Does dynamic tradeoff theory explain high-frequency debt issuers?

Transcription

1 Does dynamic tradeoff theory explain high-frequency debt issuers? B. Espen Eckbo Michael Kisser January 2015 Abstract We contrast funding policies and leverage dynamics of high versus low-frequency net-debt issuers (HFIs and LFIs). These are defined, respectively, as publicly traded industrial companies in the top and bottom quartiles of the annual net-debt issue distribution (net of debt retirements). Firms tend to remain in their initial quartile for years after public listing, suggesting that HFIs persistently draw benefits from debt financing in excess of debt issue costs, which LFIs do not. HFIs are large, highly leveraged growth firms with intensive capex programs. Classical dynamic tradeoff theory suggests that, relative to LFIs, HFIs should undertake smaller net-debt issues, have less volatile leverage ratios, and greater speed-of-adjustments to target leverage ratio deviations. We find little support for these predictions. Overall, net-debt issue frequencies are driven mainly by firms capex programs, sometimes with transitory debt (debt issues when overleveraged). Leverage ratios are lowered following transitory debt issues through equity issues and retained earnings rather than by debt repurchases. We have benefitted from the comments and suggestions of Harry DeAngelo, Pierre Chaigneau, Michael Hertzel, Michael Roberts and Karin Thorburn, and seminar participants at the 2013 meetings of the Financial Management Association, the European Finance Association, the SFS Cavalcade, the Southern Finance Association, Concordia University, the Norwegian School of Economics, Tuck School of Business at Dartmouth, University of Adelaide, University of Bristol, University of Stavanger Corporate Finance Conference, and the Vienna Graduate School of Finance. This project has received funding from Lindenauer Center for Corporate Governance at the Tuck School of Business. Tuck School of Business, Dartmouth College. Norwegian School of Economics.

2 1 Introduction It is well established that the distribution of leverage ratios among public nonfinancial U.S. firms has a persistent and sizable left tail consisting of firms with zero or near-zero debt (Strebulaev and Yang, 2013). What is much less known, however, is the right tail of the distribution: the characteristics and dynamic leverage behavior of firms that persistently fund themselves by raising debt. We show that such firms exist and persist in their high-frequency debt-issue behavior over much of their lifetime as public firms. Moreover, we use these high-frequency debt issuers to provide new empirical perspective on several of the basic predictions of the class of dynamic tradeoff theory pioneered by Fischer, Heinkel, and Zechner (1989) and extended to include investment by DeAngelo, DeAngelo, and Whited (2011). All theories of debt financing presume the existence of some form of debt benefits (whether from taxes, low issue costs, agency benefits, etc.). Public companies that have been observed to persistently fund themselves with new net debt issues (net of debt retirements) therefore form a relatively powerful place to test these theories. That is, in contrast to zero- or near-zero leverage firms, we know that high-frequency net-debt issuers view debt financing as sufficiently beneficial (for whatever reason) to cover issue costs most of the time. For these firms, the remaining issue is whether the dynamic behavior (frequency and size) of net-debt issues and of leverage ratios conform to to the theoretical predictions. In dynamic capital structure theory such as Fischer, Heinkel, and Zechner (1989), issuance costs directly influence the frequency and size of capital structure adjustments. Temporarily underleveraged firms with high fixed debt issue costs remain inactive for long spells before issuing large amounts to restore leverage targets. In contrast, underleveraged firms with largely proportional issue costs issue more frequently they become high-frequency issuers and they issue relatively small amounts each time (Leary and Roberts, 2005; Korteweg and Strebulaev, 2014). This cost-optimizing behavior further implies that high-frequency debt issuers will exhibit low leverage ratio volatility relative to low-frequency debt issuers. Moreover, high-frequency net-debt issuers are expected to more quickly revert to an optimal leverage ratio over time exhibiting relatively greater speed-of-adjustment back to leverage targets in the vernacular of Fama and French (2002), Flannery and Rangan (2006) and others. We address these predictions using publicly traded industrial firms over the period , while systematically benchmarking the capital structure decisions of high-frequency net debt issuers with that of low-frequency issuers. 1

3 We define high-frequency net-debt issuers (henceforth HFIs) as firms in the upper quartile of the annual net-debt issue frequency distribution distilled from cash-flow statements (which record private and public debt issues from all sources). Low-frequency net-debt issuers (LFIs) are firms in the lower quartile of this distribution. We eliminate debt rollovers (and use net debt) because we are primarily interested in theoretical implications for dynamic debt changes. Optimal debt levels are captured by the usual factors driving empirical cross-sectional leverage targets (Parsons and Titman, 2008; Frank and Goyal, 2009). Furthermore, we require our sample firms to go public during the sample period. Conditioning the issue frequency distribution on firm age since public listing avoids the confounding effect of pooling older firms with newly listed companies, where the latter have not yet had much time to issue securities. In the process, we also provide new and interesting evidence on the lifecycle funding behavior of public industrial companies from the year of public listing. As it turns out, our simple issue-frequency classification works well as it reveals a substantial spread between HFIs and LFIs, both in terms of firm characteristics and issue dynamics. For example, across firm-years, leverage ratios average as much as 35% for HFIs versus only 10% for LFIs. Moreover, while 32% of the LFIs have zero leverage in a typical firm-year, only 1% of HFIs ever reduce leverage to zero in a firm-year. Also important, the composition of firms classified as HFIs and LFIs tend to persist over time: firms classified as either HFIs or LFIs early on are highly likely to retain that classification throughout their lifecycle as public companies. Who are the high-frequency issuers? They are large relative to LFI, both in terms of sales and total assets. More important, they have significantly higher annual average capital expenditures (capex), which results in a substantially higher asset- and sales growth. Capex is by far the most important firm characteristic predicting the net-debt issue probability and the time between net-debt issues (the issue hazard). Notwithstanding their relatively intensive capex program, HFIs exhibit significantly lower Tobin s Q and lower research and development (R&D) expenses than LFIs. In terms of corporate funding, HFIs and LFIs receive a similar contribution from operating profits, averaging about 40% of all sources funds. HFIs rely much less on sale of equity and assets, making up the difference largely in the form of debt issues. For both HFIs and LFIs, the average combined proceeds from the sale of equity and debt securities total external finance is approximately the same as proceeds from total asset sales (illiquid assets plus cash draw-downs and reductions in net working capital). In a 2

4 typical firm year, these two sources of funds (sale of securities and total asset sales) each average about 30% of total funds (median about 15%). Our evidence suggests that costly asset sales are higher up in the financing pecking order than previously anticipated by the capital structure literature. This overall funding pattern appears soon after public listing. Under dynamic tradeoff theory, the greater the fixed component of issue costs, the longer the inactive periods and the greater the net-debt issue when the firm recapitalizes. This suggests that the net-debt issue size of LFIs should on average be larger compared to those of HFIs. The data tells a different story, however, as the issue sizes of these two categories of firms are roughly equal. Second, the leverage ratio variability is higher for HFIs than for LFIs and not lower as expected. We suspect HFIs drive much of the firm-level leverage instability highlighted by DeAngelo and Roll (2013) as net-debt issuers below the top quartile have both low and relatively stable leverage ratios. Third, when using net-debt issues as the dependent variable in the estimation of speed-of-adjustment (SOA) to target leverage deviations, we fail to find a higher SOA coefficient for HFIs than for LFIs. 1 These results fail to support the dynamic tradeoff theory of Fischer, Heinkel, and Zechner (1989). To better understand this failure, we investigate several additional issues. The most important is the role of investments in driving the timing and size of net-debt issues, and which may mask tradeoff behavior. In particular, DeAngelo, DeAngelo, and Whited (2011) shows that overleveraged firms may issue transitory debt to finance investment. The debt is transitory in the sense that it is optimally retired following the investment shock. Interestingly, we do find that overleveraged firms issue debt to finance capex. However, we do not observe the predicted debt retirements. Rather, firms adjust down leverage ratios by issuing new equity and/or retaining profits. In sum, HFIs are high-grow firms funding a relatively aggressive capex program with debt. The timing of net-debt issues appears to be largely driven by the arrival of investment projects rather than by the type of recapitalization events suggested by classical tradeoff theory. The rest of the paper is organized as follows. Section 2 identifies HFIs and LFIs and their firm characteristics. Section 3.2 describes how HFIs and LFIs differ in their overall funding mix, followed by our tests of dynamic tradeoff theory in Section 4. Section?? discusses alternative hypotheses and possible reasons for our rejection of the tradeoff theory, while Section 6 concludes the paper. 1 The extant literature estimates SOA coefficients using leverage ratio changes as dependent variable. Leverage ratios change in response to active net-debt issues and retirements as well as to passive asset growth (Welch, 2004). Our estimation concentrates on the active (net-debt issue) part. 3

5 2 Lifecycle net-debt issue frequencies 2.1 Sample selection Our sample consists of 12,131 nonfinancial U.S. domiciled corporations and an unbalanced panel of 93,031 firm-years from the period To arrive at this sample, we start with the merged Crisp/Compustat (CCM) database (22,853 firms and 250,053 firm-years) and then impose seven sample requirements in the following order: U.S domiciled firms only (eliminates 2,158 firms and 20,473 firms-years); nongovernmental industrial firms only (eliminates 5,593 firms and 65,011 firm-years for utilities (SIC codes ), financial firms (SIC ), and government entities (SIC above 8999); no change of fiscal year (-0 and -1,817); no missing book values of total assets (-16, -449), firm age positive (-382, -5,463) 2 ; consistent cash flow statement data (-333 and -3,888); 3 consistent balance sheet data (-57, -2,528); 4 firm must go public during sample period (-2,411, -39,449), contiguous balance sheet data only (-0; -16,944). Table 1 displays formal definitions of the main variables employed in this paper. 2.2 Annual issue frequencies Table 2 sorts the sample firms by firm age since public listing (year 0) and displays annual cumulative number of net-debt and gross equity issues using two issue size thresholds: 2.5% and 5% of the current book value of assets, respectively. For each size threshold, the table shows the frequency of the population average firm (mean), the firm in the 25 th percentile firm (p(25)), the median firm (p(50)), and the firm in the 75 th percentile (p(75)). Throughout the paper, we identify HFIs (LFIs) as the firms in the top (bottom) quartile of the net-debt issue frequency distribution. 5 The average issue (or retirement) frequency of both HFIs and LFIs is also displayed in the Table. Panel A lists the net-debt issue frequency (NDI + ) while Panel B shows the equity issue frequency (EI). 2 Firm age is defined as the difference between the reporting date of the financial statement and the date of the first month a company is reported in the CCM monthly stock price database. The reported age is rounded to the next smaller integer. 3 We first drop observations with negative values for the following Compustat variable names (see Table 1 for variable definitions): dltis, dltr, sstk, prstkc, dv, capx, aqc, ivch, sppe and siv (eliminates 3,493 firms). We then set missing entries for items in the cash flow statement to zero and drop observations in case total sources or uses of funds equal zero (eliminates 1,165 firms) or deviate by more than 1% from each other (eliminates 230 firms). 4 We set missing entries for deferred taxes and investment tax credit (txditc) and preferred stock liquidation value (pstkl) equal to zero and subsequently require non-missing data for the market value of the firm s equity (prcc f csho), Tobin s Q (lt + pstkl - txditc + prcc f csho)/at), total debt (dltt + dlc), cash holdings (che), property plant and equipment (ppent) and further drop observations in case the book leverage ratio is outside the unit interval or cash holdings are negative. 5 Each firm is reclassified annually over the lifecycle following public listing as HFI or LFI. 4

6 Consider first the central tendency (mean and median) in the issue frequency distribution. According to Panel A, firms that have been publicly traded for ten years on average undertake 3.1 net debt issues exceeding 2.5% of current book assets (median 3). Firms that have been listed for twenty years make 5.8 issues on average (median 6). With a 5% threshold the average and median number of issues is 2.4 and 2 after ten years of listing, and 4.3 and 4 after twenty years, respectively. Thus, firms typically undertake a single positive net-debt issue exceeding 5% of total assets every five years. 6 The annual frequency in Table 2 is about half the frequency reported by Leary and Roberts (2005) based on quarterly data over the period , implying that debt issues often cluster within a single year. 7 Since the annual frequency averages roughly once every five years (Panel A of Table 2), it follows that quarterly issues tend to cluster with roughly two in a given year on average. Welch (2004, 2013) factors out annual movements in market leverage ratios caused by stock returns from total changes in market leverage ratios. This isolates what he calls the active managerial capital structure adjustment embedded in leverage ratio changes, and which occurs with a greater frequency than what is reported in Table 2. However, our direct evidence in Panel A of Table 2 shows that the actual annual net-debt issue frequency is lower than what he reports indirectly. Next, we turn to HFIs and LFIs: firms above the 75 th and below the 25 th percentiles, respectively. After ten years of public listing, and using the 5% issue threshold, HFIs have made on average 4.9 net debt issues. This exceeds the average frequency of LFIs by a factor of ten (the average net debt issue frequency for LFIs is 0.49). After twenty years of listing, HFIs have made 7.6 issues and LFIs 1.1 issues. This difference in relative issue frequency is slightly increased if we reduce the issue-size threshold to 2.5% of current assets. Panel B shows the frequency distribution of equity issues, which on average occur less frequently than net debt issues. The low number of large equity offerings is consistent with prior studies (Eckbo and Masulis, 1995; Fama and French, 2005b; Eckbo, Masulis, and Norli, 2007; Leary and Roberts, 2010). However, the lifecycle issue-frequency information is new to the literature. With the 5% threshold, after 6 Untabulated results further show that net-debt retirements are only slightly less frequent than net-debt issues: with a 5% threshold and after ten years of public listing, the average firm has undertaken 1.6 net-debt retirements (median 1). After twenty years of listing, the average number of net-debt retirements is 2.8 (median 3). Thus, firms typically undertake a net-debt retirement exceeding 5% of total assets every seven years. 7 Leary and Roberts (2005) report that firms on average issue debt once every 8 quarters. They restrict their sample firms to have at least 16 quarters (4 years) of contiguous data, which cuts the sample in nearly one half relative to our requirement of only four contiguous quarters (one year). Our less restrictive sample brings in younger and less established companies relative to theirs. The average firm in Leary and Roberts (2005) has approximately 36 quarters of data and issues net debt 4.2 times (on average one net-debt issue every 8.57 quarters). 5

7 ten years of listing, the bottom quartile have issued equity only once, the median firm twice and the top quartile three times. After twenty years, these three categories of firms have issued equity exceeding 5% of total assets once, twice and four times, respectively. Interestingly, the cumulative number of equity issues is largely similar for LFIs and HFIs. Reducing the issue threshold from 5% to 2.5% only marginally increases the number of equity issues regardless of the age since listing. 2.3 Issue-frequency persistence Table 2 indicates that the annual differences between HFIs and LFIs tend to persist across listing age and across issue-size thresholds. That is, although the typical firm is relatively inactive in terms of issuing positive net debt, HFIs are highly active regardless of age. Table 3 further indicates a high degree of persistence of HFIs (Panel A) and LFIs (Panel B). The table presents various backward and forward looking persistence measures. For example, column (1) in Panel A shows that 100% of the firms that were classified as HFI after five years of listing were also classified as HFI one year earlier, 88% were classified as HFI two years earlier, and 65% were classified as HFI three years earlier. For firms that were classified as HFI after ten years of listing, the corresponding percentages are 100%, 92% and 82%, respectively. 8 Turning to the forward looking measures, column (6) shows that, after five years of listing, only 1% of all HFIs reduce their subsequent issue frequencies sufficiently to ever be classified as LFI in the future. Furthermore, there is a 70% chance of staying HFI (with the difference of 29% becoming medium frequency issuers). After ten years of listing, these percentages are 0%, 81% and 19% respectively. Panel B also shows that the composition of LFIs hardly changes over time, with the only exception being rebalancing years. For example, five years following public listing the threshold for being classified as a LFI increases from 0 to 1 (Table 2), which changes the portfolio composition of LFIs such that, in this year, only 59% of LFIs were also classified as low frequency in the previous year. However, with the exception of those events, the issue frequency classification is highly persistent and this is also reflected in the forward looking persistence measures. Column (6) shows that only 2% of all LFIs increase subsequent issue frequencies sufficiently to be classified as high frequency issuers in the future. Eighty-two percent of the LFIs stay low frequency, while 16% become medium frequency issuers. 8 These percentages tend to shift only in the years in Table 2 showing a discrete increase in the cumulative number of issues. 6

8 3 Who are the high-frequency issuers? 3.1 Distinguishing characteristics of high-frequency issuers Table 4 displays descriptive information for HFIs (Panel A) and LFIs (Panel B) sorted by the year since public listing. There are several interesting differences between the two issuer types. First, LFIs are much less leveraged and have higher cash balances than HFIs. Focusing on the grand average values at the end of each panel, the market leverage ratio (L) in column (1) is 35% for HFIs and 10% for LFIs. Book leverage ratios in column (2) are virtually identical. This difference in average leverage ratios is also reflected in column (3) which shows the fraction of the sample firms that are all-equity financed (AE): it is a much as 32% for LFIs and 1% for HFIs. Moreover, the cash ratio C in column (6) is 31% for LFIs and 10% for HFIs. Second, the asset structure and growth rates also differ substantially among the two issuer types, with LFIs being smaller firms with greater Tobin s Q and lower asset tangibility. The book value of total assets (assets) in column (7) is $345 million for LFIs and $804 million for HFIs, while Q in column (10) is 2.56 versus 1.80 for the two issuer types, respectively. The greater Q value for LFIs is also reflected in higher research and development spending in percent of total assets (RD in column (11)), which is 8% for LFIs and 3% for HFIs. On the other hand, HFIs have greater capital expenditures (capex in column (12)): 10% of total assets versus 6% for LFIs. The greater investment rate of HFIs also shows up in the growth rate of both total assets (g A ) and total sales (g S ) in the last two columns. 3.2 Overall funding policy of high-frequency issuers This section presents evidence on the overall funding pattern of HFIs relative to LFIs. This includes the sale of equity as well as debt securities, reduction in working capital, cash draw-downs and and sale off assets. This description, which is new to the capital structure literature, serves to solidify the notion that HFIs differ from LFIs primarily in the debt-financing dimension. For this purpose, we calculate so-called financing ratios. The financing ratio R j represents the non-negative funding source S + j and is defined as: R j S+ j 7 i S+ i, (1) 7

9 where the denominator sums over the following seven available financing sources in the cash-flow statement: 7 S i + = CF + + EI + NDI + + C + I + W + O +. (2) i CF + is the positive portion of operating cash flow; EI is proceeds from equity issues; NDI + is positive net debt issues (debt issues exceeding debt retirements), C is draw-down of cash balances; I is asset divestitures including sale of property, plant and equipment (PPE); W is reduction in net working capital; 9 and O + is a small residual that maintains the cash flow identity. Table 1 shows how each of these variables are defined using Compustat cash-flow statement mnemonics, while Table 5 lists the annual values on each of these funding ratios and their components. 10 Financing ratios are interesting as they, unlike leverage ratios, measure the evolution of capital structure using market values (cash flows) and not book value. Moreover, the ratios themselves are not confounded by asset growth. While changes in asset growth drive changes in equity values and leverage ratios even when managers do not actively rebalance capital structure, our financing ratios respond exclusively to managerial issuance activity. The detailed evolution of the individual financing ratios is shown Table 5. To simplify the exposition, Figure 1 aggregates the contribution of liquid asset sales (cash draw-downs and reductions in net working capital) a single Liquid Asset Sales ratio: R AS ( C + W + O + )/ 7 i S+ i shown in the figure are, respectively, the Net-Debt Issue ratio R NDI + NDI + / 7 i S+ i ratio R EI EI/ 7 i S+ i, positive Operating Cash Flow ratio R CF + CF + / 7 i S+ i Sales ratio R I I / 7 i S+ i. By construction, these five ratios sum (vertically) to one.. The other three ratios, the Equity Issue and the Illiquid Asset Panel A of Figure 1 illustrates that net debt issues constitute a significant source of funds for HFIs. Even in the year of public listing, net debt issues are as large as equity issues (the contribution of net debt issues is 33%, equity issues account for 36% of all sources of funds). However, and this is contrary 9 In 1988, Statement of Financial Accounting Standards (SFAS) instituted a new and uniform reporting system for working capital, including its component assets and liabilities. We work with net working capital over the entire sample period. Separate analysis on the post-1988 period shows that splitting net working capital into assets and liabilities does not affect our main conclusions below. 10 Debt and equity issues derived from the cash flow statement may differ from balance sheet induced changes of debt and equity. This happens when a transaction does not have cash flow implications. Examples of such transactions include the use of stock as a payment method in take-overs, the consolidation of debt and equity in acquisitions, exercise of convertible securities and stock option related compensation policies. Computing the difference between net debt issues and balancesheet implied positive changes in debt, shows that this effect is small: the mean (median) difference (scaled by assets) is 0% (0%). For equity issues, the distribution is slightly more skewed with a mean (median) of -4% (0%). One of the main contributions of this paper is to disentangle the cash flow effects of security issues from other changes in debt and equity and we control for these effects in the dynamic target leverage ratio regressions below. 8

10 to equity issues, they continue to be highly significant in the following years, raising close to 30% of all sources of funds. For LFIs, shown in Panel B, the story is different. The contribution of net debt issues is zero in the year of public listing, whereas equity issues account for 60% of all funding sources. Over the entire life cycle, the average contribution of net debt issues is merely 2%. whereas the median contribution is zero. While the external funding policy thus differs strongly across the two groups of firms, Table 5 shows that the contribution of operating cash flow is comparable and equal to 40%. 3.3 Investment as a predictor of net-debt issues Table 6 presents estimates of the determinants of becoming HFI after T years of listing in Panel A, and of the time between successive net-debt issues (the issue hazard or financing spell) in Panel B and C. The probit estimation in Panel A uses firm characterists in the year of public listing in the following model: Y it = α + βx i0 + ɛ it (3) Y it = 1 if Y it 75 th percentile and 0 otherwise where Y it is the latent variable for the probability of firm i being a HFI after T years of listing, and Y it is the dummy variable for Y it. The vector X i,0 of firm characteristics is the same as in Table 4 but now measured in the year of going public (year 0). In addition, the estimation of (3) includes industry dummies for eight of the 12 Fama-French (FF12) industries (excluding financial firms and regulated utilities). Panel A of Table 6 shows the parameter estimates for forecasting periods of 3, 6, 9, 12 and 15 years following public listing. These estimates strongly suggest that the issue frequency classification far out in time is predictable based on year-0 values of several of the characteristics. Moreover, the most powerful characteristics increasing the probability of becoming a HFI are initial Capex and leverage ratio L. Moreover, initial R&D and cash balance C significantly reduce this probability. 11 In Panel B of Table 6 we use hazard analysis to investigate whether corporate investments also drive the time between successive net-debt issues (net-debt financing spells). Moreover, we expect this effect to differ for LFIs and HFIs. That is, if the arrival process of new investment projects requiring external 11 The average marginal effects on the probability are as follows: a 10 percentage point increase in Capex (L) increases the probability of being classified as a HFI nine years following public listing by 7.4 points (3.8 points). A similar increase in R&D decreases the probability by 4.8 percentage points. 9

11 finance is relatively slower for LFIs than for HFIs, then the sensitivity of the issue hazard to investment should be greater for LFIs. As in Hertzel, Huson, and Parrino (2012), we estimate a standard exponential hazard model model of the form: h i = h 0 exp(β 0 + βx i )α i, (4) where h 0 is the baseline hazard (when all covariates are equal to zero and assumed constant in Panel B), and α i captures unobserved heterogeneity analogous to a regression error term. 12 The firm characteristics (x i ) now enter after subtracting the median value across all firms each year (Leary and Roberts, 2005). 13 Moreover, we perform the hazard rate estimation separately for LFIs and HFIs. In Panel B, a hazard ratio which is statistically indistinguishable from unity means that the control variable does not change the likelihood of the financing event taking place the following year (up or down). The tabulated results confirm the findings in the Panel A: Capex again has a strong and significant impact on net-debt issue decision. Specifically, a unit increase in capital expenditures (relative to the median firm) raises the issue hazard by a factor of seventeen for HFIs and a factor of nineteen LFIs. In other words, for both HFIs and LFIs, investment and net-debt issues tend to be driven by investment funding needs. Panel B also shows that the availability of internal funds either C or prof reduces issue hazards with similar marginal effects for LFIs and HFIs. The exponential hazard model can easily be adapted to investigate the role of time on issue hazards by simply extending the set of control variables, h i (t) = h 0 exp(β 0 + βx i + γf(t))α i and f(t) = t + t 2 + t 3 (5) We follow Leary and Roberts (2005) and parameterize the function f(t) as a cubic function of time (f(t) = t + t 2 + t 3 ) and then fit this function to our data on actual net debt issue activity. As shown in Panel C, this estimation exacerbates the difference in the impact of Capex in the issue hazards of LFIs and HFIs. This effect reflects both the persistently lower investment activity for LFIs and the use of net-debt issues to fund new investment projects as they materialize. 12 Pr(t T <t+m T t) If T is a random variable measuring the time between issues, h(t) = lim m 0, where h(t) is the instantaneous rate at which a firm issues net debt conditional on not having done so for t periods. For example, h(t)m at t = 5 m gives the probability that a firm will issue over the next m periods, conditional on not having done so for the last five years. 13 This also implies that the baseline hazard is ĥ0 = exp( ˆβ 0) for the median firm. 10

12 4 Does dynamic tradeoff theory explain high-frequency debt issuers? Classic dynamic trade-off theories separate financing and investment decisions (Fischer, Heinkel, and Zechner, 1989; Goldstein, Ju, and Leland, 2001; Strebulaev, 2007). In those models, there is no room for externally financed investment and recapitalizations are thus purely driven by trade-off considerations. Observed issue frequencies reflect the magnitude and functional form of debt issuance costs. In general, higher costs increase the period of optimal inactivity and the existence of fixed costs further slows down issue frequencies. 14 Dynamic financing and investment models incorporate investment decisions into a dynamic tradeoff framework (Hennessy and Whited, 2005, 2007; DeAngelo, DeAngelo, and Whited, 2011). Trade-off considerations are relevant for the existence of a long-run leverage target, but the arrival of investment opportunities further drives issue behavior. For example, more persistent investment shocks result in more frequent debt issues. 15 While debt financing is optimal even in case current leverage exceeds a long-run target, the additional debt associated with such a project is only predicted to be transitory. In general, issue frequencies are jointly driven by investment shocks and issuance costs. Figure 2 displays dynamic issue hazards for our HFIs (Panel A) and LFIs (Panel B). The graphs show that estimated hazards differ considerably across the two groups of firms. For HFIs, the probability of debt issuance is highest when the firm has just issued debt in the previous period. The more time elapses since the last issue, the lower the probability that the firm will issue again in the current period. For LFIs, the picture is different as overall debt issue probabilities are lower and the hazard rate is more hump-shaped. The shape of the hazard functions is broadly consistent with Leary and Roberts (2005) who estimate issue hazards based on simulated data generated from the tradeoff framework of Fischer, Heinkel, and Zechner (1989) under fixed and proportional issue cost regimes. In their simulations, variable costs generate a downward sloping hazard rate while fixed costs imply a hum-shaped form of the hazard function. However, the visual similarity of the hazard functions is not sufficient to conclude that observed issue behavior for HFIs and LFIs is actually driven by pure trade-off considerations. For example, we can t reject the possibility that more persistent investment shocks lead to more frequent and clustered net debt 14 For details on the overall effect of issuance costs on issue frequencies, see Table IV in (Fischer, Heinkel, and Zechner, 1989). For details on the impact of the structure of issuance costs on issue frequencies, see Leary and Roberts (2005). 15 For details on the effect of investment shock persistence on issue frequencies, see Table 4 in DeAngelo, DeAngelo, and Whited (2011). 11

13 issues of HFIs (DeAngelo, DeAngelo, and Whited, 2011). In this section, we therefore develop several hypotheses to examine whether the issue behaviour of HFIs is consistent with classical dynamic trade-off theories or dynamic financing and investment models. We split the predictions into two parts. The first part deals with classic dynamic trade-off theory, as summarized in Proposition 1 below. The second part uses the predictions of dynamic financing and investment models, detailed in Proposition Classical dynamic trade-off theory Proposition: Suppose HFIs face lower total and fixed debt issuance costs than LFIs. Dynamic tradeoff theory predicts the following: (1) HFIs make smaller and more frequent net-debt issues than LFIs. (2) Irrespective of issue frequencies, over-levered firms do not issue net debt (3) HFIs have less volatile leverage ratios than LFIs. (4) HFIs will exhibit greater speed-of-adjustment to target leverage ratio deviations than LFIs. Discussion: The Proposition is a simple restatement of the comparative statics presented by Fischer, Heinkel, and Zechner (1989) under the assumption that HFIs have lower total and fixed issue costs than LFIs. The intuition is simple: let firm value follow a random walk with a positive drift, and suppose debt issue costs are fixed (our LFIs). If the firm is currently at the optimal leverage ratio, firm value has to drift upwards by at least the issue-cost amount before triggering another recapitalization, which takes time. Moreover, at the recapitalization point, the net-debt issue will be large enough to bring the firm value back to its maximum, where net marginal issue benefits and marginal issue costs both equals zero. Conversely, if issue costs are largely proportional (our HFIs), the optimal issue size is smaller, resulting in relatively frequent but small issues, where each issue drives only the movement back to the recapitalisation boundary. In both cases, over-levered firms are not expected to issue net debt. Finally, lower issuance costs result in tighter recapitalization boundaries, which directly implies lower leverage volatility and consequently a greater speed-of-adjustment to target leverage deviations See Table IV in (Fischer, Heinkel, and Zechner, 1989) for a numerical analysis producing these effects within their dynamic tradeoff theory. The table shows that the existence of even small proportional costs can lead to wide periods of inactivity. However, note that their analysis assumes that once a leverage boundary is hit, the firm retires the entire old 12

14 4.2 Dynamic financing and investment theory Proposition: Suppose HFIs face more persistent investment shocks than LFIs. Dynamic financing and investment theory predicts the following: (1) HFIs issue transitory debt more frequently than LFIs (2) Irrespective of issue frequencies, firms retire transitory debt in periods following investment (3) HFIs have lower but more volatile (net) leverage ratios than LFIs (4) Irrespective of issue frequencies, net debt issues respond to deviations from target if optimal investment outlays are low Discussion: The Proposition is a restatement of the comparative statics presented by DeAngelo, DeAngelo, and Whited (2011) under the assumption that HFIs face more persistent investment shocks than LFIs. 17 The intuition is as follows: more persistent investment shocks lead to a higher frequency of observed net debt issues. HFIs issue net debt more frequently in order to finance investment, thereby also increasing the frequency of transitory debt (i.e. cases where current leverage exceeds a long-run target) relative to LFIs. The transitory debt model of DeAngelo, DeAngelo, and Whited (2011) further shows that, irrespective of issue frequency, firms retire transitory debt in periods following investment to bring leverage back to a long-term target. 18 The high persistence of investment shocks makes debt a valuable source of external funding. As a consequence, firms optimally keep debt low in order to preserve debt capacity (resulting in low average leverage ratios) but the occurrence of transitory debt at the same time increases leverage ratio volatility relative to LFIs. Finally, dynamic financing and investment models imply that a classical adjustment to a long-run target is most likely, in case optimal investment outlays are low. 4.3 Test results of classical dynamic tradeoff theory The evidence in Panel A of Table 7 fails to support the first two predictions in Proposition 1. The table shows net-debt issue sizes (relative to the lagged market value of the firm) for the HFIs and LFIs in our debt and issues new debt. As a result, proportional issue costs are paid on the total face value of the new debt and not just on the difference between old and new debt exacerbating the period of inactivity between rebalancing events shown in their Table IV. 17 See Table 4 in DeAngelo, DeAngelo, and Whited (2011) for detailed comparative statics. 18 Note, however, that new investment shocks can lead to further debt issues even though the old transitory debt has not been fully retired. 13

15 sample. Net-debt issues are similar and, on average, substantial for both categories of firms: 16% and 17%, respectively. Results are further robust to using median values (10% versus 9%), thereby rejecting the first prediction of Proposition 1. To benchmark the size of net debt issues, Table 7 further includes information on the firm s investment outlays. Average capital expenditures equal 12% (8%) of the lagged market value of the firm for high (low) frequency issuers. While postponing a more formal examination of dynamic financing and investment models to Section 4.4, these results are indicative that the higher frequency of net debt issues for HFIs is correlated with higher investment outlays. The last column displays the fraction of net debt issues occurring in periods when HFIs (LFIs) are already over-levered. Contrary to the predictions implied by a classical dynamic trade-off model, net debt issues of already over-levered firms occur frequently for both groups of firms (30% for HFIs, 26% for LFIs), consistent with Hovakimian (2004). Turning to the third prediction of the Proposition 1, Panel A of Table 8 links issue frequencies to the stability of corporate leverage ratios. The table is structured as follows. First, Initial leverage (L 0 ) is the leverage ratio in year 0 for a firm still listed in year t (t = 1,.., 30). For example, the average initial leverage ratio for LFIs five years following public listing (t = 5) is L 0 = 10%, while it is 20% for HFIs. Second, Change in leverage is the change in the leverage ratio from year 0 to year t (L t L 0 ). For example, in listing year five, the leverage ratio change is +2 percentage point for LFIs (up to 12% from the 10% in year 0) and +20 percentage points for HFIs (up to 40% from the 20% in year 0). Third, follows the definition of instability in DeAngelo and Roll (2013), Leverage instability is the fraction of firms for which the change in leverage from year 0 to year t exceeds +/ 20 percentage points ( L t L 0 > 0.2). Finally, Leverage volatility denotes the time-varying volatility estimate of firm i s leverage ratio. 19 Panel A of Table 8 shows that the degree of leverage instability differs substantially across LFIs and HFIs, but in the wrong direction. According to Proposition 1, leverage ratios of HFIs should be less volatile than those of HFIs. Instead, Table 8 shows clearly that leverage ratios of HFIs are substantially more volatile than those of LFIs (16% HFIs versus 8% LFIs). The higher volatility of leverage ratios is further reflected in the alternative instability measure. To illustrate, for HFIs, 50% of the individual leverage changes up through listing year 5 exceed +/ 20 percentage points. The corresponding percentage 19 Unlike the analysis in DeAngelo and Roll (2013), Table 8 conditions on firm age since listing and on the net-debt issue frequency (LFI and HFI). The frequency count in Table 8 is based on the issue-size threshold of 2.5% of the current book value of assets. 14

16 for LFIs is only 16%. 20 Panel B shows that the difference in leverage ratio volatility is eliminated when net leverage ratios are used (net leverage ratio volatility equals 19% for both HFIs and LFIs). However, the leverage instability measure of DeAngelo and Roll (2013) continues to suggest that net leverage ratios of HFIs are more unstable (43% of all HFIs experience significant changes in leverage ratios relative to the year of public listing, whereas the corresponding fraction is only 26% for LFIs). In sum, HFIs do not have less volatile leverage ratios. If anything, there ratios are more volatile and more unstable. Table 9 presents corresponding information on the evolution of target leverage ratios. Contrary to above, target leverage ratio instability is the exception, not the rule: Only 4% of all firms experience a target leverage ratio change in excess of 20 percentage points and the volatility of target leverage ratios is similar for LFIs and HFIs. This is true across issue frequencies and suggest that the observed leverage ratio instability for high frequency issuers is driven by leverage ratio changes that are unrelated to changes in target leverage ratios. We next turn to the final prediction of Proposition 1 and test whether HFIs manage their net-debt issues towards a leverage target. Several studies estimate the following speed-of-adjustment parameter φ: L i,t L i,t 1 = α + η i + φ ( L i,t(βx i,t 1 ) L i,t 1 ) + ɛi,t, (6) where the determinants X i,t 1 of the target leverage ratio include asset size, profitability, Q, cash ratio, tangibility, depreciation, R&D expenses, capital expenditures and the median industry leverage ratio. Estimation of equation (6) requires taking into account the presence of a firm fixed effect, the challenge that the target leverage ratio L i,t must be estimated and that the lagged dependent variable features as a regressor. 21 Failure to account for these complications introduces a dynamic panel bias which we address through the use of system GMM estimation (Blundell and Bond, 1998; Lemmon, Roberts, and Zender, 2008; Flannery and Hankins, 2013) Given the relatively stable leverage ratios of LFIs, this suggests that much of the leverage ratio instability reported by DeAngelo and Roll (2013) is driven by the subset of HFIs. 21 The equivalent regression in leverage levels is: L i,t = α + η i + φl i,t(βx i,t 1) + (1 φ)l i,t 1 + ɛ i,t. Thus, the SOA parameter φ is identified as the negative value of the parameter estimate for the lagged book leverage ratio L i,t The dynamic panel bias can be addressed through system GMM estimation (Blundell and Bond, 1998), long difference estimates (Hahn, Hausman, and Kuersteiner, 2007; Huang and Ritter, 2009), or bias correction methods (Kiviet, 1995; Bruno, 2005). The application of these methods is often complicated by the fact that corporate finance panels are unbalanced and suffer from non-contiguous data due to missing observations. Flannery and Hankins (2013) simulate data with similar properties and compare the performance of these estimates. Their simulations suggest that bias correction methods and system GMM estimates emerge as the most accurate methodologies. Since bias correction methods are computationally 15

17 The first column in Panel A of Table 10 shows the SOA estimate for market leverage ratios when estimating regression (6) across the 80,900 available firm-years in our sample. 23 The GMM coefficient estimate of suggests that it takes the average firm about 3 years to recover half of the target leverage deviation (the half-life implied by a SOA parameter φ is ln(0.5)/ln(1 + φ)). This estimate is very close to the GMM estimates of 0.25 reported by Lemmon, Roberts, and Zender (2008) for U.S. companies. Columns 2 and 3 of Panel A display separate SOA coefficients for low and high frequency issuers. The estimates suggest that high frequency issuers have a similar SOA coefficient (32% for high frequency issuers, 27% for low frequency issuers) and the difference is statistically insignificant. 24 These findings are inconsistent with the final prediction of Proposition 1. As pointed out by Welch (2004), leverage ratio changes may be driven by passive asset growth. To understand whether active net debt issues are related to deviations from target leverage ratios, we therefore replace the dependent variable and instead estimate the following regression equation: NDI i,t A i,t = α 1 + η 1,i + φ 1 ( L i,t (βx i,t 1 ) L i,t 1 ) + ɛ1,i,t (7) NDI + i,t A i,t = α 1 + η 1,i + φ 1 ( L i,t (βx i,t 1 ) L i,t 1 ) + ɛ1,i,t (8) Note that these regressions come without a specific hypothesis. The simple reason is that speed-ofadjustment estimates to leverage targets depend on both issue size assumptions and the frequency of net debt issues. Unconditionally, we expect that HFIs make smaller issues than LFIs (see prediction 1 in Proposition 1) which, holding frequency constant, would imply lower SOA estimates of such a net debt issue regression. However, lower issue frequencies also reduce SOA estimates (by increasing the periods intensive (requiring smaller samples), we use system GMM. We implement system GMM in Stata using the command xtabond2 and treat lagged leverage and the vector X i,t 1 as predetermined and use a maximum of 3 (lagged leverage) and 5 (X it, 1) lags when constructing instruments. Changing the specification and modelling X i,t as endogenous does not change our results. 23 See Table 11 for a robustness check using book leverage ratios. 24 These findings contribute to a more recent literature on SOA estimation that uses a model such as (6) to investigate the stability of SOA-estimates over time, across firms and across countries. Halling, Yu, and Zechner (2012) find that SOAestimates are lower during recessions and that this effect is more pronounced for financially constrained firms. Oztekin and Flannery (2012) compare SOA-estimates across countries and find that higher adjustment costs (proxied by cross-country variation in trading costs) significantly reduce the speed of adjustment. Hovakimian and Li (2012) estimate adjustment speeds at refinancing points and conclude that results are inconsistent with the premise of a partial adjustment model. Faulkender, Flannery, Hankins, and Smith (2012) estimate target leverage and the associated SOA parameter in two steps (where the first step is estimated using system GMM) and find that over-levered firms exhibit higher SOA-parameters than under-levered firms. 16

18 of inactivity) such that it is not clear whether the SOA of LFIs should be higher or lower than of HFIs. The results of the GMM estimation for equation 7 are shown in Panel B of Table 10 and show that net debt issues of HFIs are more sensitive to deviations from leverage targets than for LFIs (the estimated SOA coefficient φ equal 32% for HFIs and 19% for LFIs). These findings are robust to alternative issue frequency classifications. For example, results are unchanged if we classify HFIs and LFIs based on year 10 following public listing as opposed to the annual reclassification underlying Table 10. Panel C displays results of the positive net debt issue regression (equation 8). Note that the regression estimates the sensitivity of positive net debt issues for periods in which firms issue net debt. In other words, the coefficient φ can not be interpreted as a speed-of-adjustment parameter (given that the regression excludes periods of inactivity) but rather as an observed conditional sensitivity of positive net debt issues to deviations from leverage targets. All three coefficient estimates φ are statistically insignificant from zero, suggesting that positive net debt issues do not respond to deviations from leverage targets. Summing up, all four implications of classic dynamic trade-off models are not supported by our analysis of HFIs. While those firms actively issue net debt, those issues do not seem to be driven by classic trade-off considerations. Going beyond the four predictions, the perhaps most surprising and contradictory evidence is that while the size of net debt issues of HFIs is economically substantial, the associated net debt issues do not respond to deviations from leverage targets. 4.4 Test results of dynamic financing and investment theory Dynamic financing and investment theory implies that over-levered firms may issue debt to finance investment. The associated debt is predicted to be transitory and firms are expected to start retiring debt once investment shocks level out (DeAngelo, DeAngelo, and Whited, 2011). Panel A of Table 7 has shown that debt issues of over-levered firms happen more frequently for HFIs relative to LFIs (30% versus 26%). Panel B presents corresponding issue, investment and transitory debt characteristics of those over-levered firms (those with L < 0). Both issue and investment size are similar to the values reported in Panel A. Interestingly, issue size and investment outlays are similar for HFIs, but not for LFIs. The third column displays the deviation from the target leverage ratio in the year following the net debt issue (i.e. L t+1 L t): as expected, both group of firms are substantially over-levered as leverage exceeds the target by 9 percentage points (HFIs) and 8 percentage points (LFIs). What happens over the next three periods is interesting. Superficially, the evidence seems consistent 17

19 with the transitory debt argument as the deviation from the target is reduced to 1 percentage point (HFIs) and 2 percentage points (LFIs) in three years from now. Surprisingly, however, the adjustment does not happen through net debt retirements. In fact, for HFIs cumulative net debt issues are even positive over the next three years and equal to 8% of the current market value of assets. Instead, both groups of firms engage in substantial equity issues over the coming three years: HFIs (LFIs) raise new equity equal to 14% (11%) of the market value of assets. Those external equity issues, in conjunction with retained earnings, reduce excess leverage back to the target. Consequently, the term transitory leverage may be more appropriate. These findings are interesting and difficult to frame within a narrow interpretation of the dynamic financing and investment theory. The evidence on the first two predictions of Proposition 2 are thus mixed. Over-levered firms issue debt to finance investment and reduce deviations from target leverage ratios substantially over the next three years. However, debt retirements have no role in this adjustment as firms instead issue equity and retain profits. Turning to the third prediction of Proposition 2, Table 8 presents direct evidence on average leverage ratios and leverage ratio volatility. Results are again mixed. As shown before, HFIs have indeed more volatile leverage ratios but the difference vanishes if net leverage ratios are employed. In addition, Table 8 clearly shows that HFIs have substantially higher leverage ratios. This holds true for gross leverage ratios (Panel A) and net leverage ratios (Panel B). The findings thus do not support the prediction that HFIs keep leverage ratios low in order to preserve debt capacity. 25 Finally, dynamic financing and investment theory implies that adjustment to target leverage ratios is most likely when optimal investment outlays are low. DeAngelo, DeAngelo, and Whited (2011) state that <... >, we find that firms in our model rebalance aggressively toward target in some, but not all, states of the world, most notably when optimal investment outlays are low. Panel D of Table 10 estimates the sensitivity of net debt issues to deviations from target leverage ratios for periods in which investment outlays are low. Comparing the SOA-coefficients to those in Panel B, we can see that speed-of-adjustment to target is indeed higher in periods for which observed investment outlays are low. 25 Note that equity issue costs also generate strong variation in optimal leverage ratios in DeAngelo, DeAngelo, and Whited (2011). It would be interesting to see whether lower equity issuance costs offset the effect more persistent investment shocks on optimal leverage ratios. 18

20 5 Discussion and alternative explanations 5.1 Are low-frequency issuers financially constrained? In economic terms, a financially constrained firm faces relatively high (perhaps even prohibitive) costs of external funding. With this definition, firms that are known to have raised debt at a consistently high frequency throughout their lifecycle as public companies are almost surely truly unconstrained relative to firms that rarely raise external funding. The extant literature offers several types of financial constraint classifications, including the sensitivity of investment to cash flow (Fazzari, Hubbard, and Petersen, 1988), quantitative indices based on financial statement and stock market information (Kaplan and Zingales, 1997; Lamont, Polk, and Saa-Requejo, 2001; Whited and Wu, 2006), univariate sorts on firm characteristics (Almeida, Campello, and Weisbach, 2004), and a classification based on firm size and age (Hadlock and Pierce, 2010). Table 12 displays the Kaplan-Zingales (KZ) index, the Whited-Wu (WW) index and the size-age (SA) index proposed by Hadlock and Pierce (2010) for our LFIs and HFIs. For all three indices, higher values indicate greater financial constraints. 26 The WW-index and the SA-index give similar scores to LFIs and HFIs, thus failing to detect the greater financial flexibility revealed by HFIs. Moreover, the KZ-index wrongly scores HFIs as more financially constrained than LFIs. The KZ-index attaches a relatively large weight to leverage, and HFIs are indeed highly leveraged (Table 4). The WW-index also accounts for sales growth and firm size, and we have shown that HFIs grow substantially and are relatively large. In addition, it loads substantially less on leverage. As it turns out, the negative effect of firm size and sales growth on the financial constraint score dominates the positive effect of leverage, to the point where the difference between HFIs and LFIs becomes small under the WW index. Under the SA-index, the effect of firm size is non-linear, and this index also gives older (longer listed) firms a lower score. However, it can be seen that age dominates in the construction of the SA index as dollar differences in the book value of assets are mitigated by the logarithmic transformation of the size variable. The net effect of size and age is such that the SA-index is again similar for HFIs and LFIs. 26 KZ = prof L div C Q where div is the ratio of dividends to assets. W W = prof divpos tldt size ISG SG where divpos is a dummy variable equal to one in case the firm paid dividends, tldt is the ratio of long-term debt to assets, ISG is industry sales growth (defined using 3-digit SIC codes) and SG is sales growth. SA = size size age. 19