Lecture 12. Capital Structure Theory

Transcription

1 Lecture 12 Captal Structure Captal Structure Theory Captal Structure: How a frm fnance.e., equty (E) or debt ()- ts assets Modglan-Mller Theorem (MMT): Uses a smple model of valuaton No arbtrage.e., equal rates of return for equal rsks. Rsk-free debt Under certan assumptons (perfect markets, no taxes or bankruptcy costs, no asymmetrc nformaton, etc.), the value of the frm (V) s ndependent of how the frm s fnanced. V = + E 1

2 That s, there s no optmal captal structure. If MMT s assumptons are volated, then captal structure matters. Usual volatons - Tradtonal fnance: bankruptcy costs, agency problems, non-convex taxes, asymmetrc nformaton. - Behavoral fnance: neffcent markets, manageral and nvestor behavor. Q: If MMT s volated, what s the optmal captal structure? Man Theores Trade-off Theory. Ths theory can be obtaned from dfferent perspectves. ebt s rsky. Bankruptcy costs exst. There s a tax-bankruptcy trade-off for debt: - tax beneft - bankruptcy cost. There s an agency perspectve: - ebt dscplnes manager and mtgates agency problems of free CFs, snce debt must be repad to avod bankruptcy. But, exacerbates shareholder-debtholder conflcts. References: Jensen and Mekclng (1976), Jensen (1986), and Hart and Moore (1994) 2

3 There s a stakeholder co-nvestment perspectve: - Some frms effcency requres a frm s stakeholders to make sgnfcant frm-specfc nvestments. Frms makng unque products wll lose customers f they may go bankrupt. Reference: Ttman (1984), Maksmov and Ttman (1991). Note: These perspectves dffer from the tax-bankruptcy tradeoff. The costs of debt are from dsrupton to normal busness operatons and do not depend on the arguably small drect costs of bankruptcy. As a result of these three perspectves we have a general result = > Margnal beneft of debt declnes as debt ncreases. There s an optmal captal structure, /E. 3

4 Peckng Order Theory Asymmetrc nformaton exsts and t s costly. Managers have more nformaton about the qualty of the frm. Companes select fnancng accordng to the law of least effort. (1) Internal fnancng (retaned earnngs), frst. (2) Bank debt (n dfferent levels, easest: bank debt), second (3) Equty, last resort. Adverse selecton ssues: Equty has a lot, debt a lttle, retaned earnngs none. => The choce of fnancng s a sgnal. Myers (1984): when equty s ssued, nvestors thnk frm s overvalued (managers use the last resort tool, only because frm s overvalued). Investors demand a hgher return on equty than on debt. References: onaldson (1961), Myers (1984) Market Tmng Theory Frms are ndfferent between equty or debt fnancng. But, the market makes prcng mstakes from tme to tme. Managers select the debt or equty accordng to the relatve msprcng. If nether market looks favorable, manager may defer ssuances. If condtons look unusually favorable, managers may rase funds even f the frm has no need for funds. There are no frm specfc varables ( factors ) that nfluence /E. Behavoral fnance-type story. Baker and Wrugler (2002): Theory explans hot and cold IPO perods. 4

5 Manageral Inerta Model Frms do not act to adjust captal structure. - When stock prces are hgh, frms have low /E. - When stock prces are low, frm have hgh /E There s no counter acton to counterbalance stock return nduced /E changes. Welch (2004) fnds that stock returns can explan /E dynamcs. Stylzed Facts Fact 1: Frms use debt fnancng too conservatvely Graham (2000), Strebulaev & Yang (2007). Fact 2: Negatve relaton between proftablty and leverage - Myers (1993), Myers and Shyam-Sunder (1999). Fact 3: Frms mean-revert slowly towards target leverage Fama and French (2002), Flannery and Rangan (2006). Fact 4: Changes n market leverage are largely explaned by changes n equty prces -Welch (2004)] Fact 5: Leverage largely drven by unexplaned frm-specfc fxed effect - Lemmon, Roberts, Zender (JF, 2006). Fact 6: Lnk between governance mechansms & leverage ambguous - Berger, Ofek & Yermack (JF, 1997), John and Ltov (2008). 5

6 Trade-off Theory Frms operate under a target /E rato, representng some optmal /E rato. Frms do not nstantaneously acheve ther target /E rato. They adjust the actual /E rato over tme. The model s called the dynamc trade-off model:, t, t 1 e t, t, (1) where,t s frm s realzed /E n perod t,,t s frm s target /E rato, Δ s the dfference operator, γ s the partal adjustment coeffcent (also called speed of adjustment); 0 γ 1, and e,t s a regresson error. Ths model can be estmated usng OLS. But the target debt-equty rato,,t, s unobservable: => t s not possble to drectly test the dynamc trade-off model. A (usually ad-hoc) model for the target /E rato s used to estmate the model: t X, t, where the vector X,t contans a set of drvng varables: -EAT -B/M - Margnal tax rate - Altman Z score - Industry dummy varables - Captal and/or R& expendture (2) These varables are used because they are theory related or were used before (data mnng problem). 6

7 Then, the usual estmated equaton becomes: X ( 1 ), t 1 e, t, t t, (3) Ths equaton s n terms of observables. It s the bass of emprcal work Unrestrcted estmaton of (3) s straghtforward. But, only (1-γ ) wll be estmated, γ cannot be recover from the unrestrcted coeffcent for X,t. =>The speed of adjustment s estmated through the unrestrcted coeffcent of,t-1. Note that a two-step estmaton procedure could be used: - Frst, estmate (2), usng the observed,t as a proxy for the unobserved dependent varable,,t. - Second, wth the estmated estmated,t, do an unrestrcted estmaton of (3). Equaton (3) s a standard panel data model. Usually, a common γ = γ s estmated, assumng: - Pooled data or Fama-Macbeth s method, Fama and French (2002). - Fxed-effects model, Flannery and Rangan (2006). We can gnore the panel to allow for heterogenety. (Trade-off: less powerful estmates of common parameters, but t allows for dfferent γ s. - Indvdual (=ndustry) estmaton, Roberts (2002), Zhao and Susmel (2008). Usual trade-off n panel: Panel estmaton produces more powerful estmates, but at the cost of heterogenety (the estmate may not be representatve). 7

8 Man Fndngs: - For the trade-off model: Hovakman, Opler, and Ttman (2001), Strebulaev (2004), Flannery and Rangan (2006), and Kayhan and Ttman (2007) fnd that the dynamc trade-off model domnates alternatve models. - Aganst the trade-off model: Fama and French (2002) fnd no clear cut domnant model. - Book value debt vs. Market value debt. => Marsh (1982): emprcal results are not sgnfcantly affected by the choce. => robetz, Pensa, and Wanzenred (2007): market values calculatons are cumbersome, not readly avalable and may not reflect changes ntated by frm s managers. (They call market value of debt quas-market value of debt. Frms are more concerned wth book /E ratos.) More Fndngs: - Long term or Short term debt? Long term debt s less flexble. Partal adjustment model makes more sense for long term debt. => Flannery and Rangan (2006) use dfferent defntons of debt. Results are not sgnfcantly dfferent - Heterogenety of γ s sgnfcant: Roberts (2002), Fama and French (2002), Flannery and Hankns (2007). => Fama and French (2002) report annual estmates of γ from 7% to 18%, Roberts (2002) reports for dfferent ndustres annual estmates from close to 0% to close to 100%. - Target /E rato tme-varyng? Conflctng theory and evdence. => Mean reverson n actual /E s found by Marsh (1982), Auerbach (1985) and Opler and Ttman (1995). => Jallvand and Harrs (1984) fnd that /E ratos are stable. robetz, Pensa, and Wanzenred (2007) fnd that book /E s qute stable around.6, though market /E tends to be more tme-varyng. 8

9 Issues n Trade-off (and, n almost all Captal Structure) models - The estmated model s a reduced form model => dentfcaton problem. - Correct specfcaton of model for,t. - Measurement error may be an mportant ssue. - Mssng varables n the model for,t and thus n model for,t. - Heterogeneous panel. - Autocorrelaton and cross correlaton n error terms. - Potental unt roots. Testng the Trade-off Model Notce that the test of the trade-off theory would be straghtforward f an estmate of,t were avalable. Smply rearrange the frst equaton:, t t ( 1 ), t1 e t Usually, an unrestrcted estmaton of the above equaton s done:, t,1t,2, t1 e t (4) Then, the test of the ynamc Trade-off Theory s: H 0 (ynamc Trade-off Theory true): γ,1 + γ,2 = 1. H 1 (ynamc Trade-off Theory not true): γ,1 + γ,2 1. 9

10 Ths drect test s never done. (Ths s why the emprcal lterature emphaszes the estmates of the speed of adjustment.) Usually, tests are done ndrectly. A fndng of a sgnfcant speed of adjustment s seen as evdence for the model even though other models may also post some autocorrelaton for /E; see Chen and Zhao (2005). The drect test, however, can be done. We need an estmate of,t. => Kalman flter can do t. Assume,t. follows an AR(1). Then, the model n (4) can be wrtten n a state-space representaton:, t e, t [ 1 ],, t t 1 (measurement equaton), t, t 1 0, t 1, t 1 2 u u,1t,2t (state equaton) fferent assumptons about,t can be ntroduced by placng restrctons on φ. Constant, by settng φ = 0 Random Walk, by settng φ = 1. Ths approach avods endogenety ssues (many of the X,t varables are smultaneously determned wth,t.) and jontly drectly estmates all parameters of nterest. More mportant, the drect test of the trade-off model.e., H 0 : γ,1 + γ,2 = 1- can be done. Zhao and Susmel (2008) do ths. They estmate the state-space representaton for 578 frms, under three dfferent assumptons for,t : AR(1), constant, and RW. The model s estmated for each frm. Quarterly data, usng frms wth unnterrupted hstores. 10

11 Fndngs - Model holds for 32% of frms, assumng,t follows an AR(1). - Model holds for 52% of frms, assumng,t s constant. - A lot of heterogenety n γ,1. For the AR(1) case: -Medan γ,1 s.161, the average γ,1 s.276. (Quarterly estmates) - The emprcal 95% C.I for γ,1 [.025,.951]. The IQ range s not that extreme, gong from.088 to