Unveiling the Spatial Effects of Climate Change on Economic Growth: International Evidence

Abstract

Climate change represents one of the most serious threats to the well-being of humanity. In recent decades, there has been a significant increase in the amount of research dedicated to analysing the economic impacts of climate change. Nevertheless, the spatial aspect of climate change has not been addressed. This research is the first to empirically assess both direct and indirect (spillover) effects of climate change, as measured by temperature variations, on economic growth. The empirical analysis is based on a balanced dataset for 86 countries between 1980 and 2019. The preliminary analysis suggests the presence of spatial autocorrelation and the suitability of the dynamic spatial autoregressive model to assess the spillover effects of climate change. The results for the full sample reveal that there are no substantial short- or long-run effects of climate change on economic growth. When the sample is decomposed by income, the analysis indicates that climate change has direct and indirect spillover effects on economic growth only in low–middle-income countries over the short- and long run. The decomposition according to the climate regime also yields interesting findings as climate change exerts adverse direct and indirect spillover effects on economic growth only in the hottest countries over the long run. These findings are robust since they hold regardless of whether the contiguity weight matrix or the inverse distance weight matrix is used. The research advocates for international collaboration in the design and implementation of climate change mitigation and adaptation strategies.

1. Introduction

The last few decades have demonstrated rapid climate change and global warming. According to [1], the global mean temperature recorded in 2022 was 1.15 degrees Celsius higher than between 1850 and 1900. Land and ocean average temperature anomalies have been at their highest levels on record for the past decade. Ref. [2] reported that temperature anomalies in July 2016, 2019, 2020, 2017, 2021, and 2022 amounted to the hottest month of July since 1880. National, regional, multilateral, and international institutions are increasingly concerned with implementing measures to combat climate change and global warming. Despite the ongoing efforts, the international collaboration, primarily undertaken through the Conferences of the Parties convened under the UN Framework Convention on Climate Change, has not yet resulted in concrete mitigation policies and actions [3,4].

Global warming and climate change have also been the subject of intense debate among scholars. Two main strands have emerged in the academic literature. The first strand of studies has focused on factors contributing to climate change [5,6,7,8]. A second important strand of the literature has concentrated on the repercussions of climate change. For instance, an increasing number of studies have examined the effects of climate change on agricultural production [9,10,11], employment [12,13,14], energy demand [15,16,17], financial stability [18,19,20], and economic activity [21,22,23]. The impact of climate change on economic growth has particularly attracted the attention of scholars for at least two reasons. First, economic growth is the most accurate indicator of the general health of the economy. Therefore, investigating the impact of climate change on economic growth allows accounting for all previously mentioned factors, including agricultural production, employment, and financial stability. In other words, if climate change affects agricultural production, employment, or financial stability, these adverse effects will be reflected in economic growth. Consequently, a clear picture of the economic costs of climate change will be discerned when analysing the impact of climate change on economic growth. Second, the availability of relatively long time series on economic growth has spurred scholars to focus on the implications of climate change on economic growth. This is mainly due to the fact that climate change is a long-term phenomenon [24,25].

A survey of the existing literature suggests that most empirical studies applied standard panel data techniques to examine whether climate change significantly affects economic growth. Among others, the climate change–economic growth linkage was analysed by ref. [26] using the fixed effects vector decomposition, ref. [21] using the panel cointegration technique, ref. [27] using the impulse response functions, and ref. [23] using the panel autoregressive distributed lag approach. Although the employed empirical methodologies could have some merits, their common drawbacks consist of ignoring the global dimension of climate change and failing to account for spatial interactions. Previous studies did not account for cross-sectional interrelationships and assumed that common factors do not affect the relationship between climate change and economic growth. The violation of this assumption makes the estimators insufficient as the adverse effects of climate change on economic growth will be underestimated if the global dimension of climate change is not considered. Consequently, ignoring spatial interactions increases the bias of omitted variables.

The current study aims to fill this gap by assessing the effects of climate change on economic activity in a sample of 86 countries from 1980 to 2019. The dynamic spatial model allows for estimating the direct and indirect effects of climate change on economic growth. As mentioned above, studies ignoring the spatial dimension of climate change underestimate the effects of climate change since they only account for the direct effect of climate change, i.e., the effect of climate change in country i on economic growth in country i. However, climate change is a global and transnational phenomenon that exceeds the geographic borders of nations [28,29]. For instance, a country with extensive usage of fossil fuels and a lack of environmental regulations will experience an increase in greenhouse gas emissions and, as a result, global warming, which will impact the country itself and its neighbours. Furthermore, the economic and financial connections between countries ensure that the adverse effects of climate change on economic growth will be observed in the same country and its neighbours. It is worth noting that the study by [22] is the closest to ours as the authors highlighted that climate change is a global phenomenon and the need for empirical investigations to consider this issue. However, the authors did not measure the indirect effects of climate change in their empirical analysis.

The current research has some novelties when compared to earlier literature. First, the study is the pioneer that considers the effects of climate change, as measured by temperature changes, on economic growth by accounting for the spatial effects. Therefore, in addition to the direct effects, we will provide the indirect (spillover) and total effects of climate change on economic growth. For robustness checks, the empirical analysis is based on two different spatial matrices: the contiguity weight matrix and the inverse distance weight matrix. Second, the study considers a large sample of 86 countries that will be decomposed according to the climate regime and income level. According to the climate regime, two subsamples of countries are considered: the hottest and coldest. When moving to the income level, we distinguish high-income and low–middle-income countries. Such decompositions are of great importance as they allow the detection of whether climate change affects all countries in the same way or not. The disaggregation by income level is crucial as the public and academic debate has always claimed that low-income countries are more exposed to risks from climate change [28,30,31]. Third, the study computes the spatial effects of climate change on economic growth in both the short run (instantaneous) and long run. This will allow comparing the effects of climate change on economic growth over time and provide appropriate policy recommendations.

The remainder of this research is structured as follows. Section 2 reviews the related empirical literature, while Section 3 and Section 4 are reserved for the methodology and data, respectively. The empirical findings are discussed in Section 5. Finally, Section 6 concludes the research and provides some policy recommendations.

2. Related Literature

The inverse association between temperature shocks and productivity has been recognised since the seminal writings of [32]. Despite policymakers and scholars acknowledging the detrimental effect of rising temperatures on economic growth, the subject remains a topic of interest due to ongoing discussions surrounding how to assess these effects. Indeed, measuring the macroeconomic impact of climate change presents a substantial challenge due to the plethora of mechanisms by which climate change might affect the economy. Ref. [33] argued that two types of risk are associated with climate change: physical and transition. Physical risks stem from the outcomes of extreme weather events, including earthquakes, heatwaves, floods, and windstorms, which may reduce the economic capacity for output. The transition risks consist of the cost of shifting into a low-carbon economy, which could lower economic growth. Climate change may also harm economic growth since governments face opportunity costs when spending resources to adapt to climate change rather than on R&D and capital investments.

The existing literature concentrated on the impact of climate change on different economic sectors. For instance, the influence of climate change on agriculture was investigated by [34,35,36,37], while [38] highlighted that the effects of climate change on the macroeconomy might be transmitted via its impact on agriculture, industry, and investment. In addition, Refs. [39,40] confirmed that the adverse effects of climate change on economic growth might be transmitted through tourism. Climate change may also negatively affect bank profitability and financial stability [41,42], resulting in financial losses and negative repercussions on economic growth. Another critical point to note is that the long-run effects of climate change can be different from those observed in the short run. On the one hand, climate change could exert additional long-run effects, such as changes in groundwater, soil quality, and sea level, which results in higher adverse economic impacts [43,44,45]. On the other hand, long-term effects may be smaller than short-term effects because climate change adaptation strategies may offset the adverse short-run economic effects of climate change. Therefore, analysing the short-run and long-run effects of climate change seems necessary.

On the empirical side, although a consensus has evolved about the negative consequences of climate change on economic growth, opinions have been split regarding the extent to which climate change inhibits economic growth. For instance, ref. [38] examined the impact of climate change on output based on a monthly dataset that includes temperature, precipitation, and GDP from 1950 to 2003. The results suggest significant adverse effects of temperature changes on output and output growth in low-income economies as an increase of 1 °C in the temperature reduces output by 1.3 percentage points. However, temperature changes have no significant effect on growth in high-income economies. Changes in precipitation have no substantive economic impact in all countries. A study on the US states by [46] explored the repercussions of climate change on output growth using quarterly data from 1957 to 2012. The empirical results show that temperature has significant adverse effects at the aggregate and sectoral levels. The impact was strong during the summer, where an increase of 1 °F in temperature reduces the annual output growth by 0.15 to 0.25 percentage points. Ref. [23] looked at the long-term impact of climate change on economic activity for a large sample of 174 countries during the period ranging between 1960 and 2014. The investigation considers a stochastic growth model in which country-specific climate variables drive labour productivity. The findings suggest that a persistent change in temperature above or below its historical level harms per capita real GDP growth. Furthermore, a 0.04 °C annual increase in global average temperature will reduce world GDP per capita by more than 7% by 2100 if no mitigating policies are adopted. Ref. [47] instead analysed the effects of temperature on total factor productivity during the period 1960–2006. The findings indicate that a negative association exists in low-income countries, with a 1 °C annual increase in temperature reducing TFP growth rates by roughly 1.1–1.8%. In high-income countries, the impact is not statistically significant. Ref. [48] extended the research by [47] to analyse the effects of temperature not only on total factor productivity but also on capital stock and employment. The study conducted an empirical investigation using data from 103 countries. The findings indicated that economic variables are more negatively impacted by high temperatures in low-income countries as compared to high-income countries.

Ref. [49] investigated the macroeconomic effects of extreme weather events in the United States using the Smooth Transition Vector Autoregressive model from 1963 to 2019. The study relies on the Actuaries Climate Index, which considers many features of climate change, such as temperature, rainfall, drought, wind, and sea level. The authors select the industrial production growth, inflation, the effective federal funds rate, and the unemployment rate as dependent variables. The empirical findings indicate that climate change reduces output growth and employment and that the long-term impact of extreme weather events is more substantial than short-run effects. Ref. [50] also investigated the macroeconomic effects of extreme weather events in seven Central American countries between 2001 and 2019. The country-specific VAR and panel VAR suggest that climate disasters reduce monthly economic activity by about 0.5 to 1 percentage point. Ref. [22] analysed the effects of climate change on economic growth in several countries between 1950 and 2011. The whole sample of countries was divided into three subsamples according to the climate regime: coldest, warmest, and temperate. The analysis shows that temperature changes generally have no statistically significant short- and long-run effects on economic growth. However, there is some evidence of the adverse impacts of rising temperatures in the poorest countries. Similar findings were also reached by [51], who analysed the effects of temperature changes on economic growth at the subnational level by considering a panel of 10,597 grid cells and concluded the presence of adverse effects on economic growth. Furthermore, the authors reveal that the impact of temperature changes is more elevated when considering within-country variability. Ref. [52] analysed the impact of climate change on output at the sectoral level in Chile during the period 1985–2017. The findings reveal some heterogeneity since climate change affects the various economic sectors differently. More specifically, it has been shown that precipitation changes have no significant impact on GDP, while higher summer temperatures negatively affect the agriculture–silviculture and fishing and the construction and electricity, gas, and water sectors. Ref. [53] analysed the impact of daily temperature on annual income in the United States between 1969 and 2011. The empirical findings confirmed the detrimental economic effects of temperature as productivity declines about 1.7% for each 1 °C increase in temperature above 15 °C. Ref. [54] conducted a study on the nonlinear effects of temperature on output and concluded that global economic output has a nonlinear relationship with average temperature. Indeed, the impact of temperature on output is positive and reaches a maximum at about 13 °C and then sharply decreases at higher temperature levels. Finally, ref. [55] examined the effects of temperature variations on productivity in a large sample of 114 countries between 1955 and 2016. The panel VAR estimates demonstrate that the response of economic growth to temperature varies significantly across countries. Indeed, higher temperature variations enhance productivity in Asia while causing a decline in Europe and North America and, finally, have no significant impact on productivity in Africa and South America.

Two main conclusions may be drawn from the literature review. First, there is evidence that climate change negatively affects economic growth, with higher adverse effects in low-income countries. Second, despite the growing body of literature on the topic, no previous study has examined the spatial effects of climate change on economic growth.

3. Methodology

3.1. The Dynamic Spatial Panel Data Model

The spatial dimension allows quantifying the response of economic growth not only to climate change in the same country but also to that occurring in neighbouring countries. Indeed, positive and negative macroeconomic and climatic spillovers between neighbouring countries may exist. For example, the economic growth of a country i may affect that of its neighbour j due to economic and financial interconnections. The spatial panel data analysis allows capturing three types of effects from climate change on economic growth: direct, indirect (spillover), and total. By doing so, the spatial panel data model outperforms the standard models used in previous studies, which neglected the indirect effects and consequently underestimated the impact of climate change on economic growth.

The current research employs a dynamic spatial panel data model to estimate the effects of climate change on economic growth. Such models have better explanatory power than traditional panel data models, such as fixed and random effects, due to their ability to incorporate both dynamic and spatial effects. Because of this, dynamic spatial panel data models have been popular among scholars in recent years. The dynamic spatial model may take three main variants: (i) dynamic spatial Durbin model (DSDM), (ii) dynamic spatial autoregressive model (DSAR), and (iii) dynamic spatial error model (DSEM).

The dynamic spatial Durbin model (DSDM) accounts for the spatial interdependence for the dependent and explanatory variables. The DSDM may be written as follows:

$$\begin{gathered} y_{it}=\alpha +\tau y_{it-1}+\rho {\displaystyle \sum }_{j=1}W{y}_{it}+\lambda {\displaystyle \sum }_{j=1}W{y}_{it-1}+{\displaystyle \sum }_{k}W{X}_{it}^{\left(k\right)}{\theta }_k+{\displaystyle \sum }_k{X}_{it}^{\left(k\right)}{\beta }_k+{\mu }_i+{\epsilon }_{it} \\ i=1, \dots \dots ,N; t=1,\dots \dots , T \end{gathered}$$

(1)

where $y_{it}$ represents GDP per capita growth of country $i$ at year $t$. $X_{it}$ denotes the $k\ast 1$ vector of characteristics of country $i$ that explains GDP per capita growth, $W$ is the row-normalised spatial weight matrix, $\rho$ is the spatial lag parameter that indicates the intensity of spatial autocorrelation between a country and its geographical neighbours, $\lambda$ is the spatial-time autoregressive parameter. Finally, ${\mu }_i$ represents the fixed spatial individual effect, while ε_it is the error term.

The dynamic spatial autoregressive model (DSAR) only takes into account the spatial interdependence for the dependent variable and may be written as follows:

$$\begin{gathered} y_{it}=\alpha +\tau y_{it-1}+\rho {\displaystyle \sum }_{j=1}W{y}_{it}+\lambda {\displaystyle \sum }_{j=1}W{y}_{it-1}+{\displaystyle \sum }_k{X}_{it}^{\left(k\right)}{\beta }_k+{\mu }_i+{\epsilon }_{it} \\ i=1, \dots \dots ,N; t=1,\dots \dots , T \end{gathered}$$

(2)

Finally, the dynamic spatial error model (DSEM) takes into account the spatial interdependence for the error term and may be written as follows:

$$\begin{gathered} y_{it}=\alpha +\tau y_{it-1}+{\displaystyle \sum }_k{X}_{it}^{\left(k\right)}{\beta }_k+{\mu }_i+{\epsilon }_{it} \\ i=1, \dots \dots ,N; t=1,\dots \dots , T \end{gathered}$$

(3)

$${\delta }_{it}=\varphi W{\delta }_{it}+{\epsilon }_{it}, {\epsilon }_{it}~N\left(0,{\sigma }_i^{2}I\right)$$

It is worth mentioning that the appropriate dynamic spatial panel data model (DSDM/DSAR/DSEM) will be selected based on two Lagrange multiplier (LM) tests: the simple LM test for error dependence (LMerr) and the simple LM test for a missing spatially lagged dependent variable (LMlag).

3.2. The Spatial Weight Matrix

The weighting matrix or spatial matrix can take many forms. It can be a contiguity weight matrix ($W_c)$ that takes the following form:

$$W_{ij}=\{\begin{array}{ll}1,& if countries i and j are neighbours\\ 0,& otherwise\end{array}$$

(4)

The different weights are obtained using a row-standardised contiguity matrix as described below:

$$W_{ij}=\left[\begin{array}{ccc}0& w_{12}^{*}& w_{13}^{*}\\ w_{\begin{array}{c}21\\ \\ ⋮\end{array}}^{*}& & \\ w_{n1}^{*}& w_{n2}^{*}& w_{n3}^{*}\end{array} \begin{array}{ccc}{w}_{14}^{*}& \cdots & w_{1n}^{*}\\ & & w_{\begin{array}{c}2n\\ \\ ⋮\end{array}}^{*}\\ w_{n4}^{*}& \dots & 0\end{array}\right]$$

(5)

where $w_{ij}^{*}=\frac{w_{ij}}{{{\displaystyle \sum }}_{j=1}^n{w}_{ij}}$.

On the other hand, the spatial matrix can be based on the inverse distance. In this case, the inverse distance weight matrix ($W_d)$ may be written as follows:

$$W_d=\{\begin{array}{c}{d}_{ij}^{-1}, if i\ne j\\ 0, if i=j\end{array}$$

(6)

where $d_{ij}$ is the geographic distance between country $i$ and country $j.$

3.3. The Spatial Autocorrelation Test

Before estimating the dynamic spatial panel data model, it is essential to test the presence of spatial autocorrelation for the dependent variable. Indeed, the dynamic spatial panel data model is appropriate in the presence of spatial autocorrelation for the dependent variables. To accomplish that, we perform the global Moran index, allowing us to assess the existence of autocorrelation based on previously mentioned weight matrices ($W_c$ and $W_d$). The global Moran index is defined by the following formula:

$$Z=\frac{I-E\left(I\right)}{\sqrt{var\left(I\right)}}$$

(7)

where $I$ is the Moran index of the dependent variable, $E\left(I\right)$ is the expected value of the Moran index, and $var\left(I\right)$ is the variance of the Moran index. Under the null hypothesis of the absence of autocorrelation, Moran’s I statistics follow a reduced-centered normal law. In the same vein, the local Moran index will take the following form:

$$I_{it}=\frac{\left(z_{ij}-m_t\right){{\displaystyle \sum }}_{j=1;j\ne i}^n{w}_{ij}\left(m_{ij}-z_t\right)}{S_i^2}$$

(8)

where $S_i^2=\frac{{{\displaystyle \sum }}_{j=1;j\ne i}^n{w}_{ij}{\left(z_{jt}-m_t\right)}^2}{n-1}-m_t^2$, Z_it is the variable subject to spatial autocorrelation, $m$ is the mean of $z$ at the moment $t$, $W_{ij}$ is the element of the W-weight matrix between the two countries $i$ and $j,$ and, finally, $n$ is the number of countries.

4. Data

The empirical investigation assesses the impact of climate change on economic growth using a balanced dataset for 86 developed and developing countries from 1980 to 2019. Following many previous related studies, including [22,23], the dependent variable is measured by GDP per capita growth. Regarding climate change, it is worth noting that it is a multifaceted phenomenon that can manifest in various ways, including temperature variations, precipitation, occurrences of drought, and rising sea levels. Following many previous studies, such as [38], temperature variations are used as a proxy for climate change. The choice of temperature variations to depict climate change is mainly owing to the fact that global warming has been the most apparent climate change phenomenon during the past few decades. Moreover, many studies analysing the macroeconomic effects of climate change, such as [38,51], employed temperature variations as a proxy of climate change. The data are extracted from the Climate Change Knowledge Portal of the World Bank. A set of control variables is also introduced in the specification. These variables are initial GDP to test for the convergence hypothesis, population growth, human capital, government expenditure as a share of GDP, trade openness (sum of exports and imports as a share of GDP), and foreign direct investments (as a share of GDP). The dependent variable (GDP per capita growth) and the control variables, except human capital, are extracted from the World Development Indicators of the World Bank. Human capital is finally measured by the human capital index provided by Penn World Table.

Due to data limitations, the full sample considered in this study has been reduced from an initial sample of 120 to 86 countries. The sample of countries is listed in

Table A1. The sample and classification of countries.

#	Country	Climate Regime	Income Level	#	Country	Climate Regime	Income Level
1	Albania	Coldest	LMI	44	Jordan	Coldest	LMI
2	Algeria	Hottest	LMI	45	Kenya	Hottest	LMI
3	Argentina	Coldest	LMI	46	Korea	Coldest	HI
4	Australia	Hottest	HI	47	Kuwait	Hottest	HI
5	Austria	Coldest	HI	48	Madagascar	Hottest	LMI
6	Bahrain	Hottest	HI	49	Malaysia	Hottest	LMI
7	Bangladesh	Hottest	LMI	50	Mali	Hottest	LMI
8	Belgium	Coldest	HI	51	Malta	Coldest	HI
9	Benin	Hottest	LMI	52	Mexico	Coldest	LMI
10	Bolivia	Coldest	LMI	53	Morocco	Coldest	LMI
11	Botswana	Hottest	LMI	54	Myanmar	Hottest	LMI
12	Brazil	Hottest	LMI	55	Namibia	Coldest	LMI
13	Bulgaria	Coldest	LMI	56	Netherlands	Coldest	HI
14	Burkina Faso	Hottest	LMI	57	New Zealand	Coldest	HI
15	Cameroon	Hottest	LMI	58	Nicaragua	Hottest	LMI
16	Canada	Coldest	HI	59	Niger	Hottest	LMI
17	Central African Republic	Hottest	LMI	60	Nigeria	Hottest	LMI
18	Chile	Coldest	HI	61	Norway	Coldest	HI
19	China	Coldest	LMI	62	Pakistan	Coldest	LMI
20	Colombia	Hottest	LMI	63	Panama	Hottest	HI
21	Congo	Hottest	LMI	64	Paraguay	Hottest	LMI
22	Côte d’Ivoire	Hottest	LMI	65	Peru	Coldest	LMI
23	Cyprus	Coldest	HI	66	Philippines	Hottest	LMI
24	Denmark	Coldest	HI	67	Portugal	Coldest	HI
25	Dominican Republic	Hottest	LMI	68	Saudi Arabia	Hottest	HI
26	Ecuador	Coldest	LMI	69	Senegal	Hottest	LMI
27	Egypt	Coldest	LMI	70	Sierra Leone	Hottest	LMI
28	El Salvador	Hottest	LMI	71	Singapore	Hottest	HI
29	Fiji	Hottest	LMI	72	South Africa	Coldest	LMI
30	Finland	Coldest	HI	73	Spain	Coldest	HI
31	France	Coldest	HI	74	Sri Lanka	Hottest	LMI
32	Gabon	Hottest	LMI	75	Sweden	Coldest	HI
33	Gambia	Hottest	LMI	76	Switzerland	Coldest	HI
34	Germany	Coldest	HI	77	Syrian Arab Republic	Coldest	LMI
35	Greece	Coldest	HI	78	Thailand	Hottest	LMI
36	Honduras	Hottest	LMI	79	Togo	Hottest	LMI
37	India	Hottest	LMI	80	Tunisia	Coldest	LMI
38	Indonesia	Hottest	LMI	81	Turkey	Coldest	LMI
39	Iraq	Hottest	LMI	82	Uganda	Hottest	LMI
40	Ireland	Coldest	HI	83	United Kingdom	Coldest	HI
41	Italy	Coldest	HI	84	United States	Coldest	HI
42	Jamaica	Hottest	LMI	85	Uruguay	Coldest	HI
43	Japan	Coldest	HI	86	Zimbabwe	Coldest	LMI

“LMI” stands for low–middle-income countries (low-income countries + lower middle-income countries + upper middle-income countries), while “HI” stands for high-income countries. The classification is based on the World Bank. “Hottest” stands for countries with average temperature above the median (hottest countries), while “Coldest” stands for countries with average temperature below the median (coldest countries). See Section 4 for details on classification of countries.

in the Appendix A. The effects of climate change on economic growth are first estimated for the full sample of 86 countries. Then, the sample is decomposed into two subsamples according to the climate regime and income level. When considering the climate regime, we follow many previous studies, including [56,57], using the median of a given variable as a classification tool. Indeed, countries with an average temperature above the median are considered the hottest countries (43), while those with below-median average temperature levels are considered the coldest countries (43). The decomposition of the whole sample according to income level is based on the classification of the World Bank and provides 31 high-income countries and 55 low–middle-income countries.

5. Empirical Findings

The empirical investigation consists of three stages. We first conduct a preliminary analysis to check the presence of spatial autocorrelation, stationarity, and the appropriate spatial model. In the presence of spatial effects, we move to estimate the dynamic spatial panel data model. We finally proceed to the most important stage, which entails estimating the marginal direct, indirect (spillover), and total effects of climate change on economic growth in the short- and long term.

5.1. Preliminary Analysis

Before estimating the dynamic spatial panel data model, a series of preliminary tests should be performed. We begin by checking the presence of spatial autocorrelation based on Moran’s I spatial autocorrelation test (

Table 1. Results of Moran’s I spatial autocorrelation test.

Years	Contiguity Weight Matrix (${\mathit{W}}_{\mathit{c}}$)		Inverse Distance Weight Matrix (${\mathit{W}}_{\mathit{d}}$)
	Moran’s I	p-Value	Moran’s I	p-Value
1980	0.036 ***	0.000	0.025 ***	0.000
1981	0.052 ***	0.000	0.041 ***	0.000
1982	0.035	0.172	0.063 ***	0.003
1983	0.121 ***	0.004	0.046 **	0.019
1984	0.085 **	0.024	0.033 **	0.049
1985	0.091 **	0.019	0.040 **	0.028
1986	0.010	0.330	0.004	0.277
1987	0.189 ***	0.000	0.064 ***	0.003
1988	0.046	0.123	0.043 **	0.024
1989	0.169 ***	0.000	0.114 ***	0.000
1990	−0.040	0.250	−0.040	0.114
1991	0.086 **	0.038	−0.052 **	0.044
1992	0.328 ***	0.000	0.167 ***	0.000
1993	0.238 ***	0.000	0.115 ***	0.000
1994	0.085 **	0.027	0.037 **	0.038
1995	0.014	0.305	0.043 **	0.023
1996	0.024	0.238	0.004	0.281
1997	−0.073	0.105	−0.014	0.472
1998	0.118 ***	0.002	0.059 ***	0.003
1999	0.103 **	0.010	0.039 **	0.032
2000	0.289 ***	0.000	0.112 ***	0.000
2001	0.053 *	0.096	0.031 *	0.059
2002	0.042	0.137	0.017	0.142
2003	−0.062	0.117	−0.040	0.113
2004	0.081 ***	0.005	0.034 **	0.012
2005	0.145 ***	0.001	0.078 ***	0.001
2006	0.147 ***	0.001	0.076 ***	0.001
2007	0.112 ***	0.006	0.063 ***	0.003
2008	0.026	0.216	0.013	0.172
2009	0.153 ***	0.000	0.061 ***	0.004
2010	0.115 ***	0.005	0.084 ***	0.000
2011	0.164 ***	0.000	0.078 ***	0.001
2012	0.088 **	0.014	0.063 ***	0.001
2013	0.071 **	0.028	0.063 ***	0.001
2014	0.171 ***	0.000	0.094 ***	0.000
2015	0.015	0.275	0.002	0.285
2016	0.049	0.104	0.024 *	0.093
2017	0.181 ***	0.000	0.084 ***	0.000
2018	0.107 ***	0.008	0.048 **	0.014
2019	0.083 **	0.027	0.038 **	0.034

***, **, and * denote the rejection of the null hypothesis of no spatial autocorrelation at the 1%, 5%, and 10% levels, respectively.

). Indeed, without spatial autocorrelation among the dependent variable, the dynamic spatial panel data model is no longer suitable. Then, we test the stationary properties of the variables (

Table 2. Results of unit root tests.

	IPS Unit Root Test		LLC Unit Root Test
	Constant	Constant and Trend	Constant	Constant and Trend
$GROWTH$	−18.52 ***	−16.56 ***	−4.65 ***	−4.92 ***
$LGDP$	−3.09 ***	−2.79 ***	−2.34 ***	−5.05 ***
$GOV$	−3.09 ***	−7.79 ***	−4.19 ***	−5.60 ***
$HC$	−5.81 ***	−4.72 ***	−2.51 ***	−4.50 ***
$TRADE$	−5.75 ***	−7.97 ***	−2.73 ***	−4.06 ***
$FDI$	−14.56 ***	−18.86 ***	−8.90 ***	−7.66 ***
$POP$	−6.42 ***	−6.75 ***	−6.20 ***	−9.48 ***
$TEMP$	−40.99 ***	−40.95 ***	−52.38 ***	−44.54 ***

*** stands for the rejection of the null hypothesis of a unit root at 1% level.

). This is important as the dynamic spatial model requires stationary variables. Finally, one should identify the appropriate dynamic spatial panel data model (DSDM, DSAR, or DSEM) based on the LMerr and LMlag tests (

Table 3. Results of selection model tests.

	LMerr Test (Spatial Error)	LMlag Test (Spatial Lag)	Hausman Test
Full sample
$W_c$	7.41 (0.380)	25.15 *** (0.000)	26.55 *** (0.000)
$W_d$	10.25 (0.110)	34.91 *** (0.000)	32.64 *** (0.000)
Hottest countries
$W_c$	7.94 (0.390)	37.55 *** (0.000)	38.74 *** (0.000)
$W_d$	8.59 (0.280)	24.08 *** (0.000)	48.65 *** (0.000)
Coldest countries
$W_c$	12.42 * (0.080)	9.87 ** (0.040)	34.25 *** (0.000)
$W_d$	9.66 (0.200)	25.11 *** (0.000)	34.24 *** (0.000)
Low–middle-income countries
$W_c$	5.88 (0.610)	42.15 *** (0.000)	27.25 *** (0.000)
$W_d$	11.50 (0.100)	19.44 *** (0.000)	36.12 *** (0.000)
High-income countries
$W_c$	0.01 (0.940)	27.02 *** (0.000)	28.04 *** (0.000)
$W_d$	11.15 * (0.090)	9.92 ** (0.030)	48.75 *** (0.000)

***, **, and * represent the statistical significance at 1%, 5%, and 10%, respectively. p-values are in parentheses. For country classification, see Table A1 in the Appendix A.

). This step is of great importance as it allows the selection of the best spatial model that will be estimated to obtain the direct, indirect, and total effects of climate change on economic growth.

For robustness check, we report the Moran’s I statistics using the contiguity weight matrix $(W_c)$ and the inverse distance weight matrix ($W_d)$. The results indicate the existence of a significant positive spatial autocorrelation for GROWTH in most cases when using the two matrices. This finding suggests that GDP per capita growth associated with the different countries is related. Therefore, Table 1 confirms the presence of spatial autocorrelation among countries in our sample and the suitability of the dynamic spatial panel data model.

We then test the stationarity of variables. To ensure the robustness of results, this paper employs two panel data unit root tests: the LLC test developed by [58], which supposes a common unit root process, and the IPS test developed by [59], which assumes an individual unit root process. Moreover, the two tests are implemented for specifications with constant and with trend and constant.

The results of the two panel unit root tests reported in Table 2 suggest rejecting the null hypothesis of a unit root for all variables and models. Therefore, all variables are integrated of order one and can be introduced in levels in the dynamic spatial panel data model. As shown in Table 3, the Hausman test statistics suggest that the fixed effect estimator outperforms the random effect estimator for estimating the spatial panel data model. Moreover, the simple LM test for error dependence and simple LM test for a missing spatially lagged dependent variable indicate the superiority of the DSAR model compared to the DSEM and SDDM models.

These findings are valid for the whole sample and the different subsamples. Therefore, a dynamic spatial autoregressive model having the following equation will be estimated:

$$\begin{gathered} GROWT{H}_{it}=\alpha +\tau GROWT{H}_{i,t-1}+\rho {\displaystyle \sum }_{j=1}W\ast GROWT{H}_{it}+\lambda {\displaystyle \sum }_{j=1}W\ast GROWT{H}_{i,t-1}+{\beta }_{1}TEM{P}_{it}+{\beta }_{2}LGD{P}_{it}+{\beta }_{3}GO{V}_{it} \\ +{\beta }_{4}H{C}_{it}+{\beta }_{5}TRAD{E}_{it}+{\beta }_{6}FD{I}_{it}+{\beta }_{7}PO{P}_{it}+{\mu }_i+{\epsilon }_{it} \\ i=1, \dots \dots ,N; t=1,\dots \dots , T \end{gathered}$$

(9)

5.2. Results of the Dynamic Spatial Autoregressive Model

The preliminary analysis conducted above suggested the suitability of the dynamic spatial autoregressive model to estimate the effects of climate change on economic growth. The estimation results are reported in

Table 4. Results of the dynamic spatial autoregressive model.

Variables	Full Sample	Hottest Countries	Coldest Countries	LMI Countries	HI Countries
Contiguity weight matrix ($W_c$)
$L.GROWTH$	0.097 *** (0.000)	−0.011 (0.647)	−0.012 (0.621)	0.041 * (0.063)	0.267 *** (0.000)
$L.W\ast GROWTH$	0.023 ** (0.011)	0.071 *** (0.000)	0.003 (0.887)	−0.007 (0.547)	−0.007 (0.736)
$TEMP$	−0.083 (0.284)	−0.761 ** (0.038)	−0.074 (0.407)	−0.152 *** (0.000)	0.047 (0.728)
$LGDP$	−3.746 *** (0.000)	−3.702 *** (0.000)	−6.093 *** (0.000)	−2.631 *** (0.000)	−5.512 *** (0.000)
$GOV$	−0.081 *** (0.002)	−0.029 (0.400)	0.010 (0.775)	−0.096 (0.307)	0.078 ** (0.023)
$HC$	2.459 *** (0.000)	3.378 *** (0.000)	2.972 *** (0.000)	2.750 *** (0.000)	1.984 *** (0.007)
$TRADE$	0.026 *** (0.000)	0.021 *** (0.005)	0.040 *** (0.000)	0.022 *** (0.000)	0.035 *** (0.000)
$FDI$	−0.003 (0.470)	0.050 (0.211)	−0.007 (0.223)	0.096 ** (0.010)	−0.002 (0.534)
$POP$	0.021 (0.780)	0.188 * (0.070)	0.195 ** (0.037)	0.124 (0.443)	0.120 * (0.077)
Inverse distance weight matrix ($W_d$)
$L.GROWTH$	0.089 *** (0.000)	−0.020 (0.425)	−0.016 (0.499)	0.032 (0.141)	0.268 *** (0.000)
$L.W\ast GROWTH$	0.126 ** (0.053)	0.372 *** (0.000)	0.091 (0.266)	0.043 (0.601)	−0.108 * (0.077)
$TEMP$	−0.081 (0.292)	−0.725 ** (0.048)	−0.077 (0.386)	−0.144 *** (0.000)	0.048 (0.716)
$LGDP$	−3.513 *** (0.000)	−3.508 *** (0.000)	−6.090 *** (0.000)	−2.548 *** (0.000)	−4.876 *** (0.000)
$GOV$	−0.067 * (0.092)	−0.019 (0.567)	0.017 (0.644)	−0.101 (0.279)	0.088 *** (0.008)
$HC$	2.100 *** (0.000)	2.472 *** (0.001)	3.185 *** (0.000)	2.190 *** (0.001)	2.044 *** (0.004)
$TRADE$	0.024 *** (0.000)	0.018 ** (0.015)	0.039 *** (0.000)	0.019 *** (0.008)	0.035 *** (0.000)
$FDI$	−0.004 (0.373)	0.038 (0.345)	−0.007 (0.208)	0.083 ** (0.028)	−0.001 (0.598)
$POP$	0.012 (0.868)	0.173 * (0.094)	0.203 ** (0.029)	0.158 (0.330)	0.126 * (0.054)
${\rho }_{Wd}$	0.036 ***	0.131 **	0.023	0.046 ***	0.118 ***
${\rho }_{Wc}$	0.452 ***	0.123 *	0.247 ***	0.322 ***	0.530 ***
Number of countries	86	43	43	55	31

***, **, and * represent the statistical significance at 1%, 5%, and 10%, respectively. p-values are in parentheses. For country classification, see Table A1 in the Appendix A.

.

The estimated coefficients on the lagged dependent variable and its spatial term are generally significant, suggesting the appropriateness of the dynamic spatial panel model. The table suggests that the coefficient of the spatially lagged dependent variable $\rho$ is positive and always statistically significant when using the contiguity weight matrix and the inverse distance weight matrix. Such results confirm the existence of positive spatial autocorrelation, implying that economic growth in a given country stimulates economic growth in other countries. When considering the full sample, the coefficient associated with temperature change is negative but not statistically significant for the two matrices. One potential reason for the insignificance of coefficients is the heterogeneity of the sample. To obtain a more in-depth understanding, we decompose the sample. The decomposition according to the climate regime (hottest vs. coldest) reveals some new findings. Although the coefficient of temperature change is negative for both groups of countries, the analysis suggests that only the coefficient associated with the hottest countries is significant. Consequently, these findings show that climate change affects economic growth only in the hottest countries, while such a conclusion does not hold for the coldest countries. When we decompose the sample according to the income level (low–middle-income vs. high-income), the table reveals that the detrimental effects of temperature change on economic growth are confirmed for low–middle-income countries. The coefficient of temperature changes in high-income countries is positive but statistically insignificant when using both matrices. Regarding the control variables, the table confirms the role of human capital and trade openness as drivers of economic growth in the whole sample and the different sub-samples. The population is also shown to be a vital determinant of economic growth as an increase in population induces an improvement in human capital. In addition, foreign direct investments exert a positive impact on economic growth only in low–middle-income countries, which may be explained by the existence of investment opportunities in these countries. Finally, government expenditure negatively affects economic growth in the full sample and high-income countries as a rise in government expenditure may reduce public investments and impede economic growth. It is worth noting that the aforementioned findings are robust to the spatial weight matrix.

5.3. Analysis of Direct and Indirect Marginal Effects

The current subsection aims to estimate the direct, indirect (spillover), and total marginal effects of climate change on economic growth in the short- and long term. To check the robustness of the results, the marginal effects are computed using the contiguity weight matrix and the inverse distance weight matrix. The analysis is completed for the full sample and the different subsamples.

5.3.1. The Full Sample

The direct, indirect, and total marginal impacts of temperature change on economic growth in the short and long run using the two spatial weight matrices are reported in

Table 5. Direct, indirect, and total effects of climate change on economic growth—full sample.

	Short-Run Effects			Long-Run Effects
	Direct	Indirect	Total	Direct	Indirect	Total
Contiguity weight matrix ($W_c$)
$TEMP$	−0.076 (0.315)	−0.007 (0.338)	−0.084 (0.316)	−0.086 (0.315)	−0.015 (0.327)	−0.101 (0.316)
$LGDP$	−3.717 *** (0.000)	−0.345 *** (0.000)	−4.062 *** (0.000)	−4.153 *** (0.000)	−0.765 *** (0.000)	−4.918 *** (0.000)
$GOV$	−0.082 *** (0.001)	−0.007 *** (0.008)	−0.089 *** (0.001)	−0.091 *** (0.001)	−0.016 *** (0.003)	−0.108 *** (0.001)
$HC$	2.439 *** (0.000)	0.226 *** (0.000)	2.665 *** (0.000)	2.726 *** (0.000)	0.501 *** (0.000)	3.227 *** (0.000)
$TRADE$	0.026 *** (0.000)	0.002 *** (0.001)	0.029 *** (0.000)	0.029 *** (0.000)	0.005 *** (0.000)	0.035 *** (0.000)
$FDI$	−0.003 (0.483)	−0.0003 (0.491)	−0.004 (0.482)	−0.004 (0.483)	−0.0007 (0.485)	−0.005 (0.482)
$POP$	0.018 (0.802)	0.001 (0.800)	0.019 (0.802)	0.003 (0.800)	0.003 (0.800)	0.024 (0.801)
Inverse distance weight matrix ($W_d$)
$TEMP$	−0.075 (0.324)	−0.062 (0.343)	−0.137 (0.328)	−0.083 (0.324)	−0.148 (0.353)	−0.232 (0.337)
$LGDP$	−3.506 *** (0.000)	−2.904 *** (0.000)	−6.411 *** (0.000)	−3.910 *** (0.000)	−6.880 *** (0.000)	−10.791 *** (0.000)
$GOV$	−0.068 *** (0.006)	−0.056 ** (0.016)	−0.125 *** (0.007)	−0.076 *** (0.006)	−0.134 ** (0.024)	−0.211 ** (0.012)
$HC$	2.095 *** (0.000)	1.731 *** (0.000)	3.827 *** (0.000)	2.336 *** (0.000)	4.098 *** (0.001)	6.435 *** (0.000)
$TRADE$	0.024 *** (0.000)	0.020 *** (0.001)	0.044 *** (0.000)	0.027 *** (0.000)	0.047 *** (0.002)	0.075 *** (0.000)
$FDI$	−0.004 (0.388)	−0.003 (0.398)	−0.008 (0.389)	−0.005 (0.388)	−0.009 (0.406)	−0.014 (0.394)
$POP$	0.009 (0.896)	0.008 (0.892)	0.017 (0.894)	0.010 (0.896)	0.019 (0.892)	0.030 (0.893)

*** and ** represent the statistical significance at 1% and 5%, respectively. p-values are in parentheses.

. As shown, the direct effects of temperature change on economic growth are negative but not statistically significant when using the inverse distance and contiguity weight matrices. These findings are still valid in the short- and long run. In addition, the table shows that the indirect spillover effects of climate change are not statistically significant, implying that temperature change in a given country has no significant repercussions on the economic growth of other countries in the short- and long run. Finally, the total marginal effects of climate change are not statistically significant, which is expected since the direct and indirect effects are insignificant. These findings align with those reported in Table 4, suggesting that temperature changes have no spatial effects on economic growth in the full sample.

Table 5 also corroborates the results of Table 4 regarding control variables. Indeed, there is strong evidence of direct and indirect effects for initial GDP, government expenditure, trade openness, and human capital. However, one may note that the direct effect is higher than the indirect spillover effect for the variables above in both the short- and long run. For instance, the short-run direct effect of trade openness on economic growth is about 0.024 when using the inverse distance weight matrix, while the indirect effect is 0.020, which provides a total effect of 0.044. Consequently, an increase in trade flows by 1% in country $i$ induces a rise in economic growth by 0.024%, while an increase in trade flows by 1% in its neighbours induces a surge in economic growth in country $i$ by 0.020%.

To summarise, the findings reported in Table 5 do not support the existence of direct or indirect spillover effects of climate change on economic growth in the full sample. These results are robust to the spatial weight matrix (inverse distance weight matrix and contiguity weight matrix) and the horizon (short run and long run). These findings may be due to the heterogeneity of countries included in the sample.

5.3.2. Hottest Countries vs. Coldest Countries

The short- and long-run direct, indirect, and total effects of climate change on economic growth in the hottest and coldest countries are presented in

Table 6. Direct, indirect, and total effects of climate change on economic growth—countries by climate regime.

	Short-Run Effects						Long-Run Effects
	${\mathit{W}}_{\mathit{c}}$			${\mathit{W}}_{\mathit{d}}$			${\mathit{W}}_{\mathit{c}}$			${\mathit{W}}_{\mathit{d}}$
	Direct	Indirect	Total	Direct	Indirect	Total	Direct	Indirect	Total	Direct	Indirect	Total
Hottest countries
$TEMP$	−0.729 ** (0.042)	−0.008 (0.668)	−0.737 ** (0.043)	−0.694 * (0.053)	−0.102 (0.238)	−0.796 * (0.056)	−0.729 ** (0.042)	−0.115 * (0.067)	−0.844 ** (0.043)	−0.695 * (0.053)	−0.651 * (0.094)	−1.347 * (0.063)
$LGDP$	−3.636 *** (0.000)	−0.042 (0.626)	−3.679 *** (0.000)	−3.448 *** (0.000)	−0.516 (0.119)	−3.965 *** (0.000)	−3.635 *** (0.000)	−0.575 *** (0.001)	−4.211 *** (0.000)	−3.454 *** (0.000)	−3.262 *** (0.003)	−6.717 *** (0.000)
$GOV$	−0.029 (0.371)	−0.0003 (0.770)	−0.030 (0.371)	−0.020 (0.538)	−0.002 (0.624)	−0.023 (0.540)	−0.029 (0.371)	−0.004 (0.389)	−0.034 (0.371)	−0.020 (0.538)	−0.019 (0.561)	−0.039 (0.544)
$HC$	3.336 *** (0.000)	0.037 (0.633)	3.373 *** (0.000)	2.438 *** (0.001)	0.355 (0.126)	2.794 *** (0.001)	3.334 *** (0.000)	0.526 *** (0.001)	3.860 *** (0.000)	2.442 *** (0.001)	2.279 *** (0.008)	4.721 *** (0.001)
$TRADE$	0.021 *** (0.004)	0.0002 (0.646)	0.021 *** (0.004)	0.019 ** (0.012)	0.002 (0.188)	0.021 ** (0.015)	0.021 *** (0.004)	0.003 ** (0.017)	0.025 *** (0.004)	0.019 ** (0.012)	0.017 ** (0.045)	0.037 ** (0.019)
$FDI$	0.050 (0.226)	0.0005 (0.738)	0.050 (0.226)	0.038 (0.358)	0.005 (0.485)	0.043 (0.360)	0.050 (0.226)	0.007 (0.248)	0.058 (0.226)	0.038 (0.358)	0.035 (0.389)	0.073 (0.365)
$POP$	0.192 * (0.056)	0.002 (0.674)	0.183 * (0.056)	0.168 * (0.077)	0.025 (0.275)	0.194 * (0.082)	0.183 * (0.056)	0.029 * (0.089)	0.212 * (0.057)	0.169 * (0.077)	0.160 (0.132)	0.329 * (0.092)
Coldest countries
$TEMP$	−0.067 (0.448)	−0.003 (0.550)	−0.070 (0.449)	−0.070 (0.424)	−0.024 (0.459)	−0.094 (0.427)	−0.066 (0.448)	−0.003 (0.533)	−0.069 (0.449)	−0.069 (0.424)	−0.035 (0.449)	−0.105 (0.428)
$LGDP$	−6.034 *** (0.000)	−0.282 (0.123)	−6.316 *** (0.000)	−6.048 *** (0.000)	−2.032 *** (0.004)	−8.081 *** (0.000)	−5.963 *** (0.000)	−0.316 * (0.082)	−6.279 *** (0.000)	−5.973 *** (0.000)	−3.014 *** (0.001)	−8.987 *** (0.000)
$GOV$	0.010 (0.782)	0.0004 (0.826)	0.010 (0.782)	0.016 (0.646)	0.005 (0.666)	0.022 (0.647)	0.009 (0.782)	0.0005 (0.819)	0.010 (0.782)	0.016 (0.646)	0.008 (0.660)	0.024 (0.648)
$HC$	2.933 *** (0.000)	0.137 (0.142)	3.070 *** (0.000)	3.155 *** (0.000)	1.063 ** (0.013)	4.218 *** (0.000)	2.899 *** (0.000)	0.153 (0.102)	3.052 *** (0.000)	3.116 *** (0.000)	1.575 *** (0.005)	4.691 *** (0.000)
$TRADE$	0.040 *** (0.000)	0.001 (0.148)	0.042 *** (0.000)	0.039 *** (0.000)	0.013 ** (0.012)	0.053 *** (0.000)	0.040 *** (0.000)	0.002 (0.107)	0.042 *** (0.000)	0.039 *** (0.000)	0.019 *** (0.004)	0.059 *** (0.000)
$FDI$	−0.007 (0.239)	−0.0003 (0.390)	−0.007 (0.239)	−0.007 (0.224)	−0.002 (0.272)	−0.009 (0.226)	−0.007 (0.239)	−0.0003 (0.363)	−0.007 (0.239)	−0.007 (0.224)	−0.003 (0.257)	−0.011 (0.227)
$POP$	0.191 ** (0.027)	0.009 (0.247)	0.200 ** (0.028)	0.200 ** (0.021)	0.067 * (0.085)	0.267 ** (0.025)	0.188 ** (0.027)	0.010 (0.209)	0.199 ** (0.028)	0.197 ** (0.021)	0.100 * (0.065)	0.297 ** (0.026)

***, **, and * represent the statistical significance at 1%, 5%, and 10%, respectively. p-values are in parentheses. For country classification, see Table A1 in the Appendix A.

.

The results may be summarised as follows. First, the table confirms that the effects of temperature change on economic growth are not statistically significant in the coldest countries. More specifically, neither the direct nor the indirect repercussions of climate change in the coldest countries are statistically significant in both the short- and long run. Second, the DSAR model reveals that the direct and indirect long-run effects of temperature changes on economic growth are negative and statistically significant in the hottest countries. This result confirms that the GDP per capita growth in a country $i$ (belonging to the hottest countries) is negatively affected by temperature change in the same country (direct effect) and also by temperature change in its neighbours (indirect effect). For example, when using the contiguity weight matrix, the long-run direct effect of climate change is equal to −0.729, while the indirect effect is equal to −0.115. The total effect of climate change is, therefore, equal to −0.844, meaning that a rise in temperature by 1% in the hottest countries induces a loss in GDP per capita growth by 0.844% in the long run. It is, therefore, evident that ignoring the spatial effects of climate change on economic growth will result in underestimating these effects.

Another important conclusion that may be drawn from the table is that, while the magnitude of the direct impacts of climate change is almost the same when using the contiguity or inverse distance matrices, the indirect effect is higher when using the inverse distance weight matrix. In addition, the direct effects of climate change on economic growth are higher than the indirect effects. Finally, the analysis reveals the significance of the direct effects of climate change and the insignificance of the indirect effects in the short run for the hottest and coldest countries. The short- and long-run direct effects of trade openness, population, and human capital are positive and significant, implying that an increase in trade openness, population, and human capital in country $i$ increases the economic growth of country $i$. The effect of the three variables is higher for the coldest countries. The short-run indirect (spillover) effects are positive and significant only for the coldest countries when $W_d$ is used. This result indicates that, in the short run, an increase in trade openness, population, and human capital in one country increased economic growth in neighbouring countries. The long-run indirect effects of the three above variables are positive and statistically significant for the coldest and hottest countries. In addition, the direct and indirect effects of FDI and government expenditure are insignificant for both groups of countries.

5.3.3. Low–Middle-Income Countries vs. High-Income Countries

We move to estimate the DSAR model for the two subsamples obtained according to the income level: low–middle-income countries (55) and high-income countries (31). The findings are displayed in

Table 7. Direct, indirect, and total effects of climate change on economic growth—countries by income level.

	Short-Run Effects						Long-Run Effects
	${\mathit{W}}_{\mathit{c}}$			${\mathit{W}}_{\mathit{d}}$			${\mathit{W}}_{\mathit{c}}$			${\mathit{W}}_{\mathit{d}}$
	Direct	Indirect	Total	Direct	Indirect	Total	Direct	Indirect	Total	Direct	Indirect	Total
Low–middle-income countries
$TEMP$	−0.150 *** (0.000)	−0.018 *** (0.002)	−0.168 *** (0.000)	−0.143 *** (0.000)	−0.069 *** (0.007)	−0.212 *** (0.000)	−0.157 *** (0.000)	−0.016 *** (0.005)	−0.173 *** (0.000)	−0.148 *** (0.000)	−0.091 *** (0.006)	−0.240 *** (0.000)
$LGDP$	−2.593 *** (0.000)	−0.312 *** (0.001)	−2.905 *** (0.000)	−2.513 *** (0.000)	−1.216 *** (0.005)	−3.730 *** (0.000)	−2.702 *** (0.000)	−0.279 *** (0.004)	−2.982 *** (0.000)	−2.606 *** (0.000)	−1.610 *** (0.004)	−4.216 *** (0.000)
$GOV$	−0.097 (0.291)	−0.011 (0.315)	−0.109 (0.291)	−0.103 (0.263)	−0.050 (0.301)	−0.153 (0.267)	−0.101 (0.291)	−0.010 (0.323)	−0.112 (0.291)	−0.107 (0.263)	−0.066 (0.299)	−0.173 (0.268)
$HC$	2.734 *** (0.000)	0.327 *** (0.001)	3.061 *** (0.000)	2.178 *** (0.000)	1.041 *** (0.007)	3.219 *** (0.000)	2.848 *** (0.000)	0.293 *** (0.003)	3.142 *** (0.000)	2.258 *** (0.000)	1.378 *** (0.006)	3.636 *** (0.000)
$TRADE$	0.022 *** (0.002)	0.002 ** (0.013)	0.025 *** (0.002)	0.020 *** (0.006)	0.009 ** (0.032)	0.029 *** (0.008)	0.023 *** (0.002)	0.002 ** (0.020)	0.025 *** (0.002)	0.020 *** (0.006)	0.012 ** (0.030)	0.033 *** (0.009)
$FDI$	0.096 ** (0.013)	0.011 ** (0.034)	0.108 ** (0.014)	0.083 ** (0.033)	0.040 * (0.069)	0.124 ** (0.035)	0.101 ** (0.013)	0.010 ** (0.044)	0.111 ** (0.014)	0.086 ** (0.033)	0.053 * (0.067)	0.140 ** (0.036)
$POP$	0.120 (0.431)	0.014 (0.454)	0.135 (0.432)	0.154 (0.315)	0.075 (0.356)	0.229 (0.321)	0.125 (0.431)	0.013 (0.461)	0.138 (0.432)	0.160 (0.315)	0.099 (0.354)	0.259 (0.323)
High-income countries
$TEMP$	0.060 (0.659)	0.014 (0.677)	0.074 (0.661)	0.061 (0.647)	0.065 (0.660)	0.126 (0.653)	0.084 (0.659)	0.030 (0.689)	0.114 (0.665)	0.084 (0.647)	0.108 (0.668)	0.193 (0.657)
$LGDP$	−5.577 *** (0.000)	−1.336 *** (0.000)	−6.914 *** (0.000)	−4.992 *** (0.000)	−5.387 *** (0.000)	−10.380 *** (0.000)	−7.782 *** (0.000)	−2.911 *** (0.003)	−10.694 *** (0.000)	−6.895 *** (0.000)	−9.080 *** (0.000)	−15.975 *** (0.000)
$GOV$	0.080 ** (0.019)	0.019 ** (0.038)	0.099 ** (0.020)	0.092 *** (0.006)	0.099 ** (0.016)	0.191 *** (0.009)	0.112 ** (0.019)	0.041 * (0.067)	0.154 ** (0.023)	0.127 *** (0.007)	0.167 ** (0.032)	0.294 ** (0.014)
$HC$	1.977 *** (0.000)	0.474 ** (0.017)	2.451 *** (0.006)	2.066 *** (0.003)	2.235 ** (0.010)	4.302 *** (0.005)	2.759 *** (0.006)	1.032 ** (0.037)	3.791 *** (0.007)	2.854 *** (0.003)	3.770 ** (0.020)	6.625 *** (0.007)
$TRADE$	0.036 *** (0.000)	0.008 *** (0.000)	0.045 *** (0.000)	0.036 *** (0.000)	0.040 *** (0.000)	0.077 *** (0.000)	0.050 *** (0.000)	0.019 *** (0.005)	0.070 *** (0.000)	0.051 *** (0.000)	0.067 *** (0.001)	0.118 *** (0.000)
$FDI$	−0.002 (0.544)	−0.0005 (0.549)	−0.002 (0.543)	−0.001 (0.605)	−0.002 (0.607)	−0.004 (0.605)	−0.003 (0.543)	−0.001 (0.559)	−0.004 (0.543)	−0.002 (0.605)	−0.003 (0.612)	−0.006 (0.606)
$POP$	0.119 * (0.065)	0.029 (0.102)	0.148 * (0.068)	0.128 ** (0.043)	0.139 * (0.068)	0.267 * (0.052)	0.167 * (0.065)	0.063 (0.144)	0.231 * (0.076)	0.177 ** (0.044)	0.236 * (0.097)	0.413 * (0.064)

***, **, and * represent the statistical significance at 1%, 5%, and 10%, respectively. p-values are in parentheses. For country classification, see Table A1 in the Appendix A.

.

The most crucial aspect drawn from the table is that temperature changes affect economic growth in the two groups of countries in quite different ways. Indeed, the effects of temperature changes in low–middle-income countries are negative and statistically significant in the short- and long run. On the contrary, the DSAR model reveals that the effects of temperature on economic growth in high-income countries are not statistically significant. These results are in line with many previous studies, including [22,38,49], which concluded that low-income countries are the ones that are most affected by climate change. Such findings may be explained by the fact that high-income countries have sufficient resources and knowledge to implement climate change mitigation and adaptation policies, whereas low–middle-income countries do not. Our analysis outperforms the previous studies by computing the indirect (spillover) effects of climate change on economic growth. As shown in the table, temperature change in low–middle-income countries exerts a direct adverse effect on economic growth in both the short- and long run. However, the long-run detrimental impacts are higher than those obtained in the short run. In all cases, the indirect effects are also negative and statistically significant at the 1% level. More specifically, the indirect effects of climate change on economic growth in low–middle-income countries range between −0.018 and −0.069 in the short run and between −0.016 and −0.091 in the long run. Therefore, the total effects of climate change, which are equal to the sum of direct and indirect effects, are higher than the direct effects and range between −0.168 and −0.212 in the short run and −0.173 and −0.240 in the long run. These findings imply that economic growth in low–middle-income countries is negatively affected by climate change and global warming occurring in the same country and also by climate change in neighbouring countries. Ignoring the indirect effects of climate change induces an underestimation of the harmful effects of climate change. Regarding the control variables, the table shows that human capital and trade openness are positively linked to economic growth in low–middle-income and high-income countries. The positive impacts of these two variables come from the direct and indirect effects on economic growth. Indeed, the DSAR model suggests that human capital and trade openness have indirect spillover effects on economic growth in the short- and long run, meaning that an improvement in human capital and trade openness in a given country boosts economic growth in the same country and its neighbours. However, the direct effects exceed the indirect effects in all cases. Furthermore, the analysis suggests that both short- and long-run effects of FDI are positive and statistically significant only for low–middle-income countries. In addition, the indirect effects are positive and statistically significant, indicating that a rise in FDI flows in country $i$ increases the economic growth of country $i$ and in the neighbouring countries. In high-income countries, the direct and indirect effects of population and government expenditure are positive and significant in the short- and long run. For this group of countries, increasing population and government expenditure in a given country improve economic growth in the same country and its neighbours. However, one should note that the indirect effects are always smaller than the direct, which means that economic growth in a given country is primarily affected by its specific factors (direct effects) and also by factors relative to its neighbours (indirect spillover effects), which we think is logical. Finally, the results above are robust since they hold whether the contiguity weight matrix or the inverse distance weight matrix is used.

Table 8. Climate change and economic growth: a summary of results.

	Short-Run Effects			Long-Run Effects
	Direct	Indirect	Total	Direct	Indirect	Total
Full sample
$W_c$	−0.076	−0.007	−0.084	−0.086	−0.015	−0.101
$W_d$	−0.075	−0.062	−0.137	−0.083	−0.148	−0.232
Hottest countries
$W_c$	−0.729 ***	−0.008	−0.737 **	−0.729 ***	−0.115 *	−0.844 **
$W_d$	−0.694 *	−0.102	−0.796 *	−0.695 *	−0.651 *	−1.347 *
Coldest countries
$W_c$	−0.067	−0.003	−0.070	−0.066	−0.003	−0.069
$W_d$	−0.070	−0.024	−0.094	−0.069	−0.035	−0.105
Low–middle-income countries
$W_c$	−0.150 ***	−0.018 ***	−0.168 ***	−0.157 ***	−0.016 ***	−0.173 ***
$W_d$	−0.143 ***	−0.069 ***	−0.212 ***	−0.148 ***	−0.091 ***	−0.240 ***
High-income countries
$W_c$	0.060	0.014	0.074	0.084	0.030	0.114
$W_d$	0.061	0.065	0.126	0.084	0.108	0.193

***, **, and * represent the statistical significance at 1%, 5%, and 10%, respectively.

reports the direct, indirect, and total effects of climate change on economic growth for the full sample and the different subsamples in both the short- and long run using the contiguity or inverse distance weight matrices.

6. Concluding Remarks and Policy Recommendations

Climate change and global warming have been at the forefront of discussion among policymakers and scholars. There has been a growing focus in the research on the economic consequences of climate change. However, the existing literature ignored the spatial dimension of climate change, which results in an underestimation of the impact of climate change on economic growth. This research fills this gap by investigating the spatial effects of climate change, as measured by temperature change, on GDP per capita growth in 86 countries from 1980 to 2019. The specification is also augmented by a set of control variables: initial GDP, government expenditure, trade openness, foreign direct investment, human capital, and population. To check the robustness of the findings, we employ two different spatial matrices: the contiguity weight matrix and the inverse distance weight matrix.

The empirical Investigation begins by conducting a crucial preliminary analysis. Moran’s I spatial autocorrelation test confirms the presence of spatial autocorrelation, while the IPS and LLC unit root tests suggest that all variables are stationary at levels. Finally, the LM and Hausman tests indicate the suitability of the DSAR model and fixed effect estimator for estimating the spatial panel data model, respectively. We then estimate the dynamic spatial autoregressive model for the full sample of 86 countries and the different subsamples decomposed according to the climate regime (hottest vs. coldest) and the income level (low–middle-income vs. high-income). The analysis suggests the absence of direct and indirect effects of climate change on economic growth in the whole sample. These findings hold over the short- and long term using the two spatial matrices. When disaggregating the full sample according to the climate regime, the findings reveal that temperature change negatively affects economic growth only in the hottest countries. Economic growth in the coldest countries is not affected by climate change. Furthermore, the direct effects of climate change on economic growth in the hottest countries are observed during the short- and long run. However, the indirect spillover effect of climate change is significant only in the long run. Finally, the direct effect of climate change exceeds indirect effects in the long run, resulting in a total effect ranging between −0.844 and −1.347. The disaggregation of the sample according to income level also generates some interesting findings. The impact of temperature change on economic growth is only significant for low–middle-income countries. Moreover, the analysis suggests the presence of indirect spillover effects from climate change to economic growth over the short- and long term. However, the adverse long-run effects of climate change are higher than those observed during the short run. On the contrary, there is no evidence of direct and indirect effects of climate change on economic growth in high-income countries.

The outcomes of this study may help to suggest some methodological and policy recommendations. First, one could highlight the necessity of considering the spatial dimension when analysing the impact of climate change on economic growth. Ignoring the spatial spillover effect leads to underestimating the macroeconomic consequences of climate change. Second, given the existence of significant spillover effects, international cooperation is needed to design and implement climate change mitigation policies. International cooperation is mandatory as climate change in a given country affects economic growth in surrounding countries. Therefore, even if a country implements its climate mitigation strategies individually, it will still be harmed by climate change occurring in other countries. Finally, our findings reveal that the hottest countries and low–middle-income countries are those affected by climate change, both directly and indirectly. Consequently, technical and financial assistance should be provided to assist climate change adaptation and mitigation initiatives in the poorest and most vulnerable countries impacted by climate change.

Despite its contribution towards providing new insights into the spatial effects of climate change on economic growth, this study has some limitations. First, the study employs temperature variations as a proxy for climate change. Future research may overcome this limitation by using alternative measures, such as precipitation, sea level rise, and drought. Second, the study only examines the symmetric effects of climate change on economic growth. Future studies may be conducted on the asymmetric effects of climate change. Finally, combining the spatial effect and threshold effect models is also interesting and may provide new evidence on the macroeconomic effects of climate change.