Corporate Governance and Lqudty # Kee H. Chung a,*, John Elder b, and Jang-Chul Km b a School of Management, State Unversty of New York (SUNY) at Buffalo Buffalo, NY 14260, USA b College of Busness, North Dakota State Unversty Fargo, ND 58105, USA Current verson: October 2007 # The authors are solely responsble for the content. The authors thank Insttutonal Shareholder Servce (ISS) for provdng corporate governance data. The authors are grateful to two anonymous JFQA referees and the edtor for ther detaled comments whch mprove the paper greatly. The authors also thank Xn Zhao and Robert Van Ness for programmng help, and Yang Cao, Pankaj Jan, Jacky Jang, Robert Keschnck, Dana Knyazeva, Kenneth Km, Kaye Lee, Matthew Spegel, Joakm Westerholm, Hao Zhang, and sesson partcpants at the 2007 FMA conference for valuable comments and dscusson. *Correspondng author. Tel.: 716-645-3262; Fax: 716-645-3823; e-mal: keechung@buffalo.edu.
Corporate Governance and Lqudty Abstract We nvestgate the emprcal relaton between corporate governance and stock market lqudty. We fnd that frms wth better corporate governance have narrower spreads, hgher market qualty ndex, smaller prce mpact of trades, and lower probablty of nformaton-based tradng. In addton, we show that changes n our lqudty measures are sgnfcantly related to changes n the governance ndex over tme. These results suggest that frms may allevate nformaton-based tradng and mprove stock market lqudty by adoptng corporate governance standards that mtgate nformatonal asymmetres and deter nsder tradng. Our results are remarkably robust to alternatve model specfcatons, across exchanges, and dfferent measures of lqudty. JEL Classfcaton: G10, G34 Keywords: Corporate governance, Spreads, Prce mpact, Informaton-based tradng, Lqudty
I. Introducton Corporate governance ams to protect shareholder rghts, enhance dsclosure and transparency, and facltate effectve functonng of the board. La Porta et al. (2000) defne corporate governance as a set of mechansms through whch outsde nvestors protect themselves aganst expropraton by the nsders. Pror research examnes how nternal corporate governance (e.g., board structure, manageral compensaton, and charter provsons) and external corporate governance (e.g., legal/regulatory envronments and markets for corporate control) affect frm value, cost of captal, and stock returns. 1 In ths paper, we analyze the relaton between corporate governance and stock market lqudty. Several recent studes examne the relaton between external corporate governance and lqudty utlzng cross country dfferences n legal and regulatory envronments. For example, Bacdore and Sofanos (2002) show that, among New York Stock Exchange (NYSE)-lsted companes, those based n the U.S. exhbt hgher stock market lqudty than those based outsde the U.S. Brockman and Chung (2003) show that, among companes lsted on the Stock Exchange of Hong Kong, those based n Hong Kong have narrower spreads and greater depths than those based n manland Chna. They nterpret ths fndng as evdence that poor nvestor protecton results n poor lqudty. Smlarly, Chung (2006) shows that Amercan Depostory Recepts (ADRs) of companes operatng n countres wth stronger nvestor protecton mechansms exhbt narrower spreads. Eleswarapu and Venkataraman (2006) show that companes n countres wth better judcal effcency, hgher accountng standards, and hgher poltcal stablty exhbt hgher stock market lqudty. In contrast to the above studes that focus on dfferences n lqudty due to legal and regulatory envronments, our study focuses on dfferences n lqudty due to nternal corporate governance. The proposton that nternal corporate governance s related to stock market lqudty s not orgnal to our study. Coffee (1991), for example, argues that large nvestors have ncreasngly supported measures that mprove nternal corporate governance because such measures also mprove stock market lqudty (whch 1 See Shlefer and Vshny (1997), La Porta et al. (2000), Mtton (2002), Gompers, Ish, and Metrck (2003), Bebchuk and Cohen (2005), Bebchuk, Cohen, and Ferrell (2005), Ch (2005), Ashbaugh, Collns and LaFond (2006), and Masuls, Wang, and Xe (2006). See Gllan (2006) for a recent survey of corporate governance lterature. 1
makes ther ext less costly). Bhde (1993) holds that hgh stock market lqudty dscourages nternal montorng (by actve stockholders) and the benefts of market lqudty must be weghed aganst the cost of mpared shareholder actvsm. In contrast, Faure-Grmaud and Gromb (2004) show that nformaton generated by lqud markets ncreases the large shareholder s ncentve to undertake value-enhancng actvtes, such as montorng. Usng an alternatve defnton of lqudty (.e., the ablty to trade anonymously) several authors also show that lqudty ncreases the ncentve to montor by lowerng the cost of acqurng large postons [see Kahn and Wnton (1998), Maug (2002), and Noe (2002)]. To our knowledge, however, the emprcal relaton between nternal corporate governance and stock market lqudty has not yet been establshed. 2 We conjecture that corporate governance affects stock market lqudty because poor governance mples poor fnancal and operatonal transparency, 3 whch ncreases nformaton asymmetres between nsders (e.g., mangers and large shareholders) and outsde nvestors (e.g., outsde owners and lqudty provders), as well as among outsde nvestors. Poor transparency nsulates management and mpedes the ablty of traders to dscern the extent to whch management can exproprate frm value through shrkng, empre buldng, rsk averson, and perqustes [see Gompers, Ish, and Metrck (2003) and Bebchuk, Cohen, and Ferrell (2005)]. Damond (1985) shows that such nformaton asymmetres between management and traders ncrease the latter s ncentve to acqure prvate nformaton, leadng to greater heterogenety among trader belefs and larger speculatve postons among nformed traders. Lqudty provders may therefore post wder spreads and smaller depths for stocks of poorly governed companes because they face greater adverse selecton problems n these stocks (Glosten and Mlgrom, 1985). For the same reason, the prce mpact of trades (Kyle, 1985) may tend to be greater for stocks of companes wth poor governance structure. 2 Attg et al. (2006) show that poor nformaton dsclosure by self-servng owners reduces stock market lqudty usng a sample of Canadan stocks. They do not examne the relaton between corporate governance and lqudty. 3 Transparency, as descrbed by the OECD Prncples of Corporate Governance, nvolves the tmely dsclosure of adequate nformaton concernng a company s fnancal performance, as well as commercal objectves, ownershp structures, remuneraton, related party transactons, governance structures, and nternal controls. 2
Corporate governance may nfluence lqudty also because t s easer for nsders, ncludng managers and large shareholders, to explot prvate nformaton through nformaton-based tradng when nvestor nterests are poorly protected. Identfyng such tradng s dffcult because corporate nsders rarely trade drectly on ther nformaton, and also because large shareholders may not conduct dscrete tradng that are readly dentfed wth nsde nformaton. Instead, nsders may gve ther nformaton to confederates who trade on ther behalf, or they may alter postons over long perods of tme [see Gannett and Smonov (2006) and Harrs (2003)]. Consequently, lqudty provders are lkely to mantan wder spreads and smaller depths for stocks of companes wth poor nvestor protecton (.e., poor governance) n antcpaton of greater losses to nformed traders. Theory therefore suggests that poor corporate governance may mpar stock market lqudty to the extent that poor governance s assocated wth low transparency and poor nvestor protecton. In ths broad context, we examne the effect of corporate governance on lqudty usng an ndex of governance attrbutes that are lkely to affect fnancal/operatonal transparency and nvestor protecton. 4 Our governance ndex, whch s based on data compled by Insttutonal Shareholder Servces (ISS), conssts of 24 such governance attrbutes. Our measures of lqudty nclude quoted spreads, effectve spreads, and an ndex of market qualty for a large sample of NYSE/AMEX and NASDAQ stocks. To examne the relaton between corporate governance and nformaton asymmetres more drectly, we also estmate two measures of nformaton-based tradng, the prce mpact of trades and the probablty of nformatonbased tradng as derved by Easley, Kefer, O Hara, and Paperman (1996). Our results show that stocks of companes wth better governance structure exhbt narrower quoted and effectve spreads, hgher market qualty ndex, smaller prce mpact of trades, and lower 4 In recent years many U.S. companes have adopted new standards to mprove corporate governance. Some of these standards were mandated by new lstng requrements (e.g., by the NYSE, AMEX, and NASDAQ), some were requred by addtonal regulatory structure (e.g., by the SEC and the Sarbanes-Oxley Act), and others were nether mandated nor requred. For example, our governance ndex for D.R. Horton, Inc. (DHI) ncreased sharply between 2002 and 2003 when t adopted several new governance standards, such as allowng shareholders to call specal meetngs and requrng only a majorty shareholder vote (rather than supermajorty) to approve mergers. Incdentally, DHI also experenced a dramatc mprovement n lqudty over the same perod, wth a 50% ncrease n the market qualty ndex (defned by the rato of quoted depth to quoted spread). Whether such a relaton between corporate governance and lqudty s systematc, of course, requres a more formal emprcal analyss that, n part, controls for other relevant factors. 3
probablty of nformaton-based tradng. The estmated mprovement n lqudty s economcally sgnfcant, wth an ncrease n our governance ndex from the 25 th to 75 th percentle decreasng quoted spreads on NASDAQ by about 4.7%. Our results are robust to dfferent estmaton methods (ncludng fxed effects and error component model regressons), across markets, and alternatve measures of lqudty. In addton, we fnd that changes n our lqudty measures are sgnfcantly related to changes n governance scores over tme. These results suggest that frms may allevate nformaton-based tradng and mprove stock market lqudty by adoptng corporate governance standards that mtgate nformaton asymmetres and deter nformed tradng. The paper s organzed as follows. Secton II presents the detaled descrpton of the measures of corporate governance and stock market lqudty and ther descrptve statstcs. Sectons III presents our emprcal fndngs. Secton IV concludes the paper. II. Varable Measurement, Data Sources, and Descrptve Statstcs In ths secton we dscuss our varable measurement procedures, data sources, and descrptve statstcs of the key varables used n the study. A. Corporate Governance Metrcs An ndex of corporate governance that s relevant for stock market lqudty requres data on governance standards that would, n theory, mprove fnancal/operatonal transparency and nvestor protecton. Exstng metrcs of corporate governance are not completely adequate n ths regard. For example, a well-known ndex of corporate governance developed by Gompers, Ish, and Metrck (2003) (GIM) s desgned prmarly to capture ant-takeover provsons n a frm s charter, bylaws, and state law. Because our applcaton s based on a broader nterpretaton of corporate governance, we develop our own ndex usng the data provded by Insttutonal Shareholder Servce (ISS). The ISS data are very broad, consstng of 51 governance standards n eght categores. From the ISS data, we select 24 governance standards n sx categores that are most closely related to fnancal/operatonal transparency 4
and nvestor protecton. We determne whether a partcular governance standard s met usng the mnmum standard provded n ISS Corporate Governance: Best Practces User Gude and Glossary (2003). We then create an ndex (Gov-Index) for each frm by awardng one pont for each governance standard that s met. Ths method s smlar to the codng method used n Brown and Caylor (2006) for ther ndex. Appendx A shows the 24 governance standards and ther sx categores. We use a governance standard related to the ndependence of the audt commttee (Audt #1) to capture the extent to whch governance may mprove fnancal and operatonal transparency as well as protect shareholder nterests. The audt commttee revews the adequacy and effectveness of nternal audtng, accountng, and fnancal controls of the company. The commttee also revews the audt performed by the company's ndependent audtors and makes recommendatons concernng the appontment of the ndependent audtor. We conjecture that a frm s fnancal and operatonal transparency would be hgher f the audt commttee were composed solely of ndependent drectors. The ndependent audt commttee s also lkely to protect shareholder nterests better. For the same reason, we use nne governance standards that are related to the ndependence and effectve functonng of the board, ncludng key commttees such as the nomnatng and compensaton commttees (Board #1-4, 6, 8-11). Stock compensaton and stock ownershp serve to algn the nterests of drectors and executves wth those of shareholders. The board's legal charge of fulfllng ts fducary oblgatons s put to the ultmate test through the task of settng ts own compensaton. Drector compensaton packages should be desgned to provde value to drectors for value receved. Gven that many drectors are hgh-level executves whose personal ncome levels are generally hgh, cash compensaton may hold lttle appeal. Stock-based ncentves better renforce the drectors' role of protectng and enhancng shareholder value. Lkewse, drectors and executves who own company stock are more lkely to act on behalf of shareholders because ther own nterests are algned wth those of shareholders. 5 Based on these 5 Accordng to compensaton consultant Graef Crystal, a fve-pont ncrease n the aggregate stock ownershp percentage among drectors s assocated wth a 1.5 percentage pont ncrease n annual shareholder returns. 5
consderatons, we use four governance standards that are related to executve and drector compensaton and ownershp (Compensaton #1 and Ownershp #1, 2, 3). Fnally, attrbutes of corporate governance may also mprove operatonal transparency and protect shareholder nterests to the extent that they mtgate the entrenchment of ncumbent management. We therefore nclude nne governance standards (Board #5, 7 and Charter #1-7) that are related to provsons n the frm s charter and bylaws that, f not mplemented, serve to delay or mpede takeovers. These nne standards mrror eleven standards n the GIM ndex, ncludng two that may be most relevant n ths regard (annually elected boards and a poson pll) and four of the top fve, as dentfed by Bebchuk and Cohen, and Ferrell (2005). Appendx A shows the cross reference of our governance standards to the GIM standards. Eleven of GIM s 24 governance standards are captured n our Gov-Index, wth seven of these clustered n the ISS category charter/bylaws. Gov-Index also ncludes all the standards n GIM s category Delay; four of sx standards n GIM s category Votng; and one standard that proxes for GIM s category State. There are two addtonal reasons why we use data from ISS, rather than the GIM ndex, n our study. Frst, the ISS data are avalable for a much larger number of frms n recent years, for whch we have lqudty data (the ISS data are avalable for more than 2,400 frms at the begnnng of our study perod n 2001 and over 5,000 frms at the end of our study perod n 2004). Second, the ISS data are avalable annually, rather than bannually. B. Lqudty Measures: Spreads, Prce Impact, and the Probablty of Informaton-Based Tradng We obtan data for lqudty varables from the Trade and Quote database (TAQ) provded by the NYSE. Each quote observaton n the data fle ncludes tcker symbol, the quote date, tme-stamp, bd prce, ask prce, bd depth, ask depth, and exchange code. The Natonal Best Bd and Offer (NBBO) quotes are not ncluded n the TAQ database. Hence, we frst generate NBBO quotes usng the program provded by Wharton Research Data Servces (WRDS). We then apply the followng data flters to trades and NBBO quotes, whch are standard n the mcrostructure lterature (see, e.g., Huang and Stoll, 1996), 6
to clean the data of errors and outlers: (1) delete quotes f ether the bd or ask prce s negatve; (2) delete quotes f ether the bd or ask sze s negatve; (3) delete quotes f the bd-ask spread s greater than $4 or negatve; (4) delete trades and quotes f they are out of tme sequence or nvolve an error; (5) delete before-the-open and after-the-close trades and quotes; (6) delete trades f the prce or volume s negatve; and (7) delete trades and quotes f they changed by more than 10% compared to the last transacton prce and quote. We delete unlsted stocks, stocks wth average annual share prces less than $5, and stocks not ncluded n the NYSE s TAQ, the Center for Research n Securty Prces (CRSP), Standard & Poor's COMPUSTAT, or the ISS databases. We calculate the quoted percentage spread of stock (frm) at tme τ as (1) Quoted Spread,τ = (Ask,τ Bd,τ )/M,τ ; where Ask,τ s the ask prce for stock at tme τ, Bd,τ s the bd prce for stock at tme τ, and M,τ s the mean of Ask,τ and Bd,τ. For each stock, we then calculate the tme-weghted average quoted spread durng each year from 2001 through 2004. The quoted spread s the mplct tradng cost for market orders when a trade occurs at the quoted prce wth no prce mprovement. To measure the cost of tradng when t occurs at prces nsde the posted bd and ask quotes, we also calculate the effectve percentage spread of stock at tme τ as (2) Effectve Spread,τ = 2D,τ (P,τ M,τ )/M,τ ; where P,τ s the transacton prce for stock at tme τ, M,τ s the mdpont of the most recently posted bd and ask quotes for stock, and D,τ s a bnary varable whch equals one for customer buy orders and negatve one for customer sell orders. We estmate D,τ usng the algorthm n Ells, Mchaely, and O Hara (2000). [See Bessembnder (2003) for detaled comparatve analyses of dfferent classfcaton methods.] For each stock, we then calculate the trade-weghted average effectve spread durng each year. 7
To the extent that corporate governance can affect both the spread and depth smultaneously, a more comprehensve analyss of the effect of corporate governance on stock market lqudty requres an emprcal measure that captures both dmensons of lqudty. One such measure s the market qualty ndex orgnally suggested by Bollen and Whaley (1998), whch s defned as the rato of the quoted depth to quoted spread. 6 (3) Market Qualty Index, τ = (1/2)Quoted Depth Quoted Spread,τ,τ For each stock, we calculate the tme-weghted average market qualty ndex durng each year. Note that the market qualty ndex cannot be meanngfully calculated from TAQ data for NASDAQ frms because TAQ reports only the sze of the frst nsde dealer quote for NASDAQ frms. We therefore report the market qualty ndex for NYSE/AMEX frms only. As noted earler, another frequently used measure of stock market lqudty s the extent to whch an asset can be bought or sold wthout affectng ts prce. We measure the prce mpact of trades by (4) Prce Impact,τ = 100 D,τ [(M,τ+5 M,τ ) / M,τ ], where M,τ and M,τ+5 are quote mdponts at tme τ and τ + 5 mnutes, respectvely. The mean value of prce mpact durng each year s calculated by weghtng each trade equally. The prce of mpact of trades measures the extent to whch a trade alters the share prce. If a trade does not carry new nformaton on the value of the share, ts prce mpact should be zero on average. In contrast, f the trade s nformaton motvated, t would move the prce to the drecton of the trade buyer-ntated trades rase the prce (.e., quote mdpont) and seller-ntated trades lower the prce. To estmate the probablty of nformaton-based tradng (PIN), we use the sequental trade model of Easley, Kefer, O Hara, and Paperman (EKOP) (1996), appled to each frm over each year. In the EKOP model, market makers observe trades, update ther belefs, and establsh quotes. Ths process of tradng, and learnng from tradng, results n prces convergng to full nformaton values. The EKOP 6 Ths measure assumes a lnear lqudty supply schedule (.e., a lnear tradeoff between the spread and depth), whch may not correctly capture actual preferences of lqudty provders. 8
model provdes a structure necessary to nfer nformaton based tradng from observable varables such as the number of buys and sells. The EKOP model of the trade process for frm over tradng day j s represented by the followng lkelhood functon: (5) L (B,j,S,j θ ) = (1 α )e + α δ e + α (1 δ )e εt,j (ε T B ε T,j,j )! B (µ + ε )T,j,j, j (ε T e B )! (µ + ε )T [(µ + ε )T ],j,j B B,j,j,j e [(µ + ε )T ]! ε T,j, j B,j (ε T S e S,j!,j,j ε T )!,j, j S,j S, j (ε T S,j,j )! S,j ; where B,j s the number of buyer-ntated trades for the day, S,j s the number of seller-ntated trades for the day, α s the probablty that an nformaton event has occurred, δ s the probablty of a low sgnal gven an event has occurred, µ s the probablty that a trade comes from an nformed trader gven an event has occurred, 7 ε s the probablty that the unnformed traders wll actually trade, T,j s total tradng tme for the day, and θ = (α, δ, ε, µ ) represents the vector of parameters to be estmated. We estmate these parameters θ for frm for each year by maxmzng the jont lkelhood over the J observed tradng days n a calendar year: (6) L (M θ) = L(B,j,S,j θ). We then estmate the probablty of nformaton-based tradng (PIN) for frm for each year as J j= 1 (7) PIN αˆ µˆ αˆ µˆ + 2εˆ =. C. Control Varables Although our man research queston s whether corporate governance affects lqudty, we nclude a number of control varables n our emprcal analyses. They are share prce, return volatlty, 7 The EKOP model assumes that buy and sell orders from unnformed traders are equally lkely. 9
tradng volume, frm sze, company age, analyst followng, nsttutonal ownershp, research and development (R&D) expendture, and asset tangblty. We provde further detals on these varables later n the paper. We measure return volatlty by the standard devaton of daly closng quote-mdpont returns, tradng volume by the mean daly dollar tradng volume, frm sze by the book value of total assets, analyst followng by the number of analysts followng the company, nsttutonal ownershp by the percentage of shares held by nsttutons, and R&D expendture by the rato of R&D expendture to sales. We obtan data on analyst followng from the Insttutonal Brokers' Estmate System (I/B/E/S) database, nsttutonal ownershp from the CDA/Spectrum Insttutonal (13f) Holdngs database, and all other data (.e., company age, R&D expendtures, sales, and total assets) from the COMPUSTAT or CRSP databases. Followng Berger et al. (1996) and Almeda and Campello (2007), we measure asset tangblty by [(0.715 * Recevables + 0.547 * Inventory + 0.535 * Captal) + Cash]/Assets, where Recevables s COMPUSTAT tem #2, Inventory s tem #3, Captal s tem #8, Cash s the value of cash holdngs (tem #1), and Assets s the book value of total assets (#6). D. Descrptve Statstcs and Correlaton Matrx Gven the dfferences n both market structure and governance standards for lstng, we report our results separately for NYSE/AMEX and NASDAQ frms. Table 1 shows descrptve statstcs on Gov- Index, lqudty measures, and other stock attrbutes for our study sample of frms. 8 For the NYSE/AMEX frms n our sample, the mnmum and maxmum values of Gov-Index are 3 and 20. The mean (medan) value of Gov-Index s 11.49 (12), ndcatng that, on average, our sample frms meet about half of the governance standards. The summary statstcs on the governance standards for NASDAQ frms are qualtatvely smlar, although somewhat smaller n magntude. 8 For these descrptve statstcs, we assume that a frm has no analyst followng f ts analyst followng nformaton s not ncluded n the I/B/E/S database and zero R&D expendture f ts R&D expendture s not reported n the COMPUSTAT database. See Secton III.A for further detals on our codng of these varables. 10
The descrptve statstcs show that NYSE/AMEX frms n our sample tend to be larger n total assets, have greater tradng volume, and exhbt lower return volatlty than NASDAQ frms. NYSE/AMEX frms tend to be more lqud wth lower effectve spreads and exhbt smaller prce mpact than NASDAQ frms. In addton, NYSE/AMEX frms are older, followed by more analysts, and exhbt hgher nsttutonal ownershp. For example, the mean number of analysts (6.92) for the NYSE/AMEX sample s sgnfcantly greater than the correspondng fgure (4.57) for the NASDAQ sample. Smlarly, the mean percentage (60.33%) of shares held by nsttutonal nvestors for our NYSE/AMEX sample s sgnfcantly hgher than the correspondng fgure (46.58%) for the NASDAQ sample. Note that R&D expendture ratos of some NASDAQ frms are very hgh (the maxmum value s 71.4473), ndcatng that these frms R&D spendng s much larger than ther sales n relatve terms. Table 2 reports the correlaton matrx of the varables. As expected, Gov-Index s sgnfcantly related to our stock market lqudty measures (.e., quoted and effectve spreads, market qualty ndex, prce mpact, and PIN). Not surprsngly, analyst followng s postvely correlated wth frm sze, tradng volume, and nsttutonal ownershp. The results also show that larger frms tend to have hgher governance ndces, hgher share prces, lower return volatlty, and better stock market lqudty (.e., narrower spreads, greater market qualty ndex, smaller prce mpact, and smaller PIN values). III. Regresson Results In ths secton, we examne how our lqudty measures are related to corporate governance after controllng for other possble determnants of stock market lqudty. A. Corporate Governance, Spreads, and Market Qualty Index To examne the relaton between lqudty and corporate governance, we frst regress both the quoted and effectve spreads on Gov-Index and a number of control varables usng the pooled crosssectonal and tme-seres data. Pror studes show that a sgnfcant porton of cross-sectonal and tmeseres varaton n spreads can be explaned by select stock attrbutes such as dollar tradng volume, share 11
prce, and return volatlty. 9 To solate the effect of corporate governance on spreads, we nclude 1/prce, return volatlty, and tradng volume (n log) n the regresson model as control varables. We use the recprocal of share prce (nstead of share prce) because such specfcaton captures more accurately the effect of the tck-sze nduced bndng constrant on spreads when spreads are measured n relatve terms [see Harrs (1994, p. 160)]. 10 We note that Gov-Index and our measures of market lqudty could be spurously correlated because they are related to a common set of varables. Includng the varables that are related to both Gov-Index and market lqudty n the regresson model reduces the possblty that any estmated relaton between Gov-Index and our measures of market lqudty s spurous. For example, frms that are wdely followed by analysts and/or held by nsttutonal nvestors may be pressured to adopt better corporate governance and, at the same tme, exhbt lower spreads due to greater tradng actvty. Smlarly, larger frms may smultaneously exhbt better governance structure because of hgher nvestor nterest and lower spreads because of smaller adverse selecton rsks (e.g., more nformaton s avalable on larger frms). 11 To examne whether corporate governance has an ndependent, drect mpact on lqudty, we therefore nclude analyst followng (.e., the number of analysts followng the company), nsttutonal ownershp (.e., the percentage of shares held by nsttutons), and frm sze (.e., the book value of total assets) n the regresson model. For the same reason, we also nclude company age, asset tangblty, and R&D expendture rato as addtonal control varables. Note also that asset tangblty could reduce asymmetrc nformaton problems because tangble assets payoffs are easer to observe. In contrast, hgh R&D ntensty may ncrease asymmetrc nformaton problems because payoffs from R&D are dffcult to predct. Fnally, we nclude a dummy varable for frms ncluded n the S&P 500 ndex as well as dummy varables for one-dgt SIC ndustry codes to control for any ndex membershp and ndustry effects. Based on these 9 See, e.g., McInsh and Wood (1992), Chung, Van Ness, and Van Ness (1999), and Stoll (2000). 10 We obtan qualtatvely smlar results when we use log(prce) nstead of 1/prce. The results are avalable from the authors upon request. 11 Harrs (1994) uses frm sze as a proxy for the degree of publc nformaton avalable about the stock. 12
consderatons, we estmate the followng regresson model for our study sample of NYSE/AMEX frms, NASDAQ frms, and the combned sample of NYSE/AMEX and NASDAQ frms, respectvely: Quoted Spread,t or Effectve Spread,t = β 0 + β 1 Log(Gov-Index,t ) + β 2 (1/Prce,t ) + β 3 Return Volatlty,t + β 4 Log(Tradng Volume,t ) + β 5 Log(Assets,t ) + β 6 Age,t (8) + β 7 Number of Analysts,t + β 8 Insttutonal Ownershp,t + β 9 Asset Tangblty,t + β 10 R&D,t Expendture + β 11 S&P 500 Dummy + Dummy Varables for One-Dgt SIC Industry Code + ε,t ; where Quoted Spread,t s the tme-weghted mean quoted percentage spread of stock n year t, Effectve Spread,t s the trade-weghted mean effectve percentage spread, Gov-Index,t s the governance ndex, Prce,t s the mean stock prce, Return Volatlty,t s the standard devaton of daly closng quotemdpont returns, Tradng Volume,t s the mean daly dollar tradng volume, Assets,t s the book value of total assets, Age,t s the company age, Number of Analysts,t s the number of analysts followng frm n year t, Insttutonal Ownershp,t s the percentage of shares held by nsttutons, Asset Tangblty,t s a measure of asset tangblty, and ε,t s the error term. We calculate t-statstcs usng Whte s (1980) standard errors and report them n parentheses. Because we combne data from dfferent sources (e.g., analyst followng from the I/B/E/S database, lqudty measures from the TAQ database, and R&D data from the COMPUSTAT database), some varables have many mssng observatons n the merged dataset. The frequency of mssng observatons s partcularly hgh for the number of analysts and R&D expendture because many frms are not ncluded n the I/B/E/S database and also because many frms do not report R&D expendture. 12 To maxmze data utlzaton and also to assess the senstvty of our results to dfferent varable measurement methods, we employ two approaches. In the frst approach, we assume that a frm has no analyst followng f ts analyst followng nformaton s not ncluded n the I/B/E/S database and zero 12 Chung (2000) shows that only 1,947 (62.9%) of the 3,097 NYSE/AMEX companes ncluded n the COMPUSTAT database are covered by the I/B/E/S database and only 1,782 (44.1%) of the 4,042 NASDAQ companes nclude n the COMPUSTAT database are covered by the I/B/E/S database n 1996. 13
R&D expendture f ts R&D expendture s not reported n the COMPUSTAT database. In the second approach, we smply drop all mssng observatons before we estmate the above regresson model. Panel A of Table 3 shows the OLS regresson results usng the frst approach and Panel B shows the results usng the second approach. The results show that the coeffcents on Gov-Index n the quoted spread model are all negatve and sgnfcant for NYSE/AMEX frms, NASDAQ frms, and the combned sample of NYSE/AMEX and NASDAQ frms, regardless of how we treat mssng observatons on analyst followng and R&D expendture (.e., n both Panel A and Panel B). We obtan qualtatvely smlar results for the effectve spread, except that the regresson coeffcent on Gov-Index s nsgnfcant for NASDAQ stocks and the combned sample n Panel A. We can gauge the effect of governance on lqudty by calculatng the margnal effect of an ncrease n the governance ndex from the 25 th to 75 th percentle. For NASDAQ frms, ths corresponds to an ncrease n the governance ndex from 8 to 12. Multplyng the change n the (log) governance ndex by the coeffcent on governance n the quoted spread model yelds a change n spreads that s approxmately -4.7% of the mean quoted spread for NASDAQ frms. Hence, our results suggest that ntroducng governance standards that rase a NASDAQ frm s Gov-Index from the 25 th to 75 th percentle would decrease ts quoted spread by 4.7%, whch s economcally sgnfcant. Overall, these results are n lne wth our conjecture that better corporate governance leads to hgher stock market lqudty. 13 Consstent wth the fndng of pror research, the quoted and effectve spreads are sgnfcantly and postvely related to 1/prce and return volatlty, and negatvely to tradng volume n both markets. The relaton between spreads and frm sze s negatve and sgnfcant for the NASDAQ sample, but the relaton s mxed for the NYSE/AMEX sample. The quoted and effectve spreads are postvely and sgnfcantly related to the number of analysts n both markets, regardless of how we treat the mssng observatons on analyst followng and R&D expendture. Ths result s n lne wth the fndng of Chung 13 We have dentfed a total of 30 stocks that are cross-lsted on the London Stock Exchange or Toronto Stock Exchange n our fnal study sample. Our man results reman the same when we nclude a dummy varable representng these stocks. The results are avalable from the authors upon request. 14
et al. (1995) and Van Ness, Van Ness, and Warr (2001). Chung et al. (1995) nterpret the result as evdence that fnancal analysts have a greater ncentve to follow a stock wth greater nformaton asymmetry because the value of prvate nformaton ncreases wth nformatonal asymmetry, and market makers post wder spreads for stocks that are followed by more analysts. 14 The results show that spreads are negatvely and sgnfcantly related to nsttutonal ownershp for NYSE/AMEX stocks n both panels. One possble nterpretaton of ths result s that nsttutonal nvestors provde effectve montorng of corporate managers and thus reduce the nformaton asymmetry between nsders and lqudty provders. For NASDAQ frms, however, we do not fnd a sgnfcant relaton between spreads and nsttutonal ownershp, ndcatng perhaps that the role of nsttutonal nvestors n corporate montorng s much weaker on NASDAQ. We fnd that stocks ncluded n the S&P 500 ndex have wder quoted and effectve spreads. Gven the fndng of Gompers, Ish, and Metrck (2003) that frms n the S&P 500 have, on average, poorer governance than others, our regresson may be capturng governance features n S&P 500 frms that are not ncluded n our governance ndex. We fnd mxed results for other control varables (e.g., frm age, asset tangblty, and R&D expendture). Our regresson models capture a large fracton of the varaton n quoted and effectve spreads, wth the R 2 for each regresson n excess of 0.56. If corporate governance affects the spread and depth smultaneously, then our emprcal analyss of the relaton between spreads and Gov-Index s an ncomplete characterzaton of the relaton between lqudty and corporate governance. To examne the relaton between corporate governance and lqudty more fully, we regress the market qualty ndex on Gov-Index and the control varables. Because the dependent varable s no longer the spread, we use Log(Prce) nstead of 1/Prce n the regresson model. As noted n Secton II.B, because meanngful market depth data are not avalable (from TAQ) for NASDAQ-lsted stocks, we estmate the model usng only our study sample of NYSE/AMEX stocks and report the regresson results n Table 3. 14 Van Ness, Van Ness, and Warr (2001) also show that the adverse selecton component of the spread s postvely related to the number of analysts followng the frm. 15
The results show that the coeffcents on Gov-Index are postve and sgnfcant n both panels, ndcatng that frms wth hgher governance scores exhbt hgher market qualty. The results also show that market qualty s hgher for frms wth lower return volatlty, greater tradng volume, larger assets, hgher nsttutonal ownershp, smaller R&D expendture, and lower asset tangblty. In contrast, we fnd that younger frms and frms ncluded n the S&P 500 ndex exhbt a lower market qualty ndex. Our regresson models explan a large fracton of the varaton n market qualty, wth R 2 of 0.9 and 0.88 n each panel. Our hypothess lnkng corporate governance to stock market lqudty s that poor governance gves rse to greater nformaton asymmetry between the nsders (e.g., mangers/controllng shareholders) and outsde owners. Such nformaton asymmetres, n turn, amplfy nformaton asymmetres among market partcpants, whch adversely affects lqudty. Our results show that shares of the companes wth hgher Gov-Index tend to be sgnfcantly more lqud, wth narrower spreads and hgher market qualty ndex, than shares of the companes wth lower Gov-Index. These results are remarkably robust across our sample of both NYSE/AMEX frms and NASDAQ frms and wth respect to dfferent varable measurement methods. Our emprcal results thus far support the hypothess that better corporate governance s assocated wth hgher stock market lqudty. B. Robustness Tests In ths secton, we check the robustness of our results wth respect to dfferent estmaton methods. In partcular, we analyze the relaton between lqudty and corporate governance usng two panel-data regresson methods. We frst use the fxed effects regresson method, whch controls for omtted varables that dffer across frms but are constant over tme. Ths method focuses on changes n the varables over tme to estmate the effects of the ndependent varables on the dependent varable. Because the fxed-effects regresson method estmates the relaton between stock market lqudty and corporate governance from the tme-seres varaton n these and other control varables, and because a 16
causal relaton between varables can be tested usng ther tme-seres co-varaton, ths method sheds addtonal lght on the emprcal lnk between corporate governance and stock market lqudty. We report the results of the fxed effects regresson n Panel A of Table 4. Our results are qualtatvely smlar regardless of how we treat the mssng observatons on analyst followng and R&D expendtures, so we report only the results from the nclusve data hereafter (e.g., a frm that s not ncluded n the I/B/E/S database s assumed to have no analyst followng). We fnd agan that quoted and effectve spreads are negatvely and sgnfcantly related to Gov-Index, whle the market qualty ndex s postvely and sgnfcantly related to Gov-Index for both NYSE/AMEX and NASDAQ stocks. These results provde further evdence that better corporate governance mproves stock market lqudty. Note also that the regresson coeffcents on the control varables are qualtatvely smlar to those n Table 3. To further examne the senstvty of our results to dfferent estmaton methods, we also employ the Fuller and Battese (1974) error component model that permts a more general error structure. Ths model assumes that the error term s composed of three ndependent components: one assocated wth the cross-sectonal unts, another assocated wth tme, and the thrd varyng n both dmensons (.e., ε,t = µ + υ t + ω,t ). If the behavor of the cross-sectonal error component s dfferent from the behavor of the error term of a gven cross-sectonal unt over tme, the Fuller-Battese model would gve more accurate coeffcent estmates. Because the Fuller-Battese model requres that the number tme-seres observatons be dentcal across all cross-sectonal unts (.e., companes), we nclude only those companes wth complete data durng the entre four-year study perod. Ths reduces our sample sze sgnfcantly, so we report results from the combned sample of NYSE/AMEX and NASDAQ stocks, although the results obtaned from each market separately are smlar. Panel B of Table 4 reports the results of the Fuller-Battese model. The results ndcate that the quoted and effectve spreads are sgnfcantly and negatvely related to Gov-Index, and the market qualty ndex s sgnfcantly and postvely related to Gov-Index, provdng addtonal evdence to support our hypothess that frms wth better corporate governance exhbt greater lqudty. Note agan that the coeffcents on the control varables are qualtatvely smlar to those n Table 3 and Panel A of Table 4. 17
C. Regresson Results usng Changes n the Varables To further assess the robustness of the relaton between the governance ndex and market lqudty, we also estmate our regresson models usng changes n both the dependent and ndependent varables nstead of levels. Regresson analyses usng changes n the varables have at least two advantages over those usng the level varables. Frst, these regressons are generally less lkely to show spurous relatons between the varables than the regressons usng the level varables. 15 Second, these regressons allow us to examne the longer-term effect of corporate governance on stock market lqudty. We have assumed so far that the relaton between corporate governance and lqudty s contemporaneous, at least on a yearly bass. However, the effect of corporate governance on stock market lqudty may be gradual. For example, a change n Gov-Index n year t may have an mpact on stock market lqudty n both year t and year t + 1. To examne ths possblty, we nclude both the contemporaneous and prevous year s changes n Gov-Index n the regresson model, together wth contemporaneous changes n the control varables. As n Table 5, we report the results from the combned sample of NYSE/AMEX and NASDAQ stocks. The results (see Table 5) show that the coeffcent on the change n Gov-Index s negatve and sgnfcant n both the quoted and effectve spread models, ndcatng that a decrease n spreads tends to be assocated wth an ncrease n the frm s Gov-Index. The results also show that an mprovement n the market qualty ndex occurs when the frm s Gov-Index ncreases. We fnd that the coeffcent on the prevous year s change n Gov-Index s not sgnfcantly dfferent from zero n both the quoted and effectve spread model, ndcatng that the negatve relaton between spreads and Gov-Index s contemporaneous. We fnd that the coeffcent on the prevous year s change n Gov-Index n the market qualty ndex model s postve and sgnfcant at the 10% level. However, the coeffcent (0.073) on the 15 Year-to-year changes n varables provde a stronger test of causal relatons than do levels of these varables because the levels of many varables are cross-sectonally correlated wthout any drect causal lnk. Whle correlatons n changes do not mply causalty ether, a falure to fnd correlaton n changes s lkely to ndcate no causal relaton. 18
prevous year s change n Gov-Index s much smaller than the correspondng fgure (0.2493) for the contemporaneous change n Gov-Index. Hence, t appears that the postve relaton between the market qualty ndex and Gov-Index s largely contemporaneous also. D. Corporate Governance, Prce Impact, and Informaton-Based Tradng Does good corporate governance reduce the prce mpact of trades? To address ths queston, we regress Prce Impact on Gov-Index and the control varables that are ncluded n the spread regresson model. Panel A of Table 6 shows the OLS results and Panel B shows the results of the fxed-effects regresson. Both the OLS and fxed-effects regresson results show that Prce Impact s negatvely related to Gov-Index for both NYSE/AMEX and NASDAQ stocks, although the relaton s statstcally sgnfcant only for the NASDAQ sample. These results suggest that frms wth better governance mechansms exhbt smaller prce mpacts of trades, especally for those lsted on NASDAQ. The smaller average prce mpact of trades for the companes wth hgher governance scores s lkely to be drven by ther smaller nformaton-based tradng. To examne whether better corporate governance results n lower nformaton-based tradng, we regress the probablty of nformaton-based tradng (PIN) on Gov-Index and the same control varables used above. Panel A of Table 6 shows the OLS results and Panel B shows the results of the fxed-effects regresson. Both the OLS and fxed-effects regresson results show that PIN s negatvely related to Gov-Index for both NYSE/AMEX and NASDAQ stocks and the relaton s statstcally sgnfcant only for the NASDAQ sample. These results are consstent wth the above result that companes wth better governance structure exhbt smaller prce mpacts. IV. Summary and Concludng Remarks The separaton of corporate ownershp and manageral control gves rse to a varety of ssues that are collectvely known as the prncpal-agent problem. Pror studes show that companes wth poor corporate governance have worse operatng performance and lower market values. In the present study, 19
we address another problem assocated wth the separaton of ownershp and control that can be mtgated through good corporate governance. We hold that companes wth good corporate governance are lkely to have lqud secondary markets for ther shares because good governance generally encompasses hgh transparency, whch reduces nformaton asymmetres between the nsders and outsde owners/lqudty provders. Lqudty provders are therefore lkely to post smaller spreads and larger depths for stocks of these companes. Corporate governance nfluences stock market lqudty also because t s dffcult for nsders to explot prvate nformaton when nvestor nterests are well-protected. Lqudty provders are therefore lkely to mantan smaller spreads and larger depths for stocks of companes wth good nvestor protecton n antcpaton of smaller losses to nformed traders. Whether these effects on lqudty are dscernable and economcally sgnfcant s an emprcal queston, and our study addresses ths queston. We solate a number of governance standards that are lkely to be related to stock market lqudty and, usng these standards, create a composte ndex of corporate governance for each company. Our emprcal results show that companes wth better corporate governance generally have greater stock market lqudty as measured by narrower quoted and effectve spreads, hgher market qualty ndex, smaller prce mpact of trades, and lower probablty of nformaton based tradng. We also fnd that changes n our lqudty measures are sgnfcantly related to changes n governance ndex over tme, suggestng that frms can mprove stock market lqudty by adoptng better governance standards. Our results are robust to alternatve estmaton methods, across markets, and dfferent measures of lqudty. To the extent that nvestors requre hgher returns n stocks wth lower lqudty, our fndng of a postve relaton between stock market lqudty and corporate governance qualty provdes at least a partal explanaton of why frms wth poor governance structure have lower market values. By establshng a clear emprcal lnk between corporate governance and lqudty, our study shows that t s mportant for companes to adopt corporate governance standards that mprove fnancal/operatonal transparency and better protect shareholder nterests. Our results also suggest that, all thngs beng equal, frms n countres wth more strngent and farer dsclosure rules and better legal protecton of mnorty shareholders are lkely to be valued hgher 20
(through ther enhanced lqudty) than frms n countres wth poor dsclosure rules and poor legal protecton of nvestors. In ths perspectve, the mplementaton of Regulaton Far Dsclosure (Reg FD) by the U.S. Securtes and Exchange Commsson (SEC), whch prohbts selectve dsclosure by publc companes to market professonals and certan shareholders, should ultmately prove to make U.S. corporatons more compettve n global fnancal markets. 16 The Sarbanes-Oxley Act (SOX) of 2002 requres all publcly-traded companes to submt an annual report of the effectveness of ther nternal accountng controls to the SEC. To the extent that SOX mproves corporate governance and fnancal dsclosure, the results of our study also suggest that SOX wll ultmately lkely ncrease market lqudty and frm value. 16 In the past, many frms released mportant nformaton n meetngs and conference calls where most shareholders and the general publc were excluded. The goal of Reg FD s to even the playng feld between ndvdual nvestors and nsttutonal nvestors by mandatng that publcly traded companes must dsclose materal nformaton to all nvestors at the same tme. Reg FD sought to stamp out selectve dsclosure, n whch large nsttutonal nvestors receved market movng nformaton before smaller, ndvdual nvestors. Reg FD changed fundamentally how companes communcate wth nvestors by brngng better transparency and more frequent and tmely communcatons -- perhaps more than any other regulaton n the hstory of the SEC. 21
References Almeda, H., Campello, M., 2007. Fnancal constrants, asset tangblty, and corporate nvestment. Revew of Fnancal Studes 20, 1429-1460. Ashbaugh-Skafe, H., Collns, D., LaFond, R., 2006. The effects of corporate governance on frms credt ratngs. Journal of Accountng and Economcs 42, 203-243. Attg, N., Fong, W., Gadhoum, Y., Lang, L., 2006. Effects of large shareholdng on nformaton asymmetry and stock lqudty. Journal of Bankng and Fnance 30, 2875-2892. Bacdore, J. M., Sofanos, G., 2002. Lqudty provson and specalst tradng n NYSE-lsted non-u.s. stocks. Journal of Fnancal Economcs 63, 133-158. Bebchuk, L. A., Cohen, A., 2005. The costs of entrenched boards. Journal of Fnancal Economcs 78, 409-433. Bebchuk, L. A., Cohen, A., Ferrell, A., 2005. What matters n corporate governance? Workng paper, Harvard Law School. Berger, P., Ofek, E., Swary, I., 1996. Investor valuaton and abandonment opton. Journal of Fnancal Economcs 42, 257-287. Bessembnder, H., 2003. Issues n assessng trade executon costs. Journal of Fnancal Markets 6, 233-257. Bhde, A., 1993. The Hdden Costs of Stock Market Lqudty. Journal of Fnancal Economcs 34, 31-51. Bollen, N., Whaley, R., 2004. Does net buyng pressure affect the shape of mpled volatlty functons? Journal of Fnance 59, 711-753. Bolten, P., Thadden, E., 1998. Blocks, lqudty, and corporate control. Journal of Fnance 53, 1-25. Brockman, P., Chung, D.Y., 2003. Investor protecton and frm lqudty. Journal of Fnance 58, 921-937. Brown, L., Caylor, M., 2006. Corporate governance and frm operatng performance. Avalable at SSRN: http://ssrn.com/abstract=814205. Ch, J., 2005. Understandng the endogenety between frm value and shareholder rghts. Fnancal Management 34, 65-76. Chung, K.H., 2000. Marketng of stocks by brokerage frms: The role of fnancal analysts. Fnancal Management 29, 35-54. Chung, H., 2006. Investor protecton and the lqudty of cross-lsted securtes: Evdence from the ADR market. Journal of Bankng and Fnance 30, 1485-1505. Chung, K.H, McInsh, T., Wood, R., Wyhowsky, D., 1995. Producton of nformaton, nformaton asymmetry, and the bd-ask spread: Emprcal evdence from analysts' forecasts. Journal of Bankng and Fnance 1025-1046. 22