Liquidity and Corporate Debt Market Timing

Transcription

1 Liquiity an Corporate Debt Market Timing Marina Balboa Faculty of Economics University of Alicante Phone: Fax: Belén Nieto (Corresponing author) Faculty of Economics University of Alicante Phone: Fax: Abstract This paper investigates whether firm managers time ebt issuances accoring to market liquiity conitions. Using transactions ata in the U.S. market from July 2002 to December 2009, our results show that both the moment an volume of ebt issuance are significantly associate with perios of high aggregate liquiity in the corporate bon market. The result is especially strong when liquiity is aggregate across bons in the same risk class. Splitting the sample into timers an nontimers, we fin that liquiity timers benefit from issuing in moments of ebt overpricing obtaining financing at much lower cost. Key wors: corporate bons issuances, liquiity, market timing, multi-way cluster, panel ata, rating. JEL Classification: G31, G32, G12 1

2 I. Introuction The capital structure choice of firms is one of the most extensively researche fiels in corporate finance. However, whether this choice is the result of an attempt to apply an optimal capital structure theory or to time the market is still an area of research. There has been much interest in the role of market timing in firm financial ecisions an stuies also fin empirical evience consistent with market timing behavior in financing. For example, Graham an Harvey (2001) show that market timing is a major concern of corporate executives: Two-thirs of CEOs amit that timing consierations play an important role in financing ecisions. Generally speaking, market timing refers to the benefits that a firm can obtain by issuing securities at abnormally high prices an repurchasing at abnormally low prices. Initially, the market timing analysis focuse on equity issuances. 1 Debt market timing is relatively recent an almost all papers consier interest rates as the most relevant inicator of bons mispricing in relation to issuance ecisions. The main finings can be summarize as follows. The amount of ebt issue or the probability of issuing ebt is higher when interest rates are low (Barry et al., 2008, 2009; Doukas et al., 2011; Ooi et al., 2010). Barclay an Smith (1995), Guees an Opler (1996), Kaya (2011), an Stohs an Mauer (1996) show that ebt maturities ten to be shorter when the term sprea is higher an Bancel an Mittoo (2004) an Graham an Harvey (2001) fin that managers issue short-term ebt when they believe that future long-term interest rates will fall. Finally, interest rates are also relate to the type of coupon (fixe or floating) in new ebt issues or loans (Faulkener, 2005; Vickery, 2008). Other variables such as changes in the maturity structure of government ebt (Greenwoo et al., 2010), measures of equity mispricing (Gao an Lou, 2013), an inflation or the risk sprea (Zhou et al., 2012) have also been associate with ebt market timing. However, the benefits of this apparent market timing are not clear at all. On the one han, Baker et al. (2003), Greenwoo et al. (2010), an Greenwoo an Hanson (2013) provie evience supporting the timing hypothesis; they fin that long-term an non investment-grae ebt issues are associate with lower future bon returns. On the other han, Butler et al. (2006) present evience that the managerial market timing foun by Baker et al. (2003) is inee the result of a spurious ex post relation between 1 Stylize facts are that abnormal stock returns are reuce after a firm equity offering (Baker an Wurgler, 2000; Loughran an Ritter, 2002; Ritter, 1991; Spiess an Affeck-Graces, 1995) or that future firm leverage increases when ecreases in the stock price or increases in the equity risk premium are observe (Baker an Wurgler, 2002; Huang an Ritter, 2009). 2

3 security issuance an market returns. In aition, Song (2009) shows that issuing longterm ebt in years when the term sprea is low has no effect on firm value an, therefore, this potential timing is not successful. The contribution of this paper is to consier the aggregate liquiity of financial markets as an argument for ebt timing. A commonly accepte fact is that liquiity is relevant to the price-generating process of financial assets: Illiqui securities have higher expecte returns (lower prices) than liqui ones (Amihu an Menelson, 1986, 1991). This theoretical asset pricing founation has been use as the support of recent research papers. On the one han, several stuies examine the effects of illiquiity on corporate bon prices. Bao et al. (2011), Chen et al. (2007), Covitz an Downing (2007), Downing et al. (2009), Friewal et al. (2012), Houweling et al. (2005), an Longstaff et al. (2005) fin a positive relation between the creit sprea or the yiel of a corporate bon an its illiquiity. Acharya et al. (2013), Bao et al. (2011), Bongaerts et al. (2012), an Friewal et al. (2012) show that aggregate liquiity conitions affect corporate bon prices. 2 Therefore, it seems clear that liquiity matters to corporate bon investors. On the other han, an base also on the iea that increases in liquiity are associate with increases in asset prices, there are papers that relate liquiity an corporate asset issuances, but they are scarce an focus exclusively on equity issuances. Slutz et al (2014) use an international sample of 36 countries an show that stock market liquiity (both contemporaneous an lagge values) is an important eterminant of equity issuances across the worl. Butler et al (2005) an Gao an Ritter (2010) show that the higher the market liquiity the lower the expecte costs of unerwriters. An Lipson an Mortal (2009) use the negative relation between the stock liquiity an the cost of equity as the argumentation of avantages of equity financing relative to ebt financing an show that firms with more liqui equity have lower leverage ratios. However, the role of liquiity in timing consierations is funamentally ifferent an, as far as we are aware, this is the first paper analyzing this relation on ebt issuances. We argue that liquiity can be use as an inicator of bon mispricing, so that market liquiity conitions may allow the ientification of moments of overvaluation in the corporate ebt market. If this is the case, firms may obtain benefits by acting as a ebt timer. Our argument is theoretically supporte by the moel of Baker an Stein (2004). They propose a microstructure moel that accommoates the empirical fining in the corporate finance literature of a negative relation between security offerings an their 2 Whether the aggregate illiquiity risk is price by the corporate bon market is also analyze (e.g., Bongaerts et al., 2012; Chacko, 2005; e Jong an Driessen, 2012; Lin et al., 2011;.Mahanti et al., 2008). 3

4 subsequent returns by using assets price pressures as the justification. The iea is that market liquiity can be interprete as a sentiment inicator of irrational investors that overvalue assets. Accoring to this moel, the new information conveye in an issuance or repurchase of assets by the insier (firm manager) is fully (partially) incorporate into the estimates of rational (irrational) investors or outsiers. The ifference between the estimate of irrational investors an the funamental value of the asset is the sentiment inicator. In equilibrium, pricing epens on the level of irrational investors participating in the market. If the sentiment inicator increases, with short-sales constraints for outsiers, the weight of irrational traes in the pricing function increases an, since they only partially incorporate the new information, the price impact is lower than in the case of rational investors participation. Consequently, the market becomes more liqui an firm managers may time the market by issuing in those moments. This is precisely our argument. Baker an Stein provie inirect empirical support for their moel, showing that i) aggregate measures of the turnover in the stock market an the number of equity issues are high an positively correlate, ii) high values of turnover are associate with moments of equity overvaluations (low ivien yiel), an iii) both turnover an the number of issues are negatively relate to next year s equity returns. The aim of this paper is to show that Baker an Stein s (2004) argument also applies to the ebt market. 3 Using all U.S. aily transactions registere in the Trae Reporting an Compliance Engine (TRACE) ataset covering the perio between July 1, 2002, an December 31, 2009, our results show that liquiity in both the stock an bon markets oes influence the ebt issuance moment an volume, so that managers seem to be liquiity timers. This result is very consistent an robust to ifferent ways of measuring liquiity, ifferent estimation methoologies, an the inclusion of many control variables. Aitionally, we analyze the benefits of timing an show that liquiity timers issue ebt at abnormal high prices. We fin that, on average, the yiel sprea at the moment of issuance is significantly higher for nontimers an that the yiel sprea for timers increases substantially in the three years following the issuance. The paper is structure as follows. Section II escribes the ata an methoology use in the empirical analyses. Section III presents the main results regaring the relation between market liquiity measures an the ebt issuance ecision. Section IV compares bon an firm characteristics between firms that time their ebt issuances an those that o not an analyzes the temporal evolution of yiel spreas after the issuance. Section V runs further robustness checks, incluing an alternative illiquiity measure, ifferent time winows to account for illiquiity, alternative ways 3 Similar arguments are use by Gao an Lou (2013). They also employ a proxy of stock price pressures mutual funs flow-inuce traing for investigating the timing of both equity an ebt issues. 4

5 of estimating the moels, an alternative measures for some of the control variables consiere. Finally, Section VI summarizes an conclues the paper. II. Data an methoology A. Bon ata We employ the TRACE ataset, which compiles information on all corporate bon transactions taking place in the U.S. market. TRACE is a system with which all their members must report over-the-counter corporate transactions in the seconary market, following the rules approve by the U.S. Security an Exchange Commission in January The increase in the quantity an quality of information provie by the TRACE system allows empirical stuies to be conucte relating to the microstructure of the U.S. corporate bon market. In that sense, the implementation of the new system provies an opportunity to analyze improvements in the transparency an liquiity of the market. In aition, the intraay information allows liquiity measures an their components to be estimate more efficiently (Bessembiner et al., 2006; Dick-Nielsen et al., 2012; Ewars et al., 2007; Golstein et al., 2007). We consier all intraay information transactions that occurre between July 1, 2002, an December 31, 2009, an eliminate erroneous reporting ata by applying the filters propose by Dick-Nielsen (2009). Next, we use further filters to reuce the final number of bons in our sample. First, we iscar all bons that are perpetual, with variable coupons, asset-backe issues, bons enominate in a foreign currency, an those that inclue any type of option. Secon, we require the bons in our sample to be trae frequently in orer to compute liquiity measures. In particular, we require the bons to have at least one year of ata in TRACE an to show traing activity on at least 75% of the business ays. This results in a sample of 707 bon issues. However, two aitional bons are exclue ue to outliers in their prices. Therefore, the final sample consists of 705 issues. 5 In aition to the TRACE transaction ataset, we use the FISD (Fixe Income Securities Database) ataset to obtain iniviual bon characteristics such as the issue ate, issuance size (par an ollar values), issue yiel, the coupon rate, maturity, the issuer inustry, an the historical creit ratings for each issue. 4 The actual reporting starte in July 2002, with the issemination of all traes in bons with an initial issuance above $1 billion an in the 50 high-yiel bons. In March 2003, the issemination was extene to traes in bons with an initial issuance above $100 million an a rating of at least A-. Finally, the issemination was complete in October 2004, with 99% of all traes reporte in real time. As of July 2005, in real time means that traes are reporte within 15 minutes. 5 Specifically, the two exclue issues have ID numbers an

6 Table 1 reports summary statistics for our sample of bons an for all bons in the TRACE ataset. The bons in our sample show a larger issuance volume an, given the selection criteria, are more frequently trae than the average of all bons in TRACE. Out of the 1,892 traing ays in our sample perio, the selecte bons were trae, on average, uring 873 ays, while in the whole sample the average number of ays with transactions was 172. The average coupon, yiel, an sprea regaring the benchmark Treasury issue are similar in our sample an in the whole sample. The average maturity of our bons is 7.89 years, which is lower than the average maturity of the full sample in TRACE, with much lower ispersion. Finally, the bons in our sample represent 12 out of the 33 inustries in the whole ataset. [Insert Table 1 aroun here] B. Illiquiity measure A traitional ebate in the asset pricing literature iscusses the empirical proxy that is use to measure illiquiity. Besies the well-known bi ask sprea measure, popular proxies associate the size of the trae with the size of the price change. This price impact approach is base on the classic theoretical paper of Kyle (1985), which estimates the averse selection component of liquiity by linearly relating the net orer flow to the price variation. One of the most wiely use proxies of price impact is the Amihu (2002) ratio. This proxy has been extensively employe in papers about liquiity on the stock market 6 because of its avantages from a practical point of view: It can be easily compute for long perios since it only uses information on aily rates of returns an traing volume. Its use as an iniviual or aggregate illiquiity measure has also been recently translate to the bon markets. Among others, Jacoby et al. (2009), who uses the Amihu ratio for testing Acharya an Peersen s (2005) moel, an Dick-Nielsen et al. (2012), who extract the liquiity component from bon yiels by using a combination of four liquiity variables, one of which is the Amihu ratio. For each of the 705 selecte bons, we construct the Amihu ratio to proxy for the iniviual illiquiity measure on a aily basis. This ratio relates the absolute return of the bon to its traing volume (in ollars). When a big change in asset price is neee to accommoate the magnitue of the traing volume, the asset shows low levels of liquiity. For each bon j an each ay, the Amihu ratio is obtaine as: 6 The Amihu (2002) ratio is use as a proxy for liquiity in the stock market by Acharya an Peersen (2005), Kamara et al. (2008), Korajczyk an Saka (2008), an Watanabe an Watanabe (2008), among others. 6

7 R DVol where R is the absolute aily return an DVol j, is the total ollar volume trae j, uring ay. The return is compute using close aily prices an the cumulative traing volume is obtaine by aing all transactions uring the ay, on both a par an a ollar basis. To obtain a measure representing the average illiquiity of a bon in the recent past, the aily ratios are average over all ays within a winow of D previous ays: Ami D j, R 1 1 j, D DVol D j, We consier an initial winow length of three months. To compute the average measure, we require a minimum number of aily observations to be within this winow. In particular an consistent with Amihu (2002), we require that the bon be trae on the basis of at least 68% of the business ays in the corresponing winow. 7 Since we aim to analyze the effect of illiquiity of corporate bons on the issuance ecision, we construct two aggregate measures of illiquiity. The first represents the level of illiquiity in the whole corporate bon market an is compute as the cross-sectional mean of all iniviual ratios: Ami where N inicates all bons in the sample. 1 N WD D Ami j, N j 1 The secon measure represents the level of illiquiity for a given rating category. We use the historical rating changes provie by Mergent Bon Source, which inclues Stanar &Poor s, Mooy s, an Fitch ratings. We assign the average rating between these three agencies for each bon each ay. The illiquiity measure for each rating category is compute as the average across all bons within this rating category: 8 Ami R N RD 1 D R Ami j, N j 1 7 In Amihu s (2002) original work, at least 15 prices within a month (22 business ays) are require, which represents 68% of the total number of ays. 8 When there is no rating information on the ate of the issuance an so as not to lose ata, we assign the rating level observe on the nearest ay of any of the following five working ays. 7

8 where N R represents the number of bons in the rating class R. Aitionally, we employ an aggregate illiquiity measure for the stock market. As shown in Gao an Lou (2013), there is a cross-market timing effect because both equity an ebt are affecte by the same non-funamental shocks. Moreover, Acharya et al. (2013) fin that corporate bon returns react to liquiity shocks in the stock market. Therefore, we use the Center for Research in Security Prices (CRSP) to obtain the aily stock prices, returns, an traing volumes for the 162 issuers in our sample. With this information, we estimate the Amihu illiquiity ratio for each issuer an obtain an aggregate measure as the cross-sectional average of all iniviual ratios. This measure is represente by SD Ami. Table 2 shows the usual escriptive statistics regaring liquiity measures for the seven rating categories, both for the bon market as a whole an for the stock market. As observe, there is a quasi-monotonous relation between illiquiity an risk. The exception is the BB-rate category, but this group is one of the less representative ones because a very low number of bons in our sample have this rating level. 9 Generally, liquiity measures show positive skewness an excess kurtosis for all rating classes, with the exception of AAA bons. The correlation between ifferent measures is higher for ajacent rating categories, while the time-series pattern of the liquiity measures is very ifferent between AAA bons an bons rate C or below. In fact, the liquiity of AAA bons is negatively correlate with aggregate stock market liquiity, while the remaining rating categories show a positive correlation with the stock market measure. The correlation between the liquiity of bons rate C or below an stock market liquiity is especially high. Therefore, in terms of liquiity, the riskiest bons behave similarly to stocks, so that a flight to liquiity from stocks to AAA bons takes place. [Insert Table 2 aroun here] C. Controls 1. State variables Empirical evience shows that variables representing economic an/or financial conitions can influence the timing of issuance ecisions. For example, Baker et al. (2003) fin that inflation, short-term real interest rates, an term sprea preict one-year ahea excess bon returns an these variables are also relate to the maturity of ebt, 9 The number of bons in each rating category changes over time. On average, there are 39, 80, 202, 45, 13, seven, an eight bons in each category from AAA to C an below. 8

9 suggesting some egree of timing behavior when selecting ebt maturity. Barry et al. (2008) provie evience that the amount of ebt issue is substantially higher when interest rates are low. However, the evience of interest rate-base timing behavior is not clear. Barry et al. (2009) o not fin timing ability by examining interest rate changes subsequent to corporate ebt issuances an Song (2009) shows that timers in terms of the term sprea o not show higher firm values. Aitionally, since the work of Fama an French (1993), unexpecte changes in the term structure of interest rates an in efault risk are consiere the two most common risk factors in etermining corporate bon returns. Base on this evience, we consier the following variables. Daily yiels on U.S. Treasury securities for the eleven available maturities 10 an on Mooy s Aaa an Baa corporate bons are obtaine from the Feeral Reserve s webpage ( TB1 an TB7 refer to the yiel of the one-month an seven years maturity an are use for representing the short- an longterm interest rates. 11 Term is the term sprea compute as the ifference between 10- year Treasury bons an one-month Treasury bill yiels, an Default represents a efault premium an is estimate as the ifference between the yiel on the corporate bon inex an the yiel on 10-year Treasury bons. The yiels on Treasury bons for the remaining maturities are employe to estimate the full yiel curve an compute yiel spreas each ay for all the bons in our sample. In aition to the efault sprea, we also control for VIX as an alternative proxy for risk expectations, which is a measure of implie volatility in inex options that is calculate by employing moel-free techniques. Bao et al. (2011) fin that the most relevant relation is that between aggregate bon illiquiity an VIX. Close aily VIX prices are obtaine from the Chicago Boar Options Exchange website ( Inflation refers to actual inflation. Monthly series for inflation are compute from the chaine Consumer Price Inex (all urban consumers) available in the atabase provie by the U.S. Bureau of Labor Statistics. Finally, we inclue the stock market return (RM ) an inustrial prouction growth (IPIgrowth ) as inicators of financial an economic cycles, respectively. Daily returns on the value-weighte stock market portfolio are from Kenneth French s webpage ( Monthly ata of the inustrial prouction inex are from the Feeral Reserve, with series ientifier G17/IP Major Inustry Groups. 10 One, three, an 6 months, an one, two, three, five, seven, 10, 20 an 30 years. 11 Seven years is the maturity most similar to the average maturity of the bons in our sample. 9

10 We assign the monthly values for the variables IPIgrowth an Inflation to all ays within the corresponent month. Interest rates, market return, an VIX values are in annual basis. All the state variables are reporte in percent. 2. Bon characteristics We inclue the usual characteristics consiere in the literature: Maturity j, inicates the number of years to maturity of the corresponing bon, each ay, an Yiel j is the offering yiel of each bon, in percent. 3. Firm characteristics The variable Cash j, enotes cash an short-term investment accounts (item 1); CapExp j, refers to capital expenitures (item 128); Ebita j, is the operating income before epreciation (item 13); Tangibility j, refers to the sum of property, plant, an equipment (item 8) an inventories (item 3). These four variables are normalize by total assets (item 6). Size j, refers to the log of total assets (item 6); LeverageR j, refers to the ratio between book ebt (long-term ebt, item 9, plus ebt in current liabilities, item 34) an the sum of book ebt an the market value of equity (using ata from the CRSP to obtain the latter); an TobinR j, is compute as the sum of the market value of equity (using ata from the CRSP) an book ebt ivie by total assets. With the exception of the market value of equity, obtaine from the CRSP, ata regaring all the remaining variables are obtaine from the Compustat atabase on an annual basis. All the variables are measure at the beginning of the fiscal year (as in Barry et al. (2009) an Doukas et al. (2011)). Due to the unavailability of a sufficient number of observations, we are not able to incorporate variables such as research an evelopment expenitures an the proportion of fixe assets. The expecte impact of these control variables on ebt issues is as follows. Doukas et al. (2011) inicate that firms issue ebt even when internal funs are high, so a positive impact between cash an ebt issues is expecte. Regaring capital expenitures, the previous literature has foun a positive although not significant coefficient. Earnings before interest, taxes, epreciation, an amortization (Ebita) are expecte to have a positive impact on ebt issues, since the tax euctions for the interest payments of more profitable firms are higher (Moigliani an Miller (1963)), so these firm are expecte to issue more ebt even though they have more internal funs (Doukas et al. (2011)). The traitional literature on capital structure shows a positive relation between size an ebt issues (Frank an Goyal, 2003; Rajan an Zingales, 1995; Titman an Wessels, 1988), but Doukas et al. (2011) fin a negative relation, so that larger firms ten to epen on internally generate funs. Aitionally, the literature on capital structure points out that firms with more tangible assets that can be 10

11 use as collateral are expecte to issue more ebt (Myers an Majluf (1984)), which is also in line with Gao an Lou (2013). However, Doukas et al. (2011) fin the opposite result an they state that firms in inustries with highly tangible assets ten to issue less ebt. Accoringly, firms are more likely to issue ebt when their past leverage is relatively high an the authors state that this positive relation seems to be in accor with the iea that ebt financing coincies with favorable ebt market conitions. Finally, since the higher Tobin s Q, the higher a firm s growth opportunities, firms with a high Tobin s Q ratio are expecte to issue less ebt to protect their growth options (Doukas et al., 2011; Hovakimian, 2006). D. Methoology To analyze the effect of liquiity conitions on corporate bon issuances, we regress both the issuance moment an issuance volume on the three variables measuring illiquiity, controlling for the state, bon, an firm variables specifie in the previous subsection. The most general moel takes the form IssueM f (A), Liquiity Variables, State Inicators, Bon Characteristics, Firm Characteristics j where IssueM represents the issuance moment an is a ummy variable that takes the value of one on the ay of each bon s issuance an zero otherwise an is a noise process assume to obey stanar restrictions. We also estimate the following moel: IssueV f (B), Liquiity Variables, State Inicators, Bon Characteristics, Firm Characteristics j where IssueV represents the issuance volume an is obtaine as the total amount of the issuance (bon s face value) on the ay of the issuance, normalize by total assets, an zero otherwise. This moel is estimate because issuance size allows us to exploit information that varies cross-sectionally. Moreover, Bao et al. (2011) fin that iniviual bon liquiity is significantly relate to its issuance size. Moel A is estimate using probit panel ata regressions with ranom effects. Moel B is estimate by consiering either fixe or ranom effects, as inicate by the Hausman test. In both moels, the errors are robust to autocorrelation an heteroskeasticity. The istinctive avantage of the panel ata methoology is that it allows us to control for unobservable iniviual heterogeneity in the ata, that is, it allows the most general specification, 0 i. 11

12 III. Market liquiity an ebt issuances Moels 1 to5 in Table 3 show the results from the estimation of equation (A) (Panel A) an equation (B) (Panel B) when only liquiity variables are consiere: the aggregate illiquiity of the whole corporate bon market, the aggregate illiquiity for the class of bons within the same rating, an aggregate illiquiity in the stock market. The p-values regaring iniviual significance are reporte in parentheses. The last row of Panels A an B provie the log-likelihoo values an p-values of the Hausman test, respectively. [Insert Table 3 aroun here] The results inicate that liquiity in financial markets is significantly relate to the timing of corporate bon issuances. Illiquiity measures regaring the bon market are negatively relate to the moment of issuance, inicating that the likelihoo of corporate bon issuance is significantly higher when liquiity in the corporate bon market is also high. In contrast, corporate bon issuances are significantly associate with ecreases in the liquiity of the stock market. The results are consistent when various liquiity measures are combine an also when the epenent variable is the issuance volume. 12 We also fin that the effect of stock market illiquiity on both the moment an the volume of ebt issuances is stronger than that of bon market illiquiity. Therefore, in line with Gao an Lou (2013), we fin that shocks in the stock market affect the timing ecision in the bon market. Also our finings are consistent with some of the results of Acharya et al (2013). These authors also fin that corporate bon returns respon more strongly to liquiity shocks in the stock market than in the bon market an that the illiquiity in both markets negatively affects corporate bon returns, although this is true for junk-grae bons an not for investment-grae bons. Our regressions in Table 3 o not istinguish between rating classes. However, given that AAA bons liquiity is negatively correlate to stock market liquiity, the negative parameter associate with stock market illiquiity can be interprete as a flight to liquiity phenomenon from the stock markets to the bon markets, on the one han, an from high-risk to low-risk bons, on the other han. Moels 6 an 7 inclue also a selection of the cycle inicators that respons to the finings of a preliminary analysis that evaluates the iniviual timing capacity of each of the state variables escribe in Section II.C. In this analysis we fin that the ecision to issue ebt is relate to perios of positive expectation about future economic conitions, measure by increases in the Term sprea or IPIgrowth an ecreases in 12 We o not inclue the two illiquiity measures in the bon market simultaneously to avoi multicolinearity problems. The results regaring the aggregate measure for the whole bon market lea to the same conclusions an are available upon request. 12

13 Default sprea or VIX. The level of interest rates, the inflation rate, an the stock market return o not seem to etermine the ebt issuance moment. The relations with the issuance volume are much weaker but this variable is also higher for perios of low volatility in the stock market. The results for Moels 6 an 7 in both Panels A an B of Table 3 confirm that both the moment an volume of ebt issuances are significantly relate to market liquiity conitions, inepenent of the state variables consiere. Aitionally, Default an VIX are negative an significantly relate to ebt issuances an IPIgrowth is no longer relevant when liquiity variables are inclue. It also seems that firms issue higher volumes of ebt when the Term sprea is lower, but this variable becomes significant only when Default is not inclue in the moel. 13 Finally, the complete specification to test for the relevance of the liquiity measures inclues cycle inicators, bon characteristics (issue yiel an maturity), an several firm characteristics (as escribe in Section II.C). Table 4 provies the estimation results for a selection of five representative moels. Tables 4a an 4b refer to the issuance moment an issuance volume, respectively. Consistent with previous results an inepenent of the combination of explanatory variables, the moment of issuance is positively associate with perios of high liquiity in the bon market, specifically for bons within the same risk class, an perios of low liquiity in the stock market. Again, VIX is always negative an significant while Term is no longer relevant in the presence of aitional explanatory variables. Regaring bon characteristics, maturity oes not seem to etermine the issue moment, but the bons issue show significantly higher yiels than the rest. In general, there are no relevant associations between the moment chosen for the issuance an firm characteristics. The only significant relations are with Ebita an Leverage; we fin that firms showing high earnings or a high leverage ratio at the en of the previous year are more likely to issue ebt. As Doukas et al. (2011) fin, although the result regaring the positive sign of leverage is not consistent with the trae-off hypothesis or with the view that firms actively rebalance their leverage to stay within an optimal range, it coul be consistent with the iea that ebt financing coincies with favorable ebt market conitions. Although the results of Doukas et al. (2011) refer to the volume of ebt issue an not the moment, this woul be in line with our proposition, since firms woul issue ebt at perios of favorable liquiity in the bon market. 13 The serial correlation between Default an VIX in our sample perio is 90%. For this reason we avoi the use of the two variables simultaneously an consier only VIX in the regression analyses. The results incluing Default instea of VIX are similar an available upon request. 13

14 Regaring regressions of issue volume an consistent with previous results, in all the moels, bon market illiquiity by rating is significant an the volume issue is higher when the stock market liquiity is lower. Term is negatively relate to the issuance volume but this relation is not significant when firm variables are inclue in the moel. We also fin that the volume issue is positive an significantly relate to the maturity of bons, to Ebita, as with the issuance moment, an to firm size (as for Ooi et al. (2010)). Finally an in contrast to the results for the issue moment, the leverage ratio is not relate to the issuance volume. [Insert Tables 4a an 4b aroun here] A. Is the crisis perio special? Our sample perio is characterize by a very peculiar episoe, the recent strong crisis, uring which firm financial nees grew heavily. Therefore it seems reasonable to think that both the moment an the volume of ebt issues coul have been affecte uring this particular perio. In aition, Acharya et al. (2013) show that the effects of liquiity shocks on bon prices are only relevant in perios of stress. Although we previously inclue cycle inicators in the regressions that alreay take this fact into account to some egree, it is interesting to analyze whether ebt issues are specifically associate with crisis perios. To o so, we repeat the estimations in Tables 4a an 4b but inclue a ummy variable that equals one in recession perios; it is constructe from the recession ates provie by the National Bureau of Economic Research. Tables 5a an 5b show the results for the issuance moment an issuance volume, respectively. The ummy for the crisis generally has a negative sign but is not significant. Therefore, firms o not ten to issue more ebt in recession perios. No changes worth mentioning are foun for the rest of the parameter values an their significance an, thus, the conclusions for the illiquiity variables remain unaltere. [Insert Tables 5a an 5b aroun here] B. Financial versus non-financial firms Finally, it is also interesting to investigate whether the liquiity timing effect iffers between financial an non-financial issuers. Given the liquiity requirements in the sample selection process, most of the bons in our sample are issue by financial firms (aroun 75% of our sample). Aitionally, it is known that financial firms suffere the worst financial problems uring the crisis. Thus, financial an non-financial firms coul be behaving in ifferent ways. Tables 6a an 6b provie the estimation results for the issuance moment an issuance volume, respectively, but inclue a ummy that equals 14

15 one if the issuer is not in the financial inustry an zero otherwise, an also inclues interactions between this ummy an the illiquiity variables. The ummy alone captures the ifferences between financial an non-financial firms through the constant term, while the interactions represent the ifferential effect of the liquiity variables for financial an non-financial firms. As observe in Table 6a, no special conclusions can be rawn for non-financial firms. The parameters associate with the ummy an its interactions are not significant in any moel, so we can conclue that firms in all inustries choose perios of high liquiity in the bon market an/or low liquiity in the stock market to issue ebt. The significance of the control variables remains unaltere, with the exception of the leverage ratio, which now becomes insignificant. Therefore, the positive relation foun previously between the issuance moment an firm leverage ratio coul inicate inustrial ifferences between firms, since it is known that the leverage ratio is substantially higher for financial firms. 14 Now, these ifferences are represente by the intercept an the ummy variable instea. Regaring the issuance volume, again, no significant ifferences between financial an non-financial firms are foun. The conclusions regaring the control variables remain unaltere an the liquiity variables continue to be significantly relate to ebt issuances, although the variable relate to bon liquiity is only significant at the 10% level in the moels that inclue firm variables. [Insert Tables 6a an 6b aroun here] IV. Liquiity timers an nontimers: Forwar-looking ebt timing The results in the previous analyses inicate a significant an positive relation between the moment when corporate ebt is issue an liquiity in the bon market. However, to conclue that a liquiity timing pattern exists, it is necessary to show that corporate bons are overprice in moments of high liquiity, so that firms can benefit from issuing at an abnormally low cost. To analyze if this is the case, we split our sample into two groups of bons, istinguishing between those issue in perios of high liquiity in the corporate bon market (timers) an those issue in perios of low liquiity (nontimers). We use the meian value of the aggregate illiquiity within the same rating class to classify firms as timers or nontimers The mean leverage ratios for financial an non-financial firms are 0.60 an 0.07, respectively, with meian values of 0.64 an 0.04, respectively. 15 We use the meian value for splitting the sample since it prouces a similar number of issues in each subsample: 204 timers an 183 nontimers. Other partitions, such as the first an thir quartiles, prouce highly asymmetric subsamples; however, the results are robust to this alternative efinition. 15

16 The comparison between the ifferent characteristics of the bons an issuers in each subsample is shown in Table 7 an is base on the signe-rank test for the meian values. All the variables are measure at the moment of issuance an refer to issue characteristics in Panel A ays until maturity, price, volume (in millions), yiel, rating (average rating from S&P, Fitch, an Mooy s), an yiel sprea, that is, the ifference between the yiel of the benchmark Treasury issue an the issue s offering yiel an issuer characteristics in Panel B Cash, CapExp, Ebita, Size, Tangibility, LeverageR, an TobinR, as efine in Section II. For each characteristic, we provie the mean an meian values for the two subsamples an the p-values associate with the null of equal meians. The results show that there are no relevant ifferences in the maturity, offering price, an volume of bons issue by liquiity timers an nontimers. The offering yiel is higher for timers, but it shoul be taken into account that the series of both timers an nontimers contain issuances associate with perios of ifferent interest rate levels. Regaring rating, we fin that nontimers have slightly better rating positions at the moment of issuance. 16 Finally, the most important result in Table 7 refers to the yiel sprea. We fin that liquiity timers are able to issue ebt at a much lower cost. The meian ifference between the corporate bon yiel an the equivalent Treasury bon yiel is approximately 83 basis points for bons issue in perios of high bon market liquiity. In contrast, the yiel sprea is 95 basis points if the issuance is unertaken uner poor liquiity conitions. Therefore, the use of information relate to aggregate liquiity when eciing the offering ate reuces the creit sprea. 17 The results regaring the firm characteristics o not show significant ifferences between timers an nontimers, except for Tangibility, with timers showing larger values. [Insert Table 7 aroun here] A further analysis is performe to show that the lower yiel sprea for timers is associate with the overpricing of bons issue in moments of high market liquiity. For each ay uring the three years following the ebt issuance, we collect the transaction yiels from TRACE an compute yiel spreas by subtracting the Treasury yiel with exactly the same number of ays until maturity. 18 Then, for all the ays between the issuance ate (ate 0) an 36 months after the issuance, we compute the 16 A rating of five correspons to A + an a rating of four to AA - in the S&P an Fitch rankings. 17 To confirm that the ifference in yiel spreas between timers an nontimers is not ue to the criterion employe to split the sample, we reclassify issues using two alternative illiquiity measures: the mean an the meian of the aggregate measure for the whole bon market. The results are robust to measures of liquiity compute in winows of ifferent length. 18 For each ay, the full Treasury yiel curve is estimate by linear interpolation using the yiels on Treasury securities for the 11 available maturities. 16

17 average yiel sprea for the issues classifie as timers an nontimers. Figure 1 represents the time evolution of the two series. The yiel sprea of timers is lower than that of nontimers at the moment of issuance, as shown in Table 7, as well as uring the year subsequent to the issuance ate, a perio in which the yiel spreas of both timers an nontimers rise. However, after a year from the issuance, the yiel sprea of bons issue by timers continues to rise while that of nontimers reverts to its long-run mean. 19 In the case of timers, the increasing tren inicates that bons are overvalue at the issuance moment an confirms the success of managers liquiity timing ecisions. V. Robustness issues [Insert Figure 1 aroun here] In this section we consier some alternative issues to ensure the robustness of the results. In particular, we analyze the effect of alternative winows for measuring liquiity; an alternative proxy to measure liquiity; an alternative estimation methoology (cluster methoology); an alternative measures for certain firm characteristics. Overall, we fin that the results are robust to all these consierations. 20 A. Alternative winows for measuring liquiity In this subsection we explore the sensitivity of our results to ifferent winow lengths when computing the Amihu illiquiity measures, of one, six, an 12 months. The most remarkable finings are as follows. Focusing on the issuance moment, the significance of the liquiity variables, for both bon an stock markets, hols for all winow lengths. This is the case not only when the liquiity variables are consiere alone, but also when all the control variables are inclue. Generally, the coefficients are higher for the bon illiquiity measure an lower for the stock illiquiity variable as the winow length increases. Regaring the issuance volume, the results for the onemonth winow are practically the same. For the six- an 12-month winows, we fin that the bon illiquiity measure is only significant when the firm variables are not inclue, while the stock illiquiity maintains a significant relation with the issuance volume in almost all the moels. Aitionally, VIX is not relevant (except for some moels) an the relations with the firm variables are very similar for these alternative winow lengths. Finally, the ifferent evolution over time of yiel spreas between 19 The yiel sprea changes from the moment of issuance to two an three years after are an 3.302, respectively, for timers an an 0.257, respectively, for nontimers. These values are significantly ifferent between the two groups. 20 We o not report the results for all the robustness checks to save space, but they are available upon request. 17

18 timers an nontimers provie in Figure 1 is also confirme when liquiity is measure along these other winow lengths. B. Alternative proxy for the illiquiity measure In this subsection we use an alternative proxy for measuring iniviual bon illiquiity, following that propose by Bao et al. (2011), a measure also associate with the averse selection component of liquiity. Uner the assumption that bon prices consist of two components, one representing the funamental value that follows a ranom walk an the other, which is transitory, representing the impact of illiquiity on price, these authors approximate illiquiity as the negative covariance between consecutive price changes: Cov p, p j j,t j,t 1 Consistent with the Amihu measure, Bao s illiquiity measure for bon j on a given ay is compute by consiering the price changes for all the ays in the last three months. We also construct aily aggregate measures as the average of the whole bon market, WD j, RD, an of all bons within the same rating category, j,. Aitionally, SD we compute Bao s aggregate measure for the stock market, j,. Figures 2 to 4 compare Amihu s an Bao s illiquiity measures. For the whole bon market (Figure 2) both measures show a similar pattern over time an present a strong co-movement with the business cycle. 21 Figure 3 confirms the similarities between the two illiquiity measures in the bon market in the cross-sectional pattern within rating categories. 22 The regression analysis when using the Bao s measure instea of the Amihu s one also show that both the issuance moment an volume are associate with high liquiity in the bon market (either consiering an aggregate measure for the whole market or an aggregate measure within the same rating class). The strong relation for the issuance moment also hols when both the cycle inicators an firm characteristics are inclue in the regressions. However, regaring the issuance volume, all the parameters in almost all the moels are estimate with much less precision than for the Amihu case. 21 For the subsample before August 2007 (pre-crisis perio), the mean an stanar eviation are 0.39 an 0.09 for the Amihu measure an 0.27 an 0.14 for Bao s measure, respectively. For the subsample after August 2007 (post-crisis perio), these values are 0.91 an 0.50 for the Amihu measure an 1.48 an 1.17 for Bao s measure, respectively. See Bao et al. (2011) for more etails about the cyclic behavior of this proxy for illiquiity in the corporate bon market. 22 It shoul be note that the values represente in Figure 3 are the meian values in time for each rating class. 18

19 [Insert Figure 2 aroun here] [Insert Figure 3 aroun here] Differences between Amihu s an Bao s measures are appreciable in Figure 4 for the stock market. Both measures are strongly time varying an present peaks associate with the technological boom at the en of 2002 an the recent U.S. financial crisis. However, Bao s measure is more extremely affecte by the recent 2008 crisis. This explains why Bao s illiquiity measures in the stock an bon markets are highly correlate (0.75) 23 an also that the association between the stock market Bao s measure an timing is positive but weak an tens to isappear when other variables are inclue. Nevertheless, the comparison of the log-likelihoo values between Amihu s an Bao s measures show that when only the illiquiity measures are consiere, the ajustment is always better with the Amihu ratio. C. Alternative estimation methoology [Insert Figure 4 aroun here] We also re-estimate regressions for both the issuance moment an the issuance volume by using an alternative methoology: probit an OLS poole time-series cross-sectional regressions, respectively. These both implicitly assume 0, with two-way cluster- i robust stanar errors accounting for bon an ay clusters. The istinctive characteristic of multi-way clustere errors is that this metho allows us to carry out statistical inference that is robust by esign to simultaneous epenences of unknown form in both the cross-sectional an time-series imensions of the panel. Regression errors are assume to be inepenent but not ientically istribute across a number of clusters an can have fairly general patterns of within-cluster correlation an heteroskeasticity (Cameron et al., 2011; Gow et al., 2010; Petersen, 2009; Thompson, 2011). This methoology prouces still more favorable results in relation to the relevance of the illiquiity variables. For both the issuance moment an volume, we always fin negative relations with the illiquiity measure for the bon market an positive relations with the illiquiity measure for the stock market, after incluing all the control variables an with even higher t-statistics than before. Conclusions regaring the control variables are similar in the case of the issuance moment an the most notable ifferences in explaining the issuance volume are as follows: The yiel is now positive an relevant, while maturity is not; leverage has a positive an significant 23 Moreover, the correlation between the stock an bon Bao measures increases as the winow for the computation of the covariance increases. 19

20 impact on ebt issuances (as in the case of the issuance moment); an, in contrast, we now fin that the larger firm size, the lower the ebt volume issue. D. Alternative measures of some firm characteristics The aim of this subsection is to check the robustness of the results regaring the liquiity timing evience when certain firm characteristics are measure ifferently. This will also allow us to corroborate some striking results foun regaring some firm characteristics. Overall, we fin that only a small number of firm characteristics are significant. Especially surprising are firm size an firm profitability, as well as the fact that firms with high leverage ratios are more like to issue ebt the next year. To ensure that the results are not riven by the way in which the variables are measure, we consier alternative efinitions. Now, we measure Size as the log of sales (item 12). We also consier two other alternative efinitions for Leverage, in which the enominator is either i) total assets (item 6) or ii) the sum of book ebt an the market value of equity (the latter is obtaine with ata from Compustat: common shares outstaning, item 25; by close price, item 199). In the same way, a secon proxy for Tobin s Q, TobinR, incorporates the market value of equity, compute using information from Compustat. Regaring the results for the issuance moment, the illiquiity variables continue to be highly significant, so that the results are robust to these new efinitions. The new proxies for size an Tobin s Q continue to be insignificant in all the estimate moels. The changes in the efinition of the leverage ratio o not affect the results; both alternative measures are positive an significantly associate with the issuance moment, with higher coefficients when the ebt value is normalize by total assets. The results relate to the effect of liquiity on the issuance volume are rather similar to those previously foun, so that both the bon market liquiity an the stock market liquiity are relevant variables, no matter the change in the efinition of the control variables. The only ifferences worth mentioning refer to some firm variables an are probably ue to collinearity problems cause by the new efinitions. For example, the new measure for size shows, as before, positive slopes but is only significant in some moels (those in which Ebita becomes insignificant). The leverage ratio, measure as the book ebt value normalize by total assets, is negative an significantly relate to the issuance volume, but only if Tobin s Q is also inclue in the moel. VI. Conclusions 20