Relatve Influence of Push Attrbutes and Pull Factors on Corporate Debt Issuance Subhankar Nayak Fnancal Servces Research Centre, School of Busness and Economcs, Wlfrd Laurer Unversty 75 Unversty Avenue, Waterloo, ON N2L 3C5, Canada Tel: 1-519-884-0710 ext 2206 E-mal: snayak@wlu.ca Receved: August 17, 2010 Accepted: September 18, 2010 do:10.5539/jef.v3n2p18 Abstract Many corporate events are self-selected, but the relatve mpact of nternal (frm-specfc) and external (envronmental) nfluences on dscretonary corporate decsons remans largely unknown. In ths paper, focusng on corporate debt ssuance, we apply the contemporary push-pull theory to establsh whether bond ssuance are pushed by the frms due to nternal consderatons, or pulled by the entcement of exogenous forces. We explore the mpact of eght frm-specfc push attrbutes and four systematc pull factors on the lkelhood and volume of fve types of debt ssued, and dsentangle the relatve sgnfcance of push and pull effects. We fnd three key results. Frst, uncondtonally, both frm-specfc push attrbutes and exogenous pull factors are relevant. Second, the relatve sgnfcance of push effects are always decdedly larger than those of pull effects. Fnally, pull factors have no condtonng role n how markets react to debt ssuance. Although pull effects do matter, push effects compellngly domnate. Keywords: Push attrbutes, Pull factors, Push-pull theory, Corporate debts, Condtonal event study 1. Introducton Many corporate events are self-selected; that s, these corporate decsons are not random, arbtrary, or compelled, but are delberate and calculated decsons by frms or corporate managers to self-select nto ther preferred choces whle optmzng certan objectves or opportuntes (L and Prabhala, 2007). Corporate managers retan a large degree of dscreton n decdng whether to undertake these events and the tmng of these events f undertaken. Examples of self-selected corporate decsons abound: securty ssuance, dvdend payments, corporate restructurng, stock splts, etc. However, emprcal evdence s mxed on whether the factors that condton the choce or tmng of self-selected corporate events are nternal (frm-specfc) or external (envronmental). Classcal lterature typcally models corporate decsons as push events where the drvng nfluence are endogenous and nternal: based on frm-specfc attrbutes, corporate managers tme an event to maxmze value or utlty, to mnmze rsk, to send a credble sgnal, etc. However, emergng evdence suggest that corporate decsons are also pull events nfluenced by systematc and exogenous factors: managers are motvated to beneft from prevalng market condtons, to cater to nvestor preferences or sentment, or to respond to macro-economc forces. In ths paper, focusng on corporate debt ssuance, we explore the relatve sgnfcance of push and pull factors (.e., the nternal and exogenous nfluences) n the tmng of self-selected corporate events. (Note 1) We focus specfcally on corporate debt ssuance for certan motvatng reasons. Frst, the determnant factors for debt ssuance reman less establshed and hence deally suted for approbaton of push-pull theory. Second, many emprcal fndngs from equty lterature (especally those nvolvng behavoral pull factors) can potentally be valdated based on studes of corporate bonds. Fnally, unlke equty, whch are homogenous, frms ssue dfferent types of bonds, and an emprcal evaluaton of push-pull theory may reveal what drve the choce of bond type. Theory of captal structure owes ts orgn to the semnal work by Modglan and Mller (1958) who lay down the condtons whch make captal structure decsons largely rrelevant to the objectves of maxmzng frm value and mnmzng cost of captal. However, two caveats accompany ths dstngushed result. Frst, the proposed rrelevance s vald under a very rgd set of assumptons. Second (and more pertnent to our purposes), the rrelevance theorem does not preclude frms from ssung ether equty or debt drven by certan motvatng (nternal or external) reasons. In other words, even under the Modglan-Mller rrelevance framework, push and pull varables may nfluence corporate securty ssuance decsons. Subsequent lterature models captal structure and securty ssuance decsons mostly as push events. In partcular, t 18 ISSN 1916-971X E-ISSN 1916-9728
s proposed that, motvated by valuaton, rsk, or sgnalng consderatons, frms ssue debt based exclusvely on frm-specfc performance and rsk characterstcs. Modglan and Mller (1963), Mller (1977), and DeAngelo and Masuls (1980) suggest that accompanyng tax benefts motvate a frm to ssue debt. Recognzng that debt engenders mplct dstress rsk, Leland and Pyle (1977) and Ross (1977) propose that proftable and better performng frms ssue debt to convey a credble sgnal about the frm s qualty. Jensen and Mecklng (1976) develop a framework whereby frms ssue debt to mnmze two opposte forms of agency costs. Based on cost and rsk consderatons, Myers and Majluf (1984) reveal an ssuance peckng order where debt has hgher ssuance prorty than equty. In short, push attrbutes predomnantly condton debt ssuance decson. Nevertheless, the relevance of pull varables (systematc market or macro-economc factors) to debt ssuance cannot be dscounted. It has been conclusvely establshed that bond returns demonstrate broad economc or busness cycle effects (Fama and French, 1989) and regmes (Daves, 2008), and nterest rates are predomnant macro-economc factors that nfluence the volume and type of bond ssued (Brennan and Schwartz, 1977; Brck and Palmon, 1992; Barry et al., 2008). However, very few studes explore the relevance of nterest rates n conjuncton wth frm-specfc push attrbutes. Furthermore, the recent emergence of behavoral lterature has added a new dmenson to the relevance of pull effects to securty ssuance. Evdence suggest that systematc behavoral nfluences bear sgnfcant mpact on stock returns and equty ssuance. Broadly termed as sentment, two forms of such exogenous behavoral factors have been categorzed: (a) market performance or momentum sentment (measured as the long-term return of the aggregate stock market), and (b) nvestor sentment (defned as the degree of nvestor optmsm or pessmsm). Relatve to fundamentals, stocks are overprced when sentment s hgh, and underprced when low sentment regns. Frms tme the ssuance of equty to captalze on these msvaluatons, that s, equty ssuance s sgnfcantly more frequent when market performance s bullsh (Lamont and Sten, 2006), when nvestor sentment s optmstc and stock s overprced (Baker and Wurgler, 2000), or n hot ssuance perods (Ljungqvst et al, 2006). Moreover, n the context of equty ssuance and stock returns, recent lterature emphasze the predomnant role of pull nfluences whch contrast wth the establsh earler evdence that enuncate the mportance of push varables. Although Nayak (2010) documents the sgnfcant mpact of nvestor sentment on bond prces (spreads), lttle s known on how these systematc behavoral factors condton debt ssuance, f at all, and what s the sgnfcance of these exogenous nfluences relatve to frm-specfc attrbutes. In ths paper, we explore the jont role of push and pull effects n corporate bond ssuance. In partcular, we seek to address a few specfc questons. Frst, do systematc pull factors (specally sentment and market momentum) retan drvng nfluence on corporate debt ssuance as well? Second, do both sets of varables (push and pull) reman relevant when condtoned smultaneously, or are push and pull effects partal or complete substtutes? Fnally, f both effects are relevant, then what s the relatve sgnfcance of each effect, and are the behavoral pull factors ndeed the domnant (prmary) causes for securty ssuance as clamed n recent lterature? A few compellng reasons motvate ths project. Frst, as noted, t remans unknown whether and how systematc behavoral factors (lke momentum or sentment) nfluence debt ssuance. In addton, t s also not known whch set of varables (nternal or exogenous) matter more n condtonng the decson of a frm to ssue debt. Second, push and pull effects do not necessarly reconcle and may even conflct. For example, whereas push effects are motvated by optmzaton of value or rsk n the mean-varance effcency framework, pull motvatons are largely opportunstc and orthogonal to mean-varance effcency. Dsentanglng the relatve sgnfcance of push and pull effects enables to clarfy and renforce fundamental corporate objectves. Thrd, snce stocks and bonds are complementary securtes, many emprcal fndngs from equty lterature (especally concernng behavoral pull effects) can be valdated (or challenged) based on studes of corporate bonds. For example, market tmng theory postulates that equty s more lkely ssued when prevalng sentment s postve or equty s overvalued; the corroboratory expectaton (that remans emprcally unexplored) s that bonds are more lkely ssued when sentment s negatve or equty s undervalued. Fnally, although undoubtedly equty ssuance s nfluenced by pull effects, t remans unknown whether the overall captal structure decson depends on pull varables or not. For example, Baker and Wurgler (2002) mplctly assume that bond ssuance s ndependent of sentment and conclude on market tmng effects n captal structure based on solely market tmng of equty ssuance. Such conclusons are not robust unless t s establshed that sentment effects n bond ssuance are mnmal. For our emprcal nvestgaton, we use a large sample of fve types of bond ssues, and defne eght frm-specfc push attrbutes and four exogenous pull factors. Usng condtonal event-study framework, we explore the mpact of the push and pull varables on the types and amounts of debt ssued, and based on three quanttatve metrcs, we dsentangle the relatve sgnfcance of push and pull effects n debt ssuance decson. Our tests engender three key results. Frst, uncondtonally, both frm-specfc push attrbutes and exogenous pull factors are relevant to corporate Publshed by Canadan Center of Scence and Educaton 19
debt ssuance. Second, jont econometrc tests reveal that nternal push attrbutes bear the prmary mpact on the type and amount of debt ssued; external pull forces depct only margnal secondary mpact. The relatve sgnfcance of push effects are decdedly larger than those of pull effects based on all metrcs used. Fnally, although pull factors (margnally) condton the ssuance of debt, they bear absolutely no role n how markets react to such events. To summarze, although pull effects do matter, push effects domnate. Frm-specfc characterstcs and performance attrbutes determne the tmng, type, and amount of bonds ssued to a far greater extent than the entcement of exogenous forces. Corporate debts are largely pushed by the frms than pulled by the systematc market nfluences. We proceed as follows. In Secton 2 we specfy the emprcal model for self-selected events and propose three measures of relatve sgnfcance. Secton 3 descrbes the data sample and defnes the push and pull varables. We summarze the results of emprcal tests n Secton 4. In Secton 5, we address a few robustness ssues. Secton 6 concludes. 2. Emprcal specfcaton and test desgn We adopt the two-stage condtonal event-study model (Acharya, 1988; Prabhala, 1997; L and Prabhala, 2007) as our base econometrc model and modfy t to ncorporate the push and pull effects. Ths model, whch mplements Heckman's (1979) correcton for selecton bases n event studes, relates the decson of undertakng a self-selected event to the economc benefts arsng from t and lnks market reacton to the nformaton content underlyng the event. In ths framework, frm undertakes the event (EVT) of ssung a bond of specfc type f the decson varable EBEN s postve, where EBEN may be nterpreted as the net economc beneft arsng from undertakng the event EVT. A part of EBEN s publcly known pror to the event based on an array of varables X that consttute the pre-event nformaton set. The rest of the nformaton motvatng the event,, s frm s prvate nformaton that s unknown to market. In our extenson, we defne the pre-event nformaton set X that condtons the bond ssuance decson as a cumulatve functon of frm-specfc push attrbutes (S ) and systematc pull factors (L) such that X S L. Thus, frm undertakes the bond ssuance event EVT f EBEN X S L 0 (1) where E( ), the pre-ssuance expectaton of the prvate nformaton, s zero wthout loss of generalty, and X captures the ex ante lkelhood of the event. When the event s undertaken, t reveals more about the event frm s prvate nformaton to the market. Specfcally, the ssuance of a bond by the frm mples that EBEN 0 0 X S L 0 S L. Accordngly, market forms revsed expectatons of the ssung frm s prvate nformaton when t announces the ssuance. Abnormal returns on announcement date are a reflecton of market reacton to these revsed expectatons. That s, event announcement effect AR s related to the condtonal expectaton of the frm s prvate nformaton as follows E AR EVT E S L 0 L (2) where β captures the margnal strength of market reacton, that s, the market reacton to per unt new nformaton revealed. In smple words, a potental ssuer s pror publc nformaton set conssts of frm-specfc push attrbutes and systematc pull factors. The frm decdes to ssue a bond f the net economc beneft expected based on the pror publc nformaton set and frm s own prvate nformaton s perceved to be postve. As a response to the bond ssuance, the market revses ts expectatons of the frm s latent prvate nformaton set and reacts to these revsed expectatons revealed upon the announcement of bond ssuance. Emprcal mplementaton nvolves two steps. The frst stage adopts a probt model for the ssuance of a bond (.e., EVT = 0 or 1) condtonal on a set of frm-specfc push attrbutes S and systematc pull factors L that characterze the bond ssuance decson (equaton (1)). The condtonal resdual from the probt model, computed as the nverse mlls rato (IMR), captures the expectaton of the nformaton revealed by the event, E S L 0. The second stage nvolves lnear regresson of bond ssuance announcement effects AR on the endogenously computed IMR (equaton (2)). Ths smple framework engenders straghtforward emprcal tests to explore the role of push attrbutes and pull factors n corporate debt ssuance. Whether push and pull varables bear any nfluence on corporate bond ssuance decsons and assocated market reactons can be establshed based on the econometrc sgnfcance of the correspondng coeffcents: and n the probt specfcaton (1) and n lnear regresson (2). In addton, the above formulaton also allows the development of quanttatve metrcs to assess the relatve sgnfcance of push and pull effects. We develop modfed versons of three common econometrc measures to evaluate the relatve effects. 20 ISSN 1916-971X E-ISSN 1916-9728
The frst metrc s the D1 measure whch quantfes the ncremental contrbuton of a partal nformaton set or a partal set of varables relatve to the full nformaton set or the complete set of varables. D1 measure for push (pull) effect s defned as 2 R for pull (push) model alone D 1Measure for push (pull) effect 1 (3) 2 R for jont push pull model where R 2 s refer to adjusted R 2 for regressons and pseudo R 2 for probts. For example, to compute D1 measure for push effect, we ft a probt model usng just the pull varables, and another usng jontly all push and pull varables, and plug n estmated pseudo R 2 n equaton (3). The value reveals the ncremental explanatory power of the push effect. The second and thrd metrcs are the shock and mean effects ratos whch reflect the ncremental mpacts of shocks to and shfts n the estmated model coeffcents. The two ratos are defned as Shock rato for push (pull) effect Mean effects rato for push (pull) effect Effect of 1 shock n push (pull) model Effect of 1 shock n push pull model Effect of 1 shft n push (pull) model Effect of 1 shft n push pull model where and denote the standard devatons and means of the ndependent varables used n the regresson or probt specfcaton. Effects of 1 shock (1 shft) are computed as the cumulatve change n the value of the dependent decson varable arsng from one standard devaton (one mean) postve change n each of the ndependent varables. In each rato, the numerator s the total effect usng just the push or pull varables separately and the denomnator s the cumulatve effect usng all push and pull varables jontly. It s worth emphaszng that (a) each of the three metrcs s computed separately for push and pull effects, (b) has a numercal value rangng between 0 and 1, and (c) the larger s the value of the metrc, the greater s the relatve sgnfcance of the assocated effect. D1 measure reveals the relatve statstcal sgnfcance of push and pull varables, and shock and mean effects ratos depct the relatve economc sgnfcance. (Note 2) 3. Data and varables 3.1 Event Sample We use a large 14-year sample (coverng 1994 through 2007) of corporate bond ssues obtaned from the Mergent Fxed Investment Securtes Database (FISD). We collect ssuance related nformaton (n partcular, ssuance date and offer amount) on fve types of bond ssues: straght bonds, convertble bonds, bonds wth call or put provsons, and bonds ssued under SEC Rule 144A. Adoptng standard norms, we exclude non-corporate bonds, foregn ssues, and bonds wth non-standard optons and features. We also drop bond ssues that are n default or possess close-to-default bond ratngs. For bond ratngs, we use Standard & Poor s (S&P) ratng f t exsts; otherwse we use Moody s ratng data. We convert alphanumerc ratngs nto nteger ratng values where lower ratng values denote better qualty (smaller default rsk). Based on 6-dgt CUSIP dentfers, we match the corporate bonds wth the stock prce data n the Center for Research n Securty Prces (CRSP) database, and the fnancal and accountng nformaton n Standard and Poor s Compustat database. From CRSP, for each bond ssuer, we collect the tme-seres of daly stock returns pror to and around the bond ssuance date. Compustat provdes a large array of ssuer-specfc balance sheet and ncome statement varables. We focus on corporate bonds ssued by publcly traded frms, and hence exclude bond ssues that do not have any match n ether CRSP or Compustat or those wth nsuffcent hstorcal data. 3.2 Frm-specfc push attrbutes The lst of potental frm-specfc determnants of securty ssuance decsons s long. However, motvated by the postulates of semnal theoretcal papers on captal structure (Jensen and Mecklng, 1976; Mller, 1977; Leland and Pyle, 1977; Ross, 1977; Myers and Majluf, 1984), followng the recommendatons of recent emprcal papers (Fama and French, 2002; 2005; Goyal and Frank, 2009), and based on anecdotal evdence, we dentfy eght frm-specfc push attrbutes as relevant varables. Pror evdence engender the expectaton that larger and older frms wth greater proftablty, smaller book-to-market rato, hgh current leverage, more free avalable cash, good credt qualty ratngs, and undervalued equty are more lkely to ssue debt (specally straght or callable bonds). On the other hand, bonds that provde addtonal ncentves to nvestors (convertbles or bonds wth put optons) or sem-prvately placed bonds (Rule 144A ssues) are more lkely to emanate from smaller and newer frms wth relatvely lower proftablty, poor credt qualty, and good equty performance (for tmng reasons). We compute these eght (4) (5) Publshed by Canadan Center of Scence and Educaton 21
frm-specfc push attrbutes for each bond ssuer as follows: 1. Age: lstng perod n CRSP n years; 2. Market captalzaton: logarthm of total equty captalzaton; 3. Proftablty: rato of operatng ncome before deprecaton to total assets; 4. Tobn s Q: rato of total market value of debt and equty assets to recorded book value of assets; 5. Debt rato: rato of all labltes (current and long-term) to total assets; 6. Free cash rato: rato of retaned earnngs to total assets; 7. Ratng value: numercal value of the bond s credt ratng; and 8. Pre-ssue runup: one-year pre-ssue cumulatve abnormal return (CAR) computed as the sum of excess of stock returns over market (CRSP value-weghted ndex) returns from 250 to 3 tradng days pror to the ssuance day. 3.3 Systematc pull factors Snce nterest rate s the prmary macro-economc varable that bears sgnfcance for fxed ncome securtes lke corporate bonds, the two nterest rate factors that are pervasvely adopted n extant lterature (e.g., Fama and French, 1987; 1993) are: (a) default factor (whch captures the systematc credt rsk or probablty of default), and (b) term factor (whch reflects maturty or term-structure effects of long-term nterest rate curves). Furthermore, recent emprcal evdence suggest that frms are more lkely to ssue equty when aggregate equty market demonstrates overvaluaton, postve momentum or bullsh performance, or when nvestor sentment s optmstc (e.g., Baker and Wurgler, 2002; 2006; Lamont and Sten, 2006). By extenson, these are lkely determnants of bond ssuance as well. Consequently, we defne and use the followng four systematc pull factors: 1. Default factor: computed as the dfference between Moody s BAA yeld and 10-year swap rate (obtaned from Datastream); 2. Term factor: obtaned as the dfference between 10-year and 2-year swap rates (obtaned from Datastream); 3. Market momentum: calculated as the aggregate return on the CRSP value-weghted ndex over the year pror to bond ssuance; and 4. Investor sentment ndex: Baker and Wurgler (2006) ndex that captures the systematc level of nvestor optmsm or pessmsm. 4. Emprcal results Our fnal sample conssts of 6,451 bond ssues (1,414 straght or plan-vanlla ssues, 1,892 convertble ssues, 4,262 callable bonds, 855 bonds wth put optons, and 2,377 bonds ssued under Rule 144A) over the 14-year perod from 1994 through 2007 wth an average offer amount of $292.51 mllon, coupon of 6.44%, maturty of 12.28 years and yeld spread of 73 bass ponts. Non-straght bond ssues may possess more than one opton or feature; for example, 1,448 convertbles are also callable, and 828 convertbles are ssued under Rule 144A. Based on ndustry, the sample conssts of 4,774 Industrals, 1,108 Fnancals and 569 utltes. These 6,451 ssues belong to 2,151 unque frms wth an average age of 21.92 years and equty market captalzaton of $7.79 bllon. 4.1 Relevance of push attrbutes and pull factors In the frst set of emprcal tests, we explore the uncondtonal relevance of pull factors, f any, on debt ssuance actvty. To ths end, we construct regmes based on each of the four pull varables: based on medan monthly values of default and term factors, we defne two types of low-hgh nterest rate regmes, and based on the sgn of annual market momentum and sentment ndex values, we desgnate bearsh-bullsh market performance and pessmstc-optmstc nvestor sentment regmes. For each of the fve types of bond ssues, we explore the dfferences n the number of new ssues and average volume (offer amount) under contrastng pull factor regmes. Table 1 presents the results. The number and amount of straght bonds ssued s hgher when nterest rates (term or default) are low or nvestor sentment s optmstc; bullsh equty markets also ncrease the number of such ssues. Anecdotally, convertble and putable bonds are typcally ssued by frms durng dstressed or unfavorable perods whereby the added ncentves (the rght to convert the bond nto equty or put back the bond to the frm) are ntended to nduce lqudty and nvestor demand. Accordngly, we fnd that ssuance actvty of convertbles and putables s sgnfcantly hgher when the two nterest rates are hgh, or nvestor sentment s negatve. Unlke other bond types, callables grant addtonal benefts (the call opton) to the ssuer; so frms have greater ncentve to opportunstcally tme the callables than other bond ssues. Thus, more callable bonds are ssued when ether nvestor sentment or market 22 ISSN 1916-971X E-ISSN 1916-9728
momentum s postve, and also when prevalng nterest rates are hgh. Fnally, more bonds are ssued under SEC Rule 144A when nterest rates are hgh or markets are bullsh. For all bond types, the dfferences n contrastng pull regmes are stronger for the number of bonds ssued, and weaker when offer amounts are consdered. Nevertheless, the relevance of the pull factors cannot be dscount; all four pull factors bear sgnfcant mpact on the number, volume and type of debt ssued. We start formal tests by carryng out separate regressons for bond ssue offer amounts over the push and pull varables. Table 2 presents regresson results for the fve types of bonds. Panel A dentfes the relevant frm-specfc push attrbutes. Larger frms ssue greater volumes of bonds of all types. Hghly levered frms ssue more straght, callable and Rule 144A ssues. Age, credt ratng, and pre-ssue runup condton the amounts of convertble and callable bonds. Proftablty (Tobn s Q) bear nfluence on the ssuance of straght (putable) bonds. Panel B confrms the relevance of pull factors. Larger amounts of all fve types of bonds are ssued when equty markets are bearsh (confrmng market tmng theory of equty and debt ssuance) and when nvestor sentment s optmstc (confrmng opportunstc ssuances caterng to nvestor preferences). Default and term factors too are sgnfcant n most regressons. Comparson of Panels A and B reveal that adjusted R 2 s are sgnfcantly larger for regressons based on push varables (values range between 18-47%) than those based on pull varables (values range between 4-14%). Thus, on standalone bass, push attrbutes appear to bear sgnfcantly hgher mpact on the volume of debt ssuance than pull varables. The above nference s confrmed when we conduct encompassng regressons of bond ssue offer amounts jontly over all push and pull varables; Table 3 reports the results. There s very lttle change n the sgnfcance of frm-specfc push attrbutes from the standalone regressons (Panel A, Table 2) to jont regressons (Table 3). However, the relevance of pull factors substantally dmnsh n the jont specfcaton. Term factor s never sgnfcant, and default factor s only margnally sgnfcant. The sgnfcance of momentum and sentment factors too dmnsh. For all fve bond types, the ncrease n the values of adjusted R 2 s from Panel A of Table 2 to Table 3 are very small. These results confrm that push attrbutes carry the frst order mpact on volume of debt ssuance; pull factors, though not completely rrelevant, bear weak and margnal explanatory power. Table 1 mples that the role of pull factors are more conspcuous for number of bonds ssued than ssue amounts; n other words, pull effects more lkely nfluence the propensty to ssue bonds of certan types than the actual offer szes of these bonds. To establsh the lkelhood of ssuance of debts of dfferent types, we mplement probt models for bond type (dependent varable has a value of 1 f t s an ssue of a partcular type, and 0 otherwse) separately over the push and pull varables. Table 4 presents the probt results for the fve types of bonds. On standalone bass, almost all push attrbutes (Panel A) and most pull factors (Panel B) condton the lkelhood of debt ssuance. Market sze, proftablty, credt qualty, market-to-book rato, and pre-ssuance runup are hghly sgnfcant; age, leverage, and free cash amounts too are sgnfcant for certan bond types. Pull factors too nfluence the ssuance of dfferent bonds, but the role of these factors (especally nvestor sentment) appear margnal. Estmated model R 2 s are always larger for probts usng push attrbutes (21% and 38% for straght and convertble bonds, and 6-8% for others) than for probts usng pull factors (3-5% for all bond types). Table 5 documents the results of encompassng probt specfcatons for bond ssue type mplemented jontly over all push and pull varables. Comparng Tables 4 and 5, we fnd that although there s very lttle change n the sgnfcance of frm-specfc push attrbutes from the standalone probt models to the combned probt models, the relevance of pull factors dmnsh n the jont specfcaton. More compellng s the fndng that, for all fve bond types, the ncrease n the values of estmated R 2 s from Panel A of Table 4 to Table 5 are small. These results ndcate that push effects are far more domnant n nfluencng the decson to ssue bonds of dfferent types than pull effects. 4.2 Relatve sgnfcance of push and pull effects Precedng results reveal that although both push and pull effects are materal n the number and volume of dfferent bonds ssued, the explanatory power of push varables appear greater than those of pull varables. In ths secton, we formally test the relatve sgnfcance of push and pull effects n debt ssuance based on the metrcs developed n Secton 2. These tests consttute the crux of our project. Table 6 summarzes the values of dfferent metrcs quantfyng the relatve sgnfcance of push and pull varables. Panel A of Table 6 reports the values of four measures of relatve sgnfcance correspondng to regresson models for bond ssue offer amounts. Regresson R 2 s based on push varables (18-47%) are sgnfcantly larger than those based on pull varables (4-14%); R 2 s based on the jont specfcaton usng all push and pull varables (22-48%) are only margnally hgher than those based on push varables alone. Consequently, D1 measures based on push varables are very hgh (67-89%) and those based on pull varables are compellngly small (20% for straght bonds, 1-7% for others). One postve standard devaton shocks to (or one postve mean value shfts n) dfferent push Publshed by Canadan Center of Scence and Educaton 23
varables elct sgnfcantly greater changes n the ssue offer amounts than smlar shocks to (or shfts n) dfferent pull varables. Shock ratos range between 65% to 90% for push effects and measure just 10-35% for pull effects. Mean effects ratos (91-99% for push varables and 1-9% for pull varables) ndcate that almost all changes n debt offer volumes arse from shfts n push attrbutes alone. Panel B of Table 6 presents the values of fve measures of relatve sgnfcance of push and pull varables n the probt models for bond ssue type. Probt R 2 s based on pull varables are very low (3-5%), and are larger for push varables (21% and 38% for straght and convertble bonds, 6-8% for others); augmentaton of pull varables to push-only model brngs about just a margnal ncrease n R 2 s. D1 measure s sgnfcantly larger for push effects (54-92%) than for pull effects (7-44%). Smlarly, shock and mean effects ratos are domnantly larger n magntude for push varables (54-86% and 68-92%) than for pull varables (14-46% and 8-32%). We compare push and pull effects based on one addtonal measure: the mplct probablty of a partcular bond type predcted by the probt model. Push attrbutes do a better job n predctng ssue type than pull factors; t-tests denote that probabltes based on push factors are greater than those based on pull factors for all bond types except Rule 144A ssues. However, the relevance of pull factors s not neglgble: t-tests also reveal that predcted probabltes usng all push and pull factors are always greater than those predcted based on push factors alone. All these results conclusvely confrm that although both push and pull effects are smultaneously relevant to debt ssuance, frm-specfc push attrbutes nfluence debt ssuance decsons (n terms of number, volume and type of bond ssued) to a far greater extent than exogenous pull factors. Based on three (or four for probts) dfferent metrcs, the relatve explanatory power and statstcal as well as economc sgnfcance of push varables are always greater than those of pull varables for dfferent specfcatons for all bond types. In short, push effects domnate pull effects. 4.3 Issuance announcement effects and the role of pull factors Do exogenous pull factors condton bond ssuance announcement effects, that s, how markets react to ssuance of bonds of dfferent types? To ths end, we mplement the two-stage condtonal event study framework outlned n Secton 2. In the frst stage, for each bond type, we ft a probt model usng jontly all push and pull varables (as n Table 5). The condtonal resdual from the probt model, computed as the nverse mlls rato (IMR) denote the latent nformaton content ( ) underlyng the bond ssuance event. Then we compute the three day announcement effect (AR) around the bond ssuance date as the cumulatve of excess stock returns over market ndex returns from one day pror to bond ssuance to one day subsequent to bond ssuance. In the second stage, we carry out lnear regresson of ssuance announcement effects (AR) on latent nformaton content (IMR) and the four pull varables. Table 7 reports the results. Panel A of Table 7 shows that ssuance of bonds always reveal sgnfcant amounts of latent prvate nformaton to the market: IMR value s postve and sgnfcant for all bond types. Straght bond ssues elct no market reacton (ssuance AR = 0), but markets react negatvely (AR < 0) for bond ssues of other four types; ths fndng s n conjuncton wth anecdotal evdence and extant lterature. Panel B of Table 7 documents the results of second stage regresson of AR on IMR and the pull varables. Asde from a few exceptons, most pull varables reman nsgnfcant n ths regresson framework. Hence, although systematc pull factors are relevant to the corporate decson of bond ssuance, they have no condtonng nfluence on how markets react to the event of frms ssung debt. In contrast, extant equty lterature (Baker and Wurgler, 2000; 2002) reveal that market reactons to equty ssuance reflect prevalng systematc sentment and perceptons. 5. Robustness Issues To confrm the robustness of our fndngs, and to renforce the sgnfcance and pervasveness of our conclusons, we conduct several addtonal tests. 1. We use alternate frm-specfc push attrbutes such as equty beta, prce-earnngs rato, dvdend-payng status, and more accountng measures. We defne addtonal systematc pull varables such as GDP growth rate, nflaton, and ndustral productvty. We compute default and term factors based on Treasury rates nstead of swap rates, and nstead of Baker-Wurgler sentment ndex we use the closed-end fund dscount rate. We also compute pre-ssuance runup and ssuance announcement effect based on buy-and-hold returns (BHAR) approach nstead of cumulatve abnormal returns (CAR) method. We do not tabulate these addtonal results for brevty, but all results reported n ths paper and consequent conclusons reman largely and effectvely unaltered. The addtonal varables and the modfed defntons bear almost no mpact on any of the results. 2. The chosen sample perod (1994-2007) concdes wth very dramatc and unque upheavals n the fnancal markets. We sub-dvde our sample nto four sub-perods 1994-1999 (tech bubble expanson), 2000-2002 (post-bubble collapse), 2003-2005 (hgh growth), and 2006-2007 (credt crss) and repeat all tests. Across the four 24 ISSN 1916-971X E-ISSN 1916-9728
sub-perods, we fnd very lttle dfference n the relevance of the varables (regresson and probt coeffcents n Tables II through V) or n ther relatve sgnfcance (dfferent measures n Table VI). 3. Snce bonds are nherently more llqud, recent lterature reveal that bond lqudty and equty volatlty are sgnfcant determnants of bond prces (spreads). To explore whether these varables also mpact the ssuance of bonds, we replcate our tests usng two addtonal push attrbutes (Amhud bond lqudty measure and dosyncratc equty volatlty) and two new pull factors (VIX and systematc lqudty ndex). All four varables reman nsgnfcant n all regresson and probt specfcatons and hence bear no materal mpact on the conclusons. 4. We also brefly explore the role of the selected push and pull varables n the ssuance of equty and on the overall captal structure decsons of the bond ssung frms. We fnd that the relevance of systematc pull factors s even lower for equty ssuance. Ths unque results merts detaled analyss beyond the scope of the current paper and hence s the focus of a separate ongong project. 6. Conclusons Most corporate events are self-selected; corporate managers retan a large degree of dscretonary flexblty n decdng whether to undertake these events and the tmng and accompanyng characterstcs of these events f undertaken. However, lterature s nconclusve regardng the source of the factors that drve the decson of self-selected corporate events, that s, whether the drvng nfluences are nternal (frm-specfc) or external (envronmental). Although classcal lterature typcally models corporate decsons as drven by nternal frm-specfc consderatons, emergng evdence suggest that corporate decsons are also nfluenced by systematc and exogenous varables. In ths paper, focusng on the event of corporate debt ssuance, we apply the htherto unexplored push-pull theory to corporate fnance setup: that s, whether the self-selected corporate events lke the type (and amount) of bonds ssued are pushed by the frms due to nternal consderatons, or pulled by the entcement of exogenous and systematc forces. We base our emprcal nvestgaton on a large sample of fve dfferent types of bonds, and nvestgate the role of eght frm-specfc push attrbutes and four exogenous pull factors n condtonng the type and amounts of dfferent bond ssues. Usng a modfed econometrc framework, we explore the jont mpact of push and pull varables on debt ssuance, and we adopt three quanttatve metrcs that allows to dsentangle the relatve sgnfcance of push and pull effects. Our results engender three key conclusons. Frst, uncondtonally, both frm-specfc push attrbutes and exogenous pull factors are relevant to corporate debt ssuance. Second, jont econometrc tests reveal that nternal push attrbutes bear the frst order mpact on the type and amount of debt ssued; external pull forces have only margnal secondary mpact. The relatve sgnfcance of push effects are pervasvely greater than those of pull effects based on all adopted metrcs (D1 measure, shock rato, mean effects rato, and predcted event probabltes). Fnally, although pull factors margnally condton the ssuance of debt, they bear almost no role n how markets react to such events. In short, although pull effects do matter, push effects domnate. Corporate debts are largely pushed by the frms than pulled by the exogenous nfluences. References Acharya, S. (1988). A generalzed econometrc model and tests of a sgnalng hypothess wth two dscreet sgnals. Journal of Fnance, 43, pp. 413-429. Baker, M. P. & Wurgler, J. A. (2000). The equty share n new ssues and aggregate stock returns. Journal of Fnance, 55, pp. 2219-2257. Baker, M. P. & Wurgler, J. A. (2002). Market tmng and captal structure. Journal of Fnance, 57, pp. 1-32. Baker, M. P. & Wurgler, J. A. (2006). Investor sentment and the cross-secton of stock returns. Journal of Fnance, 61, pp. 1645-1680. Barry, C. B., Mann, S. C., Mhov, V., & Rodrguez, M. (2008). Corporate debt ssuance and the hstorcal level of nterest rates. Fnancal Management, 37, pp. 413-430. Brennan, M.J., & Schwartz, E.S. (1977). Convertble bonds: Valuaton and optmal strateges for call and converson. Journal of Fnance, 32, pp. 1699-1715. Brck, I. & Palmon, O. (1992). Interest rate fluctuatons and the advantage of long-term debt fnancng: A note on the effect of the tax-tmng opton. Fnancal Revew, 27, pp. 467-474. Daves, A. (2008). Credt spread determnants: An 85 year perspectve. Journal of Fnancal Markets, 11, pp. 180-197. Publshed by Canadan Center of Scence and Educaton 25
DeAngelo, H., & Masuls, R. W. (1980). Optmal captal structure under corporate and personal taxaton. Journal of Fnancal Economcs, 8, pp. 3-29. Fama, E. F., & French, K. R. (1989). Busness condtons and the expected returns on bonds and stocks. Journal of Fnancal Economcs, 25, pp. 23-49. Fama, E. F., & French, K. R. (1993). Common rsk factors n the returns on stocks and bonds. Journal of Fnancal Economcs, 33, pp. 3-56. Fama, E. F., & French, K. R. (2002). Testng tradeoff and peckng order predctons about dvdends and debt. Revew of Fnancal Studes, 15, pp. 1-37. Fama, E. F., & French, K. R. (2005). Fnancng decsons: Who ssues stock? Journal of Fnancal Economcs, 76, pp. 549-582. Goyal, V. K., & Frank, M. Z. (2009). Captal structure decsons: Whch factors are relably mportant? Fnancal Management, 38, pp. 1-37. Heckman, J. (1979). Sample selecton bas as a specfcaton error. Econometrca, 47, pp. 153-161. Jensen, M. C., & Mecklng, W. H. (1976). Theory of the frm: Manageral behavor, agency costs and ownershp structure. Journal of Fnancal Economcs, 3, pp. 305-360. Lamont, O.A., & Sten, J. C. (2006) Investor sentment and corporate fnance: Mcro and macro. Amercan Economc Revew, 96, pp. 147-151. Lee, E. S. (1966). A theory of mgraton. Demography, 3, pp. 47-57. Leland, H. E. & Pyle, D. H. (1977). Informatonal asymmetres, fnancal structure, and fnancal ntermedaton. Journal of Fnance, 32, pp. 371-387. L, K., & Prabhala, N. R. (2007). Self-selecton models n corporate fnance. In B. E. Eckbo (Ed.), Handbook of Corporate Fnance Vol. I, North Holland: Amsterdam, pp. 37-86. Ljungqvst, A. P., Nanda, V. & Sngh, R. (2006). Hot markets, nvestor sentment, and IPO prcng. Journal of Busness, 79, pp. 1667-1702. Mller, M. H. (1977). Debt and taxes. Journal of Fnance, 32, pp. 261-275. Modglan, F., & Mller, M. H. (1958). The cost of captal, corporaton fnance and the theory of nvestment. Amercan Economc Revew, 48, pp. 261-297. Modglan, F., & Mller, M. H. (1963). Corporate ncome taxes and the cost of captal: A correcton. Amercan Economc Revew, 53, pp. 433-443. Myers, S. & Majluf, N. S. (1984). Corporate fnancng and nvestment decsons when frms have nformaton that nvestors do not have. Journal of Fnancal Economcs, 13, pp. 187-221. Nayak, S. (2010). Investor sentment and corporate bond yeld spreads. Revew of Behavoral Fnance, forthcomng. Prabhala, N. R. (1997). Condtonal methods n event-studes and an equlbrum orented justfcaton for usng standard event-study procedures. Revew of Fnancal Studes, 10, pp. 1-38. Ross, S. A. (1977). The determnaton of fnancal structure: The ncentve-sgnalng approach. Bell Journal of Economcs, 8, pp. 23-40. Shultz, K. S., Morton, K. R., & Weckerle, J. R. (1998). The nfluence of push and pull factors on voluntary and nvoluntary early retrees retrement decson and adjustment. Journal of Vocatonal Behavor, 53, pp. 45-57. Notes Note 1. Push-pull theory, although nascent n fnance setup, s popular n socology lterature [e.g., phenomena lke human mgraton (Lee, 1966) are caused by ether push forces (repulson of source regon) or pull forces (entcement of target destnaton)] and n welfare economcs [e.g., retrement decsons are push-pull decsons: older workers are ether pushed to retre by ther employer or pulled to the choce by the post-retrement entcements (Shultz et al., 1998)]. Analogously, extendng the theory to fnance, we envson corporate events beng ether pushed by the frms or pulled by the market. Note 2. For example, hypothetcally f pull factors are completely rrelevant and only push attrbutes matter, then all three metrcs wll bear a value of 1 for push effects and a value of 0 for pull effects. 26 ISSN 1916-971X E-ISSN 1916-9728
Table 1. Number and volume of bonds ssued under dfferent pull factor regmes Pull Factor Regme Bond Type Straght Convertble Callable Putable Rule 144A Number Amount Number Amount Number Amount Number Amount Number Amount Panel A: Default factor low 949 262.77 756 303.45 1,948 254.27 296 322.96 909 295.90 hgh 465 307.54 1,136 311.14 2,314 331.62 559 401.21 1,468 320.36 dfference 0.000 0.011 0.000 0.687 0.000 0.000 0.000 0.028 0.000 0.084 Panel B: Term factor low 757 312.11 797 330.35 2,146 288.56 287 380.55 1,090 321.62 hgh 657 237.62 1,095 291.85 2,116 304.09 568 370.88 1,287 302.01 dfference 0.008 0.000 0.000 0.042 0.646 0.123 0.000 0.788 0.000 0.156 Panel C: Market momentum (1-year return on CRSP value-weghted ndex) negatve 160 296.54 487 403.97 999 385.08 189 648.81 672 367.49 postve 1,254 275.07 1,405 274.83 3,263 269.08 666 296.17 1,705 288.74 dfference 0.000 0.410 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Panel D: Baker-Wurgler nvestor sentment ndex negatve 527 226.76 1,064 254.28 1,820 285.21 558 284.84 1,218 290.05 postve 887 307.64 828 377.18 2,442 304.51 297 541.87 1,159 333.03 dfference 0.000 0.000 0.000 0.000 0.000 0.058 0.000 0.000 0.226 0.002 Table 1 reports the number of new bonds ssued and the average offer amount (n $ mllon) for fve dfferent types of bonds under dfferent pull factor regmes over the perod 1994-2007. The four panels correspond to regmes formed based on four pull factors. Each month s classfed as low or hgh regme based on comparson wth medan values of monthly default and term factors. Each year s desgnated as negatve or postve regme based on the values of precedng annual stock market momentum and concurrent Baker-Wurgler nvestor sentment ndex. The last row of each panel reports the p-values correspondng to tests of sgnfcance of dfferences n the number or amount of bond ssues under contrastng regmes. Tests of dfferences are based on Pearson 2 test for number of ssues and t-test for offer amount. Publshed by Canadan Center of Scence and Educaton 27
Table 2. Regresson of ssuance volume, separate specfcatons for push & pull varables Bond Type Varable Straght Convertble Callable Putable Rule 144A Panel A: Regresson over push attrbutes Age 0.03 2.01*** -0.19 0.65-0.11 (0.06) (3.74) (-0.76) (0.74) (-0.30) Market cap 85.7*** 183.3*** 111.7*** 215.0*** 122.5*** (14.32) (30.41) (32.62) (16.69) (27.47) Proftablty -307.90* 9.20 24.10-13.90 30.40 (-2.01) (0.19) (0.60) (-0.11) (0.63) Tobn's Q 16.96 4.36 1.54 37.73*** 3.17 (1.24) (1.29) (0.52) (3.75) (0.79) Debt rato 258.7*** 43.2 125.7*** 96.9 106.8*** (6.16) (1.51) (6.01) (1.41) (4.20) Free cash rato -26.99-5.18 2.54 13.35 3.37 (-0.55) (-0.89) (0.44) (1.32) (0.49) Ratng value 0.74-3.33*** 1.63* 0.93-1.32 (0.45) (-3.60) (2.33) (0.53) (-1.45) Pre-ssue runup -0.62-0.91*** -0.18* -0.31-0.32** (-1.89) (-7.99) (-1.97) (-0.99) (-2.86) Intercept -1727.7*** -3497.3*** -2137.3*** -4357.6*** -2287.3*** (-12.83) (-26.16) (-27.73) (-15.06) (-23.06) Adjusted R 2 0.180 0.471 0.270 0.363 0.317 Panel B: Regresson over pull factors Default factor 155.7*** 65.8 85.1*** 220.2** 45.6 (5.16) (1.61) (4.17) (2.97) (1.51) Term factor -50.78* -48.18* -34.22** -52.77-48.64** (-2.45) (-2.40) (-3.17) (-1.45) (-3.22) Market momentum -1.73* -3.19*** -2.50*** -4.01** -2.77*** (-2.38) (-3.99) (-6.11) (-2.89) (-4.64) Investor sentment 155.3*** 78.9*** 51.2*** 245.1*** 42.4** (8.30) (4.73) (5.33) (7.11) (3.11) Intercept 98.7* 274.1*** 212.4*** 116.1 308.2*** (2.43) (4.92) (7.37) (1.20) (7.25) Adjusted R 2 0.073 0.053 0.048 0.141 0.043 * p-value < 0.05; ** p-value < 0.01; *** p-value < 0.001 Table 2 reports the results (coeffcents and t-statstcs n parentheses) of regressons of bond ssue offer amounts (n $ mllon) over frm-specfc push attrbutes (Panel A) and systematc pull factors (Panel B) for the perod 1994-2007. Varables are defned n sectons 3.2 and 3.3. 28 ISSN 1916-971X E-ISSN 1916-9728
Table 3. Regresson of ssuance volume, jont specfcaton for push & pull varables Bond Type Varable Straght Convertble Callable Putable Rule 144A Age -0.07 2.29*** -0.25 1.06-0.09 (-0.19) (4.26) (-1.01) (1.21) (-0.25) Market cap 82.4*** 179.5*** 108.5*** 199.1*** 119.7*** (14.09) (29.56) (31.97) (15.24) (26.68) Proftablty -250.50 13.40 42.50-33.90 35.24 (-1.67) (0.28) (1.07) (-0.28) (0.73) Tobn's Q 6.82 0.55 1.08 31.40** 0.38 (0.51) (0.16) (0.36) (3.13) (0.09) Debt rato 206.8*** 41.5 123.1*** 109.3 105.5*** (4.96) (1.45) (5.96) (1.62) (4.16) Free cash rato -50.60-8.79 0.67 8.48 0.48 (-1.05) (-1.49) (0.12) (0.85) (0.07) Ratng value 1.18-3.16*** 1.91** 2.87-0.87 (0.73) (-3.37) (2.76) (1.62) (-0.95) Pre-ssue runup -0.58-0.94*** -0.23** -0.55-0.35** (-1.83) (-7.67) (-2.60) (-1.73) (-3.08) Default factor 93.2*** -0.1 36.7* 147.3* 0.7 (3.30) (-0.00) (2.07) (2.32) (0.03) Term factor -16.77-18.67-7.80-27.09-9.08 (-0.86) (-1.23) (-0.82) (-0.87) (-0.70) Market momentum -1.72** -0.83-2.04*** -1.14-1.44** (-2.58) (-1.37) (-5.74) (-0.95) (-2.86) Investor sentment 132.4*** 30.6* 26.5** 114.8*** 30.6* (7.63) (2.15) (3.05) (3.60) (2.53) Intercept -1740.7*** -3391.7*** -2106.2*** -4241.2*** -2213.1*** (-12.91) (-24.05) (-26.29) (-14.10) (-20.64) Adjusted R 2 0.224 0.477 0.291 0.387 0.326 * p-value < 0.05; ** p-value < 0.01; *** p-value < 0.001 Table 3 reports the results (coeffcents and t-statstcs n parentheses) of regressons of bond ssue offer amounts (n $ mllon) jontly over frm-specfc push attrbutes and systematc pull factors for the perod 1994-2007. Varables are defned n sectons 3.2 and 3.3. Publshed by Canadan Center of Scence and Educaton 29
Table 4. Probt model for bond type, separate specfcatons for push & pull varables Bond Type Varable Straght Convertble Callable Putable Rule 144A Panel A: Probt usng push attrbutes Age 0.0003-0.0066*** -0.0018-0.0030* -0.0037*** (0.33) (-5.35) (-1.92) (-2.45) (-3.88) Market cap 0.197*** 0.102*** -0.198*** 0.117*** -0.098*** (12.45) (6.51) (-15.05) (7.22) (-7.76) Proftablty -0.95** -1.13*** 1.44*** 0.95*** 1.09*** (-2.97) (-5.06) (8.95) (4.97) (7.20) Tobn's Q -0.27*** 0.16*** 0.07*** -0.05*** -0.01 (-8.59) (8.41) (4.75) (-3.38) (-0.64) Debt rato 0.48*** -1.24*** -0.03-0.66*** 0.03 (4.21) (-12.26) (-0.32) (-6.61) (0.37) Free cash rato 0.25* -0.14** 0.04-0.07** -0.03 (2.50) (-3.18) (1.86) (-2.85) (-1.45) Ratng value -0.075*** 0.109*** 0.014*** 0.044*** 0.022*** (-16.49) (33.03) (5.04) (15.39) (8.86) Pre-ssue runup -0.0032*** 0.0036*** 0.0019*** 0.0009 0.0029*** (-5.41) (8.08) (4.93) (1.95) (8.03) Intercept -3.91*** -3.96*** 4.30*** -4.02*** 1.38*** (-10.85) (-11.20) (14.43) (-10.97) (4.84) Pseudo R 2 0.214 0.378 0.075 0.076 0.060 Panel B: Probt usng pull factors Default factor -0.89*** -0.09 0.67*** -0.33*** 0.66*** (-11.99) (-1.38) (10.17) (-4.14) (10.09) Term factor 0.09* 0.31*** -0.03 0.44*** -0.09* (2.18) (8.48) (-0.79) (10.08) (-2.50) Market momentum -0.0002-0.0062*** 0.0049*** -0.0032* -0.0001 (-0.15) (-4.52) (3.60) (-1.98) (-0.06) Investor sentment -0.049 0.010 0.180*** -0.121** -0.164*** (-1.35) (0.31) (5.40) (-2.77) (-4.91) Intercept 0.44*** -0.65*** -0.64*** -1.03*** -1.23*** (4.44) (-7.16) (-7.08) (-9.44) (-13.51) Pseudo R 2 0.053 0.032 0.028 0.047 0.049 * p-value < 0.05; ** p-value < 0.01; *** p-value < 0.001 Table 4 reports the results (coeffcents and Z-statstcs n parentheses) of probt models for bond ssue type (bnary 0 and 1) over frm-specfc push attrbutes (Panel A) and systematc pull factors (Panel B) for the perod 1994-2007. Varables are defned n sectons 3.2 and 3.3. 30 ISSN 1916-971X E-ISSN 1916-9728