Policy Research Working Paper 10643 Impacts and Sources of Air Pollution in Tbilisi, Georgia Sandra Baquié Patrick A. Behrer Xinming Du Alan Fuchs Natsuko K. Nozaki Poverty and Equity Global Practice & Development Research Group December 2023 Policy Research Working Paper 10643 Abstract Air pollution profoundly impacts welfare, causing more levels corresponds to a 0.24 percent increase in respira- deaths globally than malnutrition, AIDS, tuberculosis, tory hospitalization rates. A 1 percent increase in PM2.5 is and malaria combined. In the Georgian capital, Tbilisi, air also associated with a 0.2 percent decrease in rental prices. pollution levels exceed international standards and surpass All the estimates are lower bounds of the total impact of levels in other cities in the region. The average monthly air pollution as they only account for short-term conse- PM2.5 concentration in Tbilisi is 20 µg/m3, four times quences. The study shows that traffic and industrial activity higher than the World Health Organization’s annual rec- are significant drivers of air pollution in Tbilisi. The paper ommended limit. This paper uses multiple data sources also estimates the positive co-benefits of potential carbon —administrative data, satellite imagery, private real estate pricing policies from air pollution reduction. Adopting a transactions, and traffic data—to estimate the impact of carbon tax of $25 per ton would reduce hospitalizations by air pollution on the health and productivity of people 0.44 percent per district by 2036, while increasing rental in Tbilisi. It estimates that a 1 percent increase in PM2.5 prices by 0.38 percent. This paper is a product of the Poverty and Equity Global Practice the Development Research Group, Development Economics. 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 sbaquie@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 Impacts and Sources of Air Pollution in Tbilisi, Georgia∗ Sandra Baquié1 Patrick A. Behrer1 Xinming Du2 Alan Fuchs1 Natsuko K. Nozaki1 1 World Bank 2 Columbia University JEL Classifications: Q5, R4, I1 Keywords: Air pollution; Health; Real estate; Transportation; Carbon pricing; Climate mitigation; Tbilisi ∗ Corresponding authors: Sandra Baquié (sbaquie@worldbank.org) and Xinming Du (xd2197@columbia.edu). We are grateful to Norberto Pignatti (ISET), Karine Torosyan (ISET), Mariam Tsulukidze (ISET), Elene Ergeshidze (CRRC), Kristine Vacharadze (CRRC) and Dustin Gilbreath (CRRC) for their excellent research assistance and support. The team is grateful for comments and guidance from Sebastian-A Molineus (Regional Director, World Bank), Miguel Sanchez (Program Leader, World Bank), Elena Golub (Senior Environmental Economist, World Bank), Erik Illes (Head of Development Cooperation/Deputy Head of Mission, SIDA), Tina Genebashvili (Program Officer, SIDA), Paata Shavishvili and Vasil Tsakadze (Geostat), Noe Megrelishvili and Lasha Akhalaia (Ministry of Agriculture and Environmental Protection), Khatia Chkhetiani (Tbilisi City Hall), and Lela Sturua (National Center for Disease Control), Martin Heger (Senior Env. Economist, World Bank), Trang Van Nguyen (Senior Economist, World Bank), Paulina Estela Schulz Antipa (Economist, World Bank) and Alexandra Andrea Maite Campmas (Economist,World Bank). Financial support was generously provided by Sweden and the World Bank’s South Caucasus Poverty, Equity, and Gender program. The findings, interpretations, and conclusions expressed in this report 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. 1 Introduction Existing literature highlights the substantial and wide-ranging impacts of air pollution on various aspects of welfare. Globally, air pollution is responsible for a staggering 6.7 million - 10 million deaths annually, surpassing the death toll of malnutrition, AIDS, tuberculosis, and malaria combined (Fuller et al., 2022; Weichenthal et al., 2022). Research has documented significant adverse effects on mortality in countries such as India (Greenstone and Hanna, 2014), China (Chen et al., 2013), Indonesia (Jayachandran, 2009), and the United States (Deryugina et al., 2019). In Georgia, a study finds that air pollution is the 6th highest risk factor causing more deaths and disability combined (IHME, 2023). The detrimental consequences extend beyond mortality, affecting welfare through decreased productivity (Chang et al., 2016), impaired cognitive abilities (Lavy et al., 2014), deteriorated mental health (Chen et al., 2018), and reduced housing values (Davis, 2011; Currie et al., 2015). However, while the impacts of air pollution have been extensively studied in various countries, to our knowledge, the estimates presented in this paper represent the first evaluation of the adverse effects on health, real estate values, cognitive performance, and productivity in the South Caucasus. Air pollution levels in Tbilisi, the capital of Georgia, exceed international standards and surpass those of other capitals in the region. The monthly average concentration of PM2.5 is higher than 20 µg/m3 , which is four times the annual average recommended by the World Health Organization (WHO), as is shown in Figure 1. From 2017 to 2021, Tbilisi experienced higher air pollution levels than Istanbul (Türkiye), Baku (Azerbaijan), and Kyiv (Ukraine). The residents of Tbilisi are well aware of this pressing issue and prioritize it highly for their city. A 2021 Caucuses Research Resource Center (CRRC) survey revealed that 11% of interviewees in Tbilisi considered air pollution one of the capital’s most important issues (CRRC, 2021). Within Tbilisi, air pollution is most severe in the city’s central area, where elevation is the lowest and the most affluent individuals tend to reside (Figure 2). This pattern is a result of a conjunction of factors: the elevation gradient, weather patterns, and the topography of urban development. Because the center of the city lies at the bottom of a bowl-shaped area, with the center at a lower altitude compared to the suburbs, the surrounding hills tend to trap pollution in the center. Despite high pollution levels, existing papers about air pollution in Georgia are limited. A few notable reports provide valuable insights into the topic. Work by CRRC estimates that 42% of individuals residing in the capital considered pollution the most critical infrastructural concern, while only 26% of those in other urban settlements in Georgia shared the same view 2 Figure 1: Monthly air pollution level in Tbilisi Figure 2: PM2.5 concentration is negatively correlated with elevation (CRRC, 2016). Reflecting this concern Georgian governments have been regulating air pollution through various measures to mitigate its harmful effects. The Law of Georgia on Ambient Air Protection was implemented in 1999 as a comprehensive framework to address air pollution, and it has undergone several amendments since then. As part of their efforts, the Government has taken steps to introduce more environmentally friendly buses into Tbilisi’s public transport system and implemented an excise tax on old vehicles to incentivize the import of newer, less- polluting vehicles. In 2018, the Government of Georgia also reinstated mandatory periodic technical inspections, which had been partially suspended in 2004. As of June 28, 2023, the Government has also adopted Decree 238, introducing the European Union standards for permissible automobile emissions to Georgia. This reform entails the adoption of EURO- 5 technical requirements for imported automobiles, aligning Georgia with the emissions and 3 pollution standards set by the European Union. The authorities also submitted a new Law on Industrial Emissions to Parliament in January 2023 to prevent spillage from industrial activities into the environment and waste generation. Despite a relative lack of research, the self-reported importance of air pollution for Tbilisi inhabitants and its prioritization by the Government highlights the importance of further studies on Tbilisi’s air pollution, particularly its impacts and drivers. Our analysis leverages a comprehensive dataset that combines ground-based air quality monitor data with satellite-derived PM2.5 measurements, courtesy of van Donkelaar et al. (2021). Using thermal inversions and wind directions as instrumental variables to address potential endogeneity, we find statistically significant adverse effects of air pollution. Specifically, a 1% increase in PM2.5 levels corresponds to a 0.235% increase in respiratory hospitalization rates. When studying rental prices and test scores, the instrument is not as strong and we revert to an OLS specification. We find that a 1% increase in PM2.5 levels is associated with a 0.2% decrease in rental prices. Although estimates are imprecise, results also suggest a decrease in test scores. Note that these estimates are lower bounds of the total effect of air pollution. Indeed, the methodology and data do not allow us to assess the long-term effect of air pollution on the considered outcomes. Moreover, the OLS estimates likely underestimate the causal impact of air pollution. Indeed, omitted variables that could increase air pollution are likely related to increases in economic activity and, in turn, higher real estate prices. Overall, these findings underscore the substantial and detrimental impacts of air pollution across various dimensions of public health, housing markets, and, potentially, educational outcomes. Our study also seeks to quantify the contributions of industrial emissions and traffic congestion to ambient air pollution levels. By analyzing geocoded emission data and high-frequency traffic flow information, we show that industrial sources and traffic congestion significantly reduce air quality in Tbilisi, particularly in districts downwind of the primary sources. Our analysis of traffic congestion and the associated emissions reveals that congestion’s impact on pollution is relatively linear. Even in areas with existing congestion, the introduction of additional vehicles consistently results in pollution increases. These results suggest that policies reducing the number of days with high traffic, such as congestion pricing, could effectively improve air quality in Tbilisi. Finally, we present an example of how these estimates can be further used to inform policymaking. We employ the Climate Policy Assessment Tool (CPAT) to conduct simulations of the impacts of carbon taxes on emissions and pollution levels in different carbon pricing scenarios (World Bank, 2023, 2022). We then use our estimates to assess the potential co-benefits of these climate 4 mitigation strategies. Our analysis suggests that a carbon tax set at $25 per ton could reduce respiratory hospitalizations by 0.44% and increase rental prices by 0.38%. Implementing an emission trading scheme with a substantially higher implied carbon price would yield even more substantial benefits, with projected health effects of 1.38% and an impact on housing prices of 1.18%. These findings underscore the potential for significant positive co-benefits of carbon pricing policies. This paper contributes to the existing body of evidence on the externalities of air pollution by presenting empirical findings specific to Georgia, a middle-income country. While previous research has predominantly focused on high-income countries or densely populated developing nations, our study sheds light on the nuanced challenges faced by countries in the middle-income category. By doing so, we not only deepen our understanding of environmental issues but also provide insights that can inform policy decisions and interventions in similar contexts worldwide. The subsequent sections of this paper are organized as follows. Section 2 outlines our data sources for air pollution and downstream outcomes. Section 3 details the method, and sections 4 and 5 present results on air pollution impacts and sources, respectively. Section 6 delves into the simulated policy impacts of carbon taxes and emission trading schemes, assessing their associated health, hedonic, and productivity benefits. Finally, Section 7 offers concluding remarks. 2 Data 2.1 Air pollution data Our primary measure of air pollution is the estimated concentration of surface PM2.5 produced by Van Donkelaar et al. (2021). The data product is at the 0.01◦ grid-month resolution and derived from satellite observations and ground-based measurements for calibration. This product provides continuous monthly measurements of PM2.5 from 1998 to 2020 with global coverage, allowing us to examine PM2.5 in areas without air quality monitors. In our analysis, we aggregate the PM2.5 measurements to the census block-month level and merge pollution with outcomes and pollution sources at the census block level. 2.2 Weather data Weather influences air pollution levels. Temperature impacts atmospheric stability and pollutant dispersion. Wind speed and direction are crucial factors in understanding the movement and 5 transport of air pollutants, shedding light on potential local and regional sources. Precipitation data enables the examination of wet deposition, which can influence pollution concentrations by removing particulate matter from the atmosphere. We use weather data from the ERA5, a state-of-the-art reanalysis product developed by the European Centre for Medium-Range Weather Forecasts (ECMWF). ERA5 offers comprehensive and reliable meteorological information with 9km spatial and hourly temporal resolution. We extract temperature, wind speed and direction, and precipitation variables from the ERA5 dataset at the census block-day level. Thermal inversions can also significantly impact air quality by trapping pollutants close to the surface, hindering their dispersion and leading to elevated PM2.5 levels. We use thermal inversion data from the MERRA-2 (Modern-Era Retrospective Analysis for Research and Applications, Version 2) dataset. MERRA-2 is a dataset developed by the National Aeronautics and Space Administration (NASA) that captures a wide range of atmospheric variables, including thermal inversion metrics. We focus on two key aspects of thermal inversions: inversion occurrence and inversion strength. These metrics are derived from temperature profiles and serve as indicators of the stability of the lower atmosphere. We measure inversion occurrence as instances in which the temperature in an atmospheric layer above the surface layer (but below roughly 5 km from the surface) is higher than the temperature in the surface layer. Inversion strength quantifies the magnitude of the difference in temperature when an inversion occurs. Similar to other weather variables, we aggregate thermal inversion data at the census block-day level. The relevance of our instrument is supported by the fact that the technical descriptions of the model in Van Donkelaar et al. (2021) indicate that the inversion data we use is not an input in their calibration model. There also does not appear to be an overlap between our meteorological variables and those used by Van Donkelaar et al. (2021) to calibrate their air pollution data. 2.3 Hospitalization data We use hospitalization data obtained from the National Center for Disease Control (NCDC) for Tbilisi. The dataset covers the period from January 2015 to August 2022, with data from all 134 hospitals in Tbilisi, and provides information at the hospital-disease category-day level. Our analysis focuses on four categories: mental health, cardiovascular diseases, stroke, and respiratory diseases. These health conditions have been extensively studied in the context of air pollution and have been found to be influenced by elevated levels of PM2.5 (e.g., Chen et al., 2018; Filippini et al., 2019). We geocoded hospital addresses to assign census block information 6 to each hospital. Then, we merged the hospitalization data with the corresponding air pollution data at the census block-month level. 2.4 Housing price data We use real estate data from LIVO, Georgia’s primary real estate agency. The dataset includes data scraped from LIVO’s platform on a monthly basis from January 2019 to June 2021. The dataset contains 30,000 to 60,000 observations per month, suggesting a robust representation of real estate transactions during the specified period. The LIVO dataset includes rental and sale values for flats and private houses, as well as real estate coordinates and characteristics at the transaction level. Using the real estate coordinates, we merged the real estate data with the corresponding air pollution level in the month of the observed transaction. Real estate characteristics include various attributes, such as the number of bedrooms, square footage, amenities, distances to schools, hospitals, grocery stores, etc. 2.5 Education data We obtain test score data for the Unified Entrance Exams (UEE). These exams are widely recognized as crucial benchmarks for evaluating students’ academic performance and preparedness for higher education. Our dataset comprises individual-level test scores and includes information on the exam year, subject, school, and grade. The data covers exam years from 2011 to 2017 and encompasses 29 subjects across 338 schools. To examine the relationship between air pollution and academic performance, we merge the school locations and exam months with the air pollution data at the census block-month level. 2.6 Traffic data We utilize traffic data obtained from Mapbox. The data is derived from cell phone-based applications and user movements and covers a period from 2019 to 2021. This dataset has information on speed and waiting time for each road segment in Tbilisi at 5 minute intervals. To assess traffic congestion, we define instances where the speed falls below 5 km per hour. These traffic jams indicate significant congestion. We then geocode the corresponding road segments and aggregate the traffic jam data to the census-block-day level. Then, we link the traffic jam intensity with air pollution to explore the relationship between traffic sources and air quality. 7 2.7 Industrial emission data Industrial emission data is obtained from the Environment Information and Education Centre of Georgia’s Ministry of Environmental Protection and Agriculture of Georgia. The dataset comprehensively represents 141 distinct industrial facilities, encompassing a wide range of 59 sectors. Detailed annual emissions data for total hard particles, dust, and greenhouse gases have been recorded from 2016 to 2021. Additionally, the dataset includes these facilities’ coordinates and geo-coded information, allowing for their identification within specific census blocks. Table 1: Data time periods Data set Time period Air pollution 1998-2022 Meteorological data 1998-2022 Hospitalization data 2015-2022 Housing price data 2019-2021 Education data 2011-2017 Traffic data 2019-2021 Industrial data 2016-2021 3 Methods To study the relationship between air pollution and its outcomes, we first use the following specification: Yit = βP M2.5,it + γt + ηi + εit (1) where Yit represents the outcome at location i in month t. On the right-hand side of the model, P M2.5,it denotes the air pollution level in i at t. We include year and month-fixed effects denoted by γt to control for national annual differences and seasonal patterns. We also incorporate location fixed effects represented by ηi to capture time-invariant characteristics specific to each location that may affect outcomes. The coefficient of interest is β , which captures the correlation between PM2.5 levels and the considered outcome. However, the air pollution assignment may not be random. This is the case for health outcomes, for instance. Indeed, if rich people sort into clean areas and naturally have lower 8 hospitalization levels, our OLS estimate will be higher than the actual effect of pollution. If other unobserved events increase both pollution and health burden, we will overestimate the actual effect. To address these endogeneity concerns, we use thermal inversions as an instrument for air pollution. Thermal inversions occur when the air near the surface is colder than the air at higher altitudes due to meteorological conditions. In this situation, the cold, polluted air at the surface is trapped, resulting in higher air pollution levels. They are plausibly exogenous because random weather fluctuations and local seasonality influence their occurrence and strength. Table 2 presents the first stage, showing the significant impact of thermal inversions on air pollution levels. The presence of thermal inversions leads to a substantial increase in PM2.5 levels by 16µg/m3 , which represents a 69.7% increase relative to the average pollution level. Table 2: Thermal inversion and air pollution PM2.5 Inversion 1.504∗∗ 16.254∗∗∗ 16.695∗∗∗ (0.685) (1.074) (1.105) Inversion Strength 1.751∗∗∗ -7.228∗∗∗ -7.653∗∗∗ (0.290) (0.443) (0.466) Observations 28560 28560 28560 R-square 0.010 0.397 0.411 Y-mean 23.959 23.959 23.959 Y-sd 3.994 3.994 3.994 Year, month FEs Y Y District FEs Y Notes: This table shows first stage estimation results. Controls include year, month, and district fixed effects. Robust standard errors are clustered at the district level. 4 Air pollution impacts 4.1 Effects on health We study the association between PM2.5 exposure and hospitalizations for different disease types. To assess the effect on health, we apply the above methodology to our measures of health outcomes. Here, Yit is the total number of hospitalizations for each disease type in month t at hospital i. Given that the pollution measure is at the hospital level rather than at the residential address level, we assume that residents residing in the same census block as the hospital would seek medical care at that hospital. 9 Table 3 presents the OLS estimation results. The positive coefficients on β indicate that higher PM2.5 levels are associated with increased hospitalizations. The magnitudes of the coefficients differ across disease types, aligning with existing medical evidence that highlights the severe impacts of air pollution on the respiratory and cardiovascular systems (Brunekreef and Holgate, 2002; Manisalidis et al., 2020). Specifically, we show that a 1% increase in PM2.5 is associated with a 0.155% rise in respiratory disease hospitalizations and a 0.098% increase in cardiovascular disease hospitalizations. The correlation is also positive and significant when stroke is used as a dependent variable, while it is no longer precise if we study mental health. Table 3: Air pollution and hospitalization: OLS ln(Hospitalizations) Mental Cardiovascular Stroke Respiratory ln(PM2.5) 0.008 0.098∗∗ 0.067∗∗ 0.155∗∗∗ (0.007) (0.039) (0.028) (0.044) Observations 7140 7140 7140 7140 Y-mean 0.017 0.817 0.507 1.375 Y-sd 0.125 1.509 1.005 1.680 R-square 0.390 0.789 0.717 0.705 Year, month FEs Y Y Y Y Hospital FEs Y Y Y Y Notes: This table shows ols regression estimation results. Controls include year, month, and hospital fixed effects. Robust standard errors are clustered at the district level. Table 4 presents the instrumental variable estimation results. This specification tackles the potential endogenous concerns associated with the OLS specification. Column (4) shows that a 1% increase in PM2.5 , instrumented by the occurrence and strength of thermal inversions, results in a 0.235% increase in respiratory disease hospitalizations. Estimates in columns (2) and (3) estimates remain positive but are no longer statistically significant. The elasticity of the air pollution-disease relationship is 0.056 for cardiovascular diseases and 0.040 for stroke. Column (1) shows that PM2.5 , instrumented by thermal inversions, leads to a 0.017% increase in mental health issues. The magnitude is smaller than those in the other columns, suggesting that the relationship between air pollution and mental health outcomes may be ambiguous. It is also possible that the effects of air pollution on mental health may not manifest immediately and may require prolonged exposure over an extended period. 10 Table 4: Air pollution and hospitalization: IV ln(Hospitalizations) Mental Cardiovascular Stroke Respiratory ln(PM2.5) 0.017∗ 0.056 0.040 0.235∗∗∗ (0.008) (0.063) (0.050) (0.079) Observations 7140 7140 7140 7140 Y-mean 0.017 0.817 0.507 1.375 Y-sd 0.125 1.509 1.005 1.680 R-square 0.391 0.789 0.717 0.705 Year, month FEs Y Y Y Y Hospital FEs Y Y Y Y Notes: This table shows IV regression estimation results. We use the occurrence and strength of thermal inversions as instruments for air pollution. Controls include year, month, and hospital fixed effects. Robust standard errors are clustered at the district level. Existing literature has demonstrated a nonlinear relationship between air pollution and its effects on human health (Arceo et al., 2016). Hence, it is crucial to examine the heterogeneous impacts of pollution on hospitalization rates across different pollution levels. To do so, we divide districts into four quartiles based on their average pollution levels and run separate estimations of equation 1 for each quartile. Table 5: Heterogeneity across pollution subgroups ln(Hospitalizations) Panel A: pollution quartile 1 Low: 9.9 - 21.5 µg/m3 Mental Cardiovascular Stroke Respiratory ln(PM2.5) 0.028 0.022 0.046 0.010 (0.023) (0.115) (0.087) (0.149) Observations 1380 1380 1380 1380 Y-mean 0.026 0.931 0.425 1.327 Y-sd 0.142 1.580 0.972 1.706 R-square 0.260 0.667 0.643 0.738 Panel B: pollution quartile 2 Medium 1: 21.5 - 24.2 µg/m3 ln(PM2.5) 0.001 0.046 0.099 0.149 (0.010) (0.114) (0.065) (0.097) Observations 2400 2400 2400 2400 Y-mean 0.005 0.924 0.674 1.283 Y-sd 0.065 1.550 1.091 1.522 R-square 0.094 0.839 0.818 0.776 Panel C: pollution quartile 3 Medium 2: 24.2 - 27.5 µg/m3 ln(PM2.5) 0.025 0.135 0.089 0.664∗∗∗ 11 (0.019) (0.109) (0.100) (0.198) Observations 1440 1440 1440 1440 Y-mean 0.008 0.575 0.214 1.535 Y-sd 0.084 1.313 0.517 1.794 R-square 0.098 0.856 0.563 0.641 Panel D: pollution quartile 4 High: 27.5 - 44.3 µg/m3 ln(PM2.5) 0.013 0.205∗ -0.000 0.149 (0.012) (0.102) (0.084) (0.140) Observations 1920 1920 1920 1920 Y-mean 0.030 0.782 0.579 1.404 Y-sd 0.180 1.521 1.133 1.752 R-square 0.547 0.784 0.660 0.672 Year, month FEs Y Y Y Y Hospital FEs Y Y Y Y Notes: This table shows IV regression estimation results. We use the occurrence and strength of thermal inversions as instruments for air pollution. Controls include year, month, hospital fixed effects. Robust standard errors are clustered at the district level. In Table 5’s Column (1), we find that air pollution increases mental illness in Panels A and C. It suggests that both higher-pollution and relatively cleaner groups could experience similar marginal damage to mental health as a result of ambient PM2.5 exposure. This is consistent with existing work in the U.S. (Bishop et al., 2023; Persico and Marcotte, 2022). However, the lack of significance for Panel B and D also highlights the limitation of our data and the absence of a clear pattern between pollution levels and the marginal impact on mental health. In Column (2) of Table 5, we find a significant impact of air pollution on cardiovascular disease in Panel D, the most polluted quartile. For every 1% increase in PM2.5 , there is a 0.205% increase in hospitalizations related to cardiovascular disease. The effect size gradually increases as we move from Panel A to D. Our estimates are imprecise but provide some suggestive evidence that the marginal effects of air pollution on cardiovascular health are more pronounced in areas with higher pollution levels. In the cleanest group (Panel A), the elasticity between pollution and cardiovascular hospitalization is 0.022. This implies that even in areas with lower pollution levels, air pollution still has a discernible impact on cardiovascular disease, albeit to a lesser extent. We also test the heterogeneity across seasons, given the strong seasonality of emissions and hospitalization. In Table 6, we find positive and significant estimates on both P M × Summer and P M × W inter, while estimates on the single term P M are almost absorbed. This suggests that the effects of air pollution on hospitalizations are especially stronger in summer and winter compared to spring and fall. The observed heterogeneity may be attributed to increased energy 12 consumption linked to heating and air conditioning or even the prevalence of other diseases, thereby amplifying the severity of health-related consequences (Graff Zivin et al., 2023). Table 6: Heterogeneity across seasons ln(Hospitalizations) Mental Cardiovascular Stroke Respiratory ln(PM2.5) 0.005 -0.036 0.028 -0.027 (0.010) (0.051) (0.060) (0.092) ln(PM2.5) × Summer 0.069∗ 0.207∗∗ 0.098 0.265∗∗ (0.036) (0.083) (0.062) (0.126) ln(PM2.5) × Winter 0.003 0.161 -0.006 0.580∗∗∗ (0.022) (0.105) (0.069) (0.200) Observations 7140 7140 7140 7140 Y-mean 0.017 0.817 0.507 1.375 Y-sd 0.125 1.509 1.005 1.680 R-square 0.391 0.789 0.717 0.705 Year, month FEs Y Y Y Y Hospital FEs Y Y Y Y Notes: This table shows IV regression estimation results. We use the occurrence and strength of thermal inversions as instruments for air pollution. Controls include year, month, and hospital fixed effects. Robust standard errors are clustered at the district level. In addition, we test the effect of lagged pollution last month on hospitalization this month. In Table 7, estimates on P M _lag are positive and significant and the magnitudes are larger than those in Table 4. This pattern implies that the accumulated pollution from previous periods has more severe effects on health outcomes. Table 7: Effects of pollution last month ln(Hospitalizations) Mental Cardiovascular Stroke Respiratory ln(PM2.5) last month 0.001 0.073∗ 0.102∗∗ 0.267∗∗∗ (0.006) (0.041) (0.041) (0.056) Observations 7140 7140 7140 7140 Y-mean 0.017 0.817 0.507 1.375 Y-sd 0.125 1.509 1.005 1.680 R-square 0.390 0.789 0.718 0.706 Year, month FEs Y Y Y Y Hospital FEs Y Y Y Y Notes: This table shows IV regression estimation results. We use the occurrence and strength of thermal inversions as instruments for air pollution. Controls include year, month, and hospital fixed effects. Robust standard errors are clustered at the district level. 13 4.2 Effects on real estate prices Table 8 presents the results of the OLS regression of PM2.5 levels on real estate sales and rental prices. The thermal inversion instrument is relatively weak; we only present the OLS estimates. Note that they are likely to underestimate the magnitude of the adverse impact of air pollution. Indeed, omitted variables that increase air pollution are likely related to increases in economic activity and, in turn, higher real estate prices. The negative and significant estimates indicate that PM2.5 is associated with lower flat rent prices. In contrast, the relationship between PM2.5 and flat sale prices is not statistically significant, though the sign is still negative. We find a 1% increase in PM2.5 is associated with a 0.13% decrease in flat sale price and a 0.2% decrease in flat rent price. It suggests that sale prices may be less sensitive to pollution levels at the time of advertising than rent prices. Potential home buyers likely take into account an extensive range of factors when making long-term investment decisions. Pollution at the time of advertising is likely to play a less significant role compared to other determinants of housing prices, such as neighborhood amenities, infrastructure, or expectations about long-term environmental conditions, which may not be correlated with pollution levels at the time of purchase. Table 8: Air pollution and real estate price: OLS ln(Flat sale) ln(Flat rent) ln(PM2.5) -0.007 -0.132 -0.029 -0.200∗∗ (0.021) (0.272) (0.022) (0.101) Observations 69412 69412 54750 54750 R-square 0.266 0.965 0.254 0.983 Y-mean 6.747 6.747 1.853 1.853 Y-sd 0.551 0.551 0.498 0.498 Year, month FEs Y Y Y Y District FEs Y Y Y Y Real estate FEs Y Y Notes: This table shows OLS regression estimation results. Controls include year, month, and district fixed effects in all columns. We further control for real estate fixed effects in Columns (2) and (4). Robust standard errors are clustered at the district level. 4.3 Effects on test scores We estimate the impact of air pollution in the month preceding the UEE exam on students’ scores. Unfortunately, the above weather instruments are not strong enough in the test score 14 specification to be relevant. Results of the OLS specification are presented in the Appendix. Figure A1 indicates that air pollution may have an impact on student academic performances. Overall, our estimates suggest that a 1% increase in PM2.5 during the exam month leads to a decrease in test scores, although they are statistically imprecise. The decomposition by subject indicates that the impact could be worse for some subjects, such as Geography, Mathematics, and Physics. Again, the imprecision of the results prevents us from drawing strong conclusions on this aspect. 15 5 Impact of industrial emissions and traffic on air pollution 5.1 Industrial emissions Industrial emissions have a significant impact on air pollution levels. We use GHG emissions as our measure of industrial emissions. It is important to note that, although GHG emissions and local air pollutants are often two sides of the same coin, some chemicals belong exclusively to one group. Table 9 provides the correlation between industrial emissions at the source level and ambient PM2.5 in district of the emitting source. As the industrial greenhouse gas emissions increase by 1,000 tons per year, the air pollution at the census block-year level rises by 0.011 µg/m3 , which corresponds to 0.45% of the mean pollution level. This result shows the potential of climate mitigation policies to reduce air pollution while curbing industrial emissions. We estimate the positive externalities of carbon pricing policies, aiming at reducing overall emissions, in the following section. Table 9: Industrial emissions and air pollution PM2.5 at the district-year level Industry’s GHG emissions 0.023 0.014∗ 0.011∗∗ (0.041) (0.006) (0.004) Observations 2832 2832 2832 R-square 0.000 0.802 0.907 Y-mean 24.467 24.467 24.467 Y-sd 1.717 1.717 1.717 Year FEs Y Y Firm FEs Y Notes: Emissions are divided by 1000. Standard errors are clustered at the year level. Industrial emissions not only impact their district, but they also have a severe impact on air quality in districts located downwind. In Tbilisi, the four primary industrial sources of greenhouse gases are related to metallurgical production, construction material manufacturing, and food production. Figure 3 illustrates the influence of wind in propagating industrial emissions to downwind districts. For the purpose of this analysis, we define downwind as the alignment between the wind at the emission source and the direction of the district. Within 1 kilometer of the emission source, the concentration of PM2.5 increases by an average of up to 0.2 µg/m3 for each additional day downwind. In districts located farther away from industrial sites, the increase is approximately 0.05 µg/m3 per day downwind from a polluting 16 industrial site. This strong effect is partially mitigated by the wind speed at the emission source, as stronger winds help disperse the emissions. The impact of industrial emissions decreases as the distance from the source increases, but even at 3 to 5 kilometers away, there are still significant effects on air pollution, indicating widespread impacts on air quality. Addressing these emissions could lead to substantial health benefits in Tbilisi, particularly in districts located downwind from the most polluting industrial sites. Figure 3: Impact of being downwind on PM2.5 from industrial emissions: Metallurgical production (upper left), Manufacture of construction materials (upper right), Food production (bottom left), Wood processing, and waste management (bottom right) 17 5.2 Traffic jams Figure 4 illustrates the distribution of traffic jams across Tbilisi, focusing on the proportion of traffic jam days from 2019 to 2021, at the census block level. The data reveals that a higher concentration of traffic jams occur near the city center. As distance from the city center increases, the congestion intensity decreases. In Table 10, we find a positive correlation between traffic jams and air pollution. The estimated coefficients demonstrate months with an additional day characterized by traffic jams have a monthly PM2.5 level that is 0.006 µg/m3 higher. This increase represents a 0.02% relative rise compared to the average pollution levels. Column (4) shows an exploration of the non-linear effect of traffic jam days on air pollution. The estimates for the three bins are all positive and statistically significant, indicating a clear association between the number of traffic jam days and air pollution levels. Moreover, the magnitudes of these estimates are quite similar, suggesting a linear relationship between the number of delay days and the corresponding rise in air pollution. This finding implies that as the number of traffic jam days increases, the resulting increase in air pollution is consistent and proportional. This result means that no matter the initial pollution level, traffic jams increase air pollution in a similar way. However, the impact of air pollution are often non-linear. Hence, it may still be optimal from a policy perspective to target the most polluted districts or the ones leading to high exposure of population downwind. Table 10: Traffic jams and air pollution PM2.5 at the district-month level #Delay days 0.003 0.022∗∗∗ 0.006∗∗∗ (0.003) (0.003) (0.002) (0,10] 0.007∗∗ (0.003) (10,20] 0.007∗∗∗ (0.002) (20,31] 0.005∗∗∗ (0.002) Observations 36552 36552 36552 36552 R-square 0.000 0.918 0.958 0.958 Y-mean 25.650 25.650 25.650 25.650 Y-sd 4.369 4.369 4.369 4.369 District FEs Y Y Year, month FEs Y Y Y Notes: Standard errors are clustered at the district level. 18 Figure 4: Proportion of days with traffic jams (speed below 5km/h) 2019 - 2021 0.3 - 10% 10 - 20% 20 - 30% 30 - 40% 40 - 50% 50 - 60% 60 - 70% 70 - 80% 80 - 90% 90 - 97% 6 Co-benefits of carbon pricing policies In this section, we use the estimates of Sections 4 and 5 to assess the potential co-benefits of carbon pricing policies in terms of air pollution. Georgia ratified the Paris Agreement in May 2017, signifying its dedication to global climate action. Building upon this commitment, the state has consistently demonstrated an active and influential role in subsequent Conference of the Parties (COP) gatherings, further underscoring its dedication to addressing climate challenges at both national and international levels. Other countries in the region and the European Union have adopted carbon pricing policies, such as carbon taxes and Emissions Trading Systems (ETS), to achieve their emissions target. While the primary impetus behind the pursuit of carbon pricing remains the urgent need to mitigate the far-reaching effects of climate change, it is essential to underscore the co-benefits of mitigation that come in the form of reductions in local air pollution. These co-benefits can exceed the carbon reduction benefits of some climate policies (Driscoll et al., 2015). Carbon pricing enhances air quality by incentivizing a shift towards cleaner energy sources and more sustainable practices. Addressing air pollution improves human health, hedonic values, and cognitive functions, as shown in the previous sections. The reach of air pollution’s repercussions extends even further, encompassing spheres such as productivity, mental well-being, and the intricate balance of ecosystems, but this analysis falls beyond the scope of our study due to data 19 limitations. We calculate the air pollution-related co-benefits of three different carbon pricing policies: 1. A carbon tax that sets the price of carbon to $25 per metric ton of CO2 equivalent (tCO2e) by 2030, a proposed emissions reduction target for Low-Income Countries (Chateau et al., 2022). 2. A carbon tax that sets the price to $75 by 2030, a proposed emissions reduction target for High-Income Countries. 3. An Emissions Trading System with an emissions reduction target of 19% by 2030. This target corresponds to a carbon price of $90 by 2030, which is the 2023 carbon price in the European Union.1 We simulate the effects of these carbon pricing policies using the Climate Policy Assessment Tool (CPAT) (World Bank, 2022, 2023). This tool is a spreadsheet-based model that helps policymakers assess, design, and implement climate mitigation policies. It allows for the rapid estimation of the effects of climate mitigation, including impacts on energy demand and prices, CO2 and other greenhouse gas emissions, fiscal revenues, GDP, welfare, distributional impacts on households and industries, and development co-benefits. Although the CPAT tool is developed at the national level, we use it at the city level in what follows. Indeed, we combine CPAT forecasts with the above impact estimates for Tbilisi. The underlying assumption is that the carbon pricing consequences forecasted at the national level would hold for Tbilisi. The estimated impacts of the three considered carbon pricing policies on transportation are presented in Figure 5. Introducing a $25 carbon tax (light blue curve) engenders a notable reduction of 1.3% in vehicular distance compared to the baseline by its second year (2025), underscoring its initial impact. This trend gains momentum over time, culminating in a substantial 7.8% reduction by 2036. The $75 carbon tax has a much higher effect, with an estimated 22% traffic reduction by 2036. With an equivalent carbon price of $90 by 2030, the ETS would yield a reduction in vehicular distance of 25%. The expected drop in transportation will likely positively affect air pollution levels. In the first scenario, with a carbon tax of $25 and an associated reduction of traffic congestion of 7.82% 1 Note that in the current version of CPAT, the ETS system is modeled as a carbon tax with 90% efficacy. 20 Figure 5: Effects of carbon pricing on driving in Georgia for three different policy scenarios (Million of vehicle-km) Notes: Authors’ calculation using the WB-IMF Climate Policy Assessment Tool (World Bank, 2022, 2023) in 2036, the CPAT tool projects a consequential decrease in ambient pollution levels of 0.35 µg/m3 . In the second and third scenarios with a carbon tax of $50 and the ETS, the traffic- related reductions in air pollution are 1.00 µg/m3 and 1.14 µg/m3 by 2036. These forecasts are calculated by extending the TM5-FASST model, a source-receptor model with relationships between emissions and concentrations. In this model, the decrease in air pollution stems from decreased vehicular use due to increased fuel prices. The estimates in Table 10 suggest that eliminating traffic jams, either by adopting carbon pricing to reduce the number of used vehicles, incentivizing public transport, or improving urban planning, could reduce air pollution by up to 0.18 µg/m3 . Carbon pricing regulations yield co-benefits in terms of air pollution reduction. Figure 6 presents pollution reduction trajectories in the three considered carbon pricing scenarios. Our results show a discernible decline in atmospheric particulate matter concentration of 0.08 µg/m3 in the first year (2024). The $25 carbon tax yields a reduction in air pollution of 0.43µg/m3 by 2036. The $75 carbon tax decreases air pollution by 1.18µg/m3 over the same time range, and the ETS reduces PM2.5 concentration by 1.35 µg/m3 . These quantifiable improvements in air quality validate the multifaceted advantages of carbon pricing. According to CPAT, the 21 Figure 6: Effects of carbon pricing on air pollution levels in Georgia for three different policy scenarios (µg/m3 ) Notes: Authors’ calculation using the WB-IMF Climate Policy Assessment Tool (World Bank, 2022, 2023) reduction in air pollution is mainly driven by the decrease due to transportation changes (81%- 85%). The second and third most important sources of air pollution are the residential, services, and construction sectors (8.%7-9.5%) and the industrial sector (5.3%-7.5%). This is consistent with estimates in Figure 3 showing the significant impact of industrial emissions on air pollution in downwind districts. Expanding our analysis, we proceed to evaluate the positive co-benefits of carbon pricing policies using the findings presented in Tables 4 and B2. We use the above elasticities to translate the impact of the forecasted air pollution reductions from CPAT into health and real estate impacts with a back-of-the-envelope calculation. Results are presented in Figure 7. With the 0.43µg/m3 reduction in PM2.5 levels due to a carbon tax of $25 per ton, we anticipate a drop in hospitalizations of 0.44% per district day by 2036. Additionally, rental prices would increase by 0.38%. These outcomes underscore the multifaceted advantages associated with enhanced air quality due to the implementation of carbon pricing. When implementing the suggested ETS, the corresponding reduction in hospitalizations would be 1.38% by 2036. The increase in rental price would reach 1.18% by 2036. As such, our results underscore the potentially high co-benefits of carbon pricing policies due to the associated improvements in air pollution. 22 Figure 7: Effects of carbon pricing on respiratory infections (A) and rental prices (B) in Georgia for three different policy scenarios (%) Notes: Authors’ calculation using the WB-IMF Climate Policy Assessment Tool (World Bank, 2022, 2023) 7 Conclusion In summary, this study has quantified the significant externalities of air pollution using natural experiments and instrumental variables. Our results provide evidence linking air pollution to adverse outcomes, including increased hospitalization rates and mental health admissions. Results also suggest that air pollution is associated with lowered rental housing prices. These findings underscore the pressing need for comprehensive action by policymakers, researchers, and society at large to address the multifaceted challenge of air pollution. Furthermore, our research highlights several avenues for future exploration. Given the available 23 data, it is crucial for subsequent studies to investigate additional downstream consequences such as productivity, social inequalities, and environmental justice considerations. Additionally, recognizing the differential impacts of indoor and outdoor air pollution on health, well-being, and societal dynamics, future research should place particular emphasis on distinguishing between these sources. Moreover, we stress the critical importance of extending air pollution research to middle- income countries, where the issue remains relatively understudied despite the growing severity of pollution problems. These nations face unique challenges shaped by their specific economic and environmental contexts, necessitating focused research efforts. By prioritizing investigations in middle-income countries, we can gain a more comprehensive understanding of the global air pollution landscape and develop tailored solutions to mitigate pollution challenges faced by these regions. Finally, our study underscores the crucial role of industrial and traffic-related emissions in contributing to air pollution and the potential co-benefits of climate mitigation policies in terms of air pollution reduction and welfare improvement. We estimate that implementing an Emission Trade System with a carbon price in 2036 equal to the current European Union carbon price in 2023 could yield a reduction in hospitalizations of 1.38% by 2036. It could also lead to an increase in rental price of 1.18% by 2036. These positive externalities of climate policies need to be internalized in cost and benefit analysis of climate action. 24 References Arceo, E., R. Hanna, and P. Oliva (2016). Does the effect of pollution on infant mortality differ between developing and developed countries? evidence from mexico city. The Economic Journal 126 (591), 257–280. Bishop, K. C., J. D. Ketcham, and N. V. Kuminoff (2023). Hazed and confused: the effect of air pollution on dementia. Review of Economic Studies 90 (5), 2188–2214. Brunekreef, B. and S. T. Holgate (2002). Air pollution and health. The lancet 360 (9341), 1233–1242. Chang, T., J. Graff Zivin, T. Gross, and M. Neidell (2016). Particulate pollution and the productivity of pear packers. American Economic Journal: Economic Policy 8 (3), 141–169. Chateau, J., M. F. Jaumotte, and G. Schwerhoff (2022). Economic and environmental benefits from international cooperation on climate policies. International Monetary Fund. Chen, C., C. Liu, R. Chen, W. Wang, W. Li, H. Kan, and C. Fu (2018). Ambient air pollution and daily hospital admissions for mental disorders in shanghai, china. Science of The Total Environment 613, 324–330. Chen, S., P. Oliva, and P. Zhang (2018). Air pollution and mental health: evidence from china. Technical report, National Bureau of Economic Research. Chen, Y., A. Ebenstein, M. Greenstone, and H. Li (2013). Evidence on the impact of sustained exposure to air pollution on life expectancy from chinaâs huai river policy. Proceedings of the National Academy of Sciences 110 (32), 12936–12941. CRRC (2016). Environmental issues in georgia: a concern for all? CRRC (2021). A policy agenda for tbilisi: What voters want. Currie, J., L. Davis, M. Greenstone, and R. Walker (2015). Environmental health risks and housing values: evidence from 1,600 toxic plant openings and closings. American Economic Review 105 (2), 678–709. Davis, L. W. (2011). The effect of power plants on local housing values and rents. Review of Economics and Statistics 93 (4), 1391–1402. Deryugina, T., G. Heutel, N. H. Miller, D. Molitor, and J. Reif (2019). The mortality and medical costs of air pollution: Evidence from changes in wind direction. American Economic Review 109 (12), 4178–4219. Driscoll, C. T., J. J. Buonocore, J. I. Levy, K. F. Lambert, D. Burtraw, S. B. Reid, H. Fakhraei, and J. Schwartz (2015). Us power plant carbon standards and clean air and health co-benefits. Nature Climate Change 5 (6), 535–540. Filippini, M., G. Masiero, and S. Steinbach (2019). The impact of ambient air pollution on hospital admissions. The European Journal of Health Economics 20 (6), 919–931. Fuller, R., P. J. Landrigan, K. Balakrishnan, G. Bathan, S. Bose-O’Reilly, M. Brauer, J. Caravanos, T. Chiles, A. Cohen, L. Corra, et al. (2022). Pollution and health: a progress update. The Lancet Planetary Health 6 (6), e535–e547. Graff Zivin, J., M. Neidell, N. J. Sanders, and G. Singer (2023, April). When externalities collide: Influenza and pollution. Am. Econ. J. Appl. Econ. 15 (2), 320–351. 25 Greenstone, M. and R. Hanna (2014). Environmental regulations, air and water pollution, and infant mortality in india. American Economic Review 104 (10), 3038–3072. IHME (2023). Georgia. https://www.healthdata.org/research-analysis/ health-by-location/profiles/georgia. Accessed: 2023-11-30. Jayachandran, S. (2009). Air quality and early-life mortality: Evidence from indonesiaâs wildfires. Journal of Human resources 44 (4), 916–954. Lavy, V., A. Ebenstein, and S. Roth (2014). The impact of short term exposure to ambient air pollution on cognitive performance and human capital formation. Technical report, National Bureau of Economic Research. Manisalidis, I., E. Stavropoulou, A. Stavropoulos, and E. Bezirtzoglou (2020). Environmental and health impacts of air pollution: a review. Frontiers in public health , 14. Persico, C. and D. E. Marcotte (2022). Air quality and suicide. Technical report, National Bureau of Economic Research. Van Donkelaar, A., M. S. Hammer, L. Bindle, M. Brauer, J. R. Brook, M. J. Garay, N. C. Hsu, O. V. Kalashnikova, R. A. Kahn, C. Lee, et al. (2021). Monthly global estimates of fine particulate matter and their uncertainty. Environmental Science & Technology 55 (22), 15287–15300. Weichenthal, S., L. Pinault, T. Christidis, R. T. Burnett, J. R. Brook, Y. Chu, D. L. Crouse, A. C. Erickson, P. Hystad, C. Li, et al. (2022). How low can you go? air pollution affects mortality at very low levels. Science Advances 8 (39), eabo3381. World Bank (2022, May). Climate policy assessment tool (CPAT). https://www.imf.org/en/ Topics/climate-change/CPAT. Accessed: 2023-10-4. World Bank (2023, September). Carbon Pricing Assessment Tool (CPAT) Documentation. 26 A Additional Figures Test scores Figure A1: Impact of log. PM2.5 concentration on log. UEE exam’s test scores. Notes: The regression includes year-month and school-fixed effects. Points represent the point estimates and the bars show the 95% confidence intervals. B Additional Tables Table B1: Variable descriptions Variable Available time period Raw data resolution Source Air pollution 1998-2022 monthly, 0.01◦ Van Donkelaar et al. (2021) Weather 1998-2022 hourly, 9km ERA5 Hospitalization 2015-2022 daily, hospital-category NCDC for Tbilisi Housing price 2019-2021 daily, transaction LIVO Test scores 2011-2017 yearly, student Unified Entrance Exams Traffic 2019-2021 every 5 minute, road segment MapBox 27 Weak IV and real estate prices In Table B2, we use thermal inversions to instrument air pollution. This specification tackles potential biases from confounders that impact both air pollution levels and housing prices, such as monthly localized changes in economic activity. However, the instrument is not as strong as in the health specification. Our estimated coefficients become insignificant, and the magnitude is close to zero, suggesting that short-term variations in pollution have limited effects on price fluctuations. This is again consistent with a relatively low correlation between pollution levels at the moment of sale (or lease) and purchasers’ expectations about longer-term environmental conditions. Table B2: Air pollution and real estate price: IV ln(Flat sale) ln(Flat rent) ln(PM2.5) 0.008 0.008 0.015 0.015 (0.011) (0.011) (0.025) (0.025) Observations 69307 69307 54750 54750 R-square 0.266 0.266 0.254 0.254 Y-mean 6.747 6.747 1.853 1.853 Y-sd 0.551 0.551 0.498 0.498 Year, month FEs Y Y Y Y District FEs Y Y Y Y Real estate FEs Y Y Notes: This table shows IV regression estimation results. We use the occurrence and strength of thermal inversions as instruments for air pollution. Controls include year, month, district fixed effects in all columns. We further control for real estate fixed effects in Columns (2) and (4). Robust standard errors are clustered at the district level. 28