Capital structure, risk and asymmetric information

Transcription

1 Capal structure, risk and asymmetric information Nikolay Halov NYU Stern School of Business Florian Heider NYU Stern School of Business August 11, 2004 Abstract This paper argues that the standard pecking order hypothesis is only a special case of the adverse selection argument about external financing. It only applies when there is no asymmetric information about risk so that there is no adverse selection cost of debt. As soon as outside investors are imperfectly informed about risk, debt, a concave claim, will be mispriced. Using a large unbalanced panel of publicly traded US firms, we present robust and economically significant evidence i) that there is a general adverse selection in which firms issue consistently more equy and less debt if risk matters more and ii) that the special case of the pecking order, i.e. no adverse selection cost of debt, works well when risk does not matter, irrespective of firms age, size, market-to-book ratio, tangibily or the time period. We thank Heor Almeida, Dan Bergstresser, Kobi Boudoukh, Alexander Ljungqvist, Eli Ofek, Daniel Wolfenzon, Jeff Wurgler and seminar participants at NYU for helpful comments

2 The pecking order theory of capal structure, one of the most influential theories of corporate leverage, has recently fallen on hard times. On the one hand, the theory has considerable intuive appeal. Firms seeking outside finance naturally face an adverse selection and hence mispricing problem. In order to avoid mispricing, firms finance investments internally if they can, and if they cannot, the argument is that they prefer debt to equy since is argued that debt is less sensive to outside investors not knowing the value of firms investment projects (Myers (1984) and Myers and Majluf (1984)). Yet, the pecking order seems to work well empirically when should not and seems to not work well when should. Shyam-Sunder and Myers (1999) show that the pecking order is a good first order description of the time series of debt finance for large mature firms. It has however been argued that these firms should face ltle asymmetric information in capal markets. The pecking order cannot explain why young, small, nondividend paying firms, i.e. firms that supposedly should face large asymmetric information problems, issue equy. For example, Fama and French (2002) test the pecking order and compare to the main alternative, the trade-off theory. They find that the pecking order model beats the trade-off model: more profable firms have less book leverage. But they also find that the less levered nonpayers [of dividends] are typically small growth firms and that the least-levered nonpayers make large net new issues of stock [ ], even though they appear to have low-risk debt capacy. This is not proper pecking order behavior. Graham and Harvey (2001), Frank and Goyal (2003) and Leary and Roberts (2004) reach similar conclusions. We show that this tension can be resolved by recognizing that the standard pecking order is only a special case of the adverse selection argument of external financing. Stein - 1 -

3 (2003) for example points out that the same basic adverse selection argument that is used by Myers and Majluf (1984) for the equy market can be applied to the debt market, to the extent that the debt involved has some default risk. He quotes Stiglz and Weiss (1981) who in their theory of cred rationing use the adverse selection cost of debt that arises when lenders know the mean but not the variance of borrowers investments. A closer comparison of Myers and Majluf (1984) and Stiglz and Weiss (1981) reveals the following crical difference: the former implicly assumes that there is no asymmetric information about risk while the latter assumes there is only asymmetric information about risk. In fact, Myers (1984, fn.13) acknowledges that if there is asymmetric information about the variance rate, but not about firm value [ ], the pecking order could be reversed. 1 This comparison suggests the following. First, the standard pecking order is a special case of the adverse selection argument of external financing that holds only if there is no asymmetric information about risk. There is no reason to expect the pecking order to hold in suations where risk matters (and riskfree debt cannot be issued). In other words, debt is only optimal if has no adverse selection cost. Second, there is a more general adverse selection theory of external financing than the standard pecking order suggests. Since the standard pecking order is fully reversed so that equy is optimal in the oppose special case when only risk matters, one expects that not knowing risk creates an adverse selection cost for debt. The intuion is straightforward. Debt is a concave claim that is going to be mispriced by uninformed investors if risk matters, and the mispricing is more severe if risk matters more. 1 If there is asymmetric information about the variance but not mean of firms investments then these investments are mean preserving spreads. According to Rothschild and Stiglz (1970) mean preserving spreads capture differences in pure risk

4 This paper shows evidence in support of both claims: i) there appears to be a general adverse selection logic of external financing in which firms issue consistently more equy and less debt if risk matters more, and ii) the special case of the standard pecking order works indeed very well when risk does not matter irrespective of firms age, size, market-to-book ratio or the time period considered. Our empirical strategy builds on the analysis of Helwege and Liang (1996), Shyam- Sunder and Myers (1999) and Frank and Goyal (2003). Their tests are based on how firms finance their need for external capal. Using statement of cash-flow data, one constructs a measure of this need, the financing defic, and analyzes the sensivy of debt and equy issues wh respect to the financing defic. But whereas most of the existing lerature associates the adverse selection problem of external financing automatically wh the debt dominance of the pecking order, and thus implicly assumes that debt has no adverse selection cost, we test a more general adverse selection logic where debt has an adverse selection cost caused by outside investors being imperfectly informed about risk. Our approach is to condion the sensivy debt issuance wh respect to the financing defic on a measure of the role of risk in the adverse selection problem. Any measure of asymmetric information about risk must be an indirect one since something not known to investors cannot be in the econometrician s information set. We propose to capture the role of risk in the adverse selection problem by using firms recent asset volatilies. The idea is that an outside investor knows less about a firm s investment risk if the firm s asset value has fluctuated a lot. We use this measure non-parametrically by ranking firms into risk deciles and then running decile regressions of debt issuance on - 3 -

5 the financing defic. Our central hypothesis is that i) the sensivy of debt issuance wh respect to the financing defic decreases across risk deciles and ii) the sensivy is large in the lowest risk decile. Although recent asset volatily is going to be an imperfect proxy for not knowing risk, we present evidence that suggests s usefulness. First, there is more dispersion of asset volatilies whin higher risk deciles. Thus, s seems to reasonable to argue that investors know less about risk for firms in higher asset volatily deciles. Second, using option pricing data, we calculate the change of firms implied volatilies over time and verify i) that the market s assessment of future firm risk changes more of often for firms in higher risk deciles and ii) that our results go through using this alternative measure to rank firms. Third, we show that our measure of asymmetric information about risk has ltle impact on the sensivy of debt issuance for firms wh cred ratings. This is consistent wh the idea that cred ratings flatten the supply curve of capal (see Faulkender and Petersen (2004)) by bridging the informational gap about risk. Fourth, recent asset volatily does not appear to inadvertently pick up the probabily of default as measured by a firm s modified Z-score (see MacKie-Mason (1990)). We perform a series of robustness checks to see whether our empirical model is misspecified and whether alternative theories of the issuing decision can explain our results. We account for correlation of residuals across firms and time, include time and year fixed effects and control for the conventional cross-sectional trade-off determinants of leverage found in Rajan and Zingales (1995), who distill a large body of empirical research on the cross-sectional determinants of capal structure

6 To counter the argument that certain firms issue equy to finance their defic simply because they cannot issue debt, we control for debt capacy concerns. Although there is no formal theory of debt capacy, is argued that size, tangibily, age and the market to book ratio all affect a firm s abily to issue debt (Lemmon and Zender (2002)). 2 We find that the smallest firms and firms wh the highest market to book ratios in our sample rely more on equy finance. Yet is always the case that firms in higher risk deciles issue less debt to finance their defic. The interpretation is that the adverse selection of debt due to asymmetric information about risk is an important factor that lims firms debt capacies. Our results reduce the tension mentioned at the beginning between the theory of adverse section and existing empirical evidence that large mature firms issue debt and while young small firms issue equy. An outside investor presumably knows less about the risk of an investment if he faces a young small non-dividend paying firm than if he faces a large mature dividend paying firm. Hence, the former issue equy and the latter issue debt in order to minimize adverse selection costs of external financing. Linking capal structure to risk has been difficult in the past. The survey by Harris and Raviv (1991) shows that the evidence is mixed. Rajan and Zingales (1995) exclude measures of risk arguing that tradional measures of risk such as size or the volatily of earnings are too imprecise. Moreover, the standard argument of how risk affects capal structure is based on the trade-off between the tax benefs and the bankruptcy costs of 2 We view efforts to test for debt capacy as a way to let conventional trade-off determinants of the level of debt interact wh determinants of debt issuance such as the financing defic. The issue is closely related to i) the horse race between the standard pecking order and the trade-off theory and efforts to combine them empirically (see Hovakimian et al. (2001), Mayer and Sussman (2002) and Hovakimian et al. (2003)) and ii) the question whether there are target levels of leverage as predicted by the trade-off theory and if yes, what do firms do to reach them (see Welch (2003), Flannery and Rangan (2003) and Kayhan and Tman (2003))

7 debt. The tax-bankruptcy trade-off however seems unable to explain firms capal structures or issuing decisions. Graham (2000) finds that a large fraction of firms appears to forgo large tax benefs associated wh debt financing. This paper in contrast identifies a consistent link between risk and firms capal structure decisions using an adverse selection logic. The paper also contributes to recent efforts to develop and analyze dynamic capal structure models (Fischer et al. (1991), Strebulaev (2004) and Hennessy and Whed (2004)). These papers usually assume exogenous frictions such as flotation costs when firms issue securies. Our analysis suggests that information asymmetries could be an important alternative source of frictions. The organization of the paper is as follows. Section 1 illustrates the relation between the adverse selection cost of debt and not knowing risk wh a simple numerical example. Section 2 develops our empirical strategy. Section 3 describes the sample and presents some descriptive statistics. Section 4 contains the main empirical results. Their robustness and possible alternative explanations are analyzed in section 5. Section 6 concludes. 1. Risk and the adverse selection problem of external financing To illustrate that asymmetric information about risk creates an adverse selection cost of debt, we present a simple numerical example. Appendix A contains a characterization of the argument that goes beyond the stark numerical illustration below. Suppose that there only two types of firms, A and B, and that a firm is simply an investment project. Both types of firms require an investment of 100 to be raised from a competive outside capal market in order to undertake the project. Both types of firms - 6 -

8 have a chance that the investment fails, in which case the investment return is zero. Debt is a fixed future repayment F if the investment succeeds while equy is a share of the expected future returns. The return in the case of success is different across firm types. A type A project returns 300 in the case of success while a type B project returns 400. The expected net value for type A therefore is 1/2* =50, the expected net value for type B is 100. Note that the probabily of success or failure is the same for both types of firms. This can be thought of as there being no asymmetric information about risk. If someone did not know whether he was dealing wh a type A or a type B, e.g. an uninformed outside investor, he would nevertheless know that any type fails wh probabily 1/2. What he does not know is the return condional on success. Full information benchmark Let us first consider the benchmark suation when the outside capal market knows what type of firm is dealing wh. For type A the size of the future debt repayment is F A =200 since 1/2*F A =100. For type B the future repayment is the same, F B =200, since the probabily of repayment is the same for both types. If type A were to finance the investment outlay wh equy, would have to sell a fraction A=2/3 to the outside market since A*1/2*300=100. Type B however would have to sell less, B=1/2, because s investment project has a higher expected net value. Note that under this full information benchmark the Modigliani-Miller theorem holds. It does not matter how the investment is financed since both debt and equy allow a firm to always capture the entire expected net value of s investment, 50 for type A and 100 for type B

9 No adverse selection cost of debt Consider now the case when the outside capal market is imperfectly informed, i.e. cannot distinguish between a type A and a type B firm. As a result, the capal market demands a debt repayment or an equy share that reflects how likely is that the firm issuing the secury is of type A or type B. Assume that an outside investor is equally likely to face a type A or a type B firm. The debt repayment then is F =200 since 1/2*(1/2* F )+1/2*(1/2* F )=100 and the equy repayment is α =7/12 since 1/2*(α *1/2*300)+1/2*(α *1/2*400)=100. The debt repayment is exactly the same as under full information but the equy share is the average of type A s and type B s equy share under full information. Asymmetric information causes a mispricing, or adverse selection cost, of equy but not of debt. Type A s equy is overpriced while type B s equy is underpriced. 3 This mispricing of equy but not of debt is the essence of the pecking order hypothesis in which firms prefer debt to equy. As a solution to the adverse selection problem of external financing Myers (1984) proposes that firms issue securies whose future value changes least when the manager s inside information is revealed to the market. Here, the future value of debt is independent of the manager s inside information about the firm being a type A or a type B. It is always 100, the true cost of the investment. The future value of equy however increases by 18.6% when the market gets to know that the firm is of type B and decreases by 12.5% when the market gets to know that the firm is of type A. The capal market invested 100 in a firm in return for an equy stake of α =7/12. This stake 3 The underpricing of the high value firm and the overpricing of the low value firm could cause the former to drop out of the market earlier than the latter, leaving investors wh an adverse selection of low value firms (see Myers and Majluf (1984))

10 however is worth 7/12*1/2*400=700/6=118.6 if the firm turns out to be a type B and is worth 7/12*1/2*300=7175/2=87.5 if the firm turns out to be a type A. According to Myers argument, both types of firms should issue debt in this case where there is no asymmetric information about risk. Maximal adverse selection cost of debt We now change the probabily of success of a type A firm from 1/2 to 2/3. This slight change has drastic consequences since there is now asymmetric information about risk only. Due to the change, both types of investment now have the same expected net value of 100 and differ only in their probabily of success or failure. 4 Again, consider first the full information benchmark. For type B the debt repayment and the equy share remain unchanged at F B =200 and B=1/2 respectively. For type A the size of the future debt repayment now is F A =150 since 2/3*F A =100. Should a type A firm issue equy, would have to sell half of the firm: A=1/2 since A*2/3*300=100. By making the investment of type A more likely to succeed, we lower the cost of both debt and equy finance. When outside investors do not know what type of firm they are financing, they require a debt repayment of F =175 (since 1/2*(2/3* F )+1/2*(1/2* F )=100) or an equy stake of α =1/2 (since 1/2*(α *2/3*300)+1/2*(α *1/2*400)=100) in order to break even. Again, uninformed outsiders price a firm s securies on average knowing that there is a chance of facing eher a type A or a type B. Now is the debt but not the equy that is mispriced under asymmetric information. Equy finance now dominates debt finance 4 The two types investment returns now are a mean-preserving spread, a suation that captures pure differences in risk according to Rothschild and Stiglz (1971)

11 according to Myers logic that firms issues securies whose value changes least when the information asymmetry is resolved. The reason why debt now has a large adverse selection cost relative to equy is that there is asymmetric information about risk only. The example motivates the following observations. First, the standard pecking order, i.e. debt finance dominating equy finance, is a special case that is obtained under the assumption that risk plays no role in the adverse selection problem of external financing. Second, once we relax the assumption that there is no asymmetric information about risk, there will be an adverse selection cost of debt. The example shows that in the extreme, one can completely reverse the standard pecking order when risk plays a maximal role in the adverse selection problem. Since the same economic friction, i.e. outside investors being imperfectly informed about firms investments, yields very different implications about firms issuing decision depending on our assumption about how much outside investors know about the risk of firms investments, the issue becomes an empirical one. Discussion The example is an extreme illustration that the assumption about knowing or not knowing risk is central to the consequences of Myers argument about firms issuing information insensive securies. How robust is the intuion that asymmetric information about risk creates an adverse selection cost of debt? It is by no means a new observation that debt can be mispriced under asymmetric information (see Stiglz and Weiss (1981) or Myers (1984, fn. 13)). It has however been assumed in the empirical lerature that asymmetric information between firms and outside investors always implied the standard pecking order, i.e. that the adverse

12 selection cost of debt was negligible. Nachman and Noe (1994) demonstrate that the argument in the original analysis of Myers and Majluf (1984) using option pricing theory to show that debt is less sensive to mispricing is misleading. To obtain debt as the optimal secury under asymmetric information, one needs to assume that one can order firm types investments according to first-order stochastic dominance (the first case in our simple example when debt is not mispriced satisfies this condion). This condion does not hold if investment returns are distributed lognormally so that one cannot apply option pricing arguments to show that debt is optimal. An ordering of investment returns by condional first-order stochastic dominance means that uninformed outside investors can rank investments irrespective of their preferences towards risk. In other words, debt is optimal when risk does not matter. The fact that the standard pecking order is a special case that applies only when there is no asymmetric information about risk or, equivalently, when outside investors do not care about risk, has not attracted much attention when designing empirical tests of the pecking order. 5 The example focused on two polar cases: eher there is no asymmetric information about risk and debt has no adverse selection cost, or there is only asymmetric information about risk and debt has a maximal adverse selection cost. In the appendix we show that one can apply the intuion that not knowing risk creates an adverse selection cost of debt to the cases between the two extremes. A firm should issue more equy and less debt if risk plays a larger role the adverse selection problem of external financing. This reduces the 5 Some authors do point that the pecking order should only work when riskfree debt can be issued (for example Shyam-Sunder and Myers (1999)). Being able to issue riskfree debt however is a sufficient but not necessary condion that is unlikely to ever hold in practice

13 concavy of the claim so that there is less mispricing when outside investors know less about the risk of the firm s future cash-flows Empirical strategy This section presents and discusses our empirical strategy to test the adverse selection cost of debt. It builds upon the recent tests of the standard pecking order by Shyam- Sunder and Myers (1999) and Frank and Goyal (2003). Focusing on cash-flows Shyam-Sunder and Myers (1999) propose a test of the original pecking order based on how firms finance their need for external capal. A theory of capal structure based on asymmetric information at the moment at which a firm contacts the external capal market has a priori nothing to say about the level of debt, or leverage. The starting point is therefore the following accounting identy of cash flows: DEF = I + DIV + W C = D + E (4) A firm s financing defic DEF, i.e. the difference between uses of funds (dividends DIV, investment I and changes in net working capal W) and internal sources of funds (the internal cash-flow C), must be balanced by external sources of funds, i.e. eher the issuance of debt D or equy E (we follow the definions of Frank and Goyal (2003); see also Helwege and Liang (1996), Shyam-Sunder and Myers (1999), and Chang and Dasgupta (2003)). Since Shyam-Sunder and Myers (1999) and Frank and Goyal (2003) assume that the adverse selection problem of external financing automatically leads to the standard 6 We also show in the appendix that the mispricing of debt does not depend on the simplifying assumptions of our example such as there being just success and failure or there being just two types

14 pecking order in which debt dominates equy, i.e. pooled panel regression DEF = D, they run the following D = a + bdef + ε (5) and argue that there is support for the standard pecking order if a=0 and b is large. But we showed that the standard pecking order is a special case that applies only when there is no asymmetric information about risk. Thus, (5) is ill sued to test whether firms issuing decisions are driven by a fear of adverse selection. Once there is asymmetric information about risk, debt will have an adverse selection cost. In the extreme, when the asymmetric information is mostly about risk, the adverse selection cost of debt relative to equy will be large so that DEF = E. In that case would one expect to find a=0 and a very small b when running regression (5). We therefore employ (5) condionally by ranking firms into deciles, n=1,2 10, according to measures that proxy for the degree of asymmetric information about risk (we discuss these measures in the next section), and then run regression (5) separately in each decile n: D D = a + b DEF + ε (6) n n Our key hypothesis is that we expect the estimated coefficients on the financing defic to be ranked monotonically: ˆ > ˆ > ˆ. Less is known about risk for the group of D D D b1 b2 > b10 firms that have been ranked into higher deciles so that more equy and less debt should be issued to finance the defic in higher deciles. In addion to (6), we also test the extent to which equy is issued to finance the defic in each decile n:

15 E E = a + b DEF + ε (7) n n Since (4) is an accounting identy, checking that the estimated coefficients on the defic from (6) and (7) add up to one in each decile, ˆ D b bˆ = 1 for all n, is a useful test of the accuracy of the cash-flow data. We also expect that there is no constant non-zero factor other than the defic that drives the decision to issue debt and equy, a = aˆ = = aˆ 0. 7 Of course there could be n + E n ˆ = time- and/or firm-varying determinants of debt issuance other than the financing defic, e.g. the tangibily of a firm s assets. But before showing how we control for this, we should explain the creria used for ranking firms into deciles. Proxies for asymmetric about risk The hypothesis is that the outside capal market being uninformed about firms future investment risks drives up the adverse selection cost of issuing debt. Any measure of asymmetric information must be indirect since something that is not known to the market cannot be in the econometrician s information set. Recent asset volatily appears to be a useful and easily operational measure of asymmetric information about risk. Consequently, we use firms recent volatily of assets to group them into deciles and argue that the outside capal market knows less about the risk of investments for firms in higher asset volatily deciles. In other words, we expect that when raising external 7 Chirinko and Singha (2000) point out that running regression (5) on the entire sample when there is a significant amount of equy financing biases the coefficient b towards zero. This observation reinforces our argument that one should run (6), i.e. a condional version of (5). Note also that the counter-examples in Chirinko and Singha (2000) all have a ˆ 0. Checking whether a ˆ n = 0 is therefore an important element of our tests

16 financing, firms whose asset values have fluctuated a lot, face a higher adverse selection cost of debt than firms whose asset values have been stable. Of course, using recent asset volatily to group firms into deciles is going to be an imperfect measure of the role of risk in the adverse selection problem of external financing. Hence, we use a number of other measures to support our results. First of all, we show that that the dispersion of asset volatily is higher whin higher asset volatily deciles. It therefore seems reasonable to assume that uninformed investors know less about the investment risk of a firm when the firm s market value of assets has fluctuated a lot. Moreover, we use recent asset volatily only to rank firms. The actual amount of variance is not important to us, what matters is the comparison among firms. Second, we calculate the change over time of a firm s implied volatily using option pricing data (the details of this calculation are found in the appendix). Consistent wh our hypothesis we find that the volatily of the implied volatily of a firm is higher in higher asset volatily deciles. In other words, the market s assessment of future firm risk changes more often if a firm has higher current asset volatily. Our results hold if we use this alternative measure to rank firm. Recent asset volatily however is our preferred measure since the option price data is only available for a small number of firms over a short period of time. Third, ranking firms into deciles according to recent asset volatily has no impact on how the defic is financed for firms that have a cred rating, but has a strong impact on firms that don t have a cred rating. The former issue mostly debt irrespective of their recent asset volatily. This result is consistent wh the idea that cred ratings reduce

17 asymmetric information about risk and that this asymmetric information is captured adequately by our measure. Firms wh higher asset volatilies have characteristics that may reasonably be associated wh outside investors knowing less about the investment risk of these firms. They are smaller, younger, have higher market-to-book ratios, pay less dividends, have more cash and less tangible assets on their balance sheets. It is important to note that recent asset volatily does not appear to inadvertently pick up mere bankruptcy risk as measured by a firm s Z-score. Finally, in accordance wh our argument that the standard pecking order is a special case that applies only when there is no asymmetric information about risk, we find strong support for in the lowest asset volatily decile. In fact, once we limed ourselves to firms in the lowest risk decile, the support for the standard pecking order is stronger than in Shyam-Sunder and Myers (1999) irrespective of size, age or the time period that is being considered. Measuring recent asset volatily We construct two measures of asset volatily. The first one consists of unlevering the volatily of equy. Unlevering is needed since the volatily of equy mechanically increases wh leverage ceteris paribus. 8 We compute the standard deviation of the daily return on the market value of a firm. The market value of assets is defined as in Fama and 8 Note that by unlevering we make more difficult for us to find a negative relationship between risk and debt. We find that our results are indeed slightly stronger when using equy volatily

18 French (2002) (see also our appendix). 9 If there are less than 90 days of stock price data, the firm/year observation is deleted from the sample. The second measure recognizes that equy is a call option on the value of firm assets wh the exercise price being the value of the debt (Merton (1974)). From Ito s lemma, we have σ E = σ V Vt E t Et V t (8) where σ E is the instantaneous variance of the rate of return on equy (the standard deviation of daily stock returns from CRSP), σ V is the instantaneous variance of the rate of return on the firm (to be solved for), V t is the market value of the firm and E t is the market value of equy (both calculated as above). 10 The derivative of the market value of equy wh respect to the market value of the firm in the Merton model is: 1 2 E ln( Vt / Bt ) + ( rf + 2 σ V ) T t = Φ Vt σ V T (9) where Φ is the cumulative distribution function of the standardized normal distribution N(0,1), T is the time to matury of the debt (we try both 10 and 20 years) and r f is the risk free rate (from Kenneth French s webse). The Spearman rank correlation between the two measures of asset risk in our sample is The rank correlation is the appropriate measure since we use asset risk only to rank firms into deciles. Given that both measures give virtually identical rankings, we only 9 We also try the definion of Baker and Wurgler (2001), which excludes convertible debt, and also try using just total liabilies. The results are not affected. 10 An advantage of the Merton method is that we can use the CRSP return series that is adjusted for stock spls and dividends

19 report results using the simpler first measure (see also Jones et al. (1984) for a comparison of these two measures of asset volatily). To ensure that there is no contemporaneous interplay between the issue decision and asset volatily, we use last year s asset volatily. Using longer lags would weaken the link between the role of risk in the adverse selection problem and the current capal structure decision. 11 Controlling for other determinants of debt issuance An adverse selection model of firms capal structure decisions is based on information frictions at the moment when firms contact the external capal market. It uses a different set of variables than conventional, mostly cross-sectional empirical research on the level of debt that is usually rooted in the trade-off theory. The basic trade-off theory states that the level of leverage is determined by trading off the tax benef of debt against the cost of financial distress (see for example the account given by Myers (1984)). Hence, firms wh a high present value of tax benefs and/or a low present value of distress costs should have higher levels of debt (see also the classification in the survey by Harris and Raviv (1991)). Rajan and Zingales (1995) narrow the list of conventional determinants down to four main variables: profs, size, tangibily of assets and the market-to-book ratio. More tangible assets support debt because means that firms can collateralize the debt which reduces bankruptcy costs. The market-to-book ratio is usually seen as a proxy for growth opportunies that should be negatively related to leverage. The argument is that 11 There is an issue concerning the overlap or gap between the calendar year used for stock price data and the fiscal year used for financial data. This overlap or gap exists for 48% of all firms. We check the robustness of our results by using only firms whose fiscal year is the calendar year. The results are unchanged

20 leverage exposes firms to the debt overhang problem (Myers 1977). A recent alternative explanation for a negative relationship is market timing. Firms wh a high market-to-book ratio are overvalued and hence issue equy to take advantage of (Baker and Wurgler (2001)). Sales are usually posively associated wh leverage. There is no clear theoretical foundation but one normally argues that larger firms have a higher reputation or are safer so they can borrow more. Profs show up regularly as a negative determinant of leverage. Tradionally this has been seen as the strongest empirical challenge for conventional trade-off models of leverage. They predict that more profable firms should issue more debt since more profable firms have a smaller risk of bankruptcy and have more taxable income to shield (see Tman and Wessels (1988) and Fama and French (2002)). To see whether our adverse selection model falsely oms the conventional determinants of leverage, we follow Frank and Goyal (2001) and add first-differences of the conventional determinants to (6). This allows a nesting of the conventional determinants of leverage from the trade-off theory whin an adverse selection model. 12 The set of decile regressions (6) then becomes: D = a n + b + b DEF n LOGSALES n + b TANG n LOGSALES TANG + ε + b MTB n MTB + b PROF n PROF (10) If not knowing risk creates an adverse selection cost of debt then we expect the same monotonic ranking of the estimated coefficients on the financing defic across deciles once we add the conventional determinants of leverage. 12 Although using first-differences increases standard errors and biases the estimators towards zero, we nevertheless confirm the standard signs and statistical significance on the conventional variables in a version of (10) whout the financing defic and on the entire sample (i.e. all deciles together)

21 Controlling for debt capacy We are interested mainly in the relative ranking instead of the absolute level of the estimated coefficients on the financing defic across deciles. Thus, we allow for the possibily that there are other factors that affect the sensivy of debt and equy issuance wh respect to the financing defic. But once we control for those factors, should still be the case that ˆ > ˆ > ˆ, i.e. that firms in higher risk deciles issue D D D b1 b2 > b10 less debt to finance their defic. The idea is to allow for debt capacy constraints, i.e. to allow for lims to the use of debt other than s adverse selection cost. There appears to be no formal model or unanimous definion of debt capacy but seems reasonable to expect the same factors that have been found to determine levels of debt to be relevant for the sensivy of debt issuance (see Lemmon and Zender (2004)). In other words, while the previous section added the trade-off variables to adverse selection variables, we now also allow for an interaction of them. Before ranking firm into deciles according to recent asset volatily, we sort them into market-to-book, size, age or tangibily groups to allow for non-linear interactions of these factors wh the decision to finance the defic wh debt or equy. No matter how we perform this double sorting, we always find that firms in higher volatily groups issue less debt and more equy to finance their defic. Nevertheless, we also find evidence for some debt capacy concerns. In particular, the smallest firms and firms wh the highest market-to-book ratios in the sample issue less and firms wh the higher proportion of tangible assets issue more debt to finance their defic

22 3. Data Sample construction We study a large, unbalanced panel of all firms from the merged CRSP-Compustat (CCM) database from 1971 to Our sample only starts in 1971 since we mostly use cash flow data. We make the following standard adjustments. We exclude financial firms (SIC codes ), regulated utilies (SIC codes ), and firms involved in major mergers and acquisions (Compustat footnote code AB). Furthermore, we exclude firm/year observations that report cash flows data using format code (em 318) 4 or 6 (both undefined by Compustat) and 5 (for the Canadian file) or if the format code is missing. To be able to link Compustat reliably to CRSP data we use only records wh link type LC', 'LN', 'LO', 'LS', 'LU' or LX. A small number of CRSP securies that link into more than one Compustat firm have also been deleted. In order to remove outliers and misrecorded data, we remove observations for certain variables that have missing values or are in the extreme 0.5 % left or right tail of the distribution (see the appendix for the list of variables that have been treated this way). To ensure that the sample does not contain equy issues due to IPOs, we exclude observations for the year in which a firm s stock price becomes first available in the CRSP database. The maximum number of observations in our sample then is 103,351 firm-years

23 Descriptive statistics Table 1 shows balance sheets, cash flows and other descriptive statistics at the beginning and at the end of our sample period, 1971 and 2001, as well as for two intermediate dates, 1980 and Table 1: Balance sheets, cash flows and other descriptive statistics over time Panel A presents average balance sheets and panel B shows the average of the cash flows in the accounting identy (4). The key observation is that equy plays an important role in financing the defic. It contradicts the argument that most external financing uses debt (see also Frank and Goyal (2003)). 13 Note also the difference between the mean and the median of net debt and equy issues. The median is zero for both. A typical firm appears to stay out of the market for external finance most of the time, but if does seek external finance, the magnude of the market intervention is large relative to firm size. 4. Analysis The tradional pecking order The standard pecking order should not be a good description of debt issuance for all firms in the sample. It should only work well for those firms that have the smallest adverse selection cost of debt. 13 The table confirms that dividends are a disappearing use of corporate cash flows (see Fama and French (2001) and also Baker and Wurgler (2003)). A comparison of the average and the median dividend indicates that typical firms stop paying dividends and that those who continue paying them, nevertheless reduce the amount paid

24 The result from running regression (6) on the full sample is (pooled OLS standard error in brackets): Dˆ = DEF (0.000) (0.002) (R 2 = 0.36) (10) The coefficient on the financing defic is much less than the 0.75 (R 2 of 0.68) reported by Shyam-Sunder and Myers (1999) on a sample of 157 firms wh continuous reporting from 1971 to Our coefficient is only slightly larger than the 0.28 (R 2 of 0.14) reported by Frank and Goyal (2003) using an unbalanced panel from We therefore confirm the result of Frank and Goyal that the support for the standard pecking order in Shyam-Sunder and Myers does not carry over to a broader sample of firms. Our interpretation of this finding however is very different. While Frank and Goyal interpret as evidence against an adverse selection logic of capal structure decisions, we argued that the standard pecking order is only a special case so that one cannot expect to work for all firms. It should only work for those firms in the sample that have the smallest adverse selection cost of debt, i.e. those firms that face the least asymmetric information about risk. And indeed, we show below that this is the case irrespective of firm age, size, tangibily or the time period. 14 The slight difference seems to come from the fact that our requirement about the availabily of stock price data eliminates a number of small firms from the sample

25 Ranking by recent asset volatily In order to run the decile regressions (6), we rank firms each year into deciles according to their asset volatily in the previous year. Table 2 shows balance sheets, cash flows and other descriptive statistics across deciles. Table 2: Balance sheets, cash flows and other descriptive statistics across deciles Firms in higher deciles have more cash on their balance sheet whereas differences in tangibles and intangibles are small (panel A). As far as liabilies are concerned, firms in higher deciles have roughly the same amount of short-term and less long-term debt as firms in lower risk deciles. Comparing cash flows across deciles reveals a hump shaped pattern for dividends and internal cash flows (panel B). We also find that the median internal cash flow in the highest decile is larger than in the lowest decile (not shown in the table). The average financing defic of firms in higher deciles increases, but the median financing defic remains close to zero except for the three highest deciles. Average net debt and equy issues both increase for firms in higher deciles and the increase is more pronounced for equy. The medians however are mostly zero. This again indicates that a typical firm is reluctant to contact the external capal market, but if does raise external capal, the size of the intervention is large. Firms in higher deciles are younger, smaller and have higher market-to-book ratios (panel C). Profabily and unlevered Altman s Z-scores (see MacKie-Mason (1990)) first increase and then decrease across risk deciles. Firms in higher asset volatily deciles are therefore not less profable or more likely to go bankrupt than firms in lower deciles

26 Table 2 also shows that there is more dispersion of asset volatilies whin higher deciles and that firms in higher deciles have a larger variation of their implied volatilies. The central result Table 3 contains the central result of our paper. It shows the results from running regressions (6) and (7) in each decile. 15 Table 3: Financing the defic across deciles The table shows support for our hypothesis. Firms from higher deciles issue monotonically more equy and less debt to finance their defic. To illustrate the result, we plot the coefficients on the financing defic and the associated R 2 from Table 3 in Figure Figure 1: Financing the defic across deciles To get an idea of the economic significance, consider the impact of a one standard deviation change (9.3% of book assets) from the mean defic (0.5% of book assets) on net debt issues in the lowest decile. New debt issues increase from 0.1% to 8.1% of book assets which is about one standard deviation. In the highest decile, a one standard deviation change from the mean defic increases net debt issues by about a third of a standard deviation. 15 The table reports OLS standard errors. We also computed Whe standard errors that correct for heteroscedasticy. The corrected errors are about three to four times larger, which does not affect our conclusions. 16 Note that the estimated intercept is close to zero across all deciles. This suggests that there is no factor that is common to all firms in a decile throughout the sample period that could affect the pattern of net debt issues. Furthermore, the estimated coefficients on the defic from the net debt and the net equy regression add up to one across deciles. This indicates that we are not missing cash-flows

27 Note that the standard pecking order works extremely well in the lowest decile. The coefficient on the financing defic in the lowest decile is 0.87 (R 2 = 0.85). This is larger than the 0.75 obtained by Shyam-Sunder and Myers (1999) and Frank and Goyal (2003) when they look for the strongest support for the tradional pecking order. This supports the argument that the standard pecking order is a special case of an adverse selection logic of external financing that is obtained when risk plays no role. In Table 4 we show the proportion of companies that eher issue debt, equy or do nothing in each decile. 17 Table 4: Issue decisions across deciles The proportion of debt issues decreases across deciles while the proportion of equy issues increases, which lends further support for our hypothesis that firms in higher (lower) deciles rely more on equy (debt). Cred ratings bridging the information gap We spl the sample into two groups: firms wh an S&P cred rating and firms whout a cred rating. The hypothesis is that there is less of an asymmetric information problem about risk, and hence a lower adverse selection cost of debt, for firms wh a cred rating. The idea is that the service provided by rating agencies bridges the informational gap about risk between firms and the outside capal market. Moreover, rated firms are scrutinized closely by investors and analysts. Since we expect rated firms to face a 17 Issuing debt or equy is defined as a change in D or E that exceeds 1% of book assets. There are a lot of minor changes in equy due to the exercise of options or the conversion of other classes of stock into common stock

28 smaller adverse selection cost of debt, the coefficients on the defic should be constant across deciles. Table 5 and Figure 2 show that this is indeed the case. Table 5: Financing the defic of rated and unrated firms across deciles Figure 2: Financing the defic of rated and unrated firms across deciles Faulkender and Petersen (2004) show that larger firms wh more tangible assets are more likely to have a cred rating. Large firms and firms that have a lot of tangible assets are also likely to use more debt financing. To make sure that the result in figure 2 is not driven by the fact that rated firms in higher risk deciles are larger or have more tangible assets than unrated firms in the same risk decile, we spl the entire sample into firms wh and whout a cred rating and run the following regression on both subsamples: D = a + bdef + b TANG DEF + b RISK * TANG DEF + ε * LOGRISK + b SIZE DEF * LOGSIZE (11) The regression allows the coefficient on financing defic to depend on i) recent asset volatily (the same measure that we use to rank firms into deciles we take the natural logarhm since this variable is heavily skewed), ii) size (the book value of assets) and iii) the tangibily of assets. 18 The results are shown in Table 6: 18 We also controlled for size and tangibily non-parametrically by sorting firms into 25 size-tangibily groups, then whin each of these 25 groups running regression (6) for rated and unrated firms across risk deciles (i.e. 25*2*10=500 regressions) and finally averaging the 25 coefficients on the defic to obtain a version of figure 2 that controls for size and tangibily. The result is very similar to figure 2. The problem wh this method is that there only very few observations in some of the 25 groups

29 Table 6: The impact of risk on the defic coefficient for rated and unrated firms The coefficient on DEF*LNRISK is negative but is more than twice as large for unrated than for rated firms. 19 This is further support for the argument that rated firms have only a small adverse selection cost of debt. Summary Overall, the data is consistent wh our hypothesis about an adverse selection logic of capal structure where asymmetric information about risk creates an adverse selection cost of debt. The variation in the financing defic therefore explains more the decision to issue equy and less the decision to issue debt for firms from higher asset volatily deciles, where risk plays a larger role in the adverse selection problem. In addion, the proportion of firms issuing equy increases in higher deciles while the proportion of firms issuing debt decreases. There is no strong monotonic pattern in the coefficient on the financing defic across deciles for firm wh a cred rating, presumably because these firms face a less severe asymmetric information problem about risk. Finally, the standard pecking order, which is a special case that applies only when there is ltle asymmetric information about risk, works indeed very well for firms from the lowest risk decile. Once we lim ourselves to the firms from the lowest risk decile, we find even stronger support for the standard pecking order than Shyam-Sunder and Myers (1999). 19 LNRISK has roughly the same mean and standard deviation for rated and unrated firms so that one can compare the regression coefficients across different subsamples