Policy Research Working Paper 11020 Impact of Temperature Uncertainty on Firm Growth A Grid-Level Analysis Jangho Yang Christian Schoder Development Economics A verified reproducibility package for this paper is Development Policy Team available at http://reproducibility.worldbank.org, January 2025 click here for direct access. Policy Research Working Paper 1120 Abstract This study examines the impact of temperature uncertainty across industries with differing levels of investment irre- on firm fixed capital growth using a unique dataset that versibility and among countries with varying income levels. merges extensive firm-level financial data with detailed grid- Firms in industries characterized by high investment irre- level weather data. The analysis reveals a strong negative versibility and those operating in higher-income countries relationship between temperature uncertainty and fixed experience more pronounced declines in fixed asset growth capital growth. Furthermore, the impact varies significantly due to temperature uncertainty. This paper is a product of the Development Policy Team, 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 j634yang@uwaterloo.ca and cschoder@worldbank.org. A verified reproducibility package for this paper is available at http://reproducibility.worldbank.org, click here for direct access. RESEA CY LI R CH PO TRANSPARENT ANALYSIS S W R R E O KI P NG PA 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 Impact of Temperature Uncertainty on Firm Growth: A Grid-Level Analysis Jangho Yang ∗ 1,2 and Christian Schoder3 1 Management Sciences, Faculty of Engineering, University of Waterloo 2 Institute for New Economic Thinking, University of Oxford 3World Bank Keywords: temperature uncertainty, firms’ fixed asset growth, investment irreversibility, income- level heterogeneity JEL codes: Q54, D22, D25 ∗ We would like to acknowledge two research assistants at the University of Waterloo, Edward Jeong, Yuhan Zhang and Joanna Yang, who greatly helped to collect and process the data. All errors are our own. Contacts: j634yang@uwaterloo.ca 1 Introduction The effects of rising temperatures on economic activities have been extensively documented over the past few years. While quantifying the socioeconomic impacts accurately has proven difficult (Tubiello & Rosenzweig 2008, Piontek et al. 2021), and results often contain substantial uncertainty, sometimes leading to conflicting outcomes (Newell et al. 2021), there is a general consensus that rising tempera- tures have an overall adverse effect on economic activities, especially in terms of GDP levels (Newell et al. 2021).1 What is less understood is temperature variability, defined as the change in daily temperatures over a given period, and its potential impact on economic activities.2 Unlike the clear evidence of rising global temperatures, there is no general consensus on global trends in temperature variability (Huntingford et al. 2013, Olonscheck et al. 2021). Although some studies suggest a global increase (Alessandri & Mumtaz 2022), changes in temperature variability are highly region-specific, with recent evidence indicating that it may be increasing more rapidly in lower-income countries (Bathiany et al. 2018). However, the implications of this variability for economic activities remain unclear. Sectors like agriculture, where certain species are particularly vulnerable to increased temperature variability (Vasseur et al. 2014), may experience direct and measurable consequences, whereas the effects on other sectors, such as technology and services, are less predictable. To examine the cross-industry, economy-wide effects of temperature changes, this paper focuses on temperature uncertainty—the unexpected fluctuations in temperature that deviate from expected pat- terns. Our main interest lies in understanding how location-specific temperature uncertainty, as part of the broader uncertainties faced by economic agents, influences overall economic behavior across in- dustries. We focus on firm-level activities, particularly investment patterns (measured by fixed asset growth), and how the private sector, driven by profit motives, responds to climate change and its asso- ciated risks and uncertainties. To investigate the link between uncertainty and investment, we draw on Real Options Theory (ROT) (Myers 1977), which suggests that firms facing increased uncertainty are more likely to delay investment, reducing their overall investment levels. This effect is especially pro- nounced in sectors characterized by a higher degree of irreversible investments, such as infrastructure projects (Baldwin 1982). Building on ROT, this study aims to evaluate whether increased temperature uncertainty is associ- ated with slower firm growth across diverse geographical regions. Importantly, we investigate whether this adverse effect is more pronounced in sectors characterized by a higher proportion of irreversible as- sets. Additionally, recognizing the potential regional heterogeneity in temperature uncertainty across different income levels (Bathiany et al. 2018), we explore whether the negative relationship between temperature uncertainty and firms’ fixed asset growth varies according to country income levels. 1 The most common approach to measuring the economic impact of climate change is to analyze the country- or region-level differences in its potentially adverse effects on GDP and income (Dell et al. 2009, Roson et al. 2016, Cruz & Rossi-Hansberg 2021), growth (Dell et al. 2012), productivity (Burke et al. 2015, Kalkuhl & Wenz 2020), and investment rates (Willner et al. 2021). Sector-specific impacts, especially on agriculture, have also been widely documented (Lobell & Asner 2003, Schlenker & Roberts 2009, Hatfield & Prueger 2015, Albert et al. 2021). For a meta-analysis of the existing economic activities and temperature relationship, see Newell et al. (2021), which discusses substantial uncertainty in the estimates of the impact and conflicting results. 2 The overall neglect of the impact of increased temperature variation is also observed in climate change research in ecology, as noted by Vasseur et al. (2014). 2 To do this, we employ a unique dataset that merges firm-level financial data with meteorological information. Our analysis hinges on the concept of location-specific weather uncertainty for firms, measured via grid-level temperature uncertainty. We utilize financial records from the Orbis database, covering 30 million firms across 64 countries, alongside daily surface temperature data on a 1x1 grid from the Berkeley Earth Foundation dataset (BERKEARTH). The merging of these two datasets cre- ates an extensive grid-level dataset, covering around 4,000 economically meaningful grids with firm presence from 2000 to 2019. Grid-specific temperature uncertainty is measured based on the maxi- mum daily temperatures over a year. We examine whether this grid-specific temperature uncertainty is associated with the growth of firms located in the grid. Our analysis yields three key findings. First, there is a strong negative correlation between weather uncertainty and firm growth. Our results indicate that for every one-degree increase in temperature un- certainty, measured by the standard deviation of deviations from expected temperatures, firms’ growth rate of fixed assets decreases by 0.77 percentage points. Second, the impact of weather uncertainty on firm growth varies across industries with different de- grees of investment irreversibility. Industries with higher investment irreversibility experience a more pronounced negative impact on firm size growth. Doubling capital intensity intensifies the negative effect of temperature uncertainty on firm size growth by roughly 0.72 percentage points. Third, the impact of weather uncertainty on firm growth varies across countries with different in- come levels. Higher-income countries are more adversely affected by weather uncertainties in terms of firm behavior. As income levels double, the negative impact of temperature uncertainty on fixed asset growth becomes more severe by 0.63 percentage points. This paper is among the first to examine firms’ behavior in response to location-specific temper- ature uncertainty at the grid level. While climate-related uncertainty is often studied using indirect, non-weather measures like the Climate Change News Index (Engle et al. 2020) and in climate policy uncertainty literature (Gulen & Ion 2016), recent studies have increasingly focused on direct measures of weather variability, as highlighted by Kahn et al. (2021), Kotz et al. (2021), Donadelli et al. (2022), Sheng et al. (2022). Our approach builds on these recent efforts by emphasizing weather variability as a key factor influencing economic activity. In doing so, we also draw on the established tradition of using gridded output data, particularly Nordhaus (2006) and later studies employing grid-level or sub-national geographic data, such as Kalkuhl & Wenz (2020), Wenz et al. (2023), Gortan et al. (2024). Our main contribution is the construction of grid-level economic data based on firm-level activities. Unlike regional GDP data, firm-level data shows how the private sector, driven by profit motives, re- sponds to climate change and the risks it poses. This firm-level data enables us not only to capture the overall relationship between weather variability and economic activity but also to test important hy- potheses, such as the moderating effects of investment irreversibility in this relationship—something that is difficult to achieve without detailed firm-level data. By doing so, we bridge the climate change impact literature, particularly studies using grid-level output data (Nordhaus 2006, Kalkuhl & Wenz 2020), with research on firms’ behavior under uncertainty (Myers 1977). The paper is structured as follows: Section 2 reviews the real options theory. Section 3 discusses the main data sources, cleaning process, and key descriptive statistics. Section 4 outlines the empirical methodology and presents our findings. Section 5 concludes the paper. 3 2 Uncertainty and investment 2.1 Real options theory The primary research question we explore is how firms react to uncertainty associated with climate change by adjusting their investment behavior. We adopt a broader perspective by treating temperature uncertainty as a contributor to the overall uncertainty that firms face across various industries. This approach allows us to move beyond sector-specific impacts and focus on how unpredictable weather patterns generate broader uncertainty in firms’ strategic decision-making processes, affecting invest- ment choices, operational planning, and risk management strategies in both direct and indirect ways. In the firm-behavior theory, the relationship between uncertainty and a firm’s investment has been significantly shaped by the Real Options Theory (ROT), initially proposed by Myers (1977) and further developed by Abel (1983), Bernanke (1983), McDonald & Siegel (1986), Dixit et al. (1994), Abel & Eberly (1996). This theory redefines investment decision-making in the context of uncertainty. Unlike traditional accounting methods for investment decisions, such as the net present value rule—which states that firms should invest when the discounted present value of future cash flows exceeds the cost of the investment—ROT suggests that it might be optimal for firms to delay their investments until they have more information about their economic environment. The theory highlights that uncertainty is a critical factor in such decisions, suggesting that increased uncertainty raises the value of the option to wait thereby reducing the propensity to invest. The issue of the investment–uncertainty relationship is particularly important in the context of irreversible investments (Baldwin 1982). Such assets, including infrastructure like gas pipelines, cannot be easily reversed or undone without significant costs or losses. Given the high costs associated with undoing such investments, the impact of uncertainty can be substantially more severe. This observation suggests that the relationship between uncertainty and investment may be more pronounced in sectors or firms with a higher proportion of irreversible assets (Pindyck 1990, Dixit et al. 1994).3 In the broader context of the negative relationship between uncertainty and investment rates, we understand temperature uncertainty as a contributing factor of the overall uncertainty firms encounter in their decision-making. Sudden fluctuations in weather conditions can lead to unpredictable spikes in operational costs, particularly through volatile energy expenses. Additionally, weather-driven disrup- tions in market dynamics can trigger abrupt shifts in demand or create supply shortages, driving price instability and complicating firms’ ability to forecast future profitability. These unpredictable factors may increase the complexity of long-term strategic planning, indicating that temperature uncertainty could be an important component of the broader uncertainty influencing investment behavior. 2.2 Basic model To establish the mathematical relationship between uncertainty and investment rates, we draw on Mc- Donald & Siegel (1986), which derives the value of the investment option as a function of the project’s 3 The investment–uncertainty relationship has been extensively tested both at the aggregate (Ferderer 1993, Caballero & Pindyck 1992, Ghosal & Loungani 2000, Huizinga 1993) and the firm levels (Leahy & Whited 1995, Bell & Campa 1997, Bulan 2005, Bulan et al. 2009, Bloom 2014). As surveyed in Leahy & Whited (1995) and Carruth et al. (2000), the general consensus is that increased uncertainty, at both the aggregate and disaggregate levels, leads to lower investment rates. 4 volatility and cost. In this model, the firm aims to maximize the present value of the project by choosing the optimal time to invest. This is framed as a “real options” problem, where the firm holds an option to invest, but once exercised, the decision becomes irreversible. Unlike financial options, real options are not financial derivatives and therefore cannot be traded. The key behavioral assumption of the real options theory is that the firm’s objective is to maximize the present value of the project by choosing the optimal time to invest. More specifically, a firm deciding when to invest weighs the immediate benefits of investing against the potential value gained by post- poning the investment. Let V denote the present value of expected net cash flows from the project, and F represent the cost of investment. The firm’s decision rule can be formally expressed as: V V Invest if ≥ C ∗; Wait if < C ∗, F F where C ∗ represents the critical threshold of the benefit-to-cost ratio at which immediate investment becomes optimal. This threshold is determined by the elasticity ϵ, which measures how sensitive the option value is to changes in the ratio V /F . A higher elasticity indicates that fluctuations in the ratio have a greater influence on the investment decision. Let us assume that both the project value V and the cost F follow geometric Brownian motions. The stochastic differential equations governing their behavior are: dV = αv V dt + σv V dWv , (1) dF = αf F dt + σf F dWf , (2) where αv and αf are the expected growth rates of V and F , σv and σf represent their volatilities, and dWv and dWf are increments of Wiener processes. The two processes may be correlated, with correla- tion coefficient ρvf .4 2 + σ 2 − 2ρ σ σ , the key mathematical result from Denoting the combined volatility σ 2 as σ 2 = σv f vf v f (McDonald & Siegel 1986) is the derivation of the elasticity in terms of the volatilities: 1 2 −(αv − αf − 2 σ )+ (αv − αf − 1 2 2 2 2 σ ) + 2σ r ϵ= . (3) σ2 The critical threshold C ∗ is then determined using the elasticity ϵ, with the investment trigger point V = V ∗ = C ∗ F being derived from the value-matching and smooth-pasting conditions such that5 4 Introducing additional sources of volatilities, such as weather variability, is mathematically straightforward and does not alter the core implications of the model; it only adds complexity to the correlation structure. To maintain focus on the key results of real options theory, we rely on the original model by McDonald & Siegel (1986). 5 The value-matching condition, X (V ∗ ) = V ∗ − F , and the smooth-pasting condition, X (V ∗ ) = 1, are used to determine the V critical threshold for investment. Given the solution X (V ) = AV ϵ , the value-matching condition becomes A(V ∗ )ϵ = V ∗ − F , and the smooth-pasting condition yields Aϵ(V ∗ )ϵ−1 = 1. Solving these two equations simultaneously determines the constants and the expression of the critical threshold C ∗ in terms of the elasticity ϵ. 5 ϵ C∗ = . (4) ϵ−1 This result shows that C ∗ is an increasing function of ϵ, which itself depends on the combined volatility σ 2 and the growth rates αv and αf . The option value X (V , F ) can be explicitly written as (McDonald & Siegel 1986): V ϵ X (V , F ) = (C ∗ − 1)F . (5) C ∗F This expression demonstrates the dependence of the option value X on the project’s value V , the investment cost F , the critical threshold C ∗ , and the elasticity ϵ. The relationship between volatility and the critical threshold C ∗ can be further examined by differentiating C ∗ with respect to σ : dC ∗ > 0, (6) dσ dϵ since dσ > 0 and ϵ > 1. Therefore, both the value of the investment option X and the critical threshold C∗ are increasing functions of the variance σ 2 : X ∝ σ 2ϵ , C ∗ ∝ σ 2. This proportionality demonstrates that as uncertainty (volatility) increases, the value of the option to delay investment rises, similar to financial options where greater volatility expands the range of possible outcomes. This flexibility allows firms to respond more effectively to future changes, such as shifts in prices or market trends. Firms can capitalize on upside opportunities by exercising the option to invest or expand, while avoiding losses by deferring in unfavorable conditions. This asymmetry, which favors the investor, means that higher uncertainty or a broader range of potential outcomes benefits the firm by increasing the value of waiting. The following sections outline the data sources and empirical strategies used to test the real options hypotheses, focusing on temperature uncertainty at the grid level to explore the potential negative relationship between temperature uncertainty and firms’ investment decisions. 3 Data 3.1 Data source Firm-level data. We employ Orbis as our principal firm-level data source, which includes compre- hensive financial statements for organizations on a global scale. Administered by Bureau van Dijk, the Orbis database integrates information from a variety of origins, including official company registries and regulatory filings. Its extensive coverage, particularly concerning balance sheet items and profit 6 and loss accounts for private enterprises, positions Orbis as one of the most robust datasets available for analyzing financial performance at the firm level.67 The Orbis database provided by the Wharton Research Data Services (WRDS) includes latitude and longitude information at the firm level, enabling the precise determination of each firm’s geographic location. This aspect is essential for correlating firms with their respective regional weather data. How- ever, not all firms in our dataset have this geographical data. Almost 20% of firms in our sample do not have the geographical information. To address this, we employed the Google Maps API, using available country, region, and city data for these firms. The methodology for this process is detailed in Appendix A. For our study, we extracted firm-level information from 2000 to 2019, focusing on medium, large, and very large companies. The primary variables extracted included the firm’s location, fixed assets, and the NACE code for sector identification. While other financial variables were sparsely populated, fixed asset data was more consistently available. Consequently, we used fixed asset growth as a proxy for firm size growth. Grid-level weather data. In our study, we utilize the temperature gridded data for global domains, which is derived from the Berkeley Earth Foundation dataset (BERKEARTH).8 This dataset is sourced from the Climate Data Store, as part of the Copernicus Climate Change Service, managed by the Euro- pean Centre for Medium-Range Weather Forecasts on behalf of the European Union. The BERKEARTH dataset has comprehensive coverage of global temperature changes, encompassing both land and ocean regions and providing historical temperature data spanning over two centuries. The methodology em- ployed in the processing of this dataset is detailed in Rohde & Hausfather (2020). The BERKEARTH dataset offers data in three distinct grid formats: 1 × 1, 2 × 2, and 3 × 3 degrees. Each ‘1 degree’ in this context refers to a geographical measure, where 1◦ of latitude corresponds ap- proximately to 111 kilometers (69 miles) on the Earth’s surface. Similarly, 1◦ of longitude varies in distance, being wider near the equator and converging towards the poles due to the Earth’s spherical shape. For our analysis, we specifically employ the 1 × 1-degree grid format, which provides the most granular weather information. The dataset encompasses three primary variables: mean, maximum, and minimum daily temperatures. In our research, we focus on utilizing the maximum temperature readings. Income-level data. For the country-level income data, we use the World Bank’s income data.9 This data is used to investigate whether there are distinct dynamics in the impact of temperature uncertainty on firm behavior across different income levels.10 6 However, Orbis does not provide comprehensive coverage for firms in North America, so the two major countries, the US and Canada, are not fully represented in our sample. 7 See Lepore & Fernando (2023) for an analysis of physical risks from climate changes at the firm level using Orbis data. 8 Similar studies using the grid-level weather data include Nordhaus (2006) and those studies using data from the Climate Research Unit of the University of East Anglia (Harris et al. 2014), such as Auffhammer et al. (2013), Kalkuhl & Wenz (2020). 9 https://blogs.worldbank.org/opendata/new-world-bank-country-classifications-income-level-2022-2023 10 Some economies, such as Malta, Réunion, Taiwan, and Hong Kong, do not have their income information recorded in the World Bank income data. As a result, they are excluded from the regression analysis that involves income level. However, they are included in the regression analysis when income level is not a factor. 7 Data merging. A key aspect of our research methodology is the merging of the Orbis and BERKEARTH datasets using latitude and longitude information. This common element allows us to associate firm-level financial performance data from Orbis with the corresponding weather data from BERKEARTH, both mapped to the same geographic region. Given that the weather data is gridded with one degree of latitude and longitude, we first assign a unique index to each grid. For example, Index 1 is assigned to the upper leftmost grid (longitude −180 to −179 and latitude −90 to −89). Subsequently, we assign the same index to the location of each firm. For instance, a firm located at longitude −179.3 and latitude −89.3 would also be assigned Index 1. These indices are then used to merge the two datasets. As a result, the final firm-level data includes annual observations of the firm, its location, sector, and fixed assets, as well as the mean, variance and maximum of the temperature for the gridded location of the firm. 3.2 Data pre-processing We carry out several cleaning procedures to enhance the quality and consistency of the data. First, we exclude firms with non-positive assets from our dataset. This exclusion is based on the accounting principle that fixed assets should always be non-negative. While it is technically possible to have zero fixed assets, these cases are excluded to enable the calculation of the growth rate of fixed assets. Second, to avoid potential double counting, our analysis is limited to unconsolidated data. We select U1 and U2 as the consolidation codes in Orbis data.11 This choice ensures that the data reflects each firm’s inde- pendent operations, preventing distortion from inter-company transactions or accounts. This is crucial for examining the impact of location-specific factors, such as weather, on firms’ behavior, which would be limited with consolidated data. Finally, We also focus on countries with more than 500 observations and more than single observation for country-year-grid group. All nominal variables (fixed & total assets) are deflated using the GDP deflator at the country level, sourced from World Bank data.12 3.3 Construction of variables Firm-level variable From Orbis data, we construct the variable that indicates the growth rate of the firm’s size. We use the fixed assets as the main size variable. For each firm i in country c and year t , we define the size growth as the log return on the fixed asset Ki,t : k Ki,t κi,t /pc,t SizeGrowthi,t = ln = ln k (7) Ki,t −1 κi,t −1 /pc,t −1 k where κ is the nominal fixed capital and pc,t is the deflator of country c at time t . Furthermore, to construct the proxy for investment irreversibility, we use the ratio of tangible fixed assets to total assets, following Gulen & Ion (2016), who argues that companies with significant tangible fixed capital investment often undertake projects requiring substantial initial expenditures, and thus 11 U1 represents the unconsolidated record without the accompanying consolidated record, while U2 represents the uncon- solidated record with the accompanying consolidated record. 12 https://data.worldbank.org/indicator/NY.GDP.DEFL.KD.ZG 8 their investments tend to be less irreversible. We call this variable the capital intensity (CapIntensity). Tangible Fixed Asseti,t CapIntensityi,t = (8) Total Asseti,t Weather variables In our study, we calculate the weather variables for each year and grid using daily temperatures.13 Denoting the daily temperature for each grid by Ti , the mean and maximum 1 N are N i =1 Ti and max(Ti ), respectively, for each grid i , with N = 365. To construct temperature uncertainty, we aggregate all temperature observations for each unique index (spatial grid) across multiple years and fit a smoothing spline14 to capture the overall seasonal temperature pattern for each index. This spline represents the assumed constant expectation of firms regarding temperature levels and seasonality over time. For each year t within a given index, we cal- culate the residuals Rdt by comparing the actual daily temperature Tdt to the predicted values from the common spline T ˆt for day d in year t : ˆt . Rdt = Tdt − T Figure 1 presents an example spline estimate for grid index 52259 (longitude: -1.5, latitude: 63.5) covering the period from January 1, 2000, to December 31, 2019. The smoothing spline is estimated over the full 365 days, using 20 years of daily observations to predict the temperature for each day. Temperature uncertainty is defined as the standard deviation of these residuals, calculated annually to quantify the variability in temperature: N 1 Uncertaintyt = R2 dt . (9) N d =1 This standard deviation represents the degree to which the actual temperature deviates from firms’ constant expectations, capturing yearly variations in weather conditions. This assumption of constant expectations is reasonable for several reasons as firms typically base their planning and operational strategies on long-term climate averages and historical weather patterns, which tend to remain rela- tively stable over short time horizons. 3.4 Grid-level aggregation In our study, the primary unit of observation is the grid, aligning with how weather data is measured and recorded. For the firm-level data, we aggregate this data at the grid level. We implement two distinct levels of aggregation in our analysis: 13 For the daily temperature measurement, three variables are available from the BERKEARTH dataset at the Climate Data Store: minimum, mean, and maximum daily temperatures. We do not consider the minimum temperature in this study. Between the mean and maximum daily temperatures, there is a difference in data availability. The mean temperature data is accessible only up to June 2018, whereas the maximum temperature data extends through 2019. Due to this disparity in data coverage, we have chosen the maximum temperature variable as our primary variable for analysis. Both the mean and maximum daily temperature variables are highly correlated. 14 The smoothing spline is fitted using the smooth.spline() function from the base R package. The default method for es- timating the tuning parameter (λ) uses generalized cross-validation (GCV), which balances the trade-off between smoothness and fidelity to the data. 9 Common index−wide spline 52259 20 10 Temperature 0 −10 0 100 200 300 Days since start of year Figure 1: An example of a common index-wide spline estimate This figure presents a common index- wide spline estimate for grid index 52259 (longitude: -1.5, latitude: 63.5) over the period from January 1, 2000, to December 31, 2019. The blue points represent the original daily temperature data, and the red line shows the spline fit, which is estimated across all 365 days using 20 years of observations. The x-axis shows the days since the start of the year, while the y-axis represents the predicted temperature. 1. Country, Grid, Year Level Aggregation: At this level, the unit of observation is a unique combina- tion of country, grid, and year. For example, one such unit could be Spain, a specific grid index, and the year 2015. It allows for an examination of how weather patterns in a specific country, grid and year affect the firms located there. It is important to note that some grids are shared by multiple countries. In our dataset, roughly 5% of the total grids span across multiple countries. Despite this overlap, we retain the country-level grouping because the behavior of firms can vary significantly based on the country, even if they experience similar weather patterns. 2. Industry, Grid, Year Level Aggregation: The second level of aggregation employs the industry classification, using the NACE industry code (Statistical Classification of Economic Activities in the European Community). We use the highest level of aggregation (level 1), which includes 21 sectors. Unlike the country-grid level aggregation, where grids are nested within the country, the industry-grid level aggregation leads to a unique combination of industries and grids, resulting in a higher number of observations (316,324). For example, in the geographical grid ‘16191,’ we observe multiple different industries each year, such as agriculture, manufacturing, etc. This level of aggregation allows us to examine industry-level heterogeneity, especially investment irre- versibility (Baldwin 1982), and its influence on the relationship between temperature uncertainty and firm size growth. 10 For each group defined by these aggregation methods, we use the mean value of the variables. 3.5 Descriptive statistics The final dataset contains 23,437,477 firm-level observations from 64 countries. In total, there are 3,882 economically meaningful grids where firms from our Orbis dataset are present. Table 1 presents the summary statistics for the key weather and economic variables based on grid-level observations. The mean temperature uncertainty across all grids during the sample period from 2001 to 2019 is 2.09◦ C. The average temperature (MeanT) and maximum temperature (MaxT) are 16.20◦ C and 30.54◦ C, respectively. The mean fixed asset growth (SizeGrowth) is -1.45%. Further detailed descriptions can be found in the appendices, where we provide comprehensive country-level summary statistics. Table 7 in Appendix B presents these country-level statistics using country-specific firm-level observations. Table 8 in Appendix B shows the number of grids for each country for each year. Mean SD Min Q25 Q50 Q75 Max # of firms 302.89 1253.10 2.00 5.25 22.58 136.08 26070.63 Uncertainty 2.09 0.52 0.71 1.72 2.06 2.43 4.55 MeanT 16.20 9.94 -9.87 8.05 15.22 24.02 35.46 MaxT 30.54 5.97 11.96 26.42 30.51 34.73 49.27 SizeGrowth -1.45 23.20 -733.03 -10.47 -0.10 7.23 277.01 Capintensity 0.36 0.14 0.00 0.27 0.34 0.43 0.97 Income 13532.81 14080.59 1061.59 5016.33 7471.71 11146.79 66870.01 Table 1: Summary statistics. This table presents key aggregate statistics at the grid level, including the number of firms per grid, temperature uncertainty (Uncertainty) average temperature (MeanT), maxi- mum temperature (MaxT), income, growth rate of fixed assets (SizeGrowth), and the ratio of tangible fixed assets to total assets (CapIntensity). Observations are made at the grid level using the mean values of these variables over time, resulting in 3,882 observations. For each variable, key summary statistics (mean, standard deviation, minimum, 25th percentile, median, 75th percentile, and maximum) are presented. Time series of weather variables. Figure 2 presents time series plots for three key weather variables (temperature uncertainty, mean and maximum) across all the grids considered in our study. The black line in each plot represents the annual mean value of the weather variable, accompanied by a grey shaded area representing the 95% interval for each year. Since the 95% interval is wide and the overall trend in the variables is unclear, an inset plot without the 95% interval is included. Consistent with the existing evidence (IPCC 2023), our data shows an observable increasing trend in both the mean and maximum temperature values. However, the temperature uncertainty does not display a clear global pattern, aligning with findings from other studies on temperature variability (the standard deviation of temperature measures) (Huntingford et al. 2013, Olonscheck et al. 2021).15 15 Figures 5, 6, and 7 in Appendix B show the country-specific time trends of these weather variables. 11 30.5 2.4 30 15.00 30.0 2.2 14.75 35 14.50 29.5 3 Uncertainty 2.0 14.25 29.0 20 Mean Max 30 2 10 25 20 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 Figure 2: Time series of temperature variables. Each plot illustrates the annual trajectory of tem- perature variables across all grids, with separate plots for the temperature uncertainty (left) , mean (middle), and max (right). The black line in each plot denotes the annual mean value of the respective temperature variable, accompanied by a grey shaded area that represents the 95% interval. Addition- ally, an inset plot highlights the mean values without the 95% interval, emphasizing the overall trend in these mean values over the years. It is worthwhile to note that the trends and patterns identified in Figure 2 do not necessarily repre- sent regional or global averages of the weather variables. The analysis is specifically based on the grids where firms from the Orbis dataset are located. Therefore, the findings are reflective of the weather variable characteristics in these particular economic areas, rather than providing a broader, general- ized view of regional or global weather trends. Scatterplots of temperature uncertainty and asset growth. To demonstrate the overall relationship between temperature uncertainty and fixed asset growth, we present a scatterplot of these variables in Figure 3. The plot includes all country-grid-year observations; each dot represents the mean value of fixed asset growth and the temperature uncertainty for a specific country, grid, and year. The result shows that there is a strong negative relationship with a slope of -4.77, as shown by the grey linear line. Figure 8 in Appendix B displays the cross-sectional (grid-level) relationship for each year, confirm- ing a negative relationship. 4 Empirical method and results In this section, we explore the potential negative correlation between temperature uncertainty and firms’ annual fixed asset growth, structured into three parts. First, we assess the overall relationship at the grid level. Next, we examine industry-level differences by analyzing how the impact of temperature uncertainty varies with sectoral investment irreversibility, measured by the ratio of tangible fixed assets to total assets. Finally, we investigate how income levels might influence this relationship. Table 2 presents the correlation matrix for all variables in our regression models. Temperature uncertainty is negatively correlated with firm size growth, with a coefficient of -0.09, indicating that higher uncertainty is associated with lower fixed asset growth. Temperature uncertainty also shows a negative correlation with both mean temperature (-0.59) and maximum temperature (-0.28), reflecting regional patterns where warmer areas tend to exhibit less temperature uncertainty. As expected, the 12 Slope = −4.77 200 100 Fixed asset growth (%) 0 −100 −200 1 2 3 4 5 Temperature uncertainty Figure 3: Scatter plot of temperature uncertainty and fixed asset growth. This scatter plot illustrates the relationship between the temperature uncertainty and fixed asset growth. Each dot represents a grid-level-year observation. For visualization purposes, only observations with size growth between -200% and 200% are included. mean and maximum temperatures are strongly positively correlated (0.82), suggesting that regions with higher average temperatures also experience higher maximum temperatures. The income variable shows negative correlations with the weather-related variables, implying that higher-income countries tend to have lower average temperatures, lower temperature uncertainty, and lower maximum temperatures. Interestingly, capital intensity is positively correlated with temperature uncertainty (0.13), suggest- ing that more capital-intensive industries may be located in regions with greater temperature uncer- tainty. However, the slightly negative correlation between capital intensity and firm size growth (-0.08) suggests that higher capital intensity may slightly dampen growth in the presence of temperature un- certainty. Table 2: Correlation Matrix SizeGrowth Uncertainty Mean MaxT LogCapintensity SizeGrowth Uncertainty -0.09*** Mean 0.11*** -0.59*** MaxT 0.07*** -0.28*** 0.82*** LogCapintensity -0.08*** 0.13*** -0.13*** -0.06*** LogIncome 0.02** -0.03*** -0.19*** -0.32*** -0.10*** 13 4.1 Regression method and specification Baseline model The baseline regression model is structured as follows: SizeGrowth[i,c,t ] = β0 + β1Uncertainty[i,c,t ] + β2MeanT[i,c,t ] + β3MaxT[i,c,t ] + δ[c] + γ[t ] + ϵ[i,c,t ] , (10) where the subscript i, c, t denotes the grid (i), country (c), and year (t) groupings in our dataset. The variable SizeGrowth represents the growth in firm size (Eq. 7), while Uncertainty refers to the temperature uncertainty (Eq 9). The terms δ and γ represent fixed effects at the country and year levels, respectively, and ϵ is an error term. It is important to note that this regression model is relatively unconstrained. As outlined in Sec- tion 3, this approach stems from our objective to encompass a broad range of firms, we have focused primarily on location, size growth, and sector variables extracted from firm-level data.16 Industry heterogeneity: investment irreversibility across industries In this part of our analysis, we employ a second-level aggregation approach that includes the industry group, as outlined in Section 3. It tests the main assumption of real option theory, which posits that firms with more irreversible assets (e.g., high capital intensity) might experience a more pronounced negative impact of uncertainty on their investment plans (Baldwin 1982). The regression model used is as follows: SizeGrowth[i,t,j ] = β0 + β1 Uncertainty[i,t,j ] + β2 MaxT[i,c,t ] + β3 LogCapIntensity[i,t,j ] β4 Uncertainty[i,t,j ] × LogCapIntensity[i,t,j ] + κ[j ] + γ[t ] + ϵ[i,t,j ] , (11) where LogCapIntensity denotes the log capital intensity (Eq. 8) and κ[j ] is the industry fixed effect. The coefficient of the interaction term between capital intensity and temperature uncertainty enables us to assess how the effects of temperature uncertainty on size growth vary across different levels of capital intensity. Income-level heterogeneity Finally, we examine the income level heterogeneity and its impact on the relationship between temperature uncertainty and firm size. To explore this, we introduce an inter- action between the income level indicator (log-transformed), denoted as LogIncome, and temperature uncertainty variables in our regression model. The model is formalized as follows: SizeGrowth[i,c,t ] = β0 + β1 Uncertainty[i,c,t ] + β2 MaxT[i,c,t ] + +β3 LogIncome[i,c,t ] + β4 Uncertainty[i,c,t ] × LogIncome[i,c,t ] + δ[c] + γ[t ] + ϵ[i,c,t ] , (12) The inclusion of the interaction term between income level and temperature uncertainty variables allows us to examine the heterogenous impacts of temperature uncertainty on size growth across dif- ferent income levels. 16 As will be discussed in the discussion section (Section 5), integrating more firm-specific variables such as cash flow rate, capital intensity, and firm age is left to future research. 14 4.2 Estimation results Table 3: Weather uncertainty and fixed asset growth Dep. variable: Fixed asset growth (1) (2) (3) (4) (5) (6) Uncertainty −5.048∗∗∗ −0.831∗∗ −1.181∗∗∗ −0.768∗∗ −6.517∗∗∗ 7.348∗∗ (0.268) (0.357) (0.370) (0.360) (0.342) (3.337) Mean −0.154∗∗∗ (0.043) Maximum −0.064 0.236∗∗∗ −0.059 (0.048) (0.021) (0.048) Capital Intensity −2.029∗∗∗ (0.393) Income −4.300∗∗∗ (1.116) Uncertainty:Capital Intensity −1.038∗∗∗ (0.179) Uncertainty:Income −0.903∗∗ (0.366) Constant 10.012∗∗∗ 14.337∗∗∗ 19.019∗∗∗ 16.799∗∗∗ 10.741∗∗∗ 42.887∗∗∗ (0.604) (2.210) (2.559) (2.873) (1.224) (9.652) Year FE No Yes Yes Yes Yes Yes Country FE No Yes Yes Yes Yes No Industry FE No No No No No Yes Observations 50,249 50,249 50,249 50,249 200,991 49,778 R2 0.007 0.138 0.138 0.138 0.063 0.140 Adjusted R2 0.007 0.136 0.137 0.136 0.063 0.138 Notes: This table presents the regression of fixed asset growth on weather uncertainty, measured by the standard deviation of deviations from expected temperatures. Models (1) through (4) progressively expand the baseline specification (Eq 10) by incorporating fixed effects (Model 2) and adding controls for mean and maximum temperatures (Models 3 and 4). Models 5 and 6 report the results for industry heterogeneity by capital intensity (Eq 11) and income-level heterogeneity (Eq 12), respectively. Standard errors are shown in parentheses. Significance level: ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01. For the baseline regression results, we begin with the first four columns in Table 3, which summa- rizes the results from different specifications of the main model described in Equation 10. We consider four models, starting with the simplest specification that includes only temperature uncertainty as a predictor (Model 1), and gradually expanding it to test the robustness of the estimated coefficient as additional variables are included. Model 2 adds country and year fixed effects, while Models 3 and 4 incorporate maximum and mean temperatures, respectively. The results consistently show a neg- ative relationship between location-specific temperature uncertainty and firm size growth across all 15 models.17 Using Model 4 as the baseline, the coefficient of -0.768 suggests that for each one-unit in- crease in temperature uncertainty, measured by the standard deviation of deviations from expected temperatures, firms’ growth rates decrease by 0.768 percentage points. These findings align with re- cent literature on the detrimental effects of temperature uncertainty on economic activities (Kahn et al. 2021, Kotz et al. 2021, Donadelli et al. 2022, Sheng et al. 2022). The regression results on the moderating effects of capital intensity in the impact of temperature uncertainty on firms’ asset growth (Eq. 11) are shown in the fifth column (Model 5) of Table 3. The negative interaction coefficient between temperature uncertainty and capital intensity indicates that the adverse impact of temperature uncertainty on firm growth intensifies as capital intensity increases, suggesting that capital-intensive firms are more susceptible to environmental fluctuations. Specifically, in industries where log capital intensity is higher by one unit—which corresponds to a 171.8% increase in actual capital intensity (since e1 = 2.718)—the negative effect of temperature uncertainty on firm size growth is amplified by 1.038 percentage points. To illustrate this effect more intuitively, a 100% increase in capital intensity, which corresponds to a rise of approximately 0.693 (ln(2)) in log capital intensity, amplifies the negative impact of temperature uncertainty on firm size growth by about -0.718 percentage points (−1.038 × 0.693). The regression results for income-level heterogeneity in the effect of temperature uncertainty on firms’ asset growth (Eq. 11) are presented in the sixth column (Model 5) of Table 3. The interaction coefficient between Uncertainty and LogIncome is consistently negative, indicating that the adverse ef- fects of temperature uncertainty are more pronounced in higher-income countries. This finding aligns with Donadelli et al. (2022), who also observed stronger negative effects of temperature uncertainty in high-income countries. Our results show that the coefficient for the interaction term is −0.903. Since income is expressed in logarithmic terms, a 100% increase in income (doubling income) results in an increase of ln(2) ≈ 0.693 in the log income. Therefore, a 100% increase in income amplifies the nega- tive effect of temperature uncertainty on firm size growth by approximately −0.625 percentage points (−0.903 × 0.693). This indicates that as income levels double, the detrimental impact of temperature uncertainty on fixed asset growth becomes more severe by 0.625 percentage points. In summary, our analysis shows a strong negative correlation between temperature uncertainty and fixed capital growth. The severity of this impact varies significantly across industries with differing levels of investment irreversibility and across countries with different income levels. Firms in industries with higher investment irreversibility and those in high-income countries experience more pronounced negative effects on fixed asset growth due to temperature uncertainty. 4.3 Robustness checks To ensure the robustness of our regression results, we performed additional checks by incorporating 1) a lagged dependent variable and 2) grid-level fixed effects into the main regression models. 17 Although we observe a similar negative relationship when both maximum and mean temperatures are included in the model, we do not report this result due to potential multicollinearity caused by the high correlation between the two variables (as shown in Table 2). 16 Lagged dependent variable Including the lagged dependent variable controls for any underlying mo- mentum or autocorrelation in firm growth. The results, summarized in Table 4, are generally consistent with the original regression. For instance, in Model 4, the coefficient for temperature uncertainty in the baseline regression (Eq 10) decreases slightly from -0.768 to -1.078, indicating that the negative effect of temperature uncertainty on firm growth becomes more pronounced when accounting for the auto- correlation in fixed asset growth. The results for industry heterogeneity by capital intensity and income levels remain largely unchanged, with the coefficients moving from -1.038 to -0.947 for industry differ- ences and from -0.903 to -0.923 for income-level variations. Table 4: Weather uncertainty and fixed asset growth: Lagged dependent variable Dep. variable: Fixed asset growth (1) (2) (3) (4) (5) (6) Lagged Growth 0.049∗∗∗ −0.014∗∗∗ −0.014∗∗∗ −0.014∗∗∗ 0.024∗∗∗ −0.014∗∗∗ (0.004) (0.004) (0.004) (0.004) (0.002) (0.004) Uncertainty −4.348∗∗∗ −0.700∗∗ −0.630∗ −1.078∗∗∗ −5.568∗∗∗ 7.663∗∗ (0.269) (0.356) (0.359) (0.369) (0.351) (3.405) Maximum −0.070 0.203∗∗∗ −0.058 (0.048) (0.022) (0.048) Mean −0.169∗∗∗ (0.043) Capital Intensity −2.298∗∗∗ (0.405) Income −5.174∗∗∗ (1.168) Uncertainty:Capital Intensity −0.947∗∗∗ (0.184) Uncertainty:Income −0.923∗∗ (0.373) Constant 8.372∗∗∗ 19.164∗∗∗ 21.834∗∗∗ 24.536∗∗∗ 14.348∗∗∗ 53.936∗∗∗ (0.605) (2.344) (2.965) (2.708) (1.283) (10.227) Year FE No Yes Yes Yes Yes Country FE No Yes Yes Yes No Industry FE No No No No Yes Observations 46,922 46,922 46,922 46,922 183,756 46,507 R2 0.009 0.140 0.140 0.140 0.060 0.142 Adjusted R2 0.009 0.138 0.138 0.139 0.060 0.140 Notes: This table presents the regression of fixed asset growth on weather uncertainty, measured by the standard deviation of deviations from expected temperatures, with the inclusion of a lagged dependent variable. Models (1) through (4) progres- sively expand the baseline specification (Eq 10) by incorporating fixed effects (Model 2) and adding controls for mean and maximum temperatures (Models 3 and 4). Models 5 and 6 report the results for industry heterogeneity by capital intensity (Eq 11) and income-level heterogeneity (Eq 12), respectively. Standard errors are shown in parentheses. Significance level: ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01. 17 Grid-level fixed effects Adding grid-level fixed effects controls for unobserved regional characteris- tics that might influence both temperature uncertainty and firm growth, as both are measured at the grid level. This approach helps account for spatial heterogeneity, such as differences in climate, in- frastructure, or regional economic conditions, which could otherwise bias the results. The results are summarized in Table 5. The results indicate that incorporating grid fixed effects generally amplifies the impact of tem- perature uncertainty. For instance, the coefficient for temperature uncertainty in Model 1 decreases from -0.768 to -2.575. Similarly, the coefficients for the interaction terms between Uncertainty and CapIntensity and between Uncertainty and LogIncome change from -1.038 to -1.352 and from -0.903 to -1.569, respectively. Table 5: Weather uncertainty and fixed asset growth with grid fixed effects Dep. variable: Fixed asset growth (1) (2) (3) Uncertainty −2.575∗∗∗ −3.851∗∗∗ 11.360∗∗∗ (0.517) (0.464) (3.907) Maximum −0.022 −0.294∗∗∗ −0.003 (0.114) (0.079) (0.114) Capital Intensity −1.815∗∗∗ (0.412) Income −2.840∗∗ (1.258) Uncertainty:Capital Intensity −1.352∗∗∗ (0.187) Uncertainty:Income −1.569∗∗∗ (0.431) Constant 23.947 22.811 40.973 (24.165) (15.595) (26.101) Year FE Yes Yes Yes Country FE Yes Yes No Industry FE No No Yes Grid FE Yes Yes Yes Observations 50,249 200,991 49,778 R2 0.234 0.111 0.236 Adjusted R2 0.169 0.095 0.172 Notes: This table regresses fixed asset growth on weather uncertainty variables, measured by the standard deviation temper- ature. Model (1) includes the weather uncertainty, Model (2) includes the weather uncertainty interacting with the income level, and Model (3) includes the weather uncertainty interacting with the capital intensity. All models incorporate the grid- level fixed effects. Standard errors are shown in parentheses. Significance level: ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01. 18 5 Discussion and conclusion In this research, we present a unique dataset merging firm-level financial statements with weather data to explore how climate change uncertainty affects firm investment behavior. Using financial records from 64 countries from the Orbis database and daily surface temperature data on a 1x1 grid from BEKREARTH, we created a comprehensive grid-level dataset. This dataset integrates firms’ fixed asset growth with temperature uncertainty. A key objective of this study is to provide grid-level evidence on the impact of temperature un- certainty on firms’ fixed asset growth. Our findings showed a strong negative association between temperature uncertainty and firm growth. Several factors could explain this significant negative relationship. First, temperature uncertainty introduces additional complexity into firms’ operational and strategic planning, making it harder to forecast costs and returns on investment accurately. Firms might reduce or delay capital investments as they become uncertain about the economic viability of long-term projects, especially when faced with unpredictable climate conditions. Moreover, temperature uncertainty may also raise firms’ operating costs, particularly in sectors dependent on stable climatic conditions, such as manufacturing, energy, or logistics. For instance, un- predictable fluctuations in temperature could affect energy consumption, maintenance costs, or supply chain efficiency, all of which contribute to a more cautious approach to capital expenditures. Firms may allocate fewer resources to expanding fixed assets if they expect higher variability in operational expenses due to environmental uncertainties. We also documented significant heterogeneity in the impact of temperature uncertainty on firm growth across industries. Industries with higher levels of irreversible assets, as measured by the ratio of tangible fixed assets to total assets, tend to experience a more pronounced negative effect from temper- ature uncertainty. This result aligns with the pediction by ROT (Myers 1977, Baldwin 1982), which sug- gests that firms with more irreversible investments are more likely to delay or reduce their investment activities in response to uncertainty. To be more specific, firms in those sectors with high proportion of irreversible assets typically face high sunk costs, making it difficult to adjust or reverse investments once they are committed. As a result, the temperature uncertainty increases the risk of making sub- optimal investment decisions, leading firms to adopt a more cautious stance. Second, capital-intensive industries often require longer planning horizons for investment projects, which can exacerbate the im- pact of uncertainty. The unpredictability of future environmental conditions makes it harder for firms to commit to long-term investments when they are unsure how future conditions might evolve. Finally, our study revealed that the impact of temperature uncertainty on firm growth varies signif- icantly across income levels. High-income countries are more adversely affected by weather uncertain- ties in terms of firms’ size growth. It is important to note that there is no general consensus on how income levels influence the impact of weather on economic activities, as different studies use varying metrics. For example, while Donadelli et al. (2022) focus on the growth rate of total factor productiv- ity and find the negative effects of temperature variability more pronounced in high-income countries, Kotz et al. (2021) use GDP per capita growth and find that lower-income countries are more negatively affected. 19 One possible explanation for the weaker link between temperature uncertainty and firm growth in low-income countries is the presence of more binding constraints, such as limited access to finance or inadequate infrastructure. In many lower-income regions, firms may already face significant barriers to growth that are unrelated to weather conditions. For example, financial constraints can severely limit a firm’s ability to invest, expand, or respond to new opportunities, making additional uncertainty from temperature fluctuations less impactful. Further, the stronger impact of temperature uncertainty on firm growth in high-income countries can be attributed to the greater complexity and interdependence of their economies. High-income countries tend to have more integrated and globalized supply chains, with firms relying on precise planning and forecasting for productivity, energy use, and logistics. Even relatively small disruptions, such as unexpected temperature fluctuations, can ripple through these systems and have far-reaching consequences. This heightened sensitivity to external shocks means that temperature uncertainty may pose a greater risk to firm growth in wealthier economies, where economic interdependencies and higher expectations for stability amplify the effects of weather-related unpredictability. However, our research has limitations that should guide future studies. Our regression model, while extensive, lacks some key financial variables often used in testing and operationalizing real op- tion theory (Leahy & Whited 1995, Bell & Campa 1997, Bulan 2005, Bulan et al. 2009, Bloom 2014). This limitation arose from our focus on achieving maximal coverage at the expense of excluding cer- tain financial variables, particularly to include lower-income countries and assess the country-level heterogeneous impact of temperature uncertainty. Future research should aim for a more constrained standard investment model, incorporating key financial variables like cash flow rate. Additionally, the time coverage of our data spans only 20 years, with many countries lacking full coverage. The limited time span and smaller numbers of observations in earlier years restrict the time series analysis, which is crucial for studying phenomena like climate change that unfold over long periods. Future research should use longer time horizons with firm-level or regional data to capture sufficient time series variations. 20 References Abel, A. B. (1983), ‘Optimal investment under uncertainty’, The American Economic Review 73(1), 228–233. Abel, A. B. & Eberly, J. C. (1996), ‘Optimal investment with costly reversibility’, The Review of Economic Studies 63(4), 581–593. Albert, C., Bustos, P. & Ponticelli, J. (2021), The effects of climate change on labor and capital reallocation, Tech- nical report, National Bureau of Economic Research. Alessandri, P. & Mumtaz, H. (2022), The macroeconomic cost of climate volatility, BCAM Working Papers 2202, Birkbeck Centre for Applied Macroeconomics. URL: https://ideas.repec.org/p/bbk/bbkcam/2202.html Auffhammer, M., Hsiang, S. M., Schlenker, W. & Sobel, A. (2013), ‘Using weather data and climate model output in economic analyses of climate change’, Review of Environmental Economics and Policy . Baldwin, C. Y. (1982), ‘Optimal sequential investment when capital is not readily reversible’, The Journal of Fi- nance 37(3), 763–782. Bathiany, S., Dakos, V., Scheffer, M. & Lenton, T. M. (2018), ‘Climate models predict increasing temperature variability in poor countries’, Science Advances 4(5), eaar5809. Bell, G. K. & Campa, J. M. (1997), ‘Irreversible investments and volatile markets: A study of the chemical pro- cessing industry’, Review of Economics and Statistics 79(1), 79–87. Bernanke, B. S. (1983), ‘Irreversibility, uncertainty, and cyclical investment’, The Quarterly Journal of Economics 98(1), 85–106. Bloom, N. (2014), ‘Fluctuations in uncertainty’, Journal of Economic Perspectives 28(2), 153–176. Bulan, L., Mayer, C. & Somerville, C. T. (2009), ‘Irreversible investment, real options, and competition: Evidence from real estate development’, Journal of Urban Economics 65(3), 237–251. Bulan, L. T. (2005), ‘Real options, irreversible investment and firm uncertainty: new evidence from US firms’, Review of Financial Economics 14(3-4), 255–279. Burke, M., Hsiang, S. M. & Miguel, E. (2015), ‘Global non-linear effect of temperature on economic production’, Nature 527(7577), 235–239. Caballero, R. J. & Pindyck, R. S. (1992), Uncertainty, investment, and industry evolution, Technical report, Na- tional Bureau of Economic Research. Carruth, A., Dickerson, A. & Henley, A. (2000), ‘What do we know about investment under uncertainty?’, Journal of Economic Surveys 14(2), 119–154. Cruz, J.-L. & Rossi-Hansberg, E. (2021), The economic geography of global warming, Technical report, National Bureau of Economic Research. Dell, M., Jones, B. F. & Olken, B. A. (2009), ‘Temperature and income: reconciling new cross-sectional and panel estimates’, American Economic Review 99(2), 198–204. Dell, M., Jones, B. F. & Olken, B. A. (2012), ‘Temperature shocks and economic growth: Evidence from the last half century’, American Economic Journal: Macroeconomics 4(3), 66–95. 21 Dixit, R. K., Dixit, A. K. & Pindyck, R. S. (1994), Investment under uncertainty, Princeton university press. Donadelli, M., Jüppner, M. & Vergalli, S. (2022), ‘Temperature variability and the macroeconomy: A world tour’, Environmental and resource economics 83(1), 221–259. Engle, R. F., Giglio, S., Kelly, B., Lee, H. & Stroebel, J. (2020), ‘Hedging climate change news’, The Review of Financial Studies 33(3), 1184–1216. Ferderer, J. P. (1993), ‘The impact of uncertainty on aggregate investment spending: an empirical analysis’, Journal of Money, Credit and Banking 25(1), 30–48. Ghosal, V. & Loungani, P. (2000), ‘The differential impact of uncertainty on investment in small and large busi- nesses’, Review of Economics and Statistics 82(2), 338–343. Gortan, M., Testa, L., Fagiolo, G. & Lamperti, F. (2024), ‘A unified dataset for pre-processed climate indicators weighted by gridded economic activity’, Scientific Data 11(1), 533. Gulen, H. & Ion, M. (2016), ‘Policy uncertainty and corporate investment’, The Review of financial studies 29(3), 523–564. Harris, I., Jones, P. D., Osborn, T. J. & Lister, D. H. (2014), ‘Updated high-resolution grids of monthly climatic observations–the cru ts3. 10 dataset’, International journal of climatology 34(3), 623–642. Hatfield, J. L. & Prueger, J. H. (2015), ‘Temperature extremes: Effect on plant growth and development’, Weather and climate extremes 10, 4–10. Huizinga, J. (1993), ‘Inflation uncertainty, relative price uncertainty, and investment in US manufacturing’, Jour- nal of Money, Credit and Banking 25(3), 521–549. Huntingford, C., Jones, P. D., Livina, V. N., Lenton, T. M. & Cox, P. M. (2013), ‘No increase in global temperature variability despite changing regional patterns’, Nature 500(7462), 327–330. IPCC (2023), Climate Change 2023: Synthesis Report. Contribution of Working Groups I, II, and III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, IPCC, Geneva, Switzerland. Kahn, M. E., Mohaddes, K., Ng, R. N., Pesaran, M. H., Raissi, M. & Yang, J.-C. (2021), ‘Long-term macroeconomic effects of climate change: A cross-country analysis’, Energy Economics 104, 105624. Kalkuhl, M. & Wenz, L. (2020), ‘The impact of climate conditions on economic production. evidence from a global panel of regions’, Journal of Environmental Economics and Management 103, 102360. Kotz, M., Wenz, L., Stechemesser, A., Kalkuhl, M. & Levermann, A. (2021), ‘Day-to-day temperature variability reduces economic growth’, Nature Climate Change 11(4), 319–325. Leahy, J. V. & Whited, T. (1995), ‘The effect of uncertainty on investment: Some stylized facts’. Lepore, C. & Fernando, R. (2023), Global Economic Impacts of Physical Climate Risks, International Monetary Fund. Lobell, D. B. & Asner, G. P. (2003), ‘Climate and management contributions to recent trends in US agricultural yields’, Science 299(5609), 1032–1032. McDonald, R. & Siegel, D. (1986), ‘The value of waiting to invest’, The quarterly journal of economics 101(4), 707– 727. 22 Myers, S. C. (1977), ‘Determinants of corporate borrowing’, Journal of Financial Economics 5(2), 147–175. Newell, R. G., Prest, B. C. & Sexton, S. E. (2021), ‘The gdp-temperature relationship: implications for climate change damages’, Journal of Environmental Economics and Management 108, 102445. Nordhaus, W. D. (2006), ‘Geography and macroeconomics: New data and new findings’, Proceedings of the Na- tional Academy of Sciences 103(10), 3510–3517. Olonscheck, D., Schurer, A. P., Lücke, L. & Hegerl, G. C. (2021), ‘Large-scale emergence of regional changes in year-to-year temperature variability by the end of the 21st century’, Nature Communications 12(1), 7237. Pindyck, R. S. (1990), Irreversibility, uncertainty, and investment, Technical report, National Bureau of Economic Research. Piontek, F., Drouet, L., Emmerling, J., Kompas, T., Méjean, A., Otto, C., Rising, J., Soergel, B., Taconet, N. & Tavoni, M. (2021), ‘Integrated perspective on translating biophysical to economic impacts of climate change’, Nature Climate Change 11(7), 563–572. Rohde, R. A. & Hausfather, Z. (2020), ‘The Berkeley earth land/ocean temperature record’, Earth System Science Data 12(4), 3469–3479. Roson, R., Sartori, M. et al. (2016), ‘Estimation of climate change damage functions for 140 regions in the GTAP9 database.’, Policy Research Working Paper-World Bank (7728). Schlenker, W. & Roberts, M. J. (2009), ‘Nonlinear temperature effects indicate severe damages to US crop yields under climate change’, Proceedings of the National Academy of sciences 106(37), 15594–15598. Sheng, X., Gupta, R. & Çepni, O. (2022), ‘The effects of climate risks on economic activity in a panel of us states: The role of uncertainty’, Economics Letters 213, 110374. Tubiello, F. N. & Rosenzweig, C. (2008), ‘Developing climate change impact metrics for agriculture’, Integrated Assessment Journal 8(1). Vasseur, D. A., DeLong, J. P., Gilbert, B., Greig, H. S., Harley, C. D., McCann, K. S., Savage, V., Tunney, T. D. & O’Connor, M. I. (2014), ‘Increased temperature variation poses a greater risk to species than climate warming’, Proceedings of the Royal Society B: Biological Sciences 281(1779), 20132612. Wenz, L., Carr, R. D., Kögel, N., Kotz, M. & Kalkuhl, M. (2023), ‘Dose–global data set of reported sub-national economic output’, Scientific Data 10(1), 425. Willner, S. N., Glanemann, N. & Levermann, A. (2021), ‘Investment incentive reduced by climate damages can be restored by optimal policy’, Nature Communications 12(1), 3245. 23 A Data construction The initial location data was constructed from the ORBIS firm-level data, provided by the Wharton Research Data Services. The dataset includes 1,031,238 unique locations, containing city, region, coun- try, availability, latitude, and longitude data. Of these locations, 161,822 lacked latitude and longitude information. Therefore, we needed to obtain the coordinates for these 161,822 locations to complete the dataset. The following sections describe the process of acquiring the coordinates and eventually merging them with the weather data. A.1 Missing Entries and Google Maps API We utilize the Geocoding API provided by Google Maps, a service that facilitates the conversion of various forms of location data. It accepts inputs in the form of an address or a place ID. This API is capable of transforming a given address into latitude and longitude coordinates along with a place ID. Conversely, it can also convert latitude and longitude coordinates or a place ID back into an address. For the purposes of this project, the Google Maps API returns coordinates based on the location name as input. To achieve precise latitude and longitude values, the Google Maps API requires an exact address (e.g., 200 University Ave W, Waterloo, ON N2L 3G1). However, with this dataset, the inputs are limited to city + region + country, which does not constitute a full address. Nonetheless, the API can still identify the location, but the accuracy would be classified as “APPROXIMATE” since the most granular information available is the city name. If provided with a full address, the API would return “ROOFTOP,” indicating it has retrieved precise latitude and longitude values. Figure 4 illustrates the information the Google Maps API returns for Vienna, Austria. Figure 4: Google Maps API Example A.2 Location Matching We developed an algorithm to facilitate location matching within the initial dataset. The first part of the algorithm identifies the available geographical information in the original data and converts it into homogenized values. For example, latitude and longitude values, originally in Degrees, Minutes, Seconds (DMS) format, are converted to a decimal format using the formula: decimal = 24 sign(positive or negative) × (degrees from DMS + (minutes from DMS/ 60) + (seconds from DMS/ 3600)). This standardization ensures uniformity of coordinates irrespective of their originating data sources. The second part of the algorithm identifies missing geographical information and employs the Google Maps API to retrieve coordinates. The latitude, longitude, and address data are extracted from the API using city, region, and country parameters. This process continues until every record in the dataset has been processed. It is important to note that there are instances in the original dataset where the city and region names do not match (i.e., the city does not exist within the indicated region). In such cases, the Google Maps API uses the region and country names to retrieve the coordinates. Additionally, there are some mismatches between city and country names in the original data when regional information is absent. Under these circumstances, the Google Maps API utilizes the name listed under the city column to retrieve coordinates. A.3 Indexing Once the unique location coordinates are consolidated into one dataset, each coordinate is assigned a grid index. The index file consists of midpoints of coordinates, increasing by one-degree increments in both latitude and longitude. An algorithm was developed to automatically align all unique locations with their corresponding weather index. First, the algorithm retrieves the lower and upper latitude and longitude bounds, along with the indices themselves. The latitude and longitude bounds extend ±0.5 degrees from each midpoint. Next, the algorithm checks whether the latitude and longitude of a given location fall within these specified bounds. If they do, the location is assigned to that specific index; otherwise, the process iterates until an appropriate index is found. This process continues until all coordinates have been matched to an index. A.4 Merging Location and Weather Data The final step to complete the dataset is to merge the location and weather data. To prepare the location data, a ‘year’ column spanning from 2000 to 2019 is added for each location, matching the weather dataset’s year range of 2000 to 2019. The merging process relies on the ‘index’ and ‘year’ as both values together indicate unique location and weather data. The algorithm iterates through each index and year combination in the location dataset to find the corresponding values in the weather dataset. If found, the location dataset is updated with the relevant weather information for each index and year. The final columns of the dataset are listed in Table 6. 25 Table 6: Merged Location and Weather Data Columns Description Column Description CITY_INTERNAT Location’s city name REGION_IN_COUNTRY Location’s region name LATITUDE Latitude of the location LONGITUDE Longitude of the location COUNTRY Location’s country name Found 1 = if the coordinate data was found using a data source, 0 = if coordinates were not found Data_source Source of the longitude and latitude data; ’Original’ = coordinates already exist, ’Google Maps API’ = coordinates were geocoded Google_Location_Address The address used by Google Maps API to retrieve the coordinates Time The year of the weather data Mean Mean temperature from weather data Variance Variance of temperature from weather data Quantile (quantile_01, etc) Quantile information from weather data Notes Additional Notes 26 B Additional descriptive results Economy Income Obs. MeanT Uncertainty MaxT SizeGrowth Algeria 3,823 48,605 26.20 1.91 40.34 4.28 Australia 43,996 87,941 23.42 1.60 33.89 0.97 Austria 39,096 291,168 13.10 2.80 28.52 2.62 Belarus 5,129 4,098 12.18 2.23 27.13 -17.16 Belgium 33,956 520,013 14.41 2.56 29.35 4.39 Bosnia and Herzegovina 3,672 43,263 17.27 2.72 32.42 9.37 Brazil 6,992 55,905 29.26 1.82 36.09 -7.46 Bulgaria 5,388 193,635 17.45 2.28 32.27 6.46 Chile 8,629 1,864 18.57 1.58 27.68 -0.57 China 4,612 2,053,291 18.00 1.98 32.08 7.97 Colombia 5,312 211,959 29.21 1.58 34.09 9.92 Croatia 10,135 112,445 17.53 2.62 32.31 9.70 Cyprus 24,820 10,323 24.66 1.42 34.62 8.13 Czechia 12,923 262,573 13.56 2.55 28.98 5.99 Denmark 45,346 89,826 11.74 2.04 24.57 -4.12 Estonia 14,807 59,713 9.94 2.32 24.78 9.31 Finland 36,253 239,198 7.07 2.24 23.38 4.79 France 31,883 1,739,504 16.45 2.41 30.48 3.95 Germany 34,004 1,865,497 13.37 2.39 27.83 4.59 Greece 18,941 90,041 20.78 1.76 33.40 5.53 Hong Kong SAR, China NA 673 19.76 1.89 30.87 11.25 Hungary 9,851 293,512 15.69 2.96 31.74 9.84 Iceland 50,683 21,232 6.13 1.91 15.56 15.55 India 1,385 297,873 31.45 1.52 40.62 1.21 Indonesia 2,163 6,143 31.54 1.53 36.16 -2.15 Ireland 36,303 151,598 13.36 1.45 21.94 9.70 Italy 27,366 2,294,430 18.57 1.81 31.57 8.06 Japan 31,172 857,053 17.19 1.66 31.31 8.65 Kazakhstan 7,349 13,058 11.19 2.59 31.96 -12.55 Korea, Rep. 19,454 1,406,758 17.59 2.00 31.62 12.17 Latvia 9,483 82,351 10.43 2.23 25.26 7.72 Lithuania 11,304 48,117 11.55 2.25 26.38 11.00 Luxembourg 65,147 126,453 14.33 2.65 29.91 2.23 Malaysia 7,558 124,523 31.46 1.32 35.69 3.66 Malta NA 756 21.33 1.42 32.21 5.52 Mexico 7,675 10,567 27.92 1.30 34.05 7.70 Montenegro 6,282 7,018 17.99 1.95 31.49 1.34 Morocco 2,886 74,422 24.62 1.77 35.20 7.68 Netherlands 38,131 1,010,802 13.82 2.26 27.37 3.99 New Zealand 32,706 8,654 17.17 1.51 26.41 4.85 North Macedonia 3,792 29,655 17.48 2.05 31.59 13.52 Norway 60,693 568,830 6.54 2.01 20.20 3.55 Pakistan 1,185 2,779 30.90 1.62 41.90 -12.54 Philippines 2,743 72,409 31.37 1.50 36.47 8.92 Poland 9,520 425,786 15.55 2.39 30.48 3.61 Portugal 16,839 396,852 20.72 1.71 32.03 4.64 27 Republic of Moldova 2,582 10,948 15.09 2.28 30.36 -4.67 Reunion (France) NA 4,259 23.97 2.09 33.39 11.22 Romania 6,427 316,709 15.55 2.55 30.58 13.42 Russian Federation 7,707 1,610,456 6.97 2.54 27.00 -11.39 Serbia 5,121 92,976 16.86 2.60 31.94 3.76 Singapore 35,608 113,343 31.58 1.32 35.66 8.32 Slovak Republic 12,281 136,609 14.42 2.49 29.39 5.47 Slovenia 17,069 74,526 15.80 2.54 30.67 10.35 South Africa 5,561 465 24.77 1.76 32.49 -72.42 Spain 23,239 1,869,672 20.51 1.71 32.45 11.11 Sweden 42,611 558,833 8.39 2.00 22.97 2.58 Switzerland 58,425 4,070 12.24 2.75 27.34 8.86 Taiwan, China NA 32,963 26.40 1.55 34.19 3.17 Thailand 4,778 2,519 32.97 1.25 37.79 13.82 Türkiye 9,067 102,580 19.36 2.02 33.48 -1.80 Ukraine 2,213 504,554 13.60 2.27 29.48 -13.43 United Kingdom 34,945 1,671,762 12.69 1.78 23.22 0.62 Uruguay 15,090 8,313 23.86 1.99 34.47 -6.42 Table 7: Country-Level Summary Statistics. This table presents key statistics for each country, includ- ing the number of observations, the mean temperature, its standard deviation, kurtosis, and the growth rate of fixed assets. These summary statistics are derived from firm-level observations. 28 Country 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Algeria 46 46 45 45 46 49 45 8 Australia 1 1 1 2 5 5 6 10 16 19 23 37 59 101 107 109 108 108 105 Austria 1 1 1 5 15 15 16 17 17 18 21 22 22 23 23 23 23 23 23 Belarus 1 1 37 36 36 Belgium 11 11 11 10 11 11 10 10 11 11 11 11 11 11 11 11 11 11 11 Bosnia and Herzegovina 11 11 11 12 12 12 12 12 12 12 12 12 14 14 14 14 14 14 14 Brazil 5 7 14 12 11 11 20 16 16 18 19 20 59 86 214 231 241 239 240 Bulgaria 20 16 16 15 15 14 17 17 19 19 18 19 22 22 22 22 22 22 23 Chile 3 3 8 6 7 12 13 14 15 13 12 12 12 14 5 1 1 1 1 China 1 1 2 5 451 430 442 435 395 414 410 300 267 261 270 265 221 Colombia 3 3 5 5 5 26 30 31 41 37 38 56 68 68 65 67 68 64 66 Croatia 17 18 18 19 20 20 20 19 20 20 20 20 20 20 20 21 21 20 20 Cyprus 3 3 3 3 3 3 3 3 3 3 3 3 3 3 Czech Republic 16 16 16 17 17 17 17 17 17 18 18 18 18 19 19 19 19 18 18 Denmark 14 14 14 13 10 10 11 11 15 14 14 14 14 15 14 14 16 21 21 Estonia 4 3 4 4 4 2 3 2 1 1 1 14 14 14 14 14 14 14 Finland 50 50 53 53 56 54 55 58 57 56 57 57 63 67 67 67 68 68 69 France 92 91 90 90 91 90 90 90 90 90 89 89 90 90 89 89 89 89 89 Germany 57 61 61 62 62 64 64 66 67 67 66 67 68 68 69 69 69 69 69 Greece 29 30 34 34 34 31 31 31 28 28 23 35 36 36 36 36 36 36 36 Hong Kong SAR, China 1 2 2 2 2 2 Hungary 19 19 18 19 20 19 19 19 19 20 20 20 20 20 20 20 20 20 20 Iceland 1 1 2 3 3 2 2 2 2 2 3 3 3 27 27 27 27 27 27 India 29 28 30 32 31 27 45 48 58 70 65 93 107 172 236 237 239 240 239 Indonesia 1 4 7 6 6 6 8 9 8 10 9 8 8 8 7 1 Ireland 9 15 15 15 16 16 17 16 16 15 16 16 17 16 16 16 16 16 16 29 Italy 58 57 58 58 59 60 59 59 59 59 59 60 61 61 61 61 61 61 61 Japan 53 56 53 58 58 58 63 64 65 65 67 67 68 69 69 69 69 69 70 Kazakhstan 2 22 30 34 42 104 116 117 111 113 116 118 Latvia 12 12 13 13 13 13 15 14 11 12 16 17 17 17 17 16 16 15 15 Lithuania 6 8 9 10 8 8 9 9 9 9 12 12 15 13 14 15 15 15 16 Luxembourg 1 1 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 Malaysia 25 25 27 30 30 30 31 36 36 Malta 1 1 1 1 1 1 1 1 1 1 1 1 1 Mexico 1 1 1 1 2 5 9 9 12 13 22 21 41 36 54 62 79 84 54 Montenegro 2 6 2 2 3 5 4 4 4 4 5 5 5 5 5 Morocco 1 1 1 1 31 31 33 33 34 29 29 33 Netherlands 14 14 14 14 14 15 15 15 16 15 15 15 14 14 14 14 14 14 14 New Zealand 1 2 3 6 9 8 11 15 18 26 21 20 20 19 21 North Macedonia 5 1 1 3 2 1 1 1 2 4 6 6 6 6 6 6 6 Norway 69 69 70 73 73 73 69 66 65 66 63 64 79 79 79 79 80 80 80 Pakistan 1 1 2 5 2 3 4 5 6 6 6 6 9 8 Philippines 8 9 18 33 33 38 40 40 40 38 38 42 38 38 Poland 52 51 52 52 53 53 53 52 53 53 51 53 55 55 55 55 56 55 56 Portugal 11 2 16 16 17 17 17 17 17 17 17 17 18 18 18 18 18 18 18 Republic of Korea 12 19 19 19 19 19 19 19 19 18 19 19 19 19 19 19 19 19 19 Republic of Moldova 2 6 6 6 6 7 7 6 7 5 7 8 8 8 8 10 10 Reunion (France) 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 Romania 34 33 33 35 35 36 35 35 34 33 33 35 42 42 42 42 42 42 43 Russian Federation 544 585 608 653 611 609 652 674 674 676 695 737 1,019 1,035 1,040 1,035 1,043 1,046 1,054 Serbia 6 16 16 15 14 14 14 13 17 16 17 17 17 17 17 17 17 17 Singapore 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Slovakia 10 10 12 12 12 11 12 12 12 12 12 13 14 14 14 14 14 14 14 Slovenia 5 5 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 South Africa 1 2 3 5 3 3 4 4 3 7 7 4 4 5 1 1 Spain 81 82 82 82 82 82 82 82 82 82 82 82 82 82 82 82 82 82 82 Sweden 68 69 70 69 69 69 70 66 65 65 67 70 86 88 88 88 89 89 89 Switzerland 3 5 6 6 7 7 7 7 7 7 7 7 9 9 9 9 9 9 9 Taiwan, China 6 5 5 6 5 5 6 6 6 6 6 6 7 7 7 7 7 7 7 Thailand 1 2 3 14 13 13 5 4 Turkey 2 7 17 29 26 34 37 42 96 98 97 97 96 97 94 95 Ukraine 93 95 95 92 93 92 91 89 92 90 88 106 105 103 104 105 105 105 105 United Kingdom 47 47 47 48 49 50 49 49 48 47 49 50 58 58 58 58 58 58 58 Uruguay 1 1 1 1 1 1 1 1 2 14 13 16 18 20 20 Table 8: Number of grids for each country over the time period. This table displays the annual count of grids for each country from 2001-2019. 30 Algeria Australia Austria Belarus Belgium Bosnia and Herzegovina 26.8 15 19 24.0 13 16 26.4 14 18 23.5 15 26.0 13 12 17 23.0 14 25.6 22.5 12 11 16 13 Brazil Bulgaria Chile China Colombia Croatia 19 19.0 30.0 19 29.6 19.0 18 18.5 29.2 29.5 18 18.5 18.0 28.8 17 29.0 17 18.0 17.5 28.4 16 17.0 28.5 16 Cyprus Czech Republic Denmark Estonia Finland France 25.5 16 13 11 8.0 18 25.0 15 7.5 14 12 10 7.0 17 24.5 6.5 13 11 16 24.0 9 6.0 12 10 5.5 Germany Greece Hong Kong SAR, China Hungary Iceland India 22.0 21.0 18 15 6.5 31.8 21.5 20.5 17 14 21.0 6.0 31.5 16 13 20.0 20.5 15 5.5 31.2 12 20.0 19.5 14 5.0 Indonesia Ireland Italy Japan Kazakhstan Latvia 32.1 19.5 12.0 12 18.0 31.8 13.5 19.0 11.5 17.5 11 31.5 18.5 17.0 11.0 31.2 13.0 18.0 10 16.5 10.5 30.9 17.5 Temperature mean Lithuania Luxembourg Malaysia Malta Mexico Montenegro 32.00 22.0 13 16 31.75 28.5 19.0 15 21.5 18.5 12 31.50 28.0 18.0 14 31.25 21.0 17.5 11 31.00 27.5 17.0 13 20.5 30.75 16.5 Morocco Netherlands New Zealand North Macedonia Norway Pakistan 25.5 18.0 32.0 25.0 15 17.5 18 7 31.5 24.5 14 17.0 6 31.0 16.5 17 24.0 13 30.5 16.0 16 5 23.5 12 30.0 Philippines Poland Portugal Republic of Korea Republic of Moldova Reunion (France) 18 18.5 17 32.0 21.5 23.0 17 21.0 18.0 16 22.5 31.5 16 22.0 20.5 17.5 15 31.0 21.5 15 20.0 17.0 21.0 14 30.5 14 19.5 20.5 Romania Russian Federation Serbia Singapore Slovakia Slovenia 19 32.25 17 7.5 18 32.00 16 17 16 7.0 17 31.75 15 16 31.50 15 6.5 16 14 15 31.25 14 6.0 15 31.00 13 South Africa Spain Sweden Switzerland Taiwan, China Thailand 26 21.5 14 33.5 9 27.0 21.0 13 33.0 25 26.5 20.5 8 32.5 12 24 20.0 7 26.0 32.0 11 31.5 19.5 25.5 2001 2010 2019 2001 2010 2019 Turkey Ukraine United Kingdom Uruguay 13.5 24.5 15 20 13.0 24.0 14 23.5 12.5 19 13 12.0 23.0 18 22.5 2001 2010 2019 2001 2010 2019 2001 2010 2019 2001 2010 2019 Figure 5: Time series of temperature mean. Each plot shows the economy-specific annual trajectory of the temperature mean. The black line in each plot represents the annual mean value, with a grey shaded area indicating the 95% interval. 31 Algeria Australia Austria Belarus Belgium Bosnia and Herzegovina 3.0 3.00 2.1 1.8 1.7 3.0 2.75 3.0 1.9 2.5 1.6 2.50 2.5 1.7 1.5 2.5 2.0 2.25 1.5 1.4 1.5 2.0 Brazil Bulgaria Chile China Colombia Croatia 1.9 2.4 1.8 3.2 3.0 1.6 2.2 1.7 2.5 2.8 1.8 1.6 1.5 2.0 1.5 2.4 2.0 1.7 1.4 1.5 1.4 1.8 1.3 2.0 Cyprus Czech Republic Denmark Estonia Finland France 1.8 2.7 2.8 3.0 3.0 2.6 1.6 2.8 2.4 2.1 2.5 2.5 2.4 1.4 2.4 1.8 2.0 2.0 2.2 1.2 2.0 1.5 1.5 2.0 1.5 Germany Greece Hong Kong SAR, China Hungary Iceland India 2.2 4.0 2.50 1.8 2.75 2.2 2.0 3.5 2.25 1.7 2.50 2.0 1.8 1.8 3.0 2.00 1.6 2.25 1.6 1.6 1.5 2.00 2.5 1.75 1.4 1.4 1.4 Indonesia Ireland Italy Japan Kazakhstan Latvia 1.8 2.0 1.9 3.5 3.0 1.7 1.8 2.0 1.8 3.0 1.6 1.6 1.8 2.5 1.7 2.5 1.5 1.4 2.0 1.6 1.6 1.4 2.0 Temperature uncertainty 1.3 1.2 1.4 1.5 1.5 Lithuania Luxembourg Malaysia Malta Mexico Montenegro 3.0 3.1 1.6 2.9 1.5 1.75 2.4 2.5 1.6 2.7 1.4 1.50 2.0 2.0 2.5 1.3 1.4 1.25 2.3 1.2 1.6 1.5 1.2 1.00 Morocco Netherlands New Zealand North Macedonia Norway Pakistan 2.2 1.8 3.0 1.9 2.50 1.7 2.4 1.8 2.0 2.5 1.7 2.25 1.6 2.1 1.8 1.5 2.0 1.6 2.00 1.8 1.5 1.4 1.5 1.4 1.6 1.75 Philippines Poland Portugal Republic of Korea Republic of Moldova Reunion (France) 3.5 1.8 2.0 2.8 1.9 2.25 3.0 2.4 1.7 1.6 2.4 1.8 2.00 2.5 2.2 1.5 1.7 1.4 2.0 1.75 2.0 1.6 2.0 1.3 1.5 1.50 1.5 Romania Russian Federation Serbia Singapore Slovakia Slovenia 3.2 1.7 3.2 4 3.5 2.8 1.5 3.0 2.8 3 3.0 2.4 1.3 2.6 2.4 2.5 2 2.0 1.1 2.2 2.0 2.0 0.9 1.8 South Africa Spain Sweden Switzerland Taiwan, China Thailand 2.25 1.9 2.8 1.8 1.5 2.00 1.8 2.4 3.0 1.6 1.4 1.75 1.7 2.0 2.5 1.3 1.4 1.2 1.50 1.6 1.6 1.25 2.0 1.2 1.1 2001 2010 2019 2001 2010 2019 Turkey Ukraine United Kingdom Uruguay 3.0 2.25 2.2 2.25 2.00 2.0 2.5 2.00 1.75 2.0 1.8 1.75 1.50 1.6 1.5 2001 2010 2019 2001 2010 2019 2001 2010 2019 2001 2010 2019 Figure 6: Time series of temperature uncertainty. Each plot depicts the economy-specific annual vari- ation in temperature uncertainty. The black line represents the annual mean of the standard deviation, with a grey shaded area denoting the 95% interval. 32 Algeria Australia Austria Belarus Belgium Bosnia and Herzegovina 32 30 35.0 35 41 35 34 30 28 32.5 34 33 40 28 26 30.0 32 33 27.5 31 26 30 39 32 24 Brazil Bulgaria Chile China Colombia Croatia 36.5 35 33.0 35.0 35 34 34.5 34 36.0 28 32.5 33 34.0 33 35.5 32 32 27 32.0 33.5 35.0 31 31 33.0 30 34.5 30 32.5 Cyprus Czech Republic Denmark Estonia Finland France 33 36 36 27 26 34 31 27 35 25 25 24 32 29 30 34 27 23 23 22 28 21 21 20 Germany Greece Hong Kong SAR, China Hungary Iceland India 32 36 33 35 18 41.5 32 34 17 30 35 33 41.0 34 31 32 16 28 30 40.5 31 26 33 29 30 15 40.0 32 29 Indonesia Ireland Italy Japan Kazakhstan Latvia 25 34 33 34 37.0 24 33 27 23 32 33 22 32 32 25 36.5 31 21 31 31 36.0 20 30 30 23 30 Temperature maximum 19 29 29 Lithuania Luxembourg Malaysia Malta Mexico Montenegro 35.0 37.0 34 35.5 34 28 32.5 36.5 35.0 33 33 32 26 30.0 36.0 34.5 32 31 24 35.5 34.0 27.5 30 35.0 31 33.5 29 22 25.0 Morocco Netherlands New Zealand North Macedonia Norway Pakistan 37 32 29 25.0 44 28 34 36 30 33 22.5 43 27 35 28 26 32 20.0 42 26 25 31 34 24 30 17.5 41 24 Philippines Poland Portugal Republic of Korea Republic of Moldova Reunion (France) 38.0 34 36 34 34 37.5 33 32 32 34 37.0 32 32 36.5 30 30 31 32 36.0 28 30 35.5 30 28 30 Romania Russian Federation Serbia Singapore Slovakia Slovenia 29 38 33 34 32 28 37 32 31 30 27 32 36 30 29 35 28 26 30 28 South Africa Spain Sweden Switzerland Taiwan, China Thailand 35 35 39 34 34 25.0 30.0 33 33 34 38 32 32 22.5 27.5 31 37 31 20.0 25.0 33 30 30 36 2001 2010 2019 2001 2010 2019 Turkey Ukraine United Kingdom Uruguay 26 37 35 31 36 34 30 24 35 29 22 34 33 28 20 33 32 2001 2010 2019 2001 2010 2019 2001 2010 2019 2001 2010 2019 Figure 7: Time series of temperature maximum. Each plot illustrates the economy-specific annual pattern of temperature maximum. The black line indicates the annual mean of temperature maximum, accompanied by a grey shaded area representing the 95% interval. 33 2001 2002 2003 2004 2005 200 200 100 100 100 100 100 0 0 0 0 0 −100 −100 −100 −100 −100 −200 −200 −200 −200 1 2 3 4 1 2 3 4 1 2 3 1 2 3 1.5 2.0 2.5 3.0 3.5 2006 2007 2008 2009 2010 200 200 200 100 100 100 100 100 0 0 0 0 0 −100 −100 −100 Fixed asset growth (%) −100 −100 −200 −200 −200 −200 1 2 3 4 1 2 3 4 1.0 1.5 2.0 2.5 3.0 3.5 1 2 3 1 2 3 4 2011 2012 2013 2014 2015 200 200 100 100 100 100 100 0 0 0 0 0 −100 −100 −100 −100 −100 −200 −200 −200 −200 −200 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 2016 2017 2018 2019 Total 200 200 200 200 100 100 100 100 100 0 0 0 0 0 −100 −100 −100 −100 −100 −200 −200 1 2 3 4 5 1 2 3 4 1 2 3 4 5 1 2 3 4 1 2 3 4 5 Temperature uncertainty Figure 8: Scatter plot of temperature uncertainty and fixed asset growth. This scatter plot displays the relationship between the temperature uncertainty and fixed asset growth, for each year across all grids. Each dot represents a grid-level observation for a specific year, plotted according to its grid- average fixed asset growth and temperature uncertainty. 34