Global Natural Gas Market Integration: The Role of LNG Trade and Infrastructure Constraints

Farag, Markos; Jeddi, Samir; Kopp, Jan Hendrik Article — Published Version Global Natural Gas Market Integration: The Role of LNG Trade and Infrastructure Constraints The World Economy Provided in Cooperation with: John Wiley & Sons Suggested Citation: Farag, Markos; Jeddi, Samir; Kopp, Jan Hendrik (2025) : Global Natural Gas Market Integration: The Role of LNG Trade and Infrastructure Constraints, The World Economy, ISSN 1467-9701, Wiley, Hoboken, NJ, Vol. 48, Iss. 6, pp. 1405-1417, https://doi.org/10.1111/twec.13699 This Version is available at: https://hdl.handle.net/10419/323789 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/4.0/ The World Economy, 2025; 48:1405–1417 https://doi.org/10.1111/twec.13699 1405 The World Economy ORIGINAL ARTICLE OPEN ACCESS Global Natural Gas Market Integration: The Role of LNG Trade and Infrastructure Constraints MarkosFarag1 | SamirJeddi2 | JanHendrikKopp2 1Faculty of Management, Economics and Social Sciences, University of Cologne, Cologne, Germany | 2Institute of Energy Economics, University of Cologne, Cologne,Germany Correspondence: Markos Farag ([email protected]) Received: 17 September 2024 | Accepted: 9 January 2025 Funding: The authors received no specific funding for this work. Keywords: arbitrage| LNG trade| market integration| natural gas price| supply disruptions ABSTRACT This paper analyses the integration of global natural gas markets across North America, Europe, and Asia from 2016 to 2022. The analysis focuses on the impact of the United States emerging as a major liquefied natural gas (LNG) exporter and significant supply disruptions, including the sharp reduction in Russian pipeline supplies to Europe. We identify a structural break on 1 October 2021, coinciding with these supply disruptions and a tightening global LNG market. Using both linear and nonlinear cointegration techniques, we assess price convergence across the three regions in two subsamples: before and after the break. In the first subsample, we find strong integration between all three regional gas markets, driven by growing LNG trade and shared exposure to global spot market dynamics. However, in the second subsample, the degree of integration between the Asian and European markets weakens, with US prices decoupling from both. Granger causality analysis reveals that LNG infrastructure congestion, particularly in the US and Northwest Europe, significantly drives the widening price spreads between the US and European markets. These findings suggest that physical infrastructure plays a central role in energy market integration, especially during periods of tight market conditions, where infrastructure bottlenecks limit arbitrage opportunities. JEL Classification: Q37, Q41, F14, C32, L95 1 | Introduction International trade in natural gas has traditionally been divided into three main regional markets: Asia, Europe, and North America (Melamid 1994; Economides and Wood 2009; Geng etal.2014). Historically, this segmentation has been driven by limited liquefied natural gas (LNG) transport capacity. However, the literature suggests that these markets are gradually becoming more integrated (Neumann2009; Li etal.2014). Market integration refers to the extent to which regional markets share information and align prices (McNew and Fackler1997; Fackler and Goodwin2001). Investigating this phenomenon has significant implications for supply security, as market participants in one region must increasingly consider conditions in other regions to ensure their own supply. The integration process among the three regional gas markets has been driven by several key factors. First, some regions have experienced surplus natural gas production, while others have seen increasing consumption.1 This imbalance has necessitated expanding the international gas trade, with LNG emerging as a critical solution. Increasing export capacities and the growth of the LNG fleet have significantly improved the technical and economic feasibility of interregional trade (Barnes and Bosworth 2015; Li et al. 2014). Second, many commercial agreements have shifted from traditional This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2025 The Author(s). The World Economy published by John Wiley & Sons Ltd. 1406 The World Economy, 2025 oilindexed pricing in longterm contracts to greater reliance on hubbased pricing. For example, the share of GasonGas (GoG) competition2 in global gas consumption rose from 31% in 2005 to 49% in 2021, while oil indexation declined from 24% to 19% over the same period (IGU2021). The literature also suggests that the relationship between oil and natural gas prices has become more volatile, indicating a decoupling of the two commodities (Chiappini etal.2019; Neumann2009). At the same time, the GoG competition has seen a rise in spot and shortterm transactions, where shifts in regional supply and demand prompt LNG exporters to redirect spot volumes (IGU2021). These developments have increased market liquidity, enhanced opportunities for spatial arbitrage, and boosted the presence of physical traders. Third, advancements in shale gas exploration technology have fuelled a rapid increase in production in North America, commonly referred to as the shale gas revolution. As a result, the United States began exporting LNG in 2016 and has quickly become a major player in the global market, with export capacities expanding year over year (Melikoglu2014; Wiggins and Etienne2017).3 Finally, European and Asian countries have adopted supply diversification strategies that combine pipelines and LNG imports to mitigate supply risks (Farag and Zaki2024; Ritz2019; Hinchey2018). Several studies have focused on global gas market integration, primarily relying on price data to measure the degree of integration. The hypothesis is that greater convergence between gas prices signifies stronger spatial arbitrage and higher levels of market integration. The most commonly used methodological approach to test this hypothesis is the cointegration technique,4 which examines the existence of a longrun relationship between prices.5 Siliverstovs et al. (2005) investigated the integration of the North American, European, and Asian gas markets using monthly prices from November 1993 to March 2004. Their cointegration analysis provided evidence of integration between the Asian and European markets, while the North American market remained decoupled. The authors explain that the European and Asian natural gas markets are integrated due to similar longterm contracts and oilindexed pricing mechanisms, which align price movements in these regions. In contrast, the North American market operates under a different, more competitive pricing system that decouples it from the oillinked European and Asian markets, resulting in a lack of integration across the Atlantic. A similar conclusion was reached by Li etal.(2014), who examined the integration of international natural gas markets across North America, Europe, (a) and Asia from 1997 to 2011, using a convergence test and Kalman filter analysis.6 In contrast, Neumann(2009) found evidence of increasing integration between North American and European gas markets. Using the Kalman filter to analyse data from 1999 to 2008, Neumann observed rising price convergence, particularly after 2003. This trend was attributed to the role of LNG in linking previously segmented markets across the Atlantic during this period. However, Nick and Tischler (2014) pointed out that linear cointegration models, which assume symmetric adjustments, may be misspecified for natural gas markets where adjustments to price deviations can be asymmetric. Factors such as transaction costs and different responses to widening or narrowing spreads contribute to this asymmetry, making nonlinear cointegration a more appropriate approach. To address this, they examined the degree of integration between North American and European gas prices using a nonlinear cointegration approach that accounts for transaction costs. Their results provided strong evidence of nonlinearity in the subsamples analysed (2000– 2008 and 2009–2012). More recently, Chiappini etal.(2019) applied the momentumthreshold autoregression (MTAR) model of Enders and Siklos(2001) with daily price data from 2004 to 2018, confirming the presence of nonlinearities and asymmetries in price adjustments in the global gas market. Their analysis also shows that the degree of interdependence between the North American and European markets has increased, whereas this has not occurred between the North American and Asian markets. The reviewed literature shows that conclusions on regional gas market integration depend on the methods used and the key market mechanisms at play during the analysed period. Regarding market mechanisms, LNG trade offers more opportunities for spatial arbitrage, contributing to increased price convergence among the North American, European, and Asian markets. However, it remains unclear how recent developments in the global gas market—especially the emergence of the United States as a major LNG exporter since 2016 and the supply disruptions caused by geopolitical tensions between Europe and Russia, amid a tight LNG market—have impacted market integration. This paper contributes to the literature by analysing the integration of the global gas market from 2016 to 2022, using daily futures prices across the three main regional gas markets. The North American market is represented by the Henry Hub (HH) benchmark, the Northwest European market by the Title Transfer Facility (TTF) benchmark, and the East Asian market by the East Asian Index (EAX). This analysis is particularly relevant for two main reasons. First, this period coincides with the entry of the United States into the global LNG trade, a development that may have reshaped relationships within the global gas market. Previous research on market integration largely focused on periods when the United States was a net importer of gas. Therefore, this study provides new insights into interdependencies and cointegration under different market conditions. Second, this period has seen several factors that support arbitrage in the global gas market, particularly between the United States and the other two regions, driven by the expansion of US LNG export infrastructure and the rise in spot LNG trade7. However, it has also witnessed factors that hinder arbitrage, such as US LNG export infrastructure and European import infrastructure operating at maximum capacity. In this context, this study provides a formal statistical analysis of the price differentials between regional gas markets and identifies a structural break in these differentials on 1 October 2021. This timing aligns with significant market disruptions, such as Russia reducing gas flows to Europe and a tightening global LNG market due to supply outages and capacity constraints (Fulwood etal.2022; McWilliams et al. 2023). To capture the potential effects of these dynamics, we conduct the cointegration analysis over 1407 two subsamples, splitting the data on 1 October 2021. This timing aligns with significant market disruptions, including Russia's reduction of gas flows to Europe and the tightening global LNG market due to supply outages and capacity constraints (Fulwood etal.2022; McWilliams etal.2023). Our results show that during the first subsample (January 2016 to September 2021) the Asian and European gas prices are cointegrated. This finding is consistent with previous studies, such as Chiappini etal.(2019), which also identified cointegration between European and Asian gas markets in earlier periods. The persistence of this integration during our sample period can be attributed to the growth of LNG trade, which has facilitated arbitrage opportunities between Europe and Asia. Both regions are subject to global supply–demand dynamics and spot market pricing mechanisms, reinforcing their integration. While Chiappini etal.(2019) found no cointegration between American and Asian prices but did find cointegration between American and European prices, our analysis reveals that American prices were cointegrated with both European and Asian prices during this period. These differences likely reflect changes in the global gas market, particularly the transition of the United States to a net exporter, which has reshaped its relationship with the European and Asian markets. However, our findings are consistent with those of Nick and Tischler(2014) and Chiappini etal.(2019), supporting the conclusion that regional gas prices are nonlinearly cointegrated. This implies that adjustments towards equilibrium happen at different speeds based on the direction of deviation from the equilibrium. In the second subsample (October 2021—November 2022), we find no evidence of linear cointegration for any price pairs based on the Engle–Granger approach. However, when we apply the Enders–Siklos threshold cointegration method, we find evidence of threshold cointegration only for the EAX– TTF price pair. This suggests the presence of a nonlinear, asymmetric relationship between these two markets during the second subsample period, while the other pairs (HH– TTF and HH–EAX) do not exhibit such a relationship. The lack of cointegration between the American and European markets may be attributed to LNG infrastructure congestion during this period, which acts as a physical barrier to arbitrage.8 To further investigate the decoupling of the American and European gas markets observed in the second subsample, we examine the relationship between the HH–TTF price spread and LNG infrastructure congestion. Using the Toda and Yamamoto (1995) approach, we analyse the predictive relationship between LNG infrastructure congestion and the HH–TTF price spread. In the second subsample, we find significant Granger causality from congestion to the HH–TTF spread, suggesting that infrastructure constraints are influencing price differentials. These findings underscore the critical role of infrastructure capacity in facilitating or impeding market integration between regional gas markets. The remainder of the paper is organised as follows: Section2 outlines the conceptual background. Following this, Section3 discusses structural changes in the regional gas market, focusing on regional price patterns, LNG infrastructure utilisation, and LNG trade dynamics in Northwest Europe (NWE), East Asia, and the United States. Section3.4 details our methodology. Section5 presents the baseline results of our analysis, while Section6 provides further results. Finally, Section7 provides a discussion and conclusion. 2 | Conceptual Background The concept of market integration can be traced back to Cournot, who stated that it is ‘an entire territory, of which the parts are so united by the relations of unrestricted commerce, that prices take the same level throughout with ease and rapidity’ (Cournot 1838). Empirical studies have examined market integration along vertical (prices at different stages of the supply chain), horizontal (prices across locations) and intertemporal (spot and future market prices) dimensions, often employing cointegration methods (Ihle and von CramonTaubadel 2008; Roman and Žáková Kroupová2022). This study focuses on the horizontal dimension of market integration, which is theoretically motivated by the Enke–Samuelson–Takayama–Judge spatial equilibrium model (Enke1951; Samuelson1952; Takayama and Judge1971). In the presence of transaction costs, the condition for arbitrage can be represented as follows: where pA and pB denote prices in markets A and B, respectively, 𝜏B,A represents the transaction cost of exporting natural gas from market B to market A. Therefore, arbitrage activity may only be triggered if the implied gross profit of the trade covers transaction costs. However, this spatial equilibrium model does not account for the infrastructure constraints on arbitrage between markets. If the import infrastructure in market A and the export infrastructure in market B are fully utilised, the price difference cannot be mitigated through arbitrage. The impact of infrastructure constraints on price relationships fundamentally differs from that of transaction costs. While transaction costs represent tangible expenses incurred during trade—such as transportation and handling fees—infrastructure constraints act as physical barriers that limit arbitrage, regardless of the price differential or the associated transaction costs (Kuper and Mulder2016). This distinction is crucial, as it highlights a boundary to market integration: even if the price difference ( p A −p B) exceeds transaction costs ( 𝜏 B,A) , no arbitrage mechanism can equilibrate the markets if the infrastructure is fully utilised. The above equation is appropriate for understanding price arbitrage between the United States and Europe or the United States and Asia, where direct LNG trade occurs. The United States, being a net exporter, directly supplies LNG to both regions. However, between Asia and Europe, no significant direct LNG trade can be observed during the study period. Instead, arbitrage occurs through indirect trade via thirdparty LNG traders who reroute shipments based on market conditions. To (1) pA>pB+𝜏B,A 1408 The World Economy, 2025 represent this, let pA and pE be the prices in Asia and Europe, respectively, 𝜏S,A and 𝜏S,E be the transaction costs (including transportation) from swing supplier S to Asia and Europe, respectively, and qS,A and qS,E be the volumes exported from swing supplier S to Asia and Europe, respectively. Assuming that there are no binding infrastructure constraints at both the swing supplier and the importing regions, we have the following condition pA−𝜏S,A>pE−𝜏S,E . This means that swing supplier S will export LNG to Asia if the netback price in Asia (pA −𝜏 S,A) is greater than the netback price in Europe ( pE−𝜏S,E ). If the netback is higher in Europe, the supplier will prefer to export there. 3 | Structural Changes in the International Natural Gas Market This section provides a descriptive overview of key trends in the global gas market, focusing on regional price fluctuations, LNG terminal utilisation, and shifting trade dynamics. These elements are crucial for understanding the factors influencing market integration. By examining price patterns across North America, Europe, and East Asia, along with the impact of infrastructure constraints, this section lays the groundwork for the subsequent empirical analysis. 3.1 | Trends and Fluctuations in Regional Natural Gas Prices Figure1 shows the logarithmic prices for the Henry Hub (HH) in the North American market, the Title Transfer Facility (TTF) in the European market, and the East Asian Index (EAX) in the East Asian region.9 The figure demonstrates that there was a substantial decline in prices in March 2020, likely due to the outbreak of COVID19 and its impact on natural gas demand. This was further exacerbated by historically mild temperatures.10 Figure1 also shows that European and Asian gas prices began to rise in the second half of 2021. This can be attributed to the resurgence of demand from the industrial and heating sectors as economic activity rebounded and extreme weather events occurred. From this period onwards, it is also evident that the HH series was not significantly affected by these increases.11 3.2 | LNG Terminal Utilisation and Shifting Gas Trade Dynamics Figure2 highlights the major shifts in the natural gas market starting in October 2021 (indicated by the greyshaded area), coinciding with reduced gas flows from Russia to Europe FIGURE 1 | Natural gas prices in log level. (a) Log price of TTF, (b) Log price of EAX, and (c) Log price of HH. [Colour figure can be viewed at wileyonlinelibrary.com] FIGURE 2 | (a) Δ YoY (%) in Russian gas exports, (b) Gate terminal utilisation rates, and (c) U.S. export terminals utilisation rates. Own construction based on data obtained from ENTSOG(2023), GIE(2024), and EIA(2023). [Colour figure can be viewed at wileyonlinelibrary.com] 1409 (Henderson and Chyong 2023; Farag and Ruhnau 2024). Figure2a shows a sharp yearonyear (YoY) decline in Russian gas exports to Europe, reflecting a deliberate reduction in daily flows to the level of nominations from longterm contracts, with no additional volumes supplied to the European spot market (Fulwood etal.2022). This reduction forced Europe to increase its reliance on LNG imports, as seen in Figure2b, which depicts a notable increase in the utilisation rate of the Gate terminal in the Netherlands, the largest import terminal for LNG in Northwest Europe. The heightened demand for LNG also caused congestion at other European import terminals (GIE 2024). Simultaneously, as shown in Figure2c, the utilisation rate of US gas export terminals increased, nearing full capacity and reflecting a high level of exports. However, capacity constraints at both European import and US export terminals limited the ability to significantly increase LNG trade between the two regions. 3.3 | Diverging LNG Import Dynamics in Northwest Europe and East Asia This subsection presents the yearonyear changes in LNG imports from 2016 to 2022, with separate graphs for North West Europe, Japan and Korea (combined), and China. Figure 3a shows that LNG imports to Europe increased in the last quarter of 2021, likely driven by reduced Russian gas supplies, as discussed in the previous section. This reduction led to energy security concerns and efforts to diversify away from traditional pipeline sources (Aitken and Ersoy2023). In contrast, Figure3b indicates that the growth rate of LNG imports in China began to slow, while pipeline imports from Russia increased, indicating a potential stabilisation or shift in the energy consumption mix (Rystad Energy2023). Meanwhile, Figure3c shows that LNG imports in Japan and Korea remained relatively stable, reflecting steady demand in these mature markets (Rystad Energy2023). These varying import patterns underscore differing regional demand dynamics and suggest that Europe, Japan & Korea, and China are experiencing unique drivers and pressures in their LNG markets, likely influenced by geopolitical, economic and policyrelated factors. 3.4 | US LNG Export Dynamics to Northwest Europe and East Asia This subsection illustrates the yearonyear changes in LNG exports from the United States to North West Europe and East Asia. Figure4a shows a significant increase in US LNG exports to Northwest Europe in the last quarter of 2021, coinciding with efforts to replace Russian gas pipeline supplies. However, despite this increase, European gas prices continued to rise sharply, suggesting that existing infrastructure and market capacities were fully utilised, limiting the potential for further imports to stabilise or reduce prices. In contrast, Figure4b highlights a decrease in US LNG exports to East Asia during the same period. This decline reflects a shift in LNG trade dynamics, possibly due to changes in competitive pressures within the global LNG market. These trends underscore the evolving role of the United States as a key LNG supplier and the differing impacts on regional markets. FIGURE 3 | Δ YoY (%) in LNG imports for major regions: (a) Northwest Europe (NWE), (b) China, and (c) Japan & Korea. Own construction based on data obtained from JODI(2024). [Colour figure can be viewed at wileyonlinelibrary.com] FIGURE 4 | Δ YoY (%) in US LNG exports to: (a) Northwest Europe and (b) East Asian markets. Own construction based on data obtained from the EIA(2023). [Colour figure can be viewed at wileyonlinelibrary.com] 1410 The World Economy, 2025 4 | Methodology Our analysis examines the integration of the three regional gas markets using the cointegration approach. Before conducting this analysis, we apply the augmented Dickey–Fuller (ADF) test, the Phillips–Perron (PP) test, and the Kwiatkowski–Phillips– Schmidt–Shin (KPSS) test to the three gas price series to evaluate their stationarity properties. The ADF and PP tests examine the null hypothesis of nonstationarity (i.e., the presence of a unit root), while the KPSS test examines the null hypothesis of stationarity. If the results indicate that the price series exhibits a unit root, we proceed with cointegration analysis to examine equilibrium relationships. Previous studies have often utilised the traditional symmetric cointegration framework to analyse the integration of gas prices at both regional and intraregional levels (e.g., Siliverstovs etal.2005; Asche etal.2002). However, conventional cointegration tests may be misspecified when the adjustment process is asymmetric. The methodology proposed by Enders and Siklos(2001) extends the widely used Engle and Granger(1987) twostep cointegration procedure by incorporating asymmetric adjustments in the longrun relationships between gas prices. This extension, known as the Momentum Threshold Autoregressive (MTAR) model, has been shown to perform better in the presence of asymmetry, providing more reliable results than methods that assume symmetric price adjustments. This approach has been extensively applied to analyse asymmetric adjustment in cointegration relationships between various energy prices (e.g., Hammoudeh etal.2008; Chiappini etal.2019; Chang etal.2012). In both the symmetric Engle and Granger (1987) framework and the asymmetric Enders and Siklos(2001) extension, the first step is to estimate the following model, which represents the equilibrium relationship between two regional gas price series, using ordinary least squares (OLS): where P1 t and P2 t represent the logarithmic forms of two gas price series. We estimate three sets of gas price pairs: (TTF, EAX); (HH, TTF); and (HH, EAX). The residuals, 𝜀 t , obtained from Equation2, are subsequently used in the second step of the Engle and Granger (1987) linear cointegration analysis (Equation3) and in the second step of the MTAR model for nonlinear cointegration as proposed by Enders and Siklos(2001) (Equation4): The adjustment speed coefficients, ρ0, ρ1, and ρ2, correspond to the symmetric ( 𝜌 0) and asymmetric (ρ1 and ρ2) cointegration models. Additionally, the inclusion of lagged values of Δ𝜀 t helps to ensure that the residuals are serially uncorrelated. The Heaviside indicator function, It , is defined as 1 if Δ𝜀 t−1 ≥ 𝜏 and 0 if Δ�𝜀 t−1<𝜏 , where 𝜏 is the threshold value, estimated using the consistent search method of Chan(1993). We test for evidence of asymmetric adjustments using two hypotheses. First, we test the joint null hypothesis of nocointegration ( H0:𝜌1=𝜌2=0 ), with the critical values obtained from Enders and Siklos (2001). If the null hypothesis of nocointegration is rejected, we test for the null hypothesis of symmetry ( H0:𝜌1=𝜌2 ) using a standard Ftest. 5 | Empirical Results This section presents the empirical analysis of the integration of the three regional gas markets. The analysis is structured as follows: First, we test for a potential structural break on 1 October 2021, using the Chow test on the log price differentials between the price pairs, motivated by significant developments in the global gas market. Next, we examine the linear cointegration relationships between the gas price pairs using the Engle–Granger twostep approach. We then conduct a nonlinear cointegration analysis, employing the MTAR model to investigate potential asymmetries in price adjustments. Finally, we estimate symmetric and asymmetric error correction models to assess the shortterm dynamics and adjustments towards the longrun equilibrium for each price pair. 5.1 | Structural Break Analysis We hypothesise that the relationships among our variables of interest may be affected by a potential structural break on 1 October 2021. This break date is driven by major developments in the natural gas markets, as discussed in the previous section. For instance, in the latter half of 2021, geopolitical tensions—particularly Russia's deliberate reduction of gas exports to Europe— significantly disrupted supply. This caused a major shock in the global gas market, leading to tighter market conditions. To formally test for this break, we follow Büyük şahin etal.(2013) and Luong etal.(2019), conducting a Chow(1960) test on the log price differentials between the price pairs. We perform the test using the following specification: Here, TRt represents a linear trend, 𝜃 is a constant term, and 𝜙St−1 , 𝜙St−2 and 𝜙St−3 represent the lagged values of the dependent variable, where 𝜙1 , 𝜙2 and 𝜙3 are the coefficients on the lagged terms. The resulting Fstatistics are 17.413 (significant at the 1% level), 4.328 (significant at the 1% level), and 2.867 (significant at the 5% level) for the spreads EAX–TTF, HH–TTF and HH– EAX, respectively, with 5 and 1764 degrees of freedom. Given these significant statistics, we conclude that a structural break occurred around 1 October 2021. Consequently, the analysis is conducted over the period from 1 January 2016 to 1 November 2022, divided into two subsamples, with 1 October 2021, as the split date. (2) P1 t =𝛽 0 +𝛽 1P2 t +𝜀 t (3) Δ 𝜀 t=𝜌0𝜀 t−1+ p ∑ j = 1 𝛿jΔ𝜀 t−j+u t (4) Δ 𝜀 t=𝜌1It𝜀 t−1+𝜌2(1−It)𝜀 t−1+ K ∑ i=1 𝜗iΔ𝜀 t−i+u t (5) St=𝜃+𝜆TRt+𝜙1𝜙St−1+𝜙2𝜙St−2+𝜙3𝜙St−3+ϵ t, 1411 Summary statistics for the three price series over the two subsamples are provided in TableA1 in AppendixA, which shows a shift towards higher prices and greater variability in the gas markets after September 2021. The results of the unit root tests for the log levels and their differences are also presented in TableS2. The results show that all the time series in log levels are I (1) variables, meaning they are nonstationary in levels but become stationary after first differencing. Therefore, cointegration analysis is an appropriate tool to investigate their joint properties. 5.2 | Examining the Linear Cointegration In the context of testing for linear cointegration, we apply the twostep approach proposed by Engle and Granger(1987). This approach involves first estimating the equilibrium relationship for each price pair according to the specification in Equation2. In the second step, we obtain the residuals from this regression and apply the Engle–Granger residualbased cointegration test to determine whether the residuals are stationary.12 Table1 presents the results of the twostep analysis, with the last column providing the test statistics for the stationarity of the residuals, which indicate whether the variables are cointegrated. For the EAX–TTF pair, the estimated 𝛽1 is 0.973 in the first subsample, indicating that a 1% increase in the TTF price is associated with a 0.973% increase in the EAX price. However, 𝛽1 drops to 0.663 in the second subsample. The estimated 𝛽1 coefficients from the cointegration regressions of HH against EAX and TTF are relatively lower. In the first subsample, the estimated 𝛽1 is 0.397 for HH–TTF and 0.376 for HH–EAX. In the second subsample, the estimated 𝛽1 for HH–TTF remains stable at 0.385, while HH–EAX declines sharply from 0.376 to 0.214, indicating a weakening price linkage. The results also show that the estimated coefficient 𝛽0 varies across the three pairs (EAX–TTF, HH–TTF, and HH–EAX) and the two subsamples, with a notable increase in the second subsample. For example, 𝛽0 rises from 0.244 to 1.122 for EAX– TTF, from 0.323 to 0.404 for HH–TTF, and from 0.283 to 1.036 for HH–EAX. This increase suggests that baseline price levels across the three markets have risen over time, indicating a growing divergence in market conditions. This divergence may be due to differences in regional supply and demand balances, transportation costs, or marketspecific factors such as regulatory changes affecting natural gas pricing. The Engle–Granger test statistics in the last column of Table1 indicate that the three price pairs are cointegrated in the first subsample. However, in the second subsample, the test statistics are not statistically significant, providing no evidence of linear cointegration. 5.3 | Examining the Nonlinear Cointegration In the preceding subsection, the Engle–Granger test, which assumes a linear and symmetric adjustment process, indicates that there is no evidence of cointegration for any of the price pairs during the second period. This subsection examines estimates from the MTAR model proposed by Enders and Siklos(2001), which explicitly accounts for potential asymmetries in the adjustment process towards equilibrium. This analysis aimed to determine whether there is evidence of asymmetries in the first subsample, which would suggest that the adjustment process occurs at different speeds depending on the direction of deviations from equilibrium (positive vs. negative), rather than symmetrically. Additionally, we seek to establish whether the MTAR model provides evidence of cointegration for any of the price pairs in the second subsample period. Table 2 presents the results of the MTAR cointegration test. Column (1) shows the estimated threshold values, which indicate the point at which adjustments switch between regimes for positive and negative deviations from equilibrium. Although the estimated thresholds are close to zero, our analysis shows that models with an estimated threshold value perform better—according to the information criteria—than models assuming a fixed threshold of zero. Columns (2) and (3) show the estimated parameters of 𝜌1 and ρ2, as specified in Equation 4. Here, 𝜌1 represents the speed of adjustment in response to positive deviations from equilibrium, whereas 𝜌2 represents the speed of adjustment for negative deviations. If the absolute value of 𝜌1 is greater than that of ρ2, this indicates that the adjustment process is faster in response to positive deviations from equilibrium. Conversely, if ∣𝜌2∣ is greater, the adjustment is faster in response to negative deviations from equilibrium. For example, in the TABLE 1 | Linear cointegration analysis. Price pair Subsample β0β1R2EG (1987) EAXTTF First 0.244a[0.015] 0.973a[0.009] 0.895 −5.087a Second 1.122a[0.098] 0.663a[0.027] 0.687 −2.841 HHTTF First 0.323a[0.012] 0.397a[0.007] 0.682 −4.478a Second 0.404b[0.160] 0.382a[0.044] 0.215 −1.908 HHEAX First 0.284a[0.014] 0.376a[0.007] 0.645 −4.718a Second 1.036a[0.214] 0.212a[0.060] 0.043 −1.628 Note: The first subsample includes data from 1 January 2016 to 30 September 2021, while the second subsample includes data from 1 October 2021 to 1 November 2022. Standard errors of the estimated coefficients are given in the square brackets. The column titled ‘R2’ gives the goodness of fit for the regressions. The last column displays the Engle and Granger(1987) test statistic (EG (1987)) for cointegration, with a significant test statistic suggesting that the residuals are stationary, thus confirming cointegration between the variables. The number of lags for the Engle–Granger cointegration test was selected using the AIC. The critical values of this test are obtained from MacKinnon(2010). The symbols a and b denote significance at the 1% and 5% levels, respectively. 1412 The World Economy, 2025 relationship between EAX and TTF in the first subsample, the estimated threshold is −0.023, with adjustment coefficients of −0.026 for positive deviations and −0.120 for negative deviations. This result indicates that positive deviations from equilibrium (where Δ𝜀t−1≥−0.023 ) are eliminated at a relatively slower rate of 2.6% per day. In contrast, negative deviations from equilibrium are adjusted at a much faster rate of 12% per day. Consequently, there is substantially slower convergence towards equilibrium for positive deviations (above the threshold) than for negative deviations (below the threshold). These findings suggest that arbitrageurs are more active in exploiting larger profitable opportunities depending on the direction the spread is moving from its equilibrium position. This also implies that during the first subsample period, the market adjusts more rapidly when EAX prices are decreasing relative to TTF prices. This conclusion is consistent with Chiappini etal.(2019), although the estimated speeds of adjustment in both regimes during our sample period are higher than their estimates. In the second subsample, the estimated threshold is 0.015, with adjustment coefficients of −0.208 for positive deviations and −0.070 for negative deviations. This outcome indicates that positive deviations from equilibrium are eliminated rapidly, at a rate of 20.8% per day. The results for negative deviations do not show significant adjustment, as the coefficient for negative shocks is statistically insignificant. Column (4) in the table presents the test of the joint null hypothesis of no cointegration with MTAR adjustment ( H 0 :𝜌 1 =𝜌 2 =0 ) . The results indicate that this null hypothesis is rejected for each price pair in the first subsample, as the test statistic exceeds the critical values provided by (Enders and Siklos2001). Given this result, we proceed to test the null hypothesis of H0:𝜌1=𝜌2 . The results, shown in Column (5), indicate that this null hypothesis is rejected, supporting the presence of asymmetric adjustment. However, in the second subsample, Column (4) shows that the null hypothesis of no cointegration is rejected only for the EAX–TTF price pair. The lack of nonlinear cointegration for the HH–TTF and HH–EAX pairs suggests a decoupling of the US gas market from the European and Asian markets during this period. 5.4 | Results of the (a)symmetric Error Correction Model In this step, we estimate both symmetric and asymmetric error correction models (ECMs) to examine the adjustment processes of individual prices towards equilibrium. We estimate the symmetric or asymmetric ECM for each price pair based on the cointegration results from the previous subsection. Table3 presents the estimation results for the two subsamples.13 The magnitude of the error correction term (ECT) indicates the speed at which deviations from equilibrium are corrected. For instance, if the ECT is −0.250, it suggests that approximately 25% of the deviation is corrected each day, implying that full correction to equilibrium would take about 4 days. The results indicate that, for the EAX–TTF pair in the first subsample, the ECT for EAX in the high regime is −0.003 and statistically insignificant, suggesting no adjustment to positive deviations. In the low regime, the ECT for EAX is −0.090 and statistically significant, indicating a correction towards equilibrium for negative deviations. For TTF, the ECT is −0.020 and statistically significant in the high regime and −0.020 and statistically significant in the low regime, indicating adjustments in both cases. In the second subsample, the ECT for EAX is −0.116 and statistically TABLE 2 | Nonlinear cointegration analysis. Price pair Subsample (1) (2) (3) (4) (5) Threshold ρ1ρ2Φ (H0: ρ1 = ρ2 = 0) F (H0: ρ1 = ρ2) EAXTTF First −0.023 −0.026a−0.120b27.555b25.492b (−2.390) (−7.276) [0.000] Second 0.050 −0.208b−0.070 5.909c3.665c (−3.230) (−1.622) [0.057] HHTTF First 0.012 −0.008 −0.044b12.849b5.586b (−0.671) (−5.035) [0.018] Second −0.068 −0.020 −0.090a2.915 2.218 (−1.308) (−2.029) [0.138] HHEAX First 0.033 −0.083b−0.025b15.521b8.665b (−4.480) (−3.427) [0.003] Second 0.010 −0.002 −0.031a2.029 1.393 (−0.119) (−2.011) [0.239] Note: The first subsample includes data from 1 January 2016 to 30 September 2021, while the second subsample includes data from 1 October 2021 to 1 November 2022. Column (1) provides the estimated threshold values. Columns (2) and (3) provide the estimated coefficients in Equation4. tstatistics for the estimated coefficients are given in brackets. Column (4) shows the null hypothesis tests for the threshold cointegration with the critical values from Enders and Siklos(2001) as follows: C.V (1%) is 8.310; C.V (5%) is 6.050; C.V (10%) is 5.060. Column (5) gives the second null hypothesis. The symbols a, b, and c denote significance at the 1%, 5%, and 10% levels, respectively.