Journal of Business Finance & Accounting
Volume 51, Issue 9-10 pp. 2848-2883
ARTICLE
Open Access

Risk management and private debt contracts: The role of weather derivatives

Viet Do, 

Viet Do

Department of Banking and Finance, Monash University Scenic Blvd, Clayton, Victoria, Australia

Search for more papers by this author
Thu Ha Nguyen, 

Corresponding Author

Thu Ha Nguyen

Department of Banking and Finance, Monash University, Caulfield East, Victoria, Australia

Correspondence

Thu Ha Nguyen, Department of Banking and Finance, Monash University, 900 Dandenong Road, Caulfield East, VIC 3145, Australia.

Email: [email protected]

Search for more papers by this author
Tram Vu, 

Tram Vu

Department of Banking and Finance, Monash University, Caulfield East, Victoria, Australia

Search for more papers by this author
First published: 02 April 2024
https://doi.org/10.1111/jbfa.12800

Abstract

Using energy firm data and the 1997 introduction of weather derivatives as a natural experiment, we document an average 21-basis-point interest reduction in bank loans after borrowers hedge with weather derivatives. This saving increases among borrowers with higher risk or less complex financial reports, and during more uncertain market conditions or when investors pay more attention to climate risks. Our results are robust to endogeneity-corrected methods. Hedging firms are more willing to pledge collateral, accept stricter covenants and exhibit lower risks and a lower likelihood of covenant violations within 1 year following loan origination. We also find hedging firms have lower bond yields and a lower bank debt ratio, indicating that the benefits from hedging with weather derivatives extend to the public debt market. Overall, our findings demonstrate important financial implications of hedging using weather derivatives.

The use of derivative markets for hedging climate-related risk has been around for over 25 years. These instruments … are used by a wide variety of agricultural, energy, and financial-based entities from around the globe to help manage localized exposure to weather-related impact risk.

― Dominic Sutton-Vermeulen

1 INTRODUCTION

Corporate risk management is pivotal to the overall business operations. In a frictionless market setting, different risk management strategies should not affect firm value (Modigliani and Miller, 1958). It simply moves a firm to a different risk–return combination, which then suits different clientele. However, theoretical models accounting for market imperfections suggest that hedging helps reduce the variability of future earnings, risk of financial distress and costs of underinvestment or overinvestment (Mayers and Smith, 1982; Smith and Stulz, 1985; Bessembinder, 1991; Froot et al., 1993; Morellec and Smith, 2007). A large strand of literature explores the importance of hedging using derivatives and supports that hedging can increase firm value through its impact on growth opportunities, tax shields, agency costs and probability of financial distress (Géczy et al., 1997; Allayannis and Weston, 2001; Guay and Kothari, 2003; Bartram et al., 2011; Allayannis et al., 2012; Gilje and Taillard, 2017; Guo et al., 2021; Hahnenstein et al., 2021). Despite the relevance of hedging, little is known about its causal impact on external financing. This is mainly due to the challenge encountered in previous studies to identify an exogenous shock in firms’ hedging decisions, given that hedging decisions are often endogenized within the firms themselves.

Our study examines the effect of corporate hedging on the price and non-price terms of bank loans using the introduction of weather derivatives contracts (i.e., contracts whose payoffs are determined by weather-related metrics) in 1997 as a natural experiment. The introduction of weather derivatives in 1997, being exogenous to weather-sensitive firms’ risk management policies, provides an ideal setting to examine the causal effect of hedging on the cost of bank loans. Another advantage of our empirical setting is related to the low level of information asymmetry between parties in weather derivatives contracts, as it is unlikely that one energy firm possesses better weather forecasting abilities than its counterpart. This unique characteristic of weather derivatives contracts, coupled with energy firms’ high exposure to weather changes, enables us to ascertain the use of weather derivatives for hedging purposes.,

Our study engages with two streams of literature—one on the relations between risk management strategies and corporate debt contract terms and the other on the use and pricing of weather derivatives. In the first strand, past studies have explored the effects of hedging with financial derivatives on loan spreads (Norden and Wagner, 2008; Azam et al., 2022) and cost of public debt (Deng et al., 2017). Campello et al. (2011) document that firms’ hedging via interest rate and currency derivatives helps ease their access to credit as well as reduce the cost of loans and capital expenditure restrictions in their private debt arrangements. Meanwhile, Chen and King (2014) find firms that use currency, interest rate and commodity derivatives to benefit from paying a lower cost for their public bond issues. Theoretically, Rampini and Viswanathan (2010) establish a connection between collateral requirement, risk management and debt capacity in their dynamic model. Using micro-credit data, Tchakoute-Tchuigoua (2012) studies the role of active risk management in determining loan contract terms. Berg (2015) analyzes the causal effect of risk management on loan quality proxied by loan default rates, while Lee and Choi (2021) document a lower cost of debt associated with carbon risk management in South Korean firms. Huang et al. (2022) demonstrate that borrowers more exposed to climate risk face more restrictive loan terms and specific measures adopted to reduce climate risk exposure can help negate such impacts. While our study is close in spirit to Campello et al. (2011), Chen and King (2014) and Huang et al. (2022), it is distinct in several aspects. First, we focus on the use of weather derivatives in hedging weather risk—the new type of risk that has garnered significant research interest in recent years (see, for example, Weagley, 2019; Addoum et al., 2020 and 2023; Matsumoto and Yamada, 2021). Second, our empirical setting offers two clear advantages thanks to (1) the exogeneity of the 1997 introduction of weather derivatives and (2) the low level of information asymmetry in these contracts.

In the second literature strand, most prior studies have focused on the pricing, utilization and benefits of weather derivatives as risk management tools against adverse weather conditions in the energy industry. For instance, Matsumoto and Yamada (2021) examine how simultaneous hedging strategies for both electricity price and volume risk can stabilize profit or loss for electricity businesses. Härdle and Osipenko (2012) provide insights into how variations in weather conditions across different regions impact the risk premium associated with weather derivatives and how energy companies can strategically employ these derivatives to mitigate potential losses caused by unfavorable weather conditions. Yang et al. (2009) and Pérez-González and Yun (2013) demonstrate that the use of weather derivatives in hedging against weather-based risks helps increase energy firms’ value. Our study adds to this literature by exploring the implications of weather derivatives usage to borrowing firms in the form of loan costs and non-price contract terms.

Our supposition for the effect of weather derivatives hedging is primarily anchored in the fundamental theoretical benefits of traditional hedging. On the one hand, hedging helps reduce the probability and expected cost of financial distress, which subsequently lowers overall financing costs (Smith and Stulz, 1985; Huberman, 1997). In particular, this can manifest in the form of lower cost of debt (Norden and Wagner, 2008; Campello et al., 2011; Deng et al., 2017; Azam et al., 2022). On the other hand, weather derivatives, as a distinct asset class, have been shown to be an effective hedging tool that improves firm performance and firm value (Yang et al., 2009; Härdle and Osipenko, 2012; Pérez-González and Yun, 2013; Matsumoto and Yamada, 2021). Drawing from these various strands of literature, we develop the following hypothesis: All else being equal, borrowers who engage in weather derivatives hedging enjoy a lower borrowing cost.

Using data of 2091 bank loans issued to 144 electricity and natural gas firms during the 1986−2017 period, we find an average loan spread reduction of 21 basis points (bps) for borrowers who perform weather derivatives hedging. The effect of hedging with weather derivatives on the cost of loans remains robust across endogeneity-corrected regression methods, including propensity score matching, entropy balancing and instrumental variable (IV) approach. Our results remain consistent across different robustness tests, including sub-sample analyses and employing an alternative proxy for weather derivatives usage.

The impact of weather derivatives usage on loan costs is more pronounced among firms with higher risk and during more uncertain market or regulatory conditions. Borrowing cost savings associated with weather derivatives usage also increase when firms produce better-quality financial reports and when climate risks receive stronger investor attention. Our findings are also indicative of a potential trade-off between loan price and non-price terms in loan contracts originated after borrowing firms adopt weather derivatives. Loans originated after weather derivatives usage are more likely to be secured and carry more covenant restrictions. These firms appear to be more willing to pledge collateral and accept more restrictive covenants to signal their high quality in exchange for a lower loan price.

Beyond the private debt market, we also provide evidence that hedging with weather derivatives could benefit firms in the public debt market. Our results indicate that hedging firms enjoy lower bond yields at issuance and have better access to the public debt market, that is, become less reliant on bank loans.

We conduct ex-post analyses to explore if hedging using weather derivatives helps reduce firm risk and the likelihood of covenant violations. Our results suggest that firms hedging using weather derivatives are likely to experience lower volatility in return on assets (ROA) and ROE and higher Z-scores (indicating lower risk) afterwards. This result could be explained by weather derivative contracts acting as an effective risk management tool for these borrowers. We find that despite having to comply with more restrictive covenants, firms using weather derivatives are less likely to violate their covenants within 1 year of loan origination.

The contributions of our study are threefold. First, our study adds to the growing literature on the financial implications of climate risk exposure including the impact on firms’ financing policies (Huang et al., 2018; Javadi et al., 2023) as well as their financing costs (Huynh et al., 2020; Do et al., 2021; Javadi and Masum, 2021; Huang et al., 2022). Our findings suggest that the use of weather derivatives to hedge unexpected weather-induced variability in energy demand allows energy firms to smooth their future earnings, which should result in a lower borrowing cost. Consistent with previous studies, we show that reduction in bankruptcy risk is the principal channel for hedging to reduce financing costs (see, Smith and Stulz, 1985; Chen and King, 2014; among others). In addition, we report another avenue through which hedging may lower loan costs, that is, trade-off among loan terms. Upon hedging with weather derivatives, firms are more willing to pledge collateral and accept more restrictive covenants in exchange for a lower loan price. Furthermore, risk management through weather derivatives contracts appears to be effective as ex-post covenant violations become less likely following firms’ weather derivatives usage. Since renegotiation can be costly following covenant violations, this could represent another reason why hedging may add value to a firm.

Second, we investigate the use of weather derivatives in hedging weather risk—the type of risk that has drawn growing attention in recent years. We focus on energy firms owing to their great exposure to weather risk, that is, cost fluctuation (price risk) and unpredictable future sales (volume risk) associated with changing temperatures. While the hedging literature has largely focused on price risk, volume risk has received little attention due to information asymmetry concerns that prevent the development of quantity-based insurance contracts (Pérez-González and Yun, 2013). Weather derivatives are among very few products that are not subject to the information asymmetry issue and can provide a direct hedge on product demand uncertainty for electricity and natural gas firms (Brockett et al., 2005; Pérez-González and Yun, 2013). Our work adds to the growing literature that investigates how various stakeholders, including investors, banks, firm managers and fund managers, have responded to this new layer of risk in their business and investment decisions (Addoum et al., 2020; Choi et al., 2020; Huynh et al., 2020; Do et al., 2021; Du et al., 2023). In the context of weather and climate risks, we explore how weather derivatives, a new class of derivative products, provide weather-sensitive firms with a direct hedging tool to manage demand uncertainty, and how this is reflected in their financing costs. Given the fairly recent introduction of these weather derivatives contracts and their unique features, it is crucial to understand their added values, not only to the user firms but also to its various stakeholders including creditors.

Third, our study utilizes a unique setting on energy firms, in contrast to the vast majority of the debt contracting literature that routinely excludes loans for financial and energy firms. Energy firms often adopt an array of risk management tools, including diversification across geographical regions and product lines, use of forward contracts on output sales, lower leverage, higher cash holdings and weather normalization adjustments. These strategies come with their strengths and weaknesses, as detailed in Pérez-González and Yun (2013). The adoption of weather derivatives introduced in 1997 has enabled energy firms to overcome most shortcomings of the aforementioned risk management strategies, since these products can serve as a direct hedge on energy demand. Energy companies are among the most prominent users of weather derivatives. A survey by the Weather Risk Management Association documented that energy companies accounted for 69% of over-the-counter (OTC) weather derivative end users in 2004–2005 (Weather Risk Management Association, 2006). Their demand to use weather derivatives as hedging tools was evidenced in Weagley (2019), who finds 87% of the sample energy firms have revenues positively correlated with consumers’ heating demand and 65% have revenues positively correlated with cooling demand.

The remainder of the paper is organized as follows. Section 2 provides a background on hedging activities among energy firms and their weather derivatives usage. Section 3 details the data sources, sample selection and descriptive statistics. Sections 4 and 5 present our empirical results on the impact of weather derivatives usage on loan price across various regression methods and analyses. Section 6 examines the impact of weather derivatives usage on collateral and covenants. Section 7 presents evidence on how hedging with weather derivatives affects a firm's bond yields and its debt structure. The ex-post analyses of weather derivatives usage on firm risks and covenant violations are detailed in Section 8. Section 9 provides concluding remarks.

2 HEDGING USING WEATHER DERIVATIVES IN ENERGY FIRMS

Energy firms, like any other corporate entity, are presented with various risk management strategies, including financial instruments (i.e., hedging with derivatives) and non-financial approaches (i.e., diversification, product bundling, investments in tangible assets, lengthened contract terms, etc.). Among those approaches, the usage of derivatives (for hedging purposes), being standardized and easily obtained, is widely used in the literature as an indicator of hedging activities (Allayannis and Weston, 2001; Campello et al., 2011; Lievenbrück and Schmid, 2014).

As the energy sector is among those that are most sensitive to weather changes (National Research Council, 2003), our focus on this group of firms enables us to examine the value of hedging using a new type of derivatives—weather derivatives. Weather derivatives are derivatives contracts whose payoffs are determined by weather-related metrics. Notably, this asset class allows us to examine a layer of risk that is relatively unexplored, which is volume risk or the risk of unpredictable future sales. This type of risk is especially important for energy firms whose revenues decline during periods of mild temperature as individuals and businesses have lower heating or cooling demand (Paravan et al., 2004).

In the United States, the winter-month weather contracts are based on heating degree days (HDD), whereas the summer-month weather contracts are based on cooling degree days (CDD). Mathematically, these concepts can be expressed as follows:
HDD = max ( 0 , 65 ∘ F − daily average temperature ) $$\begin{equation*}{\mathrm{HDD}} = {\mathrm{max}}\;(0,\;65^\circ {\mathrm{F}} - {\mathrm{daily}}\;{\mathrm{average}}\;{\mathrm{temperature}})\end{equation*}$$
CDD = max ( 0 , daily average temperature − 65 ∘ F ) $$\begin{equation*}{\mathrm{CDD}} = {\mathrm{max}}\;(0,\;{\mathrm{daily}}\;{\mathrm{average}}\;{\mathrm{temperature}} - 65^\circ {\mathrm{F}})\end{equation*}$$

Energy companies have been known as the sellers of HDD or CDD futures contracts, and the buyers of HDD or CDD put options. Their derivatives positions are profitable when weather conditions are mild, which are then used to offset the shortfall caused by reduced energy consumption. The correlation between temperature and energy consumption quantity is directly addressed by weather derivatives (Cao and Wei, 2004; Pollard et al., 2008).

Energy firms are also exposed to price risk due to rising input costs, typically during periods of extreme temperatures when energy demand is high. However, such price risk can be partially mitigated through alternative channels such as passing on rising costs to customers (Pérez-González and Yun, 2013), relying on the state-level purchased gas adjustments (American Gas Association, 2007) and using traditional commodity derivatives (Müller and Grandi, 2000). While temporary price fluctuations (i.e., a spike in input costs) can arise from short-term extreme temperatures, the energy industry is more exposed to mild temperature shocks. Weagley (2019) reports that 80% of Compustat energy utilities have lower revenues when temperatures are less extreme in the 1977–1996 period. Addoum et al. (2023) document that energy firms’ earnings are hurt by both extremely warm winters and extremely cool summers. Brockett et al. (2005) also highlight that the wide array of instruments used for hedging price risks have limited applicability for hedging volume risk. As a result, the investigation of volume hedging with weather derivatives helps provide a more cohesive understanding of risk management practices within this unique group of highly regulated energy firms.

Pollard et al. (2008) emphasize three main reasons why weather derivatives are superior to traditional weather insurance contracts in protecting firms from unexpected weather changes. First, weather derivatives do not require firms to specify any insurable interest. Second, these firms do not need to prove that their losses directly result from a particular weather event. Third, while standard insurance contracts target high-risk low-probability (i.e., extreme) weather episodes, weather derivatives can compensate firms against low-risk high-probability events, for example, seasons or periods that are warmer, colder, drier or wetter, than usual.

Standard practice in finance research often excludes energy firms from the study sample on the grounds of regulatory differences potentially leading to biased results. However, as argued in Lievenbrück and Schmid (2014), the main facets of energy regulation are not designed to influence these firms’ hedging decisions. Hence, our findings for this group of firms should not be materially biased by its regulatory environment and can be generalized to a broader context involving the systematic drivers of corporate risk management.

3 DATA AND SAMPLE

3.1 Data sources and filters

We obtain loan-level contract data from the Loan Pricing Corporation's DealScan database. DealScan provides information for each loan contract at origination, including loan amount, purpose, facility type, interest rate spread, fees, maturity, collateral and covenant restrictions. Each loan is matched with its borrower's most recent (prior to the loan date) financial characteristics using the link file provided by Chava and Roberts (2008), which offers matches to syndicated loans between 1982 and 2017. We limit our study to loans borrowed by Compustat firms operating in electricity and natural gas industries, namely, energy firms (Standard Industrial Classification [SIC] codes 4911, 4923, 4924, 4931 and 4932). The focus on energy firms provides us with a clean identification of hedging among those firms.

We collect data on energy firms’ weather derivatives usage from the Securities and Exchange Commission (SEC) 10-K filings. Specifically, we start with the index file that contains all 10-K filing accession numbers of electricity and natural gas firms, obtained from WRDS SEC filings index data provided by the Wharton Research Data Services. Once all filings are downloaded, their contents are parsed using a list of keywords (including “weather hedging,” “weather derivatives,” “weather forwards,” “weather futures,” “weather options” and “weather swaps”) to identify the first time these firms use weather derivatives contracts in the post-1997 period. Following Pérez-González and Yun (2013), we classify firms as weather derivatives users if they describe such contracts in their 10-K filings submitted to the SEC. The firm-level weather derivatives usage indicator is then merged with our loan data to create a final sample of 2091 loan facilities originated between 1986 and 2017.

3.2 Descriptive statistics

Table 1 presents summary statistics for the key loan and borrower characteristics for the final loan sample in our study. We winsorize our variables at the 1 and 99% levels to remove extreme outliers. In the table, we report mean and standard deviation of all variables for the entire sample, and the two sub-samples that consist of loans originated to weather derivatives users and those to weather derivatives non-users. The last column of the table presents the mean differences in loan and borrower characteristics between weather derivatives users and non-users.

TABLE 1. Summary statistics.
All loans (N = 2091) Loans to weather derivatives users (N = 603) Loans to weather derivatives non-users (N = 1488)
Variable Mean SD Mean (A) SD Mean (B) SD Difference (A − B)
Loan characteristics
AISD (bps) 111.499 91.562 119.152 104.247 108.398 85.734 10.754
Maturity (months) 37.879 23.910 37.970 23.427 37.841 24.111 0.129
Facility Amount ($m) 578 886 619 887 562 885 57
Secured Dummy 0.122 0.327 0.172 0.378 0.101 0.302 0.071
Covenant Strict 0.093 0.291 0.131 0.338 0.078 0.268 0.053
Covenant Index 0.917 1.362 1.068 1.623 0.856 1.236 0.212
Revolver Dummy 0.596 0.491 0.580 0.494 0.603 0.489 −0.023
Lender Number 10.498 7.614 11.098 7.513 10.255 7.644 0.843
Borrower characteristics
Total Assets ($m adj) 5445 5481 7113 5811 4769 5193 2344
ROA 0.028 0.026 0.026 0.025 0.028 0.027 −0.002
Leverage 0.372 0.089 0.374 0.081 0.371 0.092 0.003
Current Ratio 0.909 0.383 0.894 0.340 0.915 0.399 −0.021
Interest Coverage 1.688 0.372 1.659 0.350 1.700 0.380 −0.041
Profitability 0.264 0.101 0.260 0.097 0.265 0.103 −0.005
Tangibility 1.026 0.225 0.981 0.211 1.045 0.228 −0.064
  • Note: This table presents the mean and standard deviation for loan and borrower characteristics for the final sample of 2091 loan facilities (of 144 firms) originated between 1986 and 2017. The final sample is also divided between loans to weather derivatives users (603) and loans to weather derivatives non-users (1488). All-in-Spread-Drawn (AISD) is the interest rate margin over LIBOR on the drawn loan amount plus annual fees. Maturity is length in number of months between the loan's start date and its maturity date. Facility Amount is the dollar amount of loan facility in US$ million. Secured Dummy is a binary variable taking the value of 1 if a loan has collateral and zero otherwise. Covenant Index follows Bradley and Roberts (2015)’s count variable, which takes a value between 0 and 6 based on 6 key covenant categories. Covenant Strict is a dummy variable calculated based on Bradley and Roberts (2015)’s Covenant Index and equals to 1 for loans with three or more covenant categories and zero otherwise. Revolver Dummy is a binary variable taking the value of 1 if the loan facility is a revolving facility and zero otherwise. Lender Number is the number of lenders in a loan. Total Assets is the borrower's book value of total assets in US$ million, adjusted for inflation in year 2000 dollars. ROA is the return on assets calculated as income after depreciation, divided by total assets. Leverage is calculated as long-term debt plus current liabilities, divided by book value of total assets. Current Ratio is the ratio of current assets to current liabilities. Interest Coverage is the ratio of EBITDA to interest expenses. Profitability is the ratio of EBITDA to sales. Tangibility is the ratio of property, plant and equipment to total assets. All the values are winsorized at the 1 and 99% levels. ***, ** and * represent significance at the 1, 5 and 10% level, respectively.

Overall, our final sample includes 2091 commercial loans issued to 144 US electricity and natural gas firms between 1986 and 2017. The average All-in-spread-drawn (AISD), loan maturity and facility amount are roughly 111.5 bps, 38 months and US$578 million, respectively. Secured loans account for 12.2% of our sample, whereas the average proportion of revolving loans is roughly 60%. The average Covenant Strict is 9.3%, indicating that 9.3% of the sample loans have three (3) or more covenant categories. On average, more than ten (10) lenders are involved in each syndicated loan. When compared with statistics for loan characteristics documented in the literature (see Bharath et al., 2011; Amiram et al., 2017; among others), the aforementioned statistics indicate that on average, loans to electricity and natural gas firms carry a lower spread, larger amount and shorter maturity than loans to other industries.

In terms of borrower characteristics, the average borrower's asset size in constant 2000 dollars is US$5.445 billion, with a mean ROA of 2.8% per annum. Among 2091 loans in the final sample, 603 loans (or 28.84%) are given to weather derivatives users. The univariate tests of mean differences between weather derivatives users and non-users indicate that loans to weather derivatives users are granted at a higher spread, on a more secured basis, and involve stricter covenants than those to weather derivatives non-users. Interestingly, there is no statistical difference in loan maturity and facility amount between the two groups. Regarding borrower characteristics, firms that use weather derivatives are of larger asset size, lower ROA, lower interest coverage and lower tangibility than those that do not use weather derivatives. The differences are statistically significant at the 1 or 5% levels.

4 MULTIVARIATE ANALYSES

4.1 Baseline regression

We start our main empirical analysis by examining if weather derivatives users enjoy more favorable loan price (i.e., a lower cost of loans) upon adoption of weather derivatives contracts. We estimate the following regression model to investigate the impact of weather derivatives usage onset on loan pricing:
LoanSprea d l i t = α 1 + β 1 WD _ Pos t l i t + γ Control s l i t + θ i D i + θ t D t + ε l i t $$\begin{equation}{\mathrm{LoanSprea}}{{{\mathrm{d}}}_{lit}} = {{\alpha }_1} + {{\beta }_1}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{lit}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{lit}} + {{\theta }_i}{{D}_i} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{lit}}\end{equation}$$ (1)
where l, i and t denote loans, firms and years, respectively. LoanSpread is proxied by (i) AISD, which represents the interest rate margin over the London Interbank Offered Rate (LIBOR) plus annual fee or (ii) Log(AISD), which is the natural logarithm of AISD. In all future tests on loan spread, we use Log(AISD) as the dependent variable to take into account the skewness of loan spread data following prior studies (e.g., Graham et al., 2008; Chava et al., 2009; Amiram et al., 2017).

WD_Post is our variable of interest, which is a binary variable taking a value of 1 for loans made to weather derivatives users after the first time they engage in weather derivatives contracts, and zero otherwise. In light of Campello et al. (2011) and Chen and King (2014), we hypothesize that hedging using weather derivatives can help reduce financing costs. In that case, weather derivatives usage is hypothesized to be associated with a decrease in loan spreads; therefore, we should observe a negative and statistically significant β 1 ${{\beta }_1}$ . Controls refer to the set of borrower and loan characteristics. Borrower characteristics include ROA, Log(Assets), Leverage, Current Ratio, Log(1+Coverage), Profitability and Tangibility, whereas loan characteristics include Log(Facility Amount), Log(Maturity), Secured Dummy, Covenant Strict, Revolver Dummy and Lender Number (for details, refer to the Appendix). Our regressions include borrower and year fixed effects and also control for credit rating and loan purpose. Standard errors are adjusted for heteroscedasticity and clustered at the borrower level.

The estimation output for Equation (1) is reported in Table 2, where columns (1) and (2) present results when AISD and Log(AISD) are used as dependent variables, respectively. In both columns, the coefficient of WD_Post is negative and statistically significant at the 5% level. This negative and significant coefficient is consistent with our hypothesis that banks charge lower loan spreads to energy borrowers after they use weather derivatives. This implies that when energy borrowers are engaged in weather derivatives contracts, which are most likely for hedging purposes (Pérez-González and Yun, 2013; Bartram, 2019), banks perceive this practice as effective risk management, hence lower the required loan spread.

TABLE 2. Effect of weather derivatives usage on loan price.
Dependent variable AISD Log(AISD)
(1) (2)
WD_Post −27.172 −0.212
(12.45) (0.09)
Borrower characteristics
ROA 129.208 0.976
(107.51) (0.77)
Log(Assets) 20.728 0.088
(8.62) (0.06)
Leverage −43.659 −.012
(46.48) (0.24)
Current Ratio −0.709 0.072
(5.23) (0.04)
Log(1+Coverage) −2.958 −0.072
(12.60) (0.08)
Profitability −106.691 −0.574
(38.23) (0.23)
Tangibility −25.728 −0.213
(17.67) (0.13)
Loan characteristics
Log(Facility Amount) −5.104 −0.055
(2.93) (0.02)
Log(Maturity) 13.021 0.045
(3.67) (0.02)
Secured Dummy 50.406 0.291
(12.09) (0.08)
Covenant Strict 45.231 0.237
(10.28) (0.05)
Revolver Dummy −12.602 −0.001
(5.91) (0.04)
Lender Number −1.076 −0.004
(0.39) (0.00)
Firm FE Yes Yes
Year FE Yes Yes
Credit rating Yes Yes
Loan purpose Yes Yes
Observations 2091 2091
Adjusted R2 0.616 0.728
  • Note: This table presents the OLS regression output for the effect of weather derivatives usage on the cost of bank loans using the following model: LoanSprea d l i t = α 1 + β 1 WD _ Pos t l i t + γ Control s l i t + θ i D i + θ t D t + ε l i t ${\mathrm{LoanSprea}}{{{\mathrm{d}}}_{lit}} = {{\alpha }_1}\ + {{\beta }_1}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{lit}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{lit}} + {{\theta }_i}{{D}_i} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{lit}}$ . We use two proxies for the cost of bank loans (i.e., LoanSpread) as the dependent variable. AISD in column (1) is the interest rate margin over LIBOR on the drawn loan amount plus annual fees, while Log(AISD) in column (2) is the natural logarithm of AISD. The key explanatory variable WD_Post is an indicator variable taking the value of 1 for loans made to weather derivatives user firms after they first enter weather derivatives contracts, and zero otherwise. The regressions include all borrower and loan characteristics as controls. All variables are defined in the Appendix. The numbers in parentheses are standard errors corrected for heteroscedasticity and clustered at the firm level. ***, ** and * represent significance at the 1, 5 and 10% level, respectively.

The magnitude of the coefficient on WD_Post is also economically significant. The coefficient of WD_Post of −27.172 in column (1) suggests that the initiation of using weather derivatives leads to a drop of 27.172 bps in AISD. Similarly, the coefficient of WD_Post of −0.212 in column (2) suggests that Log(AISD) drops by 0.212 following the onset of firms using weather derivatives contracts. This is equivalent to a drop of approximately 19.1% [= exp(−0.212)−1] in AISD. For an average loan spread of 111.5 bps in our sample, the cost saving equals 21.3 bps [= 111.5 × 19.1%] in the loan spread following weather derivatives usage. Given the average loan size of US$578 million, this translates to an annual interest cost saving of US$1.23 million [= $578 million × 0.213%].

Signs of control variables’ coefficients, especially those that are statistically significant, are largely consistent with the previous literature. The significant and negative coefficients of Log(Facility Amount), Revolver Dummy and Lender Number in column (1) indicate that larger loans, revolving loans and loans with more lenders in the syndicate enjoy a lower spread. We also document a higher loan spread in secured loans, loans with longer maturity and those with stricter covenants. The relationships, as explained in the literature, reflect interdependencies among contract terms that are driven by information asymmetries, contracting costs and credit risk (Dennis et al., 2000). Borrowers that are riskier than average tend to pledge collateral and accept stricter covenants (Berger and Udell, 1990; Dennis et al., 2000). Among the borrower characteristic variables, only Profitability appears to be statistically significant in both columns of Table 2. The coefficients of Profitability are negative in both columns, indicating that more profitable firms pay lower borrowing costs, which is consistent with the bank lending literature.

4.2 Endogeneity of weather derivatives usage

A key assumption underlying an unbiased OLS estimation of Equation (1) is that the distribution and timing of weather derivatives usage is a random factor. The decision for energy firms to adopt hedging via weather derivatives contracts, however, could be driven by fundamental firm characteristics, market conditions, weather conditions and/or other unobservable factors. Hence, the estimated coefficient WD_Post from the previous section might be endogenously biased. We follow prior literature on loan contract terms (Bharath et al., 2011; Cen et al., 2016; Amiram et al., 2017; Shan et al., 2019) to address the endogeneity bias via a matched sample analysis and an IV approach.

4.2.1 Matched sample approach

We perform both propensity score matching and entropy balancing in our matched sample analysis. Our approach to propensity score matching is similar to that of Chen et al. (2020). The aim is to create two sets of loans from similar borrowers, with the only difference being one hedges with weather derivatives and the other does not. We estimate a Probit model of weather derivatives usage indicator using borrower characteristics in our baseline regression. From the Probit model, we obtain the propensity score of loans made to weather derivatives users and identify the most similar non-user loans using the 1-to-1 nearest neighbor propensity score matching method (without replacement). We are able to match 344 loan facilities made to weather derivatives users with 344 loan facilities made to non-users.

We conduct a series of diagnostic tests to examine the quality of our matching pairs. In Panel A of Table 3, we compare the distributions of propensity scores for loan facilities made to weather derivatives users and those made to non-users. After the matching process, the distributions of these two groups are almost identical. This result indicates that the 1-to-1 nearest neighbor propensity score matching process is effective in identifying matched pairs with almost identical propensity scores of weather derivatives usage.

TABLE 3. Matched sample analysis.
Panel A: Summary statistics of the propensity scores for the matched sample
N Mean SD p25 Median p75
Loans to weather derivatives users 344 0.26 0.11 0.19 0.26 0.35
Loans to weather derivatives non-users 344 0.27 0.11 0.19 0.26 0.35
Difference in statistics −0.01 0.00 0.00 0.00 0.00
Panel B: Univariate statistics of the variables for the propensity matched sample
Loans to weather derivatives users Loans to weather derivatives non-users Mean difference p Value
Variable N Mean (A) N Mean (B) (B − A)
AISD (bps) 315 132.405 309 133.194 0.789 0.919
Maturity (months) 341 43.809 337 42.300 −1.510 0.377
Facility Amount ($m) 344 799.00 344 641.00 −159.00 0.025
Secured Dummy 344 0.099 344 0.186 0.087 0.001
Covenant Strict 344 0.052 344 0.096 0.044 0.029
Revolver Dummy 344 0.680 344 0.634 −0.047 0.199
Lender Number 343 11.431 344 11.875 0.444 0.431
Total Assets ($m adj) 344 7277.4 344 7234.5 −42.95 0.927
ROA 344 0.023 344 0.024 0.001 0.557
Leverage 344 0.347 344 0.353 0.006 0.355
Current Ratio 344 0.946 344 0.971 0.026 0.380
Interest Coverage 344 1.683 344 1.683 0.001 0.986
Profitability 344 0.249 344 0.252 0.003 0.705
Tangibility 344 0.941 344 0.930 −0.011 0.484
Panel C: Baseline regression using the propensity matched sample
Dependent variable AISD Log(AISD)
(1) (2)
WD_Post −80.620 −0.229
(38.83) (0.11)
Controls Yes Yes
Firm FE Yes Yes
Year FE Yes Yes
Credit rating Yes Yes
Loan purpose Yes Yes
Observations 599 599
Adjusted R2 0.712 0.758
Panel D: Baseline regression using the entropy balanced sample
Dependent variable AISD Log(AISD)
(1) (2)
WD_Post −78.058 −0.213
(40.43) (0.11)
Controls Yes Yes
Firm FE Yes Yes
Year FE Yes Yes
Credit rating Yes Yes
Loan purpose Yes Yes
Observations 599 599
Adjusted R2 0.712 0.755
  • Note: Panel A of this table presents the summary statistics for the propensity scores of loans made to weather derivatives users and their matched counterparts to non-users. The propensity score is the probability of firms using weather derivatives for each loan, generated from the estimation of a Probit model on borrower characteristics. Each loan facility made to weather derivatives users is matched to another loan made to non-users using the 1-to-1 nearest neighbor propensity scores matching method (without replacement). Panel B presents the univariate statistics of firm and loan characteristics for the propensity matched sample. Panel C presents the OLS regression output for the effect of weather derivatives usage on AISD and Log(AISD) using the propensity matched sample. Panel D presents the OLS regression output for the effect of weather derivatives usage on AISD and Log(AISD) using the entropy balanced sample. The model used for both Panel C and Panel D is as follows: LoanSprea d l i t = α 1 + β 1 WD _ Pos t l i t + γ Control s l i t + θ i D i + θ t D t + ε l i t ${\mathrm{LoanSprea}}{{{\mathrm{d}}}_{lit}} = {{\alpha }_1}\ + {{\beta }_1}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{lit}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{lit}} + {{\theta }_i}{{D}_i} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{lit}}$ . The dependent variable AISD in column (1) is the interest rate margin over LIBOR on the drawn loan amount plus annual fees, while Log(AISD) in column (2) is the natural logarithm of AISD. The key explanatory variable WD_Post is an indicator variable taking the value of 1 for loans made to weather derivatives user firms after they first enter weather derivatives contracts, and zero otherwise. The regressions include all borrower and loan characteristics as controls. All variables are defined in the Appendix. The numbers in parentheses are standard errors corrected for heteroscedasticity and clustered at the firm level. ***, ** and * represent significance at the 1, 5 and 10% level, respectively.

Panel B of Table 3 shows the univariate t-tests for the mean difference of borrower and loan characteristics between the two sets of loans. One of the indicators of a good matching process is that the two matched groups are not significantly different in their characteristics. Panel B shows that with the exception of collateral, covenant strictness and facility amount, all other loan and borrower characteristics are not statistically different between the two matched groups. Taken together, the statistics in Panels A and B of Table 3 indicate that our propensity score matching process is effective.

In Panel C of Table 3, we re-estimate our baseline regression using the propensity score matched sample. As we have shown in our diagnostic test, the two groups in our matched sample are almost identical in most aspects of loan and firm characteristics. Nonetheless, the coefficients of WD_Post continue to be negative and statistically significant at the 5% level. This confirms our baseline results and indicates that the effect of hedging with weather derivatives on loan cost remains robust even when our model is estimated on a matched sample.

In Panel D of Table 3, we re-estimate the regression using the entropy balanced sample to account for any residual differences between the treated and control group. Hainmueller (2012) introduces the entropy balancing approach with the aim to match the three moments of co-variates between treated and control firms. In our case, this approach will match the three moments for weather derivatives users and non-users. The result for the entropy balancing regression remains consistent with that of Panel C as well as the baseline regression albeit statistically significant at the 10% level.

4.2.2 IV approach

We also apply the IV approach to address the endogeneity concern of weather derivatives usage among electricity and natural gas firms. Following Pérez-González and Yun (2013), we construct an instrument called EDD Weather-Induced Volatility for each firm using pre-1997 weather and financial data. This instrument measures the sensitivity of a firm's revenues to weather fluctuations before the introduction of the weather derivatives market in 1997, hence it should influence the firm's decision to adopt weather derivatives to hedge their revenue streams once these contracts are introduced.

We construct EDD Weather-Induced Volatility using a two-step approach. In the first step, we estimate the following regression for each firm in our sample using the pre-1997 data:
Revasset s i t = α i + β i ED D i t + λ i Log Assets i t + ε i t $$\begin{equation}{\mathrm{Revasset}}{{{\mathrm{s}}}_{it}} = {{\alpha }_i} + {{\beta }_i}{\mathrm{ED}}{{{\mathrm{D}}}_{it}} + {{\lambda }_i}{\mathrm{Log}}{{\left( {{\mathrm{Assets}}} \right)}_{it}} + {{\varepsilon }_{it}}\end{equation}$$ (2)
where Revassets is the firm's revenue to assets ratio; EDD is the energy degree days for each firm based on its headquarter location; Log(Assets) is the natural logarithm of firm asset size. We obtain pre-1997 (1960−1996) monthly data for degree days, including CDD and HDD at the climate division level from the NOAA's National Centers for Environmental Information. Based on the CDD and HDD data, we calculate energy degree days (EDD) as the total of CDD and HDD and use it as a proxy for pre-1997 total energy demand. To be consistent with the data frequency of firms’ fundamentals, we aggregate EDD on a quarterly basis. Using the climate division shapefile provided by the NOAA, we identify the boundaries of all climate divisions, which then allows us to map the location of our sample firms to their corresponding climate divisions based on their Compustat headquarter's ZIP codes. We obtain the coefficient of EDD ( β i ${{\beta }_i}$ ) for each firm from estimating Equation (2). This coefficient measures the pre-1997 sensitivity of firm revenue to changes in energy demand.
In the second step, the EDD Weather-Induced Volatility for each firm is calculated as the absolute value of β i ${{\beta }_i}$ multiplied by standard deviation of EDD for the corresponding firm location.
EDD Weather − Induced Volatility = β i σ i EDD $$\begin{equation}{\mathrm{EDD}}\;{\mathrm{Weather}} - {\mathrm{Induced}}\;{\mathrm{Volatility}} = \left| {\ {{\beta }_i}} \right|{\mathrm{\ \ }}{{\sigma }_{{{i}_{{\mathrm{EDD}}}}}}\end{equation}$$ (3)

Panel A of Table 4 shows the descriptive statistics for the pre-1997 weather-related measures, including Absolute EDD Beta, EDD Standard Deviation and EDD Weather-Induced Volatility, for our sample energy firms. All statistics are similar to those reported by Pérez-González and Yun (2013). For example, the mean value for our instrument EDD Weather-Induced Volatility is 0.018 (compared with 0.022 in Pérez-González and Yun (2013)).

TABLE 4. Instrumental variable approach.
Panel A: Summary statistics for the pre-1997 weather-related measures
Obs Mean Std Dev Min Max
EDD Standard Deviation 225 0.089 0.039 0.012 0.169
Absolute EDD Beta 225 0.233 0.307 0.000 2.187
EDD Weather-Induced Volatility 225 0.018 0.030 0.000 0.277
Panel B: IV estimation
First-stage Probit at firm level Dep Var: WD_Post IV estimation on entire loan sample Dep Var: Log(AISD) IV estimation on matched loan sample Dep Var: Log(AISD)
(1) (2) (3)
WD_Post −0.448 −0.414
(0.18) (0.15)
EDD Weather-Induced Volatility 1.556
(0.08)
Controls Yes Yes Yes
Firm FE Yes Yes Yes
Year FE Yes Yes Yes
Credit rating No Yes Yes
Loan purpose No Yes Yes
Observations 3006 514 485
Centered R2 0.130 0.145
Pseudo R2 0.641
Panel C: IV estimation (adding CC_Exposure as a control variable in the first stage)
First-stage Probit at firm level Dep Var: WD_Post IV estimation on entire loan sample Dep Var: Log(AISD)
(1) (2)
WD_Post −0.340
(0.17)
EDD Weather-Induced Volatility 2.078  
(0.30)  
CC_Exposure 91.556  
(37.36)  
Controls Yes Yes
Firm FE Yes Yes
Year FE Yes Yes
Credit rating No Yes
Loan purpose No Yes
Observations 694 151
Centered R2 0.175
Pseudo R2 0.873
  • Note: This table presents the regression output for the effect of weather derivatives usage on the cost of bank loans using the instrumental variable (IV) approach. The dependent variable Log(AISD) is measured as the natural logarithm of the interest rate margin over LIBOR on the drawn loan amount plus annual fees. The key explanatory variable WD_Post is an indicator variable taking the value of 1 for loans made to weather derivatives user firms after they first enter weather derivatives contracts, and zero otherwise. The instrument for WD_Post is EDD Weather-Induced Volatility measured for the pre-1997 period, which equals Absolute EDD Beta multiplied by EDD Standard Deviation (EDD is energy degree days, a proxy for total energy demand) (Pérez-González and Yun, 2013). Panel A presents the summary statistics for the pre-1997 weather-related measures. Panel B presents the output for the first-stage Probit of WD_Post at firm level (column 1), and the second-stage IV regression on the entire loan sample (column 2) and on the propensity score matched sample (column 3). Panel C presents the IV regression output with the inclusion of CC_Exposure as a control in the first-stage regression. CC_Exposure is Sautner et al. (2023)’s firm-level climate risk exposure measure based on earnings conference calls. All variables are defined in the Appendix. The numbers in parentheses are standard errors corrected for heteroscedasticity and clustered at the firm level. ***, ** and * represent significance at the 1, 5 and 10% level, respectively.

To be a valid instrument, EDD Weather-Induced Volatility has to satisfy two conditions, (i) the relevance condition, that is, EDD Weather-Induced Volatility should be highly correlated with the likelihood of firms’ weather derivatives usage and (ii) the exogenous condition, that is, EDD Weather-Induced Volatility should not affect the cost of bank loans except through its impact on weather derivatives usage. Theoretically, firms whose historical revenues are more sensitive to weather fluctuations potentially gain more from hedging using weather derivatives, thus they are more likely to engage in these contracts upon their introduction in 1997 (Pérez-González and Yun, 2013). We test if our instrument satisfies the first condition in the first stage of our model. While there is no direct test for the second condition, we believe that the pre-1997 weather risk exposure is unlikely to directly impact the cost of bank loans after 1997 due to the significant time gap between the loan origination and the period during which our instrument is measured. In particular, over 70% of loans obtained by weather derivatives user firms in our sample were originated three (3) years or more after 1997. It is unlikely that the weather risk exposure measured during the 1960–1996 period would affect the cost of loans originated a number of years after 1997.

We perform the IV estimation using EDD Weather-Induced Volatility as an instrument for WD_Post and report the results in Panel B of Table 4. The first-stage Probit model estimates WD_Post as a function of firms’ asset size Log(Assets) and pre-1997 weather risk exposure EDD Weather-Induced Volatility at the firm level with firm and year fixed effects. Note that our dependent variable WD_Post equals zero for all firms prior to 1997 and one for weather derivatives user firms after 1997. Our first-stage sample has 3,006 firm-year observations for 225 energy firms whose EDD Weather-Induced Volatility can be measured. The output in column (1) indicates that a firm's historical weather risk exposure is positively correlated with its likelihood of weather derivatives usage, which is consistent with the finding in Pérez-González and Yun (2013), and indicates that our instrument satisfies the first validity condition.

Columns (2) and (3) of Panel B Table 4 show the results for the second stage of the IV estimation conducted at the loan level for the entire sample and the propensity-matched sample, respectively. Our sample size in columns (2) and (3) is smaller than that of the baseline regression as we only include loans to energy firms whose EDD Weather-Induced Volatility can be measured in the first stage. The coefficients of WD_Post in the two columns are −0.448 and −0.414, and significant at the 5 and 1% levels, respectively. This result further confirms the borrowing cost savings for energy firms following their adoption of weather derivatives contracts after accounting for the potential endogeneity bias of weather derivatives usage via the IV approach.

In Panel C, we control for firms’ contemporaneous climate risk exposure, CC_Exposure, in the first-stage regression model and re-estimate our IV regression. Specifically, we employ Sautner et al. (2023)’s measure of overall climate change exposure, which applies a machine learning keyword discovery algorithm using earnings conference calls to capture financial analysts’ attention to firms’ climate change exposures. As the data provided by Sautner et al. (2023) are only available from 2001 and cover fewer energy firms compared with our sample, the number of observations in our first-stage regression drops to 694 observations. In column (1), we obtain a positive and significant coefficient of CC_Exposure suggesting that firms with higher exposure to climate change are more likely to adopt weather derivatives hedging. Our instrument variable, EDD Weather-Induced Volatility, remains significant when CC_Exposure is included. The coefficient of WD_Post in column (2), albeit statistically and economically weaker, remains negative and significant. This supports our conclusion that energy firms enjoy lower borrowing costs following their adoption of weather derivatives hedging after accounting for potential endogeneity bias and including firms’ contemporaneous exposure to climate risk.

Overall, while an energy firm's decision to hedge its revenues with weather derivatives contracts is likely to be endogenous, we mitigate this concern with propensity score matching, entropy balancing and IV approach. Our results are consistent with the baseline estimation and continue to demonstrate a negative association between firms’ hedging using weather derivatives and their cost of loans.

4.3 Robustness tests

In this section, we conduct a number of robustness tests and present their regression output in Table 5.

TABLE 5. Effect of weather derivatives usage on loan price—Robustness tests.
Dependent variable: Log(AISD)
Baseline OLS Baseline OLS IV regression
Post-1997 WD_Post_Alt WD_Post_Alt
(1) (2) (3)
WD_Post −0.285
(0.09)
WD_Post_Alt −0.148 −0.480
(0.08) (0.19)
Controls Yes Yes Yes
Firm FE Yes Yes Yes
Year FE Yes Yes Yes
Credit rating Yes Yes Yes
Loan purpose Yes Yes Yes
Observations 1783 2091 514
Adjusted R2 0.720 0.727
Centered R2 0.076
  • Note: This table presents the regression output for the effect of weather derivatives usage on the cost of bank loans through various robustness checks. The dependent variable Log(AISD) is measured as natural logarithm of the interest rate margin over LIBOR on the drawn loan amount plus annual fees. The key explanatory variable WD_Post is an indicator variable taking the value of 1 for loans made to weather derivatives user firms after they first enter weather derivatives contracts, and zero otherwise. We re-estimate the baseline model Log ( AISD ) = α 1 + β 1 WD _ Pos t l i t + γ Control s l i t + θ i D i + θ t D t + ε l i t ${\mathrm{Log}}( {{\mathrm{AISD}}} )\ = {{\alpha }_1}\ + {{\beta }_1}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{lit}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{lit}} + {{\theta }_i}{{D}_i} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{lit}}$ using loans originated after the weather derivatives market activation in 1997 (column 1). Columns (2) and (3) provide the results using an alternative proxy for weather derivatives usage (WD_Post_Alt) in the OLS baseline regression and IV regression, respectively. WD_Post_Alt is defined as a binary variable taking a value of 1 for loans made to weather derivatives users after they first enter weather derivatives contracts, based on all filings (rather than 10-K filings only) submitted to the SEC. The regressions include all borrower and loan characteristics as controls. All variables are defined in the Appendix. The numbers in parentheses are standard errors corrected for heteroscedasticity and clustered at the firm level. ***, ** and * represent significance at the 1, 5 and 10% level, respectively.

In column (1), we restrict our sample size to include only loans from 1997 given the first OTC weather derivatives contract was introduced in 1997 (Pérez-González and Yun, 2013). Due to the year restriction, our sample size is reduced to 1783 from 2091 observations as in the baseline regression. Consistent with the previous finding, the coefficient on WD_Post is negative and statistically significant at the 1% level, with its magnitude slightly higher than what is reported in Table 2. This result supports our baseline argument that banks view weather derivatives usage as an effective risk management tool and reward the energy borrowers by lowering their loan spreads.

Columns (2) and (3) report the regression output of the baseline model and its IV specification when WD_Post_Alt is used as a key explanatory variable instead of WD_Post. As our weather derivatives users and thus WD_Post in the baseline regression are identified by parsing all 10-K filings of energy firms, one might argue that not all firms report their weather derivatives usage in their 10-K filings. In that case, our identification approach might not be able to detect all weather derivatives users. To address this issue, we re-conduct our data collection of weather derivatives usage by downloading and parsing all filings (rather than 10-K filings only) submitted by energy firms to the SEC. Firms that describe weather derivatives contracts in any of those filings are considered weather derivatives users. Based on the new classification of weather derivatives users, we define WD_Post_Alt as a binary variable taking a value of 1 for loans made to weather derivatives users after they first enter weather derivatives contracts. The coefficient on WD_Post_Alt is negative and statistically significant at the 10% level in column (2) and at the 5% level in column (3). This again corroborates our earlier finding that banks reward borrowers’ hedging activities by lowering the required loan spreads.

5 CROSS-SECTIONAL VARIATION ON THE BENEFITS OF HEDGING WITH WEATHER DERIVATIVES

5.1 Hedging benefits subject to market uncertainty and firm risks

In this section, we examine whether the reward of hedging with weather derivatives as reflected in a lower loan cost varies across different levels of market uncertainty and firm risks. If hedging with derivatives allows firms to smooth future revenues and reduce the risk and cost of financial distress, its benefits should increase with earnings variability associated with risk factors at both the market and firm levels. In other words, hedging benefits should increase with market uncertainty and firms’ riskiness (Manconi et al., 2018; Bartram, 2019).

5.1.1 Market uncertainty

We first examine whether the lower loan cost following weather derivatives adoption is exacerbated during more uncertain market conditions, for example, during the Global Financial Crisis (GFC). We construct a GFC dummy variable, which equals 1 for loans originated during the GFC period (July 2007 to June 2009). We re-estimate Equation (1) with GFC and the interaction term between GFC and WD_Post. The result is presented in column (1) of Table 6. We observe that the coefficient on WD_Post (−0.187) is negative and significant at the 10% level, while that of the interaction term (−0.363) is negative and significant at the 5% level. We interpret this as evidence of the cost savings from weather derivatives usage being more pronounced for loans issued during the GFC period. This result supports our conjecture that market-wide (macroeconomic) risk intensifies the perceived benefits of hedging with weather derivatives.

TABLE 6. Effect of weather derivatives usage on loan price subject to market uncertainty and firm risk.
Dependent variable: Log(AISD)
Market uncertainty Firm risk
GFC US Recession Deregulation Climate Policy Risk Beta Rating
(1) (2) (3) (4) (5) (6)
WD_Post −0.187 −0.200 −0.062 −0.035 0.338 −0.172
(0.10) (0.10) (0.12) (0.13) (0.26) (0.11)
GFC 0.550
(0.17)
WD_Post × GFC −0.363
(0.14)
USREC −0.025
(0.07)
WD_Post × USREC −0.342
(0.17)
Deregulation 0.148
(0.06)
WD_Post × Deregulation −0.257
(0.13)
High_CP_Index −0.02
(0.06)
WD_Post × High_CP_Index −0.20
(0.09)
Beta 0.143
(0.10)
WD_Post × Beta −0.568
(0.26)
Low_Rating 0.161
(0.11)
WD_Post × Low_Rating −0.739
(0.14)
Controls Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes Yes
Year FE Yes Yes Yes Yes Yes Yes
Credit rating Yes Yes Yes Yes Yes Yes
Loan purpose Yes Yes Yes Yes Yes Yes
Observations 2091 2091 2091 1718 1523 2091
Adjusted R2 0.731 0.729 0.730 0.716 0.710 0.692
  • Note: This table presents the OLS regression output for the effect of weather derivatives usage on the cost of bank loans, subject to market uncertainty and firm risk. The general form of our model is as follows: LoanSprea d l i t = α 2 + β 2 WD _ Pos t l i t + η 2 X i t + ϑ 2 WD _ Pos t l i t × X i t + γ Control s l i t + θ i D i + θ t D t + ε l i t ${\mathrm{LoanSprea}}{{{\mathrm{d}}}_{lit}} = {{\alpha }_2}\ + {{\beta }_2}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{lit}} + {{\eta }_2}{{X}_{it}} + {{\vartheta }_2}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{lit}} \times {{X}_{it}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{lit}} + {{\theta }_i}{{D}_i} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{lit}}$ . The dependent variable Log(AISD) is measured as the natural logarithm of the interest rate margin over LIBOR on the drawn loan amount plus annual fees. The key explanatory variable WD_Post is an indicator variable taking the value of 1 for loans made to weather derivatives user firms after they first enter weather derivatives contracts, and zero otherwise. X denotes a proxy for market uncertainty in columns (1)–(4), and for firm risk in columns (5) and (6). Market uncertainty is proxied by GFC, USREC, Deregulation and High_CP_Index, whereas borrower risk is proxied by Beta and Low_Rating. GFC is a dummy variable coded 1 if the loan is originated between July 2007 and June 2009 and zero otherwise. USREC is a dummy variable coded 1 if the US economy is in recession and zero otherwise. Deregulation is a dummy variable coded 1 for gas (electricity) loans originated after the state-level deregulation of its gas (electricity) market and zero otherwise. High_CP_Index is a dummy variable coded 1 for loans originated in states with more stringent climate policies (i.e., climate policy index in the top quartile) and zero otherwise. Beta is the borrower's systematic risk of the year prior to loan origination. Low_Rating is a dummy variable coded 1 if the borrower is rated BBB- or below. The regressions include all borrower and loan characteristics as controls. All variables are defined in the Appendix. The numbers in parentheses are standard errors corrected for heteroscedasticity and clustered at the firm level. ***, ** and * represent significance at the 1, 5 and 10% level, respectively.

Our second proxy for market risk is an indicator of recession in the US economy. We obtain this dataset from the Federal Reserve Bank of St. Louis, and construct a dummy variable, USREC, to indicate if the US economy is in recession. We match these data with the timing of loan contracts and interact USREC with WD_Post. The result is presented in column (2) of Table 6. Consistent with the previous proxy for market uncertainty (GFC), the coefficient of the interaction term WD_Post  × $ \times $  USREC is negative and significant at the 10% level.

The third market risk factor that we examine in this section is market competition risk. We utilize state-level deregulation of gas and electricity market as an exogenous shock to product market competition. The US Energy Information Administration provides data on deregulation of gas and electricity markets across different states. Deregulation allows utility providers from other regions to enter and compete in a local market, therefore increasing the level of competition and altering business prospects among energy firms. Not all states have deregulated their energy markets. Among the deregulated states, the timing of deregulation also varies across states and varies between gas and electricity. Hence, this can serve as a clean exogenous shock to examine the effect of market competition on the relation between weather derivatives usage and loan costs. More intense competition adds more uncertainty to the level of energy demand, hence increases a given energy firm's risk exposure. We expect the benefit from hedging with weather derivatives to increase following energy market deregulation. To test this hypothesis, we classify firms as gas or electricity firms based on their SIC codes. We construct a Deregulation dummy variable, which takes a value of 1 for gas (electricity) loans originated after the state-level deregulation of its gas (electricity) market and zero otherwise. We then re-estimate Equation (1) with Deregulation and the interaction term between Deregulation and WD_Post. The result is presented in column (3) of Table 6. Consistent with our conjecture, the coefficient of the interaction term WD_Post  × $ \times $  Deregulation is negative and significant, albeit weakly at the 10% level.

We further obtain the state-level climate policy index developed by Bergquist and Warshaw (2023). This index is constructed using information from 25 distinct climate policies to capture efforts made at the state level to mitigate the impact of climate change through the energy system. Data for this index are available in Harvard Dataverse from 2000 to 2020. We create a dummy variable High_CP_Index, which takes the value of 1 for loans originated in states with more stringent climate policies (i.e., climate policy index in the top quartile) and zero otherwise. We argue that the stringency of climate-related regulations reflects regulatory uncertainty due to the increased transition risk, rising financial distress costs and potentially severe environmental liabilities and reputational damage (Fard et al., 2020; Dang et al., 2022; Delis et al., 2023). We re-estimate our baseline regression by including High_CP_Index and its interaction term with WD_Post. The coefficient of the interaction term WD_Post  × $ \times $  High_CP_Index is negative and significant at the 5% level, as shown in column (4) of Table 6. This suggests that the impact of weather derivatives hedging on loan price is more pronounced in the states with more stringent climate policies (i.e., higher regulatory risk).

Overall, the four market-wide risk proxies provide consistent evidence that the benefit of hedging with weather derivatives is more pronounced during periods of more volatile or more competitive markets and higher regulatory uncertainty.

5.1.2 Firm risks

We examine firm-level risk factors including firms’ beta and credit ratings. Similar to the market-wide risk factors, we expect the benefit of hedging with weather derivatives as reflected in lower loan costs to be more pronounced among riskier firms. First, we estimate the systematic risk, Beta, of our sample firms using the Beta Suite provided by WRDS, and re-estimate Equation (1) with Beta and the interaction term between Beta and WD_Post. The result is reported in column (5) of Table 6. The coefficient for the interaction term WD_Post × Beta is negative and significant at the 5% level, which suggests that the borrowing cost saving from hedging with weather derivatives is larger when the borrower carries higher systematic risk.

Our second firm-level risk proxy is based on a firm's credit rating. Similar to beta, our conjecture is that lower rated (higher risk) firms should benefit more from hedging with weather derivatives. Therefore, we create a Low_Rating dummy variable taking the value of 1 if the borrower is rated BBB- or below. We then interact the dummy Low_Rating with WD_Post and present the result in column (6) of Table 6. Once again, we observe the coefficient of the interaction term being negative and statistically significant at the 1% level. This result is consistent with the findings using Beta and indicates that hedging benefits are stronger for riskier firms.

Taken altogether, the results from this section suggest that the benefits from hedging with weather derivatives as reflected in lower loan costs are amplified when hedging is adopted during more uncertain market conditions and when borrowers exhibit higher risks.

5.2 Hedging benefits across different levels of financial reporting complexity and investor attention to climate risk

Other factors that could have material effects on hedging benefits are the levels of financial reporting complexity in borrowing firms and investor attention to climate risks.

5.2.1 Financial reporting complexity

In our study, the indicator for firms’ weather derivatives usage is identified from the inclusion of specific terms (“weather hedging,” “weather derivatives,” “weather forwards,” “weather futures,” “weather options” and “weather swaps”) in their 10-K filings. Whether these firms are successful with their mentioned hedging strategies via weather derivatives, however, is not known. The literature has documented negative effects of financial reporting complexity on informational efficiency (Li, 2008; Lawrence, 2013; Guay et al., 2016). We predict that for firms whose financial reports and filings are more complex and lack clarity, loan officers may overlook their reported hedging activities, hence do not always reward them with lower loan spreads. In other words, the borrowing cost savings from weather derivatives usage are expected to be less for firms with more complex financial reports.

We test this conjecture by re-estimating Equation (1) with proxies for borrowers’ financial reporting complexity and their interaction term with WD_Post. To measure the complexity of the firms’ 10-K filings, we employ text readability measures, including Report Length, Gunning Fog Index, Flesch-Kincaid Index and RIX Index (Courtis, 1987; Li, 2008; Guay et al., 2016). Report Length is the natural logarithm of the number of words in the firm's 10-K filings. Shorter reports are argued to be more readable (Li, 2008). Gunning Fog Index (Gunning, 1952), Flesch-Kincaid Index (Kincaid et al., 1975) and RIX Index (Anderson, 1983) measure readability based on the number of syllables, words, complex words and sentences. Lower values of these indices indicate more accessible reading levels. For each year, we sort firms based on their filings’ readability and create a dummy for firms in the top quartile of relevant readability measures (i.e., more complex financial reporting). We then interact that dummy variable with WD_Post.

The results in the first four columns of Table 7 demonstrate that WD_Post loads significantly negative (at the 1% level) on loan spreads in all model specifications. The variable of interest is the interaction term between financial reporting complexity and WD_Post. In all four model specifications, the interaction terms are positive and significant (at the 10 and 5% levels). We interpret this result as evidence of the benefits of hedging via weather derivatives being less pronounced when the hedging strategies are masked by the hedging firm's complex financial reports. The complexity and quality of financial reporting are potentially endogenous and correlated to some unobserved borrower attributes. Therefore, it is not our intention to claim the causality between financial reporting complexity and cost of bank loans. Rather, financial reporting complexity could be an indicator of clarity and credibility of the hedging strategy implemented via weather derivatives contracts. This indicator, based on our data, seems to suggest that the borrowing cost savings from weather derivatives usage are lower when borrowers’ financial reports are more complex. This is in line with our expectation.

TABLE 7. Effect of weather derivatives usage subject to financial reporting complexity and investor attention to climate risk.
Dependent variable: Log(AISD)
Financial reporting complexity Attention to climate risk
Report Length Gunning Fog Flesch-Kincaid RIX CC_WSJ Stern_Review
(1) (2) (3) (4) (5) (6)
WD_Post −0.342 −0.335 −0.333 −0.333 −0.188 −0.099
(0.10) (0.10) (0.10) (0.10) (0.09) (0.11)
High Report Length 0.043
(0.05)
WD_Post × High Report Length 0.202
(0.11)
High Gunning Fog 0.009
(0.05)
WD_Post × High Gunning Fog 0.256
(0.10)
High Flesch Kincaid −0.016
(0.05)
WD_Post × High Flesch Kincaid 0.216
(0.09)
High RIX −0.029
(0.04)
WD_Post × High RIX0 0.234
(0.10)
CC_WSJ 4.674
(5.83)
WD_Post × CC_WSJ −40.737
(15.17)
Stern_Review −0.110
(0.09)
WD_Post × Stern_Review −0.185
(0.08)
Controls Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes Yes
Year FE Yes Yes Yes Yes Yes Yes
Credit rating Yes Yes Yes Yes Yes Yes
Loan purpose Yes Yes Yes Yes Yes Yes
Observations 1628 1628 1628 1628 2091 2091
Adjusted R2 0.736 0.737 0.735 0.735 0.729 0.730
  • Note: This table presents the OLS regression output for the effect of weather derivatives usage on the cost of bank loans, across different levels of financial reporting complexity and investor attention to climate risk. The general form of our model is as follows: LoanSprea d l i t = α 3 + β 3 WD _ Pos t l i t + η 3 Z i t + ϑ 3 WD _ Pos t l i t × Z i t + γ Control s l i t + θ i D i + θ t D t + ε l i t ${\mathrm{LoanSprea}}{{{\mathrm{d}}}_{lit}} = {{\alpha }_3} + {{\beta }_3}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{lit}} + {{\eta }_3}{{Z}_{it}} + {{\vartheta }_3}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{lit}} \times {{Z}_{it}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{lit}} + {{\theta }_i}{{D}_i} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{lit}}$ . The dependent variable Log(AISD) is measured as natural logarithm of the interest rate margin over LIBOR on the drawn loan amount plus annual fees. The key explanatory variable WD_Post is an indicator variable taking the value of 1 for loans made to weather derivatives user firms after they first enter weather derivatives contracts, and zero otherwise. Z denotes a proxy for financial reporting complexity in columns (1)–(4), and for investor attention to climate risk in columns (5) and (6). Borrowers’ financial reporting complexity is proxied by Report Length, Gunning Fog Index, Flesch-Kincaid Index and RIX Index, whereas investor attention to climate risk is proxied by CC_WSJ and Stern_Review. Report Length is the natural logarithm of a number of words in the firm's 10-K filings. Gunning Fog Index is a measure of grade-level readability developed by Gunning (1952). Flesch-Kincaid Index is a measure of text readability based on US school grade level (Kincaid et al., 1975). RIX Index is a measure of text readability based on the number of words of seven or more characters per sentence (Anderson, 1983). Dummy variables for high levels of reporting complexity take the value of 1 if the reporting firm belongs to the top quartile of Report Length, Gunning-Fog Index, Flesch-Kincaid Index and RIX Index each year. The continuous variable CC_WSJ is the innovation in the WSJ Climate Change News Index developed by Engle et al. (2020). The dummy variable Stern_Review is coded 1 for loans originated after the 2006 Stern Review and zero otherwise. All variables are defined in the Appendix. The numbers in parentheses are standard errors corrected for heteroscedasticity and clustered at the firm level. ***, ** and * represent significance at the 1, 5 and 10% level, respectively.

5.2.2 Attention to climate risk

We utilize two proxies for investors’ climate risk attention. First, we obtain the data for innovations in the Wall Street Journal (WSJ) Climate Change News Index to construct the variable CC_WSJ following Engle et al. (2020). As elevated climate risk receives stronger news coverage when there is indeed a cause for concern (Engle et al., 2020), therefore drawing more attention from loan officers, we argue that the benefit from hedging with weather derivatives is higher in the period of heightened media attention to climate risks. Second, we create a dummy variable Stern_Review to capture the release of the Stern Review on the Economics of Climate Change in October 2006 (Stern, 2008). The dummy variable Stern_Review is coded 1 for loans originated after the Stern Review and zero otherwise. As the Stern Review is an important event that has attracted positive attention to climate change from both regulators and investors (Painter, 2020; Javadi and Masum, 2021; Javadi et al., 2023), we conjecture that the benefit of hedging with weather derivatives is more pronounced in the period after the Stern Review. We then re-estimate Equation (1) by including each of the attention proxies and its interaction term with WD_Post. Results are presented in columns (5) and (6) of Table 7. Consistent with our conjecture, the coefficients of the interaction terms WD_Post  × $ \times \ $  CC_WSJ and WD_Post × $ \times \ $  Stern_Review are both negative and statistically significant. This suggests that the benefit of weather derivatives hedging in lowering loan costs is more pronounced when investors pay more attention to climate risks.

Overall, the results in Table 7 suggest that hedging with weather derivatives is more beneficial in firms with better-quality financial reports and when there is heightened investor attention to climate risks.

6 HEDGING WITH WEATHER DERIVATIVES AND NON-PRICE LOAN TERMS

A loan contract is often complex and involves many loan-specific characteristics. Dennis et al. (2000) document strong substitutional effects among loan terms, including price, maturity and collateral requirements. During the loan contracting process, these terms are often simultaneously negotiated, creating a venue for borrowers to trade-off, for instance, to pledge high-quality collaterals or accept more restrictive covenants in exchange for lower spreads. This trade-off effect can be more pronounced for borrowers who adopt good hedging strategies and hence want to signal their improved quality and value. The ability to trade-off could be a channel for these firms to receive more favorable loan spreads as observed in the previous sections. We investigate if borrowers are more likely to pledge collateral and accept more restrictive covenants following their adoption of weather derivatives for hedging. We follow Shan et al. (2019) and estimate the model below:
Non − Price Loan Ter m l i t = α 4 + β 4 WD _ Pos t l i t + λ 4 WD _ Use r i + γ Control s l i t + θ t D t + ε l i t $$\begin{equation}{\mathrm{Non}} - {\mathrm{Price}}\;{\mathrm{Loan}}\;{\mathrm{Ter}}{{{\mathrm{m}}}_{lit}} = {{\alpha }_4}\ + {{\beta }_4}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{lit}} + {{\lambda }_4}{\mathrm{WD}}\_{\mathrm{Use}}{{{\mathrm{r}}}_i} + \ {\mathrm{\gamma Control}}{{{\mathrm{s}}}_{lit}} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{lit}}\end{equation}$$ (4)
where Non-Price Loan Term is proxied by Secured Dummy, Covenant Index and Covenant Strict. Secured Dummy is a binary variable taking the value of 1 if a loan has collateral and zero otherwise. Covenant Index follows Bradley and Roberts (2015)’s count variable, which takes a value between 0 and 6 based on six (6) key covenant categories. Covenant Strict is a dummy variable calculated based on Bradley and Roberts (2015)’s Covenant Index and equals 1 for loans with three or more covenant categories and zero otherwise. The key explanatory variable WD_Post is an indicator variable taking the value of 1 for loans made to weather derivatives user firms after they first enter weather derivatives contracts, and zero otherwise. The dummy WD_User takes the value of 1 for loans made to weather derivatives user firms, and zero otherwise. We include Log(Facility Amount), Log (Maturity), Log(AISD) and Lender Number as loan characteristics control variables; and Log(Assets) and credit ratings as borrower characteristics control variables. All variables are defined in the Appendix. Our regressions include year fixed effects and control for loan purpose. Standard errors are adjusted for heteroscedasticity and clustered at the borrower level.

Table 8 presents the results for the effect of weather derivatives usage on non-price contract terms. Columns (1) and (3) present the Probit output for Secured Dummy and Covenant Strict, while column (2) presents the Poisson output for Covenant Index. The coefficient of WD_Post is positive and significant at 10% in column (1) (coefficient = 0.413 or marginal effect of 0.059). It suggests that after firms adopt hedging with weather derivatives, their probability of pledging collateral in a loan contract increases by about 5.9%.

TABLE 8. Effect of weather derivatives usage on collateral and covenant restrictions.
Dependent variable Secured Dummy Covenant Index Covenant Strict
Estimation method Probit Poisson Probit
(1) (2) (3)
WD_Post 0.413 0.246 0.468
(0.25) (0.15) (0.20)
Controls Yes Yes Yes
Year FE Yes Yes Yes
Loan purpose Yes Yes Yes
Observations 2109 2137 1912
Pseudo R2 0.313 0.171 0.349
  • Note: This table presents the regression output for the effect of weather derivatives usage on collateral incidence and covenant restrictions using the following model: Non − Price Loan Ter m l i t = α 4 + β 4 WD _ Pos t l i t + λ 4 WD _ Use r i + γ Control s l i t + θ t D t + ε l i t ${\mathrm{Non}} - {\mathrm{Price}}\;{\mathrm{Loan}}\;{\mathrm{Ter}}{{{\mathrm{m}}}_{lit}} = {{\alpha }_4} + {{\beta }_4}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{lit}} + {{\lambda }_4}{\mathrm{WD}}\_{\mathrm{Use}}{{{\mathrm{r}}}_i} + \ \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{lit}} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{lit}}$ . The dependent variable Non-Price Loan Term in columns (1)–(3) is Secured Dummy, Covenant Index and Covenant Strict, respectively. Secured Dummy is a binary variable taking the value of 1 if a loan has collateral and zero otherwise. Covenant Index follows Bradley and Roberts (2015)’s count variable, which takes a value between 0 and 6 based on 6 key covenant categories. Covenant Strict is a dummy variable calculated based on Bradley and Roberts (2015)’s Covenant Index and equals 1 for loans with 3 or more covenant categories and zero otherwise. The key explanatory variable WD_Post is an indicator variable taking the value of 1 for loans made to weather derivatives user firms after they first enter weather derivatives contracts, and zero otherwise. The dummy WD_User takes the value of 1 for loans made to weather derivatives user firms, and zero otherwise. All variables are defined in the Appendix. The numbers in parentheses are standard errors corrected for heteroscedasticity and clustered at the firm level. ***, ** and * represent significance at the 1, 5 and 10% level, respectively.

The coefficient for WD_Post in column (2) (Poisson estimation of Covenant Index) is 0.246 and significant at the 10% level. It implies that after firms adopt weather derivatives for hedging, the natural logarithm of their covenant index increases by 0.246, equivalent to a 27.9% [= exp(0.246)−1] increase in the number of covenant categories. The coefficient for WD_Post is significantly positive at the 5% level in column (3) (Probit estimation of Covenant Strict). The coefficient is 0.468 (marginal effect of 0.055), suggesting that borrowers are about 5.5% more likely to accept stricter covenants after hedging with weather derivatives in their operations.

As shown in prior literature, firms more vulnerable to climate risk often face more restrictive loan terms, including higher collateral incidence and covenant intensity (Ho and Wong, 2023; Hrazdil et al., 2023; Huang et al., 2023). Meanwhile, firms facing greater climate risks have been documented to experience a higher likelihood of weather derivatives usage (Pérez-González and Yun, 2013; Weagley, 2019). Therefore, the relationship between weather derivatives hedging and collateral requirement and covenant strictness might be subject to endogeneity concerns. To mitigate these concerns, we re-estimate Table 8 using the two-stage IV approach. Specifically, similar to the estimation in Table 4, we use EDD Weather-Induced Volatility as an instrument for WD_Post. We then conduct the second stage of the IV estimation at the loan level for collateral incidence and covenant strictness and obtain robust results.

In general, the results from Table 8 suggest that borrowers are more willing to pledge collateral and accept stricter covenants after adopting weather derivatives for hedging. Consistent with the trade-off argument, this could be a channel for firms to signal better risk management strategies and improved firm value, which may help lower their borrowing costs.

7 HEDGING WITH WEATHER DERIVATIVES, DEBT CHOICE AND BOND YIELDS

In the previous sections, we have provided strong evidence that hedging with weather derivatives could benefit borrowers in both price and non-price terms of private debt contracts. In this section, we explore the possibility that benefits of hedging with weather derivatives are not limited to the private debt market but may as well extend to the public debt market, given information about the borrowing firm flows across different capital markets. We examine whether hedging with weather derivatives affects a firm's reliance on private bank loans and cost of issuing public debt.

First, we explore the effect of hedging with weather derivatives on the borrowing firm's choice of debt. Extensive research has been conducted to understand the determinants behind a firm's choice of public versus private debt, and weather derivatives hedging could add an important angle to this literature. The general consensus is that firms with higher information asymmetry tend to rely more on private debt. This is consistent with the traditional banking theories that emphasize asymmetric information mitigation as a competitive advantage of banks (Fama, 1985; Rajan, 1992; Hadlock and James, 2002; Denis and Mihov, 2003). Conversely, companies with robust corporate governance structures tend to utilize public debt more frequently (Houston and James, 1996; Lin et al., 2013). Given that hedging can mitigate risk-shifting problems (Campbell and Kracaw, 1990) and lower information asymmetries (DaDalt et al., 2002), one can infer that hedging firms are better-positioned to access the public debt market and consequently become less reliant on private bank loans. We test this conjecture using the following model:
Bank _ Debt _ Rati o i t = α 5 + β 5 WD _ Pos t i t + γ Control s i t + θ t D t + ε i t $$\begin{equation}{\mathrm{Bank}}\_{\mathrm{Debt}}\_{\mathrm{Rati}}{{{\mathrm{o}}}_{it}} = {{\alpha }_5}\ + {{\beta }_5}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{it}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{it}} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{it}}\end{equation}$$ (5)
where we follow recent literature (see, for example, Lin et al., 2013; Boubaker et al., 2018; Huang et al., 2023) and define Bank_Debt_Ratio as the percentage of bank debt over total debt of the firm. We obtain firm-level annual data for Bank_Debt_Ratio from the Capital IQ database. The key explanatory variable WD_Post is an indicator taking the value of 1 for firms after their adoption of weather derivatives contracts, and zero otherwise. The control variables include borrower characteristics similar the previous models. The result from this regression is presented in column (1) of Table 9. We observe the coefficient of WD_Post to be negative and statistically significant at the 1% level, indicating that firms have a lower bank debt ratio after hedging with weather derivatives. In other words, hedging with weather derivatives helps enhance a firm's access to public debts, or reduce its reliance on private bank debts.
TABLE 9. Effect of weather derivatives usage on bank debt ratio and bond yield spreads.
Dependent variable Bank_Debt_Ratio Bond_Yield_Spread Log_Yield_Spread
(1) (2) (3)
WD_Post −4.316 −60.863 −0.262
(1.67) (26.48) (0.15)
Controls Yes Yes Yes
Firm FE No Yes Yes
Year FE Yes Yes Yes
Credit rating Yes Yes Yes
Bond characteristics No Yes Yes
Observations 1771 592 592
Adjusted R2 0.449 0.650 0.769
  • Note: This table presents the regression output for the effect of weather derivatives usage on the bank debt ratio (column 1) and bond yield spreads (columns 2 and 3). Column (1) presents the regression output of the following model: Bank _ Debt _ Rati o i t = α 5 + β 5 WD _ Pos t i t + γ Control s i t + θ t D t + ε i t ${\mathrm{Bank}}\_{\mathrm{Debt}}\_{\mathrm{Rati}}{{{\mathrm{o}}}_{it}} = {{\alpha }_5} + {{\beta }_5}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{it}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{it}} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{it}}$ , estimated at the firm level. Bank_Debt_Ratio is the ratio of bank debt to total debt measured. Columns (2) and (3) present the regression output of the following model: Bond _ Yield _ Sprea d b i t = α 6 + β 6 WD _ Pos t b i t + γ Control s b i t + θ i D i + θ t D t + ε b i t ${\mathrm{Bond}}\_{\mathrm{Yield}}\_{\mathrm{Sprea}}{{{\mathrm{d}}}_{bit}} = {{\alpha }_6} + {{\beta }_6}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{bit}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{bit}} + {{\theta }_i}{{D}_i} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{bit}}$ , estimated at the bond level. The dependent variables in columns (2) and (3) are Bond_Yield_Spread and Log_Yield_Spread. Bond_Yield_Spread is the difference between the yield at issue and the yield of Treasury bonds with the same maturity, whereas Log_Yield_Spread is the natural logarithm of Bond_Yield_Spread. WD_Post is the key explanatory variable, which is an indicator variable taking the value of 1 for firms after their adoption of weather derivatives (column 1), or for bonds issued by weather derivatives user firms after they first enter weather derivatives (columns 2 and 3). The control variables include firm characteristics (in all three columns) and bond-specific characteristics (such as bond maturity, issuance amount, collateral and covenant dummies, and bond credit rating) (in columns 2 and 3). All variables are defined in the Appendix. The numbers in parentheses are standard errors corrected for heteroscedasticity and clustered at the firm level. ***, ** and * represent significance at the 1, 5 and 10% level, respectively.
General hedging strategies have been found to reduce bankruptcy costs through reduction of cash flow volatility (Smith and Stulz, 1985), lower information asymmetries (DaDalt et al., 2002) and lower cost of public debt (Chen and King, 2014). We further explore whether the benefits of weather derivatives usage are potentially extended to the public debt market in the form of reduced bond yields. We collect data on all bond issuance of the firms in our sample along with all bond characteristics from the Mergent Fixed Income Securities Database. This database covers US corporate bond issuance, including all bond characteristics, such as bond yield, coupon rate, maturity, issue amount, currency and covenants. Following Allman (2022), we exclude all non-standard bonds and only keep fixed coupon bonds. We then merge the bond data with borrower characteristics. The regression model is similar to our baseline model, except loan characteristics being replaced by bond characteristics:
Bond _ Yield _ Sprea d b i t = α 6 + β 6 WD _ Pos t b i t + γ Control s b i t + θ i D i + θ t D t + ε b i t $$\begin{equation}{\mathrm{Bond}}\_{\mathrm{Yield}}\_{\mathrm{Sprea}}{{{\mathrm{d}}}_{bit}} = {{\alpha }_6}\ + {{\beta }_6}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{bit}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{bit}} + {{\theta }_i}{{D}_i} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{bit}}\end{equation}$$ (6)
where b, i and t denote bonds, firms and years, respectively. Bond_Yield_Spread is defined as the difference between the yield at issue and the yield of Treasury bonds with the same maturity. The key explanatory variable WD_Post is an indicator variable taking the value of 1 for bonds issued by weather derivatives user firms after they first enter weather derivatives contracts, and zero otherwise. The control variables include issuer characteristics similar to those specified in the baseline model. We also include bond-specific characteristics such as bond maturity, issuance amount, collateral and covenant dummies and bond credit rating. The result for this regression is presented in columns (2) and (3) of Table 9. Column (2) shows the result for raw bond yield spreads expressed in basis points while column (3) presents the result for the natural logarithm of bond yield spreads. We observe the coefficient of WD_Post to be negative and significant in both columns (2) and (3). The coefficient of WD_Post of −60.863 in column (2) indicates that following initial usage of weather derivatives, firms enjoy paying approximately 61 bps less on their bond issues. These results suggest that the benefit of hedging with weather derivatives is not confined solely to the private loan market but also extends to the public debt market.

8 EX-POST ANALYSES ON THE EFFECT OF HEDGING WITH WEATHER DERIVATIVES

In the previous sections, we present empirical evidence on the relation between weather derivatives usage and loan contract (price and non-price) terms. We have documented that borrowers enjoy a lower cost of loans and are more willing to pledge collateral and accept stricter covenants after adopting weather derivatives for hedging. Consistent with the trade-off theory, this result supports a signaling mechanism where weather derivatives user firms accept more restrictive non-price loan terms to be rewarded with cheaper borrowing costs. In this section, we further explore whether the ex-ante signal through weather derivatives usage documented earlier is a credible one. We examine an ex-post channel through which hedging can add value by lowering firms’ default probability and reducing the likelihood that borrowing firms experience a technical default due to covenant violations.

8.1 Hedging with weather derivatives and risk reduction

In this section, we examine a risk reduction channel through which weather derivatives usage can lead to lower loan spreads. Pérez-González and Yun (2013) present evidence that the added value of weather derivatives usage comes from lower cash holdings and higher debt and capital expenditures. We use three proxies to measure Firm_Risk, including Change in ROA, Change in ROE and Altman Z. We adopt the two-stage IV approach with the first stage regression similar to that in Section 4.2.2. These regressions are estimated at the firm-year level with firm and year fixed effects. As the first-stage results are already presented in Panel B of Table 4, we only present results of the second stage (as per the equation below) in Table 10.
Firm _ Ris k i t = α 7 + β 7 Fitted _ WD _ Pos t i t + γ Control s i t + θ i D i + θ t D t + ε i t $$\begin{equation}{\mathrm{Firm}}\_{\mathrm{Ris}}{{{\mathrm{k}}}_{it}} = {{\alpha }_7} + {{\beta }_7}{\mathrm{Fitted}}\_{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{it}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{it}} + {{\theta }_i}{{D}_i} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{it}}\end{equation}$$ (7)
TABLE 10. Effect of weather derivatives usage on firm risks.
Dependent variable Change in ROA Change in ROE Altman Z
(1) (2) (3)
Fitted_WD_Post −0.002 −0.005 0.392
(0.00) (0.00) (0.19)
Controls Yes Yes Yes
Firm FE Yes Yes Yes
Year FE Yes Yes Yes
Observations 2980 2943 1402
Adjusted R2 0.004 0.003 0.509
  • Note: This table presents the result for the second stage of IV estimation measuring the effect of weather derivatives usage on firm risks. The estimated model is as follows: Firm _ Ris k i t = α 7 + β 7 Fitted _ WD _ Pos t i t + γ Control s i t + θ i D i + θ t D t + ε i t ${\mathrm{Firm}}\_{\mathrm{Ris}}{{{\mathrm{k}}}_{it}} = {{\alpha }_7} + {{\beta }_7}{\mathrm{Fitted}}\_{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{it}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{it}} + {{\theta }_i}{{D}_i} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{it}}$ . We use three proxies for firm risk as the dependent variable, including Change in ROA, Change in ROE and Altman Z. Change in ROA in column (1) is the yearly change in the firm's ROA. Change in ROE in column (2) is the yearly change in the firm's ROE. Altman Z in column (3) is the measure of firm risk calculated as 3.3 × (EBIT/Total assets) + 0.99*(Net sales/Total assets) + 0.6 × (Market value of equity/Total liabilities) + 1.2 × (Working capital/Total assets) + 1.4 × (Retained earnings/Total assets). Fitted_WD_Post is the fitted value of WD_Post, obtained from the first stage IV regression of WD_Post on EDD Weather-Induced Volatility measured for the pre-1997 period, which equals Absolute EDD Beta multiplied by EDD Standard Deviation. The regressions include firm characteristics as controls. All variables are defined in the Appendix. The numbers in parentheses are standard errors corrected for heteroscedasticity and clustered at the firm level. ***, ** and * represent significance at the 1, 5 and 10% level, respectively.

Column (1) of Table 10 presents the result for Change in ROA, measured as the yearly change in the firm's ROA. The coefficient of Fitted_WD_Post is negative and significant at the 10% level, consistent with a reduction in ROA volatility following weather derivatives usage. We also observe similar results for Change in ROE, presented in column (2). This suggests that hedging with weather derivatives contracts can be an effective risk management tool in reducing the volatility of revenue flows.

Column (3) of Table 10 explores the effect of weather derivatives usage on the aggregate level of firm risk measured by the standard Altman (1968)’s Z-score. The coefficient of Fitted_WD_Post is positive and significant at the 10% level and confirms the result reported in columns (1) and (2) as higher Z-scores reflect lower riskiness. Taken together, the findings from this section provide strong support for our risk reduction channel. The lower cost of loans associated with weather derivatives usage can be attributed to overall better risk management, and accordingly lower firm risks.

8.2 Hedging with weather derivatives and covenant violations

Loan covenants are often set up as an early detection mechanism for deterioration in borrower quality (Nini et al., 2009, 2012). Even though creditors seldom accelerate loan repayments, consequences following covenant violations can be substantial through management intervention. For example, Chava and Roberts (2008) show that capital investments decline sharply following a financial covenant violation. Nini et al. (2012) report that violations are followed by a reduction in capital expenditures and acquisitions plus a sharp reduction in leverage and shareholder payouts and an increase in CEO turnover. Freudenberg et al. (2017) find that covenant violations can increase loan spreads by an average of 18 bps in the subsequent loan. Billett et al. (2016) document increased borrowing costs and creditor involvement following a covenant violation. Therefore, another channel for derivatives hedging to improve firm value is that firms can better smooth their revenues through the use of weather derivatives, which then lowers the probability of covenant violations.

We examine the relationship between hedging with weather derivatives and ex-post covenant violations based on the model developed by Demerjian and Owens (2016). The model can be specified as follows:
Firs t _ Year _ Violatio n lit = α 8 + β 8 WD _ Pos t lit + λ 8 WD _ Use r i + φ 8 PVIO L lit + γ Control s lit + θ t D t + ε lit $$\begin{equation} \mathrm{Firs}\mathrm{t}\_\mathrm{Year}\_\mathrm{Violatio}{\mathrm{n}}_{\textit{lit}}={\alpha}_{8}\ +{\beta}_{8}\mathrm{WD}\_\mathrm{Pos}{\mathrm{t}}_{\textit{lit}}+{\lambda}_{8}\mathrm{WD}\_\mathrm{Use}{\mathrm{r}}_{i}\ +\ {\varphi}_{8}\mathrm{PVIO}{\mathrm{L}}_{\textit{lit}}+\ \gamma \mathrm{Control}{\mathrm{s}}_{\textit{lit}}+{\theta}_{t}{D}_{t}+{\varepsilon}_{\textit{lit}} \end{equation}$$ (8)
where First_Year_Violation is a dummy variable taking the value of 1 if the loan has a covenant violation within the first year of its origination, and zero otherwise. The key explanatory variable WD_Post is an indicator variable taking the value of 1 for loans made to weather derivatives user firms after they first enter weather derivatives contracts, and zero otherwise. The dummy WD_User takes the value of 1 for loans made to weather derivatives user firms, and zero otherwise. PVIOL denotes the measure of probability of covenant violation developed by Demerjian and Owens (2016). PVIOL_ALL is the aggregate probability of covenant violations based on 15 financial covenants. It considers both the number of covenants and the initial slack of each covenant in a loan contract. Demerjian and Owens (2016) also decompose PVIOL_ALL into probability of violation on performance covenants (PVIOL_PCOV) and that on capital covenants (PVIOL_CCOV). Following Demerjian and Owens (2016), we include Altman's Z-score among the explanatory variables to capture borrower's bankruptcy risk. Other controls for borrower characteristics include Log(Assets), ROA, Current Ratio, Log(1+Coverage), Profitability and Tangibility; and those for loan characteristics include Log(Facility Amount), Log (Maturity), Secured Dummy and Lender Number. All variables are defined in the Appendix. Our regressions control for year, loan purpose and credit rating. Standard errors are adjusted for heteroscedasticity and clustered at the borrower level.

Table 11 presents the results for the effect of weather derivatives usage on the likelihood of covenant violations within 1 year of loan origination, with PVIOL_ALL, PVIOL_PCOV and PVIOL_CCOV as additional regressors in columns (1)–(3), respectively. The coefficient on WD_Post is negative and significant at the 5% level on all three model specifications. We interpret this as evidence that borrowers using weather derivatives for hedging have a lower probability of violating covenants within the first year of their loan issuance. This is consistent with our prediction that hedging with weather derivatives helps reduce revenue variability, hence mitigating the incidence of technical default. It is also interesting to observe that the coefficients on the overall PVIOL_ALL and PVIOL_CCOV are not statistically significant while that on PVIOL_PCOV is strongly significant at the 5% level. It appears that the tightness of performance covenants is one of the main drivers of borrowers’ subsequent violations in this particular industry. Since covenant violations can adversely affect many aspects of firm operations, risk management strategies that help firms avoid violations can potentially increase firm value. Lowering the likelihood of ex-post technical defaults could therefore be another channel through which derivatives hedging improves firm value.

TABLE 11. Effect of weather derivatives usage on covenant violations.
Dependent variable: First_Year_Violation
(1) (2) (3)
WD_Post −0.865 −0.803 −0.920
(0.37) (0.38) (0.35)
PVIOL_ALL 0.401
(0.30)
PVIOL_PCOV 0.684
(0.33)
PVIOL_CCOV 0.156
(0.61)
Controls Yes Yes Yes
Year FE Yes Yes Yes
Credit rating Yes Yes Yes
Loan purpose Yes Yes Yes
Observations 270 270 270
Pseudo R2 0.209 0.217 0.204
  • Note: This table presents the Probit regression output for the effect of weather derivatives usage on first-year covenant violations using the following model: First _ Year _ Violatio n l i t = α 8 + β 8 WD _ Pos t l i t + λ 8 WD _ Use r i + φ 8 PVIO L l i t + γ Control s l i t + θ t D t + ε l i t ${\mathrm{First}}\_{\mathrm{Year}}\_{\mathrm{Violatio}}{{{\mathrm{n}}}_{lit}} = {{\alpha }_8} + {{\beta }_8}{\mathrm{WD}}\_{\mathrm{Pos}}{{{\mathrm{t}}}_{lit}} + {{\lambda }_8}{\mathrm{WD}}\_{\mathrm{Use}}{{{\mathrm{r}}}_i} + \ {{\varphi }_8}{\mathrm{PVIO}}{{{\mathrm{L}}}_{lit}} + \gamma {\mathrm{Control}}{{{\mathrm{s}}}_{lit}} + {{\theta }_t}{{D}_t} + {{\varepsilon }_{lit}}$ . The dependent variable First_Year_Violation is a dummy variable taking the value of 1 if the loan has a covenant violation within the first year of loan origination and zero otherwise. The key explanatory variable WD_Post is an indicator variable taking the value of 1 for loans made to weather derivatives user firms after they first enter weather derivatives contracts, and zero otherwise. PVIOL denotes the measure of probability of covenant violation developed by Demerjian and Owens (2016) to capture both the number of covenants and the initial slack of each covenant in a loan contract. PVIOL_ALL is the aggregate probability of covenant violations based on 15 financial covenants, while PVIOL_PCOV and PVIOL_CCOV in columns (2) and (3) are based on performance and capital covenants, respectively. All variables are defined in the Appendix. The numbers in parentheses are standard errors corrected for heteroscedasticity and clustered at the firm level. ***, ** and * represent significance at the 1, 5 and 10% level, respectively.

9 CONCLUSION

This paper explores the financial implications of firms’ usage of weather derivatives contracts for hedging through the lenses of their lenders. We focus on US energy firms, that is, electricity and natural gas firms, whose business is most directly affected by volatility in energy demand (which in turn varies with daily temperature). Weather derivatives contracts represent a relatively new class of derivative products that allow energy firms to hedge reduced demand on mild temperature days.

Our findings highlight an average spread reduction of 21 bps in loans borrowed after a firm first enters weather derivatives contracts. This cost saving can be attributed to two main channels. First, hedging via weather derivatives contracts helps lower bankruptcy risk and is therefore rewarded by banks via a lower cost of loans. Second, upon adoption of weather derivatives contracts, firms are found more likely to pledge collateral and accept more restrictive covenants due to their improved financial confidence. Signaling through these non-price terms also helps firms obtain lower loan spreads. We also observe that hedging with weather derivatives can reduce the likelihood of ex-post covenant violations, despite more restrictive covenants.

We mitigate the endogeneity concern underlying a firm's decision to hedge with weather derivatives via three approaches, that is, propensity score matching, entropy balancing and IV estimation. Our results remain robust across these endogeneity-corrected methods. Additional findings reveal that the benefits of hedging with weather derivatives as reflected in lower loan costs are amplified when the borrowing firm has less complex financial reports or carries higher systematic risk, when markets and regulatory environments are more uncertain and when investors pay more attention to climate risks. The benefit from hedging with weather derivatives is also observable in the public debt market. Hedging firms enjoy lower yield spreads on their bond issues and maintain a lower percentage of bank debts relative to total debts. Weather derivatives contracts appear to help firms effectively hedge their risk exposure as they exhibit lower volatility in ROA and ROE and higher Z-scores.

Our findings shed light on another dimension of hedging benefits by evidencing the borrowing cost saving for firms that hedge through weather derivatives. The results suggest that management actions in reducing firm risk can have a material effect on creditors’ required premium, even among a well-regulated group of borrowers. Overall, this study reinforces the validity of fundamental credit market theories even in the presence of regulatory differences.

ACKNOWLEDGEMENTS

The authors would like to express their gratitude for the helpful comments received from the editor (Viet Dang), an anonymous reviewer, Tze Chuan Ang, Tongxia Li, Alberto Manconi, Guangqian Pan and participants at the 2020 Department of Banking and Finance (Monash University) Brown Bag Seminar, 2020 ADBI-JBF-SMU Joint Conference on Green and Ethical Finance, 2021 FIRN Annual Conference, and 2022 FMCG Conference. We thank Xing Yan for her help with manual data checks. We gratefully acknowledge the financial support from the Accounting and Finance Association of Australia and New Zealand (grant number AFAANZ-2020-R2-017). Errors and omissions are our sole responsibility.

Open access publishing facilitated by Monash University, as part of the Wiley - Monash University agreement via the Council of Australian University Librarians.

    CONFLICT OF INTEREST STATEMENT

    None.

    APPENDIX: Variable description

     

    Variable Description
    Weather derivatives and weather risk measures
    WD_Post An indicator variable taking the value of 1 for observations of weather derivatives user firms after they first enter weather derivatives contracts (based on 10-K filings), and zero otherwise
    WD_User A dummy variable that takes the value of 1 for loans made to weather derivatives user firms, and zero otherwise
    WD_Post_Alt A binary variable taking a value of 1 for loans made to weather derivatives users after they first enter weather derivatives contracts, based on all filings (rather than 10-K filings only) submitted to the SEC
    Absolute EDD Beta Absolute value of the coefficient that estimates the pre-1997 sensitivity of firm revenue to changes in energy demand
    EDD Standard Deviation Standard deviation in EDD, where EDD is quarterly number of energy degree days for each firm based on its headquarter location
    EDD Weather-Induced Volatility Pre-1997 weather risk exposure, which equals Absolute EDD Beta multiplied by EDD Standard Deviation
    Loan characteristics
    AISD All-In-Spread-Drawn, which represents the interest rate margin over LIBOR on drawn loan amount plus annual fees
    Log(AISD) Natural logarithm of All-In-Spread-Drawn, which represents the interest rate margin over LIBOR on drawn loan amount plus annual fees
    Log(Facility Amount) Natural logarithm of loan facility amount
    Log(Maturity) Natural logarithm of loan maturity in number of months
    Secured Dummy A binary variable taking the value of 1 for secured loans and zero for unsecured loans
    Revolver Dummy A binary variable taking the value of 1 if the loan facility is a revolving facility and zero otherwise
    Lender Number Number of lenders in a loan
    Covenant Index Bradley and Roberts (2015)’s count variable that takes a value between 0 and 6 based on 6 key covenant categories
    Covenant Strict A binary variable based on Covenant Index that takes the value of 1 if the loan facility carries three or more covenant categories and zero otherwise
    First_Year_Violation A dummy variable taking the value of 1 if the loan has a covenant violation within the first year of loan origination and zero otherwise
    PVIOL_ALL Aggregate probability of covenant violation (Demerjian & Owens, 2016), which considers the number of covenants as well as the initial slack of each covenant in a loan contract based on 15 financial covenants
    PVIOL_PCOV Probability of covenant violation (Demerjian & Owens, 2016), which considers the number of covenants as well as the initial slack of each covenant in a loan contract based on performance covenants
    PVIOL_CCOV Probability of covenant violation (Demerjian & Owens, 2016), which considers the number of covenants as well as the initial slack of each covenant in a loan contract based on capital covenants
    Borrower characteristics
    Log(Assets) Natural logarithm of borrower's book value of total assets adjusted for inflation in year 2000 dollars
    ROA Borrower's return on assets
    Leverage Borrower's leverage ratio calculated as long-term debt plus current liabilities, divided by book value of total assets
    Current Ratio Borrower's current ratio calculated as current assets divided by current liabilities
    Interest Coverage Borrower's ratio of EBITDA over interest expenses
    Log(1+Coverage) Natural logarithm of 1 plus Interest Coverage
    Profitability Borrower's ratio of EBITDA over sales
    Tangibility Borrower's ratio of property, plant and equipment over total assets
    Beta Borrower's CAPM beta that measures systematic risk of the year prior to loan origination
    Low_Rating A dummy variable taking the value of 1 if the borrower is rated BBB- or below
    CC_Exposure Sautner et al. (2023)’s firm-level climate risk exposure measure based on earnings conference calls
    Report Length Natural logarithm of number of words in the firm's 10-K filings
    Gunning Fog Index Measure of grade-level readability, developed by Gunning (1952)
    RIX Index Measure of text readability based on number of words of 7 or more characters per sentence, developed by Anderson (1983)
    Flesch-Kincaid Index Measure of text readability based on number of words, number of syllables and number of sentences, developed by Kincaid et al. (1975)
    Bank_Debt_Ratio The ratio of the firm's bank debt to total debt
    Change in ROA Yearly change in the firm's ROA
    Change in ROE Yearly change in the firm's ROE
    Altman Z Borrower's risk calculated as 3.3 × (EBIT/Total assets) + 0.99 × (Net sales/Total assets) + 0.6 × (Market value of equity/Total liabilities) + 1.2 × (Working capital/Total assets) + 1.4 × (Retained earnings/Total assets)
    Other variables
    GFC A dummy variable taking the value of 1 if the loan is originated between July 2007 and June 2009 and zero otherwise
    USREC A dummy variable taking a value of 1 if the loan is originated when the US economy is in recession and zero otherwise
    Deregulation A dummy variable taking a value of 1 for gas (electricity) loans originated after the state-level deregulation of its gas (electricity) market and zero otherwise
    High_CP_Index A dummy variable taking the value of 1 for loans originated in states with more stringent climate policies (i.e., climate policy index in the top quartile) and zero otherwise (using Bergquist and Warshaw (2023)’s state-level climate policy index)
    CC_WSJ An innovation in the WSJ Climate Change News Index developed by Engle et al. (2020)
    Stern_Review A dummy variable taking a value of 1 for loans originated after the 2006 Stern Review and zero otherwise
    Bond_Yield _Spread The difference between the yield at issue and the yield of Treasury bonds with the same maturity
    Log_Yield_Spread Natural logarithm of the difference between the yield at issue and the yield of Treasury bonds with the same maturity
    Bond Maturity Natural logarithm of bond maturity in number of months
    Bond Issuance Amount Natural logarithm of bond issuance amount
    Bond Collateral A categorical variable indicating if the bond issue is secured, senior or subordinated
    Bond Covenant Dummy A dummy variable taking the value of 1 for bond issues with covenants and zero otherwise
    Bond Credit Rating Credit rating of the bond at issuance provided by S&P

    DATA AVAILABILITY STATEMENT

    The data that support the findings of this study are available from a number of third party providers that are cited in the paper. Restrictions apply to the availability of these data as they require a paid subscription.

    • 1 “Managing Climate Risk with CME Group Weather Futures and Options,” accessed from https://www.cmegroup.com/education/articles-and-reports/managing-climate-risk-with-cme-group-weather-futures-and-options.html.
    • 2 It is difficult to associate the use of financial derivatives with firms’ hedging purposes as firms can use financial derivatives for hedging or speculative purposes (Chernenko & Faulkender, 2011).
    • 3 We have also conducted manual checks of financial reports to confirm weather derivatives contracts are used for hedging purposes.
    • 4 For example, geographical diversification across states with differing weather conditions may help reduce weather risk exposure, but may increase energy production costs because of lower economies of scale across widespread locations. Alternatively, some firms hedge demand uncertainty with electricity futures and agricultural derivatives; however, these strategies most likely result in imperfect hedges, which increase basis risk significantly (Pérez-González & Yun, 2013).
    • 5 The first over-the-counter (OTC) derivative contract was created in 1997 between Enron and a group of utility firms.
    • 6 We thank Sudheer Chava and Michael Roberts for making the link file available via the Wharton Research Data Services. This link file has been used extensively in the literature (see, Hasan et al., 2014; Chava et al., 2017; Christensen et al., 2022; among others).
    • 7 We have manually read the filings where at least one keyword is identified to ensure the definition of weather derivatives user is correctly captured.
    • 8 It is possible that a borrower may choose to hedge with weather derivatives in 1998 but does not obtain a loan until 2005. Therefore, it is important to have our sample period long enough after 1997 to capture these observations. Notably, after the introduction of weather derivatives in 1997, electricity and natural gas firms have been active in adopting weather derivatives for hedging. By the middle of the 2010s, the take-up rate of weather derivatives hedging slowed down considerably. The number of new firms that commenced hedging with weather derivatives in 2014, 2015 and 2016 were 2, 0 and 2, respectively. Therefore, we expect that the results stay robust even if the sample could be extended beyond 2017.
    • 9 We re-estimate our baseline regression using the extended sample that includes loans granted until 2020 to matched borrowers in the Chava and Roberts (2008)’s Dealscan-Compustat link file. Our results remain robust and are presented in Table OA.1 of the Online Appendix.
    • 10 Coefficients of Log(Assets), Revolver Dummy and Lender Number in column (2), though not statistically significant, have the same signs as in column (1).
    • 11 In addition, we re-estimate our matching Probit model on the post-matched sample. The goodness of fit of the matching Probit model on the post-matched sample is another indicator of the quality of the matching process. We find that none of the control variables is statistically significant and the Pseudo-R2 is significantly smaller when compared with that of the pre-matched sample regression. This suggests that these control variables do not have significant explanatory power on weather derivatives usage in the post-matched sample. The regression output for this diagnostic test is available from the authors upon request.
    • 12 We re-estimate column (2), Panel C of Table 3 where we restrict our sample to include loans originated within 1 year of firms adopting weather derivatives contracts. This additional filter reduces our sample size to 133 loan facilities. While this is a relatively small sample, it provides us with a cleaner setting, given that firm fundamentals are less likely to experience significant shifts during the shorter time gap of 1 year before and after weather derivatives adoption. We report the results in Table OA.2 of the Online Appendix. Our results hold for this restricted sample. The coefficient of WD_Post remains negative and significant at the 1% level. The magnitude of the coefficient is −0.513, which is more than doubled that reported in the baseline regression.
    • 13 We thank the editor for suggesting this approach.
    • 14 The EDD Weather-Induced Volatility is estimated for all Compustat energy firms pre-1997. They are more than the number of borrowers in the main loan sample since loan data are not available for some energy firms.
    • 15 We thank the editor for suggesting useful measures of climate risk exposure that help enrich our analyses.
    • 16 Data on firm-level climate change exposure provided by Sautner et al. (2023) can be found at https://osf.io/fd6jq/.
    • 17 Though this argument is valid, it might not be a concern as it biases against our findings.
    • 18 See https://fred.stlouisfed.org/series/USREC
    • 19 See https://www.eia.gov/electricity/policies/legislation/california/pdf/restructure.pdf
    • 20 Tanlapco et al. (2002) document that energy market regulatory reforms result in heightened volatility in both energy prices and demand/supply factors. To mitigate this risk, energy firms have relied on a wide array of hedging instruments and strategies, involving commodity futures, commodity forwards, index futures and weather derivatives.
    • 21 Data for climate policy index are available at https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/PXWXWI.
    • 22 Gunning Fog Index = 0.4 ((# words/# sentences) + 100(# complex words/# words))Flesch-Kincaid Readability Index = 0.39 (# words/#sentences) + 11.8 (# syllables/# words) − 15.59RIX Readability Index = (#words of length 7 characters or more)/(# sentences)
    • 23 Engle et al. (2020) construct the WSJ climate change news index based on news coverage on climate change in the Wall Street Journal and estimate their innovations as the residuals from AR(1) model. Data for innovations in the WSJ Climate change news index, CC_WSJ, are available at: https://sites.google.com/view/stefanogiglio/data-code?authuser=0.
    • 24 We also adopt the data for innovations in Crimson Hexagon's negative sentiment climate change news index, which focuses on negative climate news, from Engle et al. (2020). Our results are qualitatively similar and available upon requests.
    • 25 Results are presented in Table OA.3 of the Online Appendix.
    • 26 We thank the editor for suggesting these additional tests.
    • 27 This analysis is conducted for a reduced sample size of 1402 firm-year observations due to the data requirements to measure the Altman's Z-score.
    • 28 We also examine the effect of weather derivatives hedging on borrowers’ future probability of default (measured by Rank_EDF). Rank_EDF is the ranking by quartiles of the expected default frequency (EDF) estimator, where EDF is estimated using the iterative procedure proposed by Bharath and Shumway (2008). We present results in Table OA.4 of the Online Appendix (the OLS model is employed in column 1 and the Poisson model in column 2). The coefficient of Fitted_WD_Post is negative in both columns and statistically significant in column (2) when the Poisson model is used.
    • 29 We acknowledge that the costs of covenant violations may be higher for weather derivatives user firms due to their higher climate risk exposure. Therefore, it is plausible that these firms may be more inclined to engage in manipulative behaviors such as financial engineering (Levi et al., 2021) and earnings management (DeFond and Jiambalvo, 1994) as ways to circumvent covenant violations, leading to lower incidence of covenant violations. For this alternative explanation to hold, we need to accept the premise that the cost of violations is higher for hedging firms and that hedging firms are more willing to engage in manipulative behaviors than otherwise similar firms. We thank an anonymous reviewer for highlighting this alternative possibility.
    • 30 We thank Peter Demerjian and Edward Owens for making their data available online. See Demerjian and Owens (2016) for detailed discussion of the 15 covenants used in the construction of PVIOL_ALL. As not all covenants are reported, our sample size drops to 270 loan observations.

    The full text of this article hosted at iucr.org is unavailable due to technical difficulties.