Policy Research Working Paper 10948 Financial Deepening and Carbon Emissions Intensity Evidence from a Global Sample of Countries Boris Fisera Martin Melecky Dorothe Singer Finance, Competitiveness and Innovation Global Practice October 2024 Policy Research Working Paper 10948 Abstract Financial deepening contributes to economic development, adverse effect of financial deepening: conditional local pro- but its effect on the carbon intensity of production is an jections reveal that in countries with more environmental open empirical question. If banks finance investments in regulations, a stronger rule of law, and a financial system new, greener technologies, they can contribute to lowering that is relatively more market- than bank-based, financial carbon dioxide emissions per unit of output. But if they deepening does not lead to higher carbon dioxide emis- finance investments in more traditional, carbon-intensive sions per dollar of gross domestic product. Specifically, the technologies, they can contribute to increasing carbon results show that countries with an initially lower carbon dioxide emissions per unit of output. This paper studies intensity of production can mitigate the average adverse the impact of financial deepening—an increased provision effect of financial deepening on carbon dioxide emissions of bank credit as a share of gross domestic product—on per dollar of gross domestic product by improving their carbon dioxide emissions per dollar of gross domestic prod- general institutional environment proxied by adherence to uct in a global sample of 125 economies from 1990 to 2019. the rule of law, and, to some extent, by developing their Using a local projections approach, the paper finds that, financial markets. By contrast, countries with an initially on average, financial deepening leads to a relative increase higher carbon intensity of production are better off focusing in carbon dioxide emissions per dollar of gross domestic on environmental regulations to mitigate the unconditional product, indicating that financial institutions finance rela- adverse effect of financial deepening on carbon dioxide tively more carbon-intensive investments and consumption. emissions per dollar of gross domestic product. However, a better institutional environment mitigates this This paper is a product of the Finance, Competitiveness and Innovation 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 boris.fisera@webster.ac.at, mmelecky@worldbank.org, and dsinger@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 Financial Deepening and Carbon Emissions Intensity: Evidence from a Global Sample of Countries Boris Fiseraa, Martin Meleckyb, and Dorothe Singerc Keywords: CO2 emissions, bank credit, carbon intensity of credit, financial deepening, financial institutions JEL Classification: E44, G10, Q50 a boris.fisera@webster.ac.at, Webster Vienna Private University b mmelecky@worldbank.org, World Bank c dsinger@worldbank.org, World Bank 1. Introduction The transition toward a less carbon-intensive economy is costly and requires significant investments. The Paris Agreement from December 2015 committed its signatories to rapidly reduce CO2 emissions to achieve net-zero emissions in the second half of the 21st century. It is estimated that to achieve net zero emissions by 2050, global investments in climate mitigation, both public and private, would need to increase from $0.9 trillion in 2020 to $5 trillion annually by 2030, with the private share in investments needing to increase significantly (Black et al. 2023). The financial sector can play a critical role in the transition by boosting the flow of funds toward green economic activities and investments. Indeed, the Paris Agreement was the first comprehensive climate agreement to recognize the crucial role of finance in lowering carbon emissions. Banks, the largest source of financing for firms in most economies, are a key financial sector actor in that regard. Financial deepening—measured as the increase in the ratio of bank credit to the private sector relative to GDP— has been recognized as an important driver of economic development. A long- standing literature provides empirical evidence that financial deepening contributes to long-run economic growth and development (King and Levine 1993, Beck et al. 2000, Levine et al. 2000, Gorodnichenko and Schnitzer 2013). The long-term benefits of financial deepening for growth and development are particularly significant in developing countries (Demirgüç-Kunt et al. 2013, Beck et al. 2014) and if credit flows to productive firms (as opposed to household consumption; Beck et al. 2012). But whether financial deepening has a beneficial or detrimental effect on CO2 emissions per dollar of GDP is an open empirical question: if financial institutions use the increased credit provision to finance investments in new, greener technologies, they can contribute to lowering CO2 emissions (Mongo et al. 2021, Yu and Liu 2024). By contrast, if financial institutions use it to finance investments in more traditional, carbon-intensive technologies, financial deepening can contribute to increasing CO2 emissions (Minetti 2011, Degryse et al. 2011, Brown et al. 2017). A rapidly growing literature considers the relationship between financial deepening and various dimensions of climate change, including CO2 emissions. For example, Masud et al. (2022) consider the relationship between financial deepening and environmental outcomes such as carbon 2 emissions per capita for 10 ASEAN countries. They find that credit growth can lead to increased emissions, but this effect is mitigated by improved institutional quality. Reghezza et al. (2022) show that European banks reallocated credit away from polluting firms in relative terms after the Paris Agreement. Two closely related papers consider how financial development in terms of the relative importance of (stock) markets compared to financial institutions in the financial system affects CO2 emissions: De Haas and Popov (2023) find that relatively deeper stock markets contribute to reducing CO2 emissions per unit of output faster by funding investments in green technology, especially in carbon-intensive sectors, and possibly also by reallocating investments to more energy-efficient sectors based on a sample of 48 primarily OECD countries. Similarly, Horky and Fidrmuc (2024) conclude for a sample of 32 EU and ASEAN countries that while greater depth of financial institutions reduces renewable energy consumption, deeper capital markets tend to increase renewable energy consumption due to the two parts of the financial sector funding investments in carbon-intensive and renewable energy production, respectively. In this paper, we study the relationship between credit to the private sector and CO2 emissions of an economy, both expressed relative to the dollar value of GDP. We, and the literature, do so because credit and CO2 emissions could be co-trending—as the domestic economy grows, it requires more credit and, generally, emits more greenhouse gases measured in CO2 equivalent terms. However, once scaled by the size of the domestic economy, financial deepening and CO2 emissions, on average, do not show a positive correlation, as we show in greater detail in the next section. This varying response of CO2 emissions per dollar value of GDP to financial deepening across countries is the subject of our study. We aim to condition it on relevant country-level characteristics and identify those that help explain this varying response. We contribute to the literature by studying the effect of financial deepening on CO2 emissions per dollar of GDP in a panel of 125 advanced and emerging economies spanning 1990 to 2019 using the local projections model of Jorda (2005). Taking advantage of the significant country heterogeneity in our sample, we also study the effect of the institutional environment on the response of CO2 emissions per dollar of GDP to financial deepening using conditional local projections. This can help us to identify some policy options that may contribute to reducing the carbon intensity of financial deepening. Namely, because there could be a trade-off between the effect of financial deepening on economic development and the carbon intensity of an economy, 3 we aim to provide some evidence on how this trade-off can be managed. We consider four potentially relevant dimensions of the institutional environment as candidate policy levers (management tools). First, we consider environmental regulations as proxied by the number of climate change laws passed by a country. Reghezza et al. (2022) show that environmental laws do influence bank lending. Second, we consider a broad measure of institutional quality, proxied by the rule of law index. Masud et al. (2023) show that in ASEAN countries with improved institutional quality (as measured by the World Bank’s World Governance Indicators), credit growth can contribute to mitigating carbon emissions per capita. Our third and fourth dimensions of the institutional environment are related to the structure of the financial system: the relative development of financial markets versus institutions and foreign bank ownership. De Haas and Popov (2023) and Horky and Fidrmuc (2024), for example, find that countries with financial systems that are relatively more (stock) market-based are more likely to invest in less carbon- intensive technologies and sectors. Foreign banks can make domestic banking sectors more competitive by bringing capital, expertise, and new technologies (World Bank 2018). Greater foreign bank ownership may thus help mitigate any adverse effect of financial deepening on carbon emissions when foreign banking is associated with relatively more investment in new, green technologies. Finally, we further dissect these conditional responses for countries with high and low initial carbon intensity of production to derive more nuanced policy insights. We find empirical evidence that, on average, financial deepening leads to a relative increase in CO2 emissions per dollar of GDP in our sample of 125 economies between 1990 and 2019. In particular, our results imply that a one-standard-deviation increase in credit-to-GDP leads to an increase in CO2 emissions per dollar of GDP by about 0.6 percentage points over a 5-year horizon. This means that, on average, CO2 emissions per dollar of GDP fall by 3.9 percentage points instead of 4.5 percentage points, all other things equal, over a 5-year horizon. In other words, a one- standard-deviation increase in financial deepening trims off about 13 percent of the decline in CO2 emissions per dollar of GDP that we observe in our sample. While we do not attempt to identify the mechanism for this finding, it is consistent with the literature that finds that banks are technologically conservative, that is they choose to finance existing technology instead of more innovative, green technology to protect the value of their existing, typically more carbon-intensive investments (Minetti 2011, Degryse et al. 2011, Brown et al. 2017). 4 This finding implies the existence of a trade-off: while financial deepening plays a vital role in stimulating economic growth, it can also contribute to higher CO2 emissions per dollar of GDP. Importantly, we find evidence that some dimensions of a country’s institutional environment mitigate this trade-off. Specifically, more robust environmental regulation, stronger rule of law, and a financial system that is relatively more market- than bank-based have helped countries mitigate the unconditional adverse effect of financial deepening on CO2 emissions per dollar of GDP. We find no significant role of foreign bank ownership in managing the trade-off. Moreover, our results show that countries with an initially lower carbon intensity of production can mitigate the effect of financial deepening on CO2 emissions per dollar of GDP by improving their general institutional environment as proxied by the rule of law index. There is also somewhat more limited evidence that having a relatively more market-based financial system can help mitigate the potential trade-off. By contrast, countries with an initially higher carbon intensity of production could manage the trade-off from financial deepening by improving their environmental regulations. Our results hold up in robustness checks, including using an alternative measure of financial deepening, restricting the sample to episodes of credit booms, addressing the Nickell bias and cross-sectional dependence, and using alternative empirical specifications. The remainder of the paper is organized as follows. Section 2 describes the data and presents some stylized facts. Section 3 outlines the empirical methodology, while section 4 discusses the results. Section 5 includes several robustness checks. Section 6 concludes the paper. 2. Data and stylized facts 2.1. Data We use an unbalanced panel dataset of 125 advanced and emerging economies covering 1990- 2019. 1 The frequency of the data is annual with an average time span of 20 years per country. The list of countries included in our sample is reported in Table A.1 in the Appendix. We aim to include all countries for which data is available so that we can provide comprehensive evidence for a large sample of countries. Such a large country sample also allows us to exploit its heterogeneity across institutional factors that may influence the relationship between financial deepening and CO2 1 As we include several lags of both the dependent and explanatory variables in our regressions, 1994 is the first year included in our regression analysis. 5 emissions per dollar of GDP. We exclude from our sample any countries that belong to the World Bank’s Small States Forum 2 as small countries, many of which are island countries, can exhibit significant year-to-year fluctuations in their CO2 emissions and the financial deepening that might confound our results. We also exclude any countries that are offshore financial centers due to their overly large and complex financial sectors that are largely disconnected from direct financing of the domestic real economy. We end our sample in 2019 to avoid potential confounding effects in subsequent years from (i) the COVID-19 pandemic which led to lower economic activity, and (ii) the Russian Federation’s invasion of Ukraine in February 2022 which affected energy markets and might have abruptly affected CO2 emissions per dollar of GDP in some economies (World Bank, 2022). Table A.2 provides detailed descriptions and sources for all variables used in the analysis. Our dependent variable is CO2 emissions in kg per PPP$ of GDP. Our primary explanatory variable, financial deepening, is the annual change in domestic credit to the private sector provided by the financial sector expressed as a percentage of GDP. 3 We use nominal GDP to scale both CO2 emissions and credit by the size of the economy because otherwise, these two variables might co- trend as markets and production grow. Expressing CO2 emissions per dollar of GDP enables us to interpret the dependent variable as a proxy for carbon intensity of the economy, so that we can study whether financial deepening contributes to the transition toward a less carbon-intensive economy. To build our financial deepening measure, we use credit-to-GDP, the most common proxy for financial development (see for instance Arcand et al. 2015 or Dabla-Norris and Srivisal 2013). The main benefit of credit-to-GDP lies in its simplicity, which both makes the indicator available for a large sample of countries and for a straightforward interpretation. The main drawback of this measure is that it does not reflect the developments of all components of the financial system (i.e., financial markets). Nevertheless, because most countries have a bank-based financial system, 4 not only are the developments in credit-to-GDP a relevant proxy for overall financing conditions, but also its environmental consequences are relevant from policy makers’ 2 The World Bank’s Small States Forum includes countries that ‘share common challenges associated with the small size of their economies and vulnerability to exogenous shocks’. 3 The choice of these two variables is in the style of metrics used by financial sector firms to measure their carbon intensity as suggested by the Financial Stability Board’s Task Force on Climate-related Financial Disclosures (2021). 4 This is particularly the case for emerging economies. 6 perspective. We conduct a robustness check using the comprehensive financial development index of Svirydzenka (2016) as an alternative measure of financial deepening. We include several control variables in our regressions that might affect carbon emissions per dollar of GDP, broadly following De Haas and Popov (2023). We control for the level of economic development, proxied by GDP per capita. And since the relationship between economic development and carbon emissions can be non-linear owing to the so-called environmental Kuznets curve (Hu et al. 2018), we also include the squared term of GDP per capita. To account for country and market size, which might affect the attractiveness for innovative investments, we use population (De Haas and Popov, 2023). Finally, the relationship between financial deepening and CO2 emissions per dollar of GDP might be affected by financial crises, because banks might be retreating from more risky, innovative, and potentially less carbon-intensive investments (Hardy and Sever 2021). Furthermore, financial crises, which make access to capital more difficult, might discourage investments—including investments in environmentally friendly technologies (Jalles 2024). To account for the potential effect of crises, we also include a dummy variable for banking crises from Nguyen et al. (2022) among the vector of control variables. Overall, we limit the number of control variables to retain the largest possible sample of countries. To account for time-invariant country-specific factors that might influence CO2 emissions (i.e., renewable energy potential, fossil fuel reserves, urbanization), we include country fixed effects. And, to account for time-specific factors that might influence CO2 emissions in all countries during a particular year, we add time fixed effects. We explore responses of CO2 emissions per dollar of GDP to financial deepening conditional on four variables capturing different dimensions of a country’s institutional environment. First, we use the number of climate change laws from the Climate Change Laws of the World database of the Grantham Research Institute to proxy environmental regulation. Second, we use the rule of law index taken from the World Governance Indicators (WGI) of the World Bank to capture the conditioning role of institutional quality. Third, we study whether the response varies in countries with more market-based versus bank-based financial systems. De Haas and Popov (2023) express this ‘financial structure’ as the ratio of the size of the equity market to the size of the entire financial 7 system. 5 However, we lack data for this variable for many countries that are included in our sample. Instead, we rely on the financial development index of Svirydzenka (2016). This index of financial development distinguishes between financial institutions and financial markets development. For both of these categories, the index further distinguishes between financial depth, access, and effectiveness. We express this conditioning variable as the ratio of financial markets' depth index to financial institutions' depth index. Because measures of financial depth might exhibit some volatility over time, this variable enters our regressions as a 3-year average. Fourth, we condition on the share of foreign-owned banks in the domestic banking sector. Foreign bank ownership is expressed as the share of total banking assets held by foreign banks, and the data is taken from the dataset compiled by Panizza (2020). 2.2. Stylized facts Our data show that average CO2 emissions in kg per PPP $ of GDP for the countries in our sample declined from 0.29 to 0.19 between 1994 and 2019. Over the same period, the average credit-to- GDP ratio increased from 33 percent to 57 percent. Figure A.1 shows the change in CO2 emissions per dollar of GDP and the credit-to-GDP ratio between 1999 and 2019 for 88 out of 125 countries in our sample. Because our panel dataset is unbalanced, we plot the changes for the 20 years between 1999 and 2019 to maximize the number of countries in this figure. Countries fall into one of four categories depending on whether their CO2 emissions per dollar of GDP and credit-to-GDP ratio increased or decreased. Most countries (49) experienced financial deepening coupled with a decrease in CO2 emissions per dollar of GDP over this period. This group includes primarily advanced economies. Meanwhile, 14 countries experienced a decline in CO2 emissions per dollar of GDP and a decrease in financial deepening. At the same time, 22 countries experienced an increase in CO2 emissions per dollar of GDP and financial deepening. Most countries in this group are developing economies. Finally, in 3 countries, credit-to-GDP decreased while CO2 emissions per dollar of GDP increased. 5 The size of the entire financial system was calculated as the sum of the value of publicly traded shares (i.e., size of the equity market) and of the credit extended to the private sector by banks and other credit institutions. 8 However, Figure A.1 does not control for potentially confounding factors and cannot establish causality between the two variables. In the next section, we outline the empirical approach to identify any causal effect of financial deepening on CO2 emissions per dollar of GDP. 3. Empirical methodology To study the effect of financial deepening on CO2 emissions per dollar of GDP, we use the local projections (LP) approach of Jorda et al. (2005). This approach is increasingly used in the empirical literature to obtain an impulse response function (IRF) of a variable of interest to a shock or treatment by estimating a separate regression for each forecast horizon (t+h) 6—that is, regressing the variable of interest over the entire forecast horizon (h) on a set of regressors in the initial time period (t). The IRFs are estimated directly from local projections. As a result, the LP approach enables us to identify the effect of a one-time financial deepening shock on CO2 emissions per dollar of GDP over a selected time horizon (5 years in our case). In other words, it enables us to observe how financial deepening transmits to CO2 emissions per dollar of GDP over the short to medium term in a large panel of countries. We estimate the following baseline LP regression: 2 ℎ ℎ ℎ ℎ 2,+ℎ − 2, = + 1 , + 2 � ,− + 3ℎ � ,+ℎ =1 ℎ=1 ′ +ℎ ∑3 ℎ =0�2,− − 2,−1− � + , + + ,+ℎ ℎ = 1, … ,5 (1) where 2, is our measure of CO2 emissions per dollar of GDP in country i at time t. Parameter h is the forecast horizon for impulse responses, which we set to 5 to model the development of CO2 emissions per dollar of GDP over a 5-year horizon. is our measure of financial deepening, proxied by credit-to-GDP, and our primary variable of interest or treatment indicator. is the 6 The local projections approach is primarily used in macroeconomics. However, it is now increasingly being used in the field of environmental economics: Benatti et al. (2024) employ the LP approach to study the effect of environmental regulation on environmental technology innovations using 5-year projections as we do in this paper, while Jalles (2024) used the LP approach to study the effect of financial crises on country’s resilience to climate change. 9 vector of control variables including GDP per capita, squared GDP per capita, population, and financial crisis dummy. and are country and time fixed effects. We also include two lags of financial deepening and three lags of the dependent variable in the regressions. 7 Furthermore, following Wiese et al. (2024), we also augment equation (1) with the leads of our treatment indicator because financial deepening that occurred in the past might be influencing CO2 emissions per dollar of GDP over the forecast horizon (i.e., ahead in time). Including leads within the LP framework incorporates the Teulings and Zubanov (2014) correction that enables us to avoid the bias caused by overlapping forecast horizons. We estimate equation (1) separately for h=1, …, 5, to generate IRFs of CO2 emissions per dollar of GDP over the 5 years following financial deepening (similar to Benatti et al. 2024), with the ℎ estimated {1 }ℎ=1 coefficients representing the point estimate and the estimated standard errors ℎ ℎ of the 1 coefficients being used to obtain confidence intervals for the IRFs. That is, 1 denotes the cumulative response of CO2 emissions per dollar of GDP h years after financial deepening. The error term in equation (1) might be correlated across countries, but the LP approach enables us to address this issue by clustering the standard errors at the country level. Local projections represent a flexible alternative to vector autoregression (VAR) (see, for example, Auerbach and Gorodnichenko 2012, Jorda and Taylor 2016, and Jalles 2024). The two main advantages of the LP approach over the alternative structural VAR (SVAR) include an easier estimation of non-linear impulse responses and a larger number of degrees of freedom in the estimation of IRFs since it generates the IRFs only for the dependent variable (Furceri et al. 2018). While Plagborg-Moller and Wolf (2021) show that, with lag length tending towards infinity, linear LPs and VARs estimate the same impulse responses. Montiel Olea and Plagborg-Moller (2021) prove that the LP approach robustly handles highly persistent data as well as the estimation of impulse responses over longer horizons—two issues that commonly arise in macroeconomic applications. 7 We keep the number of lags limited due to the short time span of our panel dataset because including additional lags would significantly reduce the time dimension in our analysis. Our number of lags is in line with Furceri et al. (2018) who used a panel dataset with roughly the same time span. 10 In line with Corsetti et al. (2021), we extend equation (1) to study whether the effect of financial deepening varies based on country-level institutional characteristics: 2 ℎ ℎ ℎ ℎ ℎ 2,+ℎ − 2, = + 1 , , + 1 (1 − , ), + 2 � ,− + 3ℎ � ,+ℎ =1 ℎ=1 ′ +ℎ ∑3 ℎ =0�2,− − 2,−1− � + , + + ,+ℎ ℎ = 1, … ,5 (2) where is an indicator variable indicating that the conditioning country-level characteristic is above the sample mean. We use four characteristics: the number of climate change laws, the rule of law index, the ratio of financial markets to financial institutions development, and foreign bank ownership. We express the conditioning variable as a binary indicator, as opposed to a continuous variable. Doing so has several advantages, including: (i) straightforward interpretation of the estimated coefficients, (ii) flexible extension to triple interactions, and (iii) reducing potential ℎ ℎ multicollinearity issues. Based on coefficients {1 }ℎ=1 and {1 }ℎ=1 from equation (2), we can generate the impulse responses of CO2 emissions per dollar of GDP to financial deepening at above and below the average level of the conditioning variable, respectively. Finally, we explore whether financial deepening affects CO2 emissions per dollar of GDP differently in countries with initially higher CO2 emissions intensity than in countries with initially lower CO2 intensity and whether the conditioning characteristics explored in equation (2) affect countries heterogeneously. To do so, we further interact the interaction terms from equation (2) with dummy variables for above/below average CO2 emissions per dollar of GDP. In other words, we expand equation (2) by including triple interaction terms. We measure countries’ initial CO2 intensity of production as their average level of CO2 emissions per dollar of GDP over the previous three years. 8 8 We do so because we have an unbalanced panel dataset and for many countries in our sample data on CO2 emissions per dollar of GPD are not available for the first year of our sample (i.e., 1990). If we used 1990 data to identify countries with above/below sample average level of CO2 emissions intensity, our sample of countries would be significantly reduced. Similarly, if we used the first year that a country’s CO2 emissions per dollar of GDP data become available as the initial CO2 intensity, we would not be able to identify above/below sample average as we would be comparing data across different years. 11 4. Results 4.1. Baseline results Figure 1 shows the estimated baseline IRF of CO2 emissions per dollar of GDP to financial deepening. We find that financial deepening contributes to a persistent increase in CO2 emissions per dollar of GDP over the forecast horizon: an increase in the credit-to-GDP ratio of 1 percentage point increases CO2 emissions per dollar of GDP by 0.11 percent over the next five years. The effect is most pronounced during the first year following the increase in financial deepening and further accumulating over the following four years. The detailed results for our baseline regressions are reported in Table A.3. 9 Our results imply that a one-standard-deviation increase in credit-to-GDP—in our sample about 5.3 percent—leads to a relative increase in CO2 emissions per dollar of GDP by about 0.6 percentage points over a 5-year horizon. CO2 emissions per dollar of GDP fall on average by 4.5 percentage points over a 5-year period. Therefore, after a one-standard-deviation increase in financial deepening, CO2 emissions per dollar of GDP fall by 3.9 percentage points instead of 4.5 percentage points, all other things equal. In other words, a one-standard-deviation increase in financial deepening trims off about 13 percent of the decline in CO2 emissions per dollar of GDP. 9 We find evidence that economic development increases CO2 emissions per dollar of GDP, but only over the medium- term, as the coefficient of the variable GDP per capita is positive and turns statistically significant three years after the financial deepening shock. Moreover, the coefficient of squared term of GDP per capita is negative and turns statistically significant five years after the financial deepening, providing some support for the so-called environmental Kuznets hypothesis over the medium-term. That is, at lower level of economic development, economic growth increases CO2 emissions per dollar of GDP as it is driven by carbon-intensive industries, but at a later stage of economic development, increasing economic development is driven by innovations leading to a reduction in CO2 emissions per dollar of GDP. Moreover, we find evidence that countries that experience higher growth in CO2 emissions per dollar of GDP earlier experience lower growth in CO2 emissions per dollar of GDP over the forecast horizon. 12 Figure 1: Cumulative Response of CO2 Emissions to Financial Deepening Notes: Cumulative impulse response of CO2 emissions (per PPP $ of GDP) to financial deepening. Financial deepening is proxied as an increase in credit-to-GDP. The solid line represents the point estimates and the shaded ℎ areas are 68 and 90 percent confidence bands. The point estimates correspond to 1 coefficients from equation (1). Year 1 (h=1) is the first year during which financial deepening, which occurred in year 0, influences CO2 emissions. X-axis: time in years. Y-axis: deviation in percentage points. 4.2. Conditional results Figure 2 shows the conditional IRFs from estimating equation (2). We consider four dimensions of a country’s institutional environment that might affect the baseline relationship between financial deepening and CO2 emissions per dollar of GDP: (i) the number of climate change- related laws; (ii) general adherence to the rule of law; (iii) more market-based versus bank-based financial systems; and (iv) foreign bank ownership. 10 In countries with an above-average number of climate change laws, CO2 emissions per dollar of GDP do not increase significantly following an increase in financial deepening (Figure 2 panel A). 10 Due to data availability of the conditioning variables, the baseline sample of 125 countries is reduced to 122 countries for the number of climate change laws, 123 countries for the rule of law index, 123 countries for the ratio of financial markets depth to financial institutions depth, and 122 countries for the foreign bank ownership. 13 By contrast, in economies with a below-average number of such laws, CO2 emissions per dollar of GDP increase by 0.1 percent following an increase in credit-to-GDP by 1 percentage point. This suggests that more environmental regulations can mitigate CO2 emissions per dollar of GDP due to financial deepening. This is consistent with the evidence that European banks reallocated credit away from polluting firms in relative terms after the Paris Agreement (Reghezza et al 2022). Moreover, Benatti et al. (2024) find that environmental policy tightening increases firms’ innovation activity in clean technologies that mitigate climate change. Similarly, there is no significant increase in CO2 emissions per dollar of GDP in countries with an above-average rule of law index (Figure 2 panel B). However, in economies with a below-average rule-of-law index, an increase in credit-to-GDP of 1 percentage point increases CO2 emissions per dollar of GDP by almost 0.4 percent. This holds when we consider the 68 percent confidence bands, but not the 90 percent confidence bands given their overlap. These results provide some limited evidence that in countries with a stronger institutional environment, adherence to laws, and lower legal uncertainty banks are relatively more likely to finance more innovative, less carbon- intensive investments. In countries with an above-average market-based financial system, financial deepening does not lead to a statistically significant increase in CO2 emissions per dollar of GDP. But in countries where financial institutions intermediate more credit than financial markets, an increase in credit –to-GDP by 1 percentage point triggers an increase in CO2 emissions per dollar of GDP by almost 0.2 percent. Again, this finding holds when we consider the 68 percent confidence bands, but not the 90 percent confidence bands given their overlap. Our limited finding is consistent with the recent literature that finds banks more risk-averse than capital markets and less likely to finance riskier, newer technologies that might be less carbon-intensive than existing technologies. De Haas and Popov (2023), for example, find that countries with financial systems that are relatively more (stock) market-based contribute to reducing CO2 emissions. Horky and Fidrmuc (2024) conclude that while greater depth of financial institutions reduces renewable energy consumption, deeper capital markets tend to increase renewable energy consumption. 14 Figure 2: Conditional Responses of CO2 Emissions to Financial Deepening Notes: Cumulative impulse response of CO2 emissions (per PPP $ of GDP) to financial deepening. Financial deepening is proxied as an increase in credit-to-GDP. The solid and dashed lines represent the point estimates, and ℎ the shaded areas are 68 and 90 percent confidence bands. The solid line is constructed based on 1 coefficients from equation (2) and captures the response of CO2 emission to financial deepening at above-average level of the ℎ conditioning variable. The dashed line is constructed based on 1 coefficients from equation (2) and captures the response of CO2 emission to financial deepening at below-average level of the conditioning variable. The top panel shows the conditional responses for the following conditioning variables (from left to right): number of climate change laws, rule of law index. The bottom panel shows the conditional responses for the following conditioning variables (from left to right): 3-year average of the ratio of financial markets depth index to financial institutions depth index, foreign bank ownership. Year 1 (h=1) is the first year during which financial deepening, which occurred in year 0, influences CO2 emissions. X-axis: time in years. Y-axis: deviation in percentage points. Finally, we find no evidence that above- or below-average foreign bank ownership has a statistically significantly differential effect on CO2 emissions per dollar of GDP due to financial deepening. Overall, these results suggest that a country’s institutional environment matters for the relationship between financial deepening and CO2 emissions per dollar of GDP. Strengthening the institutional environment can give policy makers important levers to mitigate the carbon intensity of production due to financial deepening. 15 4.3. Heterogeneity across countries with low and high initial CO2 emissions per dollar of GDP Figure 3 depicts the conditional IRFs including the triple interaction effect of financial deepening, distinguishing countries above/below average CO2 emissions per dollar of GDP and conditioning on the variables that produced significantly differing IRFs when we used simple interaction terms in the previous section—that is, the number of climate change laws, the rule of law value, and the ratio of financial markets to financial institutions. We find some evidence that different institutional factors matter depending on whether a country is above or below the average CO2 emissions per dollar of GDP group. Figure 3, column 1, indicates that our earlier result of more environmental regulations mitigating the increase in CO2 emissions per dollar of GDP due to financial deepening is driven by countries with initially more carbon-intensive economies. Only for countries with initially higher CO2 intensity do we observe that more climate change laws mitigate this adverse effect. In less carbon- intensive economies, where it is presumably more challenging to reduce CO2 emissions per dollar of GDP, more environmental regulations no longer have a significant differential effect. Figure 3, column 2, shows that only in countries with a less carbon-intensive economy, more adherence to the rule of law helps mitigate the adverse effect of financial deepening on CO2 emissions. Based on this finding, we conjecture that, for economies which are already less carbon- intensive and for which it is thus more challenging to reduce the CO2 emissions per dollar of GDP further, more adherence to the rule of law is important because it creates a stable environment for firms to invest in newer technologies with lower carbon footprint and for banks to finance such investments. Figure 3, column 3, provides some limited evidence that only in countries with initially lower CO2 emissions per dollar of GDP does a relatively more market-based financial system limit the carbon intensity of production due to financial deepening. This holds when we consider the 68 percent confidence bands, but not the 90 percent confidence bands given their overlap. For countries with a high initial carbon intensity of production, we do not observe any significant heterogeneity between the conditional responses. 16 Figure 3: Conditional Responses of CO2 Emissions to Financial Deepening by CO2 Intensity Notes: Cumulative impulse response of CO2 emissions (per PPP $ of GDP) to financial deepening. Financial deepening is proxied as an increase in credit to GDP. The solid and dashed lines represent the point estimates, and the shaded areas are 68 and 90 percent confidence bands. The solid line captures the response of CO2 emission to financial deepening at above-average level of the conditioning variable. The dashed line captures the response of CO2 emission to financial deepening at below-average level of the conditioning variable. From left to right, the figures show the responses of CO2 emissions to financial deepening conditional on the number of climate change laws, on the rule of law index, and on the ratio of financial markets depth index to financial institutions depth index. The top row shows the responses at below-average level of initial CO2 intensity. The bottom row shows the responses at above-average level of initial CO2 intensity. Year 1 (h=1) is the first year during which financial deepening, which occurred in year 0, influences CO2 emissions. X-axis: time in years. Y-axis: deviation in percentage points. 5. Robustness checks 5.1. Alternative measure of financial deepening While the ratio of credit-to-GDP is the most common measure of financial development (for example, Arcand et al. 2015, Dabla-Norris and Srivisal 2013, De Haas and Popov 2023), several recent empirical studies (for example, Hasan et al., 2020; Horky and Fidrmuc, 2024) have used the overall financial development index of Svirydzenka (2016) as an alternative proxy for financial 17 development. Unlike credit-to-GDP, this index captures aspects of financial development other than financial depth such as financial access and efficiency. Moreover, this index also captures financial market development. 11 Figure A.2 reports the results of this robustness check. These results corroborate our baseline findings, confirming that an increase in financial development beyond economic development triggers a persistent increase in CO2 emissions per dollar of GDP. 5.2. Alternative identification of the effect of financial deepening The next two robustness checks use alternative identifications of the effect of financial deepening. Our main measure of financial deepening, the ratio of credit-to-GDP, comprises changes that are positive or negative as well as large or small. To sharpen our identification, we focus on credit boom episodes—that is, episodes of substantial financial deepening. We identify credit boom episodes as years during which the 3-year average growth of credit-to-GDP exceeds the one standard deviation of the credit-to-GDP ratio for our sample. Moreover, because the ratio might indicate a credit boom episode even when GDP falls more than private credit, we restrict credit boom episodes to years when GDP growth is positive. We re-estimate our baseline regressions with an indicator variable for credit boom episodes and report the results in Figure A.3 (dashed line). This robustness check confirms our findings concerning the significantly adverse effect of financial deepening on CO2 emissions per dollar of GDP. When we restrict our sample to episodes involving substantial financial deepening, the magnitude of the estimated effect is more than 20 times larger than in the baseline regressions: a credit boom episode leads to a cumulative increase in CO2 emissions per dollar of GDP of about 2.5 percent in the five years after the start of the credit boom. While focusing only on credit boom episodes enables us to sharpen the identification of the effect of financial deepening, credit boom episodes might occur during years that differ from the years without a credit boom (for example, economic growth is higher). In other words, observations for boom and non-boom years might be drawn from different populations. In such a case, selection bias might arise (Wiese et al., 2024). To address this potential issue, we use the augmented inverse 11 Due to the high correlation between the overall index of financial development and GDP per capita (0.79), we exogenize the financial development index relative to GDP per capita by regressing the financial development index on GDP per capita and year time effects (see for example Fisera 2024). We then use the residuals from this regression as an exogenized financial development index. 18 probability weighted (AIPW) estimator of Jorda and Taylor (2016). The AIPW estimator involves estimating a first-stage probit regression that explains the probability of the credit boom episode (i.e., treatment) occurring in the future. 12 Subsequently, the predicted probabilities obtained from the first-stage regression are used to re-weight the sample on which the local projections regressions are estimated. This re-weighting assigns higher weights to credit boom episodes that are unlikely based on the first-stage regressions, allowing us to re-randomize the treatment. We report the results obtained with the AIPW estimator in Figure A.3 (solid line). Interestingly, when addressing the potential selection bias, we find the effect of credit boom episodes on CO2 emissions per dollar of GDP to be even more adverse: a credit boom episode leads to a cumulative increase in CO2 emissions per dollar of GDP of more than 4 percent in the five years after the boom started. 5.3. Addressing the Nickell bias and cross-sectional dependence The next robustness checks address two issues that might arise in panels with a relatively shorter time series: the Nickell bias and cross-sectional dependence. The Nickell bias arises in dynamic panel models with fixed effects owing to the correlation between the demeaned lagged dependent variable and the error term (Nickell 1981). To address this bias, (i) we re-estimate our baseline regression specification without the lags of the dependent variable, and (ii) re-estimate the baseline regression specification with the bootstrap-based bias-corrected fixed-effects (BCFE) estimator of Everaert and Pozzi (2007) and De Vos et al. (2015) which uses an iterative bootstrap procedure to address the Nickell bias. We report the results of these two robustness checks in the first two panels of Figure A.4. These results corroborate our baseline findings although, in the case of the BCFE estimator, the positive effect of financial deepening on CO2 emissions per dollar of GDP is reduced in size. Next, in the last panel of Figure A.4, we use the Driscoll-Kraay standard errors 13 to obtain the confidence bands for the point estimates from the baseline regression to account for possible cross- 12 We use the following control variables in the first-stage probit regression that explains the probability of the credit boom episode in the future: GDP (PPP) per capita, population, banking crisis dummy, net FDI inflows (as a share of GDP), Chinn-Ito financial openness index of Chinn and Ito (2016), annual change in the exchange rate, financial system deposits (as a share of GDP), and indicator variables for a credit boom episode occurring during the three preceding years. 13 As opposed to standard errors clustered at country-level that we use in the baseline regressions, which enable us to address the cross-correlation of errors across countries. 19 sectional dependence, which arises when error terms across different countries in the panel are correlated (i.e., due to some common factors affecting all countries or due to spillovers between countries). Once again, the results of this robustness check corroborate our baseline findings. 5.4. Alternative empirical methodology – panel VAR Finally, we explore the robustness of our baseline results by estimating our results using a panel VAR (PVAR) model as an alternative identification method. In line with Towbin and Weber (2013), we estimate a simple recursive PVAR model with fixed effects and use OLS to estimate the model equation by equation. Following Berdiev and Saunoris (2016), we construct a simple three-variable PVAR consisting of our two variables of interest (ratio of credit-to-GDP and CO2 emissions per dollar of GDP) and GDP per capita, with variables entering the model in first differences. The procedure to recursively identify the structural shocks is based on the Cholesky decomposition. The ordering of the variables assumes that the CO2 emission intensity of an economy is driven by long-term technology adoption primarily through large-scale, long-term investments in energy and transportation sectors as well as in building technologies. By contrast, financial deepening is assumed to be primarily driven by medium-term technological advancements and investments in lending technology, evolving lending standards, and gradual improvements in financial infrastructure. Therefore, we impose a zero restriction in our PVAR, preventing the financial deepening shock from influencing CO2 emissions per dollar of GDP contemporaneously. However, we keep its effect unrestricted with a one-period lag. The impulse response function reported in Figure A.5 confirms our baseline results by showing that a financial deepening shock significantly increases CO2 emissions per dollar of GDP. 6. Conclusion Addressing climate change remains a key concern for the public and policy makers alike. The transition toward a less carbon-intensive economy involves significant monetary costs and requires large investments. The financial sector can play a critical role in the transition by boosting the flow of funds toward green economic activities and investments. However, whether financial deepening increases or decreases CO2 emissions remains an open empirical question: If financial institutions finance more traditional, carbon-intensive technologies instead of newer and greener technologies, 20 financial deepening might increase carbon intensity of the economy. At the same time, financial deepening plays a vital role in facilitating economic development. In this paper, we study the average effect of financial deepening on economic growth, and how the potential trade-off between economic development and carbon intensity of an economy could be managed. We find empirical evidence that, on average, financial deepening leads to a relative increase in CO2 emissions per dollar of GDP over our sample period from 1990 to 2019. In particular, our baseline result implies that a one-standard-deviation increase in credit-to-GDP leads to a relative increase in CO2 emissions per dollar of GDP by about 0.6 percentage points over a 5-year horizon, meaning that CO2 emissions per dollar of GDP fall by 3.9 percentage points instead of 4.5 percentage points, all other things equal, over a 5-year horizon. However, we find evidence that some dimensions of a country’s institutional environment can mitigate this adverse effect. Our results show that more environmental regulation, stronger rule of law, and a financial system that is relatively more market- than bank-based can help countries mitigate the unconditional adverse effect of financial deepening on CO2 emissions per dollar of GDP. Thus, by improving the institutional environment, countries can effectively manage the trade-off between the effect of financial deepening on economic development and carbon intensity. Moreover, our results indicate that countries with an initially lower carbon intensity of production can mitigate the adverse effect of financial deepening on CO2 emissions per dollar of GDP by improving their general institutional environment as proxied by the rule of law index. There is also some more limited evidence that countries with relatively more market-based financial systems can mitigate this effect. By contrast, countries with an initially higher carbon intensity of production can mitigate the adverse effect by improving their environmental regulations. Our results hold up in robustness checks, including using an alternative measure of financial deepening, restricting the sample to episodes of credit booms, and using alternative empirical identification through a VAR model. Our results have important policy implications. They provide novel evidence that timely and decisive reforms of the institutional environment and the development of financial markets can leverage the financial system in supporting the green transition. 21 References Arcand, J-L, E. Berkes, and U. Panizza (2015). Too Much Finance? Journal of Economic Growth 20, 105-148. Auerbach, A., and Y. Gorodnichenko (2012). Fiscal Multipliers in Recession and Expansion. In Fiscal Policy After the Financial Crisis, ed. Alberto Alesina and Francesco Giavazzi. Cambridge, MA: National Bureau of Economic Research. Beck, T., R. Levine, and N. Loayza (2000). Finance and the Sources of Growth. Journal of Financial Economics 58 (1-2), 261–300. Beck, T., B. Büyükkarabacak, F. K. Rioja, and N. T. Valev (2012). Who Gets the Credit? And Does It Matter? Household vs. Firm Lending Across Countries, The B.E. Journal of Macroeconomics 12 (1), 1-46. Beck, T, H. Degryse, and C. Kneer. 2014. Is More Finance Better? Disentangling Intermediation and Size Effects of Financial Systems, Journal of Financial Stability 10 (C), 50-64. Benatti, N., M. Groiss, P. Kelly, and P. Lopez-Garcia (2024). The Impact of Environmental Regulation on Clean Innovation: Are There Crowding Out Effects? ECB Working Paper No. 2024/2946, European Central Bank. Berdiev, A. Z., and J. W. Saunoris (2016). Financial Development and the Shadow Economy: A Panel VAR Analysis. Economic Modelling 57, 197-207. Black, S., I. Parry, and K. Zhunussova (2023). Is the Paris Agreement Working? A Stocktake of Global Climate Mitigation. IMF Staff Climate Note 2023/002. Washington, DC: International Monetary Fund. Brown, J., G. Martinsson, and B. Petersen (2017). Stock Markets, Credit Markets, and Technology-led Growth, Journal of Financial Intermediation 32, 45-59. Chinn, M. D. and H. Ito (2006). What Matters for Financial Development? Capital Controls, Institutions and Interactions. Journal of Development Economics 81 (1), 163–192. 22 Corsetti, G., K. Kuester, G. J. Muller, and S. Schmidt (2021). The Exchange Rate Insulation Puzzle. ECB Working Paper No 2630, European Central Bank. Dabla-Norris, E. and N. Srivisal (2013). Revisiting the Link Between Finance and Macroeconomic Volatility. IMF Working Paper WP/13/29, International Monetary Fund. De Haas, R., and A. Popov (2023). Finance and Green Growth. The Economic Journal 133 (650), 637-668. Demirgüç-Kunt, A., E. Feyen, and R. Levine (2013). The Evolving Importance of Banks and Securities Markets, The World Bank Economic Review 27 (3), 476–490. De Vos, I., G. Everaert, and I. Ruyssen (2015). Bootstrap-based Bias Correction and Inference for Dynamic Panels with Fixed Effects. The State Journal 15 (4), 986-1018. Degryse, H., T. Roukny, and J. Tielens (2022). Asset Overhang and Technological Change. CEPR Discussion Paper 17507. Everaert, G., and L. Pozzi (2007). Bootstrap-based Bias Correction for Dynamic Panels. Journal of Economic Dynamics and Control 31 (4), 1160-1184. Fisera, B. (2024). Exchange Rates and the Speed of Economic Recovery: The Role of Financial Development. Economic Systems 48 (1), 101165. Furceri, D., P. Loungani, and A. Zdzienicka (2018). The Effects of Monetary Policy Shocks on Inequality. Journal of International Money and Finance 85 (July 2018), 168–186. Gorodnichenko, Y. and M. Schnitzer (2013). Financial Constraints and Innovation: Why Poor Countries Don’t Catch Up. Journal of the European Economic Association 11 (5), 1115–1152. Hardy, B. and C. Sever (2021). Financial Crises and Innovation. European Economic Review 138, 103856. Hasan, I., R. Horvath, and J. Mares (2020). Finance and Wealth Inequality. Journal of International Money and Finance 108, 102161. Horky, F., and J. Fidrmuc (2024). Financial Development and Renewable Energy Adoption in EU and ASEAN Countries. Energy Economics 131 (March 2024), 107368. 23 Hu, H., N. Xie, D. Fang, and X. Zhang (2018). The Role of Renewable Energy Consumption and Commercial Services Trade in Carbon Dioxide Reduction: Evidence from 25 Developing Countries. Applied Energy 211, 1229-1244. Jalles, J. T. (2024). Financial Crises and Climate Change. Comparative Economic Studies 66, 166- 190. Jorda, O. (2005). Estimation and Inference of Impulse Responses by Local Projections. American Economic Review 95 (1), 161–182. Jorda, O. and A. Taylor (2016). The Time for Austerity: Estimating the Average Treatment Effect of Fiscal Policy. The Economic Journal 126, 219-255. King, R. G. and R. Levine (1993). Finance and Growth: Schumpeter Might be Right. The Quarterly Journal of Economics 108 (3), 717–737. Levine, R., N. Loayza, and T. Beck (2000). Financial Intermediation and Growth: Causality and Causes. Journal of Monetary Economics 46 (1), 31–77. Masud, M. M., A. H. M. Noman, R. Akhtar, S. K. A. Selvarajan, and A. Al-Mamun (2024). Does Credit Growth Mitigate Emission Intensity in ASEAN Countries? Journal of International Development 36 (2), 1324–1349. https://doi.org/10.1002/jid.3857 Mongo, M., F. Belaid, and B. Ramdani (2021). The Effects of Environmental Innovations on CO2 Emissions: Empirical Evidence from Europe. Environmental Science & Policy 118, 1-9. Montiel Olea, J. L., and M. Plagborg-Moller (2021). Local Projections Inference Is Simpler and More Robust Than You Think. Econometrica 89, 1789-1823. Minetti, R. (2011). Informed Finance and Technological Conservatism. Review of Finance 15 (3), 633–692. Nguyen, T. C., V. Castro, and J. Wood (2022). A New Comprehensive Database of Financial Crises: Identification, Frequency, and Duration. Economic Modelling 108, 105770. Nickell, S. (1981). Biases in Dynamic Models with Fixed Effects. Econometrica 49 (6), 1417- 1426. 24 Panizza, U. (2023). Bank Ownership Around the World. Working Paper No. HEIDWP07-2023, Geneva Graduate Institute. Plagborg-Moller, M., and C. K. Wolf (2021). Local Projection and VARs Estimate Same Impulse Responses. Econometrica 89, 955-980. Reghezza, A., Y. Altunbas, D. Marques-Ibanez, C. R. d’Acri, M. Spaggiari (2022). Do Banks Fuel Climate Change? Journal of Financial Stability 62, 101049. Sastry, P., E. Verner, and D. Marquez-Ibanez (2024). Business as Usual: Bank Net Zero Commitments, Lending, and Engagement. NBER Working Paper No. 32402, National Bureau of Economic Research. Siddiqui., D. A. and Q. M. Ahmed (2013). The Effect of Institutions on Economic Growth: A Global Analysis based on GMM Dynamic Panel Estimation. Structural Change and Economic Dynamics 24 (March 2013), 18-33. Svirydzenka, K. (2016). Introducing a New Broad-based Index of Financial Development. IMF Working Paper WP/16/5, International Monetary Fund. Task Force on Climate-related Financial Disclosures (2021). “Implementing the Recommendations of the Task Force on Climate-related Financial Disclosures.” Teulings, C. and N. Zubanov (2014). Is Economic Recovery a Myth? Robust Estimation of Impulse Responses. Journal of Applied Econometrics 29, 497-514. Towbin, P., and S. Weber (2013). Limits of Floating Exchange Rates: The Role of Foreign Currency Debt and Import Structure. Journal of Development Economics 101, 179-194. Wiese, R., J. T. Jalles and J. de Haan (2024). Structural Reforms and Income Distribution: New Evidence for OECD Countries. Oxford Economic Papers, forthcoming. World Bank (2018). Global Financial Development Report 2017/2018: Bankers without Borders. Washington, DC: World Bank. doi:10.1596/978-1-4648-1148-7. World Bank (2022). Energy Crisis: Protecting Economies and Enhancing Energy Security in Europe and Central Asia. Washington, DC: World Bank. http://hdl.handle.net/10986/38101 25 Yu, H., and H. Liu (2024). Impact of Digitalization on Carbon Productivity: An Empirical Analysis of 136 Countries. Scientific Reports 14, 5094. 26 Appendix Table A.1: List of Countries Albania Denmark Lao PDR Romania Algeria Dominican Republic Latvia Russian Federation Angola Ecuador Lebanon Rwanda Argentina Egypt, Arab Rep. Libya Saudi Arabia Armenia El Salvador Lithuania Senegal Australia Estonia North Macedonia Serbia Austria Ethiopia Madagascar Sierra Leone Azerbaijan Finland Malaysia Singapore Bangladesh France Mali Slovak Republic Belarus Georgia Mauritania Slovenia Belgium Germany Mexico South Africa Benin Ghana Moldova Spain Bolivia Greece Mongolia Sri Lanka Bosnia and Herzegovina Guatemala Morocco Sudan Brazil Guinea Mozambique Sweden Bulgaria Haiti Myanmar Switzerland Burkina Faso Honduras Nepal Tajikistan Burundi Hungary Netherlands Tanzania Cambodia India New Zealand Thailand Cameroon Indonesia Nicaragua Togo Canada Iran, Islamic Rep. Niger Tunisia Central African Rep. Ireland Nigeria Türkiye Chad Israel Norway Uganda Chile Italy Oman Ukraine China Japan Pakistan United Arab Emirates Colombia Jordan Panama United Kingdom Congo, Dem. Rep. Kazakhstan Paraguay United States Congo, Rep. Kenya Peru Uruguay Costa Rica Korea, Rep. Philippines Uzbekistan Côte d'Ivoire Kuwait Poland Viet Nam Croatia Kyrgyz Republic Portugal Zambia Czechia 27 Table A.2: Description and Sources of Variables Variable Description Source CO2 emissions Carbon dioxide (CO2) emissions expressed in kg per 2017 PPP $ of WB WDI GDP; CO2 emissions are those stemming from the burning of fossil fuels and the manufacture of cement. They include carbon dioxide produced during consumption of solid, liquid, and gas fuels and gas flaring; the variable enters the regressions expressed in natural logarithms Financial Annual change in domestic credit to private sector expressed as a % of WB WDI deepening GDP; also referred as to credit-to-GDP GDP (PPP) per GDP per capita, PPP (constant 2017 international $); the variable enters WB WDI capita the regressions expressed in natural logarithms Population Total population expressed in millions; the variable enters the WB WDI regressions expressed in natural logarithms Banking crisis A dummy variable, which takes the value of 1 when a banking crisis Nguyen et al. takes place in the country and 0 otherwise (2022) Financial A dummy variable, which takes the value of 1 for above-average value IMF markets to of the financial markets to financial institutions ratio in our sample and financial 0 otherwise; Ratio was calculated based on the variables financial institutions markets depth index (FMD) and financial institutions depth index ratio (FID); FMD is an index of relative ranking of countries and it compiles data on stock market capitalization, stocks traded, international debt securities of government and total debt securities of corporations; FID is an index of relative ranking of countries and it compiles data on bank credit to private sector, pension fund assets, mutual fund assets and insurance premiums Number of A dummy variable, which takes the value of 1 for above-average Grantham climate change number of climate change laws in our sample and 0 otherwise; Number Research laws of climate change laws is measured as cumulative sum of climate Institute change-related laws or legislative acts (e.g. acts, laws, decree-laws), which were passed by a parliament or equivalent legislative authority Rule of law A dummy variable, which takes the value of 1 for above-average value WB WGI index of the rule of law index in our sample and 0 otherwise; Rule of Law estimate captures perceptions of the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence Foreign bank A dummy variable, which takes the value of 1 for above-average value Panizza ownership of foreign bank ownership in our sample and 0 otherwise; Foreign bank (2023) ownership is measured as the percentage of the total banking assets that are held by foreign banks. A foreign bank is a bank where 50 percent or more of its shares are owned by foreigners.; no development banks CO2 intensity A dummy variable, which takes the value of 1 for above-average value WB WDI of CO2 intensity in our sample and 0 otherwise; CO2 intensity is measured as the average CO2 emissions expressed in kg per 2017 PPP $ of GDP over the previous three years. Notes: WB = World Bank; WDI = World Development Indicators database; IMF = International Monetary Fund; WGI = World Governance Indicators; 28 Table A.3: CO2 Emissions and Financial Deepening (1) (2) (3) (4) (5) Δ CO2 emissions Δ CO2 emissions Δ CO2 emissions Δ CO2 emissions Δ CO2 emissions h=1 h=2 h=3 h=4 h=5 Financial deepening 0.067** 0.071** 0.090** 0.116** 0.107** (0.026) (0.033) (0.042) (0.046) (0.052) L1. Financial deepening 0.014 0.046 0.078** 0.074* 0.113*** (0.021) (0.031) (0.035) (0.040) (0.043) L2. Financial deepening 0.046** 0.081** 0.083** 0.107** 0.136*** (0.023) (0.031) (0.036) (0.042) (0.046) F1. Financial deepening 0.042 0.100*** 0.095** 0.119** 0.136** (0.036) (0.037) (0.048) (0.049) (0.058) F2. Financial deepening -0.005 0.050 0.067 0.083 (0.040) (0.040) (0.048) (0.060) F3. Financial deepening -0.006 0.030 0.051 (0.049) (0.046) (0.055) F4. Financial deepening 0.003 0.020 (0.057) (0.053) F5. Financial deepening 0.044 (0.059) L1. Δ CO2 emissions -0.148** -0.249*** -0.304*** -0.311*** -0.332*** (0.061) (0.073) (0.077) (0.080) (0.082) L2. Δ CO2 emissions -0.111*** -0.176*** -0.187*** -0.204*** -0.253*** (0.032) (0.047) (0.055) (0.058) (0.055) L3. Δ CO2 emissions -0.077*** -0.095*** -0.117*** -0.162*** -0.183*** (0.024) (0.036) (0.044) (0.042) (0.048) GDP (PPP) per capita 0.079 0.242 0.448* 0.682** 0.869** (0.100) (0.178) (0.238) (0.303) (0.357) GDP (PPP) per capita squared -0.000 -0.000 -0.000 -0.000 -0.000* (0.000) (0.000) (0.000) (0.000) (0.000) Population 0.736 1.249 0.039 -1.172 -1.513 (3.423) (6.084) (9.145) (11.839) (14.610) Banking crisis (0/1) 1.417 1.470 0.904 1.564 1.443 (1.538) (1.595) (1.962) (2.277) (2.580) Constant -52.729 -141.175* -239.089** -358.426*** -453.617*** (38.228) (71.333) (94.478) (118.143) (139.569) Observations 2,432 2,305 2,180 2,058 1,937 Countries 125 123 121 120 119 Notes: L stands for lags and F stands for leads. The models are based on equation (1). The coefficients of variable ℎ Financial deepening are the 1 coefficients from equation (1) and they represent the point estimates from Figure 1. Standard errors clustered at country-level are in parentheses. * indicates significance at 10 %, ** at 5 %, and *** at 1 % level. 29 Figure A.1: Change in CO2 emissions per dollar of GDP and credit-to-GDP between 1999 and 2019 Notes: The figure shows data for 88 countries in our sample for which we have data available for both the year 1999 and the year 2019. 30 Figure A.2: Cumulative Response of CO2 Emissions to Financial Deepening - with Financial Development Index of Svirydzenka (2016) Notes: Cumulative impulse response of CO2 emissions (per PPP $ of GDP) to financial deepening. Financial deepening is proxied as an increase in the overall index of financial development of Svirydzenka (2016). The solid line represents the point estimates and the shaded areas are 68 and 90 percent confidence bands. Year 1 (h=1) is the first year during which financial deepening, which occurred in year 0, influences CO2 emissions. X-axis: time in years. Y-axis: deviation in percentage points. 31 Figure A.3: Cumulative Response of CO2 Emissions to Credit Boom Episodes - AIPW Estimator vs. Simple LP Approach Notes: Cumulative impulse response function of CO2 emissions (per PPP $ of GDP) to a credit boom episode. Credit boom episode takes place when the 3-year average change in credit-to-GDP exceeds one standard deviation. The solid line represents the point estimates obtained with the AIPW estimator of Jorda and Taylor (2016). The shaded areas are the bootstrapped 68 and 90 percent confidence bands. The dashed line represents the point estimates obtained with the simple LP approach. Year 1 (h=1) is the first year during which credit boom episode, which occurred in year 0, influences CO2 emissions. X-axis: time in years. Y-axis: deviation in percentage points. 32 Figure A.4: Cumulative Response of CO2 Emissions to Financial Deepening – addressing Nickell bias and cross-sectional dependence Notes: Cumulative impulse response of CO2 emissions (per PPP $ of GDP) to financial deepening. Financial deepening is proxied as an increase in credit-to-GDP. The solid line represents the point estimates and the shaded areas are 68 and 90 percent confidence bands. From left to right, the figures show the following robustness checks: i) omitting the lags of the dependent variable from the baseline regression specification; ii) using the BCFE estimator to re-estimate the baseline regression specification; iii) using Driscoll-Kraay standard errors to calculate the confidence bounds for the point estimates from the baseline regression specification. Year 1 (h=1) is the first year during which financial deepening, which occurred in year 0, influences CO2 emissions. X-axis: time in years. Y-axis: deviation in percentage points 33 Figure A.5: Cumulative Response of CO2 Emissions to a Financial Deepening Shock - PVAR Model Notes: Cumulative impulse response of CO2 emissions (per PPP $ of GDP) to a financial deepening shock. The solid line represents the point estimates obtained with the panel VAR (PVAR) model. The shaded areas are the bootstrapped 90 percent confidence bands. X-axis: time in years. Y-axis: deviation in percentage points. 34