The real implications of commodity hedging
Evidence from the US oil and gas industry
Master thesis finance
University of Amsterdam | Business Economics: Finance
Abstract
This thesis examines the implications of commodity hedging for financing and investment. A hand-collected dataset on hedging activity in the US oil and gas industry is used to expose two channels through which hedging can affect firm value. Hedging reduces the chance on negative outcomes and thus reduces financial distress costs. In theory, this should be valued by creditors when providing loans to corporations. This thesis shows that hedging reduces loan spreads, and additional evidence suggests hedging also reduces the probability on capital expenditure restrictions in loan covenants. These favorable contracting terms allow firms with volatile income to increase investment spending. This analysis shows hedging has first-order effects on external financing and investment, and provides additional rationales for using derivative contracts.
Author J. van der Koelen*
Student number 10795529 Supervisor T. Homar Date 7 June 2015
* Email: jordy.vanderkoelen@uva.nl or j.vdkoelen@gmail.com
Faculty of Economics and Business Administration
1 Table of content 1 Introduction ... 2 2 Literature review ... 3 2.1 Hedging rationale ... 3 2.2 Commodity hedging ... 4 2.2.1 Commodity market ... 4
2.2.2 Commodity derivatives in the oil and gas industry ... 5
2.2.3 Operational hedging ... 5
2.3 Research on the impact of hedging ... 6
2.4 Hedging and contracting terms ... 7
2.5 Recapitulation ... 8
3 Methodology ... 8
3.1 Hedging and loan spread ... 8
3.2 Hedging and capital restrictions ... 9
3.3 Hedging and investments ... 10
3.4 Control variables ... 10
4 Data and descriptive statistics ... 12
4.1 Sample description ... 12
4.2 Databases ... 12
4.3 Data collection process ... 13
4.4 Descriptive statistics ... 15
5 Empirical results ... 17
5.1 Hedging and loan spread ... 17
5.2 Hedging and capital restrictions ... 20
5.3 Hedging and investment spending ... 22
5.4 Tobin’s Q ... 24
6 Robustness checks ... 26
6.1 Hedging instrument ... 26
6.2 Product-specific effects ... 27
7 Conclusions and future research ... 29
Bibliography ... 30
2
1 Introduction
Recently there have been several developments in the US oil and gas industry. First, after remaining relatively stable for four years at around $100, the price of crude oil has dropped by roughly 50 percent since mid-2014. Second, oil and gas producers have been able to issue substantial amounts of additional debt, as investors seem willing to lend against oil and gas reserves (BIS 2015). A third development has specifically constrained the US oil and gas industry. Even though prices have dropped dramatically, the OPEC1 increased production as part of a controversial strategy to maintain its market share. This has forced US oil and gas producers into the role of ‘swing-producer’, effectively making them adjust their production to reduce price fluctuations (Het Financieele Dagblad 2015). A severe drop in the oil price, increased debt levels, high volatility and an aggressive OPEC strategy put a strain on US oil and gas companies. The implications of using commodity derivative contracts to reduce uncertainty might be more important than ever.
Oil and gas – and commodities in general – share the characteristics that markets are arguably competitive and that products are fairly homogeneous. Individual firms in the commodity market have little influence on the sales price of their product, compared to other industries such as (producers of) consumer electronics or the service industry (Petersen & Thiagarajan 2000). Prices move with the market, unless instruments are used to lock in a certain price. In a Modigliani-Miller world of perfect capital markets, corporate risk management would be irrelevant. However, in real financial markets firms face a variety of frictions, making hedging strategies increasingly important in recent years. According to the Bank of International Settlement, the notional value of over-the-counter commodity derivatives contracts outstanding in June 2014 was $2.2 trillion. That is about four times the value it was in 2000 (BIS 2014).
Most prior studies focus on either the cross-sectional variance in the use of hedging instruments (e.g. Tufano (1996)), or on the effects of hedging on firm value (e.g. Allayannis & Weston (2001) and Carter et al. (2006)). The latter focuses on the effect of hedging activities on the value of the firm. This approach is a good approximation of the welfare that hedging can contribute. However, it is not a proxy of how it affects corporate wealth. This thesis extends the research on this topic, identifying two precise mechanisms through which hedging can affect firm value. Examining the access to external funding and the ability to invest freely will provide more insight into the real implications of hedging. This study focuses on the US oil and gas industry because it is particularly suitable. It allows for examination of a homogeneous risk exposure and hedging strategy, namely the reduction of risk stemming from fluctuations in oil and gas prices. This reduces the risk that results are driven by differences in firms risk exposures or hedging strategies.
To summarize the rationale and aim of the research in one sentence, a research question can thus be formed:
What are the real implications of using commodity derivatives contracts for US oil and gas producers?
1 The Organization of Petroleum Exporting Countries (OPEC) is an international organization comprised of 12 oil
producing countries. They have the ability to steer global oil prices by setting oil supply quantities. The US is not a member of this organization.
3
To investigate the channels through which hedging can affect firm value, I use a new hand-collected dataset obtained from several sources. Data is collected from credit agreements between firms and their lenders, as found in Securities and Exchange Commission (SEC) filings. Linking this to information on the use of hedging contracts allows me to identify the link between hedging, external financing and restrictions on investment. By incorporating other data sources I am able to build a dataset of around 60 firms with appropriate data on loan contracting terms, hedging programs, and firm- and loan-specific controls.
This study shows that firms with hedging programs in place pay lower interest rates on their loans. Firms with average hedging activity get a discount of around 50 basis points on their loan spread. Considering that the average loan spread is 164 basis points, this is reduction of around 30 percent. Furthermore I find evidence, albeit suggestive, that hedgers have lower probabilities on capital expenditure restrictions in their loan agreements. A third result is that hedging increases actual investment spending, if cash flow volatility is high.
My thesis confirms prior literature on the value adding capabilities of hedging programs. Campello et al. (2011) were the first to investigate specific channels through which hedging can affect firm value. But beyond currency and interest rate risk, often depicted as secondary risk (see Guay & Kothari (2003)), there is no evidence on the implications of commodity price risk. To my knowledge I am the first one to provide evidence on how hedging affects corporate outcomes. By thoroughly investigating credit agreements I unveil a creditors’ perspective on hedging, rather than the shareholders’ perspective that is upheld in prior literature. My thesis thus contributes to studies on loan contracts in the oil and gas industry by showing that hedging is an important driver of favourable contracting terms for loans.
The remainder of the thesis is laid out as follows: Section 2 reviews current literature. Section 3 defines the methodology used for the analyses, followed by a detailed description of the sample and the collection process in Section 4. Section 5 covers my results, which are tested for robustness in Section 6. Finally, Section 7 concludes.
2 Literature review
2.1 Hedging rationaleCorporate risk management is an important and widely used element in a firm´s overall business strategy. According to Stulz (1996) multiple theories of corporate risk management lead to the statement that “the primary goal of risk management is to eliminate the probability of costly lower-tail outcomes – those that would cause financial distress or make a company unable to carry out its investment strategy”. Using derivatives is one way of dealing with risk, with different instruments to cover numerous types of risk.
In the classic world of Modigliani and Miller with perfect capital markets, managing risk should be irrelevant. Anything the firm can accomplish through risk management can be easily mimicked by value-maximizing shareholders, since they have equal access to the same perfect capital market. However, in the real world firms face multiple frictions such as information asymmetry, bankruptcy cost, taxes and costly external financing. These frictions make a firm´s earnings more volatile, which is why firm´s often choose to deploy hedging strategies to reduce the volatility of their income (Jin & Jorion 2006).
4
Besides reducing earnings volatility there are other reasons why firms might want to use derivatives. Smith & Stulz (1985) show empirically that a convex tax function enables firms to reduce their tax liability by decreasing their earnings volatility. Convex tax functions arise with particular tax treatments of operational losses. As long as the cost of the hedging instrument does not exceed the tax benefits, derivatives can increase firm value. Leland (2013) confirms this and provides additional evidence for a linear relationship between tax convexity and hedging benefits. He also finds that hedging may increase a firm’s debt capacity, leading to a larger debt tax shield. Other than tax benefits, Smith & Stulz (1985) show that hedging can reduce financial distress cost. Volatility of cash flow can be the cause of a firm’s inability to meet payment obligations. If financial distress is costly then reducing the probability of this state will reduce the expected distress costs, thus increasing firm value. A third incentive for hedging is put forward by Froot et al. (1993). They say that hedging exposure reduces the reliance on external financing when a firm wants to invest in profitable investment opportunities. This is especially important when market frictions make external financing more expensive than internally generated funds.
In addition to operational issues there are agency problems that may lead to the use of derivatives. Stulz (1984) indicates that managers who receive a large portion of their income in out-of-the-money stock options have personal interest in ensuring stable future cash inflows. On the other hand, managers who receive most of their compensation in the form of cash might have the incentive to boost short-term cash flows. This means holding a large portion of undiversified firm risk, or even using derivatives to take on more risk, rather than reducing it. In general, managers who own stock might focus on the volatility of equity, while managers with earnings based bonuses might focus on earnings volatility (Guay & Kothari 2003).
2.2 Commodity hedging 2.2.1 Commodity market
Commodities are products for which there is demand, but which are supplied without significant qualitative differentiation across a market. Products classified as commodities range from iron ore and crude oil to salt and coffee beans. A particular submarket is the natural resource business, i.e. the extraction of oil and precious metals. It is a complex business for a number of reasons. First of all large investments need to be done to initiate new mining projects. Before the first extraction can be done, billions of dollars are needed to acquire land and equipment. Second, the products that companies in this sector produce are fairly homogeneous. This makes it difficult for a producer to differentiate the product from other producers. Because of this the best performing companies in the industry are those with efficient and low cost production (Sadorsky 2001). Consequently producers have little influence on the price of their product; the only impact they have is in cost reductions. Producers that do not pursue active risk management can experience severe problems with future income in the case of extreme price volatility (Fusaro 1998). Third, the natural resource business is dependent on depleting natural resources. Companies actively manage their resources and are continuously looking for new low cost deposits as it is vital for their survival.
To summarize: the natural resource business is very capital intensive, requires investments to be done long before any income is generated, and producers have little control over and influence on the price of the product. This makes the oil and gas industry an interesting field of study for hedging theories.
5
2.2.2 Commodity derivatives in the oil and gas industry
There are several hedging instruments that can be deployed in the oil and gas industry. Most common are forwards and futures contracts, swaps, options and price collars. All information in this Subsection is obtained from Hull (2006).
A forward contract is an over-the-counter (OTC) traded derivative, where the buyer and seller agree on the quantity, type of underlying asset, date of delivery, place and time. It is an agreement without exchange of product or monetary assets until the date of delivery. A futures contract is also based on a pre-specified agreement to buy or sell a commodity at a certain price. It deviates from a forward contract in that it is standardized and can be freely traded on an exchange through clearing parties. Gains and losses are realized daily on a margin account on a marking to market basis. The Commodity Futures Trading Commission regulates the futures market in the US.
A swap is an agreement between two parties to exchange a fixed price for a floating price of a commodity. Like forward contracts there is no physical exchange of the underlying. Swaps can be deployed to create synergies when companies have access to different markets or parties. When a party wants to hedge exposure against price movements it can agree to pay a fixed price, while the counterparty might want to benefit from advantageous price movements and therefore agree to pay the floating price.
An option is different from previous instruments because it gives the buyer the right, rather than the obligation to buy or sell a commodity at a specified price in the future. It can be exercised at any time until the maturity of the contract, but requires a fee that is called the option premium. In the oil and gas industry, put options are commonly used to create a price floor. This prevents future income from the sale of products from dropping below the “floor price” that was agreed in the option contract. If the oil price has not dropped below the floor price, the producer is free to let the option expire unused and sell the underlying in the regular spot market.
A price collar is a combination of a put option and a call option. It involves buying a put option to create a price floor, and selling a call option to create a price ceiling (also known as cap). The reason to choose this technique is from a cost perspective. The proceeds from selling the call option offset the premium to buy the put option. It is therefore also referred to as a zero-cost collar. If the price floor and cap are not equal the firm can benefit more from either upward or downward movement. This is known as a premium collar.
Looking at the oil and gas industry, several studies try to define determinants for the oil price movement. Stevans and Sessions (2008) find that the oil price movement is affected by the degree of exploration (supply), global demand and the value of the dollar. In addition to these findings, multiple papers find evidence for a significant effect of long-term futures prices on the spot price (French (2005) and Caporale et al. (2010)). These papers claim that futures prices are the most important driver of the spot price. Either way, producers can only steer global supply.
2.2.3 Operational hedging
As mentioned in Section 2.1 one of the reasons why firms hedge is to decrease the volalitility of earnings. Petersen & Thiagarajan (2000) perform an interesting case study that suggests gold mining firms have other possibilities to decrease the volatility of their earnings. They examine two individual gold mining firms. One exploiting an aggressive hedging strategy, the other one exclusively using so-called operational hedges. They argue that different mines hold different quantities and qualities of ore. If the price of gold drops, the firm can mine lower cost
6
or higher quality ore. This reduces incremental costs. Conversely, if the price of gold moves upward, the firm can mine higher cost or lower quality ore. This will increase incremental costs. Using this operational strategy the firm can thus decrease the impact of price movement without using derivatives.
Petersen & Thiagarajan argue that firms may actually have other incentives to hedge with commodity derivatives than decreasing sensitivity to prices. The alternative incentives mentioned in Section 2.1; tax benefits, reducing expected distress costs or decreasing reliance on external financing could be stronger for commodity producers. Kim et al. (2005) hypothesize that operational hedging is complementary to financial hedging because both strategies are used for managing different types of risk exposures. They find evidence that firms hedge long-term currency exposure with operational hedges and short-term currency exposure with financial hedges.
2.3 Research on the impact of hedging
Current literature on hedging can roughly be divided into two parts. The first part is the earlier research that focuses on explaining cross-sectional variation in the drivers for using derivatives. The second part examines whether or not hedging can increase firm value.
Concerning the drivers for using derivatives, Géczy et al. (1997) study a sample of US corporations that are exposed to currency risk and find results that are in line with the findings by Froot et al. (1993), as mentioned in Section 2.1. They find that hedging is used to reduce external financing cost. According to Haushalter (2000), who studies 100 firms in the oil and gas industry, the extent of hedging is positively associated with leverage. This builds on the assumption that firms hedge to reduce bankruptcy and financial distress costs. However, Tufano (1996) concludes that there is no evidence for a relationship between risk management and firm characteristics. In addition to the other studies, Lel (2003) investigates a sample of US listed firms and finds that corporate governance variables have a more significant impact than measures of frictional costs such as financial distress. All in all, there is only suggestive and mixed evidence for the relation between hedging activity and firm characteristics.
Research on whether hedging increases firm value also yields different results. Allayannis and Weston (2001) were the first to investigate this empirically using a sample of large US non-financial firms. They claim that hedging can increase firm value by 3 to 8 percent. Carter et al. (2004) perform a similar study among US airline companies and find an even higher increase in firm value; 12 to 16 percent. They attribute the increase to the firm characteristic of reduced external financing. In an extensive study among 7319 firms in an international sample, Bartram et al. (2009) nuance previous findings by implying that hedging is only associated with higher firm value for certain exposures, such as interest rate risk. Finally, Guay and Kothari (2003) analyse the effects of using derivatives for a sample of non-financial firms and conclude that the potential gains on derivatives are minor in comparison to cash flows and movements in equity values. They claim currency and interest exposure is secondary risk, and hedging this risk has only limited effect on the firm. Therefore they state is impossible that hedging has an effect on firm value of the magnitude claimed by earlier studies. An alternative explanation is that the increase in firm value is driven by other risk management practices such as operational hedges that might be correlated with derivate usage. If that is not the case, they claim the relationship is spurious.
All before mentioned studies try to either identify reasons why firms hedge, or measure the welfare implications of hedging in an intuitive way. They do however not say anything about
7
how hedging affects the firm. Campello et al. (2011) define two precise mechanisms through which hedging affects real and financial corporate outcomes. They do so by examining the impact of hedging on firms’ external financing cost and investment spending. Section 2.4 will elaborate further on this topic. Campello et al. examine a sample of over 1000 US firms and find that currency and interest rate hedgers pay lower interest spreads on their loans and are less likely to have capital expenditure restrictions in their loan agreements. With this they show that hedging has a first-order effect on firm financing and investment.
2.4 Hedging and contracting terms
Campello et al. (2011) state that hedging can affect the firm through two mechanisms, namely external financing and investment spending. These mechanisms are impacted by aspects in loan agreements between the firms and lenders. On the one hand the lender aligns the loan spread with the risk profile of the firm, on the other hand lenders put all kinds of restrictions in the loan covenant that prohibit a firm from doing certain investments and transactions.
The relationship between hedging and external financing is studied in several papers. As mentioned in Section 2.3 Froot et al. (1993) and Haushalter (2000) state that firm characteristics such as reducing reliance on external financing or distress cost drive the extent of hedging. Another theory is that by reducing the impact of information asymmetry with hedging, the costs of external financing drops, thus making firms increase their reliance on external financing (DeMarzo & Duffie 1995). Some empirics on this topic are provided by Magee (2009), who finds that hedgers of foreign currency risk rely more on external financing than nonhedgers do. It does not say anything about the actual cost of external financing, other than the reliance upon it. Bessembinder (1991) states that hedging improves firms’ investment activities through two channels. The first can be derived from the equity holders’ perspective. The use of derivative contracts lowers the risk of default, which decreases the chance that equity holders lose their money. This in turn results in less incentive to underinvest. The second way in which hedging influences investment is by improving contracting terms, since hedging induces cash flows in those states where the firm’s cash inflows are low. Bessembinder uses a theoretical framework, and it is dependent on whether the firm can credibly commit to maintaining the hedge over the life of senior debtors’ claims.
The findings of these studies should affect contracting terms that firms obtain from their lenders. At least, as long as managers act rational and do not engage in highly speculative hedging transactions. Campbell and Kracaw (1990) propose a model that suggests hedging makes it more difficult for managers to engage in risk-shifting. Credible commitments to hedge observable risk generate the incentive for managers to hedge unobservable risk as well. This could decrease the so-called asset-substitution problem. On the other hand it is also possible for firms to unwind their hedging positions after signing a contract with their lender. Managers can for example acquire offsetting positions or cancel their derivative program entirely. Banks try to avoid this behaviour by incorporating “performance pricing” in the loan agreement. This enables them to raise the interest rate on the loan in case of a covenant breach. Purnanandam (2008) states this behaviour is not rational. He presents a model that shows it is optimal for managers of firms in distress to keep their outstanding hedging positions after acquiring new debt.
8
2.5 Recapitulation
To summarize this Section, commodities are an interesting industry for studying hedging behaviour because actors in the market have little influence on the price of the products. Especially suppliers have little impact since markets are generally demand driven. The natural resource business (such as the oil and gas industry) is of particular interest because it is very capital intensive and requires long-term investments to generate income.
Hedging is not solely driven by income volatility. Tax incentives, reducing financial distress costs and reducing reliance on external financing are also mentioned as incentives for using derivative contracts. Besides operational drivers, agency issues are argued to have a role in hedging behaviour. The possibilities for natural hedges in the commodity market can decrease sensitivity to prices. Because of this, the additional incentives such as tax benefits may be more eminent in the commodity market.
Current literature on hedging is focused on either the determinants of hedging, or the effect of hedging on firm value. The latter yields mixed results, where some studies find that hedging can increase firm value quite significant, but others find no evidence for this. The study by Campello et al. (2011) is the first to identify specific channels through which hedging affects firm value. They focus on interest and currency exposure, whereas this thesis will focus on commodity price risk. As Guay & Kothari (2003) argue, interest and currency exposure is a secondary risk to non-financial firms. Hedging secondary risk has only a limited effect on the firm. From this perspective, the commodity market is more suitable to define the determinants on hedging because price fluctuations are a primary risk for commodity producers.
3 Methodology
This section will elaborate further on how the effects of hedging are measured and what outcomes are expected. On the one hand the first-order effect of hedging on the loan spread is investigated. On the other hand capital restrictions in loan covenants are examined. If hedging has an impact on financing terms, a next logical step is to examine if hedging actually shapes firms’ investment behaviour.
3.1 Hedging and loan spread
To examine the effect of hedging on the loan spread I will use regression analysis. There are many ways to empirically examine the pricing of loan agreements. Graham et al. (2008) have summarized most of this literature and provide an overall framework that summarizes previous literature on the topic. They put forward a detailed list of drivers for loan spread including firm and loan characteristics, but also macroeconomic controls and several robustness checks. I can use this framework as a base for my analysis.
In order to incorporate the effect of hedging into the model a proxy for hedging activity must be added. Several studies (Lookman (2004), Jin & Jorion (2006) and Carter et al. (2006)) use the percentage of production (oil and gas industry) or fuel demand (airline industry) hedged as a proxy. Firms often disclose information on their total hedging position during the past year. This is the most comprehensive measure. See Appendix 1 for an example. If on the other hand firms do not disclose adequate data on derivative usage in the last year, I will use the percentage of next year’s production hedged at the time of disclosure of the annual report as a proxy. This is
9
less optimal since firms often take new positions during the fiscal year, because contracts with longer maturities are more expensive (Jin & Jorion 2006). However, evaluating only firms that disclose adequate information on total annual derivatives usage would limit the sample size. After adding the hedging proxy to the model from Graham et al., the condensed regression formula looks as follows:
𝐿𝑛(𝐿𝑜𝑎𝑛 𝑠𝑝𝑟𝑒𝑎𝑑) = 𝛽0+ 𝛽1(ℎ𝑒𝑑𝑔𝑖𝑛𝑔 𝑝𝑟𝑜𝑥𝑦) + 𝛽2(𝑓𝑖𝑟𝑚 𝑐ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠) (1)
+ 𝛽3(𝑙𝑜𝑎𝑛 𝑐ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠) + 𝛽4(𝑚𝑎𝑐𝑟𝑜𝑒𝑐𝑜𝑛𝑜𝑚𝑖𝑐 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠) + 𝜀𝑖,𝑡
Loan spread is defined as the interest rate in excess of LIBOR2. Graham et al. (2008) take the natural logarithm of the loan spread to correct for positive skewness in the data. Several firm and loan characteristics and macroeconomic factors might affect the loan spread and should thus be controlled for.
3.2 Hedging and capital restrictions
Similar to Section 3.1, I will use regression analysis to investigate the effect of hedging on the presence of capital restrictions in loan covenants between firms and lenders. A bank will thoroughly assess the risk of a particular firm when granting a loan. Banks have all kinds of restrictive covenants in their loan agreements with firms, thereby restricting firms from doing certain investments and expenditures. These covenants should reduce the odds for a firm to get into financial distress. Consequently, restrictions on investments in loan covenants should closely reflect the bank’s assessment of the firm’s riskiness. This leads me to expect that more hedging should result in a lower probability on capital restrictions.
There are a number of interesting restrictive covenants in loan agreements, including restrictions on acquisitions, expenditures on non-capital items and changes in company ownership (Baird & Rasmussen 2006). However, I restrict my analysis to restrictions on capital expenditures for two reasons. First, a lot of restrictive covenants are fairly standard in a loan agreement, albeit with a lower or higher monetary boundary. So investigating them would yield little or no variation in the sample. Capital expenditure restrictions are less common, for example Nini et al. (2008) find them in 40 percent of the 3720 loan contracts in their sample. Second, capital expenditure restrictions refer primarily to cash capital expenditures. As cash is a primary driver of liquidity, banks should value capital expenditure restrictions as buffers against financial distress.
The condensed regression formula to test this is similar to equation (1):
{ Pr(𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑅𝑒𝑠𝑡𝑟𝑖𝑐𝑡𝑖𝑜𝑛) = 1} = 𝛽0+ 𝛽1(ℎ𝑒𝑑𝑔𝑖𝑛𝑔 𝑝𝑟𝑜𝑥𝑦) + 𝛽1(𝑓𝑖𝑟𝑚 𝑐ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠) (2)
+ 𝛽2(𝑙𝑜𝑎𝑛 𝑐ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠) + 𝛽3(𝑚𝑎𝑐𝑟𝑜𝑒𝑐𝑜𝑛𝑜𝑚𝑖𝑐 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠) + 𝜀𝑖,𝑡
The dependent variable CapitalRestriction is a dummy that equals one if there is a restriction on capital expenditures in the loan agreement, and zero otherwise. See Appendix 2 for an example of a restriction in a loan covenant. Since it is a dummy I will use a probit model. This
2 The London interbank offer rate (LIBOR) is a widely used reference rate. It is determined by estimations from
10
type of regression can be used when the dependent variable is dichotomous. It calculates the probability that the dependent variable is 1, given certain outcomes of the explanatory variables. It is based on a maximum likelihood estimation and the formula looks as follows:
Pr(𝑌 = 1|𝑋𝑖) = Φ(𝛽0+ 𝛽1𝑋𝑖) (3)
3.3 Hedging and investments
If there is reason to believe that hedging affects loan spreads and investment restrictions, the next analysis should be if hedging impacts actual investment spending. I will use an empirical investment model to investigate this relationship. As I examine restrictions on capital expenditures it is natural to use capital expenditures as a proxy for investments. To mitigate size effects I scale capital expenditures by a firm’s total assets. If I were to scale capital expenditures in year t = 0 by total assets in year t = 0 any incremental investments as a result of hedging could be mitigated. I therefore scale it by lagged total assets, i.e. I take total assets at t = -1. The dependent variables resemble those in formulas (1) and (2), except for the addition of cash flow volatility as suggested by e.g. Kaplan & Zingales (1997). See Section 3.4 for a detailed explanation on this.
𝐼𝑛𝑣𝑒𝑠𝑡𝑚𝑒𝑛𝑡 = 𝛽0+ 𝛽1(ℎ𝑒𝑑𝑔𝑖𝑛𝑔 𝑝𝑟𝑜𝑥𝑦) + 𝛽2(𝑓𝑖𝑟𝑚 𝑐ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠) (4)
+ 𝛽3(𝑙𝑜𝑎𝑛 𝑐ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠) + 𝜀𝑖,𝑡
3.4 Control variables
I control for several factors that might be correlated with contracting terms, as defined by Graham et al. (2008). The next section will briefly review each of the control variables. Loan contracts can be signed at any time throughout the fiscal year. To assume firm characteristics to be predetermined as the time of signing the loan contract, I lag them one year with respect to the loan agreement. Loan characteristics are inherently linked to the loan agreement and are thus not lagged. For the regression analyses I use a compact and an extended regression model. Because of the limited amount of observations I first try to find evidence with a compact model. As supporting evidence an extended regression model is deployed, limiting the degrees of freedom but controlling more carefully.
Firm characteristics
Chava & Roberts (2008) argue that information asymmetry goes down when a lender and firm have dealt with each other before. Because the firm has a track record with the lender, the firm has a lower perceived risk than firms with no previous loan contracts. The prior relationship variable is a dummy variable indicating one if a firm has had a loan contract with the same bank in the past 5 years, and zero otherwise.
As a proxy for firm size the natural logarithm of assets is used. Larger firms have easier access to external financing. Theory says that large firms have less information asymmetry and are therefore associated with smaller monitoring cost. This should provide firms with better contracting terms and thus lower loan spreads.
In case of a defaulting firm, lenders should be able to recover tangible assets. I expect firms with more tangible assets to have better contracting terms as it acts like collateral for the
11
loan. To control for it I include tangibility in the regression, which is defined as tangible assets divided by total assets.
More profitable firms have less risk on defaulting, so I will add profitability to the regression. It is defined as operating income before depreciation (EBITDA) scaled by total assets. I expect profitability to have a positive effect on contracting terms.
Firms with more growth opportunities are generally seen as more risky. Growth firms may suffer from information asymmetry or may be vulnerable to financial distress. I expect growth firms to have a higher loan spread and more capital expenditure restrictions. Market-to-book ratio serves as a proxy for growth opportunities as valued by the market. It is computed by dividing the market value of assets by the book value of assets.
Ceteris paribus, firms with higher leverage ratios bear more risk of default. I therefore expect leverage to have a negative effect on contracting terms. Leverage is computed by scaling short- and long term debt by total assets. One can argue that an ex ante lower loan spread enables firms to take on more leverage. But because I use the leverage from one year prior to the loan agreement I control for this potential endogeneity. It ensures that leverage is predetermined.
I expect the overall financial health of the firm to have a positive effect on contracting terms. However, the aim of my research is to measure the effect of hedging, whatever the financial health of the company may be. I therefore control for this using Altman’s Z-score. This proxy was developed by Edward I. Altman and it is widely used as a measure for corporate financial distress (Altman 1968). A higher Z-score indicates a better financial health and I therefore expect the Z-score to be positively correlated with contracting terms, i.e. a lower loan spread and less capital restrictions.
Where the Altman Z-score looks at overall financial health, credit rating agencies focus primarily on the ability to meet debt obligations. In practise, credit ratings are an important driver for the cost of debt. But the objective of my research is to find the effect of hedging, independent of credit ratings. I generate two proxies for credit rating. The first is a rating dummy with a dummy for each rating class. This dummy takes on a considerably amount of degrees of freedom. The other one is a rating variable indicating -1 if the rating is below CCC, +1 if the rating is higher than BBB, and zero otherwise. If a firm has no rating it is also assigned a zero. This proxy only consumes 1 degree of freedom but accounts less for rating variance.
If firms have volatile cash flows, larger liquidity buffers have to be kept in order to meet payment obligations. If they fail to do so, they may face costly financial distress. The greater the income volatility, the higher the expected cost of financial distress will be. Maintaining liquidity buffers has a direct effect on the quantity of investments. I therefore add it to the investment model. It is defined as the quarterly cash flow volatility in the four years preceding a loan contract.
Loan characteristics
Graham et al. (2008) use the natural logarithm of the dollar value of the loan to control for loan size. The main difference between this measure and the control firm size resides in that it captures economies of scale. I expect that a larger loan, i.e. a larger deal for the lender, has a positive effect on contracting terms. Firm size should control for the easiness of access to external financing, where loan size is deal-specific and irrespective of firm size. However, the
12
correlation between loan and firm size might be too high. To account for this, I scale the dollar value of the loan by total assets and add that to the regression, rather than the natural logarithm.
Banks require a liquidity premium for supplying a longer-term debt. This premium should have a negative effect on contracting terms. I therefore add the natural logarithm of the loan maturity in days to the regression.
Macroeconomic control
Macroeconomic conditions can have an effect on debt pricing. Graham et al. (2008) suggest using the term spread. It is defined as the difference between a 10-year and 1-year Treasury bond. This variable is a good proxy for economic prospects and I expect contract terms to be affected by the overall economy. It is added to the regression.
4 Data and descriptive statistics
4.1 Sample descriptionTo test the impact of commodity hedging on the loan spread and capital restrictions I investigate the US oil and gas industry. Investigating this industry has several advantages. First, volatility of oil prices has a substantial effect on the cash flows of oil and gas producers. The industry is heavily exposed to a primary risk. Second, Campello et al. (2011) study a large sample of US firms from all different industries. In comparison, the oil and gas industry is much more homogeneous and according to Jin & Jorion (2006) it still offers substantial variation in hedging ratios. The third reason for choosing this industry is because it has sufficient listed companies. The data on hedging and loan agreements have to be gathered from public annual reports, therefore the US oil and gas industry serves well as research subject. I use the definition of the US oil and gas industry as given by the Standard Industrial Classification (SIC). According to this system, code 13 is assigned to Oil and Gas Exploration firms; sub code 1311 specifies this further to Crude Petroleum and Natural Gas. Appendix 3 contains a detailed description of the industry.
The time frame to be used ranges from 1996 to 2002. I choose for this time frame because of two reasons. First, there is only limited information on loan agreements before 1996. Second, as of 2002 Statement of Financial Accounting Standards (SFAS) 133 was initiated. This standard requires firms to report derivative usage at “fair market values”, rather than notional values. According to Graham & Rogers (2002) fair market values contain only limited information on derivative positions. In 2009, SEC EDGAR started to work with XML tags in, making it easier to gather detailed hedging information for a large sample of firms. However, as derivative usage is reported at fair market values I chose not to examine this time frame.
4.2 Databases
To perform the regression analyses I need data from different sources and databases. I will briefly discuss each data source.
SEC Edgar
Electronic Data Gathering, Analysis and Retrieval system (EDGAR) is a database owned and run by the U.S. Securities and Exchange Commission (SEC). It performs automated
13
collection, validation, indexing, acceptance and forwarding of submissions by companies and others who are required to file forms with the SEC (SEC 2010). All of the uploaded documents are freely available to the public through the internet, both via de SEC website and their FTP server. There are all kinds of different form types that firms can file into EDGAR, ranging from annual reports to specific reports on changes in company ownership. Every form type has a fixed front page format and all chapters in the report are defined in advance. From 1996 all domestic U.S. companies with more than 10 million dollar in assets and a class of equity securities that is held by more than 500 owners, are required to file their SEC filings electronically through EDGAR. It does not matter whether the shares of the firm are traded privately or publicly. As of November 2002 the SEC also required foreign companies with business in the U.S. to file electronically via EDGAR. The database currently holds over twenty million filings. Entities that file documents via EDGAR are identified by a Central Index Key (CIK).
Thomson ONE Deals – syndicated loans
Thomson ONE Banker is a database that provides access to relevant real-time global market data, news, and authoritative content from different sources (Thomson Financial 2003). It has an additional module, Thomson ONE – Deals, which consists of five sub modules. The module of interest is Syndicated Loans. This database contains data on over 92.000 global credit agreements between public firms and lenders since 1982. It includes among other things the original pricing of the credit agreements, the maturity, the underwriters and the loan size. It contains more information than the actual credit agreements that can be found in certain documents in the SEC EDGAR database. For example, the pricing of a loan agreement is hardly ever listed in an actual credit agreement. Thomson ONE uses CUSIP as a firm identifier.
COMPUSTAT
The final database that is needed to build the dataset is COMPUSTAT. It is a database of financial, statistical and market information on active and inactive global companies throughout the world. It is a division of S&P Capital IQ and was founded in 1962 (Capital IQ 2014). I need annual fundamentals and credit ratings from the Compustat North America data set. This data is required for constructing the control variables. Another important reason for using Compustat is that it uses both CIK and CUSIP as company identifiers. Since SEC EDGAR uses CIK and Thomson One uses CUSIP, Compustat can help to merge the data from the different data sources. Unfortunately not all entities with a CIK have a CUSIP, because CUSIPs are only assigned to public financial securities while CIKs can be also assigned to privately traded firms.
4.3 Data collection process
As a starting point, I download all data on loan agreements from Thomson One Deals for industry 1311 for the years 1994 to 2007. I also download a list with an overview of all filings in EDGAR from 1994 to 2007. Finally I download a list with CUSIPs and CIKs from Compustat that are identified to be in SIC 1311. I tag all unique companies in each of the databases. After merging these three datasets I am left with an initial list of firms that have at least one data point in each database for the time period 1994 to 2007. I will refer further to this list as the base list. Note that this does not ensure that each firm can be used in the final sample; it is just an initial screening to reduce the number of observations for the next steps.
14
I then have to download all text documents that potentially contain actual credit agreements for the years 1998 to 2007. To do this I create a list with all CIKs on the base list. I design a program in Perl that downloads all SEC filings for the firms on this list for the appropriate years. Following Nini et al. (2008) I only download SEC EDGAR form types 10-K, 10-Q and 8-K, since these are the document types where credit agreements are filed. Next I have to determine which filings actually contain credit agreements, using the method from Nini et al. (2008). I design a text crawler in Unix Grep that searches in each physical document for the terms: “credit agreement, loan agreement”, “credit facility”, “loan and security agreement”, “revolving credit”, “financing and security agreement”, “financing & security agreement”, “credit and guarantee agreement” and “credit & guarantee agreement” in capital letters. If one of these terms is found I read the next 60 lines and look for the term “table of contents”. When both conditions are met I let the text crawler add the CIK to a list. This provides me with a list of all filings with credit agreements available on SEC Edgar for the years 1998 to 2007. Using another text crawler I extract the names, dates and file type from these documents and export it to a spreadsheet.
Next I have to locate documents that contain information on hedging. To ensure that the hedging information can be assumed to be predetermined, I lag the hedging documents with respect to the loan agreements. Because filings are not constant throughout the years I decide to look for hedging information in the four years preceding a loan agreement. Using a constraint of only one year lags would limit my sample size too much. Note that firm specific control variables are lagged one year to the loan agreement. I assume that there is not significant variation in derivatives usage for the firms throughout the years. So if there is only a report on hedging three years prior to the loan agreement, I assume that the firm is still hedging/seen as a hedger in the year prior the loan agreement. A graphical presentation can be found in Appendix 4. Concerning the firm specific controls; I will use data from the year prior to the loan agreement. Since firms have to disclose information on hedging I can use all K, KT, 10KSB, 10KSB40 and 10-K405 documents that were filed in the period from 1994 to 2006. Similar to the loan agreements, I use a Perl program to download all appropriate physical documents for the firms on the base list. Using a Unix Grep text crawler I extract the names, dates and file types from the filings and put this in a spreadsheet. Then I merge this list with the list on loan agreements to get a master overview for all firms with the available loan agreements, hedging documents and file types.
All previous steps were only preparation of the documents from which to gather the data. From this point I can proceed with the actual data collection. To reduce the impact of omitted variables that arise from having loan agreements and hedging information in different years, I use two priority rules. The first is choosing the documents as such that there is a minimal amount of years between the loan and hedging information. As a second priority rule I choose the hedging document closest to 2002, if the first priority rule permits this. This is the last year from which I can use hedging information and my assumption is that data quality will be best in 2002. Neither the information on capital expenditure restrictions nor hedging is reported in a similar fashion across all documents. Sometimes hedging activities are reported in Item 7A of the filing, other times it is found in the managerial discussion or in footnotes throughout the document.
Because of the irregular documentation I decide to manually read each document and go through all the sections to ensure a reliable result. Firms hedge oil and gas exposure separately, but a joint position can be calculated using the metric Barrels of Oil Equivalent (BOE). This metric expresses volume of gas in barrels of oil. It is then possible to sum the hedging positions
15
for both products. If firms hold derivatives for trading purposes then the intention of hedging market risk is diluted by the extra risk involved in actively trading both long and short positions. Firms have to indicate if they hold derivatives solely for non-trading purposes. Firms that actively trade are not added to the sample as it can bias the results. Short positions are often only used for creating price collars which are reported at the net number of barrels hedged. If short positions are held for other purposes than price collars, and are not held for active trading, then I take the total net position of long and short agreements. After finding the correct information I hand code the data to a spreadsheet that I feed into Stata. I am left with 69 firm observations.
The dataset only contains data on loan agreements and hedging: loan size, interest rate, maturity, capital expenditure restriction, and percentage of production hedged. Control variables have to be added at this point. Using data from Compustat I compute the necessary firm-specific variables and macroeconomic control as mentioned in Section 3.4; natural logarithm of assets, tangibility, profitability, market-to-book ratio, leverage, Altman Z-score, credit rating and term spread. I merge this data with the dataset. For these variables I use a one-year lag with respect to the loan agreement to ensure them to be predetermined. I now have the final sample to be used in the regression analyses.
4.4 Descriptive statistics
Panel A of table 1 contain the summary statistics for the sample. The hedging dummy indicates that 86.7 percent of all firms in the sample hedge commodity exposure. The firms that hedge, hedge on average 38.3 percent of next year’s total production. This is consistent with the results from Jin & Jorion (2006) who, even though they separate oil and gas hedging, find 33 and 41 percent, respectively. The average loan size is $376 million and the average spread is 164.2 basis points, so firms pay on average 1.64 percent over LIBOR. Campello et al. (2011) find an average loan size of 291.6 and a spread of 188.6 basis points. Since they study a sample of non-financial firms from different industries, but not the oil and gas industry, it appears that the oil and gas industry has larger loans and lower interest spreads. This is in line with earlier statements in Section 2.2.1 about the oil and gas industry being capital intensive. The average maturity of the loans is 1523 days (about 50 months) which is higher than the 1344 in Campello et al (2011). This makes sense since oil exploration firms make long-term investments in order to expand drilling sites or improve current operations. There is a restriction on capital expenditures in 42.6 percent of the loan agreements, slightly higher than the 37.3 percent in Campello et al. (2011).
Concerning firm characteristics there are proxies such as firm size (natural logarithm of assets), profitability (EBITDA over total assets), tangibility (net PP&E over total assets), growth opportunities (proxied by M/B), leverage (total debt over total assets), default risk (proxied by Z-score) and two credit rating proxies. The first proxy is a dummy for each possible credit rating outcome (D=1, AAA=22 on the S&P rating scale). If a firm has no credit rating it is assigned a zero, which leads to a skewed interpretation of the credit rating variable. For the companies that are rated, the average rating is 11.4 rather than 5.78. The second proxy is a variable that equals 1 if a firm has a superior rating, a zero if it has an average or no rating, and -1 if the firm has an inferior rating. Finally there are dummies for different types of hedging contracts used.
Panel B of table 1 compares firm and loan characteristics between hedgers and nonhedgers. Since 87 percent of the firms in the sample use hedging instruments, there is only a small subsample of around 8 non-hedging observations. With this subsample I find that hedgers pay a lower interest rate and have fewer restrictions on capital expenditures, which is both
16
intuitive. I further find that hedgers are larger and have more tangible assets; the differences are statistically significant. Hedgers are better valued by the market than nonhedgers because their market-to-book ratio is higher. The argument would be that hedgers have more growth opportunities. Theory suggests that hedgers take on more leverage; this is confirmed in the mean comparison. I also find that hedgers have higher credit ratings whilst having a higher default risk, as measured by the Z-score. Note that this finding does not take into account any firm characteristics such as profitability or size. Considering loan characteristics, hedgers borrow a significantly larger amount and have slightly shorter loan maturities.
Table 1: Descriptive statistics
The summary statistics are based on the sample that is used in the different regression analyses. The data is split into hedging information, firm- and loan characteristics and macro-controls. The data was gathered from different sources: SEC EDGAR, Thomson ONE Deals and Compustat.
Panel A: Sample summary statistics
Obs. Mean Std. Dev. Min p5 p50 p95 Max
Hedging information
Hedging dummy 68 0.867 0.341 0 0 1 1 1
Percentage hedged 63 0.383 0.214 0 0 0.382 0.743 0.84
Firm characteristics
Previous loan dummy 68 0.661 0.476 0 0 1 1 1
Log assets 66 6.346 1.474 3.535 4.013 6.336 9.375 10.200 Total assets ($M) 66 1968 4577 34 55 565 11798 27163 Profitability 66 0.167 0.124 -0.148 -0.087 0.162 0.291 0.735 Tangibility 66 0.804 0.141 0.002 0.575 0.840 0.929 0.940 M/B 61 1.389 0.461 0.631 0.781 1.372 2.182 3.103 Tobin’s Q 58 0.798 0.494 0.057 0.099 0.726 1.986 2.003 Leverage 66 0.361 0.216 0 0 0.355 0.670 0.953 Altman Z-score 62 0.421 0.841 -1.663 -1.057 0.538 1.728 2.498
Cash flow volatility 63 0.358 0.556 0.015 0.045 0.170 1.358 2.677
Investment/lagged assets 63 0.267 0.156 0.048 0.048 0.238 0.582 0.582
Credit rating dummy 68 5.784 5.977 0 0 8 15 17
Credit rating variable 68 -0.323 0.721 -1 -1 0 0 1
Distance-to-default 62 1.915 2.926 0.149 0.229 0.647 9.125 12.312 Restriction on CAPEX 68 0.426 0.498 0 0 0 1 1 Swaps 68 0.735 0.444 0 0 1 1 1 Forwards/futures 68 0.294 0.459 0 0 0 1 1 Options 68 0.308 0.465 0 0 0 1 1 Price collars 67 0.522 0.532 0 0 1 1 2 Loan characteristics
Log loan spread 68 4.981 0.558 2.833 3.806 5.090 5.616 5.926
Loan spread 68 164.2 68.5 17.0 25.0 162.5 275.0 375.0
Relative loan size 66 0.576 0.624 0.055 0.084 0.392 7.313 3.756
Loan size ($M) 68 376 451 15 25 227.5 1500 2440
Log maturity (days) 68 7.278 0.339 6.126 6.595 7.349 7.652 7.774
Maturity (days) 68 1523 442 443 732 1556 2105 2379
Macro controls
17
Table 1-continued
Some firm- and loan characteristics for hedgers and nonhedgers. A simple mean comparison between the two is denoted in column (3) and the significance is determined with sample t-tests.
Panel B: Nonhedgers vs. Hedgers
(1) (2) (3) (4)
Obs. Nonhedgers Hedgers Diff: (2) - (3)
Loan spread 68 173.4 163.2 7.636 Restriction on CAPEX 68 0.444 0.423 0.021 Log assets 66 5.289 6.471 -1.182** M/B 61 1.196 1.403 -0.206 Leverage 66 0.243 0.376 -0.132 Tangibility 66 0.692 0.818 -0.125*** Z-score 62 0.488 0.414 0.074 Credit rating 68 -0.666 -0.271 -0.395 Loan size 66 140.0 412.5 -272.5*
Log loan maturity 68 7.324 7.271 0.053
*, **, and *** denote statistical significance of the t-test at the 10%, 5%, and 1% level, respectively.
5 Empirical results
5.1 Hedging and loan spread
The regression results from equation (1) are reported in Table 2. Columns (1), (2), and (3) contain the results of a compact regression model with a basic set of firm and loan controls. Column (1) contains the base line model and yields no significant result for the percentage hedged variable. With this dataset and set of controls, hedging has on average no significant effect on the loan spread. The coefficients on firm’s previous loans, the log of assets and the credit rating are significant and all have a negative sign. This means firms with a relationship to the bank, large firms, and firms with high credit ratings pay lower spreads.
The asset substitution problem is a classic finance theory in which value is transferred from bondholders to shareholders by exchanging low-risk assets for risky investments. In this process, debt holders are exposed to more risk since default risk increases with risky investments. Following this theory, firms that hedge are less exposed to this extra risk because their future income is secured. From a lender’s perspective hedging strategies may thus be extra valuable for firms that engage in shifting through asset substitution. It is difficult to measure the risk-shifting directly, but I can use a proxy for the riskiness of the firm. According to Jensen & Meckling (1976), firms with more growth opportunities tend to be more risky. Even though hedging has on average no significant effect on the loan spread, there might be in-sample variation between low- and high risk firms. By interacting the market-to-book ratio with the hedging variable I can measure this effect and test this theory. The results of this regression are reported in column (2). The percentage hedged variable, the market-to-book ratio and their interaction are all significant. The market-to-book ratio has a positive coefficient, indicating that more risky firms pay higher spreads. This is in line with expectations. The hedging variable also has a positive sign, which is counterintuitive at first glance. The interaction term has a negative
18
Table 2: Hedging and loan spread
This table contains the results of regressing loan spread on hedging. The dependent variable is the natural logarithm of the loan spread, which is defined as the interest rate in excess of LIBOR. All results are obtained from OLS. The variable of interest is Percentage hedged, which measures the percentage of annual oil and gas production that is hedged by the firms in the US oil and gas industry. All other variables are defined in Section 3.4. Standard errors are denoted in parentheses and are robust to heteroskedasticity.
(1) (2) (3) (4) (5)
Percentage hedged -0.230 2.192*** -0.565* -0.738** -0.529
(0.32) (0.78) (0.33) (0.33) (0.46)
Percentage hedged x M/B -1.701***
(0.56)
Percentage hedged x Z-score 0.727* 0.938**
(0.40) (0.44)
Firm's previous loans -0.198* -0.143* -0.241** -0.057 -0.208*
(0.099) (0.084) (0.12) (0.096) (0.12) Log assets -0.131* -0.087 -0.257*** 0.053 -0.309*** (0.076) (0.082) (0.053) (0.070) (0.066) M/B -0.264 0.490* -0.241 -0.180 -0.270 (0.17) (0.25) (0.21) (0.11) (0.20) Leverage 0.325 0.393 0.426 0.101 0.426 (0.30) (0.27) (0.31) (0.32) (0.33)
Relative loan size 0.135 0.240 0.065 0.296* 0.021
(0.14) (0.15) (0.16) (0.15) (0.17)
Log maturity (days) -0.125 -0.301 0.140 -0.465** 0.394
(0.25) (0.22) (0.25) (0.22) (0.27)
Credit rating variable -0.249* -0.299**
(0.13) (0.14) Z-score -0.193 -0.383** (0.14) (0.16) Profitability 1.264** 1.378 (0.56) (0.83) Tangibility -1.144** -1.437** (0.48) (0.61) Term spread 0.964 -0.214 (0.85) (1.03) Constant 7.056*** 6.879*** 6.083*** 9.431*** 5.595*** (1.53) (1.39) (1.45) (1.47) (1.40)
Credit rating dummies No No No Yes No
Observations 60 60 58 57 58
Adjusted R² 0.4924 0.5695 0.4642 0.7738 0.4852
19
coefficient. When calculating the expected outcomes for the means of the variables the joint result is -0.070 (= 2.192 x 0.410 – 1.701 x 0.410 x 1.389). So for a firm with average hedging positions and average risk the loan spread is negatively impacted. This supports the theory of risk-shifting. For firms with higher risk, hedging has a larger impact on the loan spread.
As mentioned in Section 2.1 Smith & Stulz (1985) state that hedging can reduce expected financial distress cost by lowering income volatility. By reducing distress cost hedging should ease the access to external credit. Following this theory, hedging should be more valuable for firms that have high default risks since their expected distress costs are high. Similarly, hedging should be less valuable for financially solid firms, ceteris paribus. To test this hypothesis I interact the Altman Z-score with the hedging variable. A higher Z-score is associated with a higher default risk. The results for this regression are reported in column (3). Since credit rating and Z-score tap a similar concept I choose to drop credit rating to conserve degrees of freedom and avoid collinearity between the coefficients. The coefficient on the hedging variable and the interaction term are both significant and attract the expected sign. The interpretation is that even though hedging reduces the loan spread, this effect is diminishing with a lower default risk. In other words, as the theory of financial distress costs suggests, firms that are financially sound have less incremental value from hedging than firms that are close to financial distress.
The hedging variable does not yield a significant result in the condensed base line model. As mentioned in Section 3.4 Graham et al. (2008) suggest a more extended model. Appendix 4 contains a correlation matrix for all independent variables. Looking at the correlations there is no obvious reason to expect collinearity when adding profitability, tangibility and term spread to the regression. Also, Graham et al. (2008) suggest to use credit rating dummies for each rating class to capture the differences in credit ratings more accurately. Column (4) reports the results of the extended base line model. In comparison to the results in column (1) the hedging variable is significant. It appears that when controlling more carefully for firm-specific characteristics there is indeed a direct effect of hedging on the loan spread. Multiplying the average hedging percentage by the hedging estimator gives an estimate of the reduction in loan spread. A firm with average hedging activity is charged a 30.3 percent lower loan spread than nonhedgers (= -0.738 x 0.410). Based on the average loan spread in Table 1 of 164.2 basis points, the average reduction can be expressed as around 50 basis points. This is a significant cost reduction as the average loan amount is $376 million; i.e. a reduction of roughly $1.88 million in the first year of the loan agreement. Furthermore the loan size and maturity, the profitability, and tangibility coefficients are significant. Firms with a larger loan pay higher spreads which is not in line with the theory of economies of scale. A possible explanation can be that a larger loan exposes lenders to a higher risk. This effect might mitigate any economies of scale benefits. What is peculiar is the negative sign on the loan maturity coefficient, suggesting that the spread decreases with the loan horizon. Firms with more tangible assets are charged lower spreads, which is in line with the idea that tangible assets are seen as collateral. The positive coefficient on profitability is counterintuitive since profitable firms are expected to have lower spreads.
Finally column (5) reports the results of the extended base line model with an interaction term between the hedging and Z-score variables. Both the Z-score variable and interaction term show a significant result. A higher Z-score has a direct negative effect on the loan spread; i.e. firms with a lower default risk pay a lower interest rate. The interaction term can be similarly interpreted as the results in column (3); firms with a low default risk have less incremental value
20
from hedging than financially distressed firms. Other significant results attract the expect sign; larger firms, firms with former ties to the bank, and more tangible firms pay lower spreads.
5.2 Hedging and capital restrictions
The results of the probit model from equation (2) are reported in Table 3. It is a similar setup as in Table 2, where columns (1), (2), and (3) contain the compact form, and columns (4) and (5) contain results of the extended model. As the coefficients of a probit regression may only be interpreted with great caution, I compute the marginal effects at the means of each of the independent variable. The coefficient can then be interpreted as the increase in probability on a restriction on capital expenditures when the independent variable increases with one unit. None of the results for the compact model yields a significant coefficient for the hedging variable. It appears that with a limited set of controls there is no conclusive evidence that hedging reduces the probability on capital expenditure restrictions.
Introducing the extra controls like in Section 5.1 does yield a significant and negative coefficient on the hedging variable. Column (4) contains the results of the extended analysis. Following a similar calculation as in Section 5.1, the probability on a restriction on capital expenditures is reduced with 41.7 percent (= -1.018 x 0.410). Campello et al. (2011) find a slightly larger reduction of 54 percent. The leverage coefficient is positive and significant, in line with expectations. Also more tangible firms have a lower chance on capital expenditure restrictions in their loan covenants. I lose 14 observations relative to the base line model, because several credit rating dummies are omitted. When a dummy has no variation in outcome with respect to the dependent dichotomous variable, this estimator perfectly predicts failure of success and can thus not be used to explain the variation in the dependent variable. This occurs if for example the capital expenditure restriction dummy is always 1 when the credit rating dummy is 1. These dummies – and the corresponding observations – are then dropped from the regression.
Column (5) contains the results of interacting the Z-score with the hedging variable in the extended regression model to test the financial distress cost theory. It yields significant coefficients for the Z-score and the interaction term. As in column (3) a higher Z-score reduces the probability on capital restrictions. When controlling for the interaction between hedging and Z-score there seems to be no direct effect of hedging on the probability of capital expenditure restrictions. However, when a firm has a high Z-score hedging will actually mitigate the reduction in capital expenditures. In other words, healthy firms will only benefit from their financial wellbeing if they do not hedge. I also find significant coefficients for firm size, tangibility and loan maturity. Larger and more tangible firms have a smaller probability of capital expenditure restrictions. Opposite to the findings in the loan spread regressions, loan maturity increases the probability on restrictions. This is all in line with expectations.
The percentage hedged variable is only significant in column (4), which has fewest degrees of freedom. I want to be careful with interpreting the results. I find only suggestive evidence for an effect of hedging on capital expenditure restrictions.
21
Table 3: Hedging and capital restrictions
This table contains the results of regressing restrictions on capital expenditures on hedging. The dependent variable is a dummy indicating 1 if there is a restriction on capital expenditures in the loan agreement, and all results are obtained from a probit model. The variable of interest is Percentage hedged, which measures the percentage of annual oil and gas production that is hedged by the firms in the US oil and gas industry. All other variables are defined in Section 3.4. The reported coefficients are the marginal effects at the means of each variable, to increase the interpretability. Standard errors are denoted in parentheses and are robust to heteroskedasticity.
(1) (2) (3) (4) (5)
Percentage hedged -0.225 -0.290 -0.561 -1.018* -0.690
(0.349) (1.064) (0.548) (0.562) (0.535)
Percentage hedged x M/B 0.047
(0.698)
Percentage hedged x Z-score 0.773 1.230**
(0.575) (0.572)
Firm's previous loans -0.046 -0.048 -0.103 0.125 -0.047
(0.132) (0.135) (0.141) (0.187) (0.144) Log assets -0.098 -0.099 -0.123 0.270 -0.210** (0.091) (0.093) (0.078) (0.171) (0.085) M/B 0.014 -0.005 0.156 0.010 0.175 (0.164) (0.364) (0.184) (0.236) (0.181) Leverage 0.888** 0.887** 0.653 0.979* 0.732 (0.387) (0.387) (0.451) (0.565) (0.498)
Relative loan size -0.022 -0.025 -0.229 0.154 -0.357
(0.225) (0.228) (0.237) (0.324) (0.253)
Log maturity (days) -0.027 -0.023 0.131 -0.055 0.518*
(0.212) (0.228) (0.240) (0.316) (0.283)
Credit rating variable 0.026 0.028
(0.141) (0.144) Z-score -0.561** -0.887*** (0.240) (0.266) Profitability 0.313 1.573 (0.116) (1.291) Tangibility -2.053** -2.135** (0.888) (0.995) Term spread 0.004 0.200 (0.113) (1.379)
Credit rating dummies No No No Yes No
Observations 60 60 58 46 58
Pseudo R² 0.1368 0.1368 0.2427 0.3100 0.2981
