Ths s a repostory copy of The Response of Frms' Leverage to Uncertanty: Evdence from UK Publc versus Non-Publc Frms. Whte Rose Research Onlne URL for ths paper: http://eprnts.whterose.ac.uk/42606/ Monograph: Caglayan, M. and Rashd, A. (2010) The Response of Frms' Leverage to Uncertanty: Evdence from UK Publc versus Non-Publc Frms. Workng Paper. Department of Economcs, Unversty of Sheffeld ISSN 1749-8368 Sheffeld Economc Research Paper Seres 2010019 Reuse Unless ndcated otherwse, fulltext tems are protected by copyrght wth all rghts reserved. The copyrght excepton n secton 29 of the Copyrght, Desgns and Patents Act 1988 allows the makng of a sngle copy solely for the purpose of non-commercal research or prvate study wthn the lmts of far dealng. The publsher or other rghts-holder may allow further reproducton and re-use of ths verson - refer to the Whte Rose Research Onlne record for ths tem. Where records dentfy the publsher as the copyrght holder, users can verfy any specfc terms of use on the publsher s webste. Takedown If you consder content n Whte Rose Research Onlne to be n breach of UK law, please notfy us by emalng eprnts@whterose.ac.uk ncludng the URL of the record and the reason for the wthdrawal request. eprnts@whterose.ac.uk https://eprnts.whterose.ac.uk/
Sheffeld Economc Research Paper Seres SERP Number: 2010019 Mustafa Caglayan Unversty of Sheffeld Abdul Rashd Unversty of Sheffeld The Response of Frms' Leverage to Uncertanty: Evdence from UK Publc versus Non-Publc Frms October 2010 Department of Economcs Unversty of Sheffeld 9 Mappn Street Sheffeld S1 4DT Unted Kngdom www.shef.ac.uk/economcs
The Response of Frms Leverage to Uncertanty: Evdence from UK Publc versus Non-Publc Frms Mustafa Caglayan Department of Economcs, Unversty of Sheffeld, UK Abdul Rashd Department of Economcs, Unversty of Sheffeld, UK October 23, 2010 Abstract Ths paper emprcally nvestgates the effects of uncertanty on frms leverage. The analyss s carred out for a large panel of publc and non-publc UK manufacturng frms over 1999-2008. The emprcal results provde evdence that frms use less short-term debt as they go through perods of hgh uncertanty. The leverage of non-publc frms s more senstve to dosyncratc uncertanty n comparson to ther publc counterparts, yet macroeconomc uncertanty affects both types of frms smlarly. We fnally end our nvestgaton showng that the total mpact of ether type of uncertanty on frms leverage s related to the amount of the cash buffer each frm carres. Keywords: Cash holdngs; Leverage; Publc versus Non-publc frms; Idosyncratc versus Macroeconomc uncertanty; Spllover effects; System-GMM JEL classfcaton: C23, D81, G32 We would lke to thank S. Brown for her comments on earler versons. The standard dsclamer apples. Correspondng author: Mustafa Caglayan, e-mal: m.caglayan@sheffeld.ac.uk. Abdul Rashd, e-mal: ecp09ar@sheffeld.ac.uk. 1
1 Introducton Snce the semnal work of Modglan and Mller (1958), researchers have been scrutnzng the factors that affect frms fnancng decsons. Whle proposed theores ncludng the agency theory (Jensen and Mecklng (1976)), the captal sgnallng theory (Ross (1977)), peckng order theory(myers(1984)), the free cash flow theory(jensen(1986)), organzatonal behavor(myers (1993)), market tmng theory (Baker and Wurgler (2002)), manageral overoptmsm (Heaton (2002)) and nerta theory (Welch (2004)) help us to understand the role of frm-specfc factors such as proftablty, frm sze, the effectve tax rate, frm growth, tangble assets, stock returns and non-debt tax shelds n the determnaton of frms leverage, less attenton s pad to the role of varatons n frm-specfc and macroeconomc factors. 1 For nstance, we do not know to what extent frms readjust ther captal structure when faced wth macroeconomc or frmspecfc (dosyncratc) uncertanty. We know even less about how publc versus non-publc (prvate) frms leverage responds to macroeconomc and dosyncratc uncertanty and nether do we know f there s a dfference between the response of each type of frm. There are several papers n the exstng emprcal lterature whch scrutnze the role of macroeconomc condtons on frms borrowng behavor. Research shows that managers, n consderaton of the fnancal strength of the frm, desgn the captal structure of the frm n algnment wth the state of the economy to mnmze the adverse effects of busness-cycles. 2 However, there s lttle research on the effects of macroeconomc uncertanty on frms captal structure. To our knowledge, there are only two studes, Baum, Stephan, and Talavera (2009) and Hatznkolaou, Katsmbrs, and Noulas(2002), whch examne the mpact of macroeconomc uncertanty on frms leverage. These studes focus solely on publcly traded frms and show that an ncrease n macroeconomc uncertanty would lead to a decrease n frm borrowng. In contrast, when we revew the lterature on the mpact of dosyncratc uncertanty on frms captal structure, we observe that whle some studes provde evdence that dosyncratc uncertanty exerts a negatve mpact on leverage, others suggest that the effect s postve. Nevertheless, none of these studes nvestgate the behavor of non-publc frms under uncertanty as ther man focus s on publcly traded US companes. In ths study we dffer from the earler research n several dstnct ways. In contrast to the 1 Several researchers provde evdence on the emprcal valdty of these theores. See, for nstance, among others, Harrs and Ravv (1991), Myers (2001), Hovakman, Opler, and Ttman (2001), Fama and French (2002), Hennessy and Whted (2005) and Abor and Bekpe (2009). Also see Kolasnsk (2009) for an excellent survey of the emprcal lterature on captal structure. 2 See, for nstance, among others, Levy and Hennessy (2007), Hackbarth, Mao, and Morellec (2006), Korajczyk and Levy (2003), Suarez and Sussman (1999) and Gertler and Hubbard (1993). 2
exstng research, we use a large panel of UK manufacturng frms and examne the uncertantyleverage relatonshp for publc versus non-publc frms as our dataset enables us to dfferentate the frms by ther legal form. Our ntal nspecton of the data shows that non-publc frms are relatvely small n sze and sgnfcantly dffer from ther publc counterparts n the context of ther access to the captal markets. It s generally accepted that non-publc frms have less potental to absorb the negatve busness shocks and they have to overcome more hurdles to access outsde sources of fnance as there are substantal nformatonal asymmetres between managers of non-publc frms and the outsde credtors. As a consequence, banks would be more cautous tolend to them n an envronment where uncertanty s hgh as these frms possbly face shortfalls n ther expected cash flows. Furthermore, n our nvestgaton we explore the effects of both dosyncratc and macroeconomc uncertanty on frms leverage. Hence, ths study provdes answers to questons such as whether a frm s leverage decson s more senstve to dosyncratc or macroeconomc uncertanty and whether the effects of uncertanty on leverage are dfferent for publc versus non-publc frms. Research has also shown that uncertanty not only affects frm behavor on ts own but also n conjuncton wth varous frm-specfc varables. For nstance as Baum, Caglayan, and Talavera (2010) suggest, when uncertanty vares over tme, lenders may fal to evaluate the credtworthness of a frm and render the frm credt constraned by rasng the lqudty premum requred to provde funds. In such crcumstances, frm managers wll be more dependent on frms retaned earnngs or lqud assets to overcome the dffcultes that whch the frm has to go through. In ths case, Baum, Caglayan, and Talavera(2010) show that the mpact of uncertanty on a frm s fxed captal nvestment can also be gauged through ts effects on the frm s retaned earnngs (spllover effects) n addton to the own effects of uncertanty. Therefore, n ths paper, we explore whether uncertanty has spllover effects on leverage through other frmspecfc varables n addton to ts own effects. In partcular, we consder those movements n frms lqud assets and examne to what extent the effects of uncertanty spll over the frm s leverage as the lqud assets of the frm evolve wth the movements n dosyncratc and macroeconomc volatlty. In dong so, we present evdence on the ndrect effects (spllover effects) of uncertanty, an ssue that has not been examned earler, n addton to the drect (own) effects of uncertanty. Overall, our nvestgaton helps us to lay out a more complete pcture of how dosyncratc and macroeconomc varatons on ther own or n conjuncton wth movements n frms lqud assets affect publc versus non-publc frms leverage decsons. We begn our analyss by separately estmatng the effects of macroeconomc and dosyn- 3
cratc uncertanty on the target leverage of publc and non-publc manufacturng frms. We then ncorporate both types of uncertanty n the same model to scrutnze whether these measures are jontly operatonal. Once we establsh the role of each type of uncertanty on frms leverage, we ntroduce an nteracton term between our measures of uncertanty and the cash stock of the frm to examne how the mpact of uncertanty on leverage changes as cash holdngs of the company vary over tme the spllover effect. Our emprcal nvestgaton makes use of three dfferent measures to capture the mpact of both frm-specfc and macroeconomc uncertanty on publc versus non-publc frms leverage whle controllng for several frm-specfc varables. The data on publc and non-publc UK manufacturng frms are extracted from the FAME database and cover the perod 1999-2008. Our fndngs can be summarzed as follows. Usng the System-GMM estmator we fnd that the effects of frm-specfc varables such as nvestment-to-total assets rato, sales-to-total assets rato, cash-to-total assets rato on leverage are generally smlar to earler emprcal fndngs. 3 Therefore, throughout the dscusson of our fndngs we do not place too much emphass on the role of frm-specfc determnants on leverage. Instead, we manly focus on the mpact of tme-varyng dosyncratc and macroeconomc uncertanty on frms leverage, and whether the leverage of publc versus non-publc frms behaves dfferently n response to ether sources of uncertanty and whether frms cash holdngs affect the margnal mpact of uncertanty on leverage. When we examne the effects of uncertanty, we fnd that an ncrease n dosyncratc uncertanty causes frms to lower ther leverage. The negatve effect of dosyncratc uncertanty s consstent wth the fndngs of Ttman and Wessels (1988) and MacKe-Mason (1990) who show that a frm s leverage s sgnfcantly negatvely correlated wth ts earnngs volatlty. 4 We also fnd that the leverage of non-publc frms exhbts a greater senstvty to dosyncratc uncertanty as compared to ther publc counterparts. Ths fndng s consstent wth the vew that the fnancng polcy of non-publc frms depends more on ther n-house performance as external fnance s expected to dmnsh durng perods of volatlty due to the presence of frctons. We next turn to nvestgate the effects of macroeconomc uncertanty. We fnd that an ncrease n macroeconomc uncertanty also leads to both publc and non-publc frms to use less short-term debt n ther captal structure. Yet we fnd no evdence that the mpact s dfferent across each category. Ths ndcates that durng perods of macroeconomc turmol 3 See for nstance Ttman and Wessels (1988), Fama and French (2002) and Brav (2009). 4 Ths s n contrast to Myers (1977) who argues that large busness rsk may reduce the agency cost of debt leadng to an ncrease n the frm s debt. 4
debt becomes an unattractve source of fnance for ether type of frm. When we re-estmate the model ncorporatng both dosyncratc and macroeconomc uncertanty along wth the frm-specfc varables, we observe that the coeffcents assocated wth both uncertanty types mantan ther sgns and sgnfcance ndcatng the robustness of our results. Last but not least, we examne the spllover effects of uncertanty on leverage through frms cash holdngs. We show that both types of uncertanty have sgnfcant spllover effects on publc frms leverage, yet we fnd no such sgnfcant effects for non-publc frms. A follow up nvestgaton whch consders both drect and spllover effects of uncertanty provdes evdence that leverage of both publc and non-publc frms are negatvely and sgnfcantly affected as uncertanty ncreases. More nterestngly, we fnd that whle the overall mpact of dosyncratc uncertanty on leverage s negatve, ths negatve effect becomes stronger as the frm holds more cash stocks. In other words, we show that durng perods of hgher dosyncratc uncertanty, frms wth hgher levels of cash holdngs have a larger propensty to reduce ther leverage relatve to those frms that hold lower levels of cash stocks. In contrast, the total mpact of macroeconomc uncertanty on leverage s stronger when frms cash holdng s low. Furthermore, the negatve effect of macroeconomc uncertanty becomes weaker and, n fact, nsgnfcant as frms accumulate a stockple of cash. The remander of the study proceeds as follows. Secton 2 lays out the effects of macroeconomc and dosyncratc uncertanty on a frm s fnancng behavor. Secton 3 provdes a descrpton of datasets and explans the constructon of varables n conjuncton wth summary statstcs. Secton 4 dscusses the emprcal models. Secton 5 presents the emprcal results. Secton 6 concludes the study. 2 The Lnk between Uncertanty and Leverage In what follows below, we provde a bref dscusson on the role of macroeconomc and frmspecfc uncertanty n determnng a frm s leverage. 2.1 Macroeconomc Uncertanty and Frm Leverage There s an extensve emprcal lterature whch nvestgates how macroeconomc uncertanty affects frms behavor. Several researchers, ncludng Leahy and Whted (1996), Ghosal and Loungan (1996), and Baum, Caglayan, and Talavera (2010) ndcate that frms sgnfcantly reduce ther fxed nvestment expendtures durng perods of hgh uncertanty. 5 Bartram (2002) 5 Also see Azenman and Maron (1999), Beaudry, Caglayan, and Schantarell (2001), Bloom, Bond, and Reenen (2007) who present evdence on the adverse effects of uncertanty on fxed nvestment. 5
presents evdence that the measures of frm lqudty s sgnfcantly assocated wth nterest rate exposure. Studes, among others, Almeda, Campello, and Wesbach (2004) and Baum, Caglayan, Stephan, and Talavera (2008), fnd that frms ncrease ther demand for lqud assets n response to an ncrease n macroeconomc uncertanty. Collectvely, these emprcal fndngs ndcate that managers fne tune the fxed nvestment behavor and lqud assets of ther frms to sheld the frm aganst the adverse effects of uncertanty assocated wth the aggregate economc actvtes. Unfortunately, there s not much emprcal research that nvestgates the mpact of macroeconomc uncertanty on a frm s debt structure. Gertler and Hubbard (1993) dscuss how frms face both dosyncratc and macroeconomc uncertanty n ther producton and fnancal decsons. Accordng to them, although frms can mtgate the effect of the frst one, they are not able to manpulate the effects of macroeconomc uncertanty. Therefore, frms opt for equty rather than debt to shft (at least some of) the busness-cycle rsk to ther lenders durng perods of hgher macroeconomc uncertanty. In ths context, the effect of macroeconomc uncertanty on the leverage rato s expected to be negatve. To our knowledge, only two studes emprcally examne the lnk between leverage and uncertanty. Baum, Stephan, and Talavera (2009) show for a set of large U.S. nonfnancal frms drawn from COMPUSTAT that an ncrease n macroeconomc uncertanty leads to a sgnfcant decrease n frms optmal short-term leverage. In addton, splttng ther sample by lqudty and leverage they provde evdence that the mpact of macroeconomc uncertanty s stronger for hgh-lqudty frms and low leveraged frms. Hatznkolaou, Katsmbrs, and Noulas (2002) examne the mpact of nflaton uncertanty on frms debt-equty ratos for the frms ncluded n the Dow Jones Industral Index and they fnd that nflaton uncertanty has a sgnfcant negatve effect on a frm s debt-equty rato. Gven the lack of emprcal evdence on the effects of macroeconomc uncertanty on leverage, t s of partcular nterest to nvestgate to what extent macroeconomc uncertanty affects the targetleverageoffrms. Itmustbenotedthatthsssuesnotonlyrelevantforpublccompanes but more so for non-publc companes whose man source of fnancng s bank debt. Especally, t s well documented that the frms n each group face dfferent fnancal constrants due to the presence of asymmetrc nformaton when t comes to rasng funds to fnance ther daly actvtes as well as ther captal nvestment projects. 6 All together, emprcal evdence on publc and non-publc frms captal structures wll help us to enhance our understandng of 6 In Whted (1992), one of the fundamental predctons of the asymmetrc nformaton theory s that small frms have lmted access to debt markets due to the lack of the collateral necessary to back up ther loans. 6
macroeconomc uncertanty leverage lnkages. 2.2 Frm-Specfc Uncertanty and Frm Leverage When we revew the lterature we fnd that whle some researchers report a negatve mpact of dosyncratc uncertanty on leverage others fnd no or postve effects. The trade-off theory of captal structure predcts an nverse lnk between frm-specfc uncertanty and frms optmal debt levels. The ratonale for ths predcton s that hgher busness rsk as measured by an ncrease n the volatlty of cash flows heghtens the probablty of bankruptcy. Therefore, due to the presence of postve bankruptcy costs, frms use less debt n ther captal structure when there s a large varaton n ther earnngs. To that end Bradley, Jarrell, and Km (1984) present a sngle perod corporate captal structure model and show the presence of a negatve assocaton between frm volatlty and ts optmal debt. Subsequently, Ttman and Wessels (1988) report a negatve assocaton between earnngs volatlty and leverage. Baum, Stephan, and Talavera (2009) report a sgnfcant and negatve mpact of dosyncratc uncertanty on the optmal short-term leverage for US non-fnancal publc frms. 7 Wald (1999) nvestgates how earnngs volatlty affects the target leverage by examnng the determnants of captal structure n France, Germany, Japan, the US and the UK. They fnd a sgnfcant negatve effect of frm-level rsk on the debt-to-assets rato for frms establshed n the US and Germany. For the remanng countres, however, they do not fnd any sgnfcant assocaton between frms busness rsk and ther leverage. 8 Overall smlar fndngs are reported n Baxter (1967), Ferr and Jones (1979), Frend and Lang (1988) and MacKe-Mason (1990), ndcatng the presence of a sgnfcant and negatve mpact of frm-level rsk on leverage. In contrast, Myers (1977) predcts a postve relatonshp between rsk and debt. He argues that large busness rsk may reduce the agency cost of debt and thus frms use more debt n ther captal structure. Jaffe and Westerfeld (1987) also derve a postve assocaton between rsk and the optmal debt level. Several other emprcal studes, ncludng Auerbach (1985), Km and Sorensen (1986) and Chu, Wu, and Chou (1992), report a sgnfcant and postve mpact of frm-level rsk on leverage. Earler, Toy, Stonehll, Remmers, and Wrght (1974) report the presence of a sgnfcant and postve effect of earnngs volatlty on the debt rato of manufacturng frms n Japan, Norway and the US. Kale, Noe, and Ramrez (1991) examne the mpact of busness rsk on the optmal debt level by developng a model smlar to DeAngelo 7 In addton, they show that hghly leveraged frms and small frms are more senstve to frm-specfc uncertanly as compared to relatvely low leveraged or large frms. 8 Flath and Knoeber (1980) also show that the frm s earnng volatlty does not have any sgnfcant mpact on leverage n 38 major ndustres over the perod 1957-1972 usng a dataset drawn form the IRS Statstcs of Income, Corporate Income Tax Returns database. 7
and Musuls (1980). They predct that the relatonshp s approxmately U-shaped. Usng the annual COMPUSTAT data, they show that an ncrease n busness rsk ntally leads to a declne n debt. However, once the frm s debt exceeds a certan lmt, they use more debt n ther captal structure as busness rsk ncreases. Overall, we observe that both theoretcal and emprcal research lead to conflctng conclusons on the assocaton between dosyncratc rsk and leverage. In the case of theoretcal models, results are related to the underlyng assumptons and n the case of emprcal studes, results dffer based on the sample and measure of uncertanty used n the nvestgaton. In addton, none of the studes cted above examnes ths relatonshp for non-publc companes. Snce non-publc frms fnancng optons sgnfcantly dffer from that of publc frms as they are not legally allowed to ssue debt nstruments, t s mportant to nvestgate how non-publc frms leverage evolves under uncertanty. In ths paper, we therefore test how frm-specfc uncertanty affects non-publc frms leverage, as compared to that of publcly traded frms usng UK frm-level data. 3 Data, Varable Constructon and Measurng Uncertanty To carry out our nvestgaton we construct an annual panel dataset for publc and non-publc manufacturng frms usng the FAME database whch s made avalable by Bureau van Djk (BvD) Electronc Publshng. We generate three dfferent measures of macroeconomc uncertanty based on gross domestc product (GDP), the consumer prce ndex (CPI) and Treasury bll rates (T-bll rates). The data on macroeconomc varables are extracted from the Internatonal Fnancal Statstcs (IFS), an Internatonal Monetary Fund (IMF) database. The dataset covers a ten-year perod from 1999 to 2008. 3.1 Publc versus Non-Publc Company Defnton Under the UK Companes Act, all lmted lablty companes regster themselves wth the Companes House as ether publc or non-publc companes. Companes House s bascally an executve agency of the Unted Kngdom Department for Busness, Innovaton and Sklls (BIS). The fundamental functons of the Companes House are to ncorporate and dssolve lmted lablty companes, accumulate and scrutnze company nformaton and make ths nformaton avalable to the publc. 9 Accordng to the Companes Act of 1967, n the Unted Kngdom, all publc and non-publc companes must submt ther annual fnancal statements to the Regster of Companes House. 9 For more nformaton about Companes House, see http://www.companeshouse.gov.uk/. 8
However, the Companes Act of 1981 modfed the 1967 Act allowng small frms to fle an abbrevated balance sheet wthout a proft and loss statement and medum szed companes to submt an abbrevated fnancal statement. 10 Currently, both publc and non-publc companes must fle ther fnancal statements wthn a perod of ten and seven months respectvely of ther accountng year-end date. It should be noted that all accountng statements are compled accordng to the UK accountng standards. Both non-publc and publc companes fnancal statements must be audted by a professonal and a qualfed audtng frm f the company s annual turnover s more than one mllon pounds. However, publc frms should provde some addtonal nformaton to the general publc to be lsted at the London Stock Exchange. Hence, frm-specfc nformaton compled from ths source s compatble across publc and non-publc frms. 3.2 The FAME Database As mentoned earler, accordng to the UK Companes Act, all lmted lablty companes must submt ther annual fnancal statements to Companes House durng a specfc perod of tme from the year-end date. Once a company fles ts accountng statements, Companes House carefully nvestgates and checks ths nformaton and makes t avalable to the general publc. Jordans, one of the leadng provders of legal nformaton n the UK, collects ths data from Companes House. Fnally, BvD collects the data from Jordans and makes t avalable through the FAME database. The FAME database provdes nformaton on both actve and nactve publc/non-publc lmted lablty companes n the UK up to a maxmum of a 10-year perod. The data coverage may vary n terms of the number of observatons for a gven company as there may be entry or ext from the dataset. The man advantage of the FAME database s that t ncludes both balance-sheet and off-balance sheet nformaton, such as ncome statements, cash flows statements, proft and loss accounts and ownershp nformaton. Frms n the database operate n a wde range of ndustral sectors ncludng agrculture, forestry and mnng, manufacturng, constructon, retal and wholesale, hotels and restaurants, the fnancal sector, the publc sector and the regulated utlty ndustry. FAME contans data for both non-publc and publc lmted companes and over 99% of the companes n the database are small and not traded on the stock exchange. Hence, our dataset gves us a unque opportunty 10 Accordng to the Companes Act, a company to be classfed nto medum ( small ) category based on executon of any two of the followng crtera for at least two consecutve years: () annual sales should not be more than 11.2 (2.8) mllon pounds, () book value of total assets should not be more than 5.6 (1.4) mllon pounds, and () the number of workers should not be more than 250 (50). 9
to nvestgate the behavor of non-publc versus publc lmted companes. The FAME database reports two sorts of varables n the form of statc and annual observatons. An annual varable s a varable whose values are reported for each end of accountng year. Whereas, n the case of a statc varable (a header varable), such as ownershp nformaton, company type (publc or non-publc, lsted or unlsted, etc), date of ncorporaton, regstraton number, SIC prmary and secondary codes, only the prevous year s reported value exsts. The FAME database that we use for ths study contaned nformaton for 1999-2008 on both statc varables and annual fnancal statements for approxmately 4 mllon publc and non-publc companes n the UK. All ncorporated enttes are classfed by the 2003 Standard Industral Classfcaton (SIC) codes. 3.3 Sample Selecton Crtera and Intal Screenng In ths paper we only focus on the manufacturng frms and exclude companes that have changed the date of ther accountng year-end by more than a few weeks. The dataset refers to 12-month accountng perods for all companes. As an ntal screenng, we exclude companes that have less than 3 years of consecutve data on debt, nvestment, cash and equvalence, or sales. Second, we set all negatve values for all varables n the sample as mssng. After the ntal screenng, our dataset contans a total of 120,337 frm-year observatons over a ten-year perod from 1999 to 2008. The dataset has an unbalanced panel structure where each frm contrbutes between 3 to 10 years of observatons. Snce there s both entry and ext to the panel over the sample perod, possble selecton and survvorshp bas s to some extent extenuated. We flag each frm as ether publc or non-publc based on ther Company Type as provded by FAME. In the next subsecton, we descrbe the constructon of our frm-specfc and macroeconomc condtonng varables n detal. 3.4 Varable Constructon We construct leverage as the book value of the short-term debt to total assets rato as we am to understand the behavor of publc and non-publc frms short-term debt as uncertanty evolves over tme. 11 We should note that Ttman and Wessels (1988) also use the rato of shortterm debt to total assets as one of the proxes for frm leverage and several other researchers ncludng Marsh (1982), Fama and French (2002), Rajan and Zngales (1995) and Leary and Roberts (2005), defne leverage as a rato of the book value of debt to total assets. 11 It should be noted that the market value of debt s not avalable for non-publc frms. 10
Followng the prevous emprcal studes, we nclude a number of frm-specfc control varables n our emprcal model. We defne nvestment as expendture by the frm on the purchase of fxed tangble assets durng a year. Cash s set equal to cash and equvalents. Sales are defned as the total turnover of the company durng an accountng year perod. To control for the potental nfluence of outlers n our emprcal analyss, all varables that enter nto our model n ratos are wnsorzed at the lower and upper one-percentle to purge the mpact of outlers and reportng errors on the data. 12 Further detals on the varables are gven n the Appendx. 3.5 Generatng Frm-Specfc Uncertanty Researchers mplement dfferent methods to generate a proxy for frm-specfc uncertanty. For nstance, Huznga (1993) uses the condtonal varance obtaned from a GARCH-type specfcaton on wage and materals cost. Pndyck and Solmano (1993) and Caballero and Pndyck (1996) use a geometrc Brownan model to derve the varance of the margnal revenue product of captal. Ghosal and Loungan (2000) measure the frm-level rsk by the standard devaton of the frm s unpredctable proft. Bo (2002) constructs an AR(1) model for sales and then uses the cumulatve standard devaton of the resduals obtaned from the model for each year as a measure of uncertanty. Bo and Lensnk (2005) use stock prce volatlty as well as the volatlty of the number of employees to measure frm-level uncertanty. They compute stock prce volatlty as the dfference between the hghest and lowest stock prce for each underlyng frm normalzed by the lowest prce. To construct volatlty based on employees, they use the coeffcent of varance over a seven-year perod. Baum, Stephan, and Talavera (2009) estmate frm-level uncertanty by calculatng the standard devaton of the closng prce of the frm s shares. Most of the measures descrbed above are well-suted for cases where the focus s on large publcly traded frms as these methods may ntroduce a bas nto the constructed measure of uncertanty for small frms. 13 Gven that the focus of our paper s on the behavor of publc versus non-publc frms, and non-publc frms are much smaller than the publc frms, we follow Morgan, Rme, and Strahan (2004) and compute two separate tme-varyng measures of frmspecfc uncertanty. Ther approach requres us to run a smple model on frm sales scaled by total assets (S t ) usng frm fxed-effects (f ) and year fxed-effects (f t ): 12 See, for nstance, Brav (2009) who appled smlar screenng methods. 13 For more detals on ths ssue, see Comn and Phlppon (2005). S t = f +f t +ψ t (1) 11
where and t denote frm and year, respectvely and ψ t s the whte-nose error term. The absolute value of these resduals, σ level t = ψ t, s then used as a proxy for frm-specfc uncertanty. f and f t stand for frm-specfc fxed and tme effects, respectvely. Our second measure of uncertanty s constructed by estmatng a smlar model on the growth of frm sales ( lns t ). More specfcally, we estmate the followng model: lns t = f +f t +ψ t (2) where and t are as defned above. ψ t s the error term wth zero mean and fnte varance. In partcular, the absolute value of the resduals obtaned from Equaton (2), σ growth t = ψ t, represents the fluctuatons wth respect to both the cross-frm and the cross-year average growth of sales. Smlar to the above model, f and f t stand for frm-specfc fxed and tme effects, respectvely. The nterpretaton of Equaton (1) based on the level of frm sales s smlar. Thus, σ level t = ψ t represents the fluctuatons regardng the cross-frm and the cross-year average of the level of frm sales. We construct a thrd proxy based on Bo (2002) usng sales. To do that we estmate an AR(1) model for sales normalzed by total assets. Usng the one-perod ahead resduals, we compute the cumulatve-volatlty n sales, σt cumulatve. Specfcally, the uncertanty proxy for 2000 s constructed by calculatng the standard devaton of the resduals obtaned from the AR(1) model of sales that uses data for 2000 and 1999. Smlarly, the uncertanty measure for 2001 s constructed calculatng the standard devaton of the resduals obtaned from the same model usng the data for 2001, 2000 and 1999. The process s repeated smlarly. 14 The downsde of ths approach s the loss of one observaton per frm. 3.6 Computng Macroeconomc Uncertanty Smlar to the case of generatng frm-specfc uncertanty, researchers use dfferent methodologes to construct measures of macroeconomc uncertanty. For nstance, Azenman and Maron (1999) use condtonal varances obtaned from government expendtures as a share of GDP, nomnal money growth and the real exchange rate to proxy for macroeconomc uncertanty. Drver, Temple, and Urga (2005) construct a proxy for macroeconomc uncertanty from the condtonal varance of manufacturng output obtaned from a GARCH model. Baum, Stephan, and Talavera(2009) ft a generalzed ARCH model to derve the condtonal varance of the ndex of leadng macroeconomc ndcators as a proxy for the macro-level uncertanty. 15 14 For more detals see Bo (2002). 15 Byrne and Davs (2005) also employ the same methodology to proxy for macro-level uncertanty. 12
In contrast to the researchers above, Ghosal and Loungan (2000) use the movng standard devaton of energy prces and the Federal Fund Rate (FFR) to proxy for macroeconomc fluctuatons. Korajczyk and Levy (2003) use two-year aggregate domestc nonfnancal corporate proft growth, and two-year equty market returns. Several other researchers, ncludng Kaufmann, Mehrez, and Schmukler (2005) and Graham and Harvey (2001), utlze survey-based methods based on the dsperson of forecasts, whch are collected from frm or bank managers, as a measure of macroeconomc uncertanty. In our nvestgaton we follow the ARCH/GARCH methodology to measure macroeconomc uncertanty. To generate macroeconomc uncertanty, gven that companes tend to consder ther producton as well as fnancng decsons, we use monthly observatons for the CPI and T-bll rates and quarterly observatons for GDP for the perod between 1996 and 2008. Once the condtonal varances for each seres are obtaned, we annualze the monthly or quarterly condtonal varances to match the frequency of our uncertanty measure wth that of the panel data. 16 Two measures of uncertanty based on GDP (σ GDP t ) and T-bll rates (σt T bll ) are drectly used as proxes for macroeconomc uncertanty. In addton, we compute the equal weghted condtonal varance ndex (σt Index ) usng the condtonal varance of GDP, CPI and T-bll rates as a thrd measure of macroeconomc uncertanty. 3.7 Summary Statstcs Table 1 provdes the descrptve statstcs for our varables for the full sample, and splt by publc and non-publc frms. We apply nonparametrc equalty tests to examne f the means, medans and standard devatons of those varables that we employ n our models dffer across publc and non-publc frms. We observe that the mean value of leverage for non-publc frms s sgnfcantly hgher than ther publc counterparts over our sample perod. Ths dfference mples that the non-publc frms n our dataset depend more on short-term debt to fnance ther actvtes n comparson to the publc frms. Ths observaton makes sense as debt fnancng s the only means for nonpublc frms to rase funds. Ths observaton s also n lne wth that of Brav (2009) who shows that non-publc frms use relatvely more debt to fnance ther fxed captal nvestments than publc frms. We also observe that the leverage of non-publc frms s more volatle as compared to that of publc frms. Smlarly, there s a sgnfcant dfference between non-publc and publc frms sales-to-total assets ratos. The mean value of the sales-to-total assets rato s 1.60 for 16 Table 5 n the appendx presents the estmated ARCH/GARCH specfcatons. As the table reveals, the estmates on dagnostc tests provde evdence that our models are well-specfed and there s no remanng ARCH effect n the resduals. 13
non-publc frms, whereas, t s 1.08 for the publc frms. Ths rato s also more volatle for the non-publc frms as compared to that of publc frms. The estmates on cash and equvalent do not show any sgnfcant dfference between the two groups. Non-publc frms have a cash and equvalent-to-total assets rato of 12.2% on average, whereas, ths fgure s 11.1% for publc frms. We should also note that, on average, publc frms have hgher nvestment normalzed by total assets as compared to ther non-publc counterparts. The mean value of the nvestment to asset rato s 15% and 18% for non-publc frms and publc frms, respectvely. Ths dfferental s statstcally sgnfcant for the mean and medan values. The sze of the standard devaton for ths varable provdes evdence that publc frms nvestment rates are slghtly more varable than that of non-publc frms over the perod under consderaton. Insert Table 1 about here Table 2 presents summary statstcs of our macroeconomc and dosyncratc uncertanty measures. The table reports the means, standard devatons, as well as the 25th, 50th and 75th percentles of these proxes. There are several consderable dfferences along wth a few common characterstcs across our measures of dosyncratc and macroeconomc uncertanty. We fnd that the standard devaton of the uncertanty measure based on the level of sales s hgher than that based on the growth of sales. The condtonal varance of the gross domestc product s also more volatle as compared to the condtonal varance of Treasury bll rates. To nvestgate whether our uncertanty proxes gauge smlar movements n the busness and macroeconomc envronment, we nvestgate the correlatons between our measures of uncertanty. The estmates reported n Table 3 show that the correlaton coeffcents are very low and they are not sgnfcant at any reasonable level of sgnfcance. Hence, we conclude that each of our measures captures a dfferent aspect of the uncertanty n the envronment that frms operate n. Insert Table 2 and 3 about here In Table 4 we report smple correlaton coeffcents between our man varables and leverage for non-publc and publc frms n two separate panels. For both types of frms (publc and nonpublc), leverage has a negatve correlaton wth the sales to total asset rato. Ths assocaton s weaker and statstcally nsgnfcant n the case of publc frms, reflectng that the optmal leverage may be more senstve to sales for non-publc frms as compared to publc frms. The level of cash and equvalent s sgnfcantly and negatvely correlated wth leverage for both non-publc and publc frms. Ths correlaton suggests that cash rch frms borrow less. 14
We also fnd that the correlaton between leverage and the nvestment rate s sgnfcant and postve for both groups. The ntensty of ths relatonshp s consderably hgher for publc frms as the magntude of the correlaton coeffcent s 0.45, whle, for non-publc frms, ths magntude s only 0.17. Ths evdence suggests that publc frms use relatvely more short-term debt to fnance ther nvestment opportuntes than the non-publc frms. Regardng the correlaton between uncertanty and leverage, the table provdes some mportant lnkages. In fact, Table 4 provdes prelmnary evdence on the assocaton between uncertanty and frms leverage. From the table, we can observe that there s a sgnfcant negatve assocaton between leverage and two frm-specfc uncertanty measures one measure based on level and the other based on cumulatve sales. In contrast, the measure of volatlty based on growth of sales s postvely correlated wth frm leverage. When we nspect the correlatons between macroeconomc uncertanty and frm leverage we fnd for both publc and non-publc frms that uncertanty measures based on Treasury bll rates and gross domestc product and leverage are negatvely correlated. In summary, these observatons suggest that the leverage of UK non-publc and publc manufacturng frms has a negatve relaton wth macroeconomc uncertanty. However, to properly examne the causal effects of both types of uncertanty, we need to have a well-specfed model whch ncorporates the relevant frm-specfc varables whle consderng the leverage dynamcs. 4 Econometrc Framework Insert Table 4 about here 4.1 Specfcaton of the Baselne Emprcal Model To examne the assocaton between uncertanty and leverage we estmate separately and jontly several models for publc and non-publc frms. We formulate our baselne model by augmentng a standard model that examnes leverage wth measures of uncertanty. Our model, among others smlar to Brav (2009), Baum, Stephan, and Talavera (2009) and Auerbach (1985), contans the lagged leverage rato (lagged dependant varable) to control for the persstence of debt holdngs. Specfcally, we express the model n the followng form: Lev t = λ 0 +λ 1 Lev t 1 +λ 2 Sales t +λ 3 Cash t +λ 4 Invt t +λ 5 σ frm t 1 +λ 6σ macro t 1 +f +ε t (3) where subscrpt and t denote frms and years, respectvely. Lev t s the leverage rato n year t for frm and s defned as the rato of short-term debt to total assets. Sales t, Cash t and Invt t denote sales, cash and equvalents and fxed nvestment, correspondngly, and each 15
varable s normalzed by total assets to remove scale effects. In ths model we nvestgate the mpact of the begnnng of the perod uncertanty on leverage. Hence, uncertanty enters the model wth a lag. σ frm t 1 n year t. σ macro t 1 s one of our tme-varyng frm-specfc uncertanty measures for frm denotes one of our tme-varyng macroeconomc uncertanty measures. f denotes frm-specfc fxed effects, and ε t s the error term. All estmatons are carred out for the perod 1999-2008. The key coeffcents of nterest are λ 5 and λ 6 whch capture the effects of frm-specfc and macroeconomc uncertanty on the frm s leverage, respectvely. Partcularly, we are nterested to see f these coeffcents attan a negatve or a postve sgn so that we can determne the effect of uncertanty on the leverage of publc and non-publc manufacturng frms. 4.2 Dfferental Effects of Uncertanty Whlst estmatng the effects of uncertanty on the frm s short-term leverage, Equaton (3) does not enable us to test whether the mpact of uncertanty on publc frms s statstcally dfferent from that of non-publc frms. To scrutnze ths ssue, we extend our basc model so that all varables of nterest can assume a dfferent coeffcent across publc and non-publc frms wthn the same framework. To acheve our goal we generate two sets of dummes that allow us to separate publc frms from non-publc frms and nteract them wth all varables n the model. Specfcally, we generate a publc-frm dummy (D publc ) whch s equal to one f the frm s categorzed as a publc frm and zero otherwse. We then generate a dummy for non-publc frms (D nonpublc followng form: Lev t = φ 0 +φ 1 Lev,t 1 D publc ) whch s equal to (1 D publc ). In partcular, the extended model takes the +φ 5 Cash t D publc +φ 2 Lev t 1 D nonpublc +φ 6 Cash t D nonpublc +φ 3 Sales t D publc +φ 7 Invt t D publc +φ 9 σ frm t 1 Dpublc +φ 10 σ frm t 1 Dnonpublc +φ 11 σt 1 macro D publc +φ 4 Sales t D nonpublc +φ 8 Invt t D nonpublc +φ 12 σ macro,t 1 D nonpublc +f +ε t (4) The rest of the varables are the same as above. We prefer ths approach over estmatng leverage models on separate sub-samples of publc and non-publc frms owng to the followng two reasons. Frst, our approach allows us to work wth hgher degrees of freedom. Second, our approach allows us to properly test the dfferental effects of uncertanty on leverage for both 16
groups of frms. 17 More specfcally, we test the followng two hypotheses: H 1 0 H 2 0 frms. : The mpact of σfrm t 1 on Lev t s the same across frm-years for publc and non-publc frms. : The mpact of σmacro t 1 4.3 Spllover Effects of Uncertanty on Lev t s the same across frm-years for publc and non-publc Baum, Caglayan, Stephan, and Talavera (2008) develop a partal equlbrum model of precautonary demand for lqud assets to examne how macroeconomc uncertanty and dosyncratc uncertanty affect frms cash holdngs. Ther emprcal results ndcate that uncertanty has a sgnfcant mpact on the non-fnancal US frms optmal lqudty and frms ncrease ther demand for lqud assets n response to an ncrease n ether macroeconomc uncertanty or frm-specfc uncertanty. 18 Snce a frm s fnancng polcy markedly depends on the frm s nvestment opportuntes and avalablty of nternal funds, uncertanty s lkely to have ndrect (spllover) effects, possbly through ts mpact on cash holdngs, as well whle drectly affectng frms captal nvestment or borrowng behavor. In fact Baum, Caglayan, and Talavera(2010) provde evdence that uncertanty affects frms captal nvestments on ts own (the drect effect of uncertanty) and through ts mpact on those frms cash holdngs (the ndrect effect of uncertanty). To see whether the effects of uncertanty spll over to frms leverage behavor through ts effects on frms cash holdngs, we augment our basc specfcaton by ncorporatng cash-holdng-uncertanty nteractons. In partcular, we estmate the followng augmented model: Lev t = β 1 Lev t 1 D publc +β 5 Cash t D publc +β 2 Lev t 1 D nonpublc +β 6 Cash t D nonpublc +β 3 Sales t D publc +β 7 Invt t D publc +β 4 Sales t D nonpublc +β 8 Invt t D nonpublc +β 9 σ frm t 1 Dpublc +β 10 σ frm t 1 Dnonpublc +β 11 σt 1 macro D publc +β 12 σ frm,t 1 Dnonpublc +β 13 Cash t σ frm t 1 Dpublc +β 14 Cash t σ frm t 1 Dnonpublc +β 15 Cach t σt 1 macro D publc +β 16 Cash t σ macro t 1 D nonpublc +β 0 +f +ε t (5) We assess the spllover effects of dosyncratc uncertanty on the frm s leverage by nvestgatng 17 Ths approach also allows one to test the dfferental effects of the remanng varables across publc versus non-publc frms. Nevertheless, we leave ths step to the nterested reader to save space and concentrate on the effects of uncertanty on frms leverage. 18 Almeda, Campello, and Wesbach (2004) also show that macroeconomc condtons have a sgnfcant mpact on fnancally constraned frms cash holdngs. 17
the sgnfcance of β 13 and β 14 n Equaton (5): H0 1 : β 13 = 0, for publc frms. H0 2 : β 14 = 0, for non-publc frms. To examne the spllover mpact of macroeconomc uncertanty on leverage, we test the sgnfcance of β 15 and β 16 n Equaton (5): H0 3 : β 15 = 0, for publc frms. H0 4 : β 16 = 0, for non-publc frms. The null hypotheses suggest that dosyncratc volatlty as well as macroeconomc volatlty affect leverage n conjuncton wth movements n frms cash holdngs. If the presumptons are ncorrect, the hypotheses wll be rejected. 4.4 Estmaton Procedure The endogenety problem n the data requres us to use an nstrumental varable (IV) approach. Hence we use a robust two-step system dynamc panel data (DPD) estmator (system GMM approach) developed by Blundell and Bond (1998) to estmate our models. Whle mplementng ths methodology, fxed effects are removed by desgn as the model s estmated n frstdfferences. The estmaton procedure combnes equatons n dfferences of the varables wth equatons n levels and controls for possble endogenety problems by usng the lagged values of the regressors as nstruments. Fnally, ths approach s qute flexble and allows the researcher to make use of dfferent nstruments wth dfferent lag structure for both the levels and the frst-dfferenced equatons. To test for the valdty of the nstruments we use the J-statstc of Hansen (1982). Ths statstc s asymptotcally dstrbuted as χ 2 wth degrees of freedom equal to the number of overdentfyng restrctons (.e., the number of nstruments less the number of estmated parameters). Under the null hypothess, the nstruments are orthogonal to the errors. To examne the presence of seral correlaton n the error terms, we employ the Arellano and Bond (1991) test for autocorrelaton. Under the null of no seral correlaton, the test asymptotcally follows a standard normal dstrbuton. It also provdes a further check on the correct specfcaton of the System-GMM process. In a dynamc panel data context, the frst-order seral correlaton s lkely to be present, but the resduals should not exhbt the second-order seral correlaton f the nstruments are strctly exogenous. The estmates from the J test are reported n each table that we present below. These estmates ndcate that the nstruments used n the System GMM estmatons are approprate and satsfy the orthogonalty condtons. The Arellano-Bond AR(2) tests do not provde any 18
evdence for the presence of second-order seral correlaton n the resduals. Ths ndcates the use of our nstruments are approprate. Hence, for brevty, we do not make any further comments on those aspects when we dscuss our results. 5 Emprcal Fndngs We commence our emprcal analyss estmatng the effects of dosyncratc uncertanty on leverage usng three dfferent measures. Then we carry out the same exercse wth macroeconomc uncertanty. Once we establsh the effects of each type of uncertanty separately, we ncorporate both types of uncertanty measures nto our model as n Equaton (3). Usng a smlar approach, we next nvestgate whether uncertanty has a dfferental mpact on the leverage of non-publc versus publc frms as Equaton (4) depcts. Last but not least, we estmate Equaton (5) to examne f the effects of uncertanty spll-over to leverage through ts mpact on the cash holdngs of the frms, followed by a dscusson on the total mpact of uncertanty on frms leverage. 5.1 The Impact of Uncertanty on Leverage 5.1.1 The Role of Frm-specfc Uncertanty Table 6, Panel A, presents our results on the mpact of dosyncratc uncertanty on leverage. In addton to measures of uncertanty, the regresson model ncludes lagged leverage, sales, cash and nvestment to total asset ratos as frm-specfc explanatory varables. Lagged leverage attans a postve sgn provdng evdence on the persstence of leverage: frms that borrowed n the prevous perod contnue to use debt fnancng. Coeffcents of Sales and Cash to total asset ratos are sgnfcant and negatve as expected mplyng that an mprovement n sales and cash holdngs enables frms to borrow less funds. The nvestment rate s postve suggestng that ncreases n captal nvestment lead to an ncrease n the short-term debt of frms. Our fndngs for the frm-specfc varables are generally consstent wth the prevous emprcal work ncludng that of Ttman and Wessels (1988), Fama and French (2002), Rajan and Zngales (1995) and Brav (2009). Hence, we do not further dscuss the sgn and sgnfcance of these varables, nstead, we concentrate on the effects of uncertanty on leverage. All these varables n the remanng tables attan smlar sgns and sgnfcance as n Table 6. Table 6 dsplays the mpact of three dfferent measures of frm-specfc uncertanty on leverage. Model1consdersthempactofuncertantybasedonthelevelofsales. Model2mplements the mpact of volatlty based on the growth of sales and Model 3 estmates the mpact of cumulatve volatlty constructed as n Bo (2002) based on the level of sales. Gven the correlatons 19
depcted n Table 3, we beleve that each measure captures a dfferent aspect of uncertanty n the busness envronment yet we expect to fnd that an ncrease n uncertanty would adversely affect leverage. That s, as uncertanty n frm s operatons ncrease, we should expect to see a reducton n the use of short-term debt causng a declne n frms leverage. Our ratonale behnd ths predcton s that a hgher busness rsk ncreases the chance of bankruptcy and, as a result frms use less debt. Equally, t s possble that banks or other fnancal nsttutons do not lend to those frms that experence hgher busness rsk to protect themselves from potental losses. The key fndng emergng from Table 6 s that there s a sgnfcant negatve assocaton between dosyncratc uncertanty and leverage. For each model depcted n the table, we observe that uncertanty attans a sgnfcant and negatve coeffcent. Overall, our fndngs are consstent wth Ttman and Wessels (1988), MacKe-Mason (1990), Wald (1999) and Baum, Stephan, and Talavera (2009) that frm-level rsk has a negatve and sgnfcant mpact on leverage. 5.1.2 The Role of Macroeconomc Uncertanty Panel A, Table 7, provdes the estmates of a model smlar to Table 6 except that we now concentrate on the effects of macroeconomc uncertanty on frms leverage. Here, too, we mplement three dfferent measures of uncertanty to capture the turmol n macroeconomy. In partcular, Models 1 and 2 use uncertanty measures based on gross domestc product, and Treasury bll rates, respectvely. Model 3 uses a weghted uncertanty ndex based on gross domestc product, Treasury bll rates and the consumer prce ndex. We expect that there s a negatve relatonshp between macroeconomc volatlty and frms borrowng behavor. Ths can be ratonalzed as follows. Hgher macroeconomc uncertanty rasng the frm s busness rsk deterorates the corporate tax shelter and ncreases the chance of nsolvency. In such an uncertan state of the economy, frms managers would generally be more cautous about the costs of fnancal dstress and they therefore reduce the level of debt as debt makes ther frms more exposed to macroeconomc rsk. In all three models, we observe that macroeconomc uncertanty has a sgnfcant and negatve mpact on frms leverage. Although, the ntensty of the estmated effects of macroeconomc uncertanty on leverage depends on the uncertanty measure used, the negatve lnk s apparent. Overall, our fndngs suggest that frms use consderably less debt n ther captal structure when the macroeconomc clmate s volatle. The negatve macroeconomc uncertanty-leverage relatonshp s n lne wth our predcton that frms reduce ther short-term debt fnancng durng 20