Why do firms actively vary the interest rate mix of their debt?

Transcription

1 Why do firms actively vary the interest rate mix of their debt? Jaideep Oberoi University of Kent Abstract There is an unresolved debate on whether regular changes in the proportion of fixed versus floating rate debt issued by non-financial firms represent hedging or speculation. In turn, this question impacts our understanding of which hedging theories are supported in the data. Utilizing a decade-long hand-collected sample of after-swap interest rate exposures, I show that firms actively vary their mix of fixed and floating rate debt to an unexpectedly high extent. However, it appears that firms with different types of hedging motives adopt seemingly contradictory approaches to activity. The more active firms are smaller in size with lower leverage and higher growth opportunities, and their operating profit margins are negatively associated with interest rates. Larger firms with lower growth opportunities and higher leverage are less active, and their cash flow margins are not as sensitive to interest rate levels. In addition, the choice of mix and changes in mix are linked to the nature of constraints faced by firms, as predicted by risk management theories. It appears that activity, by itself, does not help distinguish between hedging and speculation. I then empirically motivate an alternative indicator of speculative activity: co-variation between ex-post profitability of financial decisions and operating results. JEL Code: G32

2 I. Introduction It is well known that firms issue debt both at fixed and floating rates and that this mix affects their exposure to stochastic interest rates. Both types of debt present different forms of interest rate risk floating rate debt exposes a firm s net profits to variable interest costs, while fixed rate debt impacts the firm s future borrowing/investment capacity through changing liability or leverage levels. What motivates firms to choose one type of debt over another, and then to vary the mix over time? Further, can we interpret the choice of a particular mix, or the extent of its variation, as an indicator of risk-reducing or risk-increasing behavior by managers? In order to address these questions, I analyze a decade-long hand-collected sample of data that incorporates the net effects of interest rate swaps on the final mix of debt in the firms. In addition, the empirical corporate finance literature has sought to separate firms that hedge from those that do not by identifying hedgers. Many papers do this by using derivatives usage (as a binary indicator) or its extent as a proxy for hedging (e.g. Géczy et al. 1997). I show that such an approach is inappropriate in the case of interest rate risk. I consider instead whether some of the questions we study may be formulated based on recognizing speculative activity, and empirically motivate an approach to separating speculators. The first key observation from the data is that individual firms vary their interest rate mix to a striking extent. Yet, the average across the same firms of the share of fixed rate debt is stable over time. The somewhat idiosyncratic variation in individual firms debt mix has opened a debate on whether more activity in this regard reflects hedging or speculative conduct. The traditional link between activity and hedging is established in popular terminology. Changing the mix of debt regularly represents an active risk management style, as opposed to a passive approach where changing exposures are not regularly matched (see, e.g., Bodnar et al. [2]

3 1998). This terminology reflects the idea that a firm s level of activity is associated with a need to constantly manage changing exposures to risk. On the other hand, those, such as Chernenko and Faulkender (2011), who interpret such activity as speculation would argue that there is an optimal mix of debt types, and any deviation from it is caused by reasons other than risk management. I address this debate by seeking to understand the link between activity and risk management incentives in my dataset. To help motivate this enquiry, I first present some observations graphically. For each firm in each year, I calculate the proportion of debt that is subject to fixed rates after accounting for interest rate swaps. I then record how much this proportion varies for an individual firm (over time) and across firms (in a gi ven year). I also compare this variation with that of leverage, a financial characteristic that is commonly considered to represent risk management compulsions. In Figure 1, I show in Panel A that the distribution across firms of both the proportion of fixed rate debt and of leverage is fairly stable from year to year (as is seen from the median, interquartile range and standard deviation). Then, in Panels B and C, I look at how much these variables fluctuated for an individual firm over time the picture is now strikingly different. Over half the firms in the sample would at different times be classified both as owing mostly fixed rate debt and mostly floating rate debt. Jointly, these observations of the data present a puzzle as to the motivation for individual firms to vary their interest rate exposures to such an extent. Further, they bring into question whether we can assume the existence of a stable target proportion of fixed rate debt in individual firms. It is thus natural to ask why some firms are more active than others in varying their debt mix. [Insert Figure 1 here] Notions of optimality are widespread in terms of leverage ratios, cash holdings and other [3]

4 similar measures that trade off the costs and benefits of firms financial decisions. It is perhaps tempting to extend this idea to an optimal share of fixed rate debt. Such an optimal range is often alluded to in financial statements, but to my knowledge rarely specified. If indeed the active firms are speculating by deviating from an optimal target, there would be evidence of value destruction and of negative association with factors that represent risk management imperatives. On the other hand, if activity represents hedging then there should be evidence of shifting exposures that necessitate this activity. There is a mixed bag of conclusions found in the literature on the question of whether hedging theories are supported by the data. In the case of interest rates, this may be caused by limitations in both the theory and the data. The finance literature has not yet provided a coherent theory of how a firm should decide whether to borrow at a fixed or variable interest rate. The predominant theories of corporate risk management view risk as arising from exposure of a firm's (operating) cash flows to changes in particular variables. However, interest rate moves, regardless of whether they are up or down, could negatively impact a firm. Moves down raise the value of fixed rate debt and thus reduce the firm s borrowing capacity, while moves up raise the interest payments on floating rate debt. Both types of debt expose the firm to suboptimal investment decisions thus a firm could be seen to opt for a mix that trades off the two risks. A firm could also be opting for the same debt mix because its managers have taken a view on future interest rates. I attempt to examine both these possibilities by considering the firms changing exposures over time. Studies of interest rate exposure management among non-financial firms have also been limited by the unavailability of more detailed data. However, there was a period of transition in accounting disclosure regulations when firms were incentivized to provide details about their [4]

5 derivative positions in footnotes to their financial statements. During the 1990s, companies in general reported both the face value and direction of derivative contracts; presenting an opportunity to collect arguably the least biased data on firms' interest rate positions (see also Graham and Rogers 2002). New rules designed to bring the contracts onto the balance sheet then relaxed these disclosure requirements, creating a decade long window of data availability. I exploit this window to obtain data on interest rate derivatives and debt exposures for a sample of firms randomly selected from the Standard and Poors 500 index. Chernenko and Faulkender (2011) analyze exposure data from a similar time period, but their analysis relies on the idea that there is a stable and precise optimal mix of fixed and floating rate debt for each firm. However, a key issue in interest rate risk management is the absence of such a benchmark. I examine whether deviations from an average can be meaningfully interpreted as indicators of speculative activity, and find that this may not be the case. Some firms may switch between having fully fixed and fully floating debt such variation is seen in the data. They may be doing this to speculate or because their exposure has reversed it is difficult outright to say which. Yet, despite the firm showing no tendency to aim for an equal mix, the mean implied by the data would suggest that the optimal proportion of fixed rate debt for the firm is close to 0.5. This is a major reason why I focus on year-to-year changes in firms positions rather than attempting to explain deviations from a theoretical optimum. I group firms in a variety of ways to consider alternative concepts of activeness, and attempt to reconcile theories of risk management with empirical findings. For most of the analysis, I classify firms as being Active or Less Active based on the variation in the proportion of their debt that is at fixed rates. I find that the firms most actively changing their exposure have operating profits that [5]

6 vary negatively with interest rate levels, suggesting this activity may be linked to hedging cash flow exposures as the theory predicts. However, this activity is not linked to unconditional motivations for risk management implied by corporate finance theory. When comparing them to their Less Active counterparts, it turns out Active firms have lower leverage and higher average profitability (operating cash flow margins), both of which are associated with a lower risk of financial distress. Further, there is no evidence of higher average (over a decade) investment or research and development expenditure by these firms both motives for hedging identified in the literature. I also find no evidence of destruction in value on average for Active firms. This seems to suggest that both Active and Less Active firms are responding to their particular exposures rather than varying them to speculate to different degrees. The second theme of this paper relates to identifying speculation, or as managers euphemistically put it, taking a view on future market movements. A natural way to proceed here is to note that firms would speculate with the intention of beating the market. Due to the time series element of the data, I can examine the changes in firms positions in the context of monetary policy changes, which are in turn linked to changes in interest rates faced by firms. To do this, I study the implications of what a firm did in the previous year, in terms of increasing or decreasing the share of its fixed rate debt. They would be right ex-post if the decision has led to a favorable outcome in respect of the realized change in interest rates, e.g. fixing more of their debt just before interest rates go up. I find that firms that are right more often have no significant advantage over other firms. On the other hand, the short-term implications of being wrong include underperformance compared to other firms. Since the same firm could be right or wrong in different years, managers ability is implicitly controlled for. This supports the argument that if speculation is [6]

7 occurring, there does not appear to be a clear financial incentive for it. At the same time, this result underlines the need to explore the association between firms operational outcomes and financial choices as a potential indicator of their motivation. I argue that managers make several decisions across the firm consistent with their views or intentions. Thus, a correlation between apparently unrelated operational and financial outcomes might be observed when managers take a view on financial variables consistent with their operational decisions. Overall, I find that the high degree of variation across firms and over time in proportion of fixed rate debt cannot automatically be assumed to arise from speculative activity in the strict financial sense. That is why attempts to explain such activity with financial variables may not necessarily succeed, and one needs to look at simultaneous operational decisions to identify speculation. The rest of the paper is organized as follows. In the next section, I focus on the data. First I provide an overview of the data collection, while a related account of reporting regulations that justifies the sample period is left to an appendix. In Section III I first discuss the existing literature on the choice between fixed and floating rate debt and on potential explanations for changing the mix over time. I then examine the various hypotheses on both the levels of activity and the interest rate mix and discuss the results. In Section IV I study what happens when changes in the interest rate are profitable (or not) ex -post, and motivate an alternative indicator for identifying speculation. Section V concludes. II. Data on the mix of fixed and floating rate debt A. Sample selection The sample consists of randomly picked companies from the S&P 500 index. For inclusion of a [7]

8 firm in the database, there had to be a history of its annual financial statements available from the 1990s. The financial statements were taken from EDGAR SEC filings data (typically starting with the fiscal year 1994), as well as the Lexis-Nexis database for the early 1990s. 100 such companies were picked, and their statements were read for the financial years 1990 through Data on their derivatives usage and debt mix were then collected from the footnotes to the statements. The choice of the sample period is dictated by developments in accounting regulations that created a window in time when data on debt and swaps was jointly available in a manner to allow interpretation of net exposures. These regulations and their impact on data availability are discussed in more detail in Appendix A. B. Data collection procedure The data was collected by a team of research assistants, and our experience of organizing this process is detailed in Appendix B. We used mainly footnotes to the financial statements to collect data on debt and derivatives positions. For derivatives, we collected detailed information on all interest rate derivatives used by each company in each year. In addition, we maintained a list of the types of other derivatives used by each company. We searched the footnotes to the annual statements through the use of multiple search terms in addition to looking for certain sections of the footnotes manually to ensure detailed and accurate data. Our initial experience showed that searching for standard terms such as swap, cap, futures, forward, floor, collar, and hedg was not sufficient as companies often use terms such as interest rate protection agreement and interest rate exchange agreement to mean the same things. In some instances, the interest rate protection agreement might have been an option such as a cap or floor, but in most cases, it was determined [8]

9 to be a swap. In order to ensure completeness, we read through the sections that referred to fair value disclosures, risk management, off-balance sheet instruments, and debt, in addition to checking for references to these terms in the management's discussion section of filings. Where companies mentioned the hedging of investments as one of their reasons for using interest rate derivatives, we also read the section on investments. This distinction is important because a swap related to an investment will have precisely the opposite effect to that of a swap related to borrowing. We classified interest rate instruments as either changing the firm's exposure from floating rates to fixed rates or vice versa. By floating to fixed, I mean an instrument that changes the interest rate payments from a floating rate such as USD LIBOR to a fixed US dollar rate, directly or indirectly. The exposure is always seen from the point of view of a liability, so that an increase in interest rates is a negative development for a firm with floating rate exposure and a positive development if you have fixed rates instead. Similarly, a reduction in interest rates is good if you have floating rates and bad if you have fixed rates. We ignored the use of cross-currency swaps when they were being used to hedge foreign exchange exposure unless they were clearly shown to have an effect that would fit them into one of our two main categories, viz. floating to fixed or fixed to floating. The level and clarity of the disclosures vary significantly across companies. Some companies report individual contracts, since swaps are usually on large sums and are not very frequent. Other companies only provide aggregate level information on the notional and fair values and the direction of the swaps (fixed to floating or vice versa). This latter format is sufficient for our purposes as long as the underlying item(s), whose exposure is being hedged, are delineated. [9]

10 Effectively, the data gives us a snapshot at the end of each year of the extent to which companies altered the mix of fixed and floating rate debt with the use of derivatives. After collecting detailed information about all the instruments in the disclosures, I aggregated the exposures generated by derivatives into floating to fixed or fixed to floating. Where the disclosures were not sufficiently clear or we were not confident that we were obtaining a clear picture of the companies' positions, we dropped those companies from our sample. As a result, we have a final sample containing 82 firms over the eleven-year period. Some firms had to be dropped because they had no significant debt over the period. In the case of debt data, tabulation of existing debt was typically included in the footnotes along with further descriptions in some cases. We also checked the balance sheet when information on short-term debt was not separately provided in the notes. Classification of exposures on debt data often involved reading forward and backwards in years to where the terms of debt issues were properly reported as being at fixed or floating interest rates. Some companies report debt interest rates after the effect of swaps, and this was taken into account to avoid double counting the effect of swaps while combining the numbers. In order to obtain the final exposure I took the proportion of debt that was at fixed rates (without including the effect of derivatives) and then adjusted it by the net effect of all swaps. III Understanding the interest rate mix A. Background and Literature Interest rate risk management Initial work on the question of fixed versus floating rate debt was carried out in the 1980s, albeit [10]

11 from the point of view of a lending financial institution that faces a very different risk profile from a borrowing non-financial corporation. More recent papers are by Guedes and Thompson (1995), Chava and Purnanandam (2007), and Faulkender (2005), which are all cross-sectional in nature, and Chernenko and Faulkender (2011). Chava and Purnanandam empirically analyze the choice of fixed or floating rate debt for a cross-section of firms, and find that the vega of CFOs' compensation schemes is positively related to the proportion of floating rate debt. This is interpreted as consistent with risk-taking by CFOs who have incentives to take on more risk due to the nature of their compensation contracts. However, it is not clear that floating rate debt is riskier than fixed rate debt for all firms at all times. Further, compensation schemes would be negotiated after the business exposure to interest rates is already known, so they would take into account the various risk-taking incentives for managers. More importantly, as I show by looking at exposure choices over a length of time, firms alter their debt mix significantly from year to year, perhaps not as much as the compensation schemes in firms. Guedes and Thompson (1995) develop a signaling model whereby firms disclose their quality by issuing debt that is riskier for their cash flows. They argue that, as expected inflation volatility increases, floating rate debt becomes a better hedge for the firm. This is based on the explicit assumption that firms' operating cash flows are perfectly hedged against inflation - the nominal cash flow realization is a real cash flow adjusted for price level changes, so net income will remain stable. Further, they argue that in fact the low-risk debt issuance decision (floating during high inflation volatility) is a negative signal of firm quality, leading to a fall in stock price relative to a high risk issuer. An alternative explanation for this phenomenon was provided by discussions in Campbell (1978) and Santomero (1983), who were considering debt issuance choice from the perspective of bank liquidity dynamics. Since their introduction in response to [11]

12 high inflation in the 1970s, the share of long-term floating rate instruments in the markets has grown rapidly and varied significantly at the aggregate level, and this may be linked to banks negotiating power and balance sheet exposures. However, the rise of liquid markets in swaps has made the debt issuance decision less important in exclusivity. Chernenko and Faulkender (2011) study a panel of interest rate exposure data and argue that they find evidence for both hedging and speculation. They also find evidence for income smoothing by managers. While this may seem like contradictory evidence, it is representative of the complexity of the interest-rate exposure decision and the difficulty in pinning down what exactly it is that firms are doing in the absence of a proper benchmark rule. Also, as discussed in Appendix A on accounting regulations, there is a danger of encountering biased data after 2001 (see also Campello et al. 2011). Further, the income smoothing motive may be of two kinds. Chernenko and Faulkender focus on an ex-post approach, which implies that managers take myopic decisions of transferring accounting income between years. However, this is not linked to the value destruction that should arise. The other approach is to consider ex-ante income smoothing, i.e. an incentive to reduce risk from interest rates at any date and across periods. Such an approach does not imply value-destroying myopic behavior, but recognizes the effect of market frictions such as bankruptcy as being inherently costly. For most of the subsequent analysis, I use the variable pfix, which stands for the proportion of fixed rate debt. I define it as the proportion of debt after the effect of swaps that is subject to fixed interest payments. Proxies for risk management needs I need to consider what variables to include as proxies for hedging incentives. Theoretical justifications for risk management include the existence of growth opportunities and of financial [12]

13 distress costs as central considerations. For instance, Stulz (1984), Smith and Stulz (1985) and Froot, Scharfstein and Stein (1993) predict a positive relationship between these va riables and hedging activity. Purnanandam (2008) predicts a U-shaped relationship whereby a firm increases its risk when it gets very close to a default boundary, but otherwise reduces it in line with the likelihood of bearing distress costs. On the other hand, Morellec and Smith (2007) show that large firms with low growth opportunities may also reduce risk in order to avoid the type of overinvestment costs described in Jensen (1986). Further, Almeida, Campello and Weisbach (2011) show that firms with future financing constraints may have a negative relationship between risk-taking and leverage. These latter papers motivate me to retain liquidity and (cash flow) profitability in the mix of predictors. Guided by existing theories on why firms hedge, I collect data from Compustat on market value, sales, leverage, liquidity, research and development expense, capital expenditure, and profitability measures including cash flow margins and return on assets. In order to examine the channels for profitability better, I also include cost variables in my dataset. Market value (mv) is calculated as the product of common shares outstanding in millions (csho) and share price at the close of the fiscal year (prcc_f) in the industrial annual Compustat file. Similarly, I have sales (sale) and total assets (at), leverage is total liabilities (long term debt dltt, plus debt in current liabilities - dlc) divided by total assets, while research and development expenditure (xrd) and capital expenditure (capx) are normaliz ed by total assets. For liquidity, I define the variable quick ratio as cash plus short term investments (che) divided by current liabilities (lct). Cash flow margins are calculated as operating income before depreciation (oibdp) divided by sales, and return on assets as income before extraordinary items (ib) divided by total assets. Market-to-book ratios are calculated as total assets minus common equity (ceq) plus market [13]

14 value, all divided by total assets. Finally, we also have the ratios Cost of goods sold (cogs) and Sales, General and Administrative Expense (xsga) both normalized by sales. To provide an overview of the firms in the data sample, I first calculate the average for each firm over the sample period of each of the variables. I then summarize these averages in Table I. [Insert Table I here] Interpreting variation in the interest rate mix One question that arises from the previous literature is to understand whether there is truly a stable optimal mix of interest rates for hedging risk, or whether firms must actively vary their proportion of fixed rate debt in response to changing exposures. This is essential for understanding whether a higher level of variability in the mix corresponds to the risk exposure of the firm being high in general, its risk exposure changing very often, or its managers gambling in financial markets with shareholder money. In order to study the associations with this type of activity, I separate the firms into two groups. This classification is based on the standard deviation of pfix over the sample period. 1 Firms with their standard deviation of pfix in the sixtieth percentile and above are classified as Active, while those with the standard deviation below the fortieth percentile are classified as Less Active. I exclude the middle 20% of firms to ensure the two groups are distinct. In the rest of the section, I study two sets of questions. In subsection B below, I ask what explains the high variation of pfix in individual firms. In subsection C, I consider the changing 1 Using the range of pfix as the classification criterion results in categories that have an overlap of 93%, and other results are similar. I elected to use standard deviation as it may capture frequent changes better and I believe that may be important for the current analysis. [14]

15 levels of pfix in connection with changing firm characteristics and seek to reconcile the findings with the theory. B. Variation of pfix and risk management One way to examine whether Active firms are hedging or speculating is to compare the strength of their hedging incentives relative to Less Active firms. In this section, I am considering whether (average) firm characteristics that are proxies for risk exposure are associated with the split between Active and Less Active firms. I first test for differences in the firms sensitivity to interest rates and then compare them on a range of variables that represent (unconditionally) an incentive to manage risk. The latter set of tests is carried out one variable at a time, and then in a multiple regression. B.1 Are Active firms operating cash flows more sensitive to interest rates? The primary candidate for hedging motives is based on the sensitivity of the firm's cash flows to interest rates. This theory predicts that the firm would try to adopt a hedge ratio that will allow it to have sufficient internal cash in case investment opportunities arise (see Froot, Scharfstein and Stein 1993). Faulkender (2005) obtain s firm-by-firm estimates of the betas of cash flows with respect to interest rates these betas could then be used to proxy for hedge ratios and determine whether firms should be taking on fixed or floating rate debt. When I attempt a similar exercise, I find that the coefficients from the regressions carried out for one firm at a time are often not statistically different from zero. Therefore, the betas from firm-by-firm regressions used for further analysis without accounting for measurement error may lead to biased estimates. Instead, I attempt to flip the analysis around (since I am not looking for causality, just association). I [15]

16 group all firms into a single panel regression, using data over all the quarters from January 1989 through December The dependent variable is cash flows/book assets as in earlier papers, regressed on its own lag, LIBOR, a dummy for the Active group, and its interaction with LIBOR to estimate the slope differences. As the profitability is very persistent, I include its lag as a regressor to reduce the danger of spurious regression results. To account for firm-specific characteristics, I use a fixed-effects specification. The regression results are presented in Table II for the following specifications: CF it Active * LIBOR u 0, i 1CFi, t 1 2LIBORt 3 i t it (1) Since the mean cash flow dummy would lead to collinearity in the fixed effects regression, this specification provides us with estimates only of the difference between the sensitivities of the cash flows of the Active group with that of the Less Active group. Although the fixed-effects specification is supported by a Hausmann test, I also report a random-effects regression specified as CF it * 0 1CFi, t 1 2LIBORt 3Activei 4Activei LIBORt i uit (2) which is estimated by the Feasible Generalized Least Squares method. In both cases, I report robust standard errors. [Insert Table II here] I find that, conditional on past profits, Active firms profits vary negatively with interest rates, while those of Less Active firms have a marginally positive relationship with the level of LIBOR. Contrary to the idea that greater activity is a sign of speculation, these regressions suggest that Active firms have operating profits that are subject to interest rate risk. [16]

17 I examine further why firms vary their interest rate mix so widely, when I compare other predicted characteristics for hedgers between the two categories of firms. B.2 Higher activity and other proxies for risk management First I examine whether Active firms are different on average from other firms. To do this, I calculate the average over the entire sample period for each firm of the characteristics listed above. These averages are summarized in Table III. [Insert Table III here] In the last two columns of Table III, I report the differences in the average firm characteristics between the two categories of firms, i.e. Active vs. Less Active, with the p-values below each difference. I find that Active firms are smaller in terms of sales and assets but not in terms of market value of equity. They have lower leverage, higher return on assets and higher market-to-book ratios than Less Active firms. By the median measure, they may also be more liquid. The results of these univariate comparisons indicate that on average (over time) f irms with higher growth potential and lower financial distress are more active in changing their mix of debt. Here I am using leverage and quick ratio as proxies for financial distress and market-to-book ratio and research and development (R&D) expense as proxies for agency costs. We can observe that there is a slight negative relation between the level of leverage and variability of pfix. However, this seems to be contrary to the general presumption in the literature that links higher leverage to higher risk, necessitating more hedging activity through frequent adjustments. This basic picture is puzzling. It suggests that firms with lower leverage are more active in varying their interest rate exposures. From this it is no longer clear that we can interpret a greater level of activity as evidence of active hedging. Neither does this match the prediction of [17]

18 speculation due to risk shifting incentives as in Jensen and Meckling (1976) The fact that Active firms are smaller and have higher market-to-book ratios than Less Active firms would theoretically point to the level of activity as being an indicator for hedging behavior based on the costly external financing argument. This may also be consistent with the Active firms having lower leverage and higher liquidity, but is not easy to reconcile with the financial distress argument. The difficulty arises when we see that the actual rate of capital expenditure or research and development expenditure in Active firms is not different from that in Less Active firms over the eleven-year period. Perhaps the findings of lower leverage fit better with the overinvestment avoidance argument of Morellec and Smith (2007), but then we are faced with the lower size and higher market-to-book ratios of Active firms as contradictions. This may still be consistent with Almeida, Campello and Weisbach (2011), to the extent that these firms could face future financial constraints. When we look at the flip side, the Less Active firms have higher leverage and lower liquidity which could be considered consistent with hedging behavior due to financial distress costs. Does this imply that different types of hedging motives necessitate different hedging strategies? In other words, do both smaller firms with growth options and those that are more profitable have a different policy towards interest rate risk than those that may face greater financial distress costs? I now turn to a multivariate setting to try and answer the above question. In Table IV, I present the estimates from a probit regression in which the dependent variable is 1 if the firm is classified as Active, and 0 otherwise. The regressors include all the firm characteristics ( I drop market value and total assets in favor of sales as a size proxy). I use robust standard errors and report the p-values below the estimates. The estimates reported are not the raw estimates, but the [18]

19 marginal effects estimated at the mean of the explanatory variables. [Insert Table IV here] The multiple regression also provides conflicting evidence for risk management theory. Firms are likely to vary their interest rate mix more when they have lower sales and (with a weak significance) lower leverage along with higher market to book ratios. Are these firms the hedgers predicted by the combined complementary arguments in Smith and Stulz (1985) and Morellec and Smith (2007), aiming for a target mix of debt? The overinvestment avoidance argument is also supported by the fact that Less Active firms have higher sales, general and administrative expense ratios combined with weakly higher operating profit margins, a potential indicator of the firms relative efficiency. Once again, the flip side of the results points to smaller firms with more growth opportunities and lower leverage being Active the very firms predicted by asymmetric information considerations to be hedgers. It appears that Active and Less Active firms face distinct types of hedging motives. Active firms are smaller in size with lower leverage and higher growth opportunities, and their operating profit margins are negatively associated with interest rates. As firms get bigger and have lower growth opportunities and higher leverage, they are Less Active, and their cash flow margins are not very sensitive to interest rate levels. C. The level of pfix Instead of finding that Active firms are varying their exposures with speculative incentives in mind, we have seen that they just appear to be different in the nature of their exposures. Yet, we haven t been able to pin down precisely where the differences arise. Could it be that the Active firms just have more variable firm characteristics that necessitate more variable strategies? In [19]

20 this subsection, I consider whether the level of pfix over time (rather than its variation) can tell us more about the hedging motives of firms. First, however, I digress briefly to check whether there is any unconditional relationship between risk and the level of pfix an assumption often made in the literature. C.1 Do firms with hedging considerations take up more fixed rate debt on average? Is it just the case that fixed rate debt is safer than floating rate debt, so that firms wishing to hedge against interest rate risk would just employ more fixed rate debt? I consider two typical predictors of the value of managing risk the cost of financial distress and growth opportunities. To proxy for these two variables, I use the leverage and book-to-market ratio respectively, where the book-to-market ratio is naturally interpreted as inversely varying with growth opportunities. The hypothesis is that there is a positive relationship between pfix and leverage, and a negative relationship between pfix and the book-to-market ratio. Taking all firm-year observations, the two correlations are respectively 0.01 and 0.17 respectively. Further, if we calculate the average pfix, leverage and book-to-market ratio for each firm (over the entire sample period), the correlations of these averages are 0.03 and 0.32 respectively. Unconditionally, there is very little support for the idea that issuing fixed rather than floating rate debt is a measure of interest rate risk management. C.2 Changing firm exposures and pfix I have shown above that the existence of a stable optimal proportion of fixed rate debt for a firm cannot be automatically assumed. Indeed, Section C.1 also shows the absence of an unconditional relationship between pfix and average risk indicators. However, the average characterization of firm hedging incentives is inappropriate for our purposes if the exposure (and [20]

21 thus the optimal response) to interest rate risk varies over time. Other firm characteristics such as liquidity and profitability might affect the way in which the firm responds to interest rate risk. A more complete analysis of the choices determining pfix is presented in Table V. Unlike the previous literature, I pay attention to the fact that pfix is a proportion. This makes it inappropriate to use a linear regression model with pfix as the dependent variable, as it may lead to biased and inconsistent estimates (see, e.g., McCullagh and Nelder 1989; Papke and Wooldridge 1996; Cook et al. 2008). As a consequence, I follow the strategy proposed by Smithson and Verkuilen (2006) to estimate a Beta Regression Model. This involves assuming that the pfix is conditionally beta distributed with nonlinear submodels for both the location and dispersion parameters of the beta distribution. Such an approach takes account of the necessary heteroskedasticity of the data (as a bounded variable will have variance changing with loca tion), and also allows me to model the precision of the distribution as a function of the same covariates that predict its conditional mean. It is possible to estimate this model in most standard statistical software I use the betafit package (Buis et al. 2005) in Stata for the regressions in this paper. The model is specified as (following Smithson and Verkuilen 2006): Let µ be the location parameter and φ the dispersion parameter of a beta distribution. Further, let x i be a row of observations of explanatory variables, including a constant. Then, = ( ) 1 + ( ), = ( ) where β and γ are parameter vectors. Here, µ is the expected value (or location) of the beta-distributed dependent variable (pfix in this case). The variance of the dependent variable is determined jointly by µ and φ, through the relation: Var( ) = µ(1 - µ)/(1 + φ). This means that, [21]

22 for a given level of µ, higher φ is associated with lower variance. My empirical specification includes the same firm characteristics that are used to proxy risk management imperatives. As I have shown above, the mean pfix for a single firm is uninformative and potentially misleading, but there is some variation in the cross-firm average of pfix from year to year that may bias the results, so I also include year dummies among the explanatory variables. The results of this regression show that firms with higher profitability or operating cash flows tend to have a lower pfix while those with higher short-term liquidity (quick ratio) tend to have a higher pfix. This is consistent with the arguments in Acharya, Almeida and Campello (2007), and Almeida, Campello, and Weisbach (2011), which suggest that financially constrained firms would hold more liquid reserves. The notion of constraint here is the borrowing capacity of the firm. A firm with more fixed rate debt exposes itself to the risk of increased leverage, hence endangering borrowing capacity it would thus be optimal for such a firm to also hold more cash as a hedge. On the other hand, firms with high operating cash flows would face a reduced constraint. In other words, we can see that the choice of issuing more fixed or floating rate debt is linked to other factors affecting the firm, and these change over time. It is not important to say whether the choice of debt type is determined by the liquidity and profitability of the firm or vice versa the theory suggests that these would be jointly determined. From the analysis, we also see that higher leverage is associated with lower pfix. This does not contradict the arguments above, but may also be linked to an alternative explanation. Kahl, Shivdasani and Wang (2008) argue that firms tend to borrow short-term funds (such as, by issuing commercial paper) for the purpose of making large investments or financing acquisitions. [22]

23 The fact that the leverage-pfix relationship does not appear to hold unconditionally offers increases in short-term borrowing as a potential explanation. Higher leverage is also linked to higher dispersion in the pfix, suggesting that more extreme values of pfix would be seen in higher leverage firms. This conditional finding is consistent with the argument that firms jointly determine their exposure in consideration with a combination of factors and it would be wrong to assume an unconditional link between leverage and the level of pfix or its variation. Sales and quick ratio are marginally significant for the dispersion parameter, with higher sales reducing dispersion and a higher quick ratio increasing it. The interpretation of these dispersion effects is not inconsistent with the results on Active versus Less Active firms above. [Insert Table V here] To recap the findings from this section, the evidence suggests that we cannot unconditionally interpret fixed rate debt as less risky, while an optimal mix of debt can only be interpreted as being jointly determined by other policies and exposures of the firm, and hence to be varying significantly for some firms over time. Thus, activity itself, or deviation from a mean level of fixed rate debt, is just as likely to be consistent with optimal hedging strategy as maintaining a constant level. We are clearly not able to separate hedgers from speculators based on the level or variability of their interest rate mix instead, we have found evidence that firms with different types of hedging incentives adopt different approaches to varying their interest rate mix. IV Identifying hedging or speculation The question remains how we can identify whether a firm is a hedger, if we are to try and make inference on the implications of hedging on firm performance or value (as is often the goal in the [23]

24 literature). With the aid of an example, I first point out how we should not identify hedgers. Later, I consider whether we might be more successful at recognizing speculation rather than hedging, and present some arguments to show how this is possible. A. Derivatives usage and hedging (or speculation) Several empirical papers have relied on a derivatives usage indicator variable as a proxy for hedging activity. Yet others have used the volume of derivatives positions as representing the extent of hedging. Both the above uses of derivatives data may be misleading. I demonstrate this with the help of an example. Figure 2 presents the debt and swap sizes from 1990 to 2000 of Avery Dennison, a manufacturer of pressure sensitive materials. In the first five years (1990 through 1994), Avery Dennison increased the total notional value underlying its swap contracts from USD 100 million to USD 320 million. In 1990 and 1991, the firm had USD 100 million of swaps outstanding, all having the effect of converting floating payments to fixed ones. In 1992, the notional value underlying its swaps doubled to USD 200 million. However, the net effect of the outstanding swaps was still USD 100 million floating to fixed. This is because, by the end of 1992, there were USD 150 million of floating to fixed, and USD 50 million of fixed to floating swaps on the company's books. More interestingly, in the following year, as the notional value of swaps increased further to USD 300 million, the net effect of all the swaps fell to zero. [Insert Figure 2 here] Two observations can be made from this example. First, although the firm had no swap positions after 1995, the variability of the interest rate mix was similar to the earlier period. Second, the net effect of swaps had no relation to the total notional size of the swap positions. [24]

25 Overall, the picture is inconsistent with a simple interpretation of derivative usage as a proxy for either speculation or hedging. B. Can we tell if firms are timing the market? We have seen that Active firms are just as likely to be hedging interest rate risk as the Less Active firms. I now ask whether we can identify firms that are speculating, i.e. taking a view on interest rate movements. The natural way to approach this question would be to recognize that the objective of taking such a view would be financial profit. It is clearly very difficult to look at a position taken by a firm and determine it to be a hedge or a speculative bet when we are not aware of the other operational decisions the manager is taking. More so, the slope of the term structure combined with the manager s position only tells us that a manager may be smoothing their results, which has proper justifications ex-ante i.e., before they know their results. It is difficult, therefore, to simply label a position as speculative based on observation of financial forecasts. Instead, I take advantage of the time-series aspect of the data by looking at the ex-post outcomes from firms financial and operating decisions. First, I look at the variation in interest rates over the sample period. I use the Fed Funds rate, which is plotted in Figure 3a. In Figure 3b and Table VI, I present the position that a firm would have taken if it were purely speculating financially under the guise of hedging debt and could predict the rates for the next year. 2 I call this being Right. A firm could be Right in a given year, and unlucky the following year, so I also look at the average of Right (which is a binary 2 Monetary policy decisions transmit rapidly into the wider financial market and to interest rates like the LIBOR that affect swap and borrowing rates, even if they take longer to transmit to the real economy. For some evidence on the role of monetary policy, see Ehrmann and Fratzscher (2004). [25]

26 variable) for each firm over 10 years. Firms that were in the 60 th percentile or higher of being Right on average are labeled AllRight, and those that were in the 40 th percentile or below are the other group. The distribution of the average of Right, which we can consider as a success rate in terms of our current objective, is provided in Figure 4. Similarly, the proportion of firms that were Right in any given year is provided in Figure 5. [Insert Figure 3, Table VI, Figure 4 and Figure 5 here] Once again, a familiar pattern emerges, of heterogeneity across firms in any given year and over time for a given firm, but not so much on average across firms over time. Table VII presents a cross-tabulation of the two classifications: Active and AllRight. What is immediately obvious is that more of the Active firms had a value of AllRight equal to zero than one. [Insert Table VII here] If Active firms are speculating from a purely financial perspective, they are doing a poor job of it and not learning from their mistakes. Either this means that the mistakes are not costly, or that we should see some difference in the value of these firms. We have already seen that Active firms have higher market-to-book ratios than Less Active firms, and this raises further questions about the assumption that Active firms are speculators. What, then, of being Right, or even AllRight? I compare the same set of characteristics (averages) as in earlier sections in a sequence of univariate tests between firms that have AllRight values of zero or one. I also run a probit regression to check if there is any feature of a firm that would predict that it is more likely to be Right on average. Resoundingly, none of the characteristics (size, profitability, liquidity, cost management) have a significant coefficient. Being AllRight appears to be a random phenomenon. More importantly, I have compared the [26]

27 average over 10 years of the firm characteristics, and we can see that there is no particular way in which the firms that are more successful at guessing rates stand out. I would emphasize that, on average, there appears to be no advantage from speculating and being Right, and also no disadvantage from making ex-post unprofitable bets. The case for speculative activity should show either one of these results, or it is weakened. Are there any short-term consequences to being Right? I examine what happens to firms in the year following their move by comparing those that were Right and those that were wrong on a year-by-year basis. I now treat firm-year observations independently, pooling them and allowing a single firm to gain one year when it is Right and lose in another when it is wrong. The results in Table VIII show that in the year following a Right move, firms were better off compared to their Wrong counterparts due to reduced leverage and superior operating performance reflected in a better control of sales, general and admin expenses. Finally, they had to borrow less short-term debt in that year. [Insert Table VIII here] Note that the financial result (return on assets) is not significantly different among these firms, but the operating cash flow which does not take into account the interest expense is actually better. It would be a stretch of the imagination to suggest causality from the change in pfix to purely operating performance. An alternative explanation for these findings is that managers may take a view on future developments in the economy, but they then act on them across the activities of the firm, not just in a financial sense. Evidence of speculation based purely on publicly available financial forecasts falls short because it does not take into account the operational aspects of the firm. On the other hand, when [27]

28 managers act to manage costs or make changes in a firm based on their view of the future, the outcome for their firm is correlated in the financial and the operational domains. We have finally found some evidence that managers may be acting on their views when they choose their mix of fixed and floating rate debt, but the relationship is too complex to be reflected by a simple proxy such as active swaps usage, or even the level of variation in the share of fixed rate debt. While we cannot separate hedgers from speculators directly, we can estimate the correlation between operational and financial outcomes over time to recognize where managers have been taking a view on the financial markets while making their interest rate exposure decisions. This requires a series of observations over time to identify speculative behavior. Having examined carefully the proportion of fixed rate debt chosen by firms and the reasons for its variation, I can now conclude. V Conclusion If firms are speculating in anticipation of interest rate changes, there does not appear to be any significant benefit from forecasting rates correctly (in terms of value or performance), but there is evidence that guessing wrongly is related with negative short-term operating results. I have studied the proportion of fixed rate debt in firms by using hand-collected data from a specific period of time when this type of data retrieval is informative. My approach was to emphasize the nature and scope of variation of this variable, and to demonstrate how this causes weaknesses in attempts to understand interest rate risk management in non-financial corporations. In particular, the question of how to classify a firm as a hedger or a speculator still remains a difficult and open one. I have shown how both simple and sophisticated attempts at [28]

29 classifying a particular position as speculative or a hedge could fail when we do not take other operational information into account. Contrary to Chernenko and Faulender (2011), I show that firms that actively vary their exposures are not necessarily speculating, and their deviations from a theoretical average should not immediately be interpreted as speculative positions. I show instead a potential way to identify speculative activity is based on its effects if the operational decisions of a firm are further exacerbated by interest rate changes, we can hope to conclude that the manager was taking a view on interest rates as well. I find evidence consistent with Almeida, Campello and Weisbach (2011), whereby firms with lower leverage and potential financing constraints are seen to be actively hedging their exposures. While firms with high leverage and more fixed rate debt retain more liquidity, those with lower leverage and more floating rate debt tend to have higher operating cash flows. My results are not inconsistent with the survey findings in Bodnar, Hayt and Marston (1998) that a majority of managers consider their view of the economy before taking up derivative positions. Although this in itself could be viewed as speculation, I show that the overall financial decision of choosing the proportion of fixed rate debt may in fact be linked to other simultaneous operational decisions. This conclusion also brings into question whether managers use swaps to smooth income ex-ante or ex-post. If managers do take a view on interest rates in conjunction with other views on the future operational performance of the firm, ex-post measures of earnings manipulation may show up significant in the data, though we cannot tell if they are based on ex-ante or ex-post decisions. This, however, makes a great deal of difference for our conclusions about firm activity. The finding that firms that are Active and those that are more often wrong in guessing the direction of rates are not less valuable on average suggests that ex-ante incentives to smooth [29]

30 earnings may influence empirical findings on ex-post manipulation. Several questions remain unanswered. The fact that the mean across firms of pfix does not change much suggests that the demand and supply effects transmitted through the financial intermediary sector are playing a role. This argument is developed in the context of the maturity choice by Greenwood, Hanson and Stein (2010). It is important to recognize that the results in this paper are particularly relevant to the two-sided nature of interest rate risk and should not be automatically extended to other sources of financial market risk. We do not yet have a clear theoretical idea of what the optimal choice rule for firms should be for interest rates. Further, we do not yet have a way to pin down the extent to which constraints bind on firms risk management policies. My results point to non-linear and more complex factors in determining the hedging policies of firms. I leave these questions to future work. [30]

31 References Acharya, V., H. Almeida, and M. Campello, 2007, Is Cash Negative Debt? A Hedging Perspective on Corporate Financial Policies, Journal of Financial Intermediation 16, Almeida, H., M. Campello, and M. S. Weisbach, 2011, Corporate Financial and Investment Policies when Future Financing Is Not Frictionless, Journal of Corporate Finance 17, Bodnar, G.M., G. S. Hayt, and R. C. Marston, 1998, 1998 Wharton Survey of Financial Risk Management by US Non-Financial Firms, Financial Management 27, Buis, M.L., N. J. Cox, and S. P. Jenkins, 2003, BETAFIT: Stata Module to Fit a Two-Parameter Beta Distribution, Statistical Software Components S435303, Boston College Department of Economics, revised 03 Feb Campbell, T. S., 1978, A Model of the Market for Lines of Credit, Journal of Finance 33, Campello, M., C. Lin, Y. Ma, and H. Zou, 2011, The Real and Financial Implications of Corporate Hedging, Journal of Finance 66, Chava, S., and A. Purnanandam, 2007, Determinants of the floating-to-fixed rate debt structure of firms, Journal of Financial Economics 85, Chernenko, S., and M. Faulkender, 2011, The Two Sides of Derivatives Usage: Hedging and Speculating with Interest Rate Swaps, Journal of Financial and Quantitative Analysis 46, Cook, D. O., R. Kieschnick, and B. D. McCullough, 2008, Regression Analysis of Proportions in Finance with Self Selection, Journal of Empirical Finance 15, Ehrmann, M., and M. Fratzscher, 2004, Taking Stock: Monetary Policy Transmission to Equity Markets, Journal of Money, Credit, and Banking 36, Faulkender, M., 2005, Hedging or Market Timing? Selecting the Interest Rate Exposure of Corporate Debt, Journal of Finance 60, Froot, K. A., D. S. Scharfstein, and J. C. Stein, 1993, Risk Management: Coordinating Corporate Investment and Financing Policies, Journal of Finance 48, Géczy, C., B.A. Minton, and C. Schrand, 1997, Why Firms Use Currency Derivatives, Journal of Finance 52, [31]

32 Graham, J., and D. Rogers, 2002, Do Firms Hedge in Response to Tax Incentives? Journal of Finance 57, Greenwood, R., S. Hanson, and J. C. Stein, 2010, A Gap-Filling Theory of Corporate Debt Maturity Choice, Journal of Finance 65, Guedes, J., and R. Thompson, 1995, Tests of a Signaling Hypothesis: The Choice between Fixed- and Adjustable-Rate Debt, Review of Financial Studies 8, Jensen, M. C., and S. H. Meckling, 1976, Theory of the Firm: Managerial Behaviour, Agency Costs and Ownership Structure, Journal of Financial Economics 3, Jensen, M. C., 1986, Agency Costs of Free Cash Flow, Corporate Finance, and Takeovers, American Economics Review 76, Kahl, M., A. Shivdasani, and Y. Wang, 2008, Do Firms Use Commercial Paper to Enhance Financial Flexibility? Working Paper. McCullagh, P., and J.A. Nelder, 1989, Generalized Linear Models, 2 nd edn., Chapman and Hall. Morellec, E., and C. Smith, 2007, Agency Conflicts and Risk Management, Review of Finance 11, Papke, L., and J.M. Wooldridge, 1996, Econometric Methods for Fractional Response Variables with an Application to 401 (K) Plan Participation Rates, Journal of Applied Econometrics 11, Purnanandam, A., 2008, Financial Distress and Corporate Risk Management: Theory & Evidence, Journal of Financial Economics, 87, Santomero, A. M., 1983, Fixed Versus Variable Rate Loans, Journal of Finance 38, Smith, C. W. and R. M. Stulz, 1985, The Determinants of Firms' Hedging Policies, Journal of Financial and Quantitative Analysis 20, Smithson, M., and J. Verkuilen, 2006, A Better Lemon Squeezer? Maximum-Likelihood Regression with Beta-Distributed Dependent Variables, Psychological methods 11, Stulz, R. M., 1984, Optimal Hedging Policies, Journal of Financial and Quantitative Analysis, 19, [32]

33 Appendix A. Evolution of accounting standards for reporting derivatives As the use of derivatives increased in the 1980s, accounting regulations evolved in response, thereby providing a window of time when we could learn about the directional exposure of firms to interest rates. In this appendix, we recount how off-balance sheet instruments were reported in the footnotes of firms' financial statements as a result of a series of accounting regulations. In December 1989, the Federal Accounting Standards Board (FASB) issued the Statement of Financial Accounting Standards No. 104 (FAS104) that amended FAS95 (November 1987) to allow for hedge based accounting classification of derivatives as long as the accounting policy was disclosed. Hedge-based accounting classification refers to the practice of assigning derivatives based cash flows to the same category as an identifiable underlying. Under the original FAS95, such transactions were required to be classified according to the nature of the cash flows, and not under the headings of the items being hedged. The Statement was made effective for financial years ending after June 15, FAS95 was followed by FAS105 in March 1990 which was entitled: Disclosure of Information about Financial Instruments with Off-Balance-Sheet Risk and Financial Instruments with Concentrations of Credit Risks. FAS105 was introduced as a first step toward a general system of reporting financial instruments whose usage and variety was steadily increasing. It required fuller disclosure by corporations about their off-balance sheet financial transactions that included the notional principal accompanied by information/discussions about the applicable market and credit risk. FAS105 was also applicable for financial years ending after June 15, 1990 and has since been superseded by a series of other statements beginning with FAS107 (Disclosures about Fair Value of Financial Instruments) that was issued in December [33]

34 FAS107 mainly had the impact of extending and generalizing the applicability of FAS105 and it was applicable for financial years ending after December 15, In October 1994, the FASB issued FAS119: Disclosure about Derivative Financial Instruments and Fair Value of Financial Instruments, which would be effective for financial statements ending December 15, 1994 (see footnote 11). This statement extended the applicability of disclosure requirements to instruments that did not have market risk, required identification of instruments held for trading purposes, and specified further classification of instruments, while encouraging but not requiring increased quantitative disclosures. Thus, during the period beginning in 1990, companies disclosed relatively more information about their actual derivatives positions in the footnotes to their annual financial statements. The propensity to provide quantitative and transparent information about the actual positions in swaps was reduced after 1999, when FAS133 (Accounting for Derivative Instruments and Hedging Activities) was initially set to come into effect. FAS133 was deferred by a year to 2000 under FAS137. FAS133 and subsequent statements laid down rules for an accounting treatment of derivatives that would incorporate them into the balance sheet (and earnings statements). They also removed the emphasis on provision of specific position information in the footnotes as long as risk management methods and accounting policies were clearly enunciated. Thus, the period offering distinct information on the companies' derivative transactions ended around Despite requiring or incentivizing disclosure between 1990 and 2000, the FASB rules were neither very prescriptive, nor very strict. This has allowed for a wide range of terminology and detail in disclosure. For instance, both a floating to fixed swap and a fixed to floating swap could qualify for hedge accounting as long as they were sufficiently matched to some underlying [34]

35 debt. In the former case, it might be claimed that interest payments were being hedged, while in the latter that the value of debt was being hedged. [35]

36 B. Our experience of practical issues related to collection of the data Undergraduate students were selected to collect the data as much for their finance or accounting background as for their ability to read and correctly interpret long, winding or complicated sentences. The selected students were then provided a refresher on the types of derivatives used by corporations and how to classify and distinguish those with similar terms. This classification was filtered as we moved along and came across the use of basis swaps and commodity swaps, for instance. During the actual coding process, the first task for each student was to take the randomly selected firms in sequence and check if there was a public record of financial statements available for the sample period in question. As the EDGAR filings are available only after 1994, we first relied on an old set of CDs containing monthly digests of financial results until we discovered that annual reports/10k filings were also available through the Lexis-Nexis database system. The lack of a precise standard for corporate statements implied we had to learn over the course of data collection and improve our processes. To achieve this, when initially starting with reports from the early years, we began with as wide a search as possible, reading the entire sections on debt, investments, financial instruments, and using wide search terms such as interest, interest rate, risk, and hedg apart from the standard terms like swap. To our knowledge, most papers on this subject did not incorporate such a wide range of search terms. We often needed to look at earlier or future years' statements to confirm our interpretation of the disclosures. This was true particularly for the debt data, for which we also added other search terms such as borrow and financ. In addition, short term debt was verified from the balance sheet directly in addition to the disclosures in the notes. [36]

37 The students entering the data were supervised or supported so that any questions they may have were immediately answered. Initially, at least two students worked independently on the same company, and their files were compared to ensure standardization. There was also an intermediate confirmation procedure in place so that they could cross-check their interpretation of disclosures with their neighbor as they worked together. The resulting sharing of observations led to a rapid learning process about where in the notes to look and how to record the data. Also, a student would work on one company at a time, taking advantage of any continuity in the reporting and cross-checking occasional typos with the following or previous years' reports. Despite all these checks and balances, the awkward and non-standard use of terminology and differences in emphasis on disclosure by firms led to challenges for slotting all the information into clear categories. Thus, where information was reported differently, students were required to insert a new column and record the data as originally classified by the company, and incorporating a remarks column if necessary. Simultaneously as these reports were prepared, we also attempted to retain a standard format for the data classifications, ultimately compiling all the firms into a standard coded database containing the key pieces of data with separate room for recording any remarks that were needed. [37]

38 Tables and Figures Table I: Descriptive Statistics on Average Firm Characteristics I summarize the firm characteristics used in the analysis. Each characteristic is calculated as the average over eleven years for the particular firm. Sales is item sale from the Compustat Industrial Annual file, Total Assets is at, while Market Value is defined as the product of common shares outstanding in millions (csho) and share price at the close of the fiscal year (prcc_f). Leverage is total liabilities/total assets calculated as the sum of dltt and dlc, divided by at. Research and development expense ratio is computed as xrd divided by total assets and capital expenditure ratio is capx similarly normalized. Quick ratio is defined as cash and short term investments (che) divided by current liabilities (lct). Cash flow margins are calculated as operating income before depreciation (oibdp) divided by sales, return on assets as income before extraordinary items (ib) divided by total assets, market to book ratio as the sum of total assets less common equity (ceq) and market value of equity together divided by total assets. The two cost ratios are Cost of goods sold (cogs) and S, G & A expense (xsga), each divided by sales. Variable N Mean Std. Dev. Min. Max. Sales Total assets Market value of Equity Leverage Share of long term debt Quick ratio R & D expense ratio Capital expenditure ratio Cash flow margin Return on assets Asset turnover Market to book ratio Cost of Goods Sold Sales, General & Admin Expense [38]

39 Table II: Operating profit (normalized by assets) and the level of pfix variation I regress firm cash flows on interest rates, based on whether the firms are classified as Active interest rate managers. The dependent variable is operating income before depreciation divided by book assets. The data frequency is quarterly in a panel format, with a dummy for Active and its interaction term with LIBOR used to determine, if on average, the cash flows and their sensitivities to interest rates are different from the Less Active group. p-values are provided below the estimates. Random Effects Fixed Effects Constant Lag LIBOR Active Active*LIBOR No. of observations No. of groups R 2 Within Between Overall Joint test Joint test statistic Wald(4) F(3,2825) Prob>chi [39]

40 Table III: Characteristics of Active and Less Active firms I report the mean and median of firm characterics for the entire sample and for the two categories of firms based on their interest rate exposure variability. Active firms have a standard deviation of the proportion of fixed rate debt (over the 11-year sample) that exceeds the 60 th percentile of this standard deviation across firms, and Less Active firms are those with the standard deviation of pfix below the 40 th percentile. The last two columns to the right contain differences between the means and medians of the two categories and p-values of t-tests and rank-sum tests respectively in italics below. The p-values below 0.05 are highlighted in bold. I II III II-III II-III N All Active Less Active Mean Median Sales Mean Median/p-value Total assets Mean Median/p-value Market value of Equity Mean Median/p-value Leverage Mean Median/p-value Long term debt share Mean Median/p-value Quick ratio Mean Median/p-value R & D expense ratio Mean Median/p-value Cash flow margin Mean Median/p-value Return on assets Mean Median/p-value Asset turnover Mean Median/p-value Capital expenditure ratio Mean Median/p-value Market to book ratio Mean Median/p-value Cost of Goods Sold Mean Median/p-value S, G & A Expense Mean Median/p-value N [40]

41 Table IV: Which firms are likely to be Active? I report probit regression analysis based on the average characteristics of the firms over the entire sample period. Please see Table I for a description of the variables. Estimates reported are the marginal effects estimated at the mean of the explanatory variables. Dependent variable --> Active = 1, else 0 Sales p-value Leverage p-value Quick ratio p-value Capital plus R & D expense ratio 0.46 p-value Cash flow margin p-value Return on assets p-value Asset turnover p-value Market to book ratio p-value Cost of Goods Sold p-value S, G & A Expense p-value N 57 Pseudo R [41]

42 Table V: The relationship between pfix and firm characteristics I repport the results of a beta regression of pfix on a range of firm characteristics. p-values based on robust standard errors for the location and precision parameters of the beta distribution are reported, along with the marginal effects at the mean of the explanatory variables. The estimates in bold script are statistically significant at the 5% level. µ ln ϕ Marginal effect* Estimate p-value Estimate p-value Estimate Std. Error Constant Sales Leverage Quick ratio Cash flow margin Return on assets Capital plus R & D expense ratio Market to book ratio Asset turnover S, G & A Expense Year effects Included Included Number of observations 678 Wald chi 2 (19) 75 Prob > chi 2 0 [42]

43 Table VI: Profiting from changes in pfix I list the strategy at the end of each year that would have been ex-post profitable in the following year. A negative change in pfix is profitable in most years except 1993, 1996, 1998 and In order to determine the appropriate strategy, we compare the rate at the end of a year with the average rate over the following year and the rate at the end of the following year. See also Figure 3b. At year-end Starting LIBOR Rate for following year Strategy float float float float fix float float fix float fix fix float [43]

44 Table VII: Are Active firms better at anticipating rates? I present a cross-tabulation of the variables Active and AllRight. 14 of the firms that were classified as Active (33) were also classified as having been Right (changing their exposur es in an ex-post profitable way) more often than 60% of the other firms. Active 0 1 Total AllRight Total [44]

45 Table VIII: Advantages to being Right I compare the change in operational characteristics and financial results in the year following the change in pfix based on whether the change was classified as Right. Shaded variables reflect a significant difference between Right and Wrong. % Δ Sales % Δ Total Assets % Δ Market Value of Equity Δ Leverage Δ Quick ratio Δ Cash flow margin Total Right Wrong Right-Wrong Δ Return on Assets Δ Asset turnover Δ Market to Book ratio Δ Cost of Goods Sold ratio Δ S, G & A Expense ratio Δ Short-term debt Total E Right E Wrong E Right-Wrong E [45]

46 Figure 1: Variation in proportion of fixed rate debt (pfix) and leverage This figure consists of three panels. In Panel A, I show that the distributions across firms of the proportion of each firm s debt that is subject to fixed rates and their leverage are fairly stable. A. Cross-sectional (across firms) variation over time The solid line represents the median across companies in each year, while the shaded region is the inter-quartile range around the median. The columns at the base of the chart represent the standard deviation across firms in each year. Note the scales are different, but the distribution across firms of both variables is fairly stable over the sample period. PFIX Leverage 1 Standard Deviation Median 0.5 Standard Deviation Median b. Distribution of time series variation for individual firms For each firm in the sample, I calculate the range (max - min) of the proportion of fixed rate debt and leverage, respectively, in the eleven-year sample period. I then plot a histogram of this range for the cross-section of firms. The median of this sample for pfix is 0.51, suggesting that most firms switch between having fixed and floating debt as the major component of their borrowings over this time period. The median of this sample is PFIX Leverage Frequency Frequency More Range (Max-Min over sample period) of share of fixed rate debt for individual firms More Range (Max-Min over sample period) of leverage for individual firms [46]