Climate Change and Adaptation in Global Supply-Chain Networks

Abstract

Businesses face increasing pressure to address their operational exposure to physical climate hazards. While managers and investors are looking for ways to alleviate climate change risks by adapting their operations (Lin et al. 2020) and investments (Krueger et al. 2020; Ilhan et al. 2023), academic research has primarily studied how transitory weather shocks affect firms’ earnings (Addoum et al. 2020), stock returns (Cuculiza et al. 2023), and labor and capital productivity (Graff-Zivin et al. 2018; Zhang et al. 2018), among others. Much less is known about how firms adapt to gradual change, despite the fact that firms’ endogenous responses are crucial to understanding the long-run effects of climate change on financial market outcomes.

Operational risk management in response to climate change is particularly important for firms engaged in extensive international production networks. In a globalized economy, supply chains often move through parts of the world that are most vulnerable to the impact of climate change. As a result, firms might be indirectly exposed to physical risks due to their suppliers. Reflecting these concerns, over 50% of CEOs mentioned risks posed to their global supply chains by climate change as one of their primary concerns in a recent survey (PWC 2015).

However, adapting to climate change is a complex task for economic agents in general and firms in supply-chain organizations in particular. Climate change is characterized by unknowable uncertainty—particularly in the short and medium runs—as weather realizations provide a noisy signal of potential changes in the underlying distribution (Deryugina 2013; Kala 2019). Further, indirect exposure to climate change due to suppliers and customers can be challenging to identify. In this environment, it is unclear how gradually changing exposure to climate hazards could affect firms’ decisions to discontinue existing and begin new supply-chain relationships.

In this paper, we study whether firms adjust their supply-chain networks in response to perceived increases in their suppliers’ heat exposure.¹ After estimating how the financial consequences of adverse weather propagate from suppliers to their corporate customers around the world, we investigate whether and how firms adapt their supply-chain organizations in response to changes in supplier exposure. In particular, we examine how discrepancies between realized and expected exposure to heat affect the continuation of existing and the initiation of new supply-chain relationships. Our main contribution is to show that customers are more likely to terminate suppliers when their exposure increases beyond historical expectations, and switch to less exposed replacement suppliers. Thereby, the results provide new evidence that climate change could affect firms’ operational risk management and the formation of global supply chains.

We combine detailed firm-level supply-chain data from FactSet Revere with financial performance data from Worldscope, headquarters and establishment locations from FactSet Fundamentals and Orbis, and data on local temperatures and temperature projections computed in the fifth phase of the Coupled Model Intercomparison Project (CMIP5) from the European Center for Medium-term Weather Forecasts (ECMWF). Our supply-chain data set includes 5,628 (8,200) unique supplier (customer) firms across 92 (74) countries around the world over the period from 2003 to 2016.

We focus on heat as the most pervasive projected gradual trend under climate change. Previous studies have documented several channels through which temperatures affect firm productivity. For example, heat reduces worker performance (Graff-Zivin et al. 2018), labor supply (Graff-Zivin and Neidell 2014), and firm-level output (Zhang et al. 2018), with sharp declines at temperatures over 30$°$C. Further, anecdotal evidence suggests that heat can constrain water supply, distort transportation, and wreak havoc on electrical grids. The resultant risks to firm operations are expected to increase. For our sample, the number of days on which temperatures exceed 30$°$ C are projected to rise from currently 2.7% of days to more than 12% of days by 2100 without substantial efforts to reduce emissions according to CMIP5 model output.

Before examining firms’ supply-chain adaptations, we document how heat exposure affects the financial performance of suppliers and their downstream customers. While Barrot and Sauvagnat (2016) and Carvalho et al. (2021) show that the effect of natural disasters propagates through production networks, it is unclear if moderate realizations of adverse weather, which are projected to gradually increase in frequency and severity, have similar effects. We construct location-specific measures of suppliers’ heat exposure based on daily temperatures over a quarter or year at their headquarters. In robustness tests, we measure heat across all supplier locations. Consistent with Somanathan et al. (2021), Zhang et al. (2018), and Pankratz et al. (2023), we find a significant negative effect of heat on supplier operating performance. Further, we document repercussions on downstream customers: following the occurrence of high temperatures in a given firm-quarter at supplier locations, customer operating income over assets decreases by 0.6% relative to the mean. Whereas these effects appear moderate relative to large-scale natural disasters, they could be large enough as financial incentives for customers to consider adapting their supply chains to gradual change.²

Our main analysis focuses on the question of whether firms attempt to learn about climate change and adapt to gradual increases in heat exposure. Given that short- and medium-term weather realizations provide noisy signals for the underlying distribution, detecting change is challenging. However, prior research in finance and economics has proposed experience-based Bayesian updating to model learning in general (Alevy et al. 2007; Chiang et al. 2011), and about climate change in particular (Kelly et al. 2005; Deryugina 2013; Moore 2017; Kala 2019; Choi et al. 2020). In this paper, we test if observed terminations of supply-chain relationships are consistent with an attempt of corporate customers to learn about climate change based on short-run increases in the heat exposure of their suppliers.

We assume that customer firm managers carefully consider and trade off the expected costs and benefits, such as the exposure to environmental hazards, product quality, and input prices of prospective suppliers, based on observable characteristics when entering a supply-chain relations. Under this setting, customer firms have no incentive to alter their supply-chain relations in response to the occurrence of heat events as long as the realizations are in line with ex ante expectations. However, the underlying distribution of adverse weather is projected to change in heterogeneous ways across geographies due to climate change. Hence, if customers perceive an increase in the likelihood of occurrences at their supplier firm, a previously optimal supplier may no longer be optimal going forward, and existing supplier relations may be terminated more frequently.

To test this idea, we construct a new measure of realized versus expected exposure by comparing heat days before and during a given supply-chain relationship. Our results show that a supply-chain link is 7.4% more likely to be terminated in a given year if the realized exposure to heat exceeds proxies of customers’ ex-ante expectations. This result is robust to alternative periods over which customers form priors, and holds after controlling for industry and country-by-time fixed effects for suppliers and customers. Consistent with experience-based learning about climate change (e.g., Deryugina 2013), we find stronger results after the first relationship-year in which deviations from ex ante expectations would be particularly challenging to interpret. We also document that the likelihood of terminations continues to increase with the repeated exceedance of likely ex ante expectations and when deviations of priors and realizations are large.³

To explore climate change adaptation as the mechanism underlying these results, we use cross-sectional variation in adaptation readiness across supplier countries. Based on data from the Notre-Dame Global Adaptation Index (ND-GAIN) on the economic, governance, and social capacity to adapt to climate change, we find that the likelihood of supplier-termination is higher when adaptation readiness in the supplier country is low. This finding is consistent with the idea that suppliers in low adaptation capacity countries pose a higher future exposure to customers.

In line with the notion that customers actively adapt to perceived change in climate exposure, we document adjustments along other dimensions. Customers increase their inventory, cash holdings, and R&D investments when the realized exposure to heat exceeds ex ante expectations. We find weak evidence pointing toward an additional effect on supplier diversification, that is customers increase the number of other suppliers in the industry of a supplier when realizations of heat exceed plausible expectations. In contrast, we find no evidence for increases in M&A activity. Further, the effect is stronger in the cross-section for suppliers in competitive industries and weaker for closely integrated supply chains and high customer reliance on supplier inputs. The results remain similar when we implement our tests as linear probability models, logistic regressions, and as Cox proportional hazard models.

A potential concern with our interpretation of the results is that supplier terminations may be solely driven by physical supplier-disruptions without customers’ attributing those disruptions at least partly to suppliers’ increased climate risk exposure. Arguably, it is impossible to prove if and how exactly customers update their beliefs after witnessing supply-chain disruptions. However, and consistent with the notion that customers terminate relationships (only) when they believe that suppliers’ climate exposure is persistent, we find that transitory supplier exposures, that is adverse realizations of heat regardless of expectations from a benchmark period, do not lend themselves to subsequent terminations.

While our main tests reflect the idea that firms form priors and update their beliefs based on experienced change, we also consider the role of long-term temperature projections from the CMIP5 project. From the ECMWF, we obtain projections for the average number of heat days between 2040 and 2059. We then estimate our main tests for subsamples of suppliers for which long-term climate models project limited change under various emission trajectories. We find that customers respond to increases in adverse weather above ex ante expectations even when long-run projections indicate little to no future change, which may emphasize the challenges of learning about climate change based on experienced change.

Last, we examine whether firms consider suppliers’ potential exposure when switching to new suppliers. For this purpose, we identify replacement suppliers as firms with identical SIC codes as the terminated suppliers which begin a supplier-relationship with the same customer within one year. We estimate linear probability models to test whether replacement suppliers have a lower exposure to heat than terminated suppliers conditional on an increased exposure during the terminated supplier relationship. We find a positive effect of climate exceedance on the likelihood that customers choose a replacement supplier with lower ex post exposure observed both during and after the initial relationship. An unexpectedly high exposure during the initial relationship increases the probability that customers choose a less exposed replacement by 6 to 10 percentage points, controlling for industry- and country-specific time fixed effects of both suppliers and customers. We find a smaller, less precisely estimated effect when considering differences in long-term projections.

Our paper contributes to the literature on climate change, finance, and economics along several dimensions. First, we provide novel evidence on the implications of climate change for firms and investors. Previous research in finance has studied how weather shocks directly affect firms and investors (e.g., firm profitability (Zhang et al. 2018; Addoum et al. 2020; Pankratz et al. 2023), housing prices (Baldauf et al. 2020), stock returns (Cuculiza et al. 2023), financial markets (Hong et al. 2019; Schlenker and Taylor 2019), and capital structure (Ginglinger and Moreau 2022)). In contrast, our paper studies how firms respond to perceived, gradual changes in climate risk in a unique empirical setting, by focusing on firms’ adaptation in the context of global supply-chains. Further, our motivating tests show that firms can be indirectly exposed to physical climate hazards, such as heat through their global supplier network. This aspect of our findings connects to Barrot and Sauvagnat (2016), Boehm et al. (2019), and Carvalho et al. (2021), who document the propagation of natural disasters along firm linkages in the United States. Relative to these papers, our findings provide evidence that heat exposure, which causes less severe physical damages but may intensify with long-term, gradual climate change, also propagates through production networks.

Second, our paper is among the first in the finance and environmental economics literature to study how firms learn about and adapt to climate change. For example, Lin et al. (2020), Li et al. (2020), and Li et al. (2023) examine the shift of electricity providers to more flexible power plants, changes in local employment composition, and green innovations in response to changes in local climate risk, respectively. Our paper contributes to this emerging literature by studying firms’ responses to indirect climate change exposure through supply chains, guided by a conceptual framework of firm production and experience-based learning about climate change.

Third, we contribute to the literature on learning about climate change. Choi et al. (2020) find that market participants revise their beliefs about climate change when experiencing unusual weather using Google search data. Deryugina (2013) uses survey data on beliefs about global warming to show that local temperature fluctuations affect these beliefs in a Bayesian framework. Moore (2017) and Kala (2019) develop and test theoretical models on experience-based learning about climate change. Our work builds on these models, but explores learning and adaptation outside of the agricultural sector, combining both observed signals and climate projections.

Our main finding on the adaptation of supply chains has potentially important implications, as the areas of the world which are disproportionately affected by the impact of climate change are already less developed today (Burke et al. 2015; Carleton and Hsiang 2016). Our findings also speak to the growing literature on endogenous production networks and macroeconomic growth (e.g., Antràs et al. 2017; Lim 2018; Oberfield 2018; Acemoglu and Azar 2020).

1 Data Sources and Descriptive Statistics

1.1 Global supply chains

We use international data on supply-chain relationships from FactSet Revere. The data are hand-collected, for example from annual reports, SEC filings, investor presentations, websites, press releases, supply contracts, and purchase obligations. Previous research (e.g., Hertzel et al. 2008; Cohen and Frazzini 2008; Banerjee et al. 2008) has primarily relied on SEC regulation S-K, which requires U.S. firms to disclose customers representing at least 10% of their sales. Because of this cutoff, start and end dates cannot be distinguished from changes in sales around the reporting threshold. Further, the global coverage in FactSet is important for our study, as supply chains often move through parts of the world outside of the United States that are most vulnerable to climate change. In total, we observe 8,200 (5,769) customer (supplier) firms from 74 (92) countries, comprising almost 595,000 pair-year-quarter observations from 2003 to 2016. Figure 1 shows that most suppliers are located in Asia (40%), North America (39%), and Europe (17%) and operate in manufacturing (SIC 2/3) or transport and utilities (SIC 4).⁴

Fig. 1

Geography of customer-supplier linkages

This figure shows the connections of the headquarters of the customers and suppliers in our sample. Supply-chain relationships and firm locations are obtained from FactSet Revere, FactSet Fundamentals, and Orbis. Table IA.1 corresponds to the plot and reports the number of customers and suppliers by geographic regions.

Open in new tabDownload slide

1.2 Accounting data and firm characteristics

We obtain firm financial data from 2000 to 2016 from Worldscope, with quarterly operating income scaled by lagged total assets as our main measure of performance. In addition, we collect data on revenues, employees, asset tangibility, operating margins, inventories, accounts receivables, cost of goods sold (COGS), spending on research and development (R&D), and delisting dates from Worldscope and Datastream. We construct measures of industry competitiveness as the number of firms and the Herfindahl-Hirschman Index (HHI) of revenues in a given SIC two-digit industry. From the Bureau of Economic Analysis (BEA), we obtain global input-output matrices for 2012 to construct industry-level input concentration as value of sales from supplier- to customer industry and the HHI of dollar values across all input industries for each customer industry. For comparability, we convert all outcomes into U.S. dollars, and trim variables above (below) the 99th (1st) percentile to remove outliers. We exclude firms in the financial industry (SIC 6).

Worldscope covers 99% of the global market capitalization. In contrast, our sample is subset of supply chains and limited to listed firms with financial reports. Moreover, international disparities in the availability of financial data may bias the sample toward developed markets. Compared to the universe of firms in Worldscope, the observed suppliers are similar to the average firms in their home countries in size (book assets and market capitalization) but have higher revenues over assets. In contrast, customer firms are significantly larger than the average firm in their home country.⁵

We use addresses of headquarters from FactSet as our primary measure of firms’ locations. However, plants and establishments might be remote from headquarters. Hence, we collect additional facility-level locations from Orbis. In total, we obtain 1.1 million addresses of incorporated subsidiaries, branches, and establishments. We geocode city, ZIP code, and street names using Bing Maps. As a measure of concentration, we calculate the share of establishments located within a 30-km radius of firms’ headquarters. In the main tests, we exclude dispersed firms with fewer than 10% of assets within 30 km of the firms’ headquarters. The cutoff follows Barrot and Sauvagnat (2016), who limit their sample to firms with at least 10% of employees at the headquarter locations. In additional analyses, we estimate the propagation of incidents aggregated across facilities.

1.3 Local temperatures and projections

The exposure to heat is projected to gradually increase in frequency and severity in the near future (CSSR 2017). We construct measures of firms’ exposure to heat at the firm-quarter-level using location-specific information from the ERA5 reanalysis data provided by the European Center for Medium-term Weather Forecasts (ECMWF). Re-analyses are generated by interpolating local records using atmospheric models. ERA5 provides global, daily maximum temperatures on a 0.25$°$ grid starting in 1979 (Hersbach et al. 2020). We match customers and suppliers to temperatures at the closest ERA5 grid node, and convert temperatures from Kelvin to Celsius. Accounting for firms’ reporting schedules, we sum the number of days of heat per financial quarter or year. For the main measure, we use a daily maximum temperature of 30$°$C as the temperature threshold to define a day as hot. This choice is based on the literature on physiology, temperatures, and labor productivity (e.g., Graff-Zivin and Neidell (2014); Burke et al. (2015); Carleton and Hsiang (2016); Sepannen et al. (2006)), which documents a sharp decline in worker performance for temperatures above 30$°$C. Similarly, the National Weather Service defines heatwaves based on a sequence of days with temperatures exceeding 90$°$F (32$°$C). In robustness tests, we combine this absolute threshold with relative definitions of high temperatures based on day of the year- and location-specific temperature distributions from 1979 to 1999.

Next, we add temperature projections from the CMIP5 provided by the ECMWF. The CMIP5 is used in the Intergovernmental Panel on Climate Change (IPCC) Assessment Reports and described by Hurrell et al. (2011) and Taylor et al. (2012). To compare realized temperatures with projections, we calculate the projected change at supplier locations as the number of days over 30$°$ C modeled from 2006 to 2019 to projections for 2040 to 2059 based on the output of the MPI-ESM-LR model, averaged across all available ensemble members. We obtain projections following the Representative Concentration Pathway (RCP) 2.6, 4.5, and 8.5, which provide temperature projections for different levels of future emission. The RCP 8.5 comes closest to a “business as usual scenario,” assuming limited efforts to reduce emissions.

In robustness tests, we compare our data with records of natural disasters from EM-DAT from the Centre for Research on the Epidemiology of Disasters (CRED 2011). The data are at the country-level. Hence, it is less clear whether firms were directly affected. EM-DAT is commonly used for international studies in the economic literature on natural disasters (e.g., Strömberg 2007; Noy 2009; Lesk et al. 2016). Table 1f shows that less than 6% (4%) of our observations coincide with records of fires (droughts) in the suppliers’ home countries.

Table 1 reports descriptive statistics with variable descriptions in Table A.1. The average supplier firm in our sample experiences 14 days above 30$°$ C per financial quarter with an annual average temperature of 19$°$ C (Table 1a). As in prior research (e.g., Banerjee et al. 2008), we find asymmetries between suppliers and customers (Table 1b). The median customer holds 12 times the assets of the median supplier. When sales volumes are disclosed (<10% of the sample), customers represent 19% of suppliers’ sales, whereas sales from suppliers only account for 2% of customers’ cost of goods sold (Table 1c). Measures of supplier and customer industry competitiveness are described in Table 1d and 1e. Table 1f compares our data with records of natural disasters in EM-DAT.

2 Heat, Supply Chains, and Financial Incentives to Adapt

Our main research question on the adaptation of supply chains implicitly assumes that the occurrence of heat at supplier firms has direct and indirect adverse effects on suppliers and customers, providing an incentive for customer firms to adapt when temperatures increase. To validate this assumption, we examine the direct adverse effects of weather on supplier performance and the downstream propagation effects on customers. The downstream effects of adverse weather in production networks are theoretically ambiguous. On the one hand, customers might mitigate the effects, that is through multisourcing or inventory management, so that distortions dissipate in the supply chain. Similarly, suppliers’ limited bargaining power could force them to absorb higher costs. On the other hand, even small disruptions could decrease production output and cause supply-chain glitches in modern just-in-time production, when the specificity of inputs is high (Barrot and Sauvagnat 2016), or when customers’ ability to procure inputs from alternative sources is limited.

2.1 Empirical strategy

Our validation exercise proceeds in two stages. The methodology follows Barrot and Sauvagnat (2016) and Carvalho et al. (2021), who document the direct effects and shock propagation from suppliers to customers following large-scale natural disasters. In contrast, we focus on adverse weather, which is projected to heterogeneously increase in frequency and severity around the world. In the first (second) test, we focus on supplier (customer) operating income scaled by assets as our main measure of performance. We lag assets by one year to ensure that potential direct effects of heat exposure on assets does not confound the results. Since adverse weather might distort suppliers’ operations in ways not reflected in financial performance, the estimates could understate the extent to which supplier-customer relationships are challenged by high temperatures.

In both stages, we use variation in heat exposure at supplier locations, which is plausibly exogenous and randomly distributed conditional on locations and seasons. We isolate this variation through OLS regressions with firm-by-fiscal-quarter fixed effects. The specification serves two goals: First, weather and financial performance are likely endogenous in the cross-section, as firms select into locations based on various firm-level characteristics. In contrast, within-location variation in heat days is arguably exogenous, and can only be predicted with precision over very short horizons. Second, firm-by-fiscal-quarter fixed effects mitigate concerns about confounding firm-specific patterns of seasonality. For example, certain types of firms earn higher revenues in summer, which could correlate with the incidence of heat. Since seasonality differs between firms, firm-by-fiscal-quarter fixed effects address the issue better than individual firm and quarter fixed effects. In the second stage, we aggregate the quasi-random variation in exposure at supplier locations by customer-year-quarter (similar for example to Kale and Shahrur 2007; Banerjee et al. 2008; Barrot and Sauvagnat 2016; Campello and Gao 2017; Cen et al. 2017; Phua et al. 2018). As it is ex ante unclear whether the effects manifest immediately, the main tests include three lags of heat days.

2.2 Results

Table 2a reports the results for the direct effects on suppliers. From column 1 to 4, we sequentially introduce fixed effects. Across all specifications, an additional hot day significantly decreases suppliers’ quarterly $Operating Income/Assets$⁠. In the most conservative specification (column 4), we find a reduction of 0.0087 percentage points across the current and last three quarters. This effect is economically meaningful. The standard deviation of affected days is 16.2, translating the estimate into an 13.8% decrease in operating income over assets for a one-standard-deviation increase in heat exposure. As in Barrot and Sauvagnat (2016), the direct effects persist in the directly affected and two subsequent quarters.⁷

Panel 2b shows the indirect effects on customers of affected suppliers. In line with evidence on the propagation of idiosyncratic shocks in supply chains, heat at the locations of the suppliers decreases the financial performance of downstream customers. Specifically, one additional day of heat reduces quarterly operating income over assets of the indirectly affected customer by 0.0007 percentage points summed over the lags $t=−4,…,0$ (column 4). For a one-standard-deviation increase in heat days, the downstream effect translates into a 0.6% decrease in operating income over assets, corresponding to an absolute amount of $≈$355,000 US$for the average firm. The indirect effects are smaller than the direct effects at approximately 4.6% of the direct effect. For interpreting the magnitudes, it is important to note that these effects are estimated net of mitigating measures firms may already have put in place. Hence, the estimates are a lower bound, and likely to constitute financial incentives for customers to monitor potential increases in frequency and severity.

We provide additional analyses in the appendix. First, we estimate a dynamic version of Equation (2), using indicator variables to capture the effect on customer operating income in the t—4 to t + 6 quarters around the occurrence of a heatwave at the supplier firm. The corresponding dynamics plot is shown in Figure A.1. While the coefficient estimates decline between periods t—3 and t—1, all coefficients in the pre-period are statistically indistinguishable from zero, at both the 5% (shown in the figure) and the 10% levels. Some of the decline may be due to imprecise measurement of the beginning of a heatwave across fiscal quarters. The coefficients drop significantly below zero in quarters t + 1 and t + 2, and subsequently gradually return to pre-event levels. Second, we replace our definition of hot days (>30$°$ C) in Table A.2. We find similar effects for temperature thresholds that take the day of the year- and location-specific distributions of temperatures into account, as well as for indicators of heatwaves. Third, we measure heat exposure at the supplier location level using 1.1 million addresses of sites, branches, and subsidiaries from Orbis. Although this data does not include information on the nature, scope, and time variation in activity across locations, Table A.3a shows consistent significantly negative estimates. To rule out that the result is mechanical, we reestimate the tests on a sample of years during which the relationships between customers and suppliers are reportedly inactive. With the exception of one weakly positive coefficient, the results in Table A.3b show insignificant estimates. Fourth, we describe cross-sectional differences in Table A.4. Consistent with the literature, we find significantly stronger effects of heat for suppliers in the agricultural, mining, and construction sectors. We also find a significant moderating effect of industry competitiveness and an exacerbating effect of customer reliance on supplier inputs on the propagation of the effects of hot days, in line with the idea that customers’ ability to switch suppliers or low supplier bargaining power mitigate adverse effects. Fifth, we examine other firm-level outcomes in Table A.5. We find decreases in revenues over assets, revenues over employees, revenue growth, operating margins, purchase volumes (customer accounts payable and COGS) and customer inventory.

To conclude these motivating tests, we assess our results in the context of Barrot and Sauvagnat (2016) and Carvalho et al. (2021). For the tests, we adjust heat days for the number of days on which suppliers’ countries were exposed to natural disasters recorded in EM-DAT—heatwaves, droughts, and fires. In Table A.6a, we find similar results in magnitude and significance when we estimate the effects for heat days which did not coincide with disasters. Table A.6b shows the indirect effects on customer operating performance. Again, excluding heat-related disasters does not affect the magnitude or significance of the results. This finding points to a key difference between heat and other types of climate-change-related hazards: heat is much less salient, and may affect firm performance through mechanisms that are unrelated to large-scale destruction.