Policy Research Working Paper 10466 Does Hotter Temperature Increase Poverty and Inequality? Global Evidence from Subnational Data Analysis Hai-Anh H. Dang Minh Cong Nguyen Trong-Anh Trinh Development Data Group & Poverty and Equity Global Practice June 2023 Policy Research Working Paper 10466 Abstract Despite a vast literature documenting the harmful effects of also finds negative effects of colder temperature on poverty climate change on various socio-economic outcomes, little and inequality. Yet, while poorer countries—particularly evidence exists on the global impacts of hotter temperature those in South Asia and Sub-Saharan Africa—are more on poverty and inequality. Analysis of a new global panel affected by climate change, household adaptation could dataset of subnational poverty in 134 countries finds that have mitigated some adverse effects in the long run. The a one-degree Celsius increase in temperature leads to a 9.1 findings provide relevant and timely inputs for the global percent increase in poverty, using the US$1.90 daily poverty fight against climate change as well as the current policy threshold. A similar increase in temperature causes a 1.4 debate on the responsibilities of richer countries versus percent increase in the Gini inequality index. The paper poorer countries. This paper is a product of the Development Data Group, Development Economics and the Poverty and Equity Global Practice. . It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http:// www.worldbank.org/prwp. The authors may be contacted at hdang@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Does Hotter Temperature Increase Poverty and Inequality? Global Evidence from Subnational Data Analysis Hai-Anh H. Dang, Minh Cong Nguyen, and Trong-Anh Trinh* JEL Classification: Q54; I32; O1 Key words: Climate change; temperature; poverty; inequality; subnational data * Dang (hdang@worldbank.org; corresponding author) is a senior economist with the Data Production and Methods Unit, Development Data Group, World Bank and is also affiliated with GLO, IZA, Indiana University, and International School, Vietnam National University, Hanoi; Nguyen (mnguyen3@worldbank.org) is a senior economist with the Global Poverty Practice, World Bank; Trinh (trong-anh.trinh@monash.edu) is a consultant with the Data Production and Methods Unit, Development Data Group, World Bank, and a research fellow at the Centre for Health Economics, Monash University, Australia. We would like to thank Carlo Azzarri, Edward Barbier, Romina Catavassi, Andrew Dabalen, Benjamin Davis, Stephane Hallegatte, David Johnston, Wojciech Kopczuk, Anke Leroux, Paulina Oliva, Paul Raschky, Russell Smyth, Jevgenijs Steinbuks, and participants at the Frontiers in Development Policy Conference (KDI) and seminars at Food and Agriculture Organization and Monash University for useful comments on earlier versions. We would like to thank Matthias Kalkuhl for helpful advice on data and Brenan Andre for assistance with the global maps. We would also like to thank the UK Foreign Commonwealth and Development Office (FCDO) for funding assistance through a Knowledge for Change (KCP) grant for the World Development Report 2021 “Data for Better Lives” and the Data and Evidence for Tackling Extreme Poverty (DEEP) Research Program. 1. Introduction The increasingly prominent threats of climate change have inspired a significant body of economic research on a variety of outcomes, such as agriculture (Deschênes and Greenstone, 2007; Schlenker and Roberts, 2009), labor productivity (Somanathan et al., 2021), human health (Deschênes and Greenstone, 2011), and crime and conflict (Burke et al., 2015a; Heilmann et al., 2021). In particular, since climate change could reduce economic growth (Dell et al., 2012; Newell et al., 2021), this might, in turn, translate into slower progress with poverty reduction. Furthermore, climate change may unevenly affect different countries and population groups with the most harmful consequences of income decline being borne by less affluent groups; these would likely result in increasing inequality both across and within countries (Diffenbaugh and Burke, 2019; Hsiang et al., 2019). A possible explanation for the lack of empirical evidence on the impacts of global warming is the challenge of obtaining appropriate measures of poverty and inequality. While household surveys—the main source of official poverty statistics—have become increasingly available, these surveys are still unavailable or infrequently collected in many countries, particularly in poorer regions.1 Another explanation is that poverty and inequality can widely vary within (and across) countries. Consequently, ignoring subnational variations could easily mask the dynamic relationship of these outcomes with climatic conditions, which have long been known to be location specific. Indeed, recent studies suggest that analysis using spatial aggregation of data at the country level may not reveal the true effects of climate change on economic growth, which can be improved with analysis using more disaggregated data at the subnational level (Damania et al., 2020; Kalkuhl and Wenz, 2020). 1 A recent survey by Beegle et al. (2016) indicates that just slightly more than half (i.e., 27) of the 48 countries in Sub-Saharan Africa had two or more comparable household surveys for the period between 1990 and 2012. Dang et al. (2019) find that a 10-percent increase in a country’s household consumption level is associated with almost one-third (i.e., 0.3) more surveys. 1 To further illustrate, we plot in Figure 1 poverty and inequality against temperature at the subnational level for Indonesia, a populous country with a major share of the global poor. This figure shows large degrees of subnational variation in both poverty and inequality. Poverty, as measured by the headcount poverty rate at US$1.90 a day, ranges from being relatively low in the Western regions (lowest rate of 0 percent) to quite considerable in the Eastern regions (highest rate of 34 percent) (Panel C). A similar pattern is seen with inequality, as measured by the Gini index, which ranges between 26 percent and 45 percent (Panel D). Within the country, average temperature also widely varies between 21◦C and 30◦C (Panel E). Such wide- ranging subnational variations are not revealed by simply looking at Indonesia’s country-level averages of poverty, inequality, and temperature (9 percent, 36 percent, and 25◦C, respectively), suggesting that an accurate assessment of the effects of global warming on poverty and inequality would require data analysis at the subnational level. In this study, we find strong and statistically significant global effects of both higher and lower temperature on poverty and inequality, employing different identification strategies on a novel global database of subnational poverty and inequality. Our (preferred) subnational fixed effects model shows that a one-degree Celsius (i.e., 1◦C) annual increase in temperature causes headcount poverty increases of 0.9, 1.8, and 2.3 percentage points, respectively, using the daily poverty lines of $1.90, $3.20, and $5.50 (which correspond to 9.1 percent, 9.0 percent, and 6.8 percent increases). The corresponding estimated effects using the long differences model are less pronounced at 0.5, 1.2, and 2.0 percentage point increases in poverty (which correspond to 5.3 percent, 6.1 percent, and 5.9 percent increases), suggesting household adaptation to gradual warmer temperature over time. Analysis of subnational inequality data suggests that a 1◦C rise in temperature leads to 0.8 and 1.4 percent increases in the Gini and Theil indices, respectively. 2 For both poverty and inequality, we find evidence that points to larger climate change effects at the subnational level than those estimated using more aggregated, country-level data, particularly in regions where temperature change has the largest effects. Our heterogeneity analysis further shows that countries in South Asia and Sub-Saharan Africa are more vulnerable to warmer temperature, but the effects of colder weather are also observed among countries in Europe and Central Asia. Our study makes several new contributions to the literature. First and most importantly, we offer the first global assessment of warmer temperature on both poverty and inequality, exploiting a novel global subnational panel database that we constructed based on the Global Subnational Atlas of Poverty (GSAP) (World Bank, 2021). Several attempts were made to understand the direct effects of global warming on poverty and inequality, but none examines these outcomes together.2 For example, analyzing cross-sectional household survey data from 24 Sub-Saharan African countries, Azzarri and Signorelli (2020) show that a one degree increase in long-term temperature is associated with a 2.8 percentage point increase in poverty. Paglialunga et al. (2022) use data from 150 countries and find that a one percent temperature increase is associated with a 0.5 percentage point increase in the Gini index.3 This lack of evidence poses an important, and perhaps quite urgent, challenge given the recent public debate of whether richer countries should take more responsibilities for the costs of climate change that correspond to their shares of the pollutants (Birnbaum et al., 2022; Popovich and Plumer, 2021). As an example, 80 percent of global greenhouse gas emissions are currently produced by G20 economies—the world’s largest economies—but these economies can only price 49 percent of CO2 emissions from energy use (OECD, 2021). 2 There are also a number of studies investigating the effects of natural disasters on temporary (transient) poverty (e.g., Sawada and Takasaki, 2017). Our study, in contrast, focuses on chronic poverty as a result of climate change. 3 Other studies mostly focus on a country-specific context. See also Karim and Noy (2016) and Hallegatte et al. (2020) for recent reviews of the literatures on climate change, natural disasters, and poverty. Furthermore, since we analyze poverty and inequality using the same source of official household survey data, our estimates are consistent (and more comparable) than estimates that are based on different data sources. 3 Second, we offer new, disaggregated data on headcount poverty estimates and inequality indices for 1,594 subnational areas in 134 economies from 2003 to 2019, based on the GSAP database which is generated using household income and consumption surveys that underlie countries’ official poverty statistics. This helps distinguish our study from the few existing cross-national studies that predominantly focused on country-level datasets, which, although informative, were not able to adequately capture the intricate subnational dynamics of poverty, inequality and temperature change. As part of our analysis, we make this new dataset publicly available for the first time. Our results show that analysis based on subnational data yields more accurate estimates of the impacts of temperature on poverty and inequality; consequently, this new subnational dataset can contribute to further and better research on climate change and poverty and inequality on a global scale. Finally, we add fresh evidence to the emerging literature on the distributional effects of climate change. While existing studies on other development outcomes mostly focus on areas with hotter temperature, far fewer studies investigate the effects of colder temperature. Yet, no study is currently available on these distributional effects for poverty and inequality. For example, Dell et al. (2012) show that both hot and cold deviations from the average temperature have similar effects on economic growth; Deschênes and Greenstone (2011) find more cold days to be associated with higher mortality. Most recently, Cook and Heyes (2020) find that outdoor cold temperature negatively impacts indoor cognitive performance. More evidence on the potentially adverse effects of colder temperature is important since, despite global warming, unusually colder weather has become more common in many countries in the past decades. Overall, our results indicate that the distributional effects across temperature ranges (as well as across subnational regions) should be considered together with longer-term effects of temperature change as inputs for designing more effective policies aiming at fighting climate change, poverty, and inequality. 4 This paper consists of six sections. We discuss the data in the next section, and the analytical framework in Section 3. In Section 4, we report on the estimation results for poverty (Section 4.1), inequality (Section 4.2), their nonlinear effects (Section 4.3), and further robustness checks and heterogeneity analysis (Section 4.4). We offer further analysis on potential mechanisms, projected impacts under future climate change, and some back-of-the- envelope cost-benefit analysis in Section 5 and finally conclude in Section 6. We provide additional results in Appendix A, describe the data in more detail in Appendix B, offer more robustness checks and heterogeneity results in Appendix C and additional analysis on the potential mechanisms and projected impacts of temperature in Appendix D. 2. Data The data used for our analysis are derived from multiple sources. We introduce a novel dataset that provides a granular perspective on poverty and inequality at the subnational level. In particular, we draw on the Global Subnational Atlas of Poverty (GSAP) (World Bank, 2021), a collaborative effort among different teams at the World Bank over a period of time. The GSAP is built on countries’ official household income (consumption) surveys, covering over 1,594 subnational units across 134 countries, with more than 90 percent of the data ranging from 2010 to 2019. In most cases, a subnational unit refers to a province or state (i.e., first- level administrative boundaries – ADM1) but can also be a group of regions determined by the specific sampling strategy of household surveys. For the main outcomes, we utilize the (headcount) poverty rate at US$1.90 a day, as estimated by the percentage of the population living on less than $1.90 a day at the 2011 purchasing power parities (PPP) prices.4 For richer analysis, we also employ other poverty lines of $3.20 and $5.50 a day. As alternative sources of poverty data, we also utilize two other 4 The data are accessible on the Harvard Dataverse depository at https://doi.org/10.7910/DVN/MLHFAF. 5 sources: i) country-level poverty data from the World Bank World Development Indicators (WDI), which is a widely used database for global poverty measures, and ii) subnational GDP from Kalkuhl and Wenz (2020) and Kummu et al. (2020), which we further convert to poverty data. Panel A of Figure 1 shows that Sub-Saharan Africa currently has the highest poverty rates and the poorest countries include Tanzania (51.3 percent), Mozambique (54.7 percent), and the Democratic Republic of Congo (72.9 percent). For inequality, we mostly employ the Gini index and Theil index, which are the most commonly used measures of income inequality. For robustness checks, we also use the distribution of income (consumption) shares held by each decile and calculate different percentile ratios, namely the 90/10 ratio, the 80/20 ratio, and the 90/40 ratio (i.e., the Palma ratio). All income measures are converted to real terms using 2011 PPP dollars. Panel B of Figure 1 provides a global map of income inequality at the subnational level, which shows substantial variation of inequality across regions within a country. We match our poverty and inequality data with the ERA5 satellite reanalysis data from the European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ECMWF). The ERA5 provides hourly estimates of several climate-related variables at a grid of approximately 0.25 longitude by 0.25 latitude degree resolution with data available since 1979 (Dell et al., 2014). An advantage of the ERA5 data is that it combines information from ground stations, satellites, weather balloons, and other inputs with a climate model, and therefore is less prone to station weather bias.5 For robustness tests, we use the global gridded data from Climate Research Unit of the University of East Anglia (CRU) available at 0.5◦ resolution. We provide a more detailed description of the data sources, including the list of the countries in each dataset and the summary statistics of the main variables in Appendix B. 5 Auffhammer et al. (2013) find high correlations between ERA5 reanalysis data and weather station data for temperature, which further supports our study focus on temperature. However, we appear not to have similar supportive evidence for rainfall, particularly in poorer countries with limited ground station data. 6 3. Empirical Specifications Different identification strategies were employed to estimate the effects of climate change on economic outcomes (Burke and Emerick, 2016; Dell et al., 2014; Kolstad and Moore, 2020). Early studies mostly used a cross-sectional approach utilizing spatial variation at a point of time, comparing outcomes between hot and cold areas (Mendelsohn et al., 1994; Schlenker et al., 2005). Yet, a key assumption when estimating the coefficients of the climate-related variable from cross-sectional models is that climate change is not correlated with other unobservable factors. Violation of this assumption could result in an omitted variables problem, causing the estimated coefficient of interest to be biased. Therefore, our first empirical approach identifies the effects of hotter temperature on poverty and inequality by estimating the following panel data model with fixed effects (FE): , = , + , + + + , (1) where , represents the poverty rate and inequality in location i in year t. Depending on the specific specification, location i is either country in the country-level analysis or subnational unit in the subnational analysis. , is the temperature variable, and the coefficient of interest is expected to be positive (i.e., global warming likely increases poverty and inequality). Following previous studies’ suggestion that precipitation and temperature are historically correlated and should be included in the same regression to obtain unbiased coefficients (Auffhammer et al., 2013; Dell et al., 2012), we control for precipitation (, ), measured in millimeters, in all the regressions. is the location (country or sub-national) fixed effects that controls for unobserved time-invariant factors that may be correlated with location-specific climate or economic patterns; is the year fixed effects that controls for unobserved temporal changes affecting poverty and inequality each year. We cluster the errors , at the specified location level to allow for potential serial correlation over time within a region (or a country). 7 For robustness, we also report Conley standard errors that allow for spatial correlation and arbitrary serial correlation in the error term (Conley, 1999). All the regressions are weighted with population weights at the subnational (country) level. While we can causally interpret in Equation (1), it is likely derived from short-run responses to temperature change given the nature of the annual panel data analyzed in this equation. Consequently, is not necessarily representative of households’ responses to temperature change in the longer term. In other words, long-term responses to temperature change may fundamentally differ from short-term responses to weather fluctuations because the former type of responses better accounts for potential household adaptation over time. Therefore, we address the shortcoming of Equation (1) by utilizing the long differences approach to estimate the accumulated effects of temperature change over longer periods of time (see, e.g., Burke and Emerick (2016)): ∆ = ∆ + ∆ + (2) In Equation (2), ∆ represents changes in poverty (or inequality) in the same location between two periods, and ∆ and ∆ are the corresponding changes in temperature and precipitation. To provide more stable estimates that are robust to data fluctuations in any single year, we use 3-year difference averages. That is, for all the variables in Equation (2) in our study period of 2003–2019, we analyze the differences between their averages of the earliest 3-year period 2003–2005 and their averages of the latest 3-year period 2017–2019 (e.g., ∑2019 2017 , ∑2005 2003 , ∆,2003−2019 = − ). Under the long differences approach, any time-invariant 3 3 location-specific factors are differenced out. As with Equation (1), the coefficients of interest, , is expected to be positive. In both the panel FE and long differences models, we assume the effects of temperature change to be in linear form. To allow for a more flexible functional form of temperature, we 8 further employ a temperature bin approach (e.g., Chen and Gong, 2021; Mullins and White, 2020) that offers estimates of nonlinear effects: , = ∑12 =1 , ,, + , + + + , (3) Specifically, we categorize daily temperature into 13 temperature bins, where each bin captures temperature change in increments of 3◦C (e.g., the first bin is [0◦C, less than 3◦C), the second bin is [3◦C, less than 6◦C), and so on). The two extremes of low and high temperature are respectively defined as less than 0◦C and greater than 33◦C. The temperature shock variable, ,, , reflects the number of days when the daily average temperature in a region is within a specific bin in a particular year. We use the most thermally comfortable temperature bin, which is 18–21◦C, as the reference group. The coefficients of interest , are thus interpreted as the effects of exchanging a day in the 18–21◦C reference bin with a day in the other bins. Finally, we also estimate the cumulative effects of temperature on poverty and inequality with a distributed lag model. Specifically, we capture the contemporaneous effects as well as the lag effects on each temperature bin for the last four periods. The distributed lag model is specified as , = ∑12 4 12 =1 , ,, + ∑=1 ∑=1 ,,− ,,− + , + + + , (4) 4. Results 4.1. Effects of temperature on poverty We start examining the effects of temperature change on poverty using the country-level analysis (Panel A) and subnational level analysis (Panel B) in Table 1. We use the WDI database for the country-level analysis and our newly constructed database for the subnational analysis. We analyze three poverty indicators using the daily poverty lines of $1.90, $3.20, and $5.50. For each outcome, we present the results of the fixed-effects panel model in Columns (1), (3), (5), followed by the results of the long differences model in Columns (2), (4), (6). In 9 both panels, the results are strongly statistically significant and confirm the negative effects of higher temperature on poverty for all the three different poverty lines. Yet, the estimates at the subnational-level analysis (Panel B) have stronger magnitudes than those derived from the country-level analysis (Panel A). The differences between these two sets of estimates are statistically significant, which is confirmed by the t-tests for equality of the estimated coefficients shown at the bottom of Table 1. This suggests that studies using spatial aggregation of data at the country level could mask the impacts of warmer temperature. This finding is also consistent with previous studies showing more pronounced effects of temperature on economic growth at the subnational level (e.g., Damania et al., 2020; Kalkuhl and Wenz, 2020). We subsequently focus on the subnational analysis for interpreting the estimation results. In particular, Column (1) of Panel B shows that a 1◦C increase in temperature causes a 0.9 percentage point increase in poverty (at the daily $1.90 poverty line). This equals a 9.1 percent increase in poverty using the mean poverty rate of 10.1 percent. For higher poverty lines, the impact magnitudes are higher in absolute terms (i.e., 1.8 percentage point and 2.3 percentage point increases for the daily poverty lines of $3.20 and $5.50, respectively) but are somewhat weaker in relative terms (i.e., the corresponding increases in poverty for these two poverty lines are respectively 9 percent and 6.8 percent). Using the long differences model on the same data, we show the estimated longer-term effects of temperature on poverty in Columns (2), (4), and (6). The results are qualitatively similar, indicating positive and strongly statistically significant effects of higher temperature on poverty. However, the long differences coefficient estimates are smaller in absolute value than the corresponding panel FE coefficient estimates. Specifically, a 1◦C increase in temperature is estimated to result in a poverty increase of 0.5 percentage points (5.3 percent) (using the daily poverty line $1.90) (Column 2). As shown by the t-tests at the bottom of Panel 10 B, the differences between the panel FE estimates and the long differences estimates are statistically significant, implying that longer-run household adaptation appears to have offset the negative short-run impacts of temperature on poverty by 0.4 percentage point (or 3.8 percent).6 These findings are consistent with previous studies that show the role of household adaptation in mitigating the negative effects of temperature on economic production, agriculture, and human capital (e.g., Chen and Gong, 2021; Graff Zivin et al., 2018; Kalkuhl and Wenz, 2020). While we focus on the impacts of temperature on poverty, Table 1 also reveals significant, but mixed, effects of precipitation. We find higher rainfall to be associated with lower poverty rate in the long differences model (e.g., Column 2, Panel B), but the opposite is found in the panel FE model (e.g., Column 1, Panel B).7 This ambiguity is, however, perhaps consistent with previous findings showing both negative impacts (Damania et al., 2020; Kotz et al., 2022) and positive impacts (Burke et al., 2015b; Dell et al., 2012) of rainfall on economic growth. 4.2. Effects of temperature on inequality We show in Table 2 the estimates on the effects of warmer temperature on income inequality at both the country-level and subnational level, which are strongly statistically significant. In particular, a 1◦C increase in temperature is estimated to result in an increase of 0.29 percentage point (0.8 percent) in the Gini index (Column 1 of Panel B), and an increase of 0.35 percentage point (1.4 percent) in the Theil index (Column 3 of Panel B). Similar to the estimation results for poverty (Table 1), the estimates at the subnational-level analysis (Panel B) are stronger than those obtained at the country-level analysis (Panel A). These results provide supportive 6 The long differences estimation results are based on a much smaller sample size compared with the panel FE model. To address this issue, we employ the same sample sizes used in the long-differences model and rerun regressions using the panel FE model. The results, presented in Table A1 (Appendix A), are qualitatively similar to those shown in Tables 1 and 2. 7 In addition, we employ alternative rainfall functions, including the quadratic term and the deviation of rainfall from the long-term mean, but still find similar results. 11 evidence for our earlier discussion that global warming might exacerbate income inequality because poorer countries or individuals could be more vulnerable to climate change. Indeed, our findings concur with previous studies, which find negative effects of hotter temperature on various economic outcomes. For example, recent studies by Hsiang (2010), Dell et al. (2014), and Deryugina and Hsiang (2014) have discovered that a 1◦C increase in temperature are associated with losses in, respectively, industrial output (2.5 percent), average country-level GDP per capita (1.0 percent), and county-average income per capita (1.7 percent). Our results are also qualitatively similar to, but offer slightly smaller estimates than, those found in Paglialunga et al. (2022), which show a one percent temperature increase to be associated with 0.5 percentage point increase in the Gini index. To investigate the potential long-run effects, we estimate Equation (2) and present the results in Columns (2) and (4) of Table 2. We document strong and statistically significant effects of hotter temperature on income inequality for both country-level analysis and subnational-level analysis and both measures of inequality. Again, we also find the effects at the subnational level to be stronger than those at the country level. Specifically, higher temperature by 1◦C is found to increase the Gini index and the Theil index by 0.35 (1 percent) and 0.59 (2.3 percent) percentage points (Panel B, Columns 2 and 4). Comparing the panel FE and long differences models, the magnitudes of effects appear larger for the latter. This may suggest inequality could accumulate over the longer term (i.e., intensification of negative effect). But we also note that the t-tests for the differences with the two models are only statistically significant for the Theil index with the subnational analysis, but not for the Gini index (and the country-level analysis). 4.3. Nonlinear effects 12 The effects of hotter temperature on poverty and income inequality discussed earlier are linear. To allow for a more flexible functional form of temperature, we assess the potential for nonlinear effects by specifying temperature as a series of indicator variables corresponding to 3◦C bins, where coefficients can be interpreted as the effects of falling into a given bin relative to the reference “comfortable” bin (i.e., 18-21◦C). We define hotter weather as temperature being in the top decile of the temperature range (i.e., greater than 27◦C), and colder weather as temperature being in the bottom decile of the temperature range (i.e., less than 6◦C). Figure 2 displays the point estimates and the 95% confidence intervals of these temperature bins, using Equation (3). Again, the results provide strong evidence for temperature effects, suggesting that one additional day of hotter temperature will lead to higher poverty and inequality, and the estimates are statistically significant at the 5 percent level. The magnitudes of the effects are generally consistent across hotter temperature bins. These results are consistent with our earlier findings of negative effects of warmer temperature. Furthermore, the results in Figure 2 also show that colder weather worsens poverty and inequality. Our findings concur with several studies finding negative effects of colder weather on productivity, health, and economic growth (Cook and Heyes, 2020; Dell et al., 2012; Deschênes and Moretti, 2009) and add fresh evidence for the impacts of colder weather on poverty and inequality. Since adaptation to colder weather differs from those to hotter weather, our results imply that the distributional effect of temperature should be considered when designing mitigation policies. Finally, we consider the model specification that controls for a series of lag of temperature bins (Equation 4). This approach offers insight into the cumulative effects of extreme temperature on income inequality. The estimated cumulative effects of temperature remain negative and slightly increase in magnitude compared to the contemporaneous effects as shown in Figure 2. In summary, our results suggest that when accounting for non-linearity of 13 temperature effects, we find strong evidence of the adverse impacts of both colder temperature and hotter temperature on poverty and inequality, and such effects are documented in both the short-term and long-term. 4.4. Robustness tests and heterogeneity analysis To investigate the robustness of the finding of negative temperature effects on poverty and inequality, we conduct a number of additional analyses. We briefly summarize the main results here and offer more detailed discussion in Appendix C. First, we use several variants of the panel FE and long differences models, which include adding country-specific linear time trends, controlling for temperature change, adding a quadratic or cubic term of temperature, adding an interaction term between temperature and temperature change, and using difference choices of window. We also use alternative thresholds to define hotter and colder days in the temperature bin approach. Next, we exploit different subsamples and alternative data sources and measures of temperature.8 Finally, we conduct a placebo test by using within-sample randomization of temperature. Overall, the results of these exercises remain similar to our main findings. We also offer further heterogeneity analysis across regions. We employ the temperature bin approach to consider the non-linearity of temperature effects and plot the results in Figure 3. It shows that rising temperature causes higher poverty (Panel A) and inequality (Panel B) in poorer regions such as Sub-Saharan Africa, Middle East and North Africa, and South Asia, but the effects are attenuated in the other richer regions. Furthermore, we also document negative effects of colder temperature on both outcomes, particularly in Europe and Central Asia. We further estimate temperature effects for each country. Plotting the results where each country’s 8 These include using (i) log of temperature; (ii) temperature measured in degrees Fahrenheit; (iii) the temperature data from CRU; (iv) the number of days that temperature is above 28 ◦C; (v) dropping regions with temperature being above that level; and (vi) temperature shock, defined as the difference between actual temperature and long- term temperature being greater (less) than 2 (-2) standard deviations (Appendix A, Table A4). 14 marker is proportional to its real GDP per capital, Figure A5 (Appendix A) shows that countries bearing the largest effect of global warming (e.g., Uganda, Ghana, and Mozambique) also tend to be poorer or located in poorer regions. Furthermore, we explore different country characteristics that might help mitigate temperature effects. Employing a democracy index that categorizes countries into democracies, authoritarian regimes, and hybrid regimes, we find that countries with democratic regimes appear to be less vulnerable to the impacts of global warming. Additionally, we observe that temperature effects are stronger in countries with higher shares of agriculture and are weaker in countries with higher shares of manufacturing. These findings suggest that institutions could play important roles in mitigating the effects of global warming on different countries and regions.9 5. Further analysis 5.1. Potential mechanisms and projected impacts under future climate change Having demonstrated strong evidence of temperature effects on poverty and inequality at the subnational level, we further explore agriculture as a potential mechanism. Agriculture plays an important role in poverty reduction for various reasons. The majority of the global poor live in rural areas where agriculture is the predominant form of economic activity and agricultural growth is more effective at reducing poverty than non-agricultural growth; moreover, poorer households are more vulnerable to increases in food prices (Hertel and Rosch, 2010; Hallegate et al., 2016). We analyze the global dataset of historical yields from Iizumi and Sakai (2020), which provides actual crop yields at 0.5° resolution for the period 1981-2016. Using the panel FE and long differences models as with Equations (1) and (2), we find negative effects of higher 9 We also investigate the role of information and communication technologies (ICTs) in poverty reduction by providing access to markets, decreasing transaction costs, and increasing income for a significant proportion of people living in developing countries (World Bank, 2016). Our results show that regions with better access to ICT are less vulnerable to the effects of higher temperature. 15 temperature on different crop yields including rice, maze, and soybean, as shown in Table A11 (Appendix A). Again, we find the long differences model estimates to be smaller than the panel FE model estimates, which are in line with previous studies showing potential adaptation in the long run (e.g., Chen and Gong, 2021). We also document the heterogenous effects of hotter temperature and discuss the results in Appendix D.10 We next provide projections of the effects of future temperature on poverty to better understand potential effects under different scenarios. We focus on two climate change scenarios—the RCP4.5 and RCP8.5—which are two extreme emission pathways that represent opposite ends of the climate spectrum depending on the uptake of renewable energy.11 Tables A13 and A14 (Appendix A) provide a summary of the projected changes for temperature for these scenarios in the short, medium, and long terms, where temperature can increase by between 2.6◦C and 6.0◦C in 2099. These temperature increases can result in poverty increases between 1.4 and 3.1 percentage points (i.e., 13.6 and 31.1 percent increases) (Appendix A, Table A13). Similarly, the simulated effects on inequality are estimated to be between 0.4 and 2.1 percentage point increases in Gini index (i.e., 1.2 and 5.9 percent increases) (Appendix A, Table A14). In both cases, the largest poverty and inequality increase would occur in the scenario without any countervailing strategies based on renewable energy to address climate change between 2021 and 2099.12 10 As another potential mechanism, we explore subnational migration flow between the period 2005 and 2010. Using a simple OLS regression, we find some suggestive evidence that hotter temperature could lead to more migration (Appendix A, Figure A8). However, since migration could help households obtain better economic opportunities and escape poverty, our estimation results could be considered as the net impacts of hotter temperature (after factoring in the beneficial effects of migration on poverty reduction). 11 The Representative Concentration Pathway (RCP) model captures future trends in climate change under alternative scenarios of human activities. RCP8.5 tracks emissions consistent with current trends (business as usual scenario in which greenhouse gas emissions go unchecked), while RCP4.5 considers a scenario with increased reliance on renewable energy and less reliance on coal-fired power (IPCC, 2021). 12 We also note that our projection of future impacts might be influenced by various other factors. For example, changes in ecosystems or global food production and sea level rises may amplify or lessen these effects, rendering the task of projecting the potential consequences of climate change extremely complicated. 16 5.2. Back-of-the-envelope cost analysis We next perform some rough calculations to help compare the costs of impeding temperature rise against that of eradicating poverty. Among the various measures for mitigating the effects of global warming, reducing carbon emissions such as carbon dioxide (CO2), methane, and nitrous oxide, is deemed crucial. The Intergovernmental Panel on Climate Change (IPCC) reports that limiting global warming to 1.5°C by 2030 would require reducing CO2 emissions by approximately 45 percent compared to 2010 levels and achieving net-zero emissions by 2050 (IPCC, 2018). This ambitious target is estimated to require a $US50 trillion investment in zero-carbon technology (Morgan Stanley, 2019). Using the estimated effects of temperature on poverty reported in Table 1 Column (2) (i.e., 0.53 percentage point increase in poverty per one-degree increase in temperature), we estimate that the poverty headcount ratio at US$1.90 a day would increase by 0.8 percentage point (an 8 percent increase) by 2030 if the temperature increases by 1.5°C. To counteract this rise in poverty, using Lakner et al.’s (2022) estimates, we calculate that the global GDP would have to increase by approximately 1.08 percent, or around $US1.12 trillion (based on the estimated global GDP of $US103.86 trillion in 2022). Although this estimated figure only amounts to 2 percent of the investment cost of $US50 trillion, our analysis does not consider the positive externalities of poverty elimination on other welfare outcomes, such as improved health and subjective well-being, or the beneficial impacts of poverty reduction on growth in the longer term (Thorbecke and Ouyang, 2022). To further our investigation, we estimate the allocation of costs for each country according to their respective contribution to warmer temperature. To provide a back-of-the-envelope estimate of this allocation, we use the share of CO2 emissions across countries from 1975 to 2022 (Friedlingstein et al., 2022). Table A15 (Appendix A) indicates a wide range of contribution shares for countries: while low-income countries should contribute less than 1 17 percent of the total costs, this figure increases to 10 percent, 29.7 percent, and 59.5 percent for lower-middle income, upper-middle income, and high-income countries, respectively. We plot in Figure A11 (Appendix A) more detailed estimates for each country. 6. Conclusions While there is growing evidence of harmful effects of climate change on macro-economic outcomes, little evidence exists regarding the impacts of warmer temperature on poverty and inequality on a global scale. A notable challenge is the absence of data at a disaggregated level that can allow for more accurate analysis of this relationship both within and across countries. Analyzing a new global panel dataset representative of subnational areas in 134 countries that we constructed, we find both hotter and colder temperature to result in higher poverty rate and inequality. We find stronger effects at the subnational level, which implies that country-level analysis does not reveal the true estimate of the global warming consequences. We also find significant, but smaller, effects of temperature in the long run, which suggests that households likely adapt to permanent changes in weather conditions. Our findings add to the ongoing discussion of richer countries’ responsibilities in mitigating the global effects of climate change. Over the last decades, poorer countries have been calling for compensation for the costs of climate change (also known as “climate reparations”) from wealthier nations, who are generally considered to be more responsible for global climate change. Our study contributes to this discussion by offering new global evidence that poorer regions are bearing the heaviest burden of global warming, and thus richer countries could provide further support in order to reduce the effects of climate change. The availability of subnational poverty data opens other avenues for future research. While our study provides supportive evidence that agriculture is an important factor influencing climate-induced poverty and inequality, alternative channels, such as civil conflicts and labor 18 productivity, could offer another direction. Further research on these topics would help provide policy inputs for more effective actions by different countries to address the challenge of global warming. 19 References Aghion, P., & Bolton, P. (1997). A theory of trickle-down growth and development. Review of Economic Studies, 64(2), 151-172. Arouri, M., Nguyen, C., & Youssef, A. B. (2015). Natural disasters, household welfare, and resilience: evidence from rural Vietnam. World Development, 70, 59–77. Auffhammer, M., Hsiang, S. M., Schlenker, W., & Sobel, A. (2013). Using weather data and climate model output in economic analyses of climate change. Review of Environmental Economics and Policy, 7(2), 181–198. Azzarri, C., & Signorelli, S. (2020). Climate and poverty in Africa South of the Sahara. World Development, 125, 104691. Barbier, E. B., & Hochard, J. P. (2018). Land degradation and poverty. Nature Sustainability, 1(11), 623-631. Beegle, K., Christiaensen, L., Dabalen, A., & Gaddis, I. (2016). Poverty in a rising Africa. World Bank Publications. Birnbaum, M., Chiu, A., & Kaplan, S. (2022, November 7). “U.N. climate talks off to tense start as nations feud over damage.” The Washington Post. Retrieved from https://www.washingtonpost.com/climate-environment/2022/11/07/un-climate- conference-damages/ Burke, M., Hsiang, S. M., & Miguel, E. (2015a). Climate and conflict. Annual Review of Economics, 7(1), 577–617. Burke, M., Hsiang, S. M., & Miguel, E. (2015b). Global non-linear effect of temperature on economic production. Nature, 527(7577), 235–239. Burke, M., & Emerick, K. (2016). Adaptation to climate change: Evidence from US agriculture. American Economic Journal: Economic Policy, 8(3), 106–40. Cappelli, F., Costantini, V., & Consoli, D. (2021). The trap of climate change-induced “natural” disasters and inequality. Global Environmental Change, 70, 102329. Chen, S., & Gong, B. (2021). Response and adaptation of agriculture to climate change: Evidence from China. Journal of Development Economics, 148, 102557. Conley, T. G. (1999). GMM estimation with cross sectional dependence. Journal of Econometrics, 92(1), 1–45. Cook, N., & Heyes, A. (2020). Brain freeze: outdoor cold and indoor cognitive performance. Journal of Environmental Economics and Management, 101, 102318. 20 Damania, R., Desbureaux, S., & Zaveri, E. (2020). Does rainfall matter for economic growth? Evidence from global sub-national data (1990–2014). Journal of Environmental Economics and Management, 102, 102335. Dang, H. A., Jolliffe, D., & Carletto, C. (2019). Data gaps, data incomparability, and data imputation: A review of poverty measurement methods for data‐scarce environments. Journal of Economic Surveys, 33(3), 757–797. Dell, M., Jones, B. F., & Olken, B. A. (2012). Temperature shocks and economic growth: Evidence from the last half century. American Economic Journal: Macroeconomics, 4(3), 66–95. Dell, M., Jones, B. F., & Olken, B. A. (2014). What do we learn from the weather? The new climate-economy literature. Journal of Economic Literature, 52(3), 740–98. Deryugina, T., & Hsiang, S. M. (2014). Does the environment still matter? Daily temperature and income in the United States (No. w20750). National Bureau of Economic Research. Deschênes, O., & Greenstone, M. (2007). The economic impacts of climate change: evidence from agricultural output and random fluctuations in weather. American Economic Review, 97(1), 354–385. Deschênes, O., & Moretti, E. (2009). Extreme weather events, mortality, and migration. Review of Economics and Statistics, 91(4), 659-681. Deschênes, O., & Greenstone, M. (2011). Climate change, mortality, and adaptation: Evidence from annual fluctuations in weather in the US. American Economic Journal: Applied Economics, 3(4), 152–85. Diffenbaugh, N. S., & Burke, M. (2019). Global warming has increased global economic inequality. Proceedings of the National Academy of Sciences, 116(20), 9808–9813. Dollar, D., & Kraay, A. (2002). Growth is Good for the Poor. Journal of Economic Growth, 7(3), 195-225. Friedlingstein, P., O'sullivan, M., Jones, M. W., Andrew, R. M., Gregor, L., Hauck, J., ... & Zheng, B. (2022). Global carbon budget 2022. Earth System Science Data, 14(11), 4811- 4900. Gazzotti, P., Emmerling, J., Marangoni, G., Castelletti, A., Wijst, K. I. V. D., Hof, A., & Tavoni, M. (2021). Persistent inequality in economically optimal climate policies. Nature Communications, 12(1), 1-10. Graff Zivin, J., Hsiang, S. M., & Neidell, M. (2018). Temperature and human capital in the short and long run. Journal of the Association of Environmental and Resource Economists, 5(1), 77-105. 21 Hallegatte, S., Bangalore, M., Bonzanigo, L., Fay, M., Kane, T., Narloch, U., Rozenberg, J., Treguer, D., Vogt-Schilb, A. (2016). Shock waves: Managing the impacts of climate change on poverty. Washington, DC: World Bank. Hallegatte, S., Vogt-Schilb, A., Rozenberg, J., Bangalore, M., & Beaudet, C. (2020). From poverty to disaster and back: A review of the literature. Economics of Disasters and Climate Change, 4(1), 223–247. Heilmann, K., Kahn, M. E., & Tang, C. K. (2021). The urban crime and heat gradient in high and low poverty areas. Journal of Public Economics, 197, 104408. Hertel, T. W., & Rosch, S. D. (2010). Climate change, agriculture, and poverty. Applied Economic Perspectives and Policy, 32(3), 355-385. Hsiang, S. M. (2010). Temperatures and cyclones strongly associated with economic production in the Caribbean and Central America. Proceedings of the National Academy of Sciences, 107(35), 15367-15372. Hsiang, S., Oliva, P., & Walker, R. (2019). The distribution of environmental damages. Review of Environmental Economics and Policy, 13(1), 83-103. Iizumi, T., & Sakai, T. (2020). The global dataset of historical yields for major crops 1981– 2016. Scientific Data, 7(1), 1–7. Intergovernmental Panel on Climate Change. (2018). Global warming of 1.5° C: An IPCC special report on the impacts of global warming of 1.5° C above pre-industrial levels and related global greenhouse gas emission pathways, in the context of strengthening the global response to the threat of climate change, sustainable development, and efforts to eradicate poverty. Intergovernmental Panel on Climate Change. In Masson-Delmotte et al. (eds.) Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 3−24. Intergovernmental Panel on Climate Change (IPCC). (2021). “Technical Summary.” In Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. In Masson- Delmotte et al. (eds.) Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 33−144. Kalkuhl, M., & Wenz, L. (2020). The impact of climate conditions on economic production. Evidence from a global panel of regions. Journal of Environmental Economics and Management, 103, 102360. Karim, A., & Noy, I. (2016). Poverty and natural disasters: a regression meta-analysis. Review of Economics and Institutions, 7(2), 26. 22 Kolstad, C. D., & Moore, F. C. (2020). Estimating the economic impacts of climate change using weather observations. Review of Environmental Economics and Policy, 14(1), 1–24. Kotz, M., Wenz, L., Stechemesser, A., Kalkuhl, M., & Levermann, A. (2021). Day-to-day temperature variability reduces economic growth. Nature Climate Change, 11(4), 319– 325. Kotz, M., Levermann, A., & Wenz, L. (2022). The effect of rainfall changes on economic production. Nature, 601(7892), 223–227. Kummu, M., Taka, M., & Guillaume, J. H. (2018). Gridded global datasets for gross domestic product and Human Development Index over 1990–2015. Scientific Data, 5(1), 1–15. Lakner, C., Mahler, D. G., Negre, M., & Prydz, E. B. (2022). How much does reducing inequality matter for global poverty? Journal of Economic Inequality, 20(3), 559-585. Mendelsohn, R., Nordhaus, W. D., & Shaw, D. (1994). The impact of global warming on agriculture: a Ricardian analysis. American Economic Review, 84(4), 753-771. Morgan Stanley. (2019). Decarbonization: The Race to Zero Emissions. Available at: https://www.morganstanley.com/ideas/investing-in- decarbonization#:~:text=The%20Race%20to%20Net%20Zero&text=Simply%20put%2C %20net%20zero%20is,balance%20between%20emissions%20and%20decarbonization. Mullins, J. T., & White, C. (2020). Can access to health care mitigate the effects of temperature on mortality? Journal of Public Economics, 191, 104259. Newell, R. G., Prest, B. C., & Sexton, S. E. (2021). The GDP-temperature relationship: implications for climate change damages. Journal of Environmental Economics and Management, 108, 102445. OECD. (2021). Carbon Pricing in Times of COVID-19: What Has Changed in G20 Economies? OECD, Paris. Paglialunga, E., Coveri, A., & Zanfei, A. (2022). Climate change and within-country inequality: New evidence from a global perspective. World Development, 159, 106030. Popovich, N., & Plumer, B. (2021, November 12). “Who Has the Most Historical Responsibility for Climate Change?.” The New York Times. Retrieved from https://www.nytimes.com/interactive/2021/11/12/climate/cop26-emissions- compensation.html Rodriguez-Oreggia, E., De La Fuente, A., De La Torre, R., & Moreno, H. A. (2013). Natural disasters, human development, and poverty at the municipal level in Mexico. Journal of Development Studies, 49(3), 442–455. 23 Sawada, Y., & Takasaki, Y. (2017). Natural disaster, poverty, and development: An introduction. World Development, 94, 2-15. Schlenker, W., Hanemann, W. M., & Fisher, A. C. (2005). Will US agriculture really benefit from global warming? Accounting for irrigation in the hedonic approach. American Economic Review, 95(1), 395-406. Schlenker, W., & Roberts, M. J. (2009). Nonlinear temperature effects indicate severe damages to US crop yields under climate change. Proceedings of the National Academy of Sciences, 106(37), 15594–15598. Somanathan, E., Somanathan, R., Sudarshan, A., & Tewari, M. (2021). The impact of temperature on productivity and labor supply: Evidence from Indian manufacturing. Journal of Political Economy, 129(6), 1797–1827. Thorbecke, E., & Ouyang, Y. (2022). Towards a virtuous spiral between poverty reduction and growth: Comparing sub Saharan Africa with the developing world. World Development, 152, 105776. World Bank. (2016). World development report 2016: Digital dividends. World Bank: Washington. World Bank. (2021). World Bank estimates based on data from the Global Subnational Atlas of Poverty, Global Monitoring Database. World Bank: Washington. 24 Figure 1: Global and subnational poverty and temperature 25 Notes: Poverty is measured by Global Subnational Poverty Headcount Ratio using the daily threshold of US$ 1.90. Inequality is measured by the Gini index. Temperature data is taken from the European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA-5). Data on simulated weather conditions at the subnational level are from the NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP). Poverty rate, inequality and temperature data are measured in the period 2003 – 2019. 26 Figure 2: Nonlinear effects of temperature on poverty and inequality Panel A: Poverty Panel B: Inequality Notes: The figures show the point estimates and their 95 percent confidence intervals of temperature bins using regression with rainfall and subnational fixed effects. Robust standard errors are clustered at the subnational level. Regressions are weighted by region population. The reference temperature bin is [18,21). The cumulative effects are obtained by estimating the model with four lags of weather variables. Hotter temperature and colder temperature are defined as temperature being in the top decile (i.e., greater than 27◦C) and bottom decile (i.e., less than 6◦C) of the temperature range, respectively. 27 Figure 3: Heterogeneity analysis Panel A: Effects of temperature on poverty by region 28 Panel B: Effects of temperature on inequality by region Notes: The figures show the point estimates and their 95 percent confidence intervals of temperature bins using regression with rainfall and subnational fixed effects. Robust standard errors are clustered at the subnational level. Regressions are weighted by region population. Temperature bins are identified by dividing regional average temperature into deciles with the temperature bin in the 6 th decide being the reference group. Hotter temperature and colder temperature are defined as temperature being in the top decile (i.e., greater than 27◦C) and bottom decile (i.e., less than 6◦C) of the temperature range, respectively. 29 Table 1: The effects of temperature on subnational poverty Poverty: $1.90/day $3.20/day $5.50/day Panel FE – Long Panel FE – Long Panel FE – Long all countries differences all countries differences all countries differences (1) (2) (3) (4) (5) (6) Panel A: Country-level analysis Temperature 0.681*** 0.235*** 1.085*** 0.409*** 1.397*** 0.528*** (0.107) (0.029) (0.165) (0.044) (0.233) (0.058) Precipitation -0.023 0.117 -0.021 -0.011 -0.011 -0.054 (0.018) (0.279) (0.031) (0.400) (0.039) (0.441) Country FE Yes No No No Yes No Year FE Yes No No No Yes No Mean dependent var. 7.288 7.288 15.399 15.399 26.593 26.593 Observations 464 95 464 95 464 95 Equality test (Panel p = 0.000 p = 0.000 p = 0.000 vs. long differences) Panel B: Subnation-level analysis Temperature 0.920*** 0.535*** 1.834*** 1.240*** 2.298*** 2.008*** (0.160) (0.039) (0.214) (0.070) (0.308) (0.108) Precipitation 0.236*** -0.365*** 0.373** -0.486*** -0.072 -0.177 (0.069) (0.092) (0.166) (0.142) (0.160) (0.248) Subnational FE Yes No Yes No Yes No Year FE Yes No Yes No Yes No Mean dependent var. 10.061 10.061 20.327 20.327 34.009 34.009 Observations 4,972 1,109 4,972 1,109 4,972 1,109 Equality test (Panel p = 0.000 p = 0.000 p = 0.016 vs. long differences) Equality test (country vs. p = 0.014 p = 0.000 p = 0.000 p = 0.000 p = 0.000 p = 0.000 subnational) Number of countries 134 95 134 95 134 95 Number of regions 1,594 1,109 1,594 1,109 1,594 1,109 Notes: Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. Poverty data are taken from the WDI (Panel A) and GSAP (Panel B). Poverty and weather variables in the long-differences model are measured by the difference between averages of the earliest 3-year period and averages of the latest 3-year period. The long differences estimation is based on cross-sectional data with a smaller sample size compared with panel data. The data for the country-level analysis and the subnational-level analysis comes from our newly constructed database. The equality test p-values show the t-test between the panel FE results vs. the long differences results, and the country analysis results vs. the subnational analysis results. *** p<0.01, ** p<0.05, * p<0.1. 30 Table 2: The effects of temperature on subnational inequality Inequality: Gini Theil Panel FE – Long Panel FE – Long all countries differences all countries differences (1) (2) (3) (4) Panel A: Country-level analysis Temperature 0.154*** 0.171*** 0.253*** 0.275*** (0.029) (0.040) (0.049) (0.063) Precipitation 0.002 0.025** -0.002 0.031 (0.007) (0.012) (0.012) (0.021) Country FE Yes No Yes No Year FE Yes No Yes No Mean dependent var. 34.406 34.406 23.417 23.417 Observations 423 90 423 90 Equality test (Panel vs. p = 0.726 p = 0.604 long differences) Panel B: Subnation-level analysis Temperature 0.285*** 0.349*** 0.350*** 0.592*** (0.086) (0.049) (0.032) (0.082) Precipitation -0.156** 0.474*** 0.267** 0.699*** (0.072) (0.100) (0.110) (0.170) Subnational FE Yes No Yes No Year FE Yes No Yes No Mean dependent var. 35.605 35.605 25.383 25.383 Observations 4,129 1,019 4,129 1,019 Equality test (Panel vs. p = 0.383 p = 0.029 long differences) Equality test (country p = 0.000 p = 0.000 p = 0.000 p = 0.000 vs. subnational) Number of countries 128 90 128 90 Number of regions 1,484 1,019 1,484 1,019 Notes: Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. Inequality data are taken from the GSAP. Inequality and weather variables in the long-differences model are measured by the difference between averages of the earliest 3-year period and averages of the latest 3-year period. The long differences estimation is based on cross-sectional data with a smaller sample size compared with panel data. The data for the country-level analysis and the subnational-level analysis comes from our newly constructed database. The equality test p-values show the t-test between the panel FE results vs. the long differences results, and the country analysis results vs. the subnational analysis results. *** p<0.01, ** p<0.05, * p<0.1. 31 Appendix A: Additional Tables and Figures Figure A1: Alternative temperature bin Panel A: 2-degree bin Panel B: 4-degree bin Panel C: 5-degree bin Notes: The figures show the point estimates and their 95 percent confidence intervals of temperature bins using regression with rainfall and subnational fixed effects. Robust standard errors are clustered at the subnational level. Regressions are weighted by region population. Hotter temperature and colder temperature are defined as temperature being in the top decile (i.e., greater than 27◦C) and bottom decile (i.e., less than 6◦C) of the temperature range, respectively. 32 Figure A2: Alternative measures of inequality Panel A: Contemporaneous effect Panel B: Cumulative effect Notes: The figure shows the point estimates and their 95 percent confidence intervals of temperature bins using regression with rainfall and subnational fixed effects. Robust standard errors are clustered at the subnational level. Regressions are weighted by region population. The reference temperature bin is [18,21). The cumulative effects in Panel B are obtained by estimating the model with four lags of weather variables. Hotter temperature and colder temperature are defined as temperature being in the top decile (i.e., greater than 27◦C) and bottom decile (i.e., less than 6◦C) of the temperature range, respectively. 33 Figure A3: Placebo test Panel A: Poverty Notes: Results of placebo exercise using 1,000 randomizations of regions. The outcome is poverty headcount ratio at $1.90. All regressions include precipitation and subnational fixed effects. Regressions are weighted by region population. 34 Panel B: Inequality Notes: Results of placebo exercise using 1,000 randomizations of regions. The outcome is Gini index. All regressions include precipitation and subnational fixed effects. Regressions are weighted by region population. 35 Figure A4: Heterogeneity analysis using regional temperature Panel A: Poverty 36 Panel B: Inequality Notes: The figure shows the point estimates and their 95 percent confidence intervals of temperature bins using regression with rainfall and subnational fixed effects. Robust standard errors are clustered at the subnational level. Regressions are weighted by region population. Temperature bins are identified by dividing regional average temperature into deciles with the temperature bin in the 6th decide being the reference group. Hotter temperature and colder temperature are defined as temperature being in the top decile (i.e., bin 10) and bottom decile (i.e., bin 1) of the temperature range, respectively. 37 Figure A5: The effects of temperature on poverty and inequality across countries adjusted by real GDP Panel A: Poverty Panel B: Inequality Notes: Poverty rate is measured by the Subnational Poverty Headcount Ratio at $1.90 a day. Inequality is measured by the Gini index. The figure shows the point estimates of temperature and the country dummies using regression with control variable and subnational fixed effects. Each country’s marker is proportional to its real GDP per capital using the WDI database (i.e., a larger size indicates a higher GDP per capita level). 38 Figure A6: The effects of temperature on poverty/inequality and agriculture Panel A: Poverty Panel B: Inequality Notes: The figure shows the point estimates of temperature effects on poverty and inequality (y-axis) and crop yield (x-axis) using regressions with control variable and subnational fixed effects. We then use an OLS regression of the poverty (inequality) effects on crop yield effects. Standard errors are in parentheses. Poverty is measured by the headcount ratio at $1.90 a day. Inequality is measured by the Gini index. Crop yield data is provided by Iizumi and Sakai (2020). 39 Figure A7: The effects of temperature on poverty and inequality by share of agriculture Panel A: Poverty Panel B: Inequality Notes: The figure shows the point estimates of temperature effects on poverty and inequality (y-axis) and share of agriculture in GDP (x-axis) using regressions with control variable and subnational fixed effects. Poverty rate is measured by the Subnational Poverty Headcount Ratio at $1.90 a day. Inequality is measured by the Gini index. Share of agriculture in GDP is taken from WDI database. 40 Figure A8: The effects of temperature on migration Notes: The figure shows the relationship between temperature (x-axis) and migration (y-axis) using OLS regressions with rainfall as control variable and country fixed effects. Standard errors are in parentheses. Migration is measured by the internal migration flows between 2005 and 2010 (in log). Migration data is available at https://hub.worldpop.org/. 41 Figure A9: Projected impacts of temperature on poverty Panel A: RCP 4.5 Panel B: RCP 8.5 Notes: Poverty is measured by Global Subnational Poverty Headcount Ratio using the daily threshold of US$ 1.90. Temperature data is taken from the European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA-5). Data on simulated weather conditions at the subnational level are from the NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP). The projection is estimated using the coefficient on the effects of temperature on poverty reported in Column (2) of Table 1 and the average temperature of during the period 1979 – 2019. 42 Figure A10: Projected impacts of temperature on inequality Panel A: RCP 4.5 Panel B: RCP 8.5 Notes: Inequality is measured by Gini index. Temperature data is taken from the European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA-5). Data on simulated weather conditions at the subnational level are from the NASA Earth Exchange Global Daily Downscaled Projections (NEX- GDDP). The projection is estimated using the coefficient on the effects of temperature on inequality reported in Column (2) of Table 2 and the average temperature of during the period 1979 – 2019. 43 Figure A11: Carbon emission cost allocation Notes: The cost allocation ($US million) is calculated by using each country's share of carbon emissions from 1975 to 2022 and the estimated cost of achieving net-zero emissions by 2050, based on data from Morgan Stanley (2019). 44 Table A1: Results of Panel fixed effects model using the same sample sizes as in long differences model Poverty measures Inequality measures $1.90/day $3.20/day $5.50/day Gini Theil (1) (2) (3) (4) (5) Panel A: Country-level analysis Temperature 0.498** 0.797*** 0.614** 0.137*** 0.220*** (0.200) (0.276) (0.238) (0.027) (0.043) Precipitation -0.004 0.005 -0.001 0.003 0.000 (0.011) (0.015) (0.012) (0.006) (0.014) Country FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Mean dependent var. 7.288 15.399 26.593 34.406 23.417 Observations 403 403 403 364 364 Panel B: Subnation-level analysis Temperature 0.846*** 1.807*** 2.293*** 0.264*** 0.261*** (0.144) (0.232) (0.308) (0.071) (0.029) Precipitation 0.213*** 0.200 -0.143 -0.119* 0.524*** (0.066) (0.150) (0.159) (0.069) (0.111) Subnational FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Mean dependent var. 10.061 20.327 34.009 35.605 25.383 Observations 4,225 4,225 4,225 3,400 3,400 Number of countries 95 95 95 90 90 Number of regions 1,109 1,109 1,109 1,109 1,109 Notes: Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. Poverty and inequality data are taken from the GSAP. The data for the country-level analysis and the subnational-level analysis comes from our newly constructed database. *** p<0.01, ** p<0.05, * p<0.1. 45 Table A2: Alternative specifications of panel model and long-difference model Panel A: Poverty Dependent variable: Panel model Long differences model Adding Adding Adding Adding Adding Adding country- Adding Adding temperature temperature temperature 4-year 5-year time- temperature Poverty rate at $1.90 specific temperature temperature squared squared term and interaction average average invariant interaction linear time change cubic term term temperature change term covariates term trend (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Temperature 0.867*** 0.879*** 0.873*** 0.719* 0.523*** 0.524*** -0.045 (0.129) (0.211) (0.321) (0.370) (0.129) (0.129) (0.127) ∆Temperature -0.046 0.044* 0.074 0.510*** 0.601*** 0.708*** 0.266*** (0.030) (0.024) (0.047) (0.037) (0.040) (0.054) (0.077) Temperature squared -0.007 -0.003 -0.002 -0.002 0.011** (0.006) (0.012) (0.002) (0.002) (0.005) Temperature cubic 0.000 (0.001) Temperature*∆Tempearture -0.003 0.155 (0.005) (0.115) Controlling for rainfall Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Subnational FE Yes Yes Yes Yes Yes Yes No No No No Year FE Yes Yes Yes Yes Yes Yes No No No No Number of countries 134 134 134 134 134 134 95 95 95 95 Number of regions 1,594 1,594 1,594 1,594 1,594 1,594 1,109 1,109 1,109 1,109 Observations 4,972 4,972 4,972 4,972 4,972 4,972 1,109 1,109 1,109 1,109 Notes: Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. Poverty rate is measured by the Subnational Poverty Headcount Ratio at $1.90 a day. Control variables in Column (9) are taken from Kalkuhl and Wenz (2020) which include cumulative oil gas, distance to coast, distance to river, and altitude. The long differences estimation is based on cross-sectional data with a smaller sample size compared with panel data. *** p<0.01, ** p<0.05, * p<0.1 46 Panel B: Inequality Dependent variable: Panel model Long differences model Adding Adding Adding Adding Adding Adding country- Adding Adding temperature temperature temperature 4-year 5-year time- temperature Gini index specific temperature temperature squared squared term and interaction average average invariant interaction linear time change cubic term term temperature change term covariates term trend (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Temperature 0.290*** 0.537*** 0.283*** 0.282*** 0.533*** 0.512*** 0.333 (0.103) (0.099) (0.087) (0.086) (0.101) (0.104) (0.388) ∆Temperature 0.209** 0.249* 0.099 0.309*** 0.339*** 0.421*** 0.977** (0.092) (0.135) (0.164) (0.045) (0.042) (0.052) (0.455) Temperature squared -0.001 -0.003 -0.003 -0.005 -0.027*** (0.004) (0.013) (0.007) (0.007) (0.007) Temperature cubic 0.000 (0.000) Temperature*∆Tempearture 0.012 -0.464*** (0.013) (0.070) Controlling for rainfall Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Subnational FE Yes Yes Yes Yes Yes Yes No No No No Year FE Yes Yes Yes Yes Yes Yes No No No No Number of countries 128 128 128 128 128 128 90 90 90 90 Number of regions 1,484 1,484 1,484 1,484 1,484 1,484 1,019 1,019 1,019 1,019 Observations 4,129 4,129 4,129 4,129 4,129 4,129 1,019 1,019 1,019 1,019 Notes: Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. Inequality is measured by Gini index. Control variables in Column (9) are taken from Kalkuhl and Wenz (2020) which include cumulative oil gas, distance to coast, distance to river, and altitude. The long differences estimation is based on cross-sectional data with a smaller sample size compared with panel data. *** p<0.01, ** p<0.05, * p<0.1 47 Table A3: Alternative measure of poverty – Multidimensional poverty Percentage of population deprived Monetary Educational Educational Headcount Electricity Sanitation Drinking water poverty attainment enrolment ratio (1) (2) (3) (4) (5) (6) (7) Temperature 0.737*** 0.726** 0.119 1.137*** 0.682* 0.097 1.374*** (0.203) (0.282) (0.172) (0.355) (0.373) (0.199) (0.395) Precipitation -0.011 -0.192*** 0.049 0.509** 0.461** -0.133 -0.155 (0.097) (0.064) (0.044) (0.210) (0.200) (0.117) (0.120) Subnational FE Yes Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Yes Observations 2,478 2,464 2,260 2,437 2,315 2,321 2,478 R-squared 0.419 0.089 0.232 0.259 0.146 0.047 0.412 Number of regions 1,179 1,179 1,109 1,172 1,163 1,169 1,179 Mean headcount 7.651 10.51 3.516 7.311 21.83 7.909 11.31 poverty rate (percent) Notes: Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. Column (1) measures the percentage of the population living on less than $2.15 a day at 2017 international prices, Column (2) measures the percentage of population deprived of primary educational attainment; Column (3) measures the percentage of population deprived of school enrolment; Column (4) measures the percentage of population deprived of electricity; Column (5) measures the percentage of population deprived of sanitation; Column (6) measures the percentage of population deprived of drinking water; Column (7) is the share of people who are considered multidimensionally deprived. *** p<0.01, ** p<0.05, * p<0.1 48 Table A4: Robustness test – Alternative measures of temperature Panel A: Poverty Dependent variable: Poverty rate at $1.90 Number of Dropping Log Temperature Temperature days subregions with Temperature temperature (oF) from CRU temperature temperature shock above 28 above 28 (1) (2) (3) (4) (5) (6) Temperature 2.218*** 0.434*** 0.760*** 0.070*** 0.666*** 0.337*** (0.481) (0.064) (0.118) (0.025) (0.111) (0.114) Controlling for Yes Yes Yes Yes Yes Yes rainfall Subnational FE Yes Yes Yes Yes Yes Yes Number of countries 134 134 134 134 134 134 Number of regions 1,594 1,594 1,594 1,594 1,594 1,594 Observations 5,090 5,090 5,059 4,209 4,856 5,090 Notes: Results of panel fixed effects model. Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. In Column (5), temperature shock is defined as the difference between actual temperature and long-term temperature being greater (less) than 2 (-2) standard deviation. *** p<0.01, ** p<0.05, * p<0.1 Panel B: Inequality Dependent variable: Gini index Number of Dropping Log Temperature Temperature days subregions with Temperature temperature (oF) from CRU temperature temperature shock above 28 above 28 (1) (2) (3) (4) (5) (6) Temperature 1.018*** 0.158*** 0.248*** 0.064*** 0.302*** 2.594*** (0.346) (0.048) (0.086) (0.013) (0.086) (0.472) Controlling for Yes Yes Yes Yes Yes Yes rainfall Subnational FE Yes Yes Yes Yes Yes Yes Number of countries 134 134 134 134 134 134 Number of regions 1,594 1,594 1,594 1,594 1,594 1,594 Observations 5,090 5,090 5,059 4,209 4,856 5,090 Notes: Results of panel fixed effects model. Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. In Column (5), temperature shock is defined as the difference between actual temperature and long-term temperature being greater (less) than 2 (-2) standard deviation. *** p<0.01, ** p<0.05, * p<0.1 49 Table A5: Robustness test – Alternative samples Panel A: Poverty Dependent variable: Poverty rate at $1.90 Dropping Spatially- Excluding 10 Excluding 10 countries Excluding Excluding corrected percent cold percent hot with few USA India Conley countries countries subregions S.E. (1) (2) (3) (4) (5) (6) Temperature 0.922*** 0.781*** 0.786*** 1.033*** 0.521*** 0.639*** (0.178) (0.116) (0.114) (0.158) (0.100) (0.089) Controlling for rainfall Yes Yes Yes Yes Yes Yes Subnational FE Yes Yes Yes Yes Yes Yes Number of countries 134 134 134 134 134 134 Number of regions 1,594 1,594 1,594 1,594 1,594 1,594 Observations 4,055 4,580 5,020 4,679 4,806 5,089 Notes: Results of panel fixed effects model. Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. *** p<0.01, ** p<0.05, * p<0.1 Panel B: Inequality Dependent variable: Gini index Dropping Spatially- Excluding 10 Excluding 10 countries Excluding Excluding corrected percent cold percent hot with few India Brazil Conley countries countries subregions S.E. (1) (2) (3) (4) (5) (6) Temperature 0.363** 0.288*** 0.386*** 0.274** 0.258*** 0.213** (0.142) (0.086) (0.086) (0.111) (0.085) (0.087) Controlling for rainfall Yes Yes Yes Yes Yes Yes Subnational FE Yes Yes Yes Yes Yes Yes Number of countries 128 128 128 128 128 128 Number of regions 1,157 1,449 1,457 1,329 1,352 1,484 Observations 3,175 4,059 3,994 3,789 3,865 4,149 Notes: Results of panel fixed effects model. Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. *** p<0.01, ** p<0.05, * p<0.1 50 Table A6: The effects of temperature on poverty – Subnational GDP analysis Poverty rate $1.90 Poverty rate $3.20 Poverty rate $5.50 Long Long Long Panel FE Panel FE Panel FE differences differences differences (1) (2) (3) (4) (5) (6) Temperature 0.148** 0.057*** 0.206** 0.120** 0.224** 0.105* (0.064) (0.021) (0.084) (0.057) (0.095) (0.060) Controlling for rainfall Yes Yes Yes Yes Yes Yes Subnational FE Yes No Yes No Yes No Year FE Yes No Yes No Yes No Number of countries 74 61 74 61 74 61 Number of regions 3,394 1,306 3,394 1,306 3,394 1,306 Observations 138,060 1,306 138,060 1,306 138,060 1,306 Adjusted R-squared 0.334 0.204 0.350 0.369 0.385 0.254 Mean headcount 16.847 16.847 30.152 30.152 45.559 45.559 poverty rate (percent) Notes: Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Poverty rate is calculated using subnational GDP from Kalkuhl and Wenz (2020) and the poverty lines of $1.90, $3.20, and $5.50. Poverty rates and weather variables in the long-differences model are measured by the difference between averages of the earliest 10-year period (1979–1988) and averages of the latest 10-year period (2009–2018). The long differences estimation is based on cross-sectional data with a smaller sample size compared with panel data. *** p<0.01, ** p<0.05, * p<0.1 51 Table A7: The effects of temperature on poverty – Grid-level analysis Dependent variable: Panel model Long differences model Poverty rate at $1.90 Baseline Extension Baseline Extension (1) (2) (3) (4) Temperature 0.102*** -2.046*** -0.0006*** (0.022) (0.060) (0.0001) ∆Tempearture 0.870*** 0.009*** 0.005*** (0.033) (0.001) (0.001) Temperature squared 0.092*** 0.0001*** (0.002) (0.00004) Controlling for rainfall Yes Yes Yes Yes Country FE Yes Yes No No Year FE Yes Yes No No Number of countries 82 82 82 82 Observations 1,115,478 1,072,575 42,903 42,903 R-squared 0.929 0.555 0.001 0.007 Notes: Robust standard errors in parentheses. Standard errors are clustered at the country level. Poverty incidence is calculated using subnational GDP from Kummu et al. (2018) and the poverty line from WDI. The long differences estimation is based on cross-sectional data with a smaller sample size compared with panel data. *** p<0.01, ** p<0.05, * p<0.1 52 Table A8: The effects of temperature on inequality – Country level analysis using alternative data from WDI and SWIID Gini – WDI Gini – SWIID Long Long Panel FE Panel FE differences differences (1) (2) (3) (4) Temperature 0.171*** 0.194*** 0.165*** 0.255*** (0.023) (0.033) (0.037) (0.040) Precipitation 0.013** 0.022* 0.006* 0.018* (0.005) (0.013) (0.003) (0.010) Country FE Yes No Yes No Year FE Yes No Yes No Mean dependent var. 38.295 38.295 38.445 38.445 Number of countries 128 90 128 90 Observations 1,505 90 3,781 90 Notes: Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. Inequality data in Columns (1)-(2) are taken from the World Development Indicators (WDI). Inequality data in Columns (3)-(4) are taken from the Standardized World Income Inequality Database (SWIID). Inequality and weather variables in the long-differences model are measured by the difference between averages of the earliest 3-year period and averages of the latest 3-year period. The long differences estimation is based on cross- sectional data with a smaller sample size compared with panel data. *** p<0.01, ** p<0.05, * p<0.1 53 Table A9: The effects of temperature on poverty and inequality – Heterogeneity analysis Poverty $1.90 Gini index (1) (2) Panel A: Regime type (Reference group: Democracy) Temperature*Hybrid regime 0.735* -0.004 (0.444) (0.003) Temperature*Authoritarian regime 1.395*** 0.008** (0.431) (0.003) Panel B: Location (Reference group: Countries near equator) Temperature* Countries near equator 0.943*** 0.011*** (0.293) (0.004) Panel C: Share of agriculture in GDP (Reference group: Low share) Temperature*High agriculture share 0.155*** 0.001*** (0.051) (0.000) Panel D: Share of manufacturing in GDP (Reference group: Low share) Temperature*High manufacturing share -0.076** -0.001*** (0.039) (0.000) Panel E: Share of trade in GDP (Reference group: Low share) Temperature*High trade share -0.005 0.000 (0.003) (0.000) Controlling for rainfall Yes Yes Subnational FE Yes Yes Notes: Results of panel fixed effects model. Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. *** p<0.01, ** p<0.05, * p<0.1 54 Table A10: Role of information and communication technologies (ICTs) as mediator Poverty $1.90 Gini index (1) (2) Panel A: ICT Development index Temperature* ICT Index -0.178*** -0.003*** (0.031) (0.001) Panel B: Internet 2G Temperature*Internet coverage -2.980*** -0.034*** (0.997) (0.013) Panel C: Internet 3G Temperature*Internet coverage -1.594*** -0.012*** (0.428) (0.004) Panel D: Internet 4G Temperature*Internet coverage -0.762** -0.019*** (0.297) (0.006) Controlling for rainfall Yes Yes Subnational FE Yes Yes Notes: Results of panel fixed effects model. Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. *** p<0.01, ** p<0.05, * p<0.1 55 Table A11: Effects of temperature on agriculture Crop yield Rice Maize Soybean Wheat Long Long Long Long Panel FE Panel FE Panel FE Panel FE differences differences differences differences (1) (2) (3) (4) (5) (6) (7) (8) Temperature -0.197*** -0.111*** -0.183*** -0.115*** -0.042*** -0.041*** 0.010 -0.000 (0.021) (0.014) (0.013) (0.008) (0.011) (0.011) (0.014) (0.010) Controlling for rainfall Yes Yes Yes Yes Yes Yes Yes Yes Subnational FE Yes No Yes No Yes No Yes No Year FE Yes No Yes No Yes No Yes No Mean crop yield 3.215 3.215 2.412 2.412 1.719 1.719 3.350 3.350 (tonnes/hectare) Number of countries 74 74 101 100 33 33 90 90 Number of regions 660 660 955 955 189 189 670 670 Observations 8,566 660 12,392 955 2,452 189 8,663 670 Equality test (Panel vs. p = 0.000 p = 0.000 p = 0.000 p = 0.628 long differences) Notes: Robust standard errors in parentheses. Standard errors are clustered at the subnational level. Regressions are weighted by region population. Crop yield data is provided by Iizumi and Sakai (2020). *** p<0.01, ** p<0.05, * p<0.1 56 Table A12: Effects of temperature on agriculture Crop yield Rice Maize Soybean Wheat (1) (2) (3) (4) Share of agriculture in GDP (Reference group: Low share) Temperature*High share -0.127*** -0.005 -0.119*** -0.035*** (0.015) (0.016) (0.012) (0.012) Controlling for rainfall Yes Yes Yes Yes Country FE Yes Yes Yes Yes Year FE Yes Yes Yes Yes Number of countries 42 70 16 68 Number of regions 641 915 178 634 Observations 9,967 14,259 2,778 9,635 R-squared 0.648 0.619 0.682 0.726 Notes: Robust standard errors in parentheses. Standard errors are clustered at the country level. Crop yield data is provided by Iizumi and Sakai (2020). *** p<0.01, ** p<0.05, * p<0.1 57 Table A13: Simulated effect of temperature on poverty Representative Concentration Representative Concentration Pathway (RCP) 4.5 Pathway (RCP) 8.5 2030 2050 2099 2030 2050 2099 Increase in temperature 1.388 1.984 2.631 1.235 2.114 5.999 Increase in poverty rate $1.90 0.743 1.061 1.408 0.661 1.131 3.209 Increase in poverty rate $3.20 1.721 2.460 3.262 1.531 2.621 7.439 Increase in poverty rate $5.50 2.787 3.984 5.283 2.480 4.245 12.046 Notes: Data on simulated weather conditions at the postcode level are from the NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP). The projection is estimated using the coefficient on the effects of temperature on poverty reported in Columns (2), (4), and (6) (Panel B) of Table 1. 58 Table A14: Simulated effect of temperature on inequality Representative Concentration Representative Concentration Pathway (RCP) 4.5 Pathway (RCP) 8.5 2030 2050 2099 2030 2050 2099 Increase in temperature 1.388 1.984 2.631 1.235 2.114 5.999 Increase in Gini index 0.484 0.692 0.918 0.431 0.738 2.094 Increase in Theil index 0.822 1.175 1.558 0.731 1.251 3.551 Notes: Data on simulated weather conditions at the postcode level are from the NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP). The projection is estimated using the coefficient on the effects of temperature on inequality reported in Columns (2), and (4) (Panel B) of Table 2. 59 Table A15: Cost allocation by income group Contribution Contribution Income group (%) ($US billion) High income 59.53% 29,762.56 Upper-middle income 29.72% 14,862.37 Lower-middle income 10.14% 5,070.30 Low income 0.61% 304.77 Total 100% 50,000 Notes: The income group is identified using the World Bank country classifications 2022-2023, available at: https://blogs.worldbank.org/opendata/new-world-bank- country-classifications-income-level-2022-2023 60 Appendix B: Data Description B1. Poverty data To implement the analysis, we assemble the most comprehensive data on poverty taken from the Global Subnational Atlas of Poverty (GSAP), produced by the Poverty and Equity Global Practice, coordinated by the Data for Goals (D4G) team, and supported by the six regional statistics teams in the Poverty and Equity Global Practice, and Global Poverty & Inequality Data Team (GPID) in the Development Economics Data Group (DECDG). All the teams are at the World Bank. For each survey data, the geographical area choice is based on the survey representativeness based on the sampling and sample design and survey documentation when available. For most of the database, surveys are representative at the first administrative level (ADM1) or statistical regions (areas) for the purpose of survey. On average, there are 14 subnational areas for a given country and year observation. For 18 small countries (13 percent), there is no subnational data available from the surveys, thus the national level data is used. Subnational poverty rates are calculated using official household or income surveys for the purpose of global poverty monitoring. Poverty rates are provided at the subnational level that is representative for the associated household or income survey used. Overall, cross-sectional poverty statistics are shown for about 5,500 subnational areas based on survey representativeness and availability of matched spatial geographic boundaries. Geographic boundaries must match the subnational regions in these surveys. In many cases, there is a one-to-one association between the regions in a household survey and the areas defined at an administrative level in the country. In cases where there is not a one-to-one association, geographic boundaries are altered to fit the representativeness of the surveys. In some cases, the geographic representation is at the level of “urban”, or “rural”. In these cases, subnational areas in the household survey are aggregated to levels that can be appropriately represented by boundaries. Several sources of geospatial files were leveraged to construct the GSAP: GADM, GAUL, NUTS, and customized spatial files. The choice of spatial files is based on more disaggregated availability and geographic alignment with household surveys. For example, NUTS spatial files are used prominently for the European countries in GSAP, since these files are developed and regulated by the EU. Building on the cross-sectional GSAP database, we construct a new database on poverty statistics based on almost 500 available household income/expenditure survey data in the Global Monitoring Database (GMD)14 for 139 economies, with more than 90 percent of the survey data ranging from 2010 to 2019. This database consists of panel data that are representative at 1,650 subnational areas. The number of countries across regions and over time are presented in Figures B1 and B2. As both country boundary and survey representativeness can change over time, constructing a panel data of poverty at area-level is not a simple task. When there is a change 14 The Global Monitoring Database (GMD) is the World Bank’s repository of multitopic income and expenditure household surveys used to monitor global poverty and shared prosperity. The household survey data are typically collected by national statistical offices in each country, and then compiled, processed, and harmonized. The process is coordinated by the Data for Goals (D4G) team and supported by the six regional statistics teams in the Poverty and Equity Global Practice. Global Poverty & Inequality Data Team (GPID) in Development Economics Data Group (DECDG) also contributed historical data from before 1990, and recent survey data from Luxemburg Income Studies (LIS). Selected variables have been harmonized to the extent possible such that levels and trends in poverty and other key sociodemographic attributes can be reasonably compared across and within countries over time. The GMD’s harmonized microdata are currently used in Poverty and Inequality Platform (PIP), World Bank’s Multidimensional Poverty Measures (WB MPM), the Global Database of Shared Prosperity (GDSP), and Poverty and Shared Prosperity Reports. 61 in the boundary over time or survey representativeness is different, efforts are needed to maintain a long panel of data to have comparable statistics spatially and over time. Such efforts could be (1) regroup areas to a new area that matches the previous definition of areas, or (2) a higher level of geographical disaggregation over time. In this version of the panel data, on average a country has data for 14 geographical areas over the period of 3 years. We also employ poverty data from different sources available at the country level and subnational level. The first is taken from the World Development Indicator (WDI) which provides different measures including the poverty headcount ratio, poverty gap, and number of poor at both international and national poverty lines. Our measures of interest are poverty headcount ratio at US$1.90 a day. It is calculated by the percentage of the population living on less than $1.90 a day at 2011 international prices. For richer analysis, we also use other poverty lines including the poverty headcount ratio at $3.20 and $5.50 a day. As an alternative source of subnational poverty, we exploit the annual GRP data provided by Kalkuhl and Wenz (2020), which is available from 1981 to 2016 for more than 1,500 regions in 77 countries worldwide. The dataset, however, includes only a few countries in Africa. We calculate the incidence of poverty by assuming the poverty line of $1.90, $3.20, and $5.50 for all countries in our sample.15 We also exploit annual gridded datasets for GDP per capita (PPP) from Kummu et al. (2020) which covers 26-year period from 1990 to 2015 for 82 countries. In this dataset, each grid cell is recorded at 5 arc-min resolution. We then apply a similar exercise as in the dataset of Kalkuhl and Wenz (2020) and measure the incidence of poverty at different thresholds. We present the list of countries in our datasets in Table B2. B2. Inequality data The GSAP dataset is unique in terms of country-time coverage and in the breadth of the measures of inequality covered. As the main outcomes, we use the most widely accepted measures of income distribution including the Gini index and Theil index. These indices are computed on the income available to households after government taxes and transfers, excluding indirect and value-added taxes, public services, and indirect government transfers. The Gini index measures the extent to which the distribution of income (or, in some cases, consumption expenditure) among individuals or households within an economy deviates from a perfectly equal distribution. A Lorenz curve plots the cumulative percentages of total income received against the cumulative number of recipients, starting with the poorest individual or household. The Gini index measures the area between the Lorenz curve and a hypothetical line of absolute equality, expressed as a percentage of the maximum area under the line. Thus, a Gini index of zero represents perfect equality, while an index of 100 implies perfect inequality. Similarly, the Theil index measures an entropic ‘distance’ the population is away from the ‘ideal’ egalitarian state of everyone having the same income. The Theil index ranges between zero and infinity, with zero representing an equal distribution and higher values representing a higher level of inequality. As alternative measures of income inequality, we use the distribution of income share held by each decile and calculate different percentile ratios, namely the 90/10 ratio, the 80/20 ratio, and the Palma ratio (90/40 ratio). All income measures are converted to real terms using 2011 Purchasing Power Parity (PPP) dollars for comparison across survey years. We also employ Gini data from different sources available at the country level. The first is taken from the World Bank Poverty and Inequality Platform (PIP). The data are based on primary household surveys obtained from government statistical agencies and World Bank country departments. In the case of high-income economies, they are mostly derived from the 15 To illustrate, we fix the poverty line for all regions in our sample and identify a region as poor if its gross income (per day) is below the poverty line. 62 Luxembourg Income Study database. As an alternative source of country-level inequality, we exploit the Standardized World Income Inequality Database (SWIID). SWIID provides standardized Gini income inequality measures for market and net outcomes based on the same concept, and thus allows the comparison of income inequality before and after redistribution by taxation and transfers over time. B3. Weather data We match our poverty and inequality data with the ERA5 satellite reanalysis data, which is taken from ECMWF. The ERA5 provides hourly estimates of several climate-related variables at a grid of approximately 0.25 longitude by 0.25 latitude degree resolution with data available since 1979 (Dell et al., 2014). We use air temperature and precipitation, both measured as annual averages, and map the grid spacings in ERA5 to the country/region in our poverty datasets. We follow previous studies and aggregate the gridded data to the region level by computing area-weighted averages (i.e., averaging all grid cells that fall into a region) (e.g., Heyes and Saberian, 2022; Kalkuhl and Wenz, 2020). Figure B3 provides a distribution of average temperature in our sample. It shows that most regions in our sample belong to the temperature range of between 24◦C and 28◦C. Another dataset that we use in the paper is the global gridded CRU data which provides monthly estimates at 0.5◦ resolution. The CRU data, however, is subject to absence of data in regions with less coverage of weather stations. Therefore, our main analysis exploits the ERA5 data which combines information from ground stations, satellites, weather balloons and other inputs with a climate model, and therefore is less prone to station weather bias (Auffhammer et al., 2013). To examine the impacts of future climate change on poverty and inequality, we obtain climate change prediction data from the NASA Earth Exchange (NEX) Global Daily Downscaled Projections (GDDP). The NEX data provides average temperature projections for the short term (2020–2040), the medium term (2041–2060) and the long term (2061–2099). We select the representative carbon pathway RCP8.5 as a benchmark scenario of unmitigated future warming (van Vuuren et al., 2011). It represents the ensemble average of all global climate models contributing to CMIP5, the Coupled Model Intercomparison Project phase 2010–2014 that informed the fifth assessment report of the Intergovernmental Panel on Climate Change. RCP8.5 corresponds to an expected increase of 4.3◦C in global mean surface temperature by 2100 relative to pre-industrial levels (Stocker et al., 2013). For comparison purpose, we also consider the RCP4.5 scenario with increased reliance on renewable energy and less reliance on coal-fired power. B4. Other data To examine the role of agriculture as the mechanism, we utilize annual production of four major crops (maize, wheat, soybean, rice) available from Iizumi and Sakai (2020). The dataset records global gridded data of annual crop yields, measured in tonnes/hectare, at 0.5◦ resolution and covers the period 1982–2015. The dataset was created by combining agricultural census data, satellite remote sensing and information on crop calendar and crop harvested area. Although the data include only four main crops, thereby partly limiting our analysis, the trade- off permits us to assemble consistent long panel data. Finally, in some specifications, we exploit data from different sources including type of regime from The Economist Intelligence, broadband internet coverage provided by Collins Bartholomew’s Mobile Coverage Explorer, and other country-level characteristics (i.e., population density, elevation, distance to the nearest coast, and concentration of Particulate matter of 2.5 micrometers or smaller – PM2.5) from the NASA Socioeconomic Data and Applications Center (SEDAC). We provide descriptions and summary statistics of all variables in Table B1. 63 References Auffhammer, M., Hsiang, S. M., Schlenker, W., & Sobel, A. (2013). Using weather data and climate model output in economic analyses of climate change. Review of Environmental Economics and Policy, 7(2), 181–198. Dell, M., Jones, B. F., & Olken, B. A. (2014). What do we learn from the weather? The new climate-economy literature. Journal of Economic Literature, 52(3), 740–98. Heyes, A., & Saberian, S. (2022). Hot Days, the ability to Work and climate resilience: Evidence from a representative sample of 42,152 Indian households. Journal of Development Economics, 155, 102786. Iizumi, T., & Sakai, T. (2020). The global dataset of historical yields for major crops 1981– 2016. Scientific Data, 7(1), 1–7. Kalkuhl, M., & Wenz, L. (2020). The impact of climate conditions on economic production. Evidence from a global panel of regions. Journal of Environmental Economics and Management, 103, 102360. Kummu, M., Taka, M., & Guillaume, J. H. (2018). Gridded global datasets for gross domestic product and Human Development Index over 1990–2015. Scientific Data, 5(1), 1–15. 64 Figure B1: Number of economies across World Bank regions 65 Figure B2: Number of areas over time 66 Figure B3: The effects of temperature on poverty by region Notes: Temperature data is taken from the European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA-5). 67 Table B1: Data sources and summary statistics Country Variable Descriptions Obs. No. Mean S.D. Min Max No. National poverty rate (1979–2019) (percent) Source: The World Bank (https://datacatalog.worldbank.org/home) Poverty rate $1.90 Poverty Headcount Ratio at US$ 1.90 a day 134 464 7.288 14.814 0.000 78.841 Poverty rate $3.20 Poverty Headcount Ratio at US$ 3.20 a day 134 464 15.399 23.975 0.000 91.518 Poverty rate $5.50 Poverty Headcount Ratio at US$ 5.50 a day 134 464 26.593 32.088 0.051 97.485 Subnational poverty rate (Global Subnational Atlas of Poverty – GSAP) (percent) Source: The World Bank (https://datacatalog.worldbank.org/home) Poverty rate $1.90 Poverty Headcount Ratio at US$ 1.90 a day 134 4,972 10.061 19.389 0.000 98.010 Poverty rate $3.20 Poverty Headcount Ratio at US$ 3.20 a day 134 4,972 20.327 28.482 0.000 99.724 Poverty rate $5.50 Poverty Headcount Ratio at US$ 5.50 a day 134 4,972 34.009 34.700 0.000 100.000 Subnational poverty rate (Source: Kalkuhl and Wenz, 2020) Poverty rate using average gross daily income being Poverty at $1.90 77 3,394 20.443 37.990 0.000 100.000 below US$ 1.90 a day Poverty rate using average gross daily income being Poverty at $3.20 77 3,394 34.075 44.185 0.000 100.000 below US$ 3.20 a day Poverty rate using if average gross daily income Poverty at $5.50 77 3,394 57.450 46.434 0.000 100.000 being below US$ 5.50 a day Subnational poverty rate (Source: Kummu et al., 2018) Poverty rate using average gross daily income being Poverty at $1.90 82 1,811,394 24.245 42.857 0.000 100.000 below US$ 1.90 a day Poverty rate using average gross daily income being Poverty at $3.20 82 1,811,394 32.000 46.648 0.000 100.000 below US$ 3.20 a day Poverty rate using average gross daily income being Poverty at $5.50 82 1,811,394 55.000 49.749 0.000 100.000 below US$ 5.50 a day National inequality WDI (Source: The World Bank - https://datacatalog.worldbank.org/home) 68 Gini Gini index (%) 128 423 34.406 6.980 22.968 59.777 Theil Theil index (%) 128 423 23.417 10.882 8.824 70.786 WDI (Source: The World Bank - https://datacatalog.worldbank.org/home) Gini Gini index (%) 128 423 38.295 9.062 20.700 65.800 SWIID (Source: Standardized World Income Inequality Database - https://fsolt.org/swiid/) Gini Gini index (%) 128 423 38.445 8.637 17.900 65.400 Subnational inequality (GSAP) Source: The World Bank (https://datacatalog.worldbank.org/home) Gini Gini index 128 4,129 35.627 8.045 13.371 66.448 Theil Theil index 128 4,129 25.421 13.966 3.359 192.672 Ratio of the income of the 10% richest to that of the 90/10 ratio 128 4,100 2.940 9.837 0.000 131.202 10% poorest. Ratio of the income of the 20% richest to that of the 80/20 ratio 128 4,099 2.627 7.712 0.000 106.748 20% poorest. Ratio of the income of the 10% richest to that of the Palma ratio 128 4,150 0.434 1.933 0.000 64.359 40% poorest. Satellite weather data (1979–2019) Source: European Union’s Copernicus programme (https://sentinels.copernicus.eu/web/sentinel/missions/sentinel-5p) Temperature Average temperature (C) 134 5,090 18.185 7.996 -9.417 30.790 Rainfall Average rainfall (mm) 134 5,090 3.880 3.178 0.006 34.882 Source: Climatic Research Unit (https://crudata.uea.ac.uk/cru/data/hrg/) Temperature Average temperature (C) 134 5,090 18.328 8.030 -11.082 30.426 Crop yield data Source: Iizumi and Sakai (2020) Rice Average crop yield (1981–2016) 45 10,257 3.215 3.041 0.000 22.314 Maize Average crop yield (1981–2016) 76 14,870 2.412 2.480 0.000 27.743 Soybean Average crop yield (1981–2016) 19 2,953 1.719 1.494 0.000 9.518 Wheat Average crop yield (1981–2016) 66 10,178 3.350 3.142 0.000 15.636 Variables used in heterogeneity analysis 69 Regime type in 2018 (Source: The Economist - https://www.eiu.com/n/) Democracy =1 if democracy score more than 7 126 3,945 0.193 0.394 0.000 1.000 Hybrid =1 if democracy score between 4 and 7 126 3,945 0.515 0.500 0.000 1.000 Authoritarian =1 if democracy score less than 4 126 3,945 0.292 0.455 0.000 1.000 Share of agriculture in GDP (Source: The World Bank - https://datacatalog.worldbank.org/home) =1 if share of agriculture in GDP less than 10 132 4,011 0.605 0.489 0.000 1.000 Low share percent =1 if share of agriculture in GDP equal to or 132 4,011 0.395 0.489 0.000 1.000 High share greater than 10 percent Share of manufacturing in GDP (Source: The World Bank - https://datacatalog.worldbank.org/home) =1 if share of manufacturing in GDP less than 10 132 3,911 Low share percent 0.692 0.462 0.000 1.000 =1 if share of manufacturing in GDP equal to or 132 3,911 High share greater than 10 percent 0.308 0.462 0.000 1.000 Share of trade in GDP (Source: The World Bank - https://datacatalog.worldbank.org/home) Low share =1 if share of trade in GDP less than 10 percent 132 3,924 0.632 0.482 0.000 1.000 =1 if share of trade in GDP equal to or greater than 132 3,924 0.368 0.482 0.000 1.000 High share 10 percent Broadband internet (Source: https://www.collinsbartholomew.com/) ICT ICT Development Index 118 3,828 5.045 1.833 1.040 8.980 2G Internet coverage at subnational level 130 3,955 0.913 0.161 0.000 1.000 3G Internet coverage at subnational level 123 3,337 0.809 0.264 0.000 1.000 4G Internet coverage at subnational level 94 1,861 0.781 0.331 0.000 1.000 70 Table B2: List of economies Kalkuhl and Wenz Kummu et al. No. Region GSAP (2020) (2018) 1 East Asia & Pacific Australia Australia Australia 2 East Asia & Pacific China China 3 East Asia & Pacific Fiji 4 East Asia & Pacific Indonesia Indonesia Indonesia 5 East Asia & Pacific Japan Japan Japan 6 East Asia & Pacific Korea, Rep. 7 East Asia & Pacific Lao PDR Lao PDR 8 East Asia & Pacific Malaysia Malaysia Malaysia 9 East Asia & Pacific Mongolia Mongolia Mongolia 10 East Asia & Pacific Myanmar 11 East Asia & Pacific Papua New Guinea 12 East Asia & Pacific Philippines Philippines Philippines 13 East Asia & Pacific Thailand Thailand Thailand 14 East Asia & Pacific Timor-Leste 15 East Asia & Pacific Tonga Taiwan, 16 East Asia & Pacific China 17 East Asia & Pacific Vanuatu 18 East Asia & Pacific Vietnam Vietnam Vietnam 19 Europe & Central Asia Albania Albania Albania 20 Europe & Central Asia Armenia 21 Europe & Central Asia Austria Austria Austria 22 Europe & Central Asia Azerbaijan Azerbaijan 23 Europe & Central Asia Belarus Belarus 24 Europe & Central Asia Belgium Belgium Belgium Bosnia and Bosnia and 25 Europe & Central Asia Herzegovina Herzegovina 26 Europe & Central Asia Bulgaria Bulgaria Bulgaria 27 Europe & Central Asia Croatia Croatia Croatia 28 Europe & Central Asia Cyprus 29 Europe & Central Asia Czechia Czechia Czechia 30 Europe & Central Asia Denmark Denmark Denmark 31 Europe & Central Asia Estonia Estonia Estonia 32 Europe & Central Asia Finland Finland Finland 33 Europe & Central Asia France France France 34 Europe & Central Asia Georgia Georgia Georgia 35 Europe & Central Asia Germany Germany Germany 36 Europe & Central Asia Greece Greece Greece 37 Europe & Central Asia Hungary Hungary Hungary 38 Europe & Central Asia Iceland 39 Europe & Central Asia Ireland Ireland Ireland 71 40 Europe & Central Asia Italy Italy Italy 41 Europe & Central Asia Kazakhstan Kazakhstan Kazakhstan 42 Europe & Central Asia Kosovo 43 Europe & Central Asia Kyrgyz Republic 44 Europe & Central Asia Latvia Latvia Latvia 45 Europe & Central Asia Lithuania Lithuania Lithuania 46 Europe & Central Asia Luxembourg 47 Europe & Central Asia Moldova 48 Europe & Central Asia Montenegro 49 Europe & Central Asia Netherlands Netherlands Netherlands 50 Europe & Central Asia North Macedonia 51 Europe & Central Asia Norway Norway Norway 52 Europe & Central Asia Poland Poland Poland 53 Europe & Central Asia Portugal Portugal Portugal 54 Europe & Central Asia Romania Romania Romania 55 Europe & Central Asia Russian Federation 56 Europe & Central Asia Serbia Serbia 57 Europe & Central Asia Slovak Republic 58 Europe & Central Asia Slovenia Slovenia Slovenia 59 Europe & Central Asia Spain Spain Spain 60 Europe & Central Asia Sweden Sweden Sweden 61 Europe & Central Asia Switzerland Switzerland Switzerland 62 Europe & Central Asia Tajikistan 63 Europe & Central Asia Türkiye Türkiye 64 Europe & Central Asia Ukraine Ukraine Ukraine 65 Europe & Central Asia United Kingdom United Kingdom 66 Europe & Central Asia Uzbekistan Uzbekistan Uzbekistan 67 Latin America & Caribbean Argentina Argentina Latin America & 68 Bolivia Bolivia Bolivia Caribbean Latin America & 69 Brazil Brazil Brazil Caribbean Latin America & 70 Chile Chile Chile Caribbean Latin America & 71 Colombia Colombia Colombia Caribbean Latin America & 72 Costa Rica Costa Rica Caribbean Latin America & Dominican 73 Dominican Republic Caribbean Republic Latin America & 74 Ecuador Ecuador Ecuador Caribbean Latin America & 75 El Salvador Caribbean 76 Latin America & Caribbean Guatemala Guatemala Latin America & 77 Haiti Caribbean 72 Latin America & 78 Honduras Honduras Honduras Caribbean Latin America & 79 Mexico Mexico Mexico Caribbean Latin America & 80 Nicaragua Caribbean Latin America & 81 Panama Panama Panama Caribbean Latin America & 82 Paraguay Paraguay Paraguay Caribbean Latin America & 83 Peru Peru Peru Caribbean 84 Latin America & Caribbean Uruguay Uruguay Middle East & North 85 Djibouti Africa Middle East & North 86 Egypt, Arab Rep. Africa Middle East & North 87 Iran, Islamic Rep. Africa Middle East & North 88 Iraq Africa Middle East & North 89 Israel Israel Africa Middle East & North 90 Jordan Jordan Africa Middle East & North 91 Lebanon Lebanon Africa Middle East & North 92 Malta Africa Middle East & North 93 Morocco Morocco Morocco Africa Middle East & North 94 Tunisia Africa United Arab 95 Middle East & North Africa Emirates Middle East & North 96 West Bank and Gaza Africa Middle East & North Yemen, 97 Africa Rep. 98 North America Canada Canada Canada United 99 North America United States States 100 South Asia Bangladesh Bangladesh 101 South Asia Bhutan 102 South Asia India India India 103 South Asia Maldives 104 South Asia Nepal 105 South Asia Pakistan Pakistan Pakistan 106 South Asia Sri Lanka 107 Sub-Saharan Africa Angola 108 Sub-Saharan Africa Benin Benin 73 109 Sub-Saharan Africa Botswana 110 Sub-Saharan Africa Burkina Faso 111 Sub-Saharan Africa Burundi 112 Sub-Saharan Africa Cabo Verde 113 Sub-Saharan Africa Cameroon Cameroon 114 Sub-Saharan Africa Central African Republic 115 Sub-Saharan Africa Chad 116 Sub-Saharan Africa Comoros 117 Sub-Saharan Africa Congo, Dem. Rep. 118 Sub-Saharan Africa Congo, Rep. 119 Sub-Saharan Africa Côte d'Ivoire 120 Sub-Saharan Africa Eswatini 121 Sub-Saharan Africa Ethiopia Ethiopia 122 Sub-Saharan Africa Gabon 123 Sub-Saharan Africa Gambia, The 124 Sub-Saharan Africa Ghana Ghana 125 Sub-Saharan Africa Guinea 126 Sub-Saharan Africa Guinea-Bissau 127 Sub-Saharan Africa Kenya Kenya Kenya 128 Sub-Saharan Africa Lesotho 129 Sub-Saharan Africa Liberia 130 Sub-Saharan Africa Madagascar 131 Sub-Saharan Africa Malawi Malawi 132 Sub-Saharan Africa Mali 133 Sub-Saharan Africa Mauritius Mozambiqu 134 Sub-Saharan Africa Mozambique Mozambique e 135 Sub-Saharan Africa Namibia Namibia 136 Sub-Saharan Africa Niger 137 Sub-Saharan Africa Nigeria 138 Sub-Saharan Africa Rwanda 139 Sub-Saharan Africa São Tomé and Principe 140 Sub-Saharan Africa Senegal Senegal 141 Sub-Saharan Africa Seychelles 142 Sub-Saharan Africa Sierra Leone 143 Sub-Saharan Africa South Africa South Africa South Africa 144 Sub-Saharan Africa Sudan 145 Sub-Saharan Africa Tanzania Tanzania Tanzania 146 Sub-Saharan Africa Togo 147 Sub-Saharan Africa Uganda Uganda 148 Sub-Saharan Africa Zambia Zambia 149 Sub-Saharan Africa Zimbabwe 74 Appendix C: Further robustness checks and heterogeneity analysis C1. Robustness checks In this section, we explore the robustness of our results in a number of different ways. We start with the results of panel model and long difference model presented in Tables 1 and 2 and show that our results are broadly consistent when using alternative model specifications. First, we estimate several alternate specifications to assuage the reader of misspecification concerns. These are presented in Panels A and B of Table A2 (Appendix A). Our panel model with fixed effects represents a substantial improvement over the standard cross-sectional regression, but it may also be subject to bias if there are unobservable, time-varying differences across countries. We show that our estimates are insensitive to the inclusion of country specific time trends (Column 1). Another concern is related to misspecification of the functional form of temperature. Therefore, from Columns (2) to (6), we employ different functional forms of temperature including controlling for temperature change, quadratic term and cubic term of temperature, and an interaction term between temperature and temperature change. Results of these exercises strengthen our main findings. Similarly, we also apply different variants of the long differences model and present the results in Columns (7)-(10) of Table A2 (Panels A and B). We first check whether our results remain robust when using different choices of window length (i.e., 4-year and 5-year period). The results in Columns (7) and (8) show that our findings are not sensitive to the alternative windows. In Column (9), we add a number of time-invariant covariates at the regional level including cumulative oil gas, distance to coast, distance to river, and altitude. Finally, we include an interaction term between temperature and temperature change in Column (10). In overall, the results are qualitatively similar to our main finding. Second, we replicate the results in Figure 2 but using alternative thresholds to define hot and cold days. In Panel A of Figure A1 (Appendix A), we present the results of the temperature bin approach using the 2-degree bin, while Panels B and C show the results using the 4-degree bin and 5-degree bin, respectively. We find that when our definition of hot and cold days is less (or more) demanding, the implied effects on income inequality remain consistent. Third, we present the results using alternative measures of poverty and income inequality at the subnational level. In Table A3 (Appendix A), we employ the multidimensional poverty indicators, which complement the traditional measure by capturing the acute deprivations in different aspects including monetary, education, electricity, sanitation, and drinking water. Similarly, we plot in Figure A2 (Appendix A) the effects of temperature using alternative measures of income including (i) the 90/10 ratio, (ii) the 80/20 ratio, and (iii) the Palma ratio. This helps address potential concern of using Gini and Theil indices as they are more sensitive to changes in the middle-income group. In overall, the results reaffirm the negative effects of higher temperature on poverty and income inequality. Fourth, we provide further tests in Table A4 (Appendix A) to ensure that our results are not sensitive to the choice of temperature measures. We do so by using (i) log of temperature (Column 1); (ii) temperature measured in degrees Fahrenheit (Column 2); (iii) the temperature data at 0.5◦ resolution from the Climate Research Unit of the University of East Anglia (CRU) (Column 3); (iv) the number of days that temperature is above 28oC (Column 4);16 (v) dropping regions with temperature being above that level (Column 5); and (vi) temperature shock, defined as the difference between actual temperature and long-term temperature being greater (less) than 2 (-2) standard deviations (Column 6). The results show little change from the baseline specification. 16 We choose the temperature at 28oC as this is the most common temperature in our sample (see Figure B3, Appendix B). 75 Fifth, we replicate our main analysis to different subsamples to investigate the sensitivity of our finding, as shown in Panels A and B of Table A5 (Appendix A). First, there are countries in our samples that contain only a small number of regions. We show in Column (1) that our results remain consistent when excluding these countries. The same finding is found when we exclude large countries that may drive our results such as United States, India, and Brazil (Columns 2 and 3). We also employ subsamples of countries without extremely cold weather (Column 4) and extremely hot weather (Column 5) using the 10 percent threshold. In Column (6), we use Conley standard errors that allow for spatial correlation in the error term. In overall, we find the estimated coefficients and significance levels are largely unchanged compared to our main finding. Sixth, we exploit poverty and inequality data from alternative sources to check the robustness of our results. We exploit the annual (subnational/grid level) GDP data coming from Kalkuhl and Wenz (2020) and Kummu et al. (2018) to construct poverty measures. An advantage of these datasets is that we are able to use a longer period-average (10-year) in the long differences model compared to our analysis using GSAP data. Using both panel and long differences models, Tables A6 and A7 (Appendix A) show that our findings are not sensitive to the alternative datasets, and the results are consistent across different specifications. We then conduct a similar exercise for the inequality analysis using country-level data from the World Development Indicators (WDI) and the Standardized World Income Inequality Database (SWIID). The results presented in Table A8 confirm our expectation. Finally, we conduct a placebo test of our study design. It is motivated by the fact that if estimating our chosen specification, but replacing the true value of the regressor of interest with an alternative we know should be irrelevant, we should expect to see no evidence of the effects on poverty. We do this exercise by using a within-sample randomization. First, the ‘true’ temperature of a region is replaced by temperature from another, randomly chosen in our sample without replacement. Second, the specification from Column (1) of Table 1 and Table 2 was estimated using the resulting placebo temperature series and the resulting coefficient and t-statistic on the temperature variable collected. This process is repeated with 1,000 randomizations and we present in Figure A3 (Appendix A) the coefficients and t-statistics harvested. Panel A shows that none of the placebo runs generate values anywhere close to those derived under true assignment, denoted by the dashed vertical lines. In Panel B, we find that only 5 percent of these estimates are larger in magnitude than the actual coefficient. It thus provides further support to our main estimates of the effects of temperature on poverty and inequality. C2. Heterogeneity analysis Consistent with the idea that warmer temperature leads to higher poverty rate and inequality, we also expect the impacts to be heterogenous across regions. We expect that countries bearing the largest effects of global warming tend to be poorer (i.e., low-income countries) or located in poor regions. This is explained by the fact that poor countries are less prepared for the effects of climate change. They are also more likely to suffer more damages, have proportionately higher material losses, and face greater obstacles during the phases of response, recovery, and reconstruction. To explore this, we split our sample into six regions and plot the coefficient estimates of temperature in Figure 3 (Panels A and B) using the temperature bin approach. As expected, hot temperatures are found to increase poverty and income inequality in most regions relative to temperature in the reference group, particularly poorer regions such as Sub-Saharan Africa, Middle East and North Africa, and South Asia. Furthermore, we also observe the negative effects of cold temperature among countries in East Asia and Pacific and Europe and Central Asia. Given that the range of temperature varies across countries, we split the temperature distribution within each country into deciles and choose the 60th percentile as the 76 baseline group. Figure A4 (Appendix A) shows that the effects of extreme weather are similar to what we observed in Figure 3. We also provide further support to the regional heterogeneity by estimating the effect of temperature on poverty and inequality by country, adjusted by their real GDP per capital in 2018. Figure A5 (Appendix A) shows that countries bearing the largest effect of global warming are also those with the lowest income such as Uganda, Ghana, and Mozambique. Next, we further assess the heterogeneity of the effects of temperature across different country characteristics. First, we examine whether a country’s institution may affect the impacts of temperature. This is motivated by the fact that institutions may affect adaptation to climate change through which incentives for individuals and collective action are structured. We use the democracy index from the 2020 report of the Economist Intelligence Unit and categorize countries into different types of regimes: (i) democracy; (ii) authoritarian; and (iii) hybrid. The results presented in Panel A of Table A9 (Appendix A) show evidence that countries with democracy regime appear to be less vulnerable to the impacts of global warming. We also examine the heterogenous impacts of temperature by other country characteristics. For example, countries near the equator have a higher poverty rate caused by an increase in temperature (Panel B, Table A9 in Appendix A). In addition, the effect of temperature is more pronounced in those with higher share of agriculture, while the opposite is found in countries with higher share of manufacturing (Panels C and D, Table A9 in Appendix A). Finally, we find a stronger effect among countries with lower share of trade, but our estimates are not statistically significant (Panel E, Table A9 in Appendix A). In this paper, we are also interested in examining the role of information and communication technologies (ICTs). It is reasonable to argue that ICTs, particularly the Internet, may contribute to poverty reduction by providing access to markets, decreasing transaction costs, and increasing income for a significant proportion of people living in developing countries. Therefore, we expect that regions with better internet coverage will be less vulnerable to the effects of higher temperatures. To do this exercise, we exploit the ICT Development index from the International Telecommunication Union as well as the global expansion of mobile network (2G, 3G, and 4G) from Collins Bartholomew with the latter being available at the grid level which allows us to construct a regional index. We then present coefficients on the interaction between our ICT measures and temperature in Table A10 (Appendix A). Across all panels, we find strong and consistent evidence of the role of ICT as the mediator. Specifically, areas with better access to ICT/internet broadband are less vulnerable to the effects of higher temperature. References Kalkuhl, M., & Wenz, L. (2020). The impact of climate conditions on economic production. Evidence from a global panel of regions. Journal of Environmental Economics and Management, 103, 102360. Kummu, M., Taka, M., & Guillaume, J. H. (2018). Gridded global datasets for gross domestic product and Human Development Index over 1990–2015. Scientific Data, 5(1), 1–15. 77 Appendix D: Mechanism analysis and projected impacts under future climate change D1. Potential mechanisms Having demonstrated strong evidence of the effects of warming temperature on poverty at the subnational level, we further explore why impact heterogeneity exists across regions. A possible explanation is that poor countries are often located in tropical areas, where climate change occurs faster and is more intense, and their livelihoods are more dependent on the climate vulnerable agriculture sector. In fact, a growing body of evidence suggests that extreme temperature has negative effects on crop yields, particularly in poor countries (e.g., Jacoby et al., 2015; Knox et al., 2012; Schlenker and Lobell, 2010). We analyze the global dataset of historical yields from Iizumi and Sakai (2020), which provides actual crop yields for years from 1981 to 2016 at 0.5° resolution. Using the panel fixed effects model and long differences model as in Equations (1) and (2), we find consistent and negative effects of higher temperature on different crop yields including rice, maze, and soybean, as shown in Table A11 (Appendix A). Again, we find the long differences model estimates to be smaller than the panel model estimates, which are in line with previous studies showing potential adaptation in the long run (e.g., Chen and Gong, 2021). Similarly, we also find the effects of global warming to be more pronounced among regions with a higher share of agriculture (Appendix A, Table A12). Given the adverse impacts of temperature on agricultural production, we further examine whether there exists any correlation between poverty and agriculture. Specifically, we plot the effects of temperature on poverty taken from the panel model specification on the y-axis, and the effects of temperature on agriculture in Table A11 on the x-axis in Figure A6 (Appendix A). Since the unit of analysis is different across two samples, we aggregate the data at the country level for better comparison. For all the panels, we find a negative and strongly statistically significant correlation between crop yield and poverty. Consistent with our previous findings, African countries are found to be most vulnerable to the effects of global warming. In Figure A7 (Appendix A), we further plot the effects of temperature on poverty against a country’s share of agriculture in GDP and also find the effects to be stronger among countries which rely on agriculture as the main source of income. Overall, these findings suggest that by reducing crop yield, warmer temperature may directly contribute to more poverty.17 Another potential mechanism that may explain the effects of temperature is migration. Since poverty is a major driver of people’s vulnerability to climate-related shocks, it is reasonable to expect that the flow of people escaping poverty is also affected by climate change. In fact, an emerging body of literature has shown that higher temperature increases both internal and international immigration rates (e.g., Cattaneo and Peri, 2016; Missirian and Schlenker, 2017). We reaffirm findings from the literature by using migration data available at the subnational level. The data is provided by WorldPop Open Data Repository which captures internal migration flows between 2005 and 2010.18 Using a simple OLS regression, we find suggestive evidence that hotter temperature results in higher migration flow (see Figure A8, Appendix A). D2. Projected impacts under future climate change 17 For simplicity, we assume land degradation to be constant, but it could play a role in the poverty and environment nexus (Barbier and Hochard, 2018). Temperature may also affect poverty via different channels such as civil conflicts and labor productivity (for a review, see Burke et al., 2015 and Somanathan et al., 2021). 18 The data is available at https://hub.worldpop.org/. 78 We next provide projections of the effects of future temperature on poverty to better understand potential effects under different scenarios. To do this, we combine the model estimates in Tables 1 and 2 with data on simulated weather conditions at the subnational level from 2030 to 2099. We focus on RCP4.5 and RCP8.5 scenarios, which are two extreme emission pathways that represent opposite ends of the climate spectrum depending on the uptake of renewable energy.19 Following Burke and Emerick (2016) and Kalkuhl and Wenz (2020), we generate temperature projections as follows. First, we use annual temperature from ERA-5 to construct historical average temperature and probability distribution functions for the period 1979 – 2019. We then calculate the projected changes in temperature as the difference between the projected temperature, taken from NEX, and the historical average temperature. Finally, the temperature changes are used to calculate poverty rates (inequality) by multiplying with the baseline estimates in Columns (2), (4), and (6) of Table 1 (Columns (2) and (4) of Table 2). We select the estimates from the long differences model since it embodies any adaptations that farmers have undertaken to short-run change in climate, and thus projections of future climate change impacts would appear more trustworthy than those based on either panel or cross- sectional methods (Burke and Emerick, 2016). Table A13 (Appendix A) provides a summary of the projected changes for temperature and poverty for the RCP4.5 and RCP8.5 emission pathways in the short, medium, and long terms. Under the RCP4.5 and RCP8.5 pathways, temperature will increase by 2.6◦C and 6.0◦C in 2099. These temperature increases can result in poverty increases between 1.4 and 3.1 percentage points (which correspond to 13.6 and 31.1 percent changes). Similarly, the simulated effects on inequality are estimated to be between 0.4 and 2.1 percentage point increases in Gini index (which correspond to 1.2 and 5.9 percent increases) (Table A14, Appendix A). The largest increase in poverty and inequality would occur in the scenario without any countervailing strategies based on renewable energy to address climate change between 2021 and 2099. Finally, Figures A8 and A9 (Appendix A) present the projected temperature effects across regions in our sample under both emission pathways and reaffirm our previous findings that poor countries in Africa continue to be most vulnerable to hotter temperature. References Barbier, E. B., & Hochard, J. P. (2018b). Land degradation and poverty. Nature Sustainability, 1(11), 623-631. Burke, M., Hsiang, S. M., & Miguel, E. (2015). Climate and conflict. Annual Review of Economics, 7(1), 577–617. Burke, M., & Emerick, K. (2016). Adaptation to climate change: Evidence from US agriculture. American Economic Journal: Economic Policy, 8(3), 106–40. Cattaneo, C., & Peri, G. (2016). The migration response to increasing temperatures. Journal of Development Economics, 122, 127-146. Chen, S., & Gong, B. (2021). Response and adaptation of agriculture to climate change: Evidence from China. Journal of Development Economics, 148, 102557. Iizumi, T., & Sakai, T. (2020). The global dataset of historical yields for major crops 1981– 2016. Scientific Data, 7(1), 1–7. Intergovernmental Panel on Climate Change (IPCC). (2021a). “Summary for Policymakers”. In Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. In 19 RCP is the Representative Concentration Pathway, which captures future trends in climate change under alternative scenarios of human activities. RCP8.5 tracks emissions consistent with current trends (business as usual scenario in which greenhouse gas emissions go unchecked), while RCP4.5 considers a scenario with increased reliance on renewable energy and less reliance on coal-fired power (IPCC, 2021). 79 Masson-Delmotte et al. (eds.) Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 3−32. Jacoby, H. G., Rabassa, M., & Skoufias, E. (2015). Distributional implications of climate change in rural India: a general equilibrium approach. American Journal of Agricultural Economics, 97(4), 1135–1156. Kalkuhl, M., & Wenz, L. (2020). The impact of climate conditions on economic production. Evidence from a global panel of regions. Journal of Environmental Economics and Management, 103, 102360. Knox, J., Hess, T., Daccache, A., & Wheeler, T. (2012). Climate change impacts on crop productivity in Africa and South Asia. Environmental Research Letters, 7(3), 034032. Missirian, A., & Schlenker, W. (2017). Asylum applications respond to temperature fluctuations. Science, 358(6370), 1610-1614. Schlenker, W., & Lobell, D. B. (2010). Robust negative impacts of climate change on African agriculture. Environmental Research Letters, 5(1), 014010. Somanathan, E., Somanathan, R., Sudarshan, A., & Tewari, M. (2021). The impact of temperature on productivity and labor supply: Evidence from Indian manufacturing. Journal of Political Economy, 129(6), 1797–1827. 80