime transition results followed by regime-dependent copula parameter estimates for tail-dependence and contagion modelling. Diagnostic tests are given in the last column.

Table 5 MS-GARCH-copula model results
Full size table

For all marginal distributions modeled with MS-GARCH processes, parameter estimates of ARCH and GARCH terms provide information on volatility dynamics. Further, each model distinguishes between regime-specific parameters, confirming asymmetry. Overall, the parameter estimates for FIN, EC, and CO2e are statistically significant in both regimes at 5% significance level. The stability condition entails ARCH + GARCH < 1, which is achieved for each regime of each model. However, the sum is close to 1. Accordingly, external shocks will have long-lasting effects on each series in each regime. Lastly, the remaining ARCH-type heteroskedasticity tests confirm the capability of all models in capturing the volatility dynamics, since no remaining ARCH effects are present in the residuals. The RMSE, MAE, MAPE, and LogL statistics confirm the effectiveness of the models estimated in terms of goodness of fit and one-step-ahead predictions.

The transition probability, P(st|st-1), denotes the probability of the regime at t conditional on the regime at t-1. As a typical, P(1|1) indicates the probability of regime 1 at period t conditional on the previous state being also regime 1. P(1|1) and P(2|2) provide vital information for regime persistence. Considering relatively lower standard error estimates in regime 1 relative to regime 2, regimes 1 and 2 are low and high volatility regimes, respectively. For CO2e, EC, and FIN, regime-switching probability estimates are p(1|1) < p(2|2), suggesting relatively higher persistence and duration in the second regimes characterized by high volatility. However, all regimes are long-lasting since p(1|1) > 0.7. Therefore, first regimes are also persistent. For CO2e, the average duration is 11.1 days (8.3 days) in regimes 1 and 2. For FIN, for the high and low variance regimes, the duration is 16.67 days and 3.7 days. For EC, the duration of the two regimes is close to those obtained for FIN. Hence, results clearly show distinct characteristics of each regime in terms of persistence and regime duration.

In the final part of Table 5, estimation results for copula parameters are reported. The copula results are significant tools for determining the co-movements between the variables. The copula parameters are significant at 5% and 1% levels of significance. The positivity and the size of copula estimates confirm strong and positive tail dependence among the series analyzed in each regime with varying magnitudes, confirming contagion in all cases. Accordingly, findings approve strong and positive tail dependence between EC and CO2e in each regime, while the size of tail dependence varies between regimes. The results point at contagion between EC and FIN and between CO2e and FIN and EC and CO2e in all regimes and highlight asymmetric and regime-dependent contagion dynamics.

Nonlinear causality results

Following the estimation of the marginal distributions of the series with MS-GARCH-copula, regime-switching GC tests are performed. The results are given in Table 6.

Table 6 MS-GARCH-copula-causality test results
Full size table

GC test method allows evaluation of regime-dependent GC relations and their directions. Application of causality tests is necessary to determine the direction of causality, which provides vital information for policy applications. Further, results are compared with the linear counterpart approach. In the first row, the results signify evidence of unidirectional causality from EC to CO2e in both regimes. The causal links between FIN and CO2e are unidirectional, from FIN to CO2e in both regimes. Therefore, indistinct from the type of regime, unidirectional causality cannot be rejected from FIN to emissions. Regime-specific causality test results between EC and FIN are in the last section. The findings indicate causality from FIN to EC and also from EC to FIN at conventional significance levels. As a result, bidirectional causality amid two series cannot be rejected.

For comparative purposes, Table 7 reports the GARCH-copula-causality test results. In contrast to the results in Table 6, GARCH-Copula-Causality results omit regime switching type nonlinearity, and given the nonlinear structure in the series and their volatility, these single-regime results are produced to point to possible deviations due to omitting such factors. Under this deficiency, the overall look results in a tendency to accept bidirectional causality amid the series at conventional significance levels for all tested pairs of variables.

Table 7 Single regime GARCH-copula-causality test results
Full size table

Table 8 summarizes the GC test results and possible deviations under two different approaches. While the MS-GARCH-copula-causality results point to the univariate direction of causality in two out of three tests, the single-regime causality results indicate bidirectional causal links in all three. Given the nonlinearity of data, omitting nonlinear regime-switching dynamics would likely affect the causality test results. Under such conditions, MS-GARCH-copula-causality results provide more efficient findings for the connections among FinTech markets, EC, and CO2e.

Table 8 Comparison of causality results
Full size table

Discussion

The findings provide vital information for policymakers in terms of the association of FinTech, EC, and CO2e. According to our results, the tail dependence between EC, FinTech, and CO2e was determined to be strong and positive for both low-volatility and high-volatility regimes. The results exhibited deviations from the findings that ignore nonlinearity in the conditional variance processes of the series analyzed. Considering the BDS and ARCH–LM tests, nonlinearity and volatility cannot be rejected for the EC, FinTech, and CO2e series. MS-GARCH-Copula and RSGCC methods allowed regime-specific modeling and testing of the co-movements and causality under different regimes governing Markov-switching type nonlinearity. Though this was the case, the GARCH-copula-causality results given above omitted the regime-switching type nonlinearity, which led to evidence of deviations in the results. The results from this single regime variant could be characterized as over-acceptance of causality due to bidirectional causalities this method determines in all cases. Considering the nonlinear structure in the series and their volatility, the relationship between FinTech and CO2e is unidirectional in both regimes from FinTech to CO2e confirming that FinTech influences CO2e under all scenarios, i.e., low or high volatility regimes. The research also suggests causality from FinTech to EC and from EC to FinTech at statistically significant levels, in a bidirectional setting, which is also known as the feedback effect—the EC and FIN causality relations tend to lead to self-fulfilling cycles.

Market dynamics and transaction processes in FinTech also have significant negative effects on environmental sustainability. The coefficient estimates of copulae specify that FinTech is associated positively with CO2e, and the opposite is not true. Combined with the unidirectional causality links under distinct regimes, the results designate no feedback effects between FinTech and CO2e. Regarding the discussion above, the technology behind FinTech and the serious electricity-use and irreversible waste of e-waste problems require attention on mining and transaction policies.

Then again, the financial technologies and FinTech development relate to two of the central SDG goals, SDG12 and SDG7, in addition to being related to COP27 and COP28. For SDG, the first is, SDG12, which concentrates on clean technology and green innovation, and SDG7, a goal specifically focusing on clean energy technology and investments towards green energies to achieve the mitigation of CO2e. During COP27, stakeholders stressed the pressing need to pursue a low-carbon economy, which is estimated to necessitate an annual investment of nearly $6 trillion (UNFCCC 2022). COP28 served as a platform where global consensus is sought to tackle the climate crisis, aiming to cap the global temperature rise at 1.5 degrees Celsius, support vulnerable communities in adapting to climate change impacts, and attain net-zero emissions by 2050.

With a primary focus on increasing the proportion of renewable energy, some countries are adopting supportive policies to encourage innovation. For instance, the United States allocates an annual investment of $1.3 trillion in the green economy, resulting in the creation of 9.5 million full-time jobs (Georgeson and Maslin 2019). In 2020, the European Union set a determined target of achieving carbon neutrality by 2050 with the “Green Deal” policy initiative and progress has been achieved over the last years. China, the top polluter of CO2e compared to the USA and the EU, also shined in COP28, in addition China’s double carbon and low-carbon-city intervention policies are among the policies to reduce EC particularly from fossil-fuels (Jiang et al. 2022; Yang et al. 2023).

As the commitment to cleaner energies is not adequate, combined with the EC of FinTech, it is natural to expect that this new technology will hinder the efforts made on environmental sustainability. A recent empirical projection study stressed that the commitment to clean energy technology is far from being adequate to achieve the SDGs in 2050, even if all countries were to be fully committed (Fekete et al. 2021). However, it is also argued that the technological advancements in FinTech could also help make the world a better place if the greening of FinTech is achieved (Kabaklarlı 2022; Lagna and Ravishankar 2022). Our findings confirm the negative effects of FinTech stock market size on the environment coupled with positive associations with EC. Similar to FinTech, recent literature questioned the sustainability of Metaverse (Pekow 2023) and recent evidence confirmed the influence of Metaverse on the CO2e concentrations at the global level (Bildirici et al. 2024). The results of our study highlight strong commitment needed to reverse the negative effects of FinTech, that occurs through its extensive EC. The reversal of the effect of FinTech had been shown to require cleaner energy policies, at which, FinTech could be used to encourage the energy transition (Croutzet and Dabbous 2021). Given the achievement of the reduction of EC of FinTech is not feasible as the sector continues to grow, policies should aim to include specific targets for financial technologies and markets in line with SDG and COP28. Furthermore, the shift of industries to industry 4.0 also inclined the demand for FinTech and technologies including artificial intelligence, robotics, and blockchain (Bildirici et al. 2023). Energy policies favoring renewable energy and energy efficiency are expected to help in the achievement of the specific SDG goals for cleaner energy and cleaner industrial production within the Industry 4.0 revolution (Bildirici and Ersin 2023). Eco-innovation, green energy and energy efficiency as well as reduction of dependence to fossil-fuels-based EC are among policy recommendations at the national level (Adebayo 2022; Liu et al. 2023). Policies with these aspects could help support the reduction of the environmental effects of FinTech only partially, unless steps were not taken particularly focusing on FinTech.

The discussion and the findings of our study suggests that FinTech has not been successful in promoting environmental sustainability so far in the context of FinTech markets. A set of literature suggests that FinTech could help cope with environmental degradation by increasing energy efficiency, green investments to reduce GHG including CO2e (Tao et al. 2022). However, unless green algorithms and methods were not developed, it would not be possible to reverse the environmental degradation from cryptocurrencies, another pillar of FinTech (Gundaboina et al. 2022).

The relative size if FinTech industry could be criticized as being small, compared to major industries, with major contributions to environmental degradation. By focusing predominantly on the effects of FinTech in this paper, the findings suggest FinTech to be included among the policies focusing on cleaner technologies and environmental sustainability. The small industry critique is easily refutable, given the great size of the finance sector at the global economy, at which, FinTech plays a particular role. Further, the paramount role of the finance sector will continue in the future and the inclining trend of FinTech will continue with newer innovations for FinTech. The findings advocate policies to focus on the greening of FinTech to mitigate the EC of FinTech and to reverse the adverse effects on the environment.

Conclusion

The article aimed to investigate the connections among FinTech, EC, and CO2e, focusing on tail reliance, contagion, and causality using daily data spanning from June 18, 2012, to December 22, 2022. Given the ongoing upward trajectory of FinTech markets, the exploration of the nexus between FinTech, EC and CO2e is of crucial importance. The empirical findings of the study revealed compelling evidence with the associations among the variables and the examination requires the nonlinear and heteroskedastic features of the series analyzed. After confirming nonlinearity and nonlinear unit root processes with BDS and KSS tests, variables are modeled effectively with nonlinear MS-GARCH-copula and RSGCC causality approaches with Markov-switching type nonlinear forms and tail dependence at extreme levels of the dataset for contagion and causality modeling.

The findings showed that there are two distinct regimes, one with low volatility and the other with high volatility, both regimes being subject to strong levels of contagion at tails of data. This finding holds for both regimes, and at the second regime, a relatively more persistent regime with high volatility, the contagion parameters are estimated with relatively higher values. However, the low variance regime also is subject to very high persistence, also pointing to very high levels of contagion, which closely follow those obtained for the high volatility regime. Copula parameters defining the tail dependency are asymmetric and statistically significant, estimated at a range that exceeds 0.70 and 0.80 in both regimes confirming the strong level of tail dependence and contagion. Following the contagion analysis with MS-GARCH-Copula, RSGCC tests are employed to examine nonlinear Granger-causality under each regime. The RSGCC method produces robust results under nonlinearity and heteroskedasticity, in addition to tail dependence captured with copula, the three features of the data analyzed. Novel nonlinear causality tests confirmed unidirectional causal links from FinTech to EC and from EC to CO2e in both regimes in addition to bidirectional causality between EC and FinTech indicating feedback effects that particularly amplify the association between FinTech and EC in both regimes. These findings and the directions were compared to the results obtained from a single-regime GARCH-Copula causality test, for deviations of the results and in the end, the policy recommendation, if one ignores the nonlinearity in the volatility dynamics. The comparison yielded noteworthy differences and the single-regime approach has been observed to over accept causality by pointing at bidirectional causality in all tested variable pairs. The discussion section provided detailed discussion of such results if nonlinearity was to be ignored.

The empirical results led to important outcomes for the policymakers and decision makers, Accordingly, policymakers should not ignore the significant contagion and causal relationships between FinTech, EC, and environmental pollution, FinTech markets impact EC, which in turn, influences CO2e and these relationships persist across different regimes, becoming particularly evident at extreme market levels at the tails of data. Such levels are under the influence of major changes in the market driven by the major spikes due to chief FinTech innovation, which amplifies the energy demand for FinTech. While acknowledging central variables such as the renewable EC or energy efficiency in important sectors, the study shows that a comprehensive framework is necessary to include specific factors such as the FinTech markets and its technologies in addition to cryptocurrencies as they appear as another pillar of FinTech. In sum, the study emphasizes the need for future policies to address these issues for effective pollution mitigation strategies.

Policy recommendations based on the findings suggest governments regulate the FinTech market to address environmental pollution. Key proposals include:

  1. (i)

    Prioritizing efficient payment systems with a focus on greening and clean technologies to reduce EC.

  2. (ii)

    Utilizing blockchain and smart contracts to facilitate investments in renewable energy and eco-friendly initiatives.

  3. (iii)

    Promoting awareness among customers about their carbon footprint during the transition to FinTech.

  4. (iv)

    Providing loans and incentives to investors in energy efficiency and clean energy to decrease EC and CO2e.

  5. (v)

    Enhancing Risk Assessment and Insurance (RAI) models through FinTech for better management of climate-related risks.

FinTech could also have many positive sides such as fostering financial inclusion in underserved areas and such areas should also cover its extensive energy use and the effects on the environmental degradation. Finance sector has a significant size among all sectors in the globe and the commitment to green FinTech will be influential on its environmental effects. Therefore, if environmentally concerned policies were to be developed, FinTech could play a vital role in promoting cleaner energy through many channels including energy efficiency, facilitation of sustainable investments, generation of data-driven insights for awareness regarding the ecological footprint of FinTech, and supporting green initiatives across all sectors. These policies could contribute to the global fight against climate change and the quest for sustainable development in the future.

The study has provided pivotal findings regarding the contagion and causality dynamics among FinTech, energy consumption and carbon dioxide emissions. The study also has certain limitations. The data availability is a central issue for FinTech. To solve it, the paper proposes a consensus of methods that capture tail dynamics for data from FinTech stock markets. The FinTech index used is argued to have a global coverage of FinTech companies. However, it consists of Fintech companies, whose stocks are publicly traded. The selected FinTech index includes major players throughout the World, which influence the fluctuations and the major shifts in the FinTech and new FinTech innovation and such fluctuations cannot be captured adequately with yearly data at global level. As more datasets on FinTech become available, future studies are encouraged to extend the analysis with the new data. Further, future studies are advised to focus specifically on the environmental effects of the different pillars of FinTech including blockchain and cryptocurrencies, crowdfunding, and digital currencies with the proposed methods put forth in this study.