Tradeoff theory and leverage dynamics of high-frequency debt issuers

Transcription

1 Tradeoff theory and leverage dynamics of high-frequency debt issuers B. Espen Eckbo Michael Kisser October 2017 Abstract We examine the lifecycle leverage dynamics of persistent debt issuers (net of debt retirements). We show that these public industrial firms raised the bulk of the aggregate public and private corporate debts issued in the US over the past three decades. In apparent contradiction of tradeoff theory, the cross-sectional leverage-profitability correlation is non-positive at recapitalization points. Moreover, the speed-of-adjustment to target leverage deviations are no higher than for low-frequency debt issuers, of which forty percent are all-equity firms. While there is some evidence that firms eliminate excess leverage in periods with low investment activity, financing investments with debt appears to take priority over managing towards a leverage target. JEl classification: G32 Keywords: High-frequency debt issuer, issue costs and benefits, dynamic rebalancing, speed of adjustment An earlier version of this paper was titled Does tradeoff theory explain high-frequency debt issuers?. We have benefitted from the comments and suggestions of Andras Danis, Harry DeAngelo, Pierre Chaigneau, Michael Hertzel, Steve McKeon, Michael Roberts, Karin Thorburn, Toni Whited, and seminar participants at Australian National University, Concordia University, the Norwegian School of Economics, the Norwegian School of Management, Tuck School of Business at Dartmouth, Tulane University, University of Adelaide, University of Bristol, University of Edinburgh, University of St. Gallen, University of Liechtenstein, University of New South Wales, and the Vienna Graduate School of Finance. This research has also been presented at the University of Stavanger Corporate Finance Conference, and at the annual meetings of the Financial Management Association, the European Finance Association, the Society for Financial Studies (SFS) Cavalcade, and the Southern Finance Association. Financial support from Tuck s Lindenauer Center for Corporate Governance is gratefully acknowledged. Tuck School of Business at Dartmouth, b.espen.eckbo@dartmouth.edu, Norwegian School of Economics, michael.kisser@nhh.no,

2 1 Introduction Building on Merton (1974) s seminal model of corporate default, the tradeoff theory of capital structure choice, in which firms trade off marginal debt-financing benefits against expected default costs, has evolved to become an important workhorse in corporate finance. 1 In the dynamic version of this theory, fixed debt issuance costs create a wedge between the actual and target leverage ratios. That is, a firm will be at its target leverage ratio only in periods when it actively rebalances capital structure in other periods the optimal leverage policy is one of dynamic inactivity. 2 Incorporating endogenous investment and differential issue costs across debt and equity, recent financing and investment models also point to a potentially significant option value of future debt capacity and of cash holdings. 3 Dynamic financing and investment models also give rise to a pecking order financing hierarchy, reminiscent of the original but informal framework of Myers (1984) and Myers and Majluf (1984). Empirical evidence on the tradeoff theory is mixed. 4 For example, although low-volatility firms and firms with more tangible assets tend to have higher leverage as predicted, leverage ratios appear too low relative to theoretical predictions. Also, there is an ongoing debate over whether leverage ratios revert too slowly, or are sufficiently stable, to suggest the existence of a long-run target ratio. 5 Moreover, while Danis, Rettl, and Whited (2014) find a positive cross-sectional correlation between leverage and profitability in periods when firms rebalance capital structure (as required by optimal dynamic rebalancing policy), Eckbo and Kisser (2017) find that the correlation is non-positive when the equity repurchase in the rebalancing event is financed with new debt only. Finally, there is an ongoing debate over whether firms (particularly smaller growth companies) cover their internal financing deficit with too much equity for the simple financing pecking order of Myers (1984) to hold in the data. 6 At the end of their comprehensive review, Graham and Leary (2011) express concern about test power when using broad (Compustat-wide) samples to examine specific capital structure predictions. 7 With 1 For early static tradeoff models, see Robicheck and Myers (1966), Kraus and Litzenberger (1973) and Brennan and Schwartz (1984). Sundaresan (2013) provides a theoretical review. 2 See Fischer, Heinkel, and Zechner (1989) and Goldstein, Ju, and Leland (2001). The phrase is coined by Danis, Rettl, and Whited (2014). 3 Hennessy and Whited (2005, 2007), Strebulaev (2007), Gamba and Triantis (2008a), DeAngelo, DeAngelo, and Whited (2011), and Kisser (2013) 4 See Frank and Goyal (2008), Parsons and Titman (2008), and Graham and Leary (2011) for comprehensive reviews. 5 Fama and French (2002), Flannery and Rangan (2006), Lemmon, Roberts, and Zender (2008), Hovakimian and Li (2012), Faulkender, Flannery, Hankins, and Smith (2012), DeAngelo and Roll (2015), Deangelo, Goncalves, and Stulz (2016). 6 Shyam-Sunder and Myers (1999), Frank and Goyal (2003), Fama and French (2005), Leary and Roberts (2005, 2010), Lemmon and Zender (2010), DeAngelo, DeAngelo, and Stulz (2010). 7 [W]e may gain more clarity on the drivers of financing decisions by focusing on appropriate subsets of firms (Graham 1

3 the issue of test power in mind, we identify a sample of firms from the period that have been largely ignored in the extant literature: high-frequency net-debt issuers (HFIs, net of debt repurchases and rollovers). HFIs are public industrial companies that received the bulk (56%) of all public and private debts issued by public industrial firms over the past three decades. As discussed below, the persistent debt issue activity, as well as other firm characteristics, strongly suggests that these firms exhibit a combination of high debt-financing benefits and low issue costs two fundamental parameters in any dynamic tradeoff theory. Simply put, if HFIs do not actively rebalance capital structure towards a long-run target, then who does? We study the leverage dynamics of HFIs over their life-cycle as public companies. In each year following public listing, a firm is classified as HFI if its cumulative net-debt issue frequency since the listing year is in the upper quartile of the cross-sectional issue frequency distribution for that year. The issue frequency of HFIs averages twelve times the average issue frequency in the lower quartile, which we label low-frequency issuers (LFIs). 8 We demonstrate through a battery of tests (shown in the Appendix) that the issue frequency classification is highly persistent: firms classified as HFI or LFI early in their public lifecycle tend to remain in that classification. This persistence is robust to varying the length of the issue cumulation period (e.g. a fixed three-year look-back period), and it is confirmed by out-of-sample predictions of future issue activity. Our lifecycle HFI/LFI sort adds two new aspects to the sampling procedures currently available in the literature. Both aspects helps control for some types of unwanted sample heterogeneity that may have frustrated testing in the past. First and foremost, the HFI sort exclude nearly all zero- and near-zero leverage firms, which the extant literature concludes represent a major empirical challenge to capital structure theory (Strebulaev and Yang, 2013). The HFI sort also tends to exclude near-bankrupt firms who face prohibitive debt issuance costs (Myers, 1977; Shleifer and Vishny, 1992; Eckbo and Thorburn, 2008). 9 Therefore, the HFI sort differs substantially from a sort on leverage. In fact, a sort on leverage would produce an average market leverage ratio of 52% in the upper quartile of the cross-sectional leverage distribution, while the average market leverage ratio for our HFIs is only 32%. and Leary, 2011, p. 340). Myers (2001) makes a similar point: Because [capital structure] theories are not general, testing them on a broad, heterogeneous sample of firms can be uninformative (p. 99). 8 HFIs on average issue net-debt more than twice as often as the average industrial Compustat firm studied by Leary and Roberts (2005) over their sample period We also show that HFIs repurchase net-debt more often than do LFIs, while there is little difference in the net equity issuances across the two groups. 9 Financial distress has a multi-year effect on the cumulative issue frequency as signs of poor operating performance often emerge as early as three years prior to Chapter 11 filings in the U.S. (Eckbo, Thorburn, and Wang, 2016). 2

4 The second new aspect to our sampling strategy is the firm-level lifecycle restriction. Our requirement that sample firms go public during the sample period tilts the sample towards younger firms relative to samples that do not have this restriction. While this restriction is required in order to perform the HFI/LFI classification, it also provides descriptive information on leverage dynamics conditional on firms listing age that is new to the literature. An interesting and perhaps surprising finding is that leveragerelated firm characteristics differ significantly between HFIs and LFIs going all the way back to the public listing year. It appears that firm listing age is much less important than asset structure in determining the differential leverage policies of the two types of firms. Relative to LFIs, HFIs tend to have higher profitability, asset tangibility and growth, but lower cash flow volatility, R&D expenditures and Tobin s Q. This difference in firm characteristics between HFIs and LFIs emerges shortly after public listing, and it tends to persist throughout the lifecycle as public companies. Of course, greater profitability, lower cash flow volatility and higher asset tangibility are all consistent with the notion that HFIs enjoy significantly higher debt-financing benefits and lower debt-issue costs than LFIs (Frank and Goyal, 2009). We also provide independent estimates of dynamic issue-hazard curves of the type in Leary and Roberts (2005), which further support this notion that is central to this paper s empirical analysis. We apply the HFIs in three separate explorations of dynamic tradeoff theory. In the first and most important application, we address the puzzling evidence of a negative correlation between leverage and profitability first documented by Titman and Wessels (1988) and Rajan and Zingales (1995). Myers (1993) characterizes this negative correlation as the most telling evidence against the static tradeoff theory (p. 6). However, as pointed out by Danis, Rettl, and Whited (2014), in a setting with capital structure rebalancing costs, dynamic tradeoff theory (with exogenous investment) predicts a positive leverage-profitability correlation only in periods when firms actually rebalance capital structure by issuing debt and repurchasing equity. In other periods, optimal dynamic inactivity mechanically causes the correlation to be negative. Using the HFIs, we estimate the cross-sectional correlation between profitability and market leverage ratios in quarters when firms issue debt and repurchase equity. This cross-sectional correlation is particularly interesting because it tests tradeoff theory without having to also estimate the target leverage ratio. However, we find that the conditional leverage-profitability correlation is non-positive when using our HFIs as test assets. As explained below, this finding is contrary to the correlation estimates in Danis, 3

5 Rettl, and Whited (2014) but consistent with the Compustat-wide analysis in Eckbo and Kisser (2017). Second, we test for differences in the speed-of-adjustment (SOA) to target leverage ratio deviations of HFIs and LFIs, respectively. Given the low debt-issue frequency of the LFIs (about one-quarter of that of the HFIs), tradeoff theory suggests a greater SOA coefficient for HFIs than for LFIs. Instead, we find that the SOA estimates are indistinguishable across the two types of firms. This suggests that SOA estimates are to a large extent driven by changes in equity value (the denominator of the leverage ratio) rather than by debt issues, a point also made by Welch (2004). Furthermore, since the equity issue frequency is similar across HFIs and LFIs (shown below), we conclude that the large SOA coefficient estimate for the LFIs is likely driven by passive changes (drift) in the underlying equity values over time. 10 Third, as in the investment and financing model of DeAngelo, DeAngelo, and Whited (2011), the value of funding new investment projects with debt may occasionally dominate the cost of becoming temporarily over-leveraged. There is a concern that such transitory debt issues mask true tradeoff behavior in the data. To address this concern, we isolate firm-quarters where a transitory debt issue is followed by three years of investment inactivity. While this extensive conditioning results in a small sample size, we do find some evidence that the prior excess leverage is eliminated during the subsequent quiet investment period. This part of our analysis is not unlike that of Denis and McKeon (2012), who do not find evidence of leverage reductions following debt issues by over-levered firms. We return to a potential explanation for this difference in inferences below. In sum, we demonstrate that our HFIs exhibit firm characteristics suggesting a combination of high debt-financing benefits and low debt-issue costs. Notwithstanding the high debt-issue frequency, we do not find a positive leverage-profitability correlation in periods when these firms rebalance capital structure. Moreover, the SOA coefficient estimate for HFIs does not differ significantly from that of LFIs. These two findings fail to support dynamic tradeoff theory. It appears that, in the race to fund investments, managing leverage towards a target is of second-order importance to selecting the least expensive security in the pecking order. Our main contribution is to show that this conclusion, which has been suggested more generally by DeAngelo and Roll (2015) and others before us, holds for the most persistent public industrial debt issuers in the US economy over the past three decades. In addition, we provide smallsample evidence that HFIs tend to eliminate excess leverage in periods with low investment activity, i.e., 10 Our GMM estimation suggest a target leverage deviation half-life of about 2.5 years for both LFIs and HFIs. This is similar to SOA estimates reported in the extant literature for Compustat-wide samples. See, e.g., Flannery and Rangan (2006), Hovakimian and Li (2012) and Faulkender, Flannery, Hankins, and Smith (2012). 4

6 when investment funding is of second-order importance. The rest of the paper is organized as follows. Section 2 explains our classification of firms as HFIs and LFIs and shows key differences in firm characteristics. Section 3 explains two key empirical tradeoff hypotheses that are later tested in Section 4 (leverage-profitability correlations) and Section 5 (SOA estimates). Finally, Section 6 integrates debt issues with observed investment funding policies. Section 7 concludes the paper. 2 Characteristics of persistent net-debt issuers 2.1 Sample selection We use the annual merged Crsp/Compustat (CCM) file to sample firms, and quarterly CCM data to construct the annual issue frequency count, Table 1 details the sample selection, with Panel A for annual and Panel B for quarterly data. As is common in the capital structure literature, we exclude foreign firms, financial companies and regulated utilities, as well as firms with missing entries of key Compustat balance sheet and cash flow characteristics. In Panel C, we merge the quarterly and annual financial statement information and impose two additional sample restrictions. The most restrictive is to require the firm to go public during the sample period, which excludes 4,001 firms that went public prior to We condition the analysis below on public listing age in order to control for the effect of a firm s product market maturity on the debt issue frequency. That is, since older firms may have built collateralizable assets which may affect the propensity to issue debt, we structure the issue frequency analysis in event time since the year of going public. The final sample consists of 9,340 firms and an unbalanced panel of 66,056 firm-years and 240,028 firm-quarters. 2.2 Net-debt issue counts and issue frequency sorts In this section, we build the cumulative annual issue counts from the sample firms quarterly Compustat cash flow statements. All variable definitions are in Tables 2 and 3. A quarterly net-debt issue (N DI) is defined as the difference between the sum of all forms of public and private debt issues and debt retirements. This definition ensures that we are not counting debt rollovers (which appear in the cash flow statement as an equal issue and retirement). Let N it denote the cumulative number of positive 5

7 quarterly net-debt issues (NDI + ) by firm i from the public listing year (event year 0) through event year t. We have that t 4 N it = I iqτ, (1) τ=0 q=1 where I iqτ is an index that takes a value of one if firm i issues positive net-debt of at least 2.5% of total assets in quarter q of event year τ t (NDI + iqτ 2.5%). We use the 2.5% issue-size threshold rather than the more commonly used 5% threshold (Leary and Roberts, 2005, 2010) because it makes available a greater number of security issues for our cross-sectional analyses below. 11 In a given event year t, firm i is labelled a high-frequency issuer (HFI) if N it is in the upper quartile of the frequency distribution of N t. Moreover, firm i is labelled low-frequency issuer (LFI) if N it is in the lower quartile. A firm that is neither HFI nor LFI is labelled a medium-frequency issuer (MFI). Table 4 (using the 2.5% issue size threshold in Eq. 1) and Table 5 (using a 5% issue size threshold) list the average annual cumulative frequencies of net-debt issues (NDI + ), net-debt retirements (NDI ), and equity issues (EI) by the HFIs and LFIs since public listing. Panel D presents frequencies corresponding to a broader measure of equity issues (dsm) that also accounts for stock-financed acquisitions or direct stock issues to employees (Fama and French, 2005). The issue counts for NDI in Panel B, EI in Panel C and dsm in Panel D are for the firms classified as HFI or LFI in Panel A. While the tabulation stops with event year 20 for expositional simplicity, the empirical analysis below uses all contiguous firm-years in the sample. In the year of public listing, two thirds of the sample firms do not issue net-debt, while the remaining firms issue once only. Thus, in the first year, there are only two frequencies (0 and 1) and, as a result, there are no MFIs in year 0. Moreover, as the median firm age since going public is five years (average seven), ten years into the public lifecycle the annual number of sample HFIs and LFIs shown in Table 4 drops off quickly. This drop-off is, of course, a consequence of studying firms over their full public lifecycle. As discussed extensively below, this analysis reveals a high degree of persistence in the type of debt-funding policies observed already in the public listing year. With the 2.5% issue-size threshold in Table 4, HFIs (LFIs) make 61% (4%) of the total sample of 36,587 positive net-debt issues and receive 54% (7%) of the dollar value of total issue proceeds over the 11 We have verified that key test results (such as the leverage-profitability correlation estimation) hold also when using a 5% threshold. This is to be expected as the algorithm in (1) identifies much the same firms when using a 5% net-debt issue size threshold (NDI + 5%) as with the 2.5% threshold. 6

8 sample period. The annual spread between the net-debt issue frequency of HFIs and LFIs is substantial and persists over the public lifecycle. For example, over the first five years of public listing (t = 5), HFIs (LFIs) on average make 7.37 (0.41) quarterly net-debt issues. After twenty years of listing, the average number of issues is for HFIs and 2.70 for LFIs. Raising the issue size threshold to 5% in Table 5 reduces the average number of net-debt issues by HFIs (LFIs) to 4.82 (0.0) in year five, and to (1.62) in year twenty. Moreover, with the 5% threshold, HFIs raise 58% and LFIs 10% of the total issue proceeds over the sample period. Maintaining the HFI/LFI classifications from Panel A, panels B and C show that the annual spread in net-debt retirement (NDI ) and for equity issues. The tabulated frequency of net-debt retirements is interesting per se since, in classical dynamic tradeoff models, it is never optimal to reduce leverage outside of default or strategic renegotiation (Danis, Rettl, and Whited, 2014; Admati, DeMarzo, Hellweg, and Pfleiderer, 2017). In fact, both Table 4 and Table 5 show a significant number of net-debt retirements. For example, with a 2.5% threshold, the average number of net-debt retirements after five years of listing is 4.32 for HFIs and 1.19 for LFIs, and it is and 3.73, respectively, after twenty years. Moreover, in year five, the percentage of total retirement volume is 48% for HFIs and 16% for LFIs (46% versus 6% after twenty years). While HFIs issue and retire much more debt (and much more often) than LFIs, Tables 4 and 5 show that HFIs have a somewhat lower equity issue frequency than LFIs. For example, with a 2.5% equity issue size threshold and after ten years of listing, HFIs and LFIs have on average made 3.78 and 4.87 equity issues, respectively. Increasing the issue threshold to 5% causes the spread in these numbers to decrease slightly, to 3.49 versus The total sample median is 3 equity issues after ten years (2 issues with a 5% threshold, which is similar to the frequency of seasoned equity offerings reported elsewhere in the literature. 12 Panel D uses a broader measure of equity issues that also accounts for equity issues that do not raise cash, such as stock-financed acquisitions or direct stock issues to employees (Fama and French, 2005). The use of this alternative measure doubles average equity issue frequencies and increases the spread between HFIs and LFIs. For example, after ten years and using the 5% threshold, the average sample firm has made 5.88 issues (compared to an average of 3.37 equity-for-cash issues). This increase is consistent with the finding of Fama and French (2005) that stock-financed mergers account for a large fraction of 12 Eckbo and Masulis (1995), Eckbo, Masulis, and Norli (2007), Leary and Roberts (2010). 7

9 overall equity issues. Moreover, our finding that accounting for stock-financed takeovers increases the equity-issue frequency more for LFIs than for HFIs is consistent with extant evidence that the propensity to pay for the target in bidder stock increases in acquirer s market-to-book ratio (Eckbo, Makaew, and Thorburn, 2017). In the Appendix, we demonstrate that firms classified as HFI/LFI early on after public listing tend to persist in that classification. In particular, the HFI/LFI classification is shown to significantly predict future (out-of-sample) net-debt issues, which is required for there to be true forward-looking persistence in the issuance activity. Moreover, we also show directly that there is substantial (typically greater than 80%) overlap, year by year, between the firms in the HFI/LFIs sorts above and sorts with only a threeyear look-back period, or even using a within-year sort only (no annual cumulation). We consider this discovery that a certain set of firms emerge as HFIs and LFIs early in their public life-cycle and persist in that classification a core empirical contribution of this paper. 2.3 Firm characteristics and life-cycle funding of HFIs versus LFIs Differences in leverage-related firm characteristics Table 6 lists average firm characteristics of HFIs and LFIs sorted by year since public listing. As expected, the table shows significant differences in firm characteristics that the capital structure literature often associates with differential issue costs and debt financing benefits. What is more surprising is that these differences emerge already shortly after public listing and then persist over the public lifecycle. 13 HFIs have relatively high leverage ratios whether considering gross debt or debt net of cash balances. On average, the market leverage ratio (L in Column 1) is 32% for HFIs and 7% for LFIs. 14 The annual fraction of the sample firms that are all-equity financed (AE in Column 2) averages 40% for LFIs and only 3% for HFIs. Moreover, the cash ratio C in Column (3) is much lower for HFIs than for LFIs: 11% versus 40%, respectively. While not tabulated, there is evidence that the high cash holdings of LFIs reflect basic operating policy. To see this, we first estimate the coefficients of a standard cash model accounting for firm and 13 Finally, in terms of the Fama-French FF12 industries, the sample representation of HFIs and LFIs is as follows: business equipment (HFIs 14%, LFIs 39%), shops (HFIs 21%, LFIs 8%), health care (HFIs 10%, LFIs 22%), consumer non-durables (HFIs 8%; LFIs 4%), consumer durables (HFIs 4%, LFIs 2%), manufacturing (HFIs 12%, LFIs 7%), energy (HFIs 8%, LFIs 2%), chemicals (HFIs 3%, LFIs 2%), and other (HFIs 21%, LFIs 14%). 14 As stated in the introduction, our net-debt issue frequency sort is not a direct sort on leverage ratios, as the latter produces an average market leverage ratio of 52% in the upper quartile of the cross-sectional leverage distribution. 8

10 age fixed effects and using the full sample of firms (Bates, Kahle, and Stulz, 2009), and then construct separate target cash balances for LFIs and HFIs using the coefficient estimates. Defining excess cash holding as the difference between the actual and estimated target cash holdings, the level of excess cash is similar across LFIs and HFIs: 0.4% and -.01%, respectively. In other words, the firm characteristics in the empirical target cash model go a long way in explaining the differential cash policies of LFIs and HFIs. It also suggests that much of the build-up of cash balances reported elsewhere (Bates, Kahle, and Stulz, 2009) is concentrated among LFIs. As shown in columns (4)-(6), HFIs are larger than LFIs (total assets averaging $849 million versus $514 million), have greater asset tangibility (PPE/Assets of 0.32 versus 0.17), and are more profitable: P rof averages 3% and -5% of total assets, respectively (40% of the LFIs have P rof < 0 versus 24% for HFIs). 15 The higher profitability of HFIs translates into a higher propensity to pay dividends (24% versus 15% for LFIs). In Column (7), the average ratio of dividends to assets is 0.01 for both categories of firms. However, conditional on paying dividends (Div > 0), the size of the HFI dividend is lower than for LFIs (2.6% versus 4.2%, and 4.4% versus 7.4% when we add share repurchases to the total payout). Thus, consistent with the greater leverage ratio of the HFIs, when paying dividends, the dividend rate is lower than for LFIs. Furthermore, columns (9)-(12) of Table 6 reveal interesting differences between the average investment rates of HFIs and LFIs. In Column (9), capital expenditures scaled by lagged book assets (I CX ) averages 10% versus 6% for LFIs. Column (10) is based on total cash investments and also includes cash outlays for patent purchases and acquisitions, increasing the scaled investments I CF to 15% for HFIs and 9% for LFIs. Moreover, much as Lewellen and Lewellen (2016), Column (11) reports the average investment I F A, which is computed from the yearly changes in fixed assets in the firm s balance sheet. 16 I F A fully captures cash-financed investment plus any portion of corporate acquisitions that is paid for in stock. This increases the average scaled investment to 24% for HFIs and 17% for LFIs, which very close to the average growth in book assets (column 12). In sum, while HFIs on average invest more in fixed assets than do LFIs, total long-term investment is large for both groups. Moreover, the average contribution of stock-financed acquisitions to long-term investments (the difference between columns 10 and 9) is similar across HFIs and LFIs: 9% versus 8%. 15 The average profitability for LFIs turns positive only in year 13 after public listing, compared to year 2 for HFIs. 16 The yearly changes in fixed assets are adjusted for non-cash charges that affect fixed assets such as depreciation and write-downs. See Tables 2 and 3 for the exact variable definitions. 9

11 Notwithstanding the larger rate of investments in fixed assets, Tobin s Q in Column (13) is on average substantially lower for HFIs than for LFIs (1.66 vs. 2.71). The likely reason for this difference is shown in Column (14), which document a substantially higher rate of R&D expenditures for LFIs than for HFIs: 12% versus only 3%. R&D expenditures are designed to generate valuable future growth options, which likely translate into higher Tobin s Q than do investments in fixed assets (Fama and French, 1998; Carlson, Fisher, and Giammarino, 2004) Differences in lifecycle funding policies In this section, we show that the differences in firm characteristics across HFIs and LFIs also translate into significant differences in average funding policies over the public life-cycle. We begin by investigating cash funding policy using the seven sources of funds identified by the firm s annual cash flow statement. Let R j S j / 7 i S i denote the contribution of funding source S j, where 7 S i CF + + EI + NDI + + C + I + W + O +. (2) i=1 Here, CF + is the positive portion of operating cash flow, EI is proceeds from equity issues, NDI + is positive net debt issues, C is draw-down of cash balances, I is sale of investments, sale of property, plant and equipment (PPE) and cash flows from other investment activities, W is reduction in net working capital, and O + is a small residual that maintains the cash flow identity. 17 Panel A of Table 7 and Figure 1 show the annual funding pattern after combining the seven funding sources into four ratios: the Asset Sales ratio R AS ( C + W +O + +I )/ 7 i S i, the Net-Debt Issue ratio R NDI + NDI + / 7 i S i, the Equity Issue ratio R EI EI/ 7 i S i, and the positive Operating Cash Flow ratio R CF + CF + / 7 i S i. 18 ratios than LFIs, with annual values of R NDI + As expected, HFIs exhibit substantially greater net-debt funding averaging 24% for HFIs and only 2% for LFIs (median values of 13% and 0%). In contrast, the importance of equity in the overall funding mix is more similar across HFIs and LFIs. The value of R EI in Panel A of Table 7 averages 18% for HFIs and 30% for LFIs, 17 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. 18 By construction, these four ratios sum to one. 10

12 with median values of 2% and 8%. 19 In sum, HFIs rely more on external finance than do LFIs (42% vs. 31%, respectively). Notice also that, since the HFIs generate more positive operating cash flow than LFIs (R CF + averages 34% vs. 29%), asset sales must be a particularly important funding source for LFIs. This is confirmed by our data: the lifecycle funding contribution of asset sales (R AS ) is substantial for both categories of firms, and larger for LFIs than for HFIs (40% versus 24%, respectively). The large contribution of the illiquid asset sales portion of R AS for LFIs (17%) is interesting in of itself, as it not anticipated by the traditional financing pecking order (Eckbo and Kisser, 2013; Edmans and Mann, 2016; Arnold, Hackbarth, and Puhan, 2017). For relatively high-r&d firms such as LFIs, raising cash through asset sales may be attractive as it avoids the strict disclosure requirements associated with public equity issuances and which risks disclosing valuable proprietary information produced by the R&D activity (Hall and Lerner, 2010; Brown, Martinsson, and Petersen, 2012; Bena and Li, 2014). A similar argument goes for acquisitions paid in stock. In Panel B of Table 7, we replace EI in Eq. (2) with the broader equity issue measure dsm (which, as discussed above, includes stock issues to pay for acquisitions). This increases the funding contribution of equity to an average of 23% for HFIs and 34% for LFIs. Notwithstanding this increase, the contribution of (illiquid) assets sales remains a substantial 16% for LFIs, down from the 17% shown in Panel A. In the remainder of the paper, we use the HFIs as our main test assets when examining predictions of tradeoff theory. We begin with two empirical hypotheses motivated by the class of dynamic tradeoff models in which investment is exogenous (and thus does not involve an investment financing decision) and which therefore exploit debt-issue dynamics without regard to investment funding. We subsequently expand the analysis to a third hypothesis motivated by recent financing and investment models that involve additional tradeoffs in terms of differential issuance costs across debt and equity securities. 19 Excluding the year of public listing substantially reduces the contribution from equity issues over the remaining life cycle as R EI drops to 7% for HFIs and 13% for LFIs. 11

13 3 Dynamic tradeoff theory: empirical predictions 3.1 Basic theoretical setup The following are basic features of the class of dynamic capital structure models with exogenous investment first developed by Fischer, Heinkel, and Zechner (1989). Abstracting from agency conflicts and information asymmetries, firms select an optimal mix of debt tax benefits and default cost so as to maximize equity value. The underlying state variable is unlevered firm value (a scaled version of cash flow). This state variable follows a random walk (geometric Brownian motion with positive drift) generated by assets in place as there is no new investment. The periodic cash flow is then shielded from taxation by issuing infinitely-lived risky debt with a fixed coupon (a consol bond). Firms adjust leverage by retiring all outstanding bonds at face value and issuing new bonds. This refinancing incurs a transaction cost that is proportional to the debt issuance. However, since even a marginal increase in outstanding debt involves retiring the entire outstanding bonds, the assumed transaction cost is large also for small leverage adjustments akin to a fixed cost of debt issuance, which is necessary to drive a meaningful wedge between actual and target leverage ratios. Debt issue proceeds are distributed pro rata to shareholders, as is any cash flow in excess of the coupon. Liquidation of the firm incurs a deadweight loss. Outside of liquidation, shareholders costlessly inject new equity whenever the cash flow falls short of the periodic coupon payment. Within this setup, managers maximize equity value subject to a limited liability constraint. Absent transactions costs, the firm would have continually adjusted the level of debt (bond coupon) in response to changes in unlevered firm value in order to maintain the target leverage ratio. However, the assumed cost structure implies that, as unlevered firm value drifts upwards, firms maintain their current debt levels until the benefits from leveraging up covers the debt issuance costs, which occurs when firm value reaches an upper, recapitalization boundary. Conversely, as unlevered firm value drifts downwards, managers maintain current debt level and the firm defaults if the leverage ratio reaches maximum. Thus, in this model framework, firms occasionally recapitalize by issuing new debt, while they never repurchase debt. The latter follows from the assumption that debt must be repurchased at face value, which would cause a wealth transfer from equity-holders to debt-holders when the firm is over-leveraged (Admati, DeMarzo, Hellweg, and Pfleiderer, 2017). As we show above that debt repurchases occur frequently for HFIs (tables 4 and 5), these events fall outside of the model framework. 12

14 3.2 Empirical hypotheses Consistent with the above theoretical setup, Figure 2 illustrates the optimal leverage dynamics of two firms i and j. Firm value V = f(l) is an increasing and concave function of the leverage ratio L. As the unlevered firm value drifts upwards (and L hence downwards), the firm rebalances capital structure by issuing debt to maintain the target leverage ratio L (over-leveraged firms, to the right of L do not repurchase debt). A fixed rebalancing cost (C) drives a wedge between L and the point in time when it is optimal to move the leverage ratio back to L. Specifically, the firm rebalances when L reaches the endogenous recapitalization bound L < L, which is a function of both C and f (L). The recapitalization moves L back to L. Figure 2 illustrates the value functions for two firms i and j who differ only in terms of their marginal benefits of debt financing, so that f i (L) < f j (L) for L < L. We will later associate firm j with HFIs and firm i with LFIs. For expositional simplicity, however, Figure 2 assumes that the two firms have identical target leverage ratio and fixed issue cost. Allowing C i C j and L i L j, which more realistically describes HFIs versus LFIs, complicates the illustration without changing the essence of the predictions below. The following empirical implications follow intuitively from the figure: 20 (H1) Regardless of L, the leverage profitability correlation is positive at recapitalization points for both firm j and i. (H2) Relative to firm i, firm j recapitalizes more frequently and with smaller issue sizes, and exhibits lower leverage ratio volatility and greater speed-of-adjustment to deviations from L. Note that testing H1 and H2 requires an explicit estimate of L only for the last prediction of H2 (the speed-of-adjustment). All of the predictions in H2 follow because the endogenous recapitalization range of firm j, L L j, is narrower than for firm i. Also, assuming C i > C j further increases the recapitalization range (and thus the refinancing spell) of firm i relative to firm j. 20 For formal proofs, see the comparative statics in Fischer, Heinkel, and Zechner (1989). Moreover, the simulations in Danis, Rettl, and Whited (2014) show that the leverage-profitability correlation in H1 is indeed positive at recapitalization points. 13

15 4 Is the conditional leverage-profitability correlation positive? We examine H1 by estimating the following regression model on quarterly data for the HFIs and LFIs: L t = α + γ 0 P rof t 1 + γ 1 P rof t 1 Recap t + γ 2 Recap t + βz t 1 + δrisk t 1 + ɛ t. (3) where the choice of the regressors follows Danis, Rettl, and Whited (2014). As in Table 6 above, L is the market leverage ratio and P rof is operating profitability. Moreover, X t 1 is a vector of lagged control variables including firm size, asset tangibility, market-to-book ratio and Risk. We show results with and without the control variable Risk because adding this variable requires five years of contiguous financial statement information and so reduces sample size. Recap indicates that the firm actively rebalances leverage in quarter t, as follows: Recap t = 1 if De t A t > s and ERe t A t > s, (4) and Recap t = 0 otherwise. Here, D e is the quarter s increase in long-term debt in excess of debt retirement, ER e the equity retirement in excess of equity issues, and the constant issue-size threshold s is in percent of assets (below, either 2.5% or 5%). In other words, in our definition, a gross leverageincreasing recapitalization event takes place when the firm issues debt (net of debt retirements) and retires equity (net of of equity issues) in amounts that exceed the threshold s. Dynamic tradeoff theory predicts a positive leverage-profitability correlation in quarters with active rebalancing, and a negative correlation in other periods. Thus, in terms of the above regression model H1 : γ 0 + γ 1 > 0 and γ 0 < 0. (5) Table 8 shows the coefficient estimates for HFIs and LFIs. While the table report results using a debt-issue and equity-retirement size threshold of 2.5%, our results are similar with a 5% threshold (available upon request). Notwithstanding the high net-debt issue activity of HFIs, leverage increasing recapitalizations are rare (they occur in less than 2% of the firm-quarters). Because Risk requires at least five years of contiguous data, columns (2) and (4) of Table 8 tilts the sample towards older firms. The Wald statistic in Table 8 test the hypothesis that γ 0 + γ 1 = 0. Notice first that, in all regressions, 14

16 the estimate of γ 0 is negative and significant. Thus, as expected under H1, the cross-sectional correlation between profitability and leverage is negative conditional on rebalancing inactivity. For γ 1, however, the point estimate is negative in three out of four regressions. Importantly, while three out of the four Wald test statistics reject the hypothesis that γ 0 + γ 1 = 0, none of the regressions support the hypothesis that γ 0 + γ 1 > 0. In other words, notwithstanding that HFIs persistently issue and retire net-debt over the public lifecycle, Table 8 fails to support H1. Using a Compustat-wide sample, Danis, Rettl, and Whited (2014) estimate a regression such as Eq. (3) and report evidence supportive of H1. Apart from sample differences, a key methodological difference is that they permit the equity repurchase used to define a recapitalization event to be cofinanced with cash draw-downs instead of relying on debt issues only. Using a similar Compustat-wide sample and regression specification, Eckbo and Kisser (2017) show that the bulk of the recapitalization events identified by Danis, Rettl, and Whited (2014) are entirely internally financed (no new debt issued), and that the support for H1 in Danis, Rettl, and Whited (2014) is restricted to those internally financed recapitalization events. However, as shown in Appendix Table 4, re-estimating Eq. (3) with Danis, Rettl, and Whited (2014) s definition of a recapitalization event does not yield coefficient estimates supportive of H1. In Appendix Table 5, we also examine the robustness of the conclusions from Table 8 to two alternative definitions of operating profitability. While it is common to use lagged profitability as a proxy for future profitability as in Eq. (3), Panel A replaces P rofit t 1 with the lagged value-weighted average operating profitability since public listing, while the proxy in Panel B is the lagged ratio of the change in retained earnings relative to the year of public listing (which reflects cumulative earnings less total dividend distributions) scaled by book assets. In both panels, we find that the gross-leverage profitability relation is non-positive. Finally, we return to the large number of debt retirements documented in tables 4 and 5 above. While active leverage reductions fall outside of classical dynamic tradeoff theory, contracting theory provides a possible rationale for debt retirements (Smith and Warner, 1979; Aghion and Bolton, 1992; Dewatripont and Tirole, 1994). For example, as suggested by Kisser and Rapushi (2017), shareholders may optimally repurchase outstanding debt when covenant violations threatens to transfer significant control rights from equity-holders to bond-holders. In this case, one would expect the more profitable firms to have greater optimal leverage ratios that the conditional leverage-profitability correlation is again positive (as in H1). 15

17 We examine this informal hypothesis in Table 9, where we use regression (3) with leverage-decreasing recapitalizations. As shown, hypothesis H1 is again rejected by the data. 5 Financing spells, leverage volatility and speed-of-adjustment Recall that hypothesis H2 and Figure 2 compare the financing spells, issue sizes, leverage volatilities and speed-of-adjustment of two firms i and j. In this section, we represent firm i by LFIs and firm j by HFIs. That is, we assume HFIs exhibit a combination of higher debt-financing benefits than LFIs (f LF I (L) < f HF I (L)) and lower fixed debt-issue costs (C LF I > C HF I ). 21 The descriptive evidence in Section 2 above clearly supports this representation. Not only is the net-debt issue frequency and leverage ratio of an order of magnitude higher for HFIs than for LFIs, HFIs also exhibit firm characteristics such as greater profitability and asset tangibility. These are factors that extant empirical research associate with lower issuance costs (Eckbo, Masulis, and Norli, 2007) and higher leverage benefits (Frank and Goyal, 2008; Parsons and Titman, 2008; Graham and Leary, 2011). We begin the examination of H2 by estimating the financing spells of HFIs and LFIs. We already know, of course, that HFIs issue net-debt at a much greater frequency than do LFIs. However, much as in Leary and Roberts (2005), our issue hazard estimation provides further information on the likelihood that the different issue frequencies also reflect differences in fixed issue costs. We then examine the remaining predictions of hypothesis H Relative financing spells Figure 3 presents estimated shapes of dynamic net-debt issue hazards. The shapes, which account for both firm characteristics and unobservable firm-specific heterogeneity, are estimated by parameterizing the following hazard function h of the j th net-debt issuance spell for firm i: h i,j (t α i ) = α i h 0 (t)exp(β 1 x i,j (t)), (6) where t is the length of the issue spell (years from the current to the next quarterly net debt issue), h 0 (t) is the baseline hazard, and α i captures unobserved heterogeneity analogous to a random effect in a panel data model (where multiple issues by firm i may be correlated). The shared frailty term α i is assumed 21 Unlike Figure 2, we do not assume that LFIs and HFIs have similar leverage targets. 16

18 to be independent of the firm characteristics x i,j (t) and to have a zero-mean gamma distribution. The choice of a gamma distribution is standard (Leary and Roberts, 2005; Whited, 2006). The estimated hazard shapes in the figure have steps because time has been discretized to the annual frequency. The baseline hazard h 0 (t) measures the conditional issue probability when all covariates x i,j (t) are equal to zero. We follow Leary and Roberts (2005) and parametrize h 0 (t) as a cubic polynomial in the time since the last issue: h 0 (t) = exp(c+γ 1 t+γ 2 t 2 +γ 3 t 3 ). The firm characteristics x i,j (t) are time-varying and enter the estimation each year after subtracting the sample-wide median value. Panels A and B of Figure 3 plot the estimated hazard shapes for LFIs and HFIs, respectively. The horizontal axis is years since last issue (in year 0). For example, at year five, the dynamic hazard function gives the estimated probability of a debt issue in year six conditional on not having issued debt over the previous five years. The hazard function for HFIs in Panel A has a high intercept and a negative slope, while the intercept is low and the slope positive for the LFIs in Panel B. These estimated differences are consistent with HFIs facing lower issue costs and/or greater issue benefits than LFIs (to the degree that the benefits are not fully captured by the time-varying set of control variables). Beginning with Panel A, both the intercept and the slope of the estimated hazard curve are as expected if HFIs face largely variable debt issuance costs. To illustrate, suppose the HFI just issued debt. With low fixed issue costs, the firm is likely to issue soon again as firm value drifts upwards (reflecting the standard assumption of a geometric brownian motion with positive drift) and pushes down leverage to a suboptimal level. Thus, the intercept is high, as is the estimate in Figure 2A. Moreover, the further away the HFI is from the last issue (the longer the financing spell), the more likely the asset value of the HFI has drifted down (not up), so you are less likely to issue again. Thus, the slope is negative, as the figure also shows. In contrast, suppose the issue cost is fixed and consider the estimated curve for the LFIs in Panel B. If the firm just issued, the fixed costs means the firm has to wait for the assets to drift upwards sufficiently to cover the fixed costs before rebalancing the leverage ratio. Thus, the intercept is now low, as is the estimate in Figure 2B. Moreover, the further away the LFI is from the last issue, the more likely the firm s asset value has drifted up enough to cover the fixed costs, and so the more likely the firm will issue again next period the slope is positive, as the estimation also shows. In sum, while the firm characteristics of the HFIs and LFIs generally support the notion that HFIs have lower issue costs and greater debt financing benefits than LFIs, the issue hazard curves in Figure 3 further support that HFIs may be facing lower fixed issue costs, as assumed under hypothesis H2. 17

19 5.2 Relative issue size and leverage ratio volatility Beginning with relative net-debt issue size of HFIs and LFIs, the two first columns of Panels C of Table 7 show that the average net-debt issue sizes of HFIs and LFIs, conditional on firm-years where N DI 2.5%, are indistinguishable: 30% and 31% of total sources of funds, respectively (median 25% and 26%). While not tabulated, if we scale the net-debt issue by total market value of the firm lagged one period, the corresponding average net-debt issue sizes are 10% and 11%, also statistically indistinguishable. Thus, there is no evidence that the net-debt issue sizes of LFIs are larger than those of HFIs. Under the assumption that LFIs face greater fixed issue costs than do HFIs, this evidence fails to support H2. Turning to leverage ratio volatility, the first column of Table 10 shows the average (market) leverage ratio in the year of public listing (L 0 ). For example, for the HFIs in Panel A, L 0 = 21% in year 0 for the full sample, while it is L 0 = 18% for the sample of HFIs that have been listed ten years or more. For the LFIs in Panel B, L 0 = 8% in the full sample (in year 0) and 4% for the subsample who have been listed ten years or more. Column (2) provides the average leverage ratio in event year t (L t ), while column (3) computes that difference between column (1) and (2) the change in average leverage ratios across event time. For example, in year ten after listing, the HFIs in Panel A have experienced a leverage ratio change averaging L 10 L 0 = 16%. The corresponding change for the LFIs in Panel B is only 2%. Columns (4)-(6) of Table 10 summarize the cross-sectional distribution of the leverage change. First, columns (4) and (5) list the percent of the sample with leverage ratio changes of at least ±20%. DeAngelo and Roll (2015) use this leverage ratio change limit to define an unstable leverage ratio. In Panel A, 45% of the HFIs increase leverage by at least 20% over the first five years of public listing, while only 7% of the LFIs in Panel B do so. As for reductions in the leverage ratio (Column 5), after five years of listing, 5% of the HFIs (3% LFIs) have reduced their leverage ratios by at least 20%. This indicates a substantially greater leverage volatility among HFIs than among LFIs. Column (6) of Table 10 shows the average leverage ratio volatility σ L, measured as the average of the standard deviation of the firm-level leverage ratio from year 0 up to year t, using a minimum of five annual observations. For the HFIs in Panel A, the sample-wide average standard deviation is 17%, which is stable across listing age. In contrast, for LFIs, the average standard deviation of the leverage ratio is only 5%, strongly rejecting the prediction that the leverage ratio of LFIs should exceed that of LFIs This conclusion holds also if we measure leverage volatility based on market leverage ratios net of cash holdings. 18