Derivatives Supply and Corporate Hedging: Evidence from the Safe Harbor Reform of 2005

Abstract

Are supply-side fluctuations in the derivatives market important for corporate risk management? If the supply of hedging instruments is perfectly elastic, then hedging levels are determined exclusively by a company’s demand for hedging. Evidence suggests, however, that the supply of hedging instruments is not frictionless. Survey evidence in Giambona et al. (2018), for example, suggests that financial executives from around the world consider derivatives supply conditions an important factor in making corporate hedging decisions. Combined with the presence of hedging frictions, such as the risk of financial distress (Purnanandam 2008; Smith and Stulz 1985; Stulz 2013) or information asymmetry (DeMarzo and Duffie 1991, 1995; Breeden and Viswanathan 2016), these supply-side fluctuations can affect corporate hedging policies, firm value, and performance.

Establishing a causal link between supply frictions and corporate risk management is challenging because it requires an exogenous shock to derivatives supply. In this study, we exploit a regulatory change that significantly strengthened the protection granted to nondefaulting derivatives counterparties in bankruptcy, essentially allowing them to circumvent the Bankruptcy Code’s automatic stay and preference rules (Schwarcz and Sharon 2013). These regulatory innovations, which we collectively call the Safe Harbor Reform of 2005, were introduced with the Bankruptcy Abuse Prevention and Consumer Protection Act of 2005 (BAPCPA) and have been embraced by numerous bankruptcy court decisions (Levin 2015).

The stronger protection granted to nondefaulting derivatives counterparties in Chapter 11 after 2005, both in terms of the right to terminate a derivatives contract and to take the collateral if the other party files for bankruptcy, means that counterparty risk is effectively lower after the reform. Therefore, the “supply” of hedging instruments in favor of firms that could potentially face financial distress, such as firms with a low Altman (1968) z-score, should increase.

Purnanandam (2008) develops a model in which optimal ex post hedging is determined by a trade-off between the costs of financial distress and the benefits from risk shifting. The author shows that, in a dynamic setting, it is optimal for firms nearing financial distress to hedge ex post (even without a precommitment to do so) because, by hedging, such firms stabilize their financial situation and thus are able to preserve their market share.¹|$^{,}$|² The model therefore predicts that corporate hedging and operating performance will increase for financially distressed firms (treated firms) relative to financially sound firms (control firms) after the Safe Harbor Reform of 2005.

We start our analysis by focusing on scheduled airlines (SIC 4512).³ This industry provides an ideal setting for studying corporate risk management, for the following reasons. First, jet fuel is one of the main production factors for airlines. Second, airline companies report detailed information on fuel hedging in their 10-Ks, which we hand-collect. Third, because over-the-counter (OTC) derivatives are easier to customize, airlines nearly exclusively rely on OTC products for their hedging needs. This is important for our identification strategy because the 2005 reform affected mainly noncleared OTC derivatives contracts, which are negotiated privately and hence carry significant counterparty risk. Fourth, more than 60% of the airlines in our sample are near financial distress, that is, have a low Altman (1968) z-score. Because the Safe Harbor Reform facilitates access to derivatives for firms that could potentially default, we should expect the effect of the reform to be particularly strong in the airline industry. Fifth, the focus on a single industry makes it less likely for differences in economic fundamentals across industries to explain changes in risk management policies.⁴

Using a difference-in-differences approach, we find that fuel hedging by financially distressed airlines (those that benefitted most from the 2005 reform) increased by 18.7 percentage points (pp) in the 3 years following the reform, compared to financially sound firms (the control group) (Purnanandam 2008). A common concern with any difference-in-differences estimates is whether treated and control group outcomes followed a parallel trend prior to the treatment. In support of our experiment validity, we find no indication of treated-firm specific trends in the pre-reform period.

A series of robustness tests show that our results hold when we use alternative proxies of financial distress, such as a firm’s distance to default. Our results are also robust to alternative sample selection criteria, such as the inclusion of foreign airlines, the exclusion of regional airlines, extension of the sample period, or the exclusion of potential outliers. Finally, our results are robust to controlling for other potential determinants of hedging (different from the supply-side shock that we consider in the paper), such as financial constraints or executive compensation.

As we have discussed, focusing on the airline industry to study risk management has several advantages. One concern with any single-industry study, however, is that it is not possible to know whether the results are generalizable to other industries. To investigate the external validity of our findings, we replicate our main results for a large sample of nonfinancial firms from Compustat. Although detailed information on hedging is not available for such samples, Compustat reports information on gains/losses associated with the use of derivatives. Following Adams-Bonaimé, Watson-Hankins, and Harford (2014), we use this information to build an indicator for whether firms hedge.

Using a linear probability difference-in-differences approach, we find that the propensity to hedge of financially distressed firms increased by 4.1 pp in the 3 years following the reform, relative to (otherwise similar) financially sound firms. There is no indication that a potential violation of the parallel trend assumption is driving our results. Further, we find that our results are robust to controlling for industry fixed effects, alternative measures of financial distress, and matching treated firms to untreated firms based on relevant characteristics.

To study whether operational hedging changed for the treated firms after the reform, we went through nearly 27,000 photos of the aircraft in our sample to collect information on the installation of winglets, a leading fuel-efficiency improvement device, in the pre- and post-reform periods. Efficiency devices, such as winglets, operate by reducing fuel consumption and thus lower the volatility of fuel expenses. We find that treated airlines installed significantly fewer winglets after 2005. We likewise find that fleet age and fuel consumption increased for the treated group after 2005. In line with recent evidence in Gilje and Taillard (2017), Almeida, Watson-Hankins, and Williams (2017), and Hoberg and Moon (2017), these findings suggest that, following the reform-induced increase in derivatives hedging, operational hedging efforts became less common for treated firms.⁵ In line with Purnanandam (2008), we further find a significant increase in the revenues and operating performance of financially distressed airlines in the years after the 2005 reform. To support their growth, these firms expanded their fleets and increased capital expenditures.

Our article belongs to the literature on the role of supply-side frictions for corporate policies. In the capital structure literature, Faulkender and Petersen (2006), Leary (2009), and Lemmon and Roberts (2010) show that credit market frictions affect corporate borrowing. Numerous empirical studies consider corporate risk management (e.g., Bessembinder 1991; Nance, Smith, and Smithson 1993; Tufano 1996; Mian 1996; Gay and Nam 1998; Geczy, Minton, and Schrand 1997; Graham and Rogers 2002; see, more recently, Adam 2009; Bartram, Brown, and Conrad 2011; Rampini, Sufi, and Viswanathan 2014; Almansour, Megginson, and Pugachev 2019); however, they focus on corporate demand for hedging. To our knowledge, our paper is the first to study the nexus between derivatives supply and corporate hedging.

Our article is also related to the literature on the real and financial effects of corporate risk management. Gilje and Taillard (2017) show that, following an exogenous increase in basis risk, oil and gas firms reduced hedging, which, in turn, led to lower investment, valuation, and financing for the affected firms. Focusing on energy firms, Pérez-González and Yun (2013) show that the introduction of weather derivatives tied directly to the energy sector (by reducing basis risk) made it easier for weather-sensitive firms to access credit and invest. Similarly, Cornaggia (2013) finds that the introduction of new crop insurance policies led to an increase in the productivity of the U.S. agricultural sector. In earlier studies, Adam and Fernando (2006) find that risk management leads to higher cash flows, whereas Adam (2002) and Campello et al. (2011) show that hedging helps firms increase investment by increasing their access to internal resources and lowering borrowing costs, respectively. Unlike previous studies, we focus on contracting frictions and how improvements in the counterparties’ legal rights in the event of default facilitate the use of derivatives by corporate end users. A central question in the passage of these laws is the welfare gains enjoyed by other parties. Our paper contributes to that debate: to the extent that financially distressed firms benefit from hedging their exposure, safe harbor provisions enhance welfare.⁶

Our findings can also help inform the current policy debate on derivatives’ margin requirements. Uncollateralized derivatives are considered to have played an important role in the global financial crisis. The sale of uncollateralized credit default swaps, for example, is considered to have contributed to the collapse of AIG in 2008. In response, the Dodd-Frank Act of 2010 required the U.S. prudential regulators to adopt rules requiring derivatives markets participants to collect margins. While this might improve the stability of financial markets, our findings suggest that, by limiting the supply of hedging instruments, more stringent margin requirements can affect corporate hedging and firm performance. Ultimately, our article can help inform the current policy debate by highlighting the need to balance market stability with the consequences that more stringent margin requirements might have for corporate risk management and firm performance.

1. The Treatment of Derivatives in Bankruptcy after the Safe Harbor Reform of 2005

To study the relation between supply-side frictions and corporate hedging, we rely on several important changes in the Bankruptcy Code’s treatment of derivatives that were introduced with BAPCPA (Pub.L. 109–8, 119 Stat. 23). The act was enacted on April 20, 2005, under President George W. Bush, and it went into effect on October 17, 2005. We call these changes, collectively, the “Safe Harbor Reform of 2005.”

In the U.S. Bankruptcy Code, all creditors, including secured creditors and lessors, are subject to the automatic stay (which halts actions by creditors to collect debts from a debtor that has filed for Chapter 11) and other provisions of the Bankruptcy Code. Derivatives counterparties are an important exception: they have been exempted from some of the provisions of the Bankruptcy Code since the early 1980s. Prior to 2005, however, there was significant uncertainty about the degree of protection derivatives counterparties should receive in Chapter 11. For example, courts were split on the extent to which nondefaulting derivatives counterparties could terminate a contract with a debtor in Chapter 11 and seize the underlying collateral. There was also uncertainty about whether newly designed derivatives securities and certain types of new financial market participants would fit the categories listed in the Bankruptcy Code and hence whether they should be granted safe harbor protection (Vasser 2005). The Safe Harbor Reform of 2005 resolved this uncertainty by clarifying the extent of the applicability of the safe harbor provisions. For example, the 2005 reform clarified that the automatic stay does not apply to pledged collateral in connection with derivatives contracts (Vasser 2005; Speiser, Olsen, and Rae 2005).

In addition to broadening the applicability and scope of the safe harbor provisions, the 2005 act also expanded the list of financial counterparties that can be granted safe harbor protection to include practically all systemically important institutions. Prior to the 2005 reform, only the types of institutions explicitly listed in the Bankruptcy Code were granted safe harbor status in Chapter 11. BAPCPA created a general definition of market participant eligible for safe harbor protection to include any entity that, at the time it enters a derivatives contract, holds a notional total of |${\$}$|1 billion in derivatives transactions or gross mark-to-market positions of not less than |${\$}$|100,000,000 (aggregated across counterparties), in one or more agreements with the debtor (or any other entity other than an affiliate) on any day in the 15 months prior to bankruptcy.⁷Internet Appendix Section I.A discusses several other provisions introduced by the 2005 reform to strengthen the protection of derivatives counterparties in bankruptcy.

In sum, the 2005 reform clarified that nondefaulting derivatives counterparties can terminate or liquidate a contract, set off and net out mutual debts and claims, and liquidate and realize upon any collateral held by the defaulting counterparty. The act also clarified that properties transferred to a nondefaulting counterparty prior to Chapter 11 in connection with a derivatives contract do not have to be returned to the bankruptcy estate (unless such a transfer was actual fraud). In addition, the Act substantially expanded the type of securities and market participants that are granted safe harbor protection in Chapter 11. As we discuss in Internet Appendix Section I.A, numerous bankruptcy courts have embraced the stronger protection of derivatives contracts in bankruptcy introduced with the 2005 reform. Further, the importance of the safe harbor reform has been echoed by many bankruptcy experts.

In our identification strategy, we argue that because the 2005 reform reduces counterparties’ risk in Chapter 11, these counterparties “expand their supply” of hedging instruments to firms that could face financial distress (firms with a low Altman z-score) and file for bankruptcy. Hence, we expect hedging to increase for financially distressed firms relative to financially sound firms after the Safe Harbor Reform of 2005, in line with the theoretical insights in Purnanandam (2008).

Finally, we note that the 2005 reform is the outcome of derivatives industry lobbying that started with the near-collapse of long-term capital management (LTCM) in August 1998. Following the LTCM event, the Working Group on Financial Markets issued a report on the LTCM crisis, urging Congress to expand the safe harbor provisions in the Bankruptcy Code in order to improve market stability⁸ (which led to BAPCPA). This is important for our identification strategy, because it suggests that the reform is not a response to an anticipated increase in the demand for hedging instruments by nonfinancial end users (which would have been problematic) but, rather, a change implemented to increase the stability of the derivatives market.

2. Empirical Design and Data

To test the effect of the 2005 reform on hedging, we hand-collect fuel-hedging data for the passenger airline industry (SIC 4512). Later, we discuss the external validity of our findings for a general sample of nonfinancial firms. The airline industry provides an ideal setting for our tests for several reasons. First, airlines report the percentage of next-year fuel expenses hedged in Item 7(A) of their 10-K SEC filings, in the section “Quantitative and Qualitative Disclosures about Market Risk.” To our knowledge, similar hedging information is not available for other industries during our sample period.⁹ Second, jet fuel is one of the main operating expenses for airlines. In 2008, for example, fuel expenses represented 31.5% of operating expenses, compared to 20.3% for labor expenses, the second largest operating expense (Source: Airlines 4 America).

Third, airlines have traditionally used OTC instruments, such as swaps, for their hedging needs because these products are more easily customizable. Using data directly collected from annual reports, we find that, for the period 2003–2008, all of the airlines in our sample rely on one or more types of OTC instruments (especially swaps, forwards, collars, and options), with only one airline (i.e., Northwest Airlines Corp.) also using futures contracts in the years 2006–2008 in addition to OTC swaps and options. These findings are important for our identification strategy, because the Safe Harbor Reform of 2005 affected mainly noncleared OTC derivatives contracts, which are traded directly between airlines and investment banks and hence incorporate significant counterparty risk.

In line with our identification strategy, aggregate data on the notional amount of all derivatives contracts (OTC and exchange traded) by U.S. commercial banks and trust companies from the Office of the Comptroller of the Currency shows that only the OTC derivatives market was affected by the reform. As Figure 1 shows, the OTC derivatives market grew, on average, by nearly |${\$}$|118 trillion in 2006–2008, compared to |${\$}$|57 trillion in 2003–2005. By contrast, the exchange traded derivatives market experienced an average negative growth of |${\$}$|0.7 trillion in 2006–2008, compared to the average positive growth of |${\$}$|4 trillion in 2003–2005.

Figure 1

Year-to-year change in the notional amount of derivatives contracts by U.S. commercial banks

This figure displays the year-to-year change in the notional amount of derivatives contracts (⁠|${\$}$| trillions) by U.S. insured commercial banks and trust companies in the 6 years around the Safe Harbor Reform of 2005. The information is presented for total derivatives, OTC derivatives (swaps, options, forwards, and credit), and exchange-traded derivatives (futures and options). Source: The data come from the Office of the Comptroller of the Currency (derivatives quarterly reports).

Open in new tabDownload slide

Fourth, about 62% of the airlines in our sample could face financial distress, compared to about 22% of nonfinancial firms. Because the Safe Harbor Reform is expected to facilitate access to derivatives for financially distressed firms, the reform should have a stronger effect in the airline industry. Fifth, focusing on one industry makes it less likely that differences in economic fundamentals across industries are the reason for changes in risk management.

Table I.B.1 in the Internet Appendix lists the twenty airlines in our sample, their average fuel hedged and fuel expenses during the period 2003–2008, information on whether the airline obtains fuel from another (usually larger) airline through a pass-through agreement, and information on the first and last years the airline is in the sample during the period 2003–2008.

We combine hand-collected data on fuel hedging with data from several other sources. We gather stock return and accounting data from the Center for Research in Security Prices (CRSP) and Compustat. Airline segment data are from Compustat’s Industry Specific Annual database. Derivatives exposure data of the airline’s lenders are from DealScan’s and Standard & Poor’s (S&P) Global SNL Financial databases. Fleet-level data are hand-collected from 10-K filings, Item 2, Properties, while airline-level information on flights, routes, and pilot salaries is from the Web site of the Bureau of Transportation Statistics of the U.S. Department of Transportation. We manually collect wingtip-device data from JetPhotos.com, while obtaining purchase obligation information for the nonfinancial sample by parsing 10-K footnotes. Airline cost structure data are from Airlines 4 America, while aggregate derivatives data are from the Office of the Comptroller of the Currency. Finally, we hand-collect fuel hedging data for foreign airlines from the annual reports on the airlines’ Web sites or the Web sites of the stock exchanges in which the airlines are listed. We combine the fuel hedging data for foreign airlines with accounting data from Compustat Global.

Post2005 is an indicator equal to one for the fiscal years after 2005,¹⁰ and |$y_{i}$| and |$z_{t}$| are firm and year fixed effects, respectively. Our main analysis focuses on the sample period 2003–2008: a 6-year time window centered on 2005. In our robustness tests (Internet Appendix Section I.C), we also perform our analysis for the sample periods 2002–2009 and 2004–2007. The focus of our analysis is on Fin. Distress |$\times $| Post2005 (our difference-in-differences estimator), which we expect to enter the estimation with a significantly positive coefficient.

Our basic set of control variables includes the following company characteristics: (1) Size is the natural logarithm of sales; (2) Fuel expenses is the ratio of fuel expenses to total operating expenses; (3) Profitability is the ratio of operating income before depreciation and amortization to book assets; (4) Cash is the ratio of cash and cash equivalents to book assets; (5) Tangibility is the ratio of property, plant, and equipment to book assets; and (6) Net worth is the ratio of stockholders’ equity to book assets. These control variables are defined following standard practice in risk management studies. See Table I.B.2 in the Internet Appendix for detailed definitions of all the variables.

Table 1 reports basic descriptive statistics of fuel hedging and control variables for financially distressed (i.e., treated firms) and financially sound airlines (control firms). The table shows that financially distressed airlines hedge significantly less than control firms (24.8% vs. 72.7%). Table 1 also shows that treated airlines and control firms are similar in terms of size, fuel expenses, and profitability but differ with respect to cash and other firm characteristics. As we discuss later, we perform several robustness tests to mitigate the concern that some of these differences could bias our results. Table I.B.3 in the Internet Appendix reports detailed descriptive statistics of all the variables used in the article for the combined sample, as well as for treated and control airlines.

3. Fuel Hedging for Financially Distressed Airlines after the Safe Harbor Reform of 2005

In this section, we examine the effect of the Safe Harbor Reform of 2005 on corporate hedging for financially distressed airlines (treated firms) relative to financially sound airlines (control firms) by estimating Equation (1), a difference-in-differences model.

Across all three estimations in Table 2, panel A, the coefficient on the interaction term of interest, Fin. Distress |$\times $| Post-2005, is positive and statistically significant at the 5% level or higher. In line with our prediction, this finding indicates that the stronger protection granted to nondefaulting derivatives counterparties in Chapter 11 after 2005 led to an increase in fuel hedging for airlines that were less financially sound. The effect is also economically large. We focus on Column 3 (estimation with all control variables) and find that the coefficient of 0.187 (statistically significant at the 1% level) suggests that, following the 2005 reform, financially distressed airlines increased the fraction of fuel expenses hedged by 18.7 pp relative to control firms.

Turning briefly to the control variables, we note that they generally have the sign predicted by theory (e.g., Froot, Scharfstein, and Stein 1993; Holmström and Tirole 2000) but lack statistical significance. The lack of statistical significance is perhaps not that surprising given that we are performing a within-firm estimation for a sample of 20 airlines over a relatively short period (i.e., 2003–2008).

In our identification strategy, the Safe Harbor Reform of 2005 facilitates access to hedging for firms nearing financial distress, but not necessarily firms with weak operating performance. To investigate this issue, in addition to the Altman z-score (which combines financial and operating elements), we use leverage and operating cash flows and build two indicators. Fin. distress-lev., is an indicator equal to one if, in 2005 (the last pre-reform year), leverage for an airline is above the sample median, and zero otherwise, and Ops. distress-cash flows, is an indicator equal to one if, in 2005, operating cash flows for an airline are below the sample median, and zero otherwise. We then interact each of these two indicators with Post-2005 and reestimate our fuel-hedging regression, Equation (1).

Table 2, panel B, presents the results from this estimation. Across all three estimations, the coefficient on Fin. distress-lev. |$\times$|Post-2005 is positive, economically sizable, and statistically significant at the 5% level or higher, while the coefficient on Ops. distress-cash flows |$\times$|Post-2005 is economically small and never statistically significant. In line with our identification strategy, these findings suggest that the 2005 reform facilitated access to fuel hedging for the airlines nearing financial distress, but not necessarily firms with weak operating performance.

3.1 Experiment validity

A key assumption of any difference-in-differences estimations is that the outcome variable for treated and control firms follows a parallel trend prior to the treatment. In our setting, the parallel trend assumption requires that, prior to the 2005 reform, fuel hedging for treated and control airlines follows a parallel trend. A violation of this assumption could be problematic because it would suggest that a trend specific to financially distressed firms, rather than the reform, is the reason that hedging increased for treated firms. To assess this assumption formally, we estimate Equation (1) by adding interaction terms of the Fin. Distress indicator with dummy variables for the years 2004–2008, with 2003 as the omitted case (e.g., Autor 2003; Gormley and Matsa 2011). Figure 2, panel A, plots the coefficients on these interaction terms together with 95% confidence intervals. There is no indication of a change in the fuel hedging of treated firms relative to control firms prior to the 2005 reform. However, after the reform hedging increased for financially distressed airlines relative to the control firms. This increase is slower (but sizable and statistically significant) in the year immediately after the reform and accelerates in the second and third years following the reform.

Figure 2

Fuel hedging in the period around the Safe Harbor Reform Act of 2005: Fin. distress versus nonfin. distress airlines

This figure reports the point estimates from fuel-hedged regressions. The sample includes all firms with SIC 4512 (scheduled airlines). Refer to Table I.B.2 for detailed variable definitions. The regression specifications used in panels A, B, and C are the same to those reported in the panels to Table 2, respectively, except that the effects of Fin. Distress, Fin. Distress-Lev., and Ops. Distress-Cash Flows are allowed to vary by year for each year starting 2 years prior to the Safe Harbor Reform and ending 3 years after the reform. Ninety-five percent confidence intervals are also plotted.

Open in new tabDownload slide

Panels B and C of Figure 2 plot, respectively, the coefficients on the interactions of Fin. distress-lev. and Ops. distress-cash flows with year dummies from a regression similar to Equation (1). In panel B, there is no indication of a trend in fuel hedging for Fin. distress-lev. airlines prior to the reform. However, after the reform, fuel hedging increased for the treated firms. In panel C, there is no indication of a trend in fuel hedging for Ops. distress-cash flows airlines either prior to or after the reform. Overall, the evidence in Figure 2 mitigates the concern that a trend in the hedging policies of treated firms relative to control firms could explain the findings in Table 2.

As an additional check, we hand-collect additional fuel-hedging data and reestimate our base hedging model over the following 6-year windows: 1998–2003, 1999–2004, 2000–2005, 2001–2006, and 2002–2007. If there were a trend in hedging specific to financially distressed airlines prior to 2005, we should find this effect to be economically sizable in these “placebo” pre-reform windows (Roberts and Whited 2012). We find that the coefficients on the interaction terms of interest are always insignificant in these placebo estimations with either the Fin. distress indicator (Table 3, Columns 2–6) or the Fin. distress-lev. and Ops. distress-cash flows indicators (Internet Appendix Table I.B.4, Columns 2–6). Overall, this analysis allows us to rule out any positive trend in hedging for financially distressed airlines prior to the Safe Harbor Reform of 2005.

3.2 Effect of preexisting lending relationships

As we have discussed, the 2005 reform facilitated access to hedging for firms that could face financial distress. One could argue that airlines in a lending relationship with banks that are active in the commodity derivatives market could have had access to fuel hedging derivatives even prior to the reform. Alternatively, these banks could have used their pre-reform relationship with airlines to increase the supply of fuel derivatives following the reduction in the cost of financial distress associated with the Safe Harbor Reform.¹¹

To test the effect of the reform on hedging for treated firms in a lending relationship, we use DealScan to identify banks involved in syndicated loans granted to our sample airlines (e.g., Ivashina and Scharfstein 2010; Aragon and Strahan 2012; Fernando, May, and Megginson 2012; and, more recently, Carvalho, Ferreira, and Matos 2015). We focus on loans originated in 2005 or in the previous 5 years, but still outstanding in 2005. For each of these banks, we then collect information from the S&P Global SNL Financial database on the notional amount in commodity derivatives reported on the balance sheet at the end of 2005.¹² We categorize an airline as having a lending relationship with a commodity derivatives lender if its syndicated loan lenders report a nonzero notional amount in commodity derivatives at the end of 2005. We identify four such airlines in 2005. Next, we interact our measure Fin. distress |$\times $|Post-2005 with the commodity derivatives-lender indicator and estimate a difference-in-difference-in-differences version of Equation (1).

As Table 4 shows, the coefficient on Fin. distress |$\times$|Commodity-derivative lender |$\times $|Post-2005 is positive, significant, and economically sizable. Focusing on Column 4, we find a coefficient of 0.277 on the triple interaction term, which suggests that financially distressed airlines in a lending relationship with banks active in the commodity derivatives market increased hedging by nearly 28 pp after the reform relative to control firms. The table also shows that financially distressed airlines increased hedging by 13 pp (coefficient on Fin. distress |$\times $|Post-2005), while hedging decreased on average by almost 12 pp for control airlines in a lending relationship with a commodity derivative lender. Overall, these findings suggest that the expansion of hedging instruments for airlines nearing financial distress was stronger if these firms had a preexisting lending relationship with a commodity derivatives lender.

Our results are consistent with the following argument in the banking literature. Relationship banks have an information advantage over nonrelationship banks about end users. This could help explain why relationship banks are able to increase the supply of derivatives in favor of riskier end users more than nonrelationship banks after the reform, in line with evidence in the banking literature that the availability of credit is higher for firms in a lending relationship (e.g., Petersen and Rajan 1994; Mester, Nakamura, and Renault 2007; Drucker and Puri 2009) and the theoretical argument in Leland and Pyle (1977) and Diamond (1984, 1991).

Furthermore, if supplying derivatives to riskier end users is more beneficial than supplying derivatives to safer end users (for instance, because default risk for riskier end users goes down if they hedge), relationship lenders might increase the supply of derivatives to the riskier firms by supplying less derivatives to the safer firms in order to contain risk-based capital requirements (which are typically high for OTC derivatives exposures). This could help explain the evidence in Table 4 that hedging decreased for financially sound end users in a relationship with a commodity derivatives lender after 2005.

3.3 Robustness analysis

In this section, we discuss tests performed to assess the robustness of our results to (a) alternative proxies of financial distress, (b) sample selection issues, and (c) other potential determinants of hedging.

3.3.1 Alternative proxy of financial distress

In Table 5, we use Merton’s (1974) distance to default (Vassalou and Xing 2004; Bharath and Shumway 2008), instead of Altman (1968) z-score to assess financial distress. In Columns 1–3, High-default-probability is an indicator equal to one if, in 2005, the distance to default for an airline is less than the sample’s first decile, and zero otherwise. In Columns 4–6, we partition Merton’s distance to default into its leverage and asset volatility components. We use these partitions to build two indicators. High-default-probability (Leverage) is an indicator equal to one if, in 2005, the leverage component of distance to default—the ratio of the sum of current liabilities and 0.5 |$\times $| long-term debt to assets (Vassalou and Xing 2004)—for an airline is above the sample’s 90th percentile, and zero otherwise. High-default-probability (Asset volatility) is an indicator equal to one if, in 2005, the asset volatility component of distance to default—the estimated volatility of the market value of the firm (Vassalou and Xing 2004)—for an airline is above the sample’s 90th percentile, and zero otherwise. In Columns 1–3, High-default-probability |$\times $|Post-2005 is positive and statistically significant at the 5% level or higher. Focusing on Column 3, we note a coefficient of 0.136, which suggests that, after the reform, fuel hedging increased by 13.6 pp for airlines with a high probability of default relative to control firms. In Columns 4–6, High-default-probability (Leverage) |$\times $|Post-2005 is positive, economically sizable, and highly statistically significant, whereas High-default-probability (Asset volatility) |$\times $|Post-2005 is negative and statistically insignificant. In line with our identification strategy and the leverage/operating cash flow partition results, this combined evidence suggests that financial distress (but not necessarily asset volatility) is the channel for the increase in fuel hedging for the treated group after the 2005 reform.

3.3.2 Sample selection issues

In our regressions, we include airline fixed effects to account for time-invariant unobservable differences between treated and control firms. In the regressions, we also control for observable time-varying firm characteristics by relying on a set of control variables that have traditionally been used in risk management papers. To further investigate whether unobservable time-varying differences between treated and control firms, rather than the reform, could be driving our results, we hand-collect fuel hedging and fuel expenses data for the foreign airlines (SIC 4512) in Compustat Global for which we are able to find an English annual report on the airline’s Web site (investor relations section) or the Web site of the stock exchange in which the airline is listed. The foreign sample with available data in Compustat Global that is necessary to estimate our hedging model includes 35 airlines from 27 countries and six macro regions, including Africa, Asia, the European Union, Extended Europe (European Union and airlines from Israel, Russia, and Turkey), Oceania, and South America (see Table I.B.5, panel A, in the Internet Appendix). Table I.B.5, panel B, contains the list of 14 airlines from Compustat Global without English annual reports. Table I.B.6 reports descriptive statistics for the foreign sample (with and without English annual reports).

Using these data, we estimate our difference-in-differences fuel hedging model for all of the airlines in the foreign sample, as well as separately for Asia, the European Union, Extended Europe, and all the other regions combined. We combine firms from Africa, Oceania, South America, and Canada, because we do not have sufficient observations to estimate our difference-in-differences model separately for each of these regions. Across all five estimations in Table 6, panel A, the coefficient on Fin. distress |$\times$|Post-2005 is economically small, often negative, and always statistically insignificant. These findings indicate that hedging did not increase for financially distressed foreign airlines after 2005, which is what one should expect given that the Safe Harbor Reform only affected U.S. airlines.

We also combine the U.S. airlines with the foreign sample. Using the combined sample, we estimate a difference-in-difference-in-differences fuel hedging model in which the variable of interest is Fin. distress |$\times $|Post-2005 |$\times $|U.S. airline. Table 6, panel B, shows that the coefficient for the triple interaction term is positive, economically sizable, and statistically significant across all five estimations. This finding suggests that after 2005, hedging increased for financially distressed U.S. airlines relative to financially distressed foreign airlines, which, again, is what one should expect, given that the reform only applied to U.S. airlines. By comparing hedging for financially distressed airlines in the U.S. sample and foreign samples, these findings further mitigate the concern that unobserved differences between treated and control airlines could be the reason hedging increased for financially distressed airlines in the U.S. sample after the Safe Harbor Reform of 2005.

To further assess the extent to which time-varying unobservable differences between treated and control firms could bias the coefficient on Fin. distress |$\times $|Post-2005, we rely on the estimator developed by Altonji, Elder, and Taber (2005).¹³ Using the notion that selection on unobservables is similar to selection on observables, this estimator can be used to calculate an upper bound on the extent of unobservable variable bias (for the coefficient on Fin. distress |$\times $|Post-2005) and identify the degree of selection on unobservables that would be needed to alter the economic significance for the effect of interest.

Following this technique, we find that the bias potentially introduced by time-varying unobservables is equal to 0.009 (with standard errors equal to 0.127). The size of the bias is small (and statistically insignificant) compared to the coefficient of 0.187, in Table 2, panel A, Column 3, suggesting that accounting for time-varying unobservables would not lead to any significant change in our coefficient of interest. This evidence has been interpreted in the literature as mitigating concerns about selection on unobservables. Further, the ratio between coefficient estimate and bias takes the value of |$20.8$| (⁠|$=0.187/0.009$|⁠). This result indicates that the role of unobservables that determine hedging for the treated group after 2005 would have to be more than 20.8 times the role of observables for the entire reform effect to be explained away by unobservables, which is not very likely. To put this in perspective, note that Altonji, Elder, and Taber (2005) estimate a ratio of 3.55, which they interpret as evidence of unobservables being unlikely to explain the entire effect that they document.

As we discuss in Internet Appendix Section I.C, we find further that our results hold when we keep in the sample only firms with a z-score either consistently below or consistently above 1.81 throughout 2005–2008, when we control for firms with a z-score between 1.81 and 2.99 (i.e., firms that are neither financially sound nor distressed in the original Altman’s (1968) classification), when we consider different sample periods (i.e., 2002–2009 and 2004–2007), when we exclude pass-through airlines, when we exclude one airline at a time, and when we add several other control variables to our basic hedging model.

3.3.3 Other potential determinants of hedging

Finally, we find that our results are robust when we account for other factors that could potentially affect hedging. Namely, we find that our results hold when we account for heterogeneity in leasing practices, when we control for pilot/copilot salaries, when we account for CEO vega/delta, and when we control for financial constraints. Internet Appendix Section I.C discusses these additional robustness tests.

3.3.3.1. Hedging propensity for a general sample of nonfinancial firms after the Safe Harbor Reform of 2005.

Column 1 of Table 7 shows that the coefficient on Fin. Distress |$\times $| Post-2005 (our variable of interest) is positive and statistically significant at the 1% level. The effect is also economically sizable. The coefficient of 0.041 suggests that, following the expansion in the supply of derivatives associated with the Safe Harbor Reform of 2005, the propensity to hedge for the treated group increased by 4.1 pp compared to the control group. Because the industry in which a firm operates is an important driver of hedging, we report, in Column 2, estimations of Equation (2) after adding industry fixed effects (2-digit SIC). The coefficients on the interaction term of interest increases slightly to 0.044 (although statistical significance decreases to the 5% level) in the specification with industry fixed effects compared to the base estimation in Column 1.