What are the effects of derivatives on firm risk?

Transcription

1 Tilburg School of Economics and Management Master Thesis in Finance What are the effects of derivatives on firm risk? An empirical study on S&P 500 manufacturing firms for the years Author R. Kamphuis BSc. ANR s Faculty Tilburg School of Economics and Management Program Master Finance Department Finance Supervisor Prof. Dr. J.J.A.G. Driessen Second reader Dr. F. Castiglionesi Defense date August 29, 2013

2 Abstract This master thesis investigates the effects of financial derivatives on firm risk using a sample of S&P manufacturing firms for the years 2007 to There are several theoretical studies which have shown that derivative usage by corporations is rational and consistent with hedging behavior. Though, in past empirical evidence it remains undetermined whether derivatives were used for hedging (risk decreasing) or speculating (risk increasing). The results of this thesis suggest that for the entire sample, there is a risk increase due to use of derivatives, controlled for firm fixed effects. However, it also can be concluded that the relationship between derivative usage and firm risk is nonlinear. Specifically, the use of hedge derivatives is associated with a risk reduction for moderate derivative users and with a risk increase for extensive derivative users. Therefore, this implies that hedging is possible when derivative notional amounts stay relative low.

3 Table of Contents 1. Introduction Literature overview Theories of corporate hedging Costs of Financial Distress Contracting costs Capital market imperfections Taxes Managerial risk aversion and hedging Surveys Prior research Accounting standards Sample Construction Methodology Measures of firm risk Total risk Systematic risk Idiosyncratic risk Derivative usage Control variables Regression analyses Portfolio analysis Ordinary least squares regressions Panel data regressions Interaction regressions Empirical findings Summary statistics of derivative usage Summary statistics for the dependent- and control variables Portfolio analysis Ordinary least squares regressions Regressions for separated years without control variables... 28

4 Regressions for separated years with control variables Pooled regression Fixed-effects model Derivative product regressions Interaction regressions Conclusion References Appendix... 45

5 1. Introduction Since the beginning of the seventies the financial world has undergone an intense and continuous flow of innovations, according to Capelle-Blancard (2010). Some of the causes of the major structural changes are increasing volatility, the boom in new information and communication technologies, and the deregulation of the economy. These changes have had a huge positive impact on the financial world we know today. However, whenever a crisis appears, once again the recurring debate emerges about benefits and dangers of financial innovations. The most important financial innovation in the last thirty years is undoubtedly derivatives. Simply said by Capelle-Blancard (2010), derivatives are financial contracts between two parties whose value is contingent on the future price of an asset such as a share, a currency, a commodity, or an index. There are many classes of derivatives, with the most notable being forwards, swaps, futures and options. These derivatives can be used for two purposes: risk management which reduces volatility, (frequently termed as hedging) and risk management which increases return volatility, (called speculating). Derivatives are the result of long evolution in commercial and financial practices. According to Capelle-Blancard (2010), the very first transactions with delayed delivery (and thus similar to futures or options) go back to antiquity. There is evidence of the use of derivatives in the 17 th century in Holland and in Japan. In the 19 th century the increase of derivative usage did not improved that much, despite of the creation of organized markets in Chicago in This was due to the constraining regulatory system at the time. Therefore, derivatives markets stayed rather illiquid and small and most of the contracts involved commodities. Derivative markets definitively moved to the next level due to the introduction of currency futures and stock options on organized exchanges in Chicago in the early 1970s (Capelle-Blancard, 2010). Black & Scholes (1973) and Merton (1974) published their seminal articles on option pricing and since then the derivative markets have been experiencing a major boom. Due to lack of disclosure and accounting standards, research stayed behind until the early nineties. In 1995 the first research was done by Bodnar, Hayt, & Marston (1995), however this was a survey and not an empirical research. Nevertheless for the first time there was some visibility in the usage of derivatives. One of the first empirical investigations about derivative usage for the S&P was done 1

6 by Hentschel & Kothari (2001) for the years 1992 and Nguyen & Faff (2010) did almost the same research for Australia for the years 1999 and The main conclusions of these empirical investigations were that there is no evidence that the use of financial derivatives is associated with a reduction in firm risk. As one possible explanation for these results it was argued be that the notional amounts outstanding relative to the firm size was low in the examined years (Hentschel & Kothari, 2001). This means that the effect of derivative usage was very small and it is hard to say if there is an increase or reduction in firm risk. However, there are some studies which conclude that firm risk decreases after the adoption of derivatives, by for example Guay (1999), using a sample of 254 nonfinancial firms that begin to use derivatives. Allayannis & Ofek (2001) also report a reduction in foreign exchange exposure of the sample S&P 500 nonfinancial firms as a result of foreign currency derivative usage. So the effect of derivative usage remains undetermined, though several studies indicate that corporate motives to use financial derivatives are rational and consistent with hedging behavior 1. The bank of international settlements (BIS) 2 states that the derivative usage has grown exponentially, since the early nineties until today. This could solve the problem of the relative low amount of derivatives outstanding. Therefore, there is no research for the years after 2000, so probably the largely grown notional amounts and the past financial crisis changed the use of derivatives. Therefore the primary objective of this thesis is to empirically investigate whether the use of derivatives by S&P 500 manufacturing corporations is significantly related to their overall stock return risk, for the years 2007 to That is why the main question is: what are the effects of derivatives usage on firm risk? The sector manufacturing is chosen, because of the large outstanding derivative amounts with respect to other industries (Bodnar et al., 1995). The time period 2007 to 2009 is chosen to investigate whether there is any effect of the credit crisis on derivatives usage. This investigation will help to provide an answer to the long-standing question whether firms use derivatives to hedge or to speculate. The remainder of this thesis is organized as follows. Section 2 is the literature overview with the main reasons to hedge, past surveys and empirical research. Section 3 describes the data and the sample descriptions. Section 4 describes the methodology of the different regression techniques. Section 5 presents the empirical findings. The concluding remarks are offered in Section 6. 1 See, Smith and Stulz (1985), Nance et al. (1993), Bartram et al. (2009), Gay and Nam (1998). 2 BIS Quarterly Review (June 2010): Semiannual OTC derivatives statistics at end-december 2009 (Table 19) and Statistics on exchange traded derivatives (Table 23A). 2

7 2. Literature overview The total size of the derivatives market in 2010 is about $700 trillion, while it did not exceed $1 trillion in the 1980s, according to the Bank of International Settlements (BIS). In the period from 1986 to 2010 notional amounts have increased by 30% per year on average, i.e. doubling every three years, with a speeding up since The growth of the derivative markets is not only exceptional, but it is not about to slow down significantly, despite the global financial crisis of (Capelle-Blancard, 2010). These figures are extremely large and therefore derivatives are increasingly important in the commercial world. 2.1 Theories of corporate hedging Obviously there must be some good reasons for companies to use more derivatives year after year. In a perfect world created by Modigliani & Miller (1958), there would be no justification of corporations to engage in hedging, including those strategies that use derivatives (Gay & Nam, 1998). However we do not live in a perfect world, so financial economics offers different theories why corporate hedging can be rational or value-enhancing, each of which relies on some form of market imperfection. There are four main theories according to Mian (1996) that took advantage of market imperfections by hedging, and therefore create value for the stockholders. This can be done by lowering expected costs of financial distress ((Mayers & Smith, 1982) and (Smith & Stulz, 1985)), lower contracting costs ((Mayers & Smith, 1987) and (Nance, Smith, & Smithson, 1993)), lowering costs of raising external capital (Froot, Scharfstein, & Stein, 1993) and lower expected taxes ((Mayers & Smith, 1982) and (Smith & Stulz, 1985)). We will add another fifth theory: managerial risk aversion and hedging, mentioned by (Smith & Stulz, 1985). Before the elaboration of the different theories, We will give a definition of hedging as made by Smith & Stulz (1985). A firm can hedge by trading in a particular futures-, forward-, or option market even though it has no identifiable cash position in the underling commodity. Furthermore, a firm can hedge by altering real operating decisions; for instance, a merger can produce effects similar to those of hedging through financial contracts. This is the most general definition of hedging in terms of the market value of the firm. 3

8 2.1.1 Costs of Financial Distress In the Modigliani and Miller (MM) world, financial distress is assumed to be costless. Hence, altering the probability of financial distress does not affect firm value. If financial distress is costly, firms have incentives to reduce the probability of financial distress (Mian, 1996). Smith & Stulz (1985) argue that hedging is one method by which a firm can reduce the volatility of its earnings. Based on this model, probability of hedging is higher for firms with higher expected costs of financial distress. Probably firms with higher expected costs of financial distress are larger firms, because of more bankruptcy cases. However, Nance et al. (1993) argue that if there is a fixed cost component to financial distress costs, then smaller firms are more likely to hedge, because the fixed component is relatively larger for smaller firms than for large firms. So the empirical association between hedging and firm size is indeterminate, according to Nance et al. (1993). Nevertheless, Bartram, Brown, & Fehle (2009) say that, because of bankruptcy costs are less than proportional to firm size (Warner, 1977), so smaller firms should be more likely to hedge. There are also other factors that cause financial distress costs for example high leverage, short debt maturity, low interest coverage, and less liquidity (Bartram et al., 2009). Therefore firms with these characteristics are also more likely to use derivatives to hedge these financial risks. In contrast, firms with higher dividend yield are less likely to be financially constrained since these firms probably have stable cash flows and lower financial constraints, and therefore hedge less. Finally, lowering the chance of financial distress can also increase the optimal debt-equity ratio and therefore the associated tax shield of debt. So firms with higher tax loss carry forwards in their tax shields are more likely to hedge Contracting costs Myers (1977) characterizes firms potential investment opportunities as options and demonstrates that, with fixed claims in the firm s capital structure, taking a positive Net Present Value (NPV) can reduce shareholders wealth if the gains accrue primarily to the debt holders. Consequently, the shareholders can have incentives to forego positive NPV projects. Myers calls this the underinvestment problem. By restricting the states in which the firm would default on bond payments, hedging can control this problem. Hence, firms with more growth options in their investment opportunity set (higher market to book value) are more likely to undertake a hedging program aimed at reducing the variance in firm value. Since the underinvestment problem is more pronounced with 4

9 more debt in the firm s capital structure, firms with higher leverage are more likely to hedge (Nance et al., 1993). Also, prior work has shown that regulation is an important determinant of corporate financing and dividend policy choices (Smith & Watts, 1992), (Barclay & Smith, 1995a, 1995b). Managers of firms in regulated industries are likely to have less discretion in their choice of investments policies. Regulation also makes it easier for fixed claim holders to observe managerial action. As a consequence, firms in regulated industries face lower contracting costs and therefore have less incentive to hedge. Hence, the incentive contracting hypothesis predicts that hedging is less likely in the regulated utilities industry Capital market imperfections Froot et al. (1993) examine the role of capital market imperfections in determining the demand for corporate hedging. If access to external financing (debt and/or equity) is costly, firms with investment projects requiring funding will hedge their cash flows to avoid a shortfall in their funds that could precipitate a costly visit to the capital markets. So if firms are more likely to hedge, the more closely correlated are their cash flows with collateral values (and hence with their ability to raise external finance). Second, since it is likely to have more asymmetric information about the quality of new projects for firms with high market-to-book ratios and for firms that are not in regulated industries. Therefore, their model predicts that hedging is more likely for firms with higher market-to-book ratios and for firms that are not in a regulated utilities industry. Third, fixed costs associated with capital market visits are likely to make financing more expensive for smaller firms, therefore leading to the prediction that smaller firms are more likely to hedge Taxes Mayers & Smith (1982) and Smith & Stulz (1985) argue that hedging can reduce the expected tax liability for a firm facing a progressive corporate tax structure over the range of possible income outcomes. If hedging reduces the variability of pre-tax firm values, then the expected corporate tax liability is reduced and the expected post-tax value of the firm is increased, as long as the cost of the hedge is not too large (Smith & Stulz, 1985). In addition to progressivity in the tax schedule, which admittedly applies to a very narrow range of pre-tax income, other aspects of the corporate tax structure can also influence the hedging decision. Specifically, tax shields (tax loss carry forwards and foreign tax credits) introduce convexities in the corporate tax schedule. If firms do not hedge 5

10 their cash flows, the utilization of these tax shields may be postponed to a later date, therefore reducing their present value. Hedging increases the present value of these tax shields by smoothing out corporate earnings (Mian, 1996). So it is more likely that hedgers have tax-related progressivity than non-hedgers Managerial risk aversion and hedging The corporation s managers, employees, suppliers, and customers are frequently unable to diversify risks specific to their claims on the corporation. Because they are risk averse, the individuals require extra compensation to bear the non-diversifiable risk of the claims (Reagan & Stulz, 1983). With limited liability, the amount of risk that can be allocated to the stockholders is restricted by the company s capital stock. The firm can reduce the risk imposed on other claimholders by hedging. Thus, as long as the reduction in compensation of managers and employees and other suppliers plus the increased revenues from customers exceed the costs of hedging, hedging increases the value of the firm (Smith & Stulz, 1985). Merton (1974) also shows that equity value is an increasing function of asset volatility, so managers who are acting on behalf of the shareholders might have an incentive not to hedge. However, most senior managers have a very undiversified financial position because they derive substantial monetary and nonmonetary wealth from their firm. Consequently, risk aversion may cause managers to deviate from acting purely in the best interest of shareholders, expending resources to hedge diversifiable risk. Thus, we expect firms which are closely held, will use derivatives (Bartram et al., 2009). These are the main theories of the control variables and their relationship with derivative usage. The variables have all different directions and a strong influence the use of derivatives and therefore ideal control variables to put in the regressions. The predicted signs are summarized in the table below. Predicted Signs of Coefficient Estimates Prediction Independent variables S&S (1985) NSS (1993) BBF (2009) Leverage Size -? - MTBV + + Liquidity - - Current ratio - - Dividend yield + - TLCF Executive stock + + This table: Predicted signs of coefficient estimates, shows the predicted signs of coefficient estimates for variables used as proxy for incentives to use derivatives based on the testable implications of Smith and Stulz (1985), Nance Smith and Smithson (1993) and Bartrand, Brown and Fehle (2009). + (-) indicates a positive (negative) relationship between the variable and the use of derivatives and? 6 indicates that the predicted sign is indeterminate.

11 2.2 Surveys Since the nineties, derivatives are getting more popular and the variety and complexity of the available derivatives increased markedly. The broad array of derivatives enhances the ability of firms to manage their financial risk exposure in time of highly volatile exchange rates, interest rates, and commodity prices. Unfortunately, the same derivatives that permit firms to reduce risk also provide opportunities for risk-taking by firms, therefore complicating the task of overseeing financial activities within firms. Moreover, the reporting and disclosure requirements in the early nineties are not very comprehensive around the world. Therefore, different surveys are conducted and the most important U.S. non-financial firm survey is the Wharton survey (Bodnar et al., 1995). This survey draws a random sample of 2000 firms over 40 industries around the world in November The response yield is a respectable 26.5% and out of this group of firms 35% responds that they use derivatives. Almost 65% of the large firms, defined as those with market value above $250 million, use derivatives. At small firms (market value below $50 million) the percentage lies around 13%. The firm size results are consistent with the existence of significant fixed costs associated with starting and managing a derivatives program. Besides that, the risk exposures of smaller firms may be small relative to standard contract sizes. On the other hand it is striking with basis theory that small firms have more volatile cash flows, have more restricted access to capital, and thus presumably have more reason to buy protection against financial trouble (Stulz, 1996). Traditionally, commoditybased industries, such as agriculture, refining, and mining, show the highest usage at about 50%. Manufacturing industries report derivatives use by 40% of the firms. The regulated industry segment transportation and utilities shows 32% which is reasonably lower than other sectors and consistent with the theory in paragraph Derivatives use is least common in service industries with 14% and following by retail and wholesale trade at 29%. These results are consistent with the findings of other research examining derivatives use, like Nguyen & Faff (2010). In this Wharton survey, the firms were also asked to indicate the kinds of derivatives they use. They could indicate the derivatives (forwards, futures, swaps and options) into one of the four exposures: foreign exchange, interest rates, commodities and equities. The dominance of swaps as a vehicle for interest rate risk management stands out clearly. Forwards dominate the foreign exchange category, but the use of swaps and OTC (over the counter) options are also significant for managing foreign 7

12 exchange risk. None of the contract types dominates for firms hedging commodity and equity exposures. As mentioned in paragraph 2.1, in theory firms could have a variety of reasons why they should use derivatives. In the Wharton survey, firms are also asked to indicate the frequency of derivatives usage to accomplish eight commonly-given objectives. The results are that hedging firmcommitment transaction exposures are used by 80% of the firms, which use derivatives, while only 44% of the firms use derivatives to hedge the balance sheet. Taking a view, which indicates that firms use derivatives to speculate on the direction of prices or rates, is less common. The survey shows that 34% of firms seldom use derivatives to take a view, and 57% never do. This is consistent with the fact that the widely publicized cases of derivative disasters are the exception, not the rule. Finally, there was the question to rank the importance of three different risk management goals: minimizing fluctuations in accounting earnings, minimizing fluctuations in cash flows, and protecting the appearance of the balance sheet. As a risk management objective, minimizing fluctuations in cash flows is the best overwhelming choice, with 67% of firms assigning this objective primary importance. In contrast, managing quarterly accounting earnings is the primary choice of only 28% of firms, and protecting the appearance of the balance sheet is the primary goal for only 5% of the firms. 2.3 Prior research While there is survey evidence and some empirical evidence of derivative usage in the nineties, research examining derivative usage and the consequences for firms are surprisingly absent. Hentschel & Kothari (2001) say that such research is important because since the nineties the possibility that firms use derivatives to increase their risk exposures is a principal concern guiding regulatory agencies in their considerations of derivatives regulation. Derivatives users have consistently argued that regulators concerns about the dangers of derivative usage (i.e., speculation) are misplaced. They contend that the direct and indirect costs of excessive regulation will reduce derivatives usefulness. However, in the nineties, a number of end users were the victims of their risky strategies. Capelle-Blancard (2010) has enumerated the most outstanding cases in the nineties. First, Metallgesellschaft, one of the twenty largest German companies with losses of US$ 1340 million in 1994 due to using oil forwards (Culp & Miller, 1995; Verleger, 1999). Second, Orange 8

13 county in California with losses of US$ 1810 million in 1994 due to reverse repo operations (Culp & Miller, 1995; Miller & Ross, 1997) and Ashanti Goldfields with losses of US$ 570 million in 1999 on exotic gold products (Gilbert, 2001). Thanks to bankruptcies and regulators that caught up in rebellion in 1997, there was an intense debate about the appropriate level of derivative disclosure. Though, this was a difficult debate in the absence of systematic empirical evidence about the effects of derivative usage on firms stock return or cash flow volatility. Therefore, Hentschel & Kothari (2001) decided to do a widespread research on 425 large U.S. firms. The main conclusion of that result is that they cannot detect an economically or statistically significant relation between firms risk characteristics and the extent of their participation in derivative markets. Even for firms that hold large derivative positions relative to overall firm size, they cannot detect an economically significant link between derivatives and increased volatility. Their results are inconsistent with the hypothesis that firms use derivatives to speculate on a large scale. So, this complements to one conclusion of the Wharton survey that speculating is less common. Their results for non-financial firms complement also Koski & Pontiff's (1996) findings for mutual funds. Koski and Pontiff (1996) compare risk measures and higher moments of the return distributions of mutual funds that do or do not use derivatives. In the paper of (Hentschel & Kothari, 2001), endogenous use of derivatives, restrict the possibility to unmask the marginal effect of derivative usage on return volatility using simple regression procedures. Therefore they also estimate instrumental variable regressions. Unfortunately, this analysis also leads to the conclusion that firm s derivatives usage does not measurably increase or decrease their return volatility. There is another more recent study from Nguyen & Faff (2010), which examines almost the same relationship between financial derivatives and firm risk, but then for Australia for the years 1999 and Their results are consistent with the results of Hentschel & Kothari (2001). Employing the ordinary least squares (OLS) estimation technique, they found that at the aggregate level, there is no evidence that the use of financial derivatives is associated with a reduction in firm risk. There is neither any evidence that such usage leads to an increase in firm risk. The OLS technique does not give any results, but they use also the fixed effects model. The results from this model support the hypothesis that most firms use derivatives to hedge. A majority of firms that use derivatives enjoy a lower level of stock return variance as opposed to firms that do not. Additionally, while the use of financial derivatives among excessive users appears to lead to an increase in firm risk, there is no evidence suggesting that, ex post, derivative users are exposed to a level of risk that is higher than that of non-derivative users. 9

14 Notwithstanding these two papers with insignificant results, there are also papers with significant results. One paper of Guay (1999), finds that firm risk decreases in the period following the adoption of derivatives, using a sample of 254 nonfinancial firms that begin to use derivatives. The mean risk reduction (control-sample adjusted) ranges from 5% of stock return volatility across all sample firms, to 22% of interest-rate exposure and 11% of exchange-rate exposure for firms using interest-rate and exchange-rate derivatives, respectively. Therefore, the chosen types of derivative instruments are shown to depend on the risk exposures face by these firms in ways that are, on average, consistent with hedging behavior. There also is a smaller scale study from Allayannis & Ofek (2001) that focuses entirely on foreign currency derivatives (FCDs). Their results also report a reduction in foreign exchange exposure of the sample S&P 500 nonfinancial firms as a result of FCD usage. A third paper contains a sample of Swedish firms where Hagelin & Pramborg (2004) report evidence of foreign exchange exposure reduction associated with FCDs and foreign currency debt. Similarly, de Jong, Ligterink, & Macrae (2006) find that on-balance-sheet hedging seems to induce a decline in exchange-rate exposure on Dutch Firms. 2.4 Accounting standards Research regarding derivatives in early 90s, are almost all surveys, because companies do not have disclosure for off-balance-sheet financial instruments. Therefore it is difficult to obtain data yourself, that is why it was all done by questionnaires. However this changes after June 1992 by the Financial Accounting Standard Board s Statement of Financial Accounting Standards (SFAS) No This means Disclosure of information about financial instruments with off-balance-sheet risk and financial instruments with concentration of credit risk, including footnote disclosure of the face, contract, notional, or principal amount of derivative contracts in financial statements. Therefore, SFAS No. 105 requires that the information is reported separately for option, futures, and forward contracts. The standards are many times revised and expanded with fair values and whether these instruments are being issued or held for trading purposes or for other purposes. However, it is not relevant for this thesis to provide more precise information for derivatives, because only notional amounts for the separated instruments are necessary. 10

15 3. Sample Construction This thesis focuses on the largest non-financial firms of the U.S. for the period 2007 to The initial list of firms contains the S&P 500 firms in The 79 financial firms are excluded, because they are users and providers of derivatives, and use them for other objectives than non-financial firms do. The sector is based on the Wharton Survey (Bodnar et al., 1995) as mentioned in paragraph 2.2, and their results were that commodity based firms use derivatives for 50% of the time. The second largest sector of derivative usage is the manufacturing sector of durables and nondurables, with a percentage around 40% of derivative usage. This thesis will focus on the manufacturing sector, because of the large use of derivatives and by the lack of usable data for the commodity driven firms. For each firm, we searched manually for the annual reports in the Edgar database for the three years 2007 to As mentioned in paragraph 2.4, it is required for U.S. public firms to disclose off-the-balance-sheet financial instruments, and report this separately for the different instruments in their footnotes. Important, it is mandatory to disclose the use of the derivatives. This can vary between cash flow hedge, accounting hedge or fair value hedge, and no designation of hedge. In this thesis a distinction between derivatives used for hedging purposes and non-hedging purposes. So, if the derivatives were used for cash flow hedging, accounting hedging or fair value hedging it is under the heading hedge derivatives. All other derivatives not mentioned for hedging purposes are under the heading non-hedging derivatives. In the dataset all closing dates for the fiscal years are allowed. The earliest possible year end is September 30 and the latest possible year end is August. So the first possible year-end is September and this contains data from 1 October 2006 until 30 September If the firm has a fiscal year-end in August the first year contains data from September until August All other variables (like the different measures of risk and control variables) are taken on the fiscal yearend and are therefore a measure of the same past fiscal firm year. For example, the daily stock prices where the returns and variance are calculated from, is for a fiscal year-end in September, picked from 1 October 2006 until 30 September The notional amounts of derivatives are collected from the annual report of 30 September 2007 and the control variables are from Compustat at fiscal year-end This ensures that all variables are measures of the past year; therefore there is no comparison between two completely different time periods. This is taken into account for every firm in the sample. 11

16 The derivative usage is looked up using the search terms: derivatives and notional. Some firms have no notional amounts outstanding and only some fair values of realized gains or losses in the income statement. These derivatives are not included and the firms are dropped completely. If firms have notional amounts of commodity derivatives mentioned in volumes and besides that other derivatives outstanding, the company is completely dropped out of the sample. Only if there are significant notional amounts in dollars or there are no indications of derivative usage, they are included. If firms have missing stock data for each of the years between 2007 and 2009 or there is no annual report available, or there was a spin-off or acquisition then the firm is dropped from the sample. The firms with a two digit SIC code between 20 and 39 are known as manufacturing firms and counted in 2009 for 191 firms. The potential sample size is therefore 573 firm year observations using the initial 191 firms. 17 firms are dropped because of missing data or acquisitions/spin-offs in the years between ; this makes the total sample size of 174 firms with 522 firm years. The firms included in this sample are on the company list in the appendix. The other variables for the research can be obtained by WRDS (Wharton Research Data Services). Specifically, the stock price data for the different risk measures, market model betas and annual returns can be found in WRDS by CRSP (Center for Research in Securities Prices). The control variables for the regressions can be found in Compustat and Execucomp. 12

17 4. Methodology In this section, there will be explained how all dependent-, independent- and control variables are selected and formed. Thereafter, the different methods for the regressions will be explained as well the hypotheses for the variables. 4.1 Measures of firm risk In this thesis, there are three dependent variables taken into account. These three variables are based on the study of Nguyen & Faff (2010), namely total risk, systematic risk, and idiosyncratic risk. In the following paragraphs the formation and intuition will be explained Total risk The first and most important dependent variable is total risk. This risk contains both systematic and idiosyncratic risk, and is known by the market as stock volatility. In this thesis the total risk will be measured as the variance of the daily stock returns for the separated years the derivatives are reported. So the variance is taken of all daily returns in the specific fiscal year, which results in three variances per firm, namely for the years 2007, 2008 and It is therefore considered that every firm has his own fiscal year-end Systematic risk The systematic risk variable, formed by the market model is specified as follows: The unknown variable in this equation (1), is the systematic risk variable (. The variable is the daily return on stock i, is the daily return on the index and is the error term. One has to regress the market (index) returns, in this specific case the returns of the S&P 500, on the returns of the individual firms to estimate the beta. Beta measures the sensitivity of a security to market-wide risk factors. For a stock, this value is related to how sensitive its underlying revenues and cash flows are to general economic conditions. The average beta of a stock in the market is about 1; that is, the average stock price tends to move about 1% for each 1% move in the overall market. Safe utilities 13

18 most of the time have a beta smaller than one, but high tech or pharmaceutical firms have a beta larger than one (Berk & Demarzo, 2011). Unfortunately, there is no direct empirical evidence about the relationship between systematic risk and derivatives usage. However, Schrand & Unal (1998) have evidence that firms coordinate their risk levels by decreasing homogeneous risk to shift it to core-business risk. This is done, because firms earn zero economic rents in efficient markets for bearing financial risks such as unexpected changes in interest rates, foreign currency exchange rates, or commodity prices (homogeneous risk). Nevertheless, firms earn rents or economic profits for bearing risk related to activities in which the firm has a comparative information advantage (corebusiness risk). This implies that firm specific risk should decrease, because there is no compensation for this kind of risk and that systematic risk increase, because firms are compensated for this risk Idiosyncratic risk The last dependent variable is idiosyncratic risk, also known as specific firm risk or unsystematic risk. This risk cannot be diversified away and therefore it is a good reason to have this risk as low as possible. This risk can be calculated by taking the variance of the residuals from equation (1). The expectation is, according to Nguyen & Faff (2010), that idiosyncratic risk is a large part of total risk and that they are therefore highly correlated. 4.2 Derivative usage This thesis will look from different angles to make the relationship between derivative usage and risk clear. The main variable is derivative usage (all derivatives) in general, where all derivatives outstanding are put together for the effect in general. The notional amounts in millions of dollars are collected from each firm and scaled by size. This is also done by the other independent derivative usage variables, to compare the derivative usage among firms, because it is transformed from an absolute number into a relative number expressed in percentages. Derivative usage can also be split in several ways, for example in hedge derivatives and non-hedge derivatives. Firms have to show this difference in their annual reports and the sum of these two notional amounts is exactly the same as derivative usage in general. Under the heading hedge derivatives are counted derivatives for cash flow hedges, fair value hedges and accounting hedges. For firms it is mandatory to make this distinction in the notes of the annual report. If there is mentioned that the derivatives are used for non-hedge purposes the amounts fall under the heading 14

19 non-hedge derivatives. The distinction between hedge and non-hedge derivatives is made, to visualize the fact if firms are reducing risk with derivatives used for hedging, and increasing risk by speculating with non-hedge derivatives. Since, this is the expectation because firms do not take a real position in derivatives, when they are used for hedging. They use derivatives for hedging to cover the risk of, for example a price change in the underlying. When derivatives are used for speculating (non-hedging) the position in the derivatives is naked and there is a gamble on an increase or decrease of the underlying. The third distinction in derivative usage is the underlying. In this thesis three underlying instruments are reported called commodities, foreign currency (FCD) and interest rate risk (IRD). In this case also the sum of these variables is the derivative usage in general. With this distinction it hopefully will be visible on which underlying; the most risk reduction or increase can be made. Finally, there is made a distinction between derivative products like forwards, futures, swaps, options. If the firm reports for example foreign currency contracts, but don t make any distinction in products, than it falls under the heading derivative contracts in general. This distinction is made to visualize on which product the most risk reduction or increase there can be made. 4.3 Control variables To identify the impact of derivative usage in general and the different distinctions, it is necessary to isolate other factors that also contribute to firms total level of risk. As a result, there are some control variables selected for the multivariate regressions. These control variables are for many papers more or less the same, and therefore the eight most significant variables from the papers Nguyen & Faff (2010), Hentschel & Kothari (2001) and Gay & Nam (1998) are chosen. Leverage: The variable leverage is measured as the sum of long term debt, debt mortgages and other secured and debt notes scaled by firm size. As announced in paragraph there could be an underinvestment problem, when there is a large amount of debt and gains primarily goes to debt holders. Therefore, the shareholders can have incentives to forego positive NPV projects. Hence, firms with more growth options in their investment opportunity set are more likely to undertake a hedging program, aimed at reducing the variance in firm value. Therefore, since the underinvestment problem is more pronounced 15

20 with more debt in the firm s capital structure, firms with higher leverage are more likely to hedge (Nance et al., 1993). Size: The variable size is measured as the sum of market value of equity and the book value of debt (long term debt plus debt mortgages, other secured and debt notes). It can also be seen as a proxy for economies of scale in the costs of hedging. As mentioned in paragraph 2.1.1, the relationship between firm size and total risk is indeterminate. According to Nance et al. (1993), there are alternative arguments for either a positive or negative relation between firm size and hedging activity. For example, smaller firms should hedge more, ceteris paribus, because of the inverse relationship between firm size and bankruptcy costs (Warner, 1977). We might also expect a negative relation between firm size and the use of derivatives if smaller firms have greater information asymmetries. So it is unclear whether firm size is going to have a positive or negative impact on firm risk. However, the latest empirical evidence shows that larger firms are associated with a lower level of risk (Guay, 1999; Hentschel & Kothari, 2001; Nguyen & Faff, 2010). Market to book Value (MTBV): The variable MTBV is measured as market value of equity divided by stockholders equity (book value of equity). As mentioned in paragraph firms with more growth options in their investment opportunity set are more likely to undertake a hedging program. This is because high-growth firms are perceived to be more risky as they are more exposed to the risk of underinvestment. Therefore, underinvestment is costly as it represents the failure to maximize shareholders wealth (Froot et al., 1993). Liquidity: The variable liquidity is measured as cash and short term investments scaled by size. As mentioned by paragraph and by different hedging theories, the prediction is that firms having ample internal funds can avoid the risk of obtaining costly external funds and hedge less. As a result, liquidity is associated with relatively low underinvestment risk and hence low overall risk (Nguyen & Faff, 2010). However, the flipside to the coin mentioned also by Nguyen & Faff (2010) is that extreme liquidity, also known as free cash flow, triggers management entrenchment and therefore induces additional volatility. This is based on the agency theories by Jensen & Meckling (1976) and Shleifer & Vishny (1989). So a reasonably degree of liquidity is expected to be associated with a lower level of firm risk, due to a lower level of underinvestment risk. However, extreme liquidity leads to an increased level of risk due to agency problems, so the relationship between liquidity and firm risk appears to be nonlinear. 16

21 Current ratio: The variable current ratio is measured as current liabilities divided by current assets. Current ratio is also a measure of liquidity and has therefore the same intuition. Dividend yield: The variable dividend yield is measured as dividend per share at pay date at fiscal year-end divided by the price of the stock at fiscal yearend. The expectation is, mentioned in paragraph and according to Bartram et al. (2009), that firms with higher dividend yield are less likely to be financially constrained since these firms probably have stable cash flows and lower financial constraints. In contrast to that paper, dividend yield is not a dummy to be equal to unity if they pay dividend and zero if they do not, but a real percentage. Hopefully, this gives a better insight in the use of dividend yield and total risk. Tax loss carry forwards (TLCF): The variable tax loss carry forwards is measured as the absolute number tax loss carry forwards scaled by firm size. As mentioned in paragraph when firms do not hedge their cash flows, the utilization of the tax shields may be postponed to a later date, and therefore reducing their present value. So hedging increases the present value of these tax shields by smoothing out corporate earnings. Therefore, the expectation is that firms with high tax loss carry forwards as a percentage of firm size are hedging more than firms with no tax loss carry forwards. Executive stock percentage: The variable executive stock percentage is measured as the absolute number of executive shares owned (excluded options) scaled by common shares issued. This variable common shares issued contains all common shares outstanding and treasury stock. As mentioned in paragraph and according to Bartram et al. (2009), there are many theoretical models (e.g., Merton, 1974) which show that equity value is an increasing function of asset volatility, so managers who are acting on behalf of the stockholders might have an incentive not to hedge. However, most senior managers have a much undiversified financial position because they derive substantial monetary and nonmonetary wealth of their firm. Consequently, risk aversion may cause managers to deviate from acting purely in the best interest of shareholders, expending resources to hedge diversifiable risk. Thus, Bartram et al. (2009) expect that firms that are closely held will use derivatives. In paragraph 5.2 the summary statistics of these control variables are shown. There is also made a distinction between the statistics of non-derivative users and derivative users. If the difference between control variables of the two groups is significant, is tested by the Wilcoxon rank sum test. 17

22 There is chosen for a Wilcoxon rank sum test instead of a parametric t-test, because of no normality in the control variables. The normality is tested by the Shapiro-Francia W test and for all variables the null hypothesis of normality is rejected. Since the t-test maximizes power when the distribution is normal this test is dismissed, according to the book statistics for business and economics (McClave, Benson, & Sincich, 2005). The nonparametric Wilcoxon rank sum test is used to test the hypothesis that the probability distributions associated with the two populations are equivalent just is done as in the paper of Bartram et al. (2009). According to McClave et al. (2005), the idea behind this test is to compare the probability distributions of the samples by ranking the sample observations as though they were all drawn from the same populations. That is, we pool the measurements from both samples and then rank the measurements from the smallest (a rank of 1) to the largest (a rank of 13). If the two populations were identical, we would expect the ranks to be randomly mixed between the two samples. If, on the other hand, one population tends to have larger percentage changes than the other, we would expect the larger ranks to be mostly in one sample and the smaller ranks mostly in the other. Thus, the test statistic for the Wilcoxon test is based on the totals of the ranks for each of the two samples, that is, on the rank sums. The greater the difference in rank sums the greater the evidence to indicate a difference between the populations. The only two conditions that must be met for a valid test is that the two samples are random and independent, and that the two probability distributions from which the samples are drawn are continuous. These assumptions are met, so the test can be done and the results are shown in paragraph Regression analyses In this section the methodologies of the different regressions will be shown in the form of a formula with the corresponding explanation. The first paragraph explains the portfolio analysis, followed by the ordinary least squares regression, fixed effects regressions and finally the interaction regressions Portfolio analysis The preliminary investigation of the role of financial derivatives in reducing stock return variance will be a univariate portfolio analysis. The meaning of this analysis is to check whether there is a linear relationship between derivative usage and total risk. All firm years were divided into decile portfolios according to the level of derivative use, as measured by the total notional value of all derivative contracts scaled by firm size. It does not matter if a firm has a large derivative usage one year and a small usage the year after, because all firms are separated in standalone years. For this 18

23 analysis it does not give any problems, because of the yearly determined risk and control variables. All years that do not have any derivatives outstanding are grouped into portfolio 0. Portfolio 1 consists the least intensive derivative user years while portfolio 10 contains the most intensive derivative user years. If firm risk is a linear function of financial derivative use, it is expected that firm years in portfolio 10 have the highest level of risk and for portfolio 1 otherwise. This thesis does not only look at firm risk as a function of derivative use, but also at the relations of the control variables and derivative usage. The results can be found in paragraph Ordinary least squares regressions In this paragraph the linear relationship of derivatives usage and the different measures of risk will be formulated. Every firm has three points in time namely 2007, 2008 and 2009 where data is collected. Therefore, the relationship will be tested for each year separated, with and without control variables and for all three measures of risk. When control variables are used, three regressions per risk measure are made: (6) These regressions can be repeated for the risk measures systematic risk and idiosyncratic risk. When the control variables are dropped, these equations are the regressions for the separated years without control variables. Total risk, systematic risk and idiosyncratic risk are defined in the paragraph 4.1 measures of firm risk ; deriv is the total notional amount of derivatives outstanding scaled by firm size defined in paragraph 4.2 derivative usage. Hedge and non-hedge stands 19

24 respectively for hedge derivatives and non-hedge derivatives also defined in paragraph 4.2 Com, FCD and IRD stand respectively for commodity derivatives, foreign currency derivatives and interest rate derivatives also defined in paragraph 4.2 Fwrd, Swap, Option, Futures and Contract stands respectively for forward derivative usage, swap derivative usage, option derivative usage, futures derivatives usage and non-specified derivatives contracts, all defined in paragraph 4.2. The other variables are control variables and are defined in paragraph 4.3 control variables. In short lev (Leverage), size (size), MTBV (market to book value), liq (liquidity), CR (current ratio), DY (dividend yield), TLCF (tax loss carry forwards), Exes (executive stock percentage). Equation (2) aims at discovering the impact of the overall derivative position on firm risk every year, while equation (3), (4) and (5) examines the impact at the individual derivative level by partitioning the use of financial derivatives into hedge and non-hedge derivatives, three different underlying instruments and the five derivative products. In Equation (6) an interaction term is added to check for example whether more levered firms which use derivatives have more impact on firm risk. This can also be done for the different risk measures and derivative usage variables with all control variables. To control for possible outliers in the sample there is done some robust regressions. It could be that the results are more reasonable, because outliers do not have any impact anymore. The data points are no data entry errors and the firm observations are from reasonable firms, therefore there is no reason to exclude these points from analysis. Robust regressions can be a good method to compromise between excluding these points entirely and including all the data points and treat them all equally in OLS regression. The method that is used to control for outliers is to run an OLS regression and then get the Cook s D for each observation. After that, drop any observation with Cook s distance greater than 1. This process repeats until the tolerance level is reached. This form of regression can also be called a weighted and reweighted least squares regression. The confidence level is 95%, so only the most influential points are dropped 3. These regressions give the relationship for every year separated. The next ordinary least squares (OLS) regression is the pooled regression where all firm years are combined to one sample. Thereafter the regressions in equation (2), (3), (4), (5) and (6) are done for the whole sample for all risk measures. The aim of this regression is to check if there is an effect on firm risk by using derivatives over the time period 2007 to 2009, without controlling for firm fixed effects. This is also the main problem of OLS that it is not corrected for firm fixed effects, what provides heterogeneity 3 Introduction to SAS UCLA: Statistical Consulting Group. From: 20

25 and lower explanatory power of the coefficients. This could be solved partially by the regressions in the next paragraph Panel data regressions OLS regressions are probably not the best method to examine the relationship between derivative usage and total risk, because of the multiple observations per individual firm and probably a nonlinear relationship between risk and derivatives. This regression does not correct for the specific effects within every firm. Therefore, panel data regressions are developed to deal with this problem. The definition according to the book analysis of panel data of Hsiao (2003) is as follows: A panel data set is one that follows a given sample of individuals over time, and thus provides multiple observations on each individual in the sample. This is exactly the case for this sample and therefore these techniques are used. There are some benefits by using panel data compared to the ordinary least squares regression. First, with panel data it is possible to do a cross-sectional time series regression. Second, the results are more informative because of more variability, less collinearity (high correlation between independent variables), more degrees of freedom, so in other words estimates are more efficient. Finally, and most important benefit is to control for individual unobserved heterogeneity. The case is that most of the time empirical studies find (or does not find) certain effects, because of the presence of omitted (miss measured or unobserved) variables that are correlated with explanatory variables (Hsiao, 2003). So, with panel data it is possible to control for this unobserved heterogeneity what results in better estimates. The idea behind this panel data model is that the error term can be decomposed in two components, according to (Brüderl, 2005; Hsiao, 2003): A person-specific error and an idiosyncratic error,. For simplicity, the assumption is that there are no time-specific effects, but only individual-specific effects. In section five it will be clear after some tests, that only the fixed effects model is effective in contrast to the random effects model, therefore only the fixed-effects model will be explained and used. The model according to Hsiao (2003) is as follows: (7) where the value of the dependent variable for the ith unit at time t,, depends on K exogenous variables, ( ), is a 1 x K vector of constants and is a 1x1 scalar constant representing the effects of those variables peculiar to the ith individual in more or less the same fashion over 21

26 time. The person-specific error does not change over time. Every person has a fixed value on this latent variable (fixed-effects). Represents person-specific time-constant unobserved heterogeneity. The idiosyncratic error varies over individuals and time. This all sounds as a perfect model, nevertheless there are some restrictions which are good to know. According to Brüderl (2005), the first and most important restriction is that with fixed effects regressions it is not possible to estimate the effects of time-constant covariates. These are all cancelled out by the within transformation. Consequently, panel data do not help to identify the causal effect of a time-constant covariate. Second, there must be some variation in X. Otherwise; it is not possible to estimate its effect. This could be a problem, if only a few observations show a change in X. Finally according to Brüderl (2005), there is also one important assumption that must be met to use the fixed effects model. This assumption is: for all t and s. This is the assumption of (strict) exogeneity. If it is violated, there is an endogeneity problem what means that the independent variable and the idiosyncratic error term are correlated. Under endogeneity the fixed effects estimator will be biased. Endogeneity could be produced inter alia by omitted variables, measurement errors or simultaneity and reverse causality Interaction regressions The regressions in the past paragraphs are all examinations with regard to a linear relationship between the use of derivatives and firm risk. However there is past evidence by for example Guay (1999) which indicates that one of the reasons leading to an insignificant coefficient is the nonlinearity between firm risk and the explanatory variables. Therefore this paragraph aims for examining for a nonlinear relationship between firm risk and derivative usage. The way to test this is based on the paper of Nguyen & Faff (2010), where this test is also done. Into equation (2) there will be put two dummy variables. The first dummy variable is set equal to unity if the size of derivative usage is less than a threshold that will be defined later. Similarly, is set equal to unity if the size of derivative usage is above that threshold and zero otherwise. The threshold will be chosen on the basis of the portfolio analysis from paragraph The threshold level in percentages is the 22

27 average amount of derivatives usage scaled by size from the portfolio where maximum risk reduction is achieved. More specifically, the following equation is estimated: [ ] [ ] This regression (equation 8) will be repeated also for the derivative usage distinctions. Anyway, the coefficients of primary interest is in this equation are and. It is expected based on the results of Nguyen & Faff (2010) that moderate users in this case with a threshold of 20% will experience risk reduction in risk and hence a negative sign is predicted for. Vice versa for it is expected that it has a positive sign because of excessive derivative usage and therefore probably firm risk increases. For the variables hedge derivatives and non-hedge derivatives the expectation is that hedge derivative users under the threshold will decrease risk and above increase risk and for non-hedge derivatives that it always increase risk, because of speculating. There is no specific expectation for the different underlying instruments, only that under the threshold they all will decrease risk and above increase risk. 23

28 5. Empirical findings 5.1 Summary statistics of derivative usage As can be seen in the Wharton survey of Bodnar et al. (1995), the percentage of responding firms that use derivatives for manufacturing firms is around 40% in Between 2007 and 2009, almost 79% of the 174 manufacturing firms in the S&P 500 used derivatives for one year or more. Besides, the users of derivatives are increased in 2007 from almost 57% to almost 74% in An explanation could be that firms are eager to cover risk after the credit crisis. There are also large differences in usage between the different derivative groups for hedging purposes. Only 17.8% of the firms report notional amounts of any commodity contract for one year or more, in contrast to 50.6% for interest rate contracts and almost 87.4% for foreign currency contracts. Figure 1 (appendix) shows these firm derivative usages for hedging in a more specific way. This figure is quite similar to the Wharton survey result, and therefore in this respect not much has changed. However, table 1 (appendix) summarizes the basic characteristics of the sample firms derivative positions and shows that there is an enormous growth in notional amounts with respect to the Hentschel & Kothari (2001) paper. Where the total average derivatives outstanding, for nonfinancial firms with derivatives in general, were $1.7 billion in 1992 to 1993, the average for 2007 to 2009 was more than $3 billion for manufacturing firms. Moreover only derivatives for hedging purposes are included in this figure. A comparison of the mean and median shows that average derivative holdings are influenced by a few very active firms. Therefore, it can be seen that there are very few firms which use derivatives for other purposes than hedging. This result is also quite similar to the Wharton Survey conclusion. For every derivative group there are many descriptive statistics shown like, notional amount, mean, median, standard deviation, minimum and maximum. However for the remainder there are no striking results, but only confirmations for figure 1 that interest rate swaps and foreign currency forwards are the most frequently used derivatives. Tables 2 and 4 (appendix) show that derivative usage in general have an increasing trend between 2007 and 2009, and that all derivative groups are separately increasing in notional amounts and in amount of firms. Finally, table 3 (appendix) shows descriptive statistics of the different derivative instruments. Forwards, swaps, and options are the most commonly used instruments. However the 24

29 report style of derivative contracts in general is most commonly used. These tables show very large notional amounts even in the trillions. These amounts are all absolute numbers and therefore difficult to compare, so in table 5 the notional amounts averages scaled by firm size are shown. The average amount of derivatives outstanding is 6.71% of the firm size, where the largest part consists of hedge derivatives and a small part of non-hedge derivatives. Foreign currency derivatives and interest rate derivatives are the most popular, which also can be seen in figure 1 (appendix). In table 6 the correlations between the different derivative usage variables are shown. The results are consistent with table 5, that the largest part of all derivatives is hedge derivatives and the correlation between those two variables is almost one. Therefore interest rate derivatives have a correlation of almost 0.90 with respect to all derivatives even as swaps, so probably these variables have the largest share in total derivatives. Considerably, the notional principal of the derivatives positions only gives a rough indication of the size of the exposures. This is well illustrated in the Hentschel & Kothari (2001) paper: if two call options are identical, except that one is deep out of the money, while the other one is deep in the money, the deep-in-the-money option provides a much higher exposure to the underlying asset at the same notional principal. Similarly, swaps with longer maturities have higher exposures than similar swaps with shorter maturities. Consequently, the reported notional principal values have to be interpreted with care. Therefore in the next paragraphs there will be a more specific analysis if there are more or less risks for firms with derivative exposures outstanding. 5.2 Summary statistics for the dependent- and control variables In table 7 the descriptive statistics are shown for the three dependent- and all control variables. In this table the amount of observations, mean, median, standard deviation, minimum and maximum are mentioned. First, total risk has an average of , which means that the average of all firms has a yearly stock return variance of The median is not far removed from the mean, and is therefore a reasonable figure. Second, the average beta (systematic risk) for all firms is almost 1 and varies between 0.25 and 2.8. Third, the average of the idiosyncratic risk variable is with a median of and varies between almost zero and Forth, the average leverage ratio is 23%, the median 19% and varies between 0 and almost 100%. In this case there are probably some outliers to look after. Fifth, the average firm size is million dollar, so almost 29 billion dollar. The size is very varied, but that does not matter that much, as it is mainly used as scaling 25

30 factor. Six, the average market to book value is 4 and the median is 2.8. These are very large figures and are probably formed by the outliers. Seven, the variable liquidity does not look strange and varies between 0.5% and 81%. For the variables current ratio and dividend yield the same applies. The variable tax loss carry forwards has a much larger mean than median, probably through some outliers at the right side. Finally, the executive shares percentage variable shows nothing strange with a mean of 8% and a median of 2%. In table 8 there is made a distinction between derivative users and non-derivative users. In this way the summary statistics can be seen for the separated sample groups. If there are any significant differences between the means of the two groups is tested by a Wilcoxon rank sum test. The null hypothesis is that the distributions are the same and therefore there is no difference between the two samples. As can be seen in the table there is only a significant difference at a 1% level in the probability distribution of leverage. All other variables have the same probability distribution and therefore there is no difference between the variables between users of derivatives and non-users. These different control variables are chosen because of significant results in past studies. When we look at the different control variables in table 9, the correlations are very low and most of the time even negative. So probably they explain all different parts of the risk measures, and collinearity should not cause problems. 5.3 Portfolio analysis When firms change its status from a nonuser of derivatives to a less intensive user (portfolio 1), on average its stock return variance reduce by approximately 10%, as can be seen in table 10 and figure 2. The portfolio with the lowest average stock return variance is portfolio 8 with a variance of and with the second lowest variance is portfolio 2. These two portfolios have respectively 13.35% and 1.53% derivatives outstanding scaled by firm size. The largest risk reduction takes place between portfolio 7 and 8 with a risk reduction of almost 40%. The derivatives usage increase from 9.68% of portfolio 7 to 13.35% of portfolio 8. There are only four portfolios (6, 7, 9, and 10) with a larger average risk return variance than the nonusers in portfolio 0. The maximum risk reduction achieved in portfolio 8 suggests that, in the current sample, there exist an optimal percentage of derivatives outstanding which appears to be 13.35%. In figure 3 total risk and idiosyncratic risk are once again shown, but now measured in stock return volatility. The figure looks almost the same as in figure 2 where it is measured in variance. The only difference is that it now more flatten and it 26

31 looks that the volatility of the firms do not differ that much. Comparing to the paper of Nguyen & Faff (2010) with a sample of the 500 largest Australian firms the results differ something, but not very much. Nguyen & Faff (2010) report the largest risk reduction in portfolio 6 with an optimal percentage of derivatives outstanding of 19.55%. For portfolios 7, 8, 9 and 10 firm risk are rising as a function of derivative usage intensity. So it could be said that the results per sample differs, nevertheless the existence of an optimal level of derivative usage provides an indicator that the relationship between the use of financial derivatives and firm risk may not be linear. In figure 2 the relationship is shown for total risk, but also for the other measures of risk. They all show a trough at portfolio 8, so all risks are the least in portfolio 8. Leverage decreases as derivative usage increases, however this is not consistent with the expectation that firms with higher leverage are more likely to hedge, according to Nance et al. (1993). The control variable market to book value has one peek and trough when the average of each portfolio is used. In paragraph 5.2 is already mentioned that there are some outliers, so when the median of each portfolio is used it is almost a flat line. However, this is not the expectation that firms with more growth options in their investment opportunity set are more likely to undertake a hedging program. The dividend yield is for portfolio 0 and 10 almost the same with one trough and peek at portfolio 4 and 5. This behavior is inconsistent with the expectation that stable cash flow firm with lower risk and higher dividend yield hedge less. The variable size is almost flat, except for portfolio 0 where the average firm size is much higher than in all other portfolios. So the largest firms do not use derivatives on average to reduce firm risk. So the hypothesis that smaller firms are more likely to hedge could be true with this results. The last variable liquidity goes up and down, where in the end the liquidity for portfolio 0 is larger than portfolio 10. This is not the expectation, because firms with less liquidity could have more distress so firms are therefore more hedging. These are the first global results of the portfolio analysis. In the next paragraphs the research will be more specified. 5.4 Ordinary least squares regressions In this paragraph the results of the ordinary least squares (OLS) regressions are presented. First, the direct effect of derivatives on the three measures of firm risk for the years 2007, 2008 and 2009 will be shown. Second, the results of these OLS regressions corrected by control variables will be presented. Finally, all separated years will be pooled to one sample and will be regressed on different derivative usage measures on firm risk, corrected by the control variables. 27

32 Regressions for separated years without control variables In table 11 the results are shown for the first year (2007). All derivative usage variables are regressed on their own and in the case for hedge and non-hedge derivatives and for the three underlying instruments they are also combined. The reported t-statistics are not adjusted for heteroscedasticity in table 11, and also for table 12 and 13, because the Breusch-Pagan (1980) test has proven that there is no heteroscedasticity for these separated samples. In column (1) the independent variable derivative usage (all derivatives) is regressed on the three measures of firm risk. The conclusion that can be drawn is that derivative use does not appear to be associated with any change in firm risk. Therefore, derivative usage has a very low explanatory power for the variability in firm risk as indicated by the R-squared (<0.01). In column (2) and (3) the hedge and non-hedge derivatives are not associated with any change in risk for 2007, although the R-squared is a bit higher. Columns (4), (5) and (6) show the results for the separated underlying instruments: commodity, foreign currency and interest rate with respect to the measures of risk. These results are not significant and the R-squared goes to zero except for interest rate derivatives. Column (7) show the results of these underlying s combined with respect to the measures of risk. Unfortunately the results are not significant, although the R-squared go up for every risk measure. Finally in column (8) the hedge and non-hedge derivatives are combined and also these results are not significant. Looked at the various risk measures, it can be said that systematic risk can be the least explained by derivative usage in 2007, although all estimates are not significant. In table 12 the results are shown for year The body of this table is exactly the same as in table 11, however for this year there are some significant results. Only the significant results are mentioned and all other results can be seen in the appendix. The most striking results can be found in column (5) and (7) for the underlying foreign currency. In column (5) the relationship between foreign currency derivatives and the risk measures are shown. By one standard deviation increase in foreign currency derivative usage the total risk (stock variance) increases with This result is statistically significant at a 5% level and has an R-squared of The same is happening for systematic risk where the use of one standard deviation increase in foreign currency derivatives results in a beta increase of , statistically significant at a 10% level. Finally, the foreign currency derivatives variable is also significant at a 5% level in the regression with idiosyncratic risk. In column (7) with the regression of all underlying s combined is the foreign currency derivatives variable also significant at the same levels and almost the same coefficient. So for this year only, the 28

33 amount of foreign currency derivatives outstanding scaled by firm size have a statistically significant result. Although, these results are not a very economically significant, because one standard deviation increase in FCD s result in a total risk increase of (table 5) times , in other words The other two FCD coefficients are also not economically significant with respect to systematic risk and idiosyncratic risk. In table 13 the results are shown for the last year (2009). The body of this table is the same as the past two tables. More precisely, this table looks like table 11, because there are no significant results to present. The foreign currency derivatives t-statistics are now very small and far from significant. The only variable with almost some significant results is the variable interest rate derivatives with a t-stat of 1.58 for total risk. So it can be concluded that there are no direct relationships between derivative usage and the different firm measures. There is a possibility that the OLS model is wrong for this data set or, there are some variables omitted what ensures in wrong results. Therefore some control variables are added to the model Regressions for separated years with control variables The results in the last paragraph were most of the time not significant and the R-squared was very low. So the univariate linear model does not explain the different measures in firm risk. This is due to omitted variables, or there is no real relationship. In this paragraph the equations (2), (3), (4) and (6) from paragraph are regressed, which includes control factors to get hopefully better results. The reported t-statistics of these models were adjusted by the White (1980) method for heteroscedasticity, because the Breusch- Pagan (1980) test has proven that there is heteroscedasticity. In table 14 the results are shown for the year The first three columns are the results for total risk, column (4) to (6) are the results for systematic risk and, finally column (7) to (9) are the results for idiosyncratic risk. In every first column of each risk measure all derivatives are regressed and controlled by eight variables (equation 2). In every second column of each risk measure there is made a distinction between hedge and non-hedge derivatives, like in equation (3). In every third column of each risk measure there is made a distinction between the underlying instruments, commodity, foreign currency and interest rate, like in equation (4). First take a look at columns (1) to 29

34 (3) for an explanation of derivative usage and total risk. However, all independent derivative usage variables were insignificant, just like the regressions in table 11 without control variables. On the other hand there are four control variables significant, which explain some of the total risk. The first control variable is size, with a coefficient of almost zero, so the economic effect is almost zero, but statistically significant at a 5% level. The second control variable is dividend yield with a coefficient of , significant at a 1% level. So a one standard deviation increase in dividend yield, results in a decrease of total risk by The third control variable is tax loss carry forwards (TLCF) with a coefficient of and statistically significant at a 1% level. This is also a small economic effect, where firm risk increases with a very small amount as the TLCFs relatively increase with respect to firm size. The last significant control variable is executive stock percentage with a coefficient of at a 5% significant level. So, just like TLCFs total firm risk increases when executive stock percentage increases. The R-squared for these regressions is around which much larger than for the regressions without control variables. This means that this model sounds better and more plausible; however any increase in the amount of control variables ensures a larger R-squared. The columns (4) to (5) show the relationship between derivative usage and systematic risk. None of the independent variables about derivatives usage are statistically significant. Only the control variable dividend yield is statistically significant at a 5% level with a coefficient of So, one standard deviation increase in dividend yield results in a decrease in beta by 0.1. The R-squared for these regressions is around The columns (7) to (9) show the relationship between derivative usage and idiosyncratic risk, however just like all other columns none of the independent variables are significant. On the other hand the same control variables as in columns (1) to (3) are significant, with almost the same coefficients and an R-squared of around It can be said that total risk is explained by the same variables as idiosyncratic risk and that these variables do not have any effect on systematic risk, except for dividend yield. In table 15 the results are shown for the year The body is exactly the same as in table 14 and only the significant results will be mentioned. None of the independent derivative usage variables are statistically significant. The control variables that explain something of total risk are leverage, liquidity, dividend yield and TLCFs. Leverage has a coefficient of and is statistically significant at a 1% level. In contrast to year 2007, liquidity is statistically significant at a 5% level with a coefficient of Dividend yield is statistically significant at a 5% level with a coefficient of and TLCF is significant at a 1% level with a coefficient of So leverage, liquidity and tax loss carry forwards increase total firm risk and only dividend yield decrease total risk. The R-squared 30

35 of these models (0.257) are slightly higher than for Systematic risk can be explained by the variables leverage, liquidity and TLCF. Leverage is significant at a 1% level with a coefficient of , so one standard deviation increase in leverage ensures a higher beta of Liquidity is statistically significant at a 1% level with a coefficient of and TLCF is significant at a 10% level with a coefficient of Idiosyncratic risk can be explained by leverage, size, liquidity, dividend yield, TLCF and executive stock. Leverage is statistically significant at a 1% level with a coefficient of Size has a coefficient of almost zero and is significant at a 10% level. Liquidity has a coefficient of and is statistically significant at a 5% level. Dividend yield has a coefficient of and is statistically significant at a 5% level. TLCF has a coefficient of and is statistically significant at a 1% level and finally executive stock has a coefficient of and is statistically significant at a 5% level. The R-squared of these models for idiosyncratic risk is around In table 16 the results are shown for the year For this table the body is exactly the same as the past two tables and only the significant results will be mentioned. For this year there are some independent derivative usage variables significant. The first variable is non-hedge derivatives in column (8) with a coefficient of and statistically significant at a 5% level. This effect is only not economic significant, because one standard deviation increase in non-hedge derivatives decrease firm risk with The strange thing about this result is that the use of non-hedge derivatives ensures for lower firm specific risk, although it is very small. While the expectation is that non-hedge derivatives are used for speculating and therefore increase risk. The second statistically significant variable is the commodity usage on all different measures of firm risk. The coefficient of commodity usage is and statistically significant at a 5% level for total risk. So it can be said that total risk decreases with the use of commodity derivatives. Systematic risk also decreases with the use of commodity derivatives, because of a coefficient of and statistically significant at a 10% level. Finally and most powerful result is commodities on idiosyncratic risk with a coefficient of and statistically significant at a 1% level. However the economic impact stays very low. Therefore there are some control variables significant like leverage, size, MTBV, liquidity, dividend yield and TLCF. Leverage is in all nine columns significant at a 1% level and has by one standard deviation increase an impact of on total risk. The variable size is almost every time statistically significant, but the coefficient goes to zero. The same applies to market to book value which is most of the time significant, but the economic effect is also almost zero. However, liquidity is the most of the time statistically significant at a 5% level and has a small economic but non zero effect on every risk measure. Dividend yield is only significant at a 1% level for the risk measures 31

36 total risk and idiosyncratic risk and has a large economic effect on both risk measures. Finally, tax loss carry forwards are significant at a 1% level for systematic risk. The R-squares lies between 0.25 and 0.338, which is pretty good, however it is almost all accomplished by the control variables and not the independent derivative usage variables. Finally, there are some regressions done, like in equation (6) in paragraph We made some interactions with almost all control variables multiplied by the derivative usage variables. However, none of the interaction coefficients are statistically significant in the regressions, and therefore these tables are left out of this thesis. There are also some robust regressions done as well to check, whether the outliers have some wrong impact on the dependent variables. The results are almost the same as in the normal regressions. The dependent variables have almost the same coefficients and t-stats, only the control variable MTBV is now more significant with a higher coefficient. This was also expected in paragraph 5.2 and table 7. Specific to this sample, the power of the independent variables in explaining variations in firm risk appears to be relatively poor. This is also the case in the Nguyen & Faff (2010) paper and the Hentschel & Kothari (2001) paper, where between 18% and 50% of volatility can be captured by only a few independent control variables. So, derivative usage in these studies seems to be both statistically and economically unimportant in explaining the cross-sectional variation in stock return variance, just like in this sample. A possible problem for the low t-stats and explanatory power of the variables could be reverse causality an example of endogeneity. The dependent variables could have a causal relationship with the risk measures, but it could also be the other way around, for example that the amount of derivatives outstanding is determined by total risk. This also applies for the control variables, and therefore all these variables are called endogenous variables, or determined within the system. Probably the problem is solved partly by doing the fixed effects regressions in paragraph

37 Pooled regression In this paragraph all firm years were pooled and regressed at the same time, without taking into account that there are three years of data for every firm. So there could be a correlation between some individual observations and the t-stats could be higher than they really are. The results can be seen in table 17 without control variables and table 18 with control variables. Again, all t-stats are made robust using the White (1980) method, since there is proved heterescedasticity by the Breusch-Pagan (1980) test. First, table 17 shows the results for the individual effects of independent derivative variables with respect to the risk measures. Total risk can be explained by three derivatives variables including all derivatives usage with a coefficient of , statistically significant at a 10% level. So a one standard deviation increase of derivatives usage ensures an increase of in total risk. This effect is very small and is therefore not economically significant. Hedge derivatives are statistically significant at a 5% level with a coefficient of , this sounds plausible because of a correlation of almost 1 between those two variables. Finally, the use of foreign currency derivatives is also significant at a 5% level with a coefficient of The use of derivatives and hedge derivatives explain for 0.8% the total risk, which is relatively low even just like foreign currency derivatives with an R-squared of None of the derivative variables have a significant relationship with systematic risk. Idiosyncratic risk can be explained only by foreign currency derivatives with a coefficient of at a significant level of 1% and an R-squared of Table 18 shows the relationships of derivatives with respect to the different risk measures. In contrast to table 17 the variable all derivatives is not significant anymore with respect to total risk. However, non-hedge derivatives are significant at a 10% level and a coefficient of This is the same strange results as in table 13 where non-hedge derivatives decrease the idiosyncratic risk, where the expectation is a risk increase, because of speculating. Again none of the derivative variables have a significant effect on systematic risk. Non-hedge derivatives have also a significant relationship to idiosyncratic risk with a coefficient of at a significant level of 1%. Commodity derivatives have also a significant relationship to idiosyncratic risk with a coefficient of at a significant level of 5%. This is the same result as what can be seen in table 13. For total risk all control variables except for the current ratio have a significant contribution to total risk with an R- squared of So these control variables explain 21.3% of the variation in total risk where total 33

38 derivative usage and hedge derivatives explain 0.8%. For systematic risk only leverage, liquidity and dividend yield are significant; however for idiosyncratic risk the same variables are significant as for total risk. The exact coefficients and significant levels can be found in table 18. In general some conclusions can be drawn from these pooled regressions. First, derivatives usage seems to be statistically unimportant in explaining the cross-sectional variation in stock return volatility as the addition of the variable only increases the explanatory power by 0.8%. This was also mentioned for the separated year regressions and in common with the two papers from Hentschel & Kothari (2001) and Nguyen & Faff (2010). Second, it seems that in the regressions without control variables the relationships between derivatives and risk is positive, however in the regressions with control variables the relationships are negative. So with OLS regressions it is difficult to say what the real relationship between derivatives and firm risk is. This is also due to the fact that some variables are significant without control variables, but insignificant with control variables. Third, it is clear that derivatives have no effect on systematic risk. This result can be found in almost all separated year regressions and in this pooled regression too. Nguyen & Faff (2010) said in their paper that it is not obvious from past literature whether corporate use of financial derivatives is aimed at hedging systematic risk or nonsystematic risk. Except for Smithson, Smith & Wilford (1995) who argue that it is a widely held belief that the use of financial derivatives targets hedging nonsystematic risk. Fourth, the most important problem is that there is probably an endogeneity problem and there is no control for firm fixed effects. Every firm has three observations and these could correlate very much, because of the small changes in the firm control variables over one year. The t-stats will therefore be higher than they really are. The reverse causality like in the separated year regressions also applies to these regressions. So, because of these two reasons it is difficult to draw any conclusion out of these pooled regressions. Finally, according to Nguyen & Faff (2010) idiosyncratic risk accounts for a large proportion of total risk, so these two measures of risk are expected to be highly correlated. Therefore, this is the reason why almost the same variables with the same relation are significant for total risk and idiosyncratic risk. 34

39 5.5 Fixed-effects model In this paragraph the results of the fixed effects regressions are presented. In these regressions the data is also pooled, but now firm fixed effects are taken into account. These effects control for unobserved heterogeneity and take into account that there are three observations per firm, which possibly correlate with each other. The standard errors are corrected by heteroscedasticity using the White (1980) method. Table 19 shows the results of the fixed effects regressions without control variables. The amount of derivative variables that explain total risk is now four. The variable total derivative usage has a coefficient of at a significant level of 5%. So the effect is now much larger than shown at the OLS regressions. Hedge derivatives are also significant at a 5% level with a coefficient of Foreign currency derivatives and interest rate derivatives are both significant at a 1% level with respectively a coefficient of and Not only the marginal effects are larger, also the R- squares are much larger than 0.008, namely around for the variables total derivatives and hedge derivatives and even for FCDs and IRDs. However, the economic results stay small for example one standard deviation increase in all derivative usage increase total risk with This is a volatility increase of 1.5%. For systematic risk the results are almost the same as for the pooled regression, therefore zero significant variables that explain systematic risk. For idiosyncratic risk the same derivative variables are significant as for total risk. Table 20 shows the results of the fixed regressions with control variables. The same derivative variables are significant as in table 18, but the non-hedge derivatives variable can be included. For total risk the total derivatives outstanding has a positive effect. The coefficient is and significant at a 5% level. Hedge derivatives are also significant at a 1% level with a coefficient of and non-hedge derivatives are significant at a 1% level with a coefficient of In contrast to table 18, the non-hedge derivatives have a positive effect on firm risk. Foreign currency derivatives and interest rate derivatives have a coefficient of respectively and at a significant level of 1% and have therefore a positive effect on total risk. All kind of derivatives have a positive effect on total risk. None of the derivatives variables have an effect on systematic risk like the past tables have shown. For idiosyncratic risk the same five derivatives variables are significant as for total risk. So, in general derivative usage increases total risk and therefore idiosyncratic risk, 35

40 because idiosyncratic risk is a large part of total risk. The control variables that have a significant relationship are leverage, size, MTBV, liquidity, current ratio and tax loss carry forwards. 5.6 Derivative product regressions What are the best derivative products to achieve risk reduction? That is the main question of this paragraph and the results of these regressions are shown in table 21 and 22. Where in table 21 only the different derivative products are regressed on the different measures of risk; and in table 22 the control variables are added. The first three columns of both tables present the OLS pooled regressions and the last three columns the fixed effect regressions. Beside, column (1) and (4) are results for total risk, (2) and (5) for systematic risk and (3) and (6) for idiosyncratic risk. The standard errors are corrected by heteroscedasticity using the White (1980) method. First, the results of table 21 where can be seen that the use of forward contracts results in a total risk increase. Both the OLS regressions and fixed effects regressions show that total risk and idiosyncratic risk increases. Systematic risk increases by forward contracts only for the OLS regressions. Swap contracts have a significant positive relationship with total risk and idiosyncratic risk in the firm fixed effects regressions. Option contracts do not have any significant relationships. Futures contracts have significant relationships with all risk measures in the fixed effects regressions. Where total risk and idiosyncratic risk increase by the use of futures contracts at a significant level of 1% and systematic risk decreases at a significant level of 1%. Unspecified derivative contracts in the annual reports tend to increase total risk and idiosyncratic risk, where the effects are larger for the fixed effects regressions than for the OLS regressions. So it looks like that only options do not have any significant relationship to the risk measures and that all other derivative products only have effects on total and idiosyncratic risk. Second, in table 22 the results are shown with control variables and there can be seen that the effects are almost the same as the regressions without control variables. Forwards tend to have a positive relationship with all risk measures in the OLS regressions and only for total risk in the fixed effects regressions. Swaps only have a significant effect on total risk and idiosyncratic risk in the fixed effects regressions. Options again do not have any effect on risk and futures contracts tend to have a negative relationship on total- and idiosyncratic risk. However, these results are no longer significant when controlled for firm fixed effects. Nevertheless, futures decrease systematic risk when controlling for firm fixed effects, because one standard deviation increase in futures usage 36

41 decrease beta with The unspecified derivative contracts increase total and idiosyncratic risk when controlled for firm fixed effects. This figure is statistically significant at a 1% level and one standard deviation increase in unspecified derivative contracts results in a total risk increase of Therefore the economic effect is not that large. Again, higher leverage and higher liquidity results in more risk and in addition more tax loss carry forwards increase risk too. The negative effects of size and market to book value on risk measures are again very small but statistically significant. In short, most derivative products in general increase firm risk, except options which do not have any effect and futures tend to have a negative effect, but this is not very significant. However, the effects are not largely economic significant, so the explanatory power of derivative usage on firm risk stays low. 5.7 Interaction regressions In the past paragraphs is shown that there is a small positive relationship between derivatives usage and total- and idiosyncratic risk. However, the R-squared stays relatively low and probably the relationship between the use of derivatives and firm risk, measured by the daily stock return variance is not linear. The paragraph 4.3 portfolio analysis suggests that there might be a chance that the relationship is not linear, because firm risk is not increasing or decreasing as derivative use increases. This can also be seen in figure 2 where it looks like a parabola. Unfortunately, there are many reasons why this relationship could not be linear, but this is difficult to estimate. Though different papers indicate, for example Guay (1999) and Nguyen & Faff (2010), that one of the reasons leading to insignificant coefficients is the nonlinearity between firm risk and the explanatory variables. Therefore, this paragraph aims at testing for a nonlinear relationship between firm risk and derivative usage. The test is explained in paragraph and the corresponding regression is presented in equation (7). The only thing left to be determined are the thresholds of the dummy variables. There are two thresholds chosen from table 10, because in two deciles there is a large risk reduction, because of the usage of derivatives. The average notional amounts chosen are from decile 2 and decile 8 where the risk reduction 17% and almost 40% is. These notional amounts are 1.53% and 13.35%, respectively, and therefore the thresholds for the dummies. Compared to Nguyen & Faff (2010) and Tufano (1996), this amount is much lower than 20% of and 40%, respectively. The standard-errors are corrected for heteroscedasticity using the White (1980) method, proved by the Breusch-Pagan test. The regression results of equation (7) are presented in table 23 and provide supporting evidence for a nonlinear relationship, at least for the 1.53% 37

42 threshold. The coefficient of derivatives usage is and statistically significant at a 5% level for low derivative users. This result suggests that for a total notional amount oustanding scaled by firm size of less than 1.53%, corporate use of financial derivatives is associated with a reduction in firm risk and could have the motive of hedging. For high derivative users the opposite is happening, because the coefficient is and statistically significant at a 5% level. This result suggests that for a notional amount scaled by firm size of more than 1.53%, corporate use of financial derivatives is associated with an increase in firm risk and could have the motive of speculating. When nonderivative users are excluded the effect is even larger for both low- and high derivative users. The results for the threshold of 13.35% are not significant at all. Therefore the conclusions are all about the threshold of 1.53%. First, there is evidence that a risk reduction with derivatives is possible; however this is determined by the notional amount outstanding. Second, it looks like there is a significant relationship between the use of derivative and firm risk. Third, the R-squared is lower than for the fixed effects regressions, but higher than the pooled OLS regression. So for this sample it appears that this is not the best explanation for the variation in stock return variance. Finally, it is important to mention that firm years are separated, so it is possible that one firm year is a low derivative user and the other firm year is a high derivative user. There is no correction made for firm fixed effects, what could solve the problem, however by doing this regression controlled for firm fixed effects, there are no significant results. The final conclusions are about the relationship between firm risk and the control variables. Stock return variance is higher for firms that have more leverage and are more liquid. Smaller firms tend to have more risk, however this effects is very small for this sample. These results are consistent with the results for fixed effects regressions and the OLS regressions. The same regression is done with the distinction in hedge derivatives and non-hedge derivatives. This regression gives the result shown in table 24 and provides supporting evidence for a difference in hedging and speculating. For derivative users with an amount outstanding of less than 1.53% hedge derivatives lowers total risk and non-hedge derivatives increase total risk. The coefficients of hedge and non-hedge derivatives are respectively and , statistically significant at a 5% and a 1% level. These variables are also economically significant because one standard deviation increase in hedge derivatives results in a total risk decrease of For high derivative users hedge derivatives increase risk and non-hedge derivatives decrease risk with coefficients of and significant at a 5% and 1% level. This suggests that when firms have more than 1.53% in hedge derivatives outstanding scaled by total firm size, there is a risk increase and it is not hedging, 38

43 but speculating. Non-hedge derivatives for speculating increase risk for small amounts, but when the amounts become larger the risk decrease. When nonusers are excluded the non-hedge derivatives variable is not significant anymore, but the hedge derivatives variable is even more significant. For the 13.35% threshold these independent variables are not significant anymore. The last test is on which underlying the largest risk reduction can be made? The results are shown in table 25 and suggest that for low derivative users with a threshold of 1.53%, on interest rate risk the largest risk reduction is achieved. The coefficient for interest rate risk derivatives is and is statistically significant at a 1% level. The total risk increase is due to the use of commodity derivatives with a coefficient of and significant at a 1% level. Derivatives on foreign currency derivatives tend to decrease total risk; however this variable is not significant. For high derivative users things turnaround, so now one standard deviation increase in commodity derivatives decrease risk with of , significant at a 1% level and one standard deviation increase in interest rate derivatives increase rick with 0,00518, significant at a 1% level. This economic effect is much larger than for the OLS and fixed effects regressions. Foreign currency derivatives tend to increase risk, but this variable is again not significant. For the threshold 13.35%, when nonusers are excluded and we speak about low derivative users foreign currency derivatives increase total risk with a coefficient of significant at a 10% level. The other variables are not significant, so this threshold seems also be too large to separate low derivative users from large derivative users, just like in the other tables. In short, when a threshold of 1.53% is taken into account the risk reduction by derivative usage becomes visible. Hedge derivatives tend to decrease risk for low users and increase risk when this threshold is passed. Non-hedge derivatives are used for speculating and therefore increase risk, when used for small amounts, but decrease risk when the threshold is passed. So, to use the derivatives where they meant for, do not take very large positions in derivatives. 39

44 6. Conclusion The main goal of this thesis is to determine whether there are effects of hedging and speculating with financial derivatives on firm risk, where firm risk is measured by the daily stock return variance. The sample used for this thesis contains the largest manufacturing firms from the S&P500 for the years 2007 to There are many theories why and when firms should use derivatives, however empirical studies do not give a clear answer of the effects of those financial derivatives on a reduction or increase in firm risk. The expectation is that firms reduce firm risk with hedge derivatives and increase risk by non-hedge derivatives, because of speculating. The portfolio analysis shows the first global impression, that the relationship between derivative usage and total risk is not linear and that there is an optimal amount of 1.53% scaled by firm size. The ordinary least square regressions show that there is no association between risk reduction and derivative usage, for the separated years. The only significant result is a total risk increase by foreign currency derivatives usage for the year However, when the control variables are added this effect is negligible. The use of commodities results on the other hand in a total risk reduction for For the pooled OLS regression without control variables the use of all derivatives, hedge derivatives and foreign currency derivatives ensures a total risk increase. However, by the addition of the control variables all significant effects are disappeared and only non-hedge derivatives result in a total risk reduction. Therefore, the results are not consistent for total risk and this could be caused by the fact that the relationship is nonlinear. The effects on idiosyncratic risk corresponds most of the time with the effects on total risk due to the fact that idiosyncratic risk is a large part of total risk. For systematic risk the effects are not significant and it seems that derivative usage do not have any impact on systematic risk. One possible explanation for the insignificant results by the OLS regressions could be an endogeneity problem, namely reverse causality. All variables are endogenous and could affect total risk, but total risk could also affect the variables. This results in low standard errors and few significant coefficients. To control for this problem, firm fixed effects could rather help. 40

45 When controlling for firm fixed effects the use of derivatives, hedge derivatives, foreign currency derivatives and interest rate derivatives results in a total risk increase. Even with control variables, all kinds of derivative usage results in a risk increase, however the effects are just a little smaller with respect to the regressions without control variables. Consequently, there is no question of risk reduction or hedging by derivative usage. When looking at the different derivative products, there can be said, that the use of forwards, swaps and the unspecified reported contracts have a positive relationship to firm risk. The effect of options is negligible and the effects of futures contracts are not consistent for all regressions, therefore it is difficult to draw any conclusion about this product. The last analysis is to test whether there is a nonlinear relationship between derivative usage and total firm risk. The results from these tests are that firms with a derivative usage of 1.53% or less (low derivative users), leads to a risk reduction and firms with a higher amount outstanding (high derivative users) experience a risk increase. Beside, hedge derivatives used by low derivative users lead to a risk reduction and non-hedge derivatives lead to a risk increase. For high derivative users it is exactly the other way around. This is consistent with the expectations of hedging and speculating with hedge and non-hedge derivatives. The largest risk reduction created by low derivative users is due to interest rate derivatives. For commodity usage it is the other way around, because of the negative relationship with total risk for high derivative users. So the optimal amount of commodities outstanding is probably higher than the 1.53% threshold. In general there is a positive relationship between derivative usage and total risk. However, when looking for a nonlinear relationship, firms with derivatives outstanding of 1.53% or less experience a negative relationship and probably a risk reduction. Hedging and speculating with derivatives is possible for large manufacturing firms, but only when the outstanding amounts are 1.53% or less, because otherwise it overshoots the mark. 41

46 7. References Allayannis, G., & Ofek, E. (2001). Exchange rate exposure, hedging, and the use of foreign currency derivatives. Journal of International Money and Finance, 20(2), Barclay, M. J., & Smith, C. W. (1995). The Maturity Structure of Corporate Debt. The Journal of Finance, 50(2), Bartram, S. M., Brown, G. W., & Fehle, F. R. (2009). International evidence on financial derivatives usage. Financial Management, 38(1), Berk, J., & Demarzo, P. (2011). Corporate Finance (Global edi., p. 1104). Pearson Education Limited. Black, F., & Scholes, M. (1973). The Pricing of Options and Corporate Liabilities. Journal of Political Economy, 81(3), Bodnar, G. M., Hayt, G. S., & Marston, R. C. (1995). Wharton Survey of Derivatives Usage by US Non- Financial Firms. Financial Management, Summer 24(2), Breusch, T. S., & Pagan, A. R. (1980). Lagrange Multiplier Test and to Model Applications Specification in Econometrics. The Review of Economic Studies1, 47(1), Brüderl, J. (2005). Panel data analysis. University of Mannheim. Retrieved from Capelle-Blancard, G. (2010). Are Derivatives Dangerous? a literature Survey, 24(November), Culp, C., & Miller, M. H. (1995). Metallgesellchaft and the Economics of Synthetic Storage. Journal of Applied Corporate Finance, 7, Froot, K. a, Scharfstein, D. S., & Stein, J. C. (1993). Risk Management: Coordinating Corporate Investment and Financing Policies. Journal of Finance, 48(5), Gay, G. D., & Nam, J. (1998). The Underinvestment Problem and Corporate Derivatives Use. Financial Management, 27(4 Winter), Gilbert. (2001). Has the Ashanti Goldfields Loss Discredited Collar Hedges?. Working Paper, Vrije Universiteit. Guay, W. R. (1999). The impact of derivatives on firm risk: An empirical examination of new derivative users. Journal of Accounting and Economics, 26(1-3), Hagelin, N., & Pramborg, B. (2004). Hedging Foreign Exchange Exposure: Risk Reduction from Transaction and Translation Hedging. Journal of International Financial Management and Accounting, 15(1),

47 Hentschel, L., & Kothari, S. P. (2001). Are Corporations Reducing or Taking Risks with Derivatives? Journal of Financial and Quantitative Analysis, 36(1), Hsiao, C. (2003). Analysis of Panel Data (Second.). The press syndicate of the University of Cambridge. Jensen, M. C., & Meckling, W. H. (1976). Theory of the firm: Managerial behavior, agency costs and ownership structure. (M. C. Jensen, Ed.)Journal of Financial Economics, 3(4), Jong, A. De, Ligterink, J., & Macrae, V. (2006). A Firm-Specific Analysis of the Exchange-Rate Exposure of Dutch Firms. Journal of International Management and Accounting, 17(1). Koski, J. L., & Pontiff, J. (1996). Evidence from the Mutual Fund. Wharton Financial Institutions Center. Mayers, D., & Smith, C. W. (1982). On the Corporate Demand for Insurance. The Journal of Business, 55(2 April), Mayers, D., & Smith, C. W. (1987). Corporate Insurance and the Underinvestment Problem. The Journal of Risk and Insurance, 54(1), McClave, Benson, & Sincich. (2005). Statistics for Business and Economics (9th ed.). Merton, C. (1974). On the Pricing of Corporate Debt: The Risk Structure on Interest Rates. The Journal of Finance, 29(2), Mian, S. L. (1996). Evidence on Corporate Hedging Policy. The Journal of Financial and Quantitative Analysis, 31(3 september), Miller, M. H., & Ross, D. (1997). The Orange County Bankruptcy and its Aftermath: Some New Evidence. Journal of Derivatives, 4(4), Modigliani, F., & Miller, M. H. (1958). The Cost of Capital, Coporation Finance and the Theory of Investment. The American Economic Review, 48(3), Myers, S. (1977). Determinants of Corporate Hedging. Journal of Financial Economics, 5, Nance, D. R., Smith, C. W., & Smithson, C. W. (1993). On the Determinants of Corporate Hedging. The Journal of Finance, 48(1), Nguyen, H., & Faff, R. (2010). Are firms hedging or speculating? The relationship between financial derivatives and firm risk. Applied Financial Economics, 20(10), Reagan, P. B., & Stulz, R. M. (1983). Risk Sharing, Labor contracts and capital markets. Unpublished Manuscripts, (University of Rechester). Schrand, C., & Unal, H. (1998). Hedging and Coordinated Risk Management : Evidence from Thrift Conversions, 53(3),

48 Shleifer, A., & Vishny, R. W. (1989). Management Entrenchment The Case of Manager-Specific Investments. Journal of Financial Economics, 25, Smith, C. W., & Stulz, R. M. (1985). The Determinants of Firms Hedging Policies. The Journal of Financial and Quantitative Analysis, 20(4), Smith, W., & Watts, R. L. (1992). The investment opportunity set and corporate financing, dividend, and compensation policies. Journal of Financial Economics, 32(1991), Smithson, C. W., Smith, C. W., & Wilford, D. S. (1995). Managing Financial Risk: A Guide to Derivative Products, Financial Engineering and Value Maximization. Irwin Professional Publishin, Chicago. Stulz, R. M. (1996). Rethinking Risk Management. Journal of Applied Corporate Finance, 9(3), Tufano, P. (1996). Who Manages Risk? An Empirical Examination of Risk Management Practices in the Gold Mining Industry. The Journal of Finance, 51(4), Verleger, P. (1999). Was Metallgesellchaft s Use of Petroleum Futures Part of a Rational Corporate Strategy?. Journal of Energy Finance and Development, 4, Warner, J. B. (1977). Bankruptcy Costs: Some Evidence. The Journal of Finance, 32(2), White, H. (1980). A Heteroskedasticity Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity. Econometrica, 48(4),

49 8. Appendix 45

50 Notional Amount Derivative Holdings for Manufacturing firms in the S&P 500 for the years (in Million $) Derivatives Designated for Hedging Derivatives Not Designated for Hedging Mean Median Std. Min Max Firm Years Notional Amount Mean Median Std. Min Max Firm Years Commodity Commodity Forwards Forwards Swaps Swaps Options Options Futures Futures Forward & Options Forward & Options Swaps & Options Swaps & Options Commodity contracts Commodity contracts Total Total Interest Rate Interest Rate Forwards Forwards Swaps Swaps Options Options Futures Futures Interest Rate contracts Interest Rate contracts Total Total Foreign Currency Foreign Currency Forwards Forwards Swaps Swaps Options Options Futures Futures Forwards & Swaps Forwards & Swaps Forwards & Options Forwards & Options Foreign Currency contracts Foreign Currency contracts Total Total Grand Total Grand Total Table 1 shows the notional amount, mean, median, standard deviation (std.), min, max in each derivative category across firm years. The notional amounts are measured in millions of dollars. There is made a distinction between derivatives designated for any hedging purpose and derivatives that are not designated for hedging purposes. The last subset of each derivative, for example commodity contracts means that the annual report made no distinction which class is used between commodities. This is the same for interest rate and foreign currency contracts. The sample consists of 522 firm years from 2007 to These firms are selected from the S&P 500 and are all manufacturing firms according to the two digits SIC code between 20 and 39.

51 Derivatives outstanding for Manufacturing firms in the S&P 500 per Year (in Million $) Total Derivatives outstanding Total Notional Amount Mean Median Standard deviation Firms Total Derivatives outstanding for Hedging Total Notional Amount Mean Median Standard deviation Firms Total Derivatives outstanding for non Hedging Total Notional Amount Mean Median Standard deviation Firms Table 2 shows the total notional amounts of derivatives outstanding for each year and besides the mean, median, standard deviation and firm numbers. There is also made a distinction between the derivatives outstanding for hedging and for non-hedging purposes. The main result is that the use of derivatives for whatever reason is growing. Derivative Activity by Instrument (In Million $) Forwards Swaps Options Futures Derivative contracts Amount Mean Median Std Min Max Table 3 shows the descriptive statistics for derivative activity by instrument across underlying products. The products most commonly reported are forwards, swaps and derivative contracts in general. For derivative contracts it is not clear which instrument or combination of instruments is used.

52 Total Notional Amount per Year for all Manufacturing Firms in the S&P 500 (in Million $) Commodity Forwards Swaps Options Futures Futures & Options Forward & Options Swaps & Options Commodity contracts Total Interest Rate Forwards Swaps Options Futures Swaptions Interest Rate contracts Total Foreign Currency Forwards Swaps Options Futures Forwards & Swaps Forwards & Options Foreign Currency contracts Total Table 4 shows the total notional amounts per year, underlying and derivative type. Some derivative types are grouped together, because these are as pairs reported in annual statements. If there was no specific contract reported, it is categorized in the overall group contracts. For commodities the most common derivative contracts are forwards and swaps. For hedging interest rates, the most common derivatives are swaps. For foreign currency the most used derivatives are all of them except futures.

53 Summary Statistics for Derivative Usage Mean Std. Dev. All derivatives 6.71% 14.2% Hedgederivatives 6.12% 13.6% Nonhedge derivatives 0.61% 2.6% Commodity derivatives 0.19% 1.1% Foreign Currency derivatives 3.95% 6.8% Interest rate derivatives 2.59% 12.1% Forwards 1.74% 4.4% Swaps 2.29% 12.0% Options 0.06% 0.4% Futures 0.01% 0.1% Unspecified contracts 2.35% 6.3% Table 5 shows the average notional amounts of different derivative usage variables in percentages of the total size of the firm and the corresponding standard deviations. All derivatives 1 All derivatives Hedge Nonhedge Com FCD IRD Forwards Swaps Options Futures Contracts Hedge derivatives Nonhedge derivatives commodity Correlation Matrix of the Derivative Usage Variables FCD IRD Forwards Swaps Options Futures Contracts Table 6 shows the correlations of the derivative usage variables between each other. Where the all derivatives variable is the total amount, commodity is commodity derivatives, FCD is foreign currency derivatives, IRD is interest rate derivatives and contracts means unspecified contract in the annual report.

54 Descriptive statistics for the dependent- and control variables of the whole sample Stats Risk Systematic Idiosync Lev Size MTBV Liq CR DY TLCF Exes N Mean Median Sd Min Max Table shows the descriptive statistics of the dependent variables and control variables of all firm years Risk, systematic, idiosync, lev, size, MTBV, liq, cr, dy, TLCF and exes stands respectively for total risk, systematic risk, idiosyncratic risk, leverage, size, market to book value, liquidity, current ratio, dividend yield, tax loss carryforwards and executive share percentage. N are the total observations where 522 firm years the most is, so only for the variable current ratio there are fewer observations. Sd stands for standard deviation of the variable. Univariate Tests of determinants of Derivatives Usage Derivative Users Nonderivative Users N Mean Median Sd N Mean Median Sd Wilcoxon p-value Risk Systematic Idiosync Lev Size MTBV Liq CR DY TLCF Exes Table 8 reports the mean, median and standard deviation for the group derivative users (N=338) and nonderivative users (N=184). The last column presents p-values of Wilcoxon rank sum tests for differences between derivative users and non-derivative users.

55 Leverage 1 Leverage Size MTBV Liquidity Current ratio Dividend Yield TLCF Executive stock Size MTBV Correlation Matrix of the Control Variables Liquidity Current ratio Dividend Yield TLCF Executive stock Table 9 shows the correlations between the control variables. The exact meaning of the control variables can be found in paragraph 4.3 control variables. The most control variables have very low or even negative correlations with each other. Portfolio Analysis between Risk and Derivatives Usage Deciles Total Risk Mean Risk Change Derivatives outstanding Change % 0.44% 0.44% % 1.53% 1.09% % 2.67% 1.14% % 4.09% 1.42% % 5.41% 1.32% % 7.23% 1.82% % 9.68% 2.46% % 13.35% 3.67% % 18.56% 5.21% % 41.41% 22.85% Table 10 shows the portfolios based on derivatives outstanding where portfolio 0 contains 184 non-users and portfolio 1 (34 users like all other portfolios) the least intensives users to portfolio 10 the most intensive users. The largest risk reduction is between portfolios 7 and 8 with a risk reduction of almost 40% with respect to portfolio 7. The lowest risks can be found in portfolio 2 and 8 with respectively 1.53% and 13.35% derivatives outstanding scaled to firm size.

56 Firm Risk and the Use of Derivatives Total Risk 2007 (1) (2) (3) (4) (5) (6) (7) (8) Constant *** *** *** *** *** *** *** *** (13.58) (13.64) (14.56) (14.51) (12.98) (14.42) (12.67) (13.54) All derivatives (0.45) Hedge (0.52) (0.54) Nonhedge (-0.49) (-0.51) Commodity (0.29) (0.18) FCD (0.99) (0.96) IRD (0.01) (-0.00) R-squared Systematic Risk 2007 (1) (2) (3) (4) (5) (6) (7) (8) Constant *** *** *** *** *** *** *** *** (39.87) (40.10) (41.88) (42.14) (38.88) (41.82) (37.98) (39.51) All derivatives (-0.19) Hedge (-0.25) (-0.27) Nonhedge (0.41) (0.42) Commodity (0.29) (0.34) FCD (-0.4) (-0.43) IRD (-0.04) (-0.04) R-squared Idiosyncratic Risk 2007 (1) (2) (3) (4) (5) (6) (7) (8) Constant *** *** *** *** *** *** *** *** (11.47) (11.51) (12.34) (12.30) (10.84) (12.22) (10.61) (11.45) All derivatives (0.44) Hedge (0.51) (0.53) Nonhedge (-0.54) (-0.57) Commodity (0.09) (-0.04) FCD (1.21) (1.20) IRD (-0.11) (-0.11) R-squared Table 11 shows the effect of derivative usage on total risk, systematic risk and idiosyncratic risk for the year Columns (1) to (6) shows the univariate results of derivative usage variables and risk. Column (7) shows the results of a common effect of commodity, foreign currency (FCD) and interest rate (IRD) derivatives. Column (8) shows the results of a common effect of hedge and non-hedge derivatives. The figures in the brackets are t-stats not adjusted for heteroscedasticity. *, **, *** denote significance of 10, 5, 1%, respectively.

57 Firm Risk and the Use of Derivatives Total Risk 2008 (1) (2) (3) (4) (5) (6) (7) (8) Constant *** *** *** *** *** *** *** *** (16.20) (16.30) (17.41) (17.96) (14.68) (17.76) (14.38) (16.02) All derivatives (1.10) Hedge (1.05) (0.88) Nonhedge (0.80) (0.58) Commodity (-0.48) (-0.54) FCD ** ** (2.40) (2.40) IRD (0.12) (0.03) R-squared Systematic Risk 2008 (1) (2) (3) (4) (5) (6) (7) (8) Constant *** *** *** *** *** *** *** *** (35.39) (35.58) (37.74) (38.48) (33.08) (38.25) (32.17) (35.06) All derivatives (1.16) Hedge (1.12) (0.96) Nonhedge (0.80) (0.56) Commodity (0.03) (-0.01) FCD * * (1.90) (1.87) IRD (0.39) (0.32) R-squared Idiosyncratic Risk 2008 (1) (2) (3) (4) (5) (6) (7) (8) Constant *** *** *** *** *** *** *** *** (14.13) (14.22) (15.13) (15.67) (12.60) (15.52) (12.42) (13.95) All derivatives (0.84) Hedge (0.78) (0.60) Nonhedge (0.85) (0.68) Commodity (-0.57) (-0.65) FCD ** ** (2.47) (2.48) IRD (-0.19) (-0.29) R-squared Table 12 shows the effect of derivative usage on total risk, systematic risk and idiosyncratic risk for the year Columns (1) to (6) shows the univariate results of derivative usage variables and risk. Column (7) shows the results of a common effect of commodity, foreign currency (FCD) and interest rate (IRD) derivatives. Column (8) shows the results of a common effect of hedge and non-hedge derivatives. The figures in the brackets are t-stats not adjusted for heteroscedasticity. *, **, *** denote significance of 10, 5, 1%, respectively.

58 Firm Risk and the Use of Derivatives Total Risk 2009 (1) (2) (3) (4) (5) (6) (7) (8) Constant *** *** *** *** *** *** *** *** (11.01) (10.57) (14.06) (14.51) (12.16) (12.91) (11.17) (10.54) All derivatives (0.55) Hedge (0.64) (0.69) Nonhedge (-0.08) (-0.27) Commodity (-0.52) (-0.55) FCD (-0.53) (-0.66) IRD (1.53) (1.58) R-squared Systematic Risk 2009 (1) (2) (3) (4) (5) (6) (7) (8) Constant *** *** *** *** *** *** *** *** (21.70) (21.05) (27.06) (27.88) (22.81) (25.77) (21.48) (20.98) All derivatives (0.29) Hedge (0.18) (0.18) Nonhedge (0.02) (-0.03) Commodity (-0.47) (-0.48) FCD (0.03) (0.00) IRD (0.31) (0.32) R-squared Idiosyncratic Risk 2009 (1) (2) (3) (4) (5) (6) (7) (8) Constant *** *** *** *** *** *** *** *** (10.98) (10.63) (13.73) (14.21) (11.87) (12.79) (11.05) (10.60) All derivatives (0.15) Hedge (0.14) (0.16) Nonhedge (-0.07) (-0.11) Commodity (-0.69) (-0.70) FCD (-0.51) (-0.58) IRD (0.89) (0.94) R-squared Table 13 shows the effect of derivative usage on total risk, systematic risk and idiosyncratic risk for the year Columns (1) to (6) shows the univariate results of derivative usage variables and risk. Column (7) shows the results of a common effect of commodity, foreign currency (FCD) and interest rate (IRD) derivatives. Column (8) shows the results of a common effect of hedge and non-hedge derivatives. The figures in the brackets are t-stats not adjusted for heteroscedasticity. *, **, *** denote significance of 10, 5, 1%, respectively.

59 Determinants of Firm Risk and the Use of Financial Derivatives Total Risk Systematic Risk Idiosyncratic Risk 2007 (1) (2) (3) (4) (5) (6) (7) (8) (9) Constant *** *** *** *** *** *** *** *** *** (5.74) (5.76) (5.56) (14.34) (14.23) (14.10) (5.10) (5.13) (4.90) All derivatives 1.26E E-06 (0.10) (-0.50) (-0.02) Hedge 5.64E E-05 (0.41) (-0.58) (0.29) Nonhedge (-1.62) (0.46) (-1.63) Commodity (0.08) (0.98) (-0.32) FCD (0.35) (-0.60) (0.49) IRD -2.95E E-05 (-0.64) (-0.42) (-1.22) Leverage -8.55E E E E E E-06 (-0.05) (0.05) (-0.10) (-0.72) (-0.73) (-0.73) (0.03) (0.14) (-0.01) Size ** ** ** -1.90E E E *** *** ** (-2.31) (-2.32) (-2.29) (-0.41) (-0.40) (-0.40) (-2.64) (-2.65) (-2.60) MTBV -7.80E E E E E E-07 (-0.35) (-0.42) (-0.23) (-1.57) (-1.56) (-1.52) (0.16) (0.08) (0.28) Liquidity 4.45E E E E (0.12) (0.06) (0.09) (0.91) (0.92) (0.93) (-0.43) (-0.52) (-0.48) Current ratio 1.42E E E E E E-06 (0.65) (0.62) (0.70) (1.11) (1.12) (1.06) (0.30) (0.26) (0.38) Dividend yield *** *** *** ** ** ** *** *** *** (-5.47) (-5.56) (-5.34) (-2.52) (-2.49) (-2.46) (-5.50) (-5.61) (-5.37) TLCF *** *** *** *** *** *** (3.33) (3.35) (3.19) (0.22) (0.22) (0.23) (4.43) (4.45) (4.18) Executive stock ** ** ** ** ** * (2.24) (2.39) (2.17) (0.66) (0.63) (0.67) (2.00) (2.13) (1.93) R-squared Table 14 shows the effect of derivative usage on total risk, systematic risk and idiosyncratic risk controlled by the variables leverage, size, MTBV (market-to-book value), liquidity, current ratio, dividend yield, tax loss carry forwards and executive stock percentage, for the year The explanations of these control variables can be found in paragraph 4.3 control variables. Columns (1) to (3) show that relationship between all derivatives, hedge and non-hedge derivatives and the three underlying s commodity, foreign currency (FCD) and interest rate (IRD) derivatives on total risk. Columns (4) to (6) show the relationship for systematic risk and columns (7) to (9) for idiosyncratic risk. The figures in the brackets are t-stats adjusted for heteroscedasticity using the White (1980) method. *, **, *** denote significance of 10, 5, 1%, respectively.

60 Determinants of Firm Risk and the Use of Financial Derivatives Total Risk Systematic Risk Idiosyncratic Risk 2008 (1) (2) (3) (4) (5) (6) (7) (8) (9) Constant *** *** *** *** *** *** *** *** *** (3.76) (3.75) (3.59) (11.29) (11.25) (10.92) (3.61) (3.60) (3.37) All derivatives E-05 (0.56) (0.77) (0.36) Hedge E-05 (0.67) (0.66) (0.58) Nonhedge (-0.49) (0.25) (-0.82) Commodity (-1.32) (-0.19) (-1.50) FCD (0.81) (0.74) (0.93) IRD -2.86E E-05 (-0.19) (0.38) (-0.85) Leverage *** *** *** *** *** *** *** *** *** (3.61) (3.62) (3.48) (3.05) (3.03) (2.92) (3.76) (3.79) (3.64) Size E E E * * * (-1.09) (-1.09) (-1.08) (-0.09) (-0.09) (-0.08) (-1.77) (-1.77) (-1.74) MTBV 2.57E E E E E E-07 (1.56) (1.59) (1.64) (1.60) (1.59) (1.62) (0.89) (0.96) (0.98) Liquidity ** ** ** *** *** ** ** ** ** (2.51) (2.51) (2.32) (2.74) (2.73) (2.50) (2.59) (2.58) (2.46) Current ratio 1.85E E E E E E-06 (0.41) (0.40) (0.53) (0.71) (0.71) (0.82) (-0.21) (-0.22) (-0.09) Dividend yield ** ** ** ** ** ** (-2.46) (-2.44) (-2.23) (-1.39) (-1.40) (-1.24) (-2.49) (-2.47) (-2.29) TLCF *** *** *** * * * *** *** *** (3.76) (3.73) (3.64) (1.81) (1.80) (1.77) (3.39) (3.36) (3.22) Executive stock ** ** * (1.64) (1.62) (1.44) (0.74) (0.74) (0.62) (2.01) (1.99) (1.79) R-squared Table 15 shows the effect of derivative usage on total risk, systematic risk and idiosyncratic risk controlled by the variables leverage, size, MTBV (market-to-book value), liquidity, current ratio, dividend yield, tax loss carry forwards and executive stock percentage, for the year The explanations of these control variables can be found in paragraph 4.3 control variables. Columns (1) to (3) show that relationship between all derivatives, hedge and non-hedge derivatives and the three underlying s commodity, foreign currency (FCD) and interest rate (IRD) derivatives on total risk. Columns (4) to (6) show the relationship for systematic risk and columns (7) to (9) for idiosyncratic risk. The figures in the brackets are t-stats adjusted for heteroscedasticity using the White (1980) method. *, **, *** denote significance of 10, 5, 1%, respectively.

61 Determinants of Firm Risk and the Use of Financial Derivatives Total Risk Systematic Risk Idiosyncratic Risk 2009 (1) (2) (3) (4) (5) (6) (7) (8) (9) Constant *** *** *** *** *** *** *** *** *** (3.15) (2.72) (3.17) (6.66) (6.43) (6.52) (3.72) (3.32) (3.69) All derivatives (-1.04) (-0.93) (-1.48) Hedge (-0.34) (-0.66) (-0.70) Nonhedge ** (-1.55) (-0.67) (-2.09) Commodity ** * *** (-2.46) (-1.90) (-3.53) FCD (-0.79) (0.04) (-1.10) IRD (-0.46) (-1.04) (-0.77) Leverage *** *** *** *** *** *** *** *** *** (4.61) (4.57) (5.14) (3.56) (3.54) (4.52) (4.90) (4.89) (5.08) Size * * *** ** ** * * (-1.84) (-1.60) (-1.82) (-2.63) (-2.54) (-2.46) (-1.75) (-1.53) (-1.75) MTBV *** *** *** ** ** ** * * * (-2.66) (-2.71) (-2.66) (-2.45) (-2.46) (-2.25) (-1.74) (-1.76) (-1.80) Liquidity ** ** ** ** ** ** *** *** ** (2.36) (2.45) (2.20) (2.42) (2.46) (2.02) (2.77) (2.82) (2.52) Current ratio 9.43E E E E E-06 (0.27) (0.32) (0.28) (-0.52) (-0.52) (-0.30) (0.30) (0.34) (0.34) Dividend yield *** *** *** *** *** *** (-3.12) (-3.10) (-3.53) (-0.83) (-0.84) (-1.18) (-4.41) (-4.39) (-4.79) TLCF *** *** *** E E-05 (1.13) (1.11) (1.14) (4.49) (4.43) (4.64) (0.72) (0.72) (0.74) Executive stock * (0.48) (0.64) (0.47) (1.53) (1.63) (1.66) (0.32) (0.49) (0.31) R-squared Table 16 shows the effect of derivative usage on total risk, systematic risk and idiosyncratic risk controlled by the variables leverage, size, MTBV (market-to-book value), liquidity, current ratio, dividend yield, tax loss carry forwards and executive stock percentage, for the year The explanations of these control variables can be found in paragraph 4.3 control variables. Columns (1) to (3) show the relationship between all derivatives, hedge and non-hedge derivatives and the three underlying s commodity, foreign currency (FCD) and interest rate (IRD) derivatives on total risk. Columns (4) to (6) show that relationship for systematic risk and columns (7) to (9) for idiosyncratic risk. The figures in the brackets are t-stats adjusted for heteroscedasticity using the White (1980) method. *, **, *** denote significance of 10, 5, 1%, respectively.

62 Firm Risk and the Use of Financial Derivatives without Control Variables Total Risk Systematic Risk Idiosyncratic Risk Pooled OLS (1) (2) (3) (4) (5) (6) (7) (8) (9) Constant *** *** *** *** *** *** *** *** *** (21.53) (21.55) (19.07) (55.44) (55.05) (49.44) (21.35) (21.39) (19.60) All derivatives * (1.81) (0.97) (1.49) Hedge * (1.69) (0.66) (1.40) Nonhedge E-05 (0.10) (0.67) (-0.06) Commodity (-0.45) (-0.37) (-1.29) FCD ** *** (2.35) (1.13) (2.61) IRD E-05 (0.94) (0.34) (0.19) R-squared Table 17 shows the effect of derivative usage on total risk, systematic risk and idiosyncratic risk for the pooled sample. Columns (1) to (3) show the relationship between all derivatives, hedge and non-hedge derivatives and the three underlying s commodity, foreign currency (FCD) and interest rate (IRD) derivatives on total risk. Columns (4) to (6) show that relationship for systematic risk and columns (7) to (9) for idiosyncratic risk. The figures in the brackets are t-stats adjusted for heteroscedasticity using the White (1980) method. *, **, *** denote significance of 10, 5, 1%, respectively.

63 Firm Risk and the Use of Financial Derivatives with Control Variables Total Risk Systematic Risk Idiosyncratic Risk Pooled OLS (1) (2) (3) (4) (5) (6) (7) (8) (9) Constant *** *** *** *** *** *** *** *** *** (4.11) (4.07) (3.85) (18.07) (18.16) (17.81) (5.40) (5.36) (5.10) All derivatives E-05 (0.67) (-0.63) (0.18) Hedge E-05 (1.08) (-0.70) (0.80) Nonhedge * *** (-1.69) (-0.06) (-2.62) Commodity ** (-1.47) (-0.90) (-2.08) FCD (0.95) (-0.07) (0.86) IRD 1.07E E-05 (0.01) (-0.76) (-0.65) Leverage *** *** *** *** *** *** *** *** *** (5.55) (5.69) (5.49) (3.71) (3.73) (3.69) (5.43) (5.62) (5.39) Size ** ** ** -5.70E E E *** *** *** (-2.59) (-2.52) (-2.57) (-1.54) (-1.54) (-1.55) (-3.74) (-3.69) (-3.70) MTBV *** ** ** *** *** *** (-2.59) (-2.51) (-2.52) (-0.61) (-0.61) (-0.60) (-2.73) (-2.66) (-2.67) Liquidity *** *** *** *** *** *** *** *** *** (4.54) (4.61) (4.32) (3.71) (3.70) (3.65) (3.91) (3.96) (3.78) Current ratio 1.55E E E E E E-07 (0.67) (0.67) (0.81) (0.72) (0.71) (0.71) (-0.14) (-0.14) (-0.01) Dividend yield * * * *** *** *** *** *** *** (-1.87) (-1.91) (-1.79) (-2.86) (-2.87) (-2.84) (-3.13) (-3.18) (-3.04) TLCF ** ** ** *** *** *** (2.40) (2.39) (2.43) (1.45) (1.45) (1.45) (2.75) (2.73) (2.77) Executive stock * ** * ** *** ** (1.93) (2.14) (1.78) (1.52) (1.54) (1.50) (2.49) (2.71) (2.36) R-squared Table 18 shows the effect of derivative usage on total risk, systematic risk and idiosyncratic risk controlled by the variables leverage, size, MTBV (market-to-book value), liquidity, current ratio, dividend yield, tax loss carry forwards and executive stock percentage, for the pooled sample. The explanations of these control variables can be found in paragraph 4.3 control variables. Columns (1) to (3) show the relationship between all derivatives, hedge and non-hedge derivatives and the three underlying s commodity, foreign currency (FCD) and interest rate (IRD) derivatives on total risk. Columns (4) to (6) show that relationship for systematic risk and columns (7) to (9) for idiosyncratic risk. The figures in the brackets are t-stats adjusted for heteroscedasticity using the White (1980) method. *, **, *** denote significance of 10, 5, 1%, respectively.

64 Fixed Effects Regression without Control Variables Total Risk Systematic Risk Idiosyncratic Risk Fixed Effects (1) (2) (3) (4) (5) (6) (7) (8) (9) Constant *** *** *** *** *** *** *** *** *** (15.90) (18.54) (18.86) (84.60) (70.60) (58.18) (26.23) (23.84) (20.84) All derivatives ** ** (2.09) (0.01) (2.07) Hedge ** ** (2.58) (0.13) (2.09) Nonhedge (1.37) (0.65) (0.46) Commodity (1.10) (0.25) (1.09) FCD *** *** (5.22) (0.52) (3.88) IRD *** *** (4.79) (-0.29) (3.61) R-squared Table 19 shows the effect of derivative usage on total risk, systematic risk and idiosyncratic risk corrected for firm fixed effects using panel data regressions. Columns (1) to (3) show the relationship between all derivatives, hedge and non-hedge derivatives and the three underlying s commodity, foreign currency (FCD) and interest rate (IRD) derivatives on total risk. Columns (4) to (6) show that relationship for systematic risk and columns (7) to (9) for idiosyncratic risk. The figures in the brackets are t-stats adjusted for heteroscedasticity using the White (1980) method. *, **, *** denote significance of 10, 5, 1%, respectively.

65 Total Risk Fixed Effects Regression with Control Variables Systematic Risk Fixed Effects (1) (2) (3) (4) (5) (6) (7) (8) (9) Constant *** *** *** *** *** *** (1.02) (1.02) (0.85) (14.32) (14.34) (14.19) (3.46) (3.45) (3.30) Derivatives ** ** (2.51) (-0.30) (2.35) Idiosyncratic Risk Hedge *** *** (3.20) (-0.11) (2.93) Nonhedge * ** (1.84) (0.60) (2.13) Commodity (0.76) (1.03) (0.81) FCD *** *** (3.21) (0.11) (2.85) IRD *** *** (5.38) (-0.57) (4.11) Leverage *** *** *** ** ** ** *** *** *** (7.54) (7.50) (7.45) (2.42) (2.34) (2.32) (4.74) (4.71) (4.70) Size ** ** ** -7.00E E E ** ** ** (-2.24) (-2.24) (-2.23) (-0.48) (-0.48) (-0.48) (-2.19) (-2.19) (-2.18) MTBV *** *** *** ** *** ** E E-07 (-2.69) (-2.69) (-2.71) (2.59) (2.62) (2.44) (-1.47) (-1.46) (-1.41) Liquidity ** ** ** (2.31) (2.30) (2.13) (1.47) (1.46) (1.37) (1.17) (1.15) (1.03) Current ratio * * * E E E-05 (-1.92) (-1.92) (-1.85) (-1.34) (-1.33) (-1.25) (-1.65) (-1.64) (-1.58) Dividend yield *** *** *** * * * (-1.27) (-1.29) (-1.38) (-3.60) (-3.59) (-3.73) (-1.78) (-1.80) (-1.95) TLCF *** *** *** *** *** *** (5.80) (5.89) (5.84) (-0.47) (-0.51) (-0.50) (5.56) (5.63) (5.61) Executive stock E (0.08) (-0.01) (0.06) (-0.04) (-0.12) (-0.05) (0.40) (0.34) (0.42) R-squared Table 20 shows the effect of derivative usage on total risk, systematic risk and idiosyncratic risk corrected for firm fixed effects using panel data regressions and controlled by the variables leverage, size, MTBV (market-tobook value), liquidity, current ratio, dividend yield, tax loss carry forwards and executive stock percentage, for the pooled sample. The explanations of these control variables can be found in paragraph 4.3 control variables. Columns (1) to (3) show the relationship between all derivatives, hedge and non-hedge derivatives and the three underlying s commodity, foreign currency (FCD) and interest rate (IRD) derivatives on total risk. Columns (4) to (6) show that relationship for systematic risk and columns (7) to (9) for idiosyncratic risk. The figures in the brackets are t-stats adjusted for heteroscedasticity using the White (1980) method. *, **, *** denote significance of 10, 5, 1%, respectively.

66 Derivative Products and the Relation to Firm Risk without Control Variables OLS Regressions Fixed Effects Regressions (1) (2) (3) (4) (5) (6) Constant *** *** *** *** *** *** (18.84) (49.03) (19.46) (20.89) (72.84) (19.32) Forwards ** ** ** *** *** (2.23) (2.11) (2.01) (6.07) (1.29) (2.68) Swaps E *** *** (0.61) (-0.48) (0.18) (6.06) (-0.12) (3.55) Options (0.31) (-0.04) (0.54) (0.84) (0.39) (1.07) Futures *** *** *** (0.06) (-0.53) (-1.61) (3.67) (-4.70) (8.14) Contracts ** * *** *** R-squared Table 21 shows the effect of different derivative products like forwards, swaps, options, futures, and unspecified contracts on the risk measures total-, system- and idiosyncratic risk. Columns (1) to (3) show the relationship obtained by OLS regressions, where column (1) represents the effect to total risk, column (2) to systematic risk and column (3) to idiosyncratic risk. Columns (4) to (6) show the relationship obtained by fixed effects regressions, where column (4) represents the effect to total risk, column (5) to systematic risk and column (6) to idiosyncratic risk. The figures in the brackets are t-stats adjusted for heteroscedasticity using the White (1980) method. *, **, *** denote significance of 10, 5, 1%, respectively.

67 Derivative Products and the Relation to Firm Risk with Control Variables OLS Regressions Fixed Effects Regressions (1) (2) (3) (4) (5) (6) Constant *** *** *** *** *** (3.40) (17.37) (4.90) (0.82) (13.99) (3.07) Forwards ** ** ** *** (2.23) (2.01) (1.99) (3.44) (0.79) (1.47) Swaps -6.30E E *** *** (-0.28) (-0.97) (-0.86) (5.26) (-0.43) (2.72) Options (0.36) (-0.33) (0.59) (-0.26) (-0.07) (0.85) Futures * ** *** (-1.93) (-0.61) (-2.01) (-1.20) (-4.70) (0.67) Contracts *** *** (-0.28) (-1.48) (-0.56) (3.86) (0.85) (3.64) Leverage *** *** *** *** ** *** (5.84) (3.92) (5.80) (7.41) (2.23) (4.80) Size ** -4.60E *** ** -5.70E * (-2.18) (-1.22) (-3.27) (-2.16) (-0.39) (-1.79) MTBV *** *** *** *** -5.00E-07 (-2.63) (-0.55) (-2.83) (-2.66) (2.77) (-1.45) Liquidity *** *** *** ** (4.55) (3.73) (4.24) (2.10) (1.38) (1.01) Current ratio 1.69E E * * (0.73) (0.66) (-0.15) (-1.84) (-1.27) (-1.72) Dividend yield * *** *** *** * (-1.74) (-2.79) (-3.00) (-1.23) (-3.53) (-1.72) TLCF *** *** *** *** (2.60) (1.52) (3.00) (5.91) (-0.54) (5.51) Executive stock ** * *** (2.22) (1.75) (2.73) (0.17) (-0.00) (0.48) R-squared Table 22 shows the effect of different derivative products like forwards, swaps, options, futures, and unspecified contracts on the risk measures total-, system- and idiosyncratic risk controlled by the variables leverage, size, MTBV (market-to-book value), liquidity, current ratio, dividend yield, tax loss carry forwards and executive stock percentage. The precise explanations of these control variables can be found in paragraph 4.3 control variables. Columns (1) to (3) show the relationship obtained by OLS regressions, where column (1) represents the effect to total risk, column (2) to systematic risk and column (3) to idiosyncratic risk. Columns (4) to (6) show the relationship obtained by fixed effects regressions, where column (4) represents the effect to total risk, column (5) to systematic risk and column (6) to idiosyncratic risk. The figures in the brackets are t-stats adjusted for heteroscedasticity using the White (1980) method. *, **, *** denote significance of 10, 5, 1%, respectively.

68 Nonlinear relationship: Firm Risk and Derivatives Nonusers included Nonusers excluded 1.53% 13.35% 1.53% 13.35% Low derivative users Constant *** *** *** * (3.74) (3.67) (2.92) (1.91) All derivatives ** *** (-2.11) (-0.42) (-2.70) (1.22) Leverage *** *** *** (2.82) (5.05) (-0.68) (4.13) Size * ** ** *** (-1.83) (-2.02) (-2.10) (-2.80) MTBV * ** -2.69E * (-1.69) (-2.34) (-1.21) (-1.78) Liquidity *** *** *** (1.33) (3.48) (3.19) (3.04) Current ratio 1.42E E E E-05 (0.53) (1.06) (-1.56) (1.29) Dividendyield ** ** (-1.31) (-2.28) (0.08) (-2.39) TLCF ** (1.56) (2.31) (0.86) (1.57) Executive stock ** ** *** (2.28) (2.46) (3.07) (1.30) High derivative users Dummy * * (-0.91) (1.58) (-1.83) (1.71) All derivatives ** *** (2.14) (0.63) (2.72) (-1.00) Leverage 4.27E ** (0.07) (-0.83) (2.21) (-0.58) Size * (-1.60) (-1.91) (1.07) (-1.42) MTBV 1.11E ** 2.50E ** (0.57) (-2.00) (1.12) (-2.04) Liquidity E (1.32) (0.72) (-0.01) (0.22) Current ratio 9.83E ** 9.81E ** (0.21) (-2.00) (1.58) (-2.03) Dividendyield (0.23) (0.45) (-0.54) (0.54) TLCF (-1.08) (0.47) (-0.75) (0.50) Executive stock ** *** * (-1.22) (-2.53) (-2.92) (-1.88) R-squared Table 23 shows the effect of derivative usage on total risk, where the sample is split in low derivative users and large derivative users with a threshold of 1.53% and 13.35% and controlled by the variables leverage, size, MTBV (market-tobook value), liquidity, current ratio, dividend yield, tax loss carry forwards and executive stock percentage. The explanations of these control variables can be found in paragraph 4.3 control variables. In the left columns nonusers are included and the two right columns non-users were excluded. The exact formula (equation 7) for this regression can be found in paragraph interaction regressions. The figures in the brackets are t-stats adjusted for heteroscedasticity using the White (1980) method. *, **, *** denote significance of 10, 5, 1%, respectively.

69 Nonlinear relationship: Firm Risk and hedging/speculating Nonusers included Nonusers excluded 1.53% 13.35% 1.53% 13.35% Low derivative users Constant *** *** *** * (3.66) (3.70) (2.83) (1.95) Hedge derivatives ** *** (-2.16) (-0.49) (-2.70) (0.96) Nonhedge derivatives *** (2.78) (-0.08) (-1.52) (0.70) Leverage *** *** *** (2.84) (5.05) (-0.73) (4.13) Size * ** ** *** (-1.81) (-2.02) (-2.09) (-2.77) MTBV * ** -2.83E * (-1.67) (-2.36) (-1.25) (-1.87) Liquidity *** *** *** (1.30) (3.47) (3.18) (3.06) Current ratio 1.57E E E E-05 (0.58) (1.05) (-1.54) (1.26) Dividendyield ** ** (-1.30) (-2.28) (0.11) (-2.37) TLCF ** (1.57) (2.30) (0.83) (1.56) Executive stock ** ** *** (2.26) (2.44) (2.93) (1.30) High derivative users Dummy * (-0.97) (1.26) (-1.86) (1.39) Hedge derivatives ** *** (2.20) (0.71) (2.73) (-0.74) Nonhedge derivatives *** (-2.87) (-0.54) (1.48) (-1.23) Leverage 8.08E ** (0.14) (-0.53) (2.21) (-0.29) Size * (-1.43) (-1.70) (1.12) (-1.19) MTBV 1.06E ** 2.64E ** (0.53) (-2.22) (1.16) (-2.25) Liquidity E (1.36) (0.77) (-0.05) (0.24) Current ratio 1.10E (0.23) (-1.58) (1.62) (-1.60) Dividendyield (0.21) (0.63) (-0.56) (0.71) TLCF (-1.09) (0.62) (-0.73) (0.64) Executive stock ** *** * (-1.12) (-2.23) (-2.77) (-1.66) R-squared Table 24 shows the effect of hedge and non-hedge derivatives on total risk, where the sample is split in low derivative users and large derivative users with a threshold of 1.53% and 13.35% and controlled by the variables leverage, size, MTBV (market-tobook value), liquidity, current ratio, dividend yield, tax loss carry forwards and executive stock percentage. The explanations of these control variables can be found in paragraph 4.3 control variables. In the left columns nonusers are included and the two right columns non-users were excluded. The exact formula (equation 7) for this regression can be found in paragraph interaction regressions. The figures in the brackets are t-stats adjusted for heteroscedasticity using the White (1980) method. *, **, *** denote significance of 10, 5, 1%, respectively.

70 Nonlinear relationship: Firm Risk and underlyings Nonusers included Nonusers excluded 1.53% 13.35% 1.53% 13.35% Low derivative users Constant *** *** ** ** (3.59) (3.64) (1.97) (1.97) Commodity *** *** (9.98) (-1.10) (3.14) (-0.73) FCD * (-1.38) (0.63) (-1.05) (1.76) IRD *** *** (-3.21) (-1.52) (-2.93) (-0.38) Leverage *** *** 9.89E *** (2.93) (5.14) (0.11) (4.24) Size * ** *** (-1.71) (-2.10) (-1.43) (-3.01) MTBV -2.89E ** -2.11E * (-1.60) (-2.29) (-0.97) (-1.79) Liquidity *** *** *** (1.28) (3.46) (2.78) (3.07) Current ratio 1.44E E E E-05 (0.53) (1.02) (-0.96) (1.19) Dividendyield ** ** (-1.36) (-2.12) (-0.64) (-2.01) TLCF ** (1.56) (2.19) (0.72) (1.48) Executive stock ** ** *** (2.34) (2.40) (3.63) (1.27) High derivative users Dummy (-1.05) (1.51) (-1.08) (1.63) Commodity *** *** (-10.04) (0.10) (-3.22) (-0.16) FCD (1.48) (0.11) (1.12) (-0.55) IRD *** *** (3.22) (1.62) (2.94) (0.48) Leverage 1.20E (0.00) (-0.81) (1.42) (-0.56) Size * * (-1.80) (-1.87) (0.31) (-1.37) MTBV 1.13E * 1.94E * (0.57) (-1.65) (0.89) (-1.68) Liquidity (1.17) (0.64) (0.12) (0.16) Current ratio 1.57E ** 7.06E ** (0.33) (-2.12) (1.21) (-2.13) Dividendyield (0.36) (0.36) (0.16) (0.40) TLCF (-1.06) (0.32) (-0.60) (0.35) Executive stock ** *** * (-1.38) (-2.43) (-3.47) (-1.90) R-squared Table 25 shows the effect commodity-, foreign currencyand interest rate derivatives on total risk, where the sample is split in low derivative users and large derivative users with a threshold of 1.53% and 13.35% and controlled by the variables leverage, size, MTBV (market-tobook value), liquidity, current ratio, dividend yield, tax loss carry forwards and executive stock percentage. The explanations of these control variables can be found in paragraph 4.3 control variables. In the left columns nonusers are included and the two right columns non-users were excluded. The exact formula (equation 7) for this regression can be found in paragraph interaction regressions. The figures in the brackets are t-stats adjusted for heteroscedasticity using the White (1980) method. *, **, *** denote significance of 10, 5, 1%, respectively.

71 45% Firm Percentage of Derivatives Used for Hedging 40% 35% 30% 25% 20% 15% 10% Commodity Interest Rate Foreign Currency 05% 00% Foreign Currency Interest Rate Commodity Figure 1 shows the firm percentage of derivatives used for hedging. This figure gives a clear view of table 4, where can be seen that foreign currency forwards and interest rate swaps are the most commonly used derivatives.