Heterogeneous effects of climate change on real GDP per capita: evidence from selected MENA countries and Türkiye

1 Introduction

Rising global temperatures associated with climate change are exerting an increasingly significant pressure on economic growth and social welfare. The effects of rising temperatures on economic output exhibit a nonlinear pattern, and it is projected that global per capita income could decline by approximately 23% by 2100 if no measures are taken to address climate change (Burke et al., 2015). Similarly, it has been empirically demonstrated that temperature shocks permanently reduce economic growth rates, particularly in poor countries, and that a one-degree Celsius increase in annual temperature reduces economic growth in these countries by approximately 1.3 percentage points (Dell et al., 2012). Long-term projections indicate that persistent deviations from historical trends in climate variables such as temperature and precipitation could negatively affect not only agricultural production but also overall macroeconomic growth dynamics through labor productivity and investment (Kahn et al., 2021).

The Middle East and North Africa (MENA) region is among the areas that will be most severely affected by this negative global economic outlook. The MENA region is already the area with the most severe water scarcity worldwide; due to low rainfall, high population growth, and a geographically concentrated demographic structure, it is extremely vulnerable to climate shocks (Waha et al., 2017). The fact that approximately 70% of agricultural production in the region relies directly on rainfall, and that the agricultural sector accounts for 22% of total employment, sets the stage for rising temperatures and drought to have adverse effects on food security and rural livelihoods (Sivakumar et al., 2013). Furthermore, climate scenarios indicate that warming in the MENA region may proceed at a faster rate than the global land average, and that extreme heat and droughts could pose significant risks to regional economies, infrastructure, and human health (Niang et al., 2014). In this context, evidence on how climate conditions relate to economic performance in the region is important both for macroeconomic stability and for sustainable development objectives. While SDG 13, part of the United Nations Sustainable Development Goals, emphasizes combating climate change and increasing resilience to climate-related risks, SDG 8 aims to support sustainable and inclusive economic growth (United Nations, 2015). Accordingly, the study is relevant to both goals and to the broader 2030 Agenda. It examines how climate conditions are related to real GDP per capita across selected MENA countries and Türkiye, and whether these relationships vary across structurally distinct country groups.

The empirical literature examining the relationship between climate change and economic growth shows that temperature shocks and persistent deviations in climate variables can have significant effects on economic performance (Dell et al., 2012; Burke et al., 2015; Kahn et al., 2021). Studies at the global level reveal that these effects may vary depending on a country’s income level, initial temperature conditions, sectoral structure, and institutional capacity. While studies on the MENA region offer important findings, differences in resource structure, sectoral vulnerability, and climate sensitivity between Gulf and non-Gulf countries have often not been addressed as a separate analytical dimension (Waha et al., 2017; Meyghani et al., 2023; Namri et al., 2026). Whether a single regional panel average adequately captures the relationship between climate change and real GDP per capita across structurally different country groups therefore remains an open question. This matters because conclusions drawn from the regional average may not apply equally across groups with different resource structures and sectoral vulnerabilities.

Although the MENA region shares similar geographical and climatic conditions, it does not present a homogeneous structure in terms of economic structure, dependence on hydrocarbon revenues, sectoral composition, and capacity to adapt to climate shocks. One of the most striking examples of this disparity is the structural difference between the Gulf Cooperation Council (GCC) member states, namely, Kuwait, Qatar, Saudi Arabia, and the United Arab Emirates, and other MENA countries. The existing literature suggests that the fiscal capacity provided by high hydrocarbon revenues may enable these countries to finance adaptation investments such as water management, desalination, and climate-resilient infrastructure; conversely, in MENA countries where agriculture accounts for a larger share of economic activity and fiscal resources are more limited, climate shocks may have a more significant impact on economic activity (Waha et al., 2017; Sivakumar et al., 2013; Namri et al., 2026). In this context, the distinction between Gulf and non-Gulf countries is treated not as a descriptive classification derived from empirical findings, but as a theoretical distinction based on structural differences discussed in the literature. This distinction aims to test the expectation that the effects of climate change on real GDP per capita may vary across country groups. Adaptation capacity, meanwhile, is not a directly measured variable in this study but rather one of the potential mechanisms invoked in interpreting the structural differences among country groups.

Against this background, this study empirically examines the relationship between climate change and real GDP per capita in selected MENA countries and Türkiye using a panel data set comprising 11 countries, namely, Türkiye, Egypt, Morocco, Tunisia, Algeria, Jordan, Iran, Saudi Arabia, the United Arab Emirates, Qatar, and Kuwait, covering the period 1995–2024.

More specifically, the study addresses two research questions: (RQ1) How, if at all, are temperature anomalies and drought conditions related to real GDP per capita, including possible nonlinearities in the temperature relationship? (RQ2) To what extent, if any, do these relationships vary between the Gulf and non-Gulf subsamples and across individual countries?

The countries were selected based on data availability, economic representativeness, and climate vulnerability. The sample includes high-income Gulf economies as well as North African and Eastern Mediterranean economies with different income levels and economic structures. Türkiye is included alongside the MENA countries because it shares the Eastern Mediterranean climate system and has strong trade and financial links with the region. The study employs second-generation panel methods that account for cross-sectional dependence and slope heterogeneity; it uses the CS-ARDL model as the baseline estimator while supporting the robustness of the findings with CCEMG and Driscoll-Kraay estimates.

The study contributes to the literature in three related ways. Conceptually, whereas earlier work on the climate-economy relationship in the MENA region has largely reported panel average effects or concentrated on specific country groups, this study incorporates the Gulf-non-Gulf distinction into the research design in advance and treats it as a structural classification grounded in differences in resource structure, income levels, and economic structure. Empirically and methodologically, the climate–real GDP per capita relationship is examined at the full-panel, predefined subgroup, and country levels, so that the regional average estimate is assessed alongside evidence obtained at lower levels of aggregation. In practical terms, the study provides a more differentiated basis for climate-related policy discussion in the region. Given the variation in direction and magnitude of country-level estimates and the important limitations of subgroup inference, policy conclusions for the MENA region should consider countries’ resource structures and sectoral vulnerabilities rather than rely solely on a single regional coefficient. This policy perspective is closely related to SDG 8 and SDG 13.

Figure 1 summarizes the conceptual framework underlying the analysis.

2 Literature review

2.1 Global climate-economy evidence

The international empirical literature examining the relationship between climate change and macroeconomic performance focuses on the asymmetric and persistent effects of weather fluctuations (particularly temperature and precipitation) on economic growth. Key studies using global datasets have shown that temperature increases have a statistically significant and negative impact on economic growth, particularly in poor countries, whereas temperature shocks do not have a significant effect in rich countries (Dell et al., 2012). Similarly, long-term global projections examining the macroeconomic losses resulting from deviations from historical norms in climate variables indicate that, under scenarios where greenhouse gas emissions are not reduced, global per capita income could decline by more than 7% by 2100 (Kahn et al., 2021). More recent evidence also points to substantial variation in climate-related economic impacts across countries and across the distribution of economic growth. Using updated climate scenarios for 174 countries, Mohaddes and Raissi (2025) estimate country-specific annual per capita GDP losses that vary considerably across countries and climate pathways. Apergis and Rehman (2025), using panel quantile estimates for 148 countries, find that temperature deviations from normal are associated with lower GDP growth at higher quantiles. A similar pattern appears in Kömürcüoğlu et al. (2025), whose MMQR estimates for the E7 countries show heterogeneous effects of a composite climate change index on economic growth across the conditional distribution. Rainfall conditions can also affect economic performance through agricultural outcomes (Deschênes and Greenstone, 2007). Using data for 1,554 subnational regions, Kotz et al. (2022) find that increases in wet days and extreme daily rainfall are associated with lower economic growth, alongside nonlinear responses to annual rainfall totals.

2.2 Nonlinear temperature effects

One of the key findings of the relevant global literature is that the relationship between temperature and economic output is nonlinear. In their study covering 166 countries, Burke et al. (2015) identified an inverted U-shaped relationship, showing that economic output reaches its peak at an average annual temperature of approximately 13 °C, while temperature increases above this threshold significantly reduce output. This nonlinear relationship is observed in both the agricultural and non-agricultural sectors and suggests that countries with already high initial temperatures may be more severely affected by further warming (Pretis et al., 2018). The form of this relationship may also differ across climatic settings. Using 44 African countries across six climatic zones, Zhao and Liu (2023) report different nonlinear temperature–output patterns across climate regimes, including inverted U-shaped and U-shaped relationships as well as regions where no significant relationship is found. These findings are consistent with current debates in climate econometrics, which argue that temperature effects may not only cause temporary losses in production levels but also have a lasting impact on economic growth rates (Newell et al., 2021; Chang et al., 2023). In addition, recent reviews of the climate−economy literature emphasize that the economic consequences of climate change should be evaluated not only through contemporaneous weather shocks but also through dynamic adjustment mechanisms, long-run damage functions, and uncertainty surrounding future climate impacts (Kolstad and Moore, 2020; Tol, 2021). This perspective supports the use of empirical frameworks that distinguish between average panel effects and heterogeneous country-level responses, particularly in structurally diverse regions such as MENA. These global findings necessitate a separate examination of the MENA region, which is one of the areas with high climate vulnerability.

2.3 MENA-specific climate-economy evidence

Developing countries are becoming more vulnerable to the economic costs of climate change due to both their geographic exposure and their low adaptive capacity, as well as their high dependence on climate-sensitive sectors such as agriculture (Abidoye and Odusola, 2015). In this context, the Middle East and North Africa (MENA) region—where extreme heat, drought events, and water stress conditions are particularly pronounced—holds special significance in regional climate studies. An updated regional assessment by Zittis et al. (2022) identifies the Eastern Mediterranean and Middle East as a prominent climate-change hotspot characterized by rapid warming and increasing risks from heat extremes and drought. The fact that approximately 70% of agricultural production in the MENA region is directly dependent on rainfall, coupled with the gradual depletion of water resources, indicates that rising temperatures could place significant pressures on food security and rural livelihoods (Waha et al., 2017). Additionally, it has been shown that extreme heat reduces labor productivity in countries across the region and negatively impacts fiscal balances by increasing budget expenditures, which could lead to increases in public debt stocks of up to 18% in the long term (Giovanis and Ozdamar, 2022).

Empirical studies focusing directly on the MENA region suggest that climate−economy relationships can differ across countries and subregions. Meyghani et al. (2023), using DOLS and FMOLS methods, found that a 1% increase in temperature reduces economic growth by 0.25%–0.28% across 14 MENA countries. For North African countries, Habib (2022), using a panel VAR model over 1980–2016, reports a negative effect of interannual meteorological variability, measured through changes in precipitation and temperature, on real GDP per capita. At the country level, Islam et al. (2021) apply an ARDL model to Saudi Arabia over 1990–2019 and report opposite long-run signs for rainfall and temperature in relation to economic growth, with rainfall negative and temperature positive. In a more recent study, Namri et al. (2026) found a nonlinear, inverted U-shaped relationship between temperature and GDP per capita for the MENA region. They highlighted structural heterogeneity within the region, noting that North Africa’s climate sensitivity has different threshold values compared to Middle Eastern countries. This evidence points to substantial variation in climate-economy relationships within the region and indicates that both temperature and rainfall conditions can matter for macroeconomic outcomes. However, the MENA region does not exhibit a homogeneous structure; there are distinct differences in climate vulnerability between the oil-rich Gulf countries and other countries in the region.

2.4 Gulf-non-Gulf and resource heterogeneity

The MENA region exhibits distinct differences in terms of economic diversity and institutional capacity. This situation differentiates climate vulnerability between the oil-exporting Gulf countries (Qatar, Kuwait, the United Arab Emirates, and Saudi Arabia) and the countries of North Africa and the Levant. The literature emphasizes that, thanks to their high per capita income levels and strong capital accumulation, Gulf countries may have the potential to finance adaptation measures against the short-term physical impacts of climate change (Waha et al., 2017). However, the fact that this high income relies heavily on oil exports leaves these countries vulnerable to transition risks—such as declining fiscal revenues—during the global transition to a low-carbon economy (Giovanis and Ozdamar, 2022; Emmanuel et al., 2024).

Cross-country heterogeneity in climate vulnerability and economic responses is also documented in recent work, although this does not in itself establish a formal difference between the Gulf and non-Gulf groups used in this study. Namdar et al. (2021), using vulnerability indicators and cluster analysis, document substantial differences across MENA countries in exposure, sensitivity, and adaptive capacity. In a wider international sample, Benhamed et al. (2023) report no substantial short- or long-run temperature effects for the full sample, while adverse effects emerge in separately estimated low–middle-income and hottest-country subsamples. Berg et al. (2024) report similar cross-country variation in real GDP per capita growth responses to temperature shocks, based on country-specific estimates.

In Gulf countries, the climate-growth relationship is shaped not only through agricultural impacts but also through the channels of “dependence on natural resource rents” and “urban heat stress.” Analyses by Namri et al. (2026) show that oil-rich countries in the Middle East are less exposed to agricultural shocks, but that institutional capacity and government effectiveness play a significant role in mitigating climate shocks. In this context, it is argued that in economies with high government effectiveness, the temperature threshold (tolerance level) at which economic growth begins to suffer could be higher. These findings suggest that the financing capacity for adaptation tools—such as cooling systems, green infrastructure investments, and water treatment facilities—in oil-rich countries could be considered a potential mechanism for limiting the economic costs of climate shocks. This structural heterogeneity is taken into account in the empirical strategy described in Section 3. More broadly, recent evidence from 33 African countries indicates that effective governance can shape sustainability transitions by supporting environmental sustainability and strengthening the environmental contribution of energy transition (Mesagan et al., 2026).

2.5 Research gap and study objectives

A general review of the existing literature reveals that, while significant empirical findings regarding the effects of climate change on economic growth have been presented, structural heterogeneities within the MENA region have not been sufficiently distinguished. In particular, differences in resource structure, income levels, sectoral vulnerabilities, and the potential to mitigate climate shocks between Gulf and non-Gulf countries make it difficult to interpret the climate-growth relationship based on a homogeneous regional average. Furthermore, when cross-sectional dependence and slope heterogeneity are not taken into account, the effects of climate variables on growth may be estimated incompletely or misleadingly. This study addresses this gap by estimating full-panel, subsample, and country-level specifications with second-generation panel methods. The analysis examines how climate variables are related to real GDP per capita and whether these relationships vary between Gulf and non-Gulf subsamples and across individual countries.

3 Data set and methodology

3.1 Data set and empirical strategy

This study empirically examines the relationship between climate change and real GDP per capita using an annual panel data set covering 11 countries in the Middle East and North Africa (MENA) region and Türkiye. The countries included in the analysis are Türkiye, Egypt, Morocco, Tunisia, Algeria, Jordan, Iran, Saudi Arabia, the United Arab Emirates (UAE), Qatar, and Kuwait. Three criteria were used in selecting the countries. First is data availability: countries such as Libya, Iraq, Yemen, and Syria do not provide reliable and uninterrupted macroeconomic data for the 1995–2024 period due to prolonged political instability and armed conflicts. Second, economic representativeness: the 11 selected countries account for a large portion of the MENA region’s total GDP; they encompass both the oil-exporting Gulf economies and the North African and Eastern Mediterranean countries with diversified economic structures. Third is climate vulnerability: all of these countries are identified in IPCC (2022) reports as regions with high climate risk. While Türkiye stands apart from the region due to its European Union accession process and diversified economic structure, it has been included in the analysis because it shares the Eastern Mediterranean climate system and maintains strong trade and financial ties with MENA countries. When assessing whether the exclusion of certain countries creates selection bias in the sample, and considering the data constraints regarding those countries and the UAE together, it is concluded that the current sample represents the region’s relatively stable economies with comparable data.

The analysis period was set as 1995–2024. Annual near-surface air temperature (TAS) data were obtained from the ERA5/CCKP dataset (Hersbach et al., 2020; World Bank, 2026), and the temperature anomaly was constructed by subtracting each country’s 1995–2024 mean annual TAS from its annual TAS value. The anomaly calculation itself involved no additional spatial aggregation or area/population weighting. The SPEI-12 data used as the drought index were obtained from the 0.5° gridded SPEIbase v2.11 database (Spanish National Research Council, 2026; Vicente-Serrano et al., 2010). For each country, monthly values from one grid cell were averaged to obtain annual values, without area or population weighting. The 12-month accumulation scale was selected because it captures persistent moisture conditions over an annual horizon, consistent with the annual frequency of the macroeconomic panel. In SPEIbase v2.11, the index is standardized over the full period covered by the dataset (1901–2024). Macroeconomic variables were obtained from the World Development Indicators database (World Bank, 2026), while capital formation and human capital variables were obtained from the Penn World Table (Feenstra et al., 2015; University of Groningen Groningen Growth and Development Centre, 2026). Since trade openness data for the UAE were not available in international databases for the period prior to 2001, the six observations from the 1995–2000 period were imputed using a linear backward extrapolation method. In this process, the level of the variable was extrapolated backward from 2000 to 1995 using the average annual rate of change calculated from the actual trade openness series for the 2001–2005 period. Since the data imputation process affected only the trade openness variable for the UAE, the complete exclusion of the UAE was used as an additional robustness check regarding the potential impact of this process on the results. For this purpose, the CCEMG model was re-estimated with the UAE excluded. The estimates obtained by excluding the UAE are reported in Appendix Table A1, and it is observed that the climate variables do not exhibit a statistically significant effect. This result is consistent with the main finding that climate variables do not exhibit a strong and consistent average effect across the panel. However, since some changes were observed in the signs and magnitudes of the coefficients, these robustness findings should be evaluated in the context of overall result consistency rather than directional consistency.

The natural logarithm of real GDP per capita (lngdppc) was used as the dependent variable. This choice follows the empirical literature, where real GDP per capita is widely used as an indicator of per capita economic performance. A log-log specification was not used because the temperature anomaly and SPEI series include zero and negative values, so they cannot be directly transformed into logarithms. The control variables were retained in their original percentage, ratio, or index form. In the baseline semi-log specification, coefficients on variables that enter linearly are interpreted as semi-elasticities. Accordingly, a one-unit change in a regressor is associated with an approximate percentage change in real GDP per capita equal to 100 times the relevant coefficient. The reported short-run and long-run coefficients refer to their respective horizons. Since temperature enters the model in both linear and squared form, its effect is evaluated using the marginal-effect expression specified below rather than from the linear temperature coefficient alone. The GDP per capita growth rate (growth) was used as an alternative dependent variable in the robustness tests to distinguish effects on the level of real GDP per capita from effects on its annual growth rate. To maintain consistency with the study’s heterogeneous-panel robustness framework, this specification was estimated using CCEMG. Given the integration properties reported in Table 1, growth and the I (0) regressors were used in levels, whereas the I (1) controls cap, lab, and open entered in first differences.

TABLE 1. CADF and CIPS panel unit root test results.

*10%, 5%, *1% significance levels. The critical values are −2.14, −2.25, and −2.45, respectively. The growth tests use the 1995–2024 sample, while the CO₂ tests use 1995–2021. For oil rents, the unit-root tests use the balanced 1995–2020 sample because the 2021 observation for Kuwait is missing. At level, the CIPS test rejects the unit-root null at the 5% level while the CADF test does not; both tests reject after first differencing.

Temperature anomaly (temp), the squared temperature anomaly (temp²), and the drought index (SPEI) were included in the model as key climate variables. The temperature anomaly (temp) measures deviations from each country’s 1995–2024 mean annual TAS. The term “temp²” was added to the model to account for the nonlinear relationship specific to the MENA region, and its coefficient will be interpreted in conjunction with the first-order term. Joint Wald tests are used to assess the null hypothesis H₀: β_temp = β_temp2 = 0. For the targeted Gulf interaction specification, the corresponding joint null is H₀: β_Gulf×temp = β_Gulf×temp2 = 0. Accordingly, the marginal effect of the temperature anomaly is calculated as ∂lnGDPpc/∂temp = β₁ + 2β₂ temp, with standard errors and confidence intervals obtained using the delta method. Gulf-non-Gulf differences in this marginal effect are evaluated at the same anomaly points, and the null hypothesis of no between-group difference at each point is tested using delta-method standard errors. In the SPEI index, positive values indicate moist conditions, while negative values indicate drought; this variable is theoretically expected to have a positive effect on real GDP per capita.

The share of investment expenditures in GDP (cap), the labor force participation rate (lab), trade openness (open), and human capital (hum) were used as control variables. For the capital and human capital variables, which had missing observations in the WDI, the csh_i and hc series from the PWT database were used. Oil rent (oil) and CO₂ emissions (CO₂), however, were not included in the main model but were used in robustness tests to assess sensitivity to potential omitted variable problems. Due to missing observations for the oil rent variable in recent years, robustness analyses including oil rent were limited to the 1995–2021 period. Summary information on the variables used in the study is presented in Table 2.

TABLE 2. Information on variables.

The basic empirical model established in this study to examine the relationship between climate change and macroeconomic variables and real GDP per capita is presented in Equation 1.

l n g d p p c i t = α i + β 1 t e m p i t + β 2 t e m p i t 2 + β 3 s p e i i t + β 4 c a p i t + β 5 l a b i t + β 6 o p e n i t + β 7 h u m i t + ε i t (1)

For the purpose of robustness analysis, the extended model, which includes the oil rent and CO₂ emissions variables, is presented in Equation 2.

l n g d p p c i t = α i + β 1 t e m p i t + β 2 t e m p i t 2 + β 3 s p e i i t + β 4 c a p i t + β 5 l a b i t + β 6 o p e n i t + β 7 h u m i t + β 8 c o 2 i t + β 9 o i l i t + ε i t (2)

Equation 2 was estimated using the 1995–2021 subsample due to missing observations in the oil rent variable; consequently, the CO₂ variable is also evaluated using the same subsample. In both equations, i represents the cross-sectional dimension (countries), t represents the time dimension (years), α i denotes country-specific unobservable fixed effects, and εit represents the error term.

For the baseline estimation, the specification in Equation 1 is represented in a dynamic CS-ARDL form. Let y i t denote the logarithm of real GDP per capita and x i t denote the vector of explanatory variables (temp, temp², spei, cap, lab, open, and hum). The estimated specification in Equation 3 is an ARDL (1,0) model:where z ¯ t = y ¯ t , x ¯ t ′ ′ contains the cross-sectional averages of the dependent variable and all explanatory variables. Thus, the specification includes one lag of the dependent variable, contemporaneous explanatory variables, and the contemporaneous and first-lagged values of the full set of cross-sectional averages.

y i t = α i + ϕ i y i , t − 1 + β i ′ x i t + ∑ l = 0 1 δ i l ′ z ¯ t − l + ε i t (3)

The corresponding short-run representation is given in Equation 4:

Δ y i t = α i + ϕ i − 1 y i , t − 1 + β i ′ x i t + ∑ l = 0 1 δ i l ′ z ¯ t − l + ε i t (4)

Defining λ i = ϕ i − 1 and θ i = β i/(1 − ϕ i, the conditional error-correction representation is given in Equation 5:

Δ y i t = α i + λ i y i , t − 1 − θ i ′ x i t + ∑ l = 0 1 δ i l ′ z ¯ t − l + ε i t (5)

Here, λ i is the country-specific adjustment coefficient and θ i contains the country-specific long-run coefficients. The reported mean-group long-run coefficients are averages of the country-specific ratios θ i; therefore, they need not equal the ratio of the mean-group short-run coefficients to the mean-group adjustment coefficient. The estimates are based on an ARDL (1,0) specification in which the cross-sectional averages enter with one lag. This structure leaves approximately four degrees of freedom per country; a longer full-CSA lag structure is not feasible in this sample. The choice of estimators was based on the diagnostic results reported in Sections 4.3–4.5. The findings indicated cross-sectional dependence, rejection of slope homogeneity (Pesaran and Yamagata, 2008), and a mixed I (0)/I (1) integration structure, with no I (2) variables. Given these characteristics, CS-ARDL was used as the baseline estimator because it allows for heterogeneous slopes, accounts for common factors through cross-sectional averages, and can accommodate the mixture of I (0) and I (1) variables while estimating short-run dynamics and conditional long-run coefficients (Chudik and Pesaran, 2015). This is also consistent with the study’s aim of examining differences in climate responses across countries and country groups. CCEMG was used as a complementary heterogeneous estimator that also controls for common factors, while Driscoll-Kraay fixed-effects (DK-FE) was included as an additional robustness check. Since DK-FE imposes homogeneous slopes, it was not treated as the baseline estimator or used to draw conclusions about country-specific effects (Eberhardt and Teal, 2010). Related evidence on long-run climate impacts in emerging economies is provided by Kumar and Maiti (2024). The relatively small cross-sectional dimension of the full sample (N = 11), and especially of the Gulf (N = 4) and non-Gulf (N = 7) subsamples, limits the reliability of mean-group inference. These estimators are nevertheless retained because slope homogeneity is rejected and cross-sectional dependence is present in the panel. The subgroup results are therefore interpreted with caution, with the Gulf estimates treated as exploratory because residual cross-sectional dependence remains in that subsample (Table 3).

TABLE 3. CCEMG subsample analysis: Gulf and non-Gulf countries.

The symbols *, , and * denote significance levels of 10%, 5%, and 1%, respectively. Since the Gulf subsample consists of only four countries and cross-sectional dependence persists in this group, the coefficients in the Gulf column are reported for exploratory and descriptive purposes. Joint Wald tests of H₀: β_temp = β_temp2 = 0 are reported for the CCEMG specifications in the text below. Differences in statistical significance across the subgroup estimates should not be interpreted as evidence of a statistically significant Gulf-non-Gulf difference. Between-group differences are assessed separately using the targeted CCE interaction tests. Because the model includes both temp and temp², the coefficient on temp is the marginal effect evaluated at temp = 0 and should not be interpreted as the general effect of a 1 °C increase.

To obtain empirical results, the following tests and estimators were used: a) descriptive statistics; b) correlation analysis and VIF analysis; c) Pesaran (2004), Pesaran (2015) CD test, Juodis and Reese (2022) CDw and CDw + tests, and Pesaran and Xie (2021) CD* cross-sectional dependence tests; d) the Pesaran and Yamagata (2008) Delta slope homogeneity test; e) the Pesaran (2007) Cross-Sectionally Augmented IPS (CIPS) unit root test and Cross-Sectionally Augmented Dickey–Fuller (CADF) unit root tests; f) the Westerlund (2007) error-correction model (ECM) panel cointegration test, implemented using the xtwest command (Persyn and Westerlund, 2008); g) Chudik and Pesaran (2015) CS-ARDL estimator, implemented using xtdcce2 (Ditzen, 2021), Pesaran (2006) CCEMG estimator, and Driscoll and Kraay (1998) fixed-effects estimators; and h) the Chudik et al. (2016) cross-sectionally augmented distributed lag (CS-DL) estimator, used as an additional sensitivity check for the non-Gulf subsample. The CCEMG estimates were obtained using the same xtdcce2 command, specifying the cross-sectional averages of the dependent variable and all regressors through the crosssectional () option with cr_lags (1). The Driscoll-Kraay estimator was implemented using xtscc with the fe option. The resulting maximum lag was three. Empirical analyses were conducted using the Stata 17 software package. Given the mixed integration structure of the variables, the Westerlund (2007) ECM panel cointegration test was applied to the I (1) variables of the baseline model: lngdppc, cap, lab, and open. The I (0) variables temp, temp², spei, and hum were not treated as members of the cointegrating vector, and CO₂ emissions and oil rents were excluded because they enter only the robustness specifications. Because cross-sectional dependence was detected in the panel, inference was based on bootstrap p-values using 499 replications (seed = 20260831). The specification included a constant, one lag, one lead, and a Bartlett long-run variance window of three, with no deterministic trend. The null hypothesis is no cointegration. The Gt and Ga statistics are group-mean tests, whereas Pt and Pa are panel tests.

4 Results

4.1 Descriptive statistics

Descriptive statistics for the variables used in the study are presented in Table 4. Table 4 presents the mean, standard deviation, minimum, and maximum values for the panel dataset covering 11 countries and the period 1995–2024. These statistics provide preliminary information on the distribution characteristics of the variables and the levels of variability within the sample; they also highlight differences across countries and periods.

TABLE 4. Descriptive statistics.

Table 4 presents descriptive statistics for the variables used in the study. The logarithmic value of real GDP per capita (lngdppc) has a mean of 9.125, with minimum and maximum values reflecting the income heterogeneity between high-income Gulf countries and middle-income countries in the sample. Among the climate variables, the mean of the temperature anomaly (temp) is zero by construction because each country’s annual TAS series was centered on its own 1995–2024 mean. The SPEI has a sample mean of −0.739 over the study period. An examination of the control variables reveals that the labor force participation rate varies significantly, particularly between the Gulf countries and North African countries. The high variability of the oil rent (oil) and per capita CO₂ emissions (CO₂) variables reflects the structural differences between oil-exporting and oil-importing countries in the sample. The average per capita GDP growth rate (growth), used as an alternative dependent variable, is 1.29%, fluctuating between −11.86% and 24.94% over the period, reflecting differences in vulnerability among countries. Descriptive statistics reveal significant heterogeneity among the countries in the sample in terms of climate and economic indicators. This supports the use of heterogeneous panel data methods in subsequent stages.

4.2 Correlation matrix and VIF analysis

A correlation matrix and Variance Inflation Factor (VIF) analysis were conducted to examine the linear relationships among the independent variables in the model and to identify potential multicollinearity issues. While the correlation matrix reveals the bivariate relationships among the variables, the VIF analysis determines whether multicollinearity is present at a level that could affect the estimation results. The results are presented in Table 5 and Table 6, respectively.

TABLE 5. Correlation matrix.

TABLE 6. Variance inflation factor (VIF) results.

The Correlation Matrix and Variance Inflation Factor (VIF) tests were used to examine whether there was a multicollinearity problem among the independent variables in the model that could affect the estimates; the results are presented in Table 5 and Table 6. Upon examining the correlation coefficients, it is observed that the strongest relationship among the independent variables is between trade openness (open) and human capital (hum). This value is well below the critical thresholds (typically 0.80 or 0.90) that could indicate multicollinearity. In particular, the temperature anomaly (temp) is centered within country by construction, as each country’s annual TAS is expressed relative to its own 1995–2024 mean. In the sample, the correlation between temp and temp² is low at 0.150, while the VIF values are 1.54 for temp and 1.08 for temp². Across all explanatory variables, the VIF values range from 1.08 to 1.76, with an average of 1.42. These values do not indicate a serious multicollinearity concern. Overall, the correlation matrix and VIF findings indicate that there is no serious multicollinearity issue among the independent variables, allowing us to proceed with the regression analysis.

4.3 Cross-sectional dependence

In panel data analyses, the presence of common shocks or unobserved common factors across countries necessitates testing for cross-sectional dependence to ensure the reliability of the estimation results. Therefore, cross-sectional dependence among the variables was examined in this study using various tests. The results are presented in Table 7.

TABLE 7. Cross-sectional dependence test results.

The ‘+' in CDw+ denotes the power-enhanced version of the weighted CD test, while the ^‘*' in CD^* denotes the bias-corrected CD test. Values in parentheses are p-values.

The presence of cross-sectional dependence in the series included in the analysis was examined using four different tests, and the results are presented in Table 7. The standard CD statistic was statistically significant for all variables except labor force participation (lab), while the augmented CDw + statistic was statistically significant for all series. This result indicates the presence of strong cross-sectional dependence in the panel, likely related to the MENA region and Türkiye’s exposure to common climatic shocks and regional macroeconomic fluctuations. From the perspective of the study’s analytical methodology, the CD* test results developed by Pesaran and Xie (2021) are also informative for the study’s methodological choice. When unobserved common factors are filtered out using principal components, it is observed that no cross-sectional dependence remains in any variable except for lab. This finding indicates that the cross-sectional dependence in the model stems significantly from unobserved common factors. Therefore, in subsequent stages of the analysis, second-generation panel data methods that model this dependence through common factors—rather than addressing it solely at the error term level—were preferred.

4.4 Homogeneity test

In panel data models, determining whether the slope coefficients are the same across all cross-sectional units is important for selecting the appropriate estimation method. Therefore, the homogeneity of the slope coefficients was examined using the Pesaran and Yamagata (2008) Delta test, and the results are presented in Table 8.

TABLE 8. Results of the Pesaran and Yamagata (2008) slope homogeneity test.

The results of the Pesaran and Yamagata (2008) slope homogeneity test presented in Table 8 indicate that both the standard Delta and the adjusted Delta statistics are significant at the 1% level. This finding indicates that the null hypothesis assuming homogeneous slope coefficients has been rejected and that the effects of climate variables and other explanatory variables on real GDP per capita may vary across countries. When evaluated in conjunction with the cross-sectional dependence identified in the previous stage, this finding demonstrates that the use of pooled first-generation estimators based on the assumption of homogeneous coefficients would produce biased estimates. It supports the preference for second-generation panel data estimators, such as CS-ARDL and CCEMG, which allow for heterogeneous slope coefficients, in the subsequent analysis.

4.5 Unit root and panel cointegration tests

To determine the appropriate estimation method in panel data analyses, it is necessary to examine the stationarity properties of the series. In this study, the CADF and CIPS tests—which are second-generation panel unit root tests—were used, taking into account the cross-sectional dependence identified earlier. The results are presented in Table 1.

The stationarity properties of the variables included in the analysis were examined using second-generation CADF and CIPS panel unit root tests, which account for the cross-sectional dependence identified in the previous stage, and the results are presented in Table 1. The findings indicate that the results of both tests are generally consistent with each other and that a mixed integration structure exists among the series. Accordingly, the variables temperature anomaly (temp), the squared temperature anomaly (temp²), the drought index (spei), and human capital (hum) exhibit stationarity at the level. In contrast, the variables real GDP per capita (lngdppc), capital formation (cap), labor force participation rate (lab), and trade openness (open) contain unit roots at the level but become stationary when first differences are taken. Among the variables used in the robustness specifications, GDP per capita growth (growth) exhibits stationarity at the level, whereas CO₂ emissions become stationary after first differencing. For oil rents, the CADF and CIPS tests provide mixed evidence at the level, while both tests indicate stationarity after first differencing. The analysis revealed that none of the variables were stationary at the I (2) level. The coexistence of variables that become stationary at the I (0) and I (1) levels in the model supports the use of the CS-ARDL method, which controls for cross-sectional dependence and slope heterogeneity while allowing for the simultaneous modeling of series with a mixed integration structure.

To provide an independent assessment of cointegration, the Westerlund (2007) ECM panel cointegration test was applied to the I (1) variables lngdppc, cap, lab, and open. The null hypothesis of no cointegration could not be rejected by any of the four statistics: Gt = −1.083 (robust p = 0.970), Ga = −3.725 (0.846), Pt = −4.240 (0.637), and Pa = −3.199 (0.631). Because slope homogeneity is rejected in Table 8, the group-mean statistics Gt and Ga are the primary reference, and the panel statistics Pt and Pa agree with them. The test therefore does not provide independent evidence of panel cointegration, although the modest panel dimensions mean that this non-rejection should not be read as proof that no cointegrating relationship exists. The bootstrap inference accounts for cross-sectional dependence in the test, whereas the CS-ARDL specification directly incorporates cross-sectional averages in estimation. Accordingly, the error-correction term and the long-run coefficients reported below are interpreted as conditional estimates within the CS-ARDL framework rather than as independent evidence of a cointegrating relationship.

4.6 Regression analysis

Regression analyses were conducted to examine the effects of climate change and macroeconomic factors on real GDP per capita. The CS-ARDL approach, which accounts for cross-sectional dependence, slope heterogeneity, and mixed integration orders, was used as the primary estimation method; the robustness of the findings was tested using the CCEMG and Driscoll-Kraay standard-error fixed-effects (DK-FE) models. The estimation results are presented in Table 9.

TABLE 9. Panel regression estimation results (dependent variable: lngdppc).

The symbols *, , and * denote significance levels of 10%, 5%, and 1%, respectively. Values in parentheses are standard errors. CCEMG refers to Pesaran’s (2006) Common Correlated Effects Mean Group estimator, while DK-FE refers to the Driscoll-Kraay fixed-effects estimator with robust standard errors. The CS-ARDL estimates are based on an ARDL (1,0) specification, with one lag of the dependent variable, contemporaneous explanatory variables, and one lag of the full set of cross-sectional averages. The CS-ARDL and CCEMG estimations use 319 observations, whereas the DK-FE estimation uses 330. This difference reflects the lag structure of the CS-ARDL and CCEMG specifications.

The effects of climate change and macroeconomic factors on real GDP per capita were analyzed using the CS-ARDL, CCEMG, and Driscoll-Kraay (DK-FE) estimators, and the findings are presented in Table 9. In the CS-ARDL model, which serves as the baseline estimator, the error correction term (ECT) coefficient was found to be negative and statistically significant at the 1% level, indicating conditional adjustment dynamics within the estimated model. This coefficient indicates that approximately 63% of short-term deviations are corrected in the following period. However, the ECT coefficient alone is not treated as independent evidence of cointegration, and the Westerlund test reported in Section 4.5 does not supply such evidence either. The ECT and the long-run coefficients are accordingly interpreted as conditional estimates within the CS-ARDL framework. The stationarity of the key climate variables—temperature anomaly (temp), the squared temperature anomaly (temp²), and the SPEI drought index—also means that these variables are not interpreted as members of an I (1) cointegrating vector. Therefore, the study’s main findings are interpreted in conjunction with the CCEMG, Driscoll-Kraay, and subsample estimates. The residual CD statistic in the CS-ARDL estimate is significant at the 10% but not at the 5% level (p = 0.089). The result should therefore not be interpreted as evidence that residual cross-sectional dependence has been fully eliminated. In the CCEMG model, by contrast, the CD statistic is not statistically significant, indicating no evidence of remaining cross-sectional dependence at conventional levels. The DK-FE model, meanwhile, was evaluated as an additional robustness check under the homogeneous slope constraint.

When the estimation results are evaluated overall, it is observed that climate variables do not exhibit a strong and consistent effect on real GDP per capita at the panel mean level across estimators. Temperature anomaly (temp) exhibits a weak positive effect at the 10% level in the CCEMG model and a weak negative effect at the 10% level in the DK-FE model. However, neither coefficient is statistically significant at the 5% threshold and is not supported by the CS-ARDL estimates. In contrast, the squared temperature anomaly (temp²) and the drought index (spei) do not show a statistically significant effect in the baseline estimates presented in Table 9. These findings indicate that the average effect of climate variables on real GDP per capita at the full-panel level remains limited and sensitive to the estimator. In contrast, structural factors such as physical capital accumulation appear to play a more consistent explanatory role across models. Indeed, capital formation (cap) has a positive and statistically significant coefficient in the CS-ARDL long-run, CCEMG, and DK-FE estimates, representing one of the consistent and prominent findings supporting neoclassical growth theory. The absence of a strong and consistent average effect of climate variables across the panel should not be interpreted as meaning that climate shocks have no impact on real GDP per capita. On the contrary, this result may reflect significant differences among MENA countries in terms of resource structure, sectoral vulnerability, and the potential to mitigate climate shocks, and average panel coefficients may mask these heterogeneous effects.

The fact that the positive but weak coefficient in the CCEMG and the negative but weak coefficient in the DK-FE point in opposite directions indicates that the temperature effect at the full panel level is estimator-sensitive and that there is no strong average effect. Table 3 shows a positive but statistically insignificant temperature coefficient in Gulf countries and a negative and significant coefficient in non-Gulf countries.

The economic structure of the MENA region varies significantly between Gulf economies—which have high per capita hydrocarbon rents and high income levels—and other economies, which have relatively lower hydrocarbon rents and more diversified production structures. Given the fiscal space provided by high hydrocarbon revenues and the ability to finance infrastructure investments, it can be argued that the impact of climate shocks on real GDP per capita in Gulf countries may differ from that in other MENA countries. The primary criterion for forming these subgroups is not merely whether countries are oil exporters, but rather whether they share similar characteristics in terms of Gulf Cooperation Council (GCC) membership, per capita hydrocarbon rent, and economic structure. Within this framework, the Gulf group consists of Kuwait, Qatar, Saudi Arabia, and the United Arab Emirates—all of which are members of the Gulf Cooperation Council (GCC) and have notably high per capita hydrocarbon rents and income levels. Although Iran is a major oil exporter, it has been classified in the non-Gulf group because it is not a GCC member, its per capita income is lower than that of the Gulf economies, and its economic structure is more diversified. This classification does not imply that adaptation capacity is measured directly; rather, adaptation capacity is treated as one of the possible explanatory mechanisms used in interpreting the empirical findings. As a first step in examining this hypothesis, the sample was divided into two subgroups: Gulf countries (Kuwait, Qatar, Saudi Arabia, UAE) and non-Gulf countries (Algeria, Egypt, Iran, Jordan, Morocco, Tunisia, Türkiye), and the CCEMG estimator was applied to each subgroup. The results are presented in Table 3.

The results in Table 3 show different coefficient patterns across the two groups. The absence of a strong and consistent average effect of climate variables across the panel, together with the lack of consistency across estimators, may reflect structural heterogeneity stemming from the sample composition. In the non-Gulf subsample, the coefficient on temp was estimated at −0.0192 (p = 0.047). Because the specification includes temp², this coefficient is the marginal effect evaluated at temp = 0 rather than a general effect of a 1 °C increase. Marginal effects were therefore also evaluated at common points of the temperature-anomaly distribution, as reported in Table 10.

TABLE 10. CCEMG marginal effects of temperature anomaly at selected evaluation points.

Marginal effects are calculated as β₁ + 2β₂ temp using the delta method. The evaluation points are the pooled CCEMG estimation-sample p10 (−0.5833), p50 (−0.0620), and p90 (0.8290) temperature-anomaly values and lie within the common support of the Gulf and non-Gulf samples. The estimation samples contain 319, 116, and 203 observations for the full, Gulf, and non-Gulf models, respectively. Given the small Gulf subsample (N = 4) and residual cross-sectional dependence, Gulf estimates remain exploratory. These subgroup marginal effects are not formal tests of Gulf-non-Gulf differences; formal between-group inference is based on the CCE interaction tests reported later in this section.

The marginal effects in the non-Gulf subsample are negative at all three pre-specified evaluation points, but none is statistically significant at the 5% level and all corresponding 95% confidence intervals include zero. In the full sample, only the median marginal effect is positive and weakly significant at the 10% level (0.0309, p = 0.072), while the Gulf marginal effects are statistically insignificant at all three evaluation points.

Joint Wald tests were also conducted for the linear and squared temperature terms in the CCEMG specifications. The null hypothesis H₀: β_temp = β_temp2 = 0 was not rejected for the full sample (F (2,55) = 1.66, p = 0.199) or for the Gulf subsample (F (2,20) = 1.97, p = 0.165). The Gulf result remains exploratory given the small number of countries and residual cross-sectional dependence. In the non-Gulf subsample, the joint null was rejected at the 5% level, F (2,35) = 3.36, p = 0.046. This result indicates joint significance of the two temperature terms in that specification, but it should not be interpreted as evidence of statistically significant curvature: the squared temperature term itself is not statistically significant, and the marginal effects at the pre-specified p10, p50, and p90 evaluation points are not significant at the 5% level.

To formally assess whether the temperature response differs between Gulf and non-Gulf countries, a targeted CCE interaction specification was estimated using Gulf × temp and Gulf × temp² terms. In this specification, the temperature terms and their Gulf interactions were restricted to common coefficients, while the remaining regressors were allowed to remain heterogeneous. This specification is used only for the formal between-group comparison and does not replace the heterogeneous CCEMG subgroup estimates. Neither interaction term was statistically significant (Gulf × temp: coefficient = −0.0002, SE = 0.0373, F (1,73) = 0.00, p = 0.996; Gulf × temp²: coefficient = −0.0417, SE = 0.0351, F (1,73) = 1.41, p = 0.239). The joint Wald test did not reject the null that both interaction terms are zero, F (2,73) = 1.00, p = 0.372. Pointwise Gulf-non-Gulf marginal-effect differences at the pooled estimation-sample p10, p50, and p90 temperature-anomaly values were 0.0485 (p = 0.483), 0.0050 (p = 0.900), and −0.0693 (p = 0.163), respectively. All three evaluation points lie within the common support of the two groups. Thus, the formal tests do not provide evidence of a statistically significant Gulf-non-Gulf difference in the temperature-response function. Given the small Gulf subsample (N = 4), however, failure to reject the null should not be interpreted as evidence that the two response functions are identical.

As an additional sensitivity check, the non-Gulf relationship was estimated using CS-DL. The temperature terms were not jointly significant, F (2,42) = 0.75, p = 0.477, and the long-run marginal effects at the pre-specified p10, p50, and p90 anomaly values were positive and statistically insignificant. A separate growth-based specification also yielded an insignificant joint temperature test, F (2,91) = 0.06, p = 0.946, although the estimates were imprecise. Because the growth specification uses a different dependent variable, it is treated as an outcome-based sensitivity check rather than as a direct replication of the level relationship. Overall, these checks do not support treating the negative non-Gulf CCEMG result as a robust finding.

Conversely, the fact that the coefficients related to temperature variables were not statistically significant in Gulf countries does not imply that climate shocks have no effect on real GDP per capita in these countries. Since the Gulf subsample consists of only four countries and cross-sectional dependence persists within this group, the coefficients for the Gulf group should be interpreted with caution. Therefore, the findings regarding the Gulf countries should be regarded not as definitive evidence of the absence of climate effects, but as limited and exploratory results that need to be validated with larger samples and alternative modeling strategies. Given this limitation, fiscal space and adaptation capacity should be treated only as possible explanatory mechanisms rather than as explanations established by the subgroup results. Adaptation capacity was not measured directly in this study, and the formal interaction tests do not establish that Gulf economies are less climate-sensitive than non-Gulf economies.

The non-Gulf marginal effects at the three pre-specified evaluation points are negative but not statistically significant at the 5% level. The formal interaction tests also do not establish a statistically significant difference between the Gulf and non-Gulf temperature-response functions, and the non-Gulf estimate should therefore be interpreted as a subgroup finding rather than as evidence of a significant between-group difference. More broadly, the Pesaran-Yamagata Delta test and the country-level estimates provide separate evidence of slope heterogeneity across countries. From the perspective of control variables, the fact that labor force participation (lab) exhibits a statistically significant and positive effect at the 1% level in non-Gulf countries highlights the importance of demographic dynamics in the per capita economic performance process for this group of countries. The fact that control variables exhibit different levels of significance across models can be attributed to differences in the assumptions made by the estimators. While CCEMG estimates heterogeneous coefficients on a country-by-country basis and calculates their average, DK-FE assumes a common coefficient for all countries. Therefore, particularly when interpreting the long-run effects of slowly changing variables such as human capital (hum) and labor force participation (lab), the CS-ARDL and CCEMG results—which allow for heterogeneous coefficients—provide a more appropriate reference.

4.7 Country-level analysis results

Average coefficients obtained at the panel level may not always adequately reflect differences across countries. Therefore, to highlight the heterogeneity in the effects of climate variables and control variables on real GDP per capita, coefficients were estimated on a country-by-country basis using the CCEMG method. The results are presented in Table 11.

TABLE 11. Individual coefficients by country in the CCEMG model (dependent variable: lngdppc).

The symbols *, , and * indicate significance levels of 10%, 5%, and 1%, respectively, based on the z-statistics and corresponding p-values reported by xtdcce2. Country-specific coefficients are obtained from a CCEMG specification that includes contemporaneous cross-sectional averages of the dependent variable and all regressors but no lagged cross-sectional averages (330 observations; residual CD = −2.19, p = 0.028). Their simple averages therefore differ from the full-sample CCEMG estimates in Table 9, which include one lag of the cross-sectional averages.

As a complementary descriptive comparison, averaging the country-specific coefficients reported in Table 11 gives a mean temp coefficient of 0.0329 for Gulf countries and −0.0016 for non-Gulf countries; the corresponding mean temp² coefficients are −0.0050 and −0.0061. These descriptive averages are derived from the full-sample country-specific specification reported in Table 11 and are not directly comparable with the separately estimated subgroup coefficients in Table 3, which use a different cross-sectional-average and lag specification. Given the very small group sizes and the limitations of second-stage inference based on estimated country-specific coefficients, these averages are treated as descriptive only. Formal Gulf-non-Gulf inference is based on the interaction and Wald tests reported in Section 4.6. The CCEMG estimation results presented in Table 11 reveal that while climate variables do not exhibit a consistent average effect at the panel level, they can have effects of varying directions and magnitudes at the country level. The drought index (SPEI) has a positive and statistically significant coefficient in Morocco, whereas it has a negative and statistically significant coefficient in Saudi Arabia. In contrast, the squared temperature anomaly (temp²) has a negative coefficient at the 10% level in Algeria and Tunisia, suggesting a possible nonlinear relationship between temperature and real GDP per capita in these country-specific estimates. An examination of the control variables reveals that the effects of capital formation, trade openness, and human capital also differ across countries in terms of direction and significance. The fact that the SPEI coefficient is negative in Saudi Arabia does not appear, at first glance, to be fully consistent with expectations, as it suggests that humid conditions (positive SPEI) negatively affect real GDP per capita. However, this finding can be explained by the fact that the Saudi Arabian economy relies on oil exports rather than agricultural activity, and thus increases in precipitation levels do not have a direct channel through which to affect real GDP per capita.

These findings support the result of slope heterogeneity identified by the Pesaran-Yamagata Delta test and indicate that the relationship between climate change and real GDP per capita varies across the MENA region depending on country-specific economic and structural characteristics. However, even in cases where the squared temperature term (temp²) is significant in Algeria and Tunisia, the fact that the corresponding linear term (temp) remains statistically insignificant limits the reliable determination of an optimal temperature threshold (−β₁/2β₂) that could be calculated theoretically. In full-sample estimates, while the temperature anomaly is marginally significant at the 10% level in only some estimators, the squared temperature term is not statistically significant. Therefore, the lack of strong and consistent support for both the linear and squared temperature terms makes it difficult to interpret a threshold value calculated within this framework in a reliable economic sense. Furthermore, the fact that the temperature variable is defined in the study as a temperature anomaly—measuring deviation from the historical average—rather than as an absolute level (°C) also makes it difficult to interpret any calculated threshold directly in terms of temperature levels. For these reasons, no optimal temperature threshold is reported in the study.

4.8 Robustness analysis

To assess the robustness of the baseline model, three alternative specifications were estimated. The results are presented in Table 12. In the first robustness test, the per capita GDP growth rate (growth) was used as the dependent variable. In this specification, temp, temp², and SPEI were individually statistically insignificant, and the temperature terms were also jointly insignificant (F (2,143) = 1.70, p = 0.1864). In the second and third robustness tests, CO₂ emissions and oil rent variables, respectively, were added to the baseline model. In the CO₂ model, while the squared temperature anomaly (temp²) was marginally significant at the 10% level, the climate variables did not yield statistically significant results at the 5% level. In the OIL model, however, climate variables do not exhibit a statistically significant effect. These results support the interpretation that climate variables do not exhibit a strong and consistent full-panel average effect, rather than confirming exact coefficient stability across all specifications. Marginally significant climate coefficients should be interpreted with caution, as they are sensitive to the estimator and model specification.

TABLE 12. Robustness analysis results.

The symbols *, , and * indicate significance levels of 10%, 5%, and 1%, respectively. Values in parentheses are standard errors. The Growth Model was estimated using CCEMG; in this specification, cap, lab, and open enter in first differences. The CO₂ and OIL models were estimated using the 1995–2021 subsample.

5 Discussion

Across estimators, the full-panel results do not reveal a strong and consistent average relationship between the climate variables and real GDP per capita. Subgroup and country-level estimates, however, vary in direction and magnitude, suggesting that the regional average should be interpreted together with evidence at lower levels of aggregation.

The weak and estimator-sensitive full-panel result differs from MENA studies reporting clearer temperature relationships, including the negative association reported by Meyghani et al. (2023) and the nonlinear pattern identified by Namri et al. (2026). It is closer to Benhamed et al. (2023), who report no substantial temperature effect for their full sample while finding adverse effects in separately estimated subsamples, and to Berg et al. (2024), who document substantial cross-country variation in temperature responses. The full-panel result does not mean that climate conditions are unimportant. It instead suggests that a single average relationship is not stable across estimators in this sample.

In the non-Gulf subsample, the CCEMG estimates suggest a negative temperature response, but the pre-specified marginal effects are not statistically distinguishable from zero and the CS-DL sensitivity check does not reproduce the negative response. The separate growth-based specification also provides no supporting evidence, although its estimates are imprecise. The non-Gulf result is therefore interpreted as suggestive and specification-dependent rather than as a robust temperature effect. This pattern is broadly consistent with adverse climate-economy relationships reported for the wider MENA region by Meyghani et al. (2023) and, for climate variability in North Africa, by Habib (2022).

The Gulf estimates are even more tentative because the subsample contains only four countries and residual cross-sectional dependence remains. Their lack of statistical significance does not imply that climate effects are absent, and the formal interaction tests do not establish a statistically significant Gulf-non-Gulf difference. Previous studies also do not suggest a uniform response across Gulf economies. Islam et al. (2021), for example, report a positive long-run temperature association for Saudi Arabia. Differences in the capacity to finance adaptation (Waha et al., 2017) and in institutional capacity (Namri et al., 2026) may help explain some cross-country variation, but adaptation capacity is not directly measured in this study and is therefore only a possible mechanism.

At the country level, the signs and significance of individual climate coefficients vary, but many estimates are imprecise, and isolated SPEI or squared-temperature coefficients are not sufficient to identify country-specific transmission mechanisms or a common nonlinear threshold. This is consistent with Zhao and Liu (2023), who find that nonlinear temperature–output patterns differ across climatic regimes rather than following a single common form. Methodologically, the detected cross-sectional dependence and slope heterogeneity support the use of estimators that allow for common shocks and heterogeneous responses, consistent with Lanzafame (2014). The contribution of the study is to consider the regional average together with subgroup and country-level evidence, while keeping the interpretation of these differences within the limits of the formal tests.

6 Conclusion

This study examines the relationship between climate change and real GDP per capita in an 11-country panel of selected MENA countries and Türkiye from 1995 to 2024. The analysis accounts for cross-sectional dependence and slope heterogeneity using second-generation panel estimators, including CS-ARDL, CCEMG, and Driscoll-Kraay fixed effects.

The results indicate that a single regional average should be interpreted with caution when subgroup and country-level estimates are also considered. Full-panel estimates do not show a strong and consistent average relationship between the climate variables and real GDP per capita across estimators. Subgroup and country-level estimates vary in direction and magnitude, but the negative temperature pattern in the non-Gulf subsample is suggestive and specification-dependent rather than robust, and formal tests do not establish a statistically significant Gulf-non-Gulf difference. Gulf estimates remain exploratory because of the small subsample and residual cross-sectional dependence. Differences in adaptation-related capacity may help explain some of the variation across countries, but adaptation capacity is not directly measured in this study and should therefore be viewed only as a possible explanation.

Given this variation, climate policies in the MENA region should avoid one-size-fits-all approaches and take into account differences in resource structures, sectoral vulnerabilities, and climate sensitivities. For non-Gulf countries, relevant areas for policy discussion include climate-smart agricultural practices, more efficient water management, and drought-resilient infrastructure. The Gulf results do not imply that these economies are unaffected by climate shocks and should be interpreted with caution given the limited subsample and residual cross-sectional dependence. For Gulf countries, relevant areas for long-term resilience planning include urban heat stress, water security, energy demand, and the transition to a low-carbon economy. These areas reflect the structural characteristics discussed in the study; the effects of specific policy interventions are not estimated here.

These findings also have implications for environmental economic policy and climate resilience. Climate adaptation is relevant to both economic resilience and environmental protection, linking the study to the broader objectives of SDG 8 and SDG 13.

6.1 Limitations and future research

This study has several limitations. First, the exclusion of MENA countries such as Syria, Yemen, Libya, and Iraq because of prolonged instability and data gaps limits the generalizability of the findings to the entire MENA region. The findings therefore apply to the selected sample rather than to all MENA countries. Second, trade openness data for the UAE for the 1995–2000 period were imputed using linear backward extrapolation. Although this issue was examined through the UAE-exclusion sensitivity check reported in Appendix Table A1, the imputation remains a data limitation. Third, the small subgroup sizes, particularly the Gulf subsample of four countries, limit inference, and residual cross-sectional dependence remains in the Gulf group. The Gulf estimates should therefore be treated as exploratory and require confirmation in broader samples. More generally, the Gulf-non-Gulf comparison is a theoretically motivated heterogeneity exercise rather than a causal identification strategy.

Future research could re-examine these subgroup patterns using larger country samples and investigate the mechanisms discussed here using direct measures of adaptation capacity, institutional quality, and infrastructure resilience.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. The author used ChatGPT (GPT-5.6 Sol, OpenAI) for spelling and grammar checks, language editing, and reference-format checking. All AI-assisted changes were reviewed and approved by the author, who takes full responsibility for the final manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Mohammad Haseeb, Wuhan University, China

Reviewed by: Precious Muhammed Emmanuel, University of Ibadan, Nigeria

Raga Elzaki, King Faisal University, Saudi Arabia