ARTICLE IN PRESS

How does law affect nance? An examination of equitytunneling in Bulgaria$

Vladimir Atanasov a,, Bernard Black b, Conrad Ciccotello c, Stanley Gyoshev d

a Mason School of Business, College of William and Mary, P.O. Box 8795, Williamsburg, VA 23187, USAb University of Texas at Austin, Austin, TX 78712, USAc Georgia State University, Atlanta, GA 30302, USAd Exeter University, Exeter, Devon EX4 4PT, UK

a b s t r a c t

We model and test the mechanisms through which law affects tunneling and tunneling

affects rm valuation. In 2002, Bulgaria adopted legal changes which limit equity

tunneling through dilutive equity offerings and freezeouts. Following the changes,

minority shareholders participate equally in equity offerings, where before they

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/jfec

Journal of Financial Economics

rshey,

Andrei

w and

opean

search

porate

trade data and to Grzegorz Trojanowski for providing the Matlab code to compute oceanic Shapley Values. This project was funded in part through Grant

Number S-LMAQM-00-H-0146 provided by the United State Department of State and administered by the William Davidson Institute. The opinions,

Journal of Financial Economics 96 (2010) 1551730304-405X/$ - see front matter & 2009 Elsevier B.V. All rights reserved.

doi:10.1016/j.jneco.2009.12.005ndings, conclusions, and recommendations expressed herein are those of the authors and do not necessarily reect those of the Department of State or

the William Davidson Institute. Corresponding author. Tel.: +1 7572212954; fax: +1 7572212937.E-mail address: vladimir.atanasov@mason.wm.edu (V. Atanasov).Governance (2008); College of William and Mary, Georgia State University, New Economic School (Moscow, Russia), Southern Methodist University,

Stockholm School of Economics, University of Amsterdam, and University of Kansas for helpful comments. We are indebted to Apostol Apostolov, Roumen

Nikolov, Vassil Golemanski, Vladimir Lazarov and Kamen Dikov for kindly providing rm ownership, earnings, leverage and Bulgarian Stock ExchangeBennedsen, Erik Berglof, Mike Burkart, Petko Dimitrov, Alexander Dyck, Vladimir Gatchev, Mariassunta Gianetti, Martin Grace, Greg Hebb, Mark He

Clifford Holderness, Wendy Liu, Marina Martynova, David Mauer, Chris Muscarella, Enrico Perotti, Jose Luis Peydro-Alcalde, Dimana Rankova,

Shleifer, James Smith, David Stolin, Aris Stouraitis, Per Stromberg, Ajay Subramanian, Josef Zechner, and seminar participants at the American La

Economics Association Annual Meeting (2006), Fourth Asian Corporate Governance Conference, Conference on Empirical Legal Studies (2007), Eur

Finance Association Annual Meeting (2007), Financial System Modernization Conference organized by the European Central Bank, International Re

Conference on Corporate Governance in Emerging Markets (Istanbul, 2007), European Corporate Governance Institute Conference on CorEmerging markets

$We would like to thank Luc Laeven (the referee) for greatly improving the paper. We are also grateful to Franklin Allen, Marco Becht, Mortena r t i c l e i n f o

Article history:

Received 5 October 2007

Received in revised form

3 April 2009

Accepted 7 May 2009Available online 24 December 2009

JEL classication:

G32

G34

K22

Keywords:

Equity tunneling

Preemptive rights

Dilution

Freezeout

Controlling shareholder

Securities lawsuffered severe dilution; freezeout offer price ratios quadruple; and Tobins q rises

sharply for rms at high risk of tunneling. The paper shows the importance of legal rules

in limiting equity tunneling, the role of equity tunneling risk as a factor in determining

equity prices, and substitution by controlling shareholders between different forms of

tunneling.

& 2009 Elsevier B.V. All rights reserved.

ARTICLE IN PRESS

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 1551731561. Introduction

Classic nance theory, motivated by the Coase theorem(Coase, 1960), often presumes that nancial markets canfunction well regardless of the legal environment. Yet, agrowing literature starting with La Porta, Lopez-de-Silanes, Shleifer, and Vishny (1997, 1998) providesevidence that the legal environment is a signicant factorin explaining capital market growth and development.Most of this research, however, says little about thespecic channels by which law affects nancial markets.Both legal protections and nancial market developmentare typically estimated in the aggregate and evaluatedacross countries. This leaves the law and nancialdevelopment literature vulnerable to concerns aboutendogeneity, measurement error, and omitted-variablebiases.

Our paper rst provides a simple model whichunbundles different forms of tunneling [the extractionof rm value by a rms controlling shareholders ormanagers; see Johnson, La Porta, Lopez-de-Silanes, andShleifer (2000)], and derives how each affects rmprotability and valuation. We develop the model partlyto extend existing models of tunneling, but primarily todevelop predictions which we can test using a naturalexperiment in Bulgaria, provided by 2002 anti-tunnelingreforms.

Prior tunneling models (e.g., Shleifer and Wolfenzon,2002) are simple two-period models that, explicitly orimplicitly, examine only a single form of tunneling(effectively, cash-ow tunneling). In our model, in con-trast, there is both cash-ow tunneling (diversion ofongoing cash ow) and equity tunneling (extraction ofvalue via nancial transactions that affect ownershipclaims, rather than the rms operations). We model andstudy empirically two avors of equity tunneling: dilutiveequity offerings (issuance of shares to insiders at below-market value) and freezeouts (forced sale of minorityshares to the controller for below-market value).

We then use the model to examine how 2002 changesto Bulgarian securities law affect tunneling. Bulgaria wentthrough mass privatization in 1998, which was followedby extensive equity tunneling. The 2002 legal changeslimit both dilution and freezeouts, and allow us toexamine how specic rules can affect specic forms oftunneling, rm valuation, and rm protability(as reported to outside shareholders).

Prior to 2002, Bulgarian rms used equity offeringsprincipally to dilute minority shareholders. When rmsissue shares, almost all are purchased by controllingshareholders, often at a large discount to market value.The new law strengthens preemptive rights by requiringpublic companies to distribute publicly tradable warrantsto all shareholders when they issue shares. After 2002,equity offerings are subscribed to roughly pro rata byminority and majority shareholders and rms begin toissue equity to raise capital.

Prior to the 2002 legal changes, freezeout offers are atlarge discounts to market value, which has often beenalready depressed by prior dilutive offerings and investoranticipation of future tunneling. Informal freezeouts(going dark transactions) are common, at effectiveprices approaching zero. During 19992001, nearly 500rms (over half of all the listed rms on the BulgarianStock Exchange) either go dark or conduct freezeouts. Thenew law adds several freezeout protections, includingregulatory approval of freezeout terms and a ban on goingdark transactions. Post 2002, freezeouts of minorityshareholders are at premiums to market value, consistentwith those in developed markets (DeAngelo, DeAngelo,and Rice, 1984), instead of at severe discounts. The ratio offreezeout offer price to sales roughly quadruples.

At the same time, there is evidence that insidersrespond to reduced opportunity for equity tunneling byengaging in more cash-ow tunneling. Post-2002, returnon assets (ROA) declines sharply for rms at high risk ofequity tunneling, relative to lower risk rms, in a rm-xed effects framework with period dummies.

The 2002 legal changes, and controllers responses tothem, also affect equity values. Tobins q levels risesharply for high-equity-tunneling-risk rms relative tolow-risk rms. The value gains from reduced equitytunneling are presumably partly offset by increasedcash-ow tunneling, but there are still large net gains invaluation ratios. These results are economically large androbust to different ways of estimating tunneling risk, anddifferent valuation measures (price/sales, price/earnings,and market/book value of equity).

We measure equity tunneling risk in two ways, withconsistent results. In our principal specication, we userm nancial and ownership characteristics and pre-lawdata on equity tunneling events to estimate each rmsdelisting and dilution propensities at year-end 2001, justbefore the law change. Second, we use ownership todirectly proxy for equity tunneling risk. We expect rmswith a controlling private owner to have high equitytunneling risk because the controlling owner has thepower and often the incentive to engage in dilution orfreezeout. The control group of lower-equity-tunneling-risk rms is rms without a large private blockholder.

Our results have implications for asset pricing researchin emerging markets. In high-tunneling-risk markets,investors must estimate not only expected cash ows(as in any market) but also tunneling risk. We ndevidence that equity tunneling risk varies widely in thecross-section, and that Bulgarian investors consider thisrisk and update their valuation estimates when legal ruleschange. Equity tunneling risk, as a factor in explainingequity prices and expected returns, can complement somecommonly used factors, such as market risk and momen-tum, and interact with others, such as rm size and book/market ratio. Size may correlate with tunneling risk(as we conrm below for Bulgaria), and high book/marketratios could reect high tunneling risk. Investor pricing ofequity tunneling risk factors can also help to explainhome country bias (Kang and Stulz, 1997), as localinvestors may be better equipped to evaluate equitytunneling risk at the rm-level (Black, 2001b; Glaeser,Johnson, and Shleifer, 2001).

Prior studies of how specic legal rulesas opposed tooverall quality of lawcan affect equity tunneling risk, orhow this risk affects rm valuations, are limited. On the

empirical side, the studies most similar in spirit to oursare Nenova (2005) and Carvalhal-da-Silva andSubramanyam (2007), who examine how changes inBrazilian rules providing takeout rights to common sharesduring freezeouts affect the premium of common sharesover economically similar non-voting preferred shares. Onthe theoretical side, Himmelberg, Hubbard, and Love(2002) allow tunneling risk to vary across rms, similarto our approach, but study only cash-ow tunneling.1 Ourpaper is also related to work which examines theconnection between the overall quality of a countryslegal system and rm behavior or nancial contractingpractices.2

The remainder of the paper is structured as follows.Section 2 develops a model of how equity tunnelingthrough dilutive offerings and below-market freezeoutsaffects minority shareholder valuations. Section 3 appliesthe model to the Bulgarian context and derives testable

2002; Djankov, La Porta, Lopez-de-Silanes, and Shleifer,

equity tunneling and derive: (1) how legal rules can affectparticular types of tunneling, and (2) how different typesof tunneling should affect rm protability and valuation.The model is partial equilibrium in natureprices forminority shares are in equilibrium, but we treat owner-ship as exogenous and do not model the controllers

ARTICLE IN PRESS

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173 1572008). A single parameter typically captures the cost oftunneling (e.g., Shleifer and Wolfenzon, 2002). This costincreases with the degree of tunneling, as it must for themodel to have an interior solution, but the cost is notdirectly connected to legal constraints. In contrast, wedevelop a multiperiod model with both cash-ow and

1 Other related studies, in addition to those cited above, include

Black (2001a) and La Porta, Lopez-de-Silanes, and Shleifer (2006), who

assess which aspects of securities law predict stronger securities

markets; Beck, Demirguc-Kunt, and Levine (2005), who examine the

effects of different aspects of legal systems on rms access to external

nance; Bhattacharya and Daouk (2002), who nd a relationship

between enforced insider trading laws and the cost of equity capital;

and Jiang, Lee, and Yue (2008), who report on, but do not directly study,

efforts by Chinese regulators to limit cash-ow tunneling by parent

companies from public subsidiaries. Studies that report evidence of

equity dilution in emerging markets include Black, Kraakman, and

Tarassova (2000) and Baek, Kang, and Lee (2006). Studies of value

extraction through formal freezeouts include Gilson and Gordon (2003),

and Bates, Lemmon, and Linck (2006). Studies of going dark transactions

include Marosi and Masoud (2007), and Leuz, Triantis, and Wang (2008).2 Examples include Qian and Strahan (2007) (effect of creditor

protection on structure of bank loans), Lerner and Schoar (2005) (effect

of legal environment on form of venture capital investment), Laeven and

Woodruff (2007) (effect of judicial quality on rm size), and Doidge,

Karolyi, and Stulz (2007) (connection between quality of legal institu-

tions and rms governance choices).predictions. Section 4 describes the data we rely on.Section 5 provides empirical results on the effect of the2002 legal changes on equity tunneling mechanisms(equity dilution and freezeout). Section 6 reports resultson the relations among tunneling risk, rm protability,and equity valuations. Section 7 concludes.

2. A model of cash-ow and equity tunneling

We begin by developing a model of cash-ow andequity tunneling. Prior tunneling models are highlystylized, and use a two-period framework, which doesnot permit one to distinguish between forms of tunneling(Burkart, Gromb, and Panunzi, 1998; Shleifer andWolfenzon, 2002; La Porta, Lopez-de-Silanes, Shleifer,and Vishny, 2002; Bertrand, Mehta, and Mullainathan,decision on how much to tunnel. Treating ownership asexogenous is appropriate for our empirical setting, inwhich ownership stakes largely ow from the 1998 massprivatization.

In our model, minority shareholders infer, based on thelegal environment and limited information about eachcompany, the extent of cash-ow tunneling and theprobability and magnitude of two forms of equitytunnelingdilutive equity offerings and freezeouts.Changes in law cause minority shareholders to updatetheir estimates and thus, the price they are willing to payfor shares.

2.1. Model setup

There are N rms in the economy, indexed by n, andthree relevant time periods, indexed by t (t=0, 1, or 2). Wegenerally focus on a typical rm, suppress the n-subscript,and assume that this rm initially has one share out-standing. We assume an all-equity rm. The rm has acontrolling shareholder C, who initially owns a0 shares,and minority shareholders who own the remaining(1a0) shares.

Dene the rms intrinsic value per share with notunneling as Vno-tun, and its intrinsic income per share asEno-tun. The rm has assets A0. We assume that cash-owtunneling affects earnings but not assets. The rmsno-tunneling return on assets is ROAno-tun=Eno-tun/A0. Thecontrolling shareholder engages in cash-ow tunneling,diverting a fraction dcf of Eno-tun. Minority shareholdersobserve only the rms income at t=0 after cash-owtunneling3:

Eobs Eno-tun1dcf Let shareholders value shares using a simple no-

growth discounted cash-ow model, in which all ob-served cash ow is paid out as dividends, cash-owtunneling is expected to be permanent, and the discountrate r does not depend on the type or level of tunneling.4

The rms per-share value to minority shareholders, withcash-ow tunneling but no equity tunneling, Vno-eq,equals:

Vno-eq Eobs 1

r Vno-eq1dcf : 1

The controller can also engage in equity tunnelingthrough share dilution or freezeout. Minority share-holders value the rm taking into account the risk offuture equity tunneling. Thus, the rms shares will trade

3 We do not model the determinants of dcf. See Durnev and Kim

(2005) for a model of cash-ow tunneling.4 In follow-up work, we extend the model presented here to a

continuous time/innite horizon framework, and allow for positive

growth. See Atanasov, Black, and Ciccotello (2009). These extensions do

not affect the model predictions we test below.

price Pdilut=Vno-eq*(1ddilut). Because P0 is less than Vno-eq,

tage ownership. If minority shareholders buy no shares,then k=0; if they buy shares pro rata, then k=1; if the rm

ARTICLE IN PRESS

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173158issues new shares principally to outside investors, thenk41. Since the shares are offered at a discount to intrinsicvalue, the controlling shareholder often has an incentiveto minimize k, to the extent permitted by law.

After the offering, the rm will have (1+ i) outstandingshares, of which minority shareholders will own afraction:

1a1 1a0k1a0i

1 i 1a01k i

1 i : 2

After the issuance, the rms no-freezeout value will be[1+ i*(1ddilut)] *Vno-eq, and the per-share value ofminority shares will drop to:

Vno-freeze 1 i 1ddilut

1 i Vno-eq

1ddiluti

1 i

Vno-eq: 3

Minority shareholder wealth, assuming no future free-zeout, will decline by a fraction equal to (all proofs are inthe discount to no-equity-tunneling value ddilut is notdirectly observed, and exceeds the observed discount tomarket price. The minority shareholders acquire a fractionk*(1a0) of the newly issued shares, where k is thefractional take-up of shares by minority shareholders,relative to the number needed to maintain their percen-at a price P0oVno-eq. We assume that minority investorscan estimate rm-level equity tunneling risk based onlegal rules, the characteristics of the rm and thecontroller, and experience at other rms. Their estimatesare correct, on average, but they do not know which rmswill engage in equity tunneling.

For simplicity, we assume that dilution happens onlyonce (at t=1) and that freezeout occurs only at t=2, andonly following a dilutive offering. The model algebrawould become more complex, but neither the intuitionnor the central empirical predictions would change if weincluded a direct to delisting channel. At t=1, thecontrolling shareholder will cause the rm, with prob-ability pd, to issue i new shares to its existing shareholdersat a discount ddilut to the rms value before equitytunneling Vno-eq. If a dilutive offering occurs, there is afurther probability pf that at t=2, the controlling share-holder will acquire all minority shares through a tenderoffer at a discount dfreeze to the value of minority sharesafter dilution but with no freezeout Vno-freeze.

2.2. Specic equity tunneling mechanisms

In this subsection we model how tunneling throughdilution and freezeout affects rm protability andvaluation.

2.2.1. Period 1: dilution

At time t=1, the rm issues i shares at a discountedAppendix A):

Ddilut ddiluti

1 i 1k: 4

If there are no legal protections against dilutive shareofferings, then ddilut can approach one, k can approachzero, i can approachN, and the controller can acquire anarbitrarily large number of shares at an arbitrarily lowprice, thus expropriating the minority shareholders initialwealthDdilutE1. Most legal systems, however, limitdilutive offerings in some way (Atanasov, Durnev, Fauver,and Litvak, 2008). Based on Eq. (4), these rules can beclassied into three main groups: (i) preemptive rightswhich attempt to bring k closer to one; (ii) minimumpricing rules that limit ddilut; and (iii) minority share-holder approval rules, which let the minority limit any ofthese parameters, but usually apply above a minimumoffering size (e.g., i40.2).

2.2.2. Period 2: freezeout

We next analyze the post-dilution scenario in which, att=2, the controlling shareholder freezes out the minorityshareholders at a price Pfreeze which is at a discount dfreezeto the no-freezeout per-share value:

Pfreeze Vno-freeze1dfreeze Vno-eq 1ddiluti

1 i

1dfreeze:

5The combination of dilution and subsequent freezeout

reduces minority shareholder wealth by a fraction (proofin Appendix A):

Dfreeze 1a1 Vnofreeze dfreeze

1a0 Vnoeq

1k i1 i 1ddiluti

1 i

dfreeze: 6

With no legal protections, the controlling shareholdercan offer an arbitrarily low freezeout price, so dfreeze canapproach one, and the minority shareholders remainingwealth is expropriated. However, law can limit discountedfreezeout offers through some or all of appraisal rights,minimum price rules, duciary duty rules, minorityshareholder approval, or regulatory approval.

2.3. Equity tunneling and equilibrium share prices

We next model how equity tunneling affects equili-brium share prices. To simplify the algebra while main-taining the intuition, we assume that a dilutive offeringwill be large (ib1). The post-dilution, no-freezeoutfractional wealth loss to minority shareholders from theoffering will be Ddilut=ddilut(1k). Investors will realizethe following payoffs:

No dilution or freezeout: Vno-eq with probability(1pd);Dilution but no freezeout: [1ddilut(1k)]Vno-eq withprobability pd(1pf);Dilution and freezeout: [1ddilut(1k)] (1dfreeze) Vno-eqwith probability pd pf.*

ARTICLE IN PRESS

param

turn

tunne

partic

rm va

ramet

un=10

Comp

to one

25 to

pd

0

nge %

2555

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173 159tunneling:We can combine these payoffs to determine theequilibrium price of minority shares at t=0.

Proposition 1. If dilutive offerings are large (ib1), themarket price of minority shares at t=0 will equal:

P0 Vno-eq 1deq Vno-tun 1dcf 1deq; 7where deq is the discount due to expected future equity

deq pd f11ddilut1k1pf dfreezeg: 8

Table 1Comparative statics and numerical example.

Panel A shows the expected sign of the marginal effect of the six model

(ROAobs) and observed Tobins q (qobs) for an all-equity rm. Observed re

dcf=fraction of (unobserved) actual earnings that is diverted via cash-ow

pre-dilution rm value. pd=probability of dilutive offering. k=fractionaldiscount at which minority shares are acquired, relative to pre-freezeout

Panel B provides a numerical example of how a shock to tunneling pa

companies: A and B. Both companies have cost of capital r=ROAno-t(pd,A=pf,A=0.75). Company B has no equity tunneling risk (pd,B=pf,B=0).Company B has dcf,B=0.25. A change in law causes k to increase from zero

increases dcf,A from zero to 0.25, while Company B increases dcf,B from 0.

Panel A: Predicted marginal effects (see Proposition 2)

dcf ddilut

ROAobs 0qobs

Panel B: Numerical example of tunneling shock

Firm A

Pre Post Cha

ROAobs 10.0% 7.5%

qobs 0.48 0.75We next derive expressions for Tobins q and ROA as afunction of the model parameters.

Proposition 2. The expressions for ROA and Tobins q are:

ROAobs EobsA0

Eno-tun1dcf A0

ROAno-tun1dcf 9

and

qobs P0A0

ROAnotun1dcf r

1deq ROAobs

r1deq:

10

We can distinguish empirically between different typesof tunneling by using a combination of these nancialmetrics. Cash-ow tunneling will affect both ROA andTobins q. Equity tunneling will affect Tobins q, but notROA. Thus, it may still be feasible empirically to estimatethe effects of equity tunneling on Tobins q (or similarmetrics such as market/book or price/sales) by controllingfor ROA as a proxy for cash-ow tunneling.

2.4. Comparative statics

From Propositions 1 and 2, we can readily derivecomparative statics expressions for how changes in dcfand the ve parameters that enter the expression fordeqk, ddilut, dfreezepf, and pdaffect ROA and Tobins q.We summarize these comparative statics in Table 1, PanelA. Decreasing the risk of dilution (captured by k, pd, andddilut) or freezeout (captured by pf and dfreeze reduces deqand increases Tobins q.

There are also important interaction effects. First, theeffect on share value from decreasing dilution (freezeout)risk is lower when freezeout (dilution) risk is higher.Indeed, if freezeout risk at t=2 is large enough, even

k dfreeze pf

0 0 0

+

Firm B

Pre Post Change %

7.5% 6.5% 150.75 0.65 13eters (assuming a dilutive offering is large) on observed return on assets

on assets ROAobs=Earnings/Assets. qobs=(market value of equity)/Assets.

ling. ddilut=fractional discount at which new shares are issued, relative to

ipation of minority shareholders in an equity offering. dfreeze=fractional

lue. pf=probability of freezeout, given that dilution has already occurred.ers affects observed ROA and observed Tobins q for two hypothetical

%; ddilut=dfreeze=0.6; k=0. Company A has high equity tunneling risk

any A originally does not engage in cash-ow tunneling (dcf,A=0), while

, and dfreeze to drop from 0.6 to zero. In response, Company As controller

0.35.strong preemptive rights and a large discount will notinduce minority shareholders to buy shares at t=1 (seeAppendix A for formal discussion). The intuition is that tobuy shares at a discount, but then get frozen out at aneven lower price, is to throw good money after bad.Thus, legal protections against dilution and freezeoutwork in tandem to reduce the overall equity tunnelingdiscount. Second, a change in legal rules that reduces thetunneling discounts ddilut and dfreeze makes tunneling lessprotable and therefore, may make it less likely (lower pdand pf). Third, cash-ow tunneling and equity tunnelingcould be either complements or substitutesif legal rulesrestrict one form of tunneling, controllers may react bydoing more or less of other forms.

2.5. A numerical example

A numerical example, using inputs which we considerto be realistic for Bulgaria, can provide perspective on themodel implications, and the potential for substitutionbetween equity and cash-ow tunneling. Consider twocompanies, A and B, with identical no-tunneling perfor-mance and cost of capital ROAno-tun=r=10%. The twocompanies are in a country with poor legal protections(Bulgaria before 2002), so that ddilut=dfreeze=0.6, k=0.

ARTICLE IN PRESS

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173160Investors see Company A as having high equity tunnelingrisk, with pd,A=pf,A=0.75, but view Company B as unlikelyto engage in equity tunneling, with pd,B=pf,B=0. Forcash-ow tunneling, the situation is reversed: CompanyA does not engage in cash-ow tunneling (dcf,A=0), whileCompany Bs managers divert 25% of its cash ow(dcf,B=0.25).

Assume next that new laws regulating dilution andfreezeout cause k to increase to one, and dfreeze to dropto zero. In response to lost equity tunneling opportu-nities, Company As controller starts diverting cash ow,so dcf,A rises to 0.25. Increased cash-ow tunneling byrms like Company A provides cover for Company Bsmanagers, who increase their cash-ow tunneling dcf,Bto 0.35.

Table 1, Panel B shows the changes in ROA and Tobinsq for the two companies. Company As ROA drops, butTobins q increases sharply due to the drop in equitytunneling risk. In contrast, Company B shows a drop inboth ROA and Tobins q due to higher cash-ow tunnelingand no change in equity tunneling risk.

2.6. Leverage and taxes

Our model assumes all-equity nancing and no taxes.We summarize here how adding leverage and taxes willchange the model predictions. Leverage will affect theextent of cash-ow tunneling, because leveraged rmshave committed some of their cash ow to debt repay-ment (Friedman, Johnson, and Mitton, 2003). Tax policiescan affect the rms overall cost of capital r, and can affectcash-ow tunneling by giving the government a stake inthe rms earnings and thus, an incentive to limit cash-ow tunneling (Dyck and Zingales, 2004).

Equity tunneling can affect the value of debt. Forexample, even a dilutive offering may inject some newequity capital into the rm, which, other things equal,should reduce default risk and increase debt value. Incontrast, a freezeout in which minority shares arepurchased with corporate cash (either directly, or throughthe controller withdrawing cash after taking the rmprivate) will remove equity capital and should decreasedebt value, other things equal. These second-order effectsaside, adding leverage and taxes will not signicantlyalter the equations above (one would replace earnings inthe model with earnings before interest and taxes (EBIT)).

3. Applying the theoretical framework toBulgariaempirical predictions

In this section we apply our theoretical framework toBulgaria both before and after legal changes enacted in2002, and develop testable predictions about the impactof these changes.

3.1. Preemptive and appraisal rights in Bulgarian law

Pre-2002 dilution. Prior to 2002, there were nominimum price rules for secondary equity issues, andthus no limits on ddilut, other than a rule requiring sharesto be issued for no less than par value. Shareholders hadpreemptive rights but notice periods were short, noticerequirements were often not enforced, and the rightswere not tradable. Moreover, freezeout risk was high, sominority shareholders would rationally not participate inequity offerings even at a large discount to intrinsic sharevalue. Before the 2002 reforms, dilutive offerings werecommon (see Table 4).

Pre-2002 freezeouts. Prior to 2002, freezeout prices hadto at least equal the rms three-month volume-weightedaverage stock price. However, the market was also veryilliquid. Controlling shareholders could freeze out minor-ity shareholders at very low prices using one or more ofseveral methods. First, they could reduce the market pricethrough a dilutive offering. Second, they could manipulatethe weighted average stock price by engaging in largetrades with related parties at low prices. Third, manyrms shares did not trade at all in a three-month period.Controllers could then simply delist the company withoutmaking any offer to minority shareholders. The delistedcompany would no longer be subject to the securities law,and would be governed only by the Bulgarian CommercialCode, which offers no protections against dilutive offers orfreezeouts, and does not require companies to providenancial statements to shareholders. A going darktransaction was likely to be equivalent to a freezeout ata close to zero price. Many going dark transactions werefollowed by dilutive offerings.

Legal reform. In the summer of 2001, Bulgaria elected anew government headed by the former Bulgarian king.One government priority was to improve capital markets.In December 2001, the government proposed, and theBulgarian Parliament soon thereafter adopted, severalchanges to the securities laws. Table 2 summarizes thesechanges. The changes became effective in June 2002, butin practice were enforced by the Bulgarian securitiesagency from the beginning of 2002.

The new rules strengthen preemptive rights by allow-ing a public company to issue shares only by distributingpreemptive rights to all shareholders, which are traded onthe Bulgarian Stock Exchange. With regard to freezeouts,the new law requires a tender offer for all minority shareswhen a shareholder reaches 50%, 67%, or 90% ownership.Delisting is allowed only after reaching 90% ownership.The tender offer price must be approved by a majority ofthe minority shareholders, and approved as fair tominority shareholders by the securities regulator. Theregulator computes a minimum fair price using acombination of discounted cash ow (DCF) valuationand comparison to peer rms. The freezeout price mustequal or exceed the greater of this fair price or the three-month weighted average share price.

The new government also reorganized and increasedthe powers of the securities regulator. The newlyestablished Financial Supervision Commission (FSC) suc-ceeded the Bulgarian Securities and Stock ExchangeCommission. The FSC has subsequently issued severalimportant regulations that strengthen enforcement andclarify minority shareholder protections, and has oftenrequired majority shareholders to increase freezeoutprices.

ARTICLE IN PRESS

ities l

t draft

came

out r

equity

lling

ares.

1. Warrants are required to be issued for any capital

increaseone warrant for each share.

holder

re pr

g wit

three

regu

prot

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173 161Table 2Changes in Bulgarian securities and tax laws in 2002.

This table summarizes the principal 2002 changes to Bulgarian secur

Parliament by the Bulgarian government on December 14, 2001. The rs

version of the law was accepted by Parliament on June 6, 2002, and be

Supervision Commission, rms were effectively subject to the new freeze

Statute Pre-2002

Preemptive rights Minority shareholders can participate in new

offerings. If they do not participate, the contro

shareholder can purchase all unsubscribed sh

Freezeout and

delisting rules

In a freezeout transaction, a controlling share

should offer at least the weighted-average sha

from the last three months of trading. Delistin

freezeout is allowed if no trading within last

months.

Tax changes Capital gains and interest income are taxed as

income up to 40% marginal tax rate. Corporate

taxed up to 40%.3.2. Empirical predictions

First, we test whether k (minority participation inequity issues) increases following the legal changes. Wecannot observe k directly. However, we observe thecontrollers pre-offering ownership (a0), post-offeringownership (a1), and the number of new shares issued(i), so we can estimate k as:

1 1a11a0 i1 i 1k: 11

The right-hand side of Eq. (11) provides a measureof dilution. It equals one if minority shareholdersare completely diluted (k=0); zero if they participate prorata (k=1); and is negative if the rm raises capitalprincipally from outside investors (k41). The term i/(1+ i)is a measure of offering size. We estimate (1k)empirically by regressing the left-hand side of Eq. (11)on i/(1+ i). We predict that (1k) declines (outsideshareholder participation increases) after the legalchanges.

Second, we test whether the legal changes affectfreezeout discounts. For going dark transactions, we canonly observe that these transactions are common pre-lawand are no longer possible post-law. We can, however,compare pre- versus post-law pricing in explicit free-zeouts. We predict that freezeout prices should increasepost-law.

Third, we test whether restrictions on equity tunnelinglead controllers to substitute from equity tunneling into2. Warrants are exercisable for a period of at least one

month, and are traded on the BSE.

ice

hout

1. A controlling shareholder must make a tender offer for all

remaining shareholders when reaching 50%, 67%, and 90%

ownership in the rm. A controlling shareholder can initiate

a freezeout only after reaching 90%.

2. Minority shareholders should receive a fair price for their

shares in tender offers and going-private transactions. A fair

price is computed using discounted cash ow and

comparable company multiples valuation methods, and is

compared to the average stock price for the last three

months, excluding block trades. Minority shareholders

should receive the higher of the two prices.

3. A majority of minority shareholders has to approve

going-private transactions.

4. The FSC has to evaluate the price in going-private

transactions and approve tender offers only if they meet the

fair value requirements.

lar

s are

1. Tax rate on interest income from bank deposits and

publicly traded bonds, and capital gains from publicly

traded securities is set to 0%.

2. Tax rate on corporate prots is reduced to 22.5%.aw and tax law. The changes to the securities law were introduced to

of the law was approved by Parliament on February 14, 2002. The nal

effective on June 21, 2002. However, due to the efforts of the Financial

ules from the beginning of 2002.

2002greater cash-ow tunneling. If so, then the protability ofhigh-equity-tunneling-risk rms should decline post-law,relative to lower-risk rms.

Fourth, we predict that Tobins q levels will increasepost-law for rms at high risk of equity tunneling, relativeto lower-risk rms. We do so both with and withoutcontrolling for changes in observed protability, mea-sured by ROA. A post-law increase in Tobins q for high-equity-tunneling-risk rms, controlling for changes inROA, offers evidence consistent with the limits on equitytunneling causing higher valuations of high-tunneling-risk rms, controlling for any change in cash-owtunneling that may result from loss of equity tunnelingopportunities. A post-law increase in Tobins q for high-equity-tunneling-risk rms, without controlling for ROA,offers evidence consistent with the limits on equitytunneling causing higher overall valuations of high-equity-tunneling-risk rms, even if controllers of theserms substitute from equity tunneling into greater cash-ow tunneling.

3.3. Empirical strategy: identifying a control group

A central empirical challenge for our pre- vs. post-lawcomparisons, is to identify an appropriate control group ofrms, which will let us separate the effects of a change inequity tunneling rules from other concurrent changes inthe economic or legal environment affecting Bulgarianrms (for example, the new government reduced the

ARTICLE IN PRESS

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173162capital gains tax, see Table 2). To study the change in thedilution parameter k, we have a partial control grouprms which delist prior to 2002. Because these rms areprivate, they are not affected by the 2002 law, so weexpect no post-law change in k for equity offerings. Thedramatic change we observe in k values, from a pre-lawmean of about zero to a post-law mean of about one, alsomakes it likely that the law change was a contributingfactor.

For change in freezeout prices, our control group isrms that do not undergo freezeouts. For ROA and Tobinsq, our principal strategy is to study changes in thesemetrics for rms at high risk of equity tunneling, usinglow-equity-tunneling-risk rms as a control group tocapture general changes in the business environment. Weneed to nd suitable proxies for equity tunneling risk. Ourprincipal approach is to use the pre-law tunneling eventsto estimate rm-level propensities to dilute and delist.We estimate logit models of dilution and freezeout usingall privatized rms. We use the logit coefcients toestimate dilution and freezeout propensities tunpropd,iand tunpropf,i for rms which remain public at the endof 2001.

We then ask whether these propensities predict post-law changes in ROA and Tobins q. We run the followingrm-xed effects regression models:

R1 : ROAit aAiBtD postlaw

Xj d;f

bj tunpropj; i postlaweit

R2 : qit aAiBtD postlawXj d;f

bj tunpropj; i postlawROAiteit :

12where i indexes companies and t indexes time; Ai arerm-xed effects which control for unobserved, time-invariant differences between rms; Bt are time-xedeffects, which control for country-wide changes (like thecapital gains tax reduction); postlaw is a post-law-changedummy variable (=one after the law change, zero before).The coefcient on postlaw gives the predicted change inROA or q for a hypothetical rm with zero tunnelingpropensities. The bd and bf predict the change in ROA or qfor a hypothetical zero-to-one change in the respectivetunneling propensity. We include ROA in the Tobins qregression if we want to control for changes in (observed)protability, and thus, control for changes in cash-owtunneling.

Our second approach is to form an explicit low-equity-tunneling-risk control group and use differences-in-differences (DiD) estimation (see Bertrand, Duo, andMullainathan, 2004). We form a control group in twoways. First, we treat rms with below-median propensityfor both dilution and freezeout as the control group, andother rms as the treatment group. Second, we predictthat rms with a controlling private owner will be morelikely to engage in equity tunneling, since this person cancontrol the rms decisions. We treat rms with no 20%private owner as the control group.We estimate the following DiD equation:

qit aAiBtD postlawb postlaw treatdummyeit ; 13

where treatdummy=one for treatment group rms, zerootherwise, and the notation is otherwise the same as inEq. (12). The coefcient b measures the after-minus-before change in Tobins q for treatment group rms,relative to control group rms.

4. Data and summary statistics

Testing our model requires rm-level stock ownership,accounting, delisting, and trading data. Each raises its ownchallenges in the Bulgarian context.

Overall sample: We obtain a list of all 1,040 Bulgariancompanies that participated in mass privatization andwere then listed on the Bulgarian Stock Exchange (BSE) in1998 from the Center for Mass Privatization. Atanasov(2005) and Miller and Petranov (2000) provide details onthe mass privatization process. Following the massprivatization auctions in 1998, more than 90% of all rmshad a 20%+blockholder.

Share ownership: We obtain year-end shares outstand-ing, ownership of the largest shareholder, ownership ofother 5% blockholders, and ownership by shareholdertype (government, private companies, individuals, andnancial institutions) for 19982003 for 1,021 of the1,040 mass privatization rms from the Bulgarian CentralDepository.

Delisting and nancial data: We obtain listing anddelisting information from the FSC. We exclude 190 rmswhich never traded or led nancial reports, and ninecompanies which were acquired or went bankrupt in1998 and have no data available. The remaining 822 rmsare required to le nancial data annually in spreadsheetform, which we obtain. We go through a series of steps(details available from the authors) to extract data itemsfrom the spreadsheets and correct obvious data entryerrors. In the end, we extract reasonably complete datafor 738 of the 822 rms, for at least one year from 1999to 2003.

Stock prices: We obtain trade-by-trade volume andprice data for 19992003 from the BSE. The market is thinwith a mean (median) of 15 (three) trades per rm peryear over this period, and some rms have zero trades.We hand-collect data on freezeout offers from news tapesprovided by the BSE. These are only available beginningin 2001.

Equity tunneling events: We trace equity tunnelingevents (dilution and delisting) over 19982003 for the822 rms with ownership and delisting data. We deneequity dilution as a rm issuing 20% or more new sharesin a single year. Of the 822 initial post-privatization rms,68 experience dilution only and 492 are delisted. Of thedelisted rms, 60 experience dilution in one year followedby freezeout in the same or a subsequent year (if dilutionand freezeout occur in the same year, we cannot tellwhich came rst). Altogether, 560 of the 822 rms (68%)experience an equity tunneling event. Firms that remainpublic also steadily transition away from having no

minority investors.5 Anecdotal evidence supports theregression results. We are unaware of a single genuineequity offering (intended to raise capital, rather thandilute minority shareholders) by a Bulgarian rm prior to2002. Genuine offerings begin to appear after the lawchange.

The last column of Table 4 reports results for a controlgroup of 79 rms which delist rst, and then engage indilutive offerings. We observe these offerings becausemost delisted rms continue to use the Bulgarian CentralDepository as a share registry after delisting. These 79rms have a high estimated value of (1k), and thus,negative estimated k. This indicates that in addition to thedilutive offering, the controllers buy additional sharesfrom minority shareholders. The law change shouldnotand indeed does notsignicantly affect the esti-mate of (1k), since these rms are not public and thus,not subject to the new law. This is consistent with the lawchange, rather than some unobserved factor, causing thepost-law decline in (1k) for public rms.

The degree of harm to minority shareholders fromequity offerings involves a combination of exclusion fromparticipation (1k), offering size, and the offering price.Most offerings are very large, the number of outstanding

ARTICLE IN PRESS

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173 163majority owner. This pattern is consistent with Bebchuks(1999) argument that dispersed ownership is unstable ifprivate benets of control are high. Table 3 shows theyear-by-year number of equity tunneling events.

State-owned rms: We exclude rms with majoritystate ownership in our main results but include theserms in robustness checks. There are 124 such rms atyear-end 1998, but only 13 at year-end 2001, largely dueto the state selling its controlling stake as part of anongoing privatization process.

5. Dilution and freezeout: pre- versuspost-law differences

In this section, we study how the 2002 legal changeaffect equity tunneling. Section 5.1 examines equityofferings. Section 5.2 examines freezeout offers.

5.1. Dilutive equity offerings

In Table 4, we estimate minority shareholderparticipation in equity offerings using Eq. (11). Weconstruct the dilution measure 1(1a1)/(1a0) basedon year-end ownership by the largest shareholder. Weconstruct the equity increase measure EquityInc= i/(1+ i)based on year-end shares outstanding: i=(sharestsharest1)/sharest1. We regress the dilution measure on

Table 3Equity tunneling events across time.

Yearly number of delistings (number of these involving a formal

freezeout tender offer) and dilutive equity issues, proxied by a 20% year-

over-year increase in outstanding shares (100% increase) for the 822

Bulgarian rms which participated in the mass privatization process and

have ownership and delisting data. The data for delistings end in 2002,

while the equity issue data include 2003. Formal freezeout offers are

offers announced on the BSE online news system.

Year Delistings (formal freezeout

offers)

Equity issues Z20%(Z100%)

1999 4 (n.a.) 72 (63)

2000 358 (n.a.) 36 (31)

2001 130 (10) 20 (15)

2002 45 (17) 9 (6)

2003 n.a. (4) 16 (9)EquityInc and an interaction term postlaw*EquityInc.We pool the pre- and post-2002 observations to reportan F-test for whether the pre- and post-2002 estimates of(1k) are statistically different.

The rst two columns of Table 4 report regressionresults for the 153 rms which make large offerings(dened as a 20% or greater year-over-year increase inshares outstanding). The results are consistent with thelaw change greatly reducing minority shareholder dilu-tion. Our pre-2002 estimate for (1k) is close to one.Thus, k is roughly zero and minority shareholders are(on average) completely excluded from new equity issues.After 2002, (1k) is close to zero, so kE1 and, on average,minority shareholders participate pro rata in equityofferings. This is consistent with the 2002 changes largelyeliminating dilutive offerings, and companies issuingshares primarily to raise capital, rather than to diluteshares more than doubles for 109 of the 128 pre-lawofferings. The mean (median) increase is 400% (177%). Welack data on offer prices for most offerings. However, weare able to nd a limited number of public announce-ments. For these, pre-law prices cluster at the minimumlawful price of one lev per share regardless of the marketprice of the shares.6

5.2. Delistings

About 60% (492/822) of the privatized rms withtrading data were delisted in the pre-law period. Most of

5 In robustness checks, we obtain similar results if we estimate

separate regressions for the pre-law and post-law samples or use cutoff

levels of a 50% or 100% increase in outstanding shares. We cannot

determine whether controllers increased their stake between t1 and tby purchasing shares on the open market or through the equity issuance.

This data limitation is unlikely to be signicant because average annual

share turnover for the rms in the dilution regressions is 5%, while the

average increase in number of shares is 400%. Still, as a robustness check,

we re-estimate the regressions with a conservative dilution measure

which assumes that all traded shares during the year were purchased by

the controlling shareholder. The pre-law estimate for (1k) becomes 0.6(signicant) and the post-law change is 0.85 (signicant), which isconsistent with the post-law change estimated in Table 5. Another

concern in interpreting Table 5 is whether different types of rms issue

equity pre-2002 and post-2002. In unreported robustness checks, we

address this concern by separating rms into three categories: rms that

did not issue equity, rms that issued equity pre-2002, and rms that

issued equity post-2002. We estimate multinomial logit models to

determine whether these three groups differ in percent owned by the

largest shareholder, percent owned by the state, and ln(market

capitalization), and do not nd signicant post-law differences in which

rms issue shares.6 For example, the shares of one of the most liquid companies on the

BSESofarma, AD traded at the beginning of December 2000 around 13

lev per share. During that month, Sofarma announced an equity issue

priced at one lev a share, which increased the number of shares

outstanding by 536%. We estimate k=0.67 for this issue; the controllers

stake increased from 66% to 77%.

ARTICLE IN PRESS

that

form

areho

equity

the B

share

* post

which

the p

n (fra

ring ye

s, are

1

**)***0)

)

***3)0

8)

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173164Table 4Dilution and equity issuance.

Beginning with a database that includes all 1,040 Bulgarian companies

Stock Exchange (BSE) in 1998, we examine equity offerings by listed and

outstanding during 19992003. The dependent variable is minority sh

fractional ownership by the largest owner at year-end after and before the

rms are rms which delist before the year of dilution, but continue to use

equity capital as EquityInc= i/(1+ i), where i=(shares1shares0)/shares1 andoffering. We regress minority shareholder dilution on EquityInc, EquityInc

and controls. The coefcient on EquityInc provides an estimate of (1k),offering) pre-2002. The coefcient on EquityInc * postlaw is an estimate of

(fractional controllers ownership prior to the offering), stateown fractio

(number of pre-offering shares times share price at the end of the pre-offe

are measured at year-end. t-Statistics, using Whites robust standard error

respectively. Signicant results, at 5% level or better, are in boldface.

Model

Estimate for (1k) pre-2002 1.132*(2.75

Change in (1k) post-2002 0.920(2.7

Resulting estimate for (1k) post-2002 0.211(0.18

Controls:

a0 1.143(3.0

Stateown fraction 0.49(0.4these rms (386 out of 492) do not trade during the threemonths before delisting date. They simply go dark, withno formal freezeout offer to minority shareholders. Theremaining 106 rms have at least one trade in the threemonths before delisting, and therefore, must make atender offer for minority shares at a price no lower thanthe weighted average trading price of the shares over thethree months before delisting.

At least 60 rms engage in dilutive offerings prior todelisting; the freezeout can then take place at a pricealready depressed by the dilutive offering. In Table 5,Panel A, we report seven representative examples ofdilution followed by delisting. In each, minorityshareholders either do not participate at all (k=0) orparticipate minimally.

Some rms employ large block trades at low priceswith related parties during the three-month measure-ment period, to reduce the freezeout price. In Table 5,Panel B, we report seven examples of apparent manip-ulative block trades (at least 50 times averagedaily trading volume) in the three months before delist-ing. These large trades represent an average of 89% oftrading volume during the three months precedingdelisting and are at an average discount of 81% to thevolume-weighted price of other trades during this period.The freezeout price will thus be at an average discount

Market capitalization

Intercept 0.081

(0.24)

No. of offerings (pre-2002) 128

No. of offerings (post-2002) 25

Prob. (from F-stat) 0.000

Adjusted R2 0.12participated in mass privatization and were then listed on the Bulgarian

erly listed Bulgarian rms that result in at least a 20% increase in shares

lder dilution, dened as (1(1a1)/(1a0)), where a1 and a0 are theissuance. Listed rms are publicly traded in the year of dilution. Delisted

ulgarian Central Depository as a share registry. We dene the increase in

s1 (shares0) is shares outstanding at the end of the year after (before) the

law dummy (=one for offerings during 2002 and 2003, zero otherwise),

is a measure of the exclusion of minority shareholders from an equity

ost-2002 change in (1k). The control variables in Models 1 and 2 are a0ctional state ownership prior to the offering), and market capitalization

ar, in millions of Bulgarian lev). Shareholder stakes and number of shares

in parentheses. *, **, And *** denote signicance at 10%, 5%, and 1% levels,

Listed rms Delisted rms

Model 2 Model 1

1.041* 2.355**(1.94) (2.49)

1.015** 0.503(2.57) (0.85)0.025 2.858***(0.00) (9.46)

1.551*** 4.474***(3.18) (4.86)0.486 0.042(0.82) (0.03)dfreeze of 0.81*0.89=72% to the market price excludingthese trades and an even larger discount to intrinsic value.Table 5 provides a rare glimpse of likely marketmanipulation in connection with a tender offer (Hermalinand Schwartz, 1996). We also see an important interac-tion: the effectiveness of a market price rule for freezeoutsdepends on legal control of dilution and of manipulativetrades.

Freezeout terms change dramatically after the 2002legal changes. We identify 31 tender offers with offerprices over 20012003 using BSE news tapes. Of these,two rms have no trades in the three months before theoffer. Table 6 provides summary data for these tenderoffers.

Pre-2002, offer premiums over the minimum allowedprice are small, and none of the ten initial offers arerevised upwards. After the law change, the mean (median)nal premium over the trade-volume-weighted averageprice for the last three months is 74% (42%). Thesepremiums compare favorably with the premiums forgoing-private transactions in the United States (e.g.,DeAngelo, DeAngelo, and Rice, 1984). Ten of the 21post-law offers are revised upwards to secure minorityshareholder and regulatory approval of the offer price. Thepost-law increases in mean and median premiums arestatistically signicant.

0.000

(0.90)

0.201 0.546

(0.49) (0.81)

101 33

21 46

0.000 0.000

0.13 0.24

ARTICLE IN PRESS

that

f com

The c

ar bef

q. (11

rading

de. The discount is computed as (weighted average price of other tradespricetrade=(size of trade)/sum(all trades during three months preceding delisting).

ice

p

1

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173 165Table 5Examples of equity tunneling.

Beginning with a database that includes all 1,040 Bulgarian companies

Stock Exchange (BSE) in 1998, this table shows representative examples o

rms that rst dilute minority shareholders and then delist prior to 2002.

the percentage ownership of the largest shareholder at the end of the ye

participation of minority shareholders in the new issue, computed using E

later year. Panel B includes companies with large trades (50+times daily t

The pre-trade price range is taken over the three months before the last tra

of block trade)/(weighted average price of other trades). Weight of large

Panel A: Examples of dilution followed by delisting

Dilution Delisting

Firm Name Year Year

Balkan 2000 2000

Galko 2000 2000

Agropolihim 2000 2001

Shvedski Kibrit 1999 2000

Himatech 2001 2001

Dunav-57 1999 2000

Soa Princess Hotel 1999 2001

Panel B: Examples of possible use of wash trades to reduce delisting pr

Firm name Trade date Delisting date

Plastimo 4 Apr 2000 5 Jul 2000

PreslavAH 19 May 2000 25 May 2000

Sintermat 46 Oct 2000 29 Nov 2000The higher post-law premiums understate the effect ofthe law change on freezeout outcomes. The premium isbased on the pre-offer share price. Pre-law prices weresometimes explicitly depressed by dilutive offerings ormanipulative block trades, and were also depressed by therisk of a future equity tunneling event. As we show below,valuation ratios for rms at high risk of equity tunnelingincrease post-law. Thus, post-law, the premium is higher,and the base price to which the premium applies is higheras well. Table 6 also reports the pre- and post-law mean(median) ratio of offer price/sales and offer price/bookvalue. These measures reect both the premium and thehigher base price. The mean (median) nal offer price/sales ratio quadruples, from 0.15 (0.07) to 0.65 (0.30),while mean (median) offer price/book doubles (quad-ruples). The mean rise in offer price/sales and the rise inboth medians are statistically signicant, despite smallsample size.

The higher post-law offer price/sales and offer price/book ratios largely reect higher offer prices relative totrading prices of shares, rather than higher trading prices.The nal two rows of Table 6 report mean and medianmarket-adjusted ratios. To compute the adjusted ratio,we divide offer price/sales (offer price/book) by the meanratio of trading price/sales (trading price/book) for non-frozen-out rms in the freezeout year. The after-minus-before changes in the mean-adjusted ratios are, in effect,

Ropotamo 9 Nov 2000 10 Jan 2001 1

Loviko Chirpan 15 Mar 2000 8 Jun 2000 1

Himatech 14 Aug 2001 4 Sep 2001 1

General Ganetzki 1 Oct 2001 15 Dec 2001

Average:Price of Discount of Weight of

Pre-trade large large large

rice range trade trade trade

5.008.00 1.00 0.857 0.663

2.683.51 1.05 0.610 0.945

1.9913.00 2.50 0.790 0.952i a0 (%) a1 (%) k

0.77 60.0 77.7 0.00

1.69 68.0 88.1 0.00

5.07 63.0 93.9 0.00

17.56 58.0 95.2 0.06

2.00 27.4 66.4 0.19

3.12 48.0 80.2 0.18

1.85 60.0 86.0 0.00participated in mass privatization and were then listed on the Bulgarian

panies listed on the BSE that engage in equity tunneling. Panel A includes

oefcient i equals the fractional increase in shares outstanding, a0 (a1) isore the dilution year (the dilution year), k is the estimated proportional

). We identify 60 rms which dilute in year t and delist in the same or a

volume) within three months before delisting, identied from BSE tapes.DiD estimates, where the control group is non-frozen-outrms. We will underestimate the effect of the law changeif the change results in higher share prices for the controlgroup. The increases in market-adjusted ratios are verysimilar, in size and statistical signicance, to the rawratios.

6. Firm characteristics and changes in protability andequity valuation

We next examine how the law change affects theprotability and valuations of high-equity-tunneling-riskrms, relative to low-equity-tunneling-risk rms. We rstestimate each rms propensity to engage in dilution ordelisting pre-law (Section 6.1). We then assess whetherthese tunneling propensities predict post-law change inROA (Section 6.2), and Tobins q (Section 6.3).

6.1. Estimates of equity tunneling propensities

We estimate the probability that a rm will engagein dilution or delist during 19992001, based on therms nancial and ownership characteristics. To reduceendogeneity issues, we use nancial data from 1999, therst post-privatization year for which data are available.

3.6620.70 1.12 0.943 0.971

0.0010.00 4.16 0.584 0.981

0.0010.00 1.10 0.890 0.762

4.004.00 0.12 0.970 0.938

0.806 0.887

ARTICLE IN PRESS

Table 6Freezeout tender offers before and after 2002 law changes.

Beginning with a database that includes all 1,040 Bulgarian companies that participated in mass privatization and were then listed on the Bulgarian

Stock Exchange (BSE) in 1998, this table shows premiums and offer/sales ratios for sample of 31 freezeout tender offers over 20012003, identied from

BSE news tapes. Market price is trade-volume-weighted average trading price for the three months before the announcement. Premium=(offer

pricemarket price)/(market price). Offer price/sales=(offer price)/(sales in year before tender offer). Offer price/book=(offer price)/(book value per sharedivide offer price/sales (offer price/book) for each freezeout offer by the mean

in the freezeout year. t-Statistics for means and z-statistics for medians are in

specti

2002

10

9

121

.85)

.001

.55)

0

121

.85)

.001

.55)

.150

.076)

.360

.123)

364

.203)

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173166at end of year before tender offer). To compute market-adjusted ratios, we

ratio of trading price/sales (trading price/book) for non-frozen-out rms

parentheses. *, **, And *** denote signicance at 10%, 5%, and 1% levels, re

signicant results (Po0.10) are indicated with italics.

Characteristics Before

Summary statistics of tender offers

Number of tender offers

Offers for rms with trading in 3 months before offer

Average premium to market price for initial tender offers 0.

(0

Median premium to market price for initial offers 0

(1

Number of offer prices that are revised upwards

Average premium for nal offers 0.

(0

Median premium for nal offers 0

(1

Raw offer price ratios

Mean (median) nal offer price/sales 0

(0

Mean (median) nal offer price/book 0

(0

Market-adjusted estimates

Mean (median) market-adjusted nal offer price/sales 0.

(0We measure ownership prior to the delisting (dilution)event, or at year-end 2001 if no delisting (dilution) occurs.

We estimate two propensity models, which differ inhow we measure ownership. In both, we use the followingnancial variables: ln(sales), leverage ((long-term debt+-short-term debt)/total assets), exports/sales in 1996 (thelast year with available data), operating margin, salesgrowth from 19981999, and a dummy variable forpositive net income. In Model 1 we control for ownershipusing the equity stake of the largest private blockholder,the difference between the ownership stake of the largestand the second-largest shareholder (see Laeven and Levine,2007), and dummy variables for presence of minority stateownership and a majority private shareholder.

In Model 2, we replace the ownership variables withthe Shapley values of the largest private owner and of thestate, which measure the likelihood that a shareholderwill be pivotal in a voting contest (Milnor and Shapley,1978; Shapiro and Shapley, 1978); and a Shapley wedgevariable as a measure of the difference between thelargest shareholders voting control and economic own-ership: Shapley wedge=Shapley value/ownership stake ofthe largest private blockholder.7 We compute Shapley

Mean (median) market-adjusted nal offer price/book 0.976

(0.338)

7 For example, if a shareholder has 50% of the shares, it has 100%

control, and will have Shapley wedge=(Shapley value=1)/(economic

ownership=0.5)=2. The Shapley wedge value will rise from one for a

small stake to two as ownership rises from zero to 50%, and will then

decline toward one as the controllers stake rises toward 100%. We usevely. Signicant results, at 5% level or better, are in boldface. Marginally

law After 2002 law Z-value (post-law - pre-law difference)

21

20

0.349*** 1.10(2.85)

0.166*** 1.42(3.31)10

0.740*** 1.84*(3.44)

0.416*** 2.17**(3.45)

0.649 2.01**(0.296) (2.80)***0.744 0.94

(0.487) (1.98)*

1.465 1.88*

(0.633) (2.80)***values using three blockholders and an ocean of othershareholders, where the third blockholder is the com-bined stake of all nancial institutions (we have data oncombined ownership by nancial institutions, but no dataon the stake of each institution). Laeven and Levine (2007)and Maury and Pajuste (2005) provide evidence on theinuence of other blockholders, not only the largestholder.

Table 7 shows marginal effects from logit regressionsof dilution and delisting propensities. In both models, thefactors that predict delisting are quite different from thosethat predict dilution. For example, if we focus on signrather than signicance, larger rms and more leveragedrms are less (more) likely to be delisted (diluted), whileprotable rms are more (less) likely to be delisted(diluted). We use the logit coefcients in Table 7 toestimate each rms propensity to dilute and delist(the parameters tunpropd and tunpropf in Eq. (12)).Delisting (dilution) propensity estimates using Model 1range from 0.04 to 0.98 (0.03 to 0.67). As suggested bytheir different loadings on rm characteristics, the

1.927 0.90

(1.292) (2.02**)

(footnote continued)

the Shapley wedge measure rather than the simple wedge variable

used in other research (dened as the difference between fractional

control rights and fractional economic rights) because Bulgaria has a

one-share, one-vote regime, and few controllers use pyramids or cross-

ownership, and thus, the traditional wedge is on average small, with

little variation across rms.

ARTICLE IN PRESS

that

on log

listed

holde

ake 1-

lockho

lders.

raction

lude i

5% lev

D

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173 167State minority stake dummy 0.192***(4.01)

Private control dummy 0.162***(2.68)

Stake of largest private shareholder 0.030(0.10)

Stake 1Stake 2 0.573***Table 7Pre-law propensity to delist or dilute.

Beginning with a database that includes all 1,040 Bulgarian companies

Stock Exchange (BSE) in 1998, this table shows marginal effects, based

Dependent variable in delisting (dilution) models equals one if a rm is de

otherwise. Private control dummy variable equals one if a non-state share

delisting (dilution), or at year-end 2001 if no delisting (dilution) occurs. St

second-largest blockholder. Shapley values are calculated using three b

combined stake of all nancial institutions) and an ocean of other shareho

growth is fractional sales growth from 1998 to 1999. Exports/sales is the f

Leverage=(Long-term+short-term debt)/Total assets. All regressions inc

signicance at 10%, 5%, and 1% levels, respectively. Signicant results, at

Logit regression of propensity to:

Model: (1)estimated delisting and dilution propensities correlatenegatively, at 0.65 (0.53) for Model 1 (Model 2). Thetwo models produce similar propensity estimatescorrelations between the estimates from the two modelsare 0.92 for delisting and 0.89 for dilution.

6.2. Equity tunneling propensities and cash-ow tunneling

The 2002 changes do not directly affect rm prot-ability or the availability of cash-ow tunneling. However,we hypothesize that the legal changes indirectly affectcash-ow tunneling, and therefore observed protability,because controllers who would have previously engagedin equity tunneling choose to substitute into greater cash-ow tunneling. If so, then observed protability shoulddecline for rms at high risk of equity tunneling, relativeto low-tunneling-risk rms. We test this hypothesis inTable 8. We use rm-xed effects regressions to estimateModel R1 of Eq. (12) and examine after-minus-before law

(2.59)Shapley value of largest private shareholder

Shapley value of state

Shapley wedge

Ln(sales) in 1999 0.113***(6.32)

Exports/sales in 1996 0.038(0.46)

Positive net income in 1999 dummy 0.147***(2.90)

Operating margin in 1999 0.335***(3.36)

Sales growth from 1998 to 1999 0.020

(0.51)

Leverage in 1999 0.113(0.62)

Pseudo R2 0.19

Firms undergoing delisting (dilution) 399

Firms in sample at start of period 647participated in mass privatization and were then listed on the Bulgarian

it regressions of propensities to delist or dilute minority shareholders.

from the BSE (issues more than 20% new equity) during 19992001, zero

r owns a majority stake; zero otherwise. Ownership is measured prior to

Stake 2 equals the difference between fractional stakes of the largest and

lders (largest and second-largest blockholder, third blockholder is the

Shapley wedge=Shapley value/ equity stake of largest blockholder. Sales

of sales outside of Bulgaria in 1996 (more recent data are not available).

ndustry dummies. t-Statistics are in parentheses. *, **, And *** denote

el or better, are in boldface.

elist Dilute

(2) (1) (2)

0.038(1.34)0.025(0.72)0.138

(0.70)

0.220changes in ROA, measured as EBITt/assetst1, winsorizedat 1% and 99%. The coefcients of principal interest are onthe interaction terms between the postlaw dummy andthe delisting and dilution propensities.

We report results separately for propensity Models 1and 2. For both models and both propensities, thecoefcients on the interactions between the propensitiesand the postlaw dummy are negative, statistically sig-nicant, and economically very large. For example, withpropensity Model 1 and full controls, a low-to-highchange in delisting (dilution) propensity predicts a 0.19(0.22) decline in ROA, compared to a sample mean of 0.04.A one standard deviation change in propensity to delist(dilute) predicts a 0.05 (0.05) decline in ROA. Our resultsare consistent with controllers of high-equity-tunneling-risk rms sharply increasing their cash-ow tunnelingafter the law changes restrict equity tunneling.

Higher cash-ow tunneling is, of course, only onereason why the observed protability of high-equity-tunneling-risk rms might decline, relative to low-risk

(1.63)

0.158 0.277***(1.49) (4.36)0.185 0.218***(1.23) (2.65)0.062 0.218***(0.64) (3.72)

0.122*** 0.007 0.012(7.12) (0.82) (1.34)0.102 0.053 0.028(1.28) (1.08) (0.58)0.125*** 0.033 0.028(2.58) (1.15) (0.98)

0.337*** 0.082 0.074(3.53) (1.17) (1.04)0.014 0.013 0.013(0.39) (0.58) (0.58)0.079 0.226*** 0.228***(0.45) (2.58) (2.60)

0.14 0.13 0.10

399 95 95

647 647 647

ARTICLE IN PRESS

that participated in mass privatization and were then listed on the Bulgarian

ns of yearly ROA, dened as EBIT(t)/Total Assets(t1) and winsorized at 1%/99%.

or after the law change. Postlaw dummy=one for 2002 and 2003, zero for 2000

1 and 2 from Table 7. For each model we run two specicationswithout and

hapley wedge are dened in Table 7. t-Statistics, using Whites robust standard

nd *** denote signicance at 10%, 5%, and 1% levels, respectively. Signicant

ts (po0.10) are indicated with italics.

ROA

(1)

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173168rms. We cannot rule out the possibility that some set ofrm characteristics is both associated with equity tunnel-ing risk and predicts a decline in protability after 2002,

Table 8Substitution between equity tunneling and cash-ow tunneling.

Beginning with a database that includes all 1,040 Bulgarian companies

Stock Exchange (BSE) in 1998, this table shows rm-xed effects regressio

We drop rms with no observation of the dependent variable either before

and 2001. Delisting and dilution propensities are estimated using Models

with controls for leverage and ownership. Leverage, Shapley values, and S

errors, are in parentheses. All regressions include year dummies. *, **, A

results, at 5% level or better, are in boldface. Marginally signicant resul

Dependent variable:

Propensities based on: Model

(1)

Postlaw*delisting propensity 0.173*(1.96)

Postlaw*dilution propensity 0.313**(2.15)

Leverage

Shapley private

Shapley state

Shapley wedge

Postlaw dummy 0.189***(2.78)

Mean (median) pre-law ratio

Number of rms (observations)for reasons unrelated to the law change. However, thepropensities to delist and dilute load very differently onrm characteristics. Thus, for example, if larger rmsbecome less protable after 2002, this would, other thingsequal, predict a negative (positive) coefcient on dilution(delisting) propensity in Table 8. If rms which were moreprotable pre-law declined in relative protability post-law, this would predict a positive (negative) coefcient ondilution (delisting) propensity.

In unreported robustness checks, we obtain similarresults if we: (i) use operating margin (EBIT/sales) as thedependent variable, instead of ROA; (ii) include state-controlled rms in the sample; (iii) limit the post-lawperiod to 2002; (iv) exclude rms which change owner-ship from majority-state ownership to non-majoritystate ownership, or from no majority private owner toa majority private owner; (v) take logs of all continuousvariables; and (vi) use the DiD approach discussed inSection 6.3. We also obtain consistent results with acombined measure of propensity to either delist ordilute.

In sum, we nd a sharp decline in post-law prot-ability for high-equity-tunneling-risk rms, relative tolow-risk rms, which is consistent with controllers ofthese rms engaging in greater cash-ow tunneling afterthe 2002 legal changes block the previously open road ofequity tunneling. The decline in protability for high-equity-tunneling-risk rms (whether or not due to highercash-ow tunneling) makes all the more striking theresults in the next section, where we nd that the Tobinsq of these rms rises signicantly post-law.

Model (2)

(2) (3) (4)

0.200** 0.213** 0.242**(2.08) (2.10) (2.21)0.346** 0.445*** 0.478***(2.27) (2.48) (2.61)0.148** 0.148**(2.17) (2.20)0.036 0.054(0.64) (0.97)0.050 0.048

(0.34) (0.32)

0.044 0.064

(0.89) (1.26)

0.215*** 0.227*** 0.255***(2.90) (2.96) (3.08)

0.04 (0.02)

156 (593)6.3. Equity tunneling propensities and rm valuation

We next explore the association between equitytunneling risk and after-minus-before law changes inrm valuation, proxied by Tobins q. We estimate Tobinsq on a quarterly basis, similar to Kaplan and Zingales(1997), as (book value of assets+market value of equitybook value of equity)/(book value of assets).8 We use thevolume-weighted average price for the quarter to mea-sure market value of equity.9

In Table 9, we estimate Model R2 of Eq. (12) using arm-xed effects specication, over 20002003, bothwith and without controls for protability, leverage, andownership. We exclude the fourth quarter of 2001 fromthe pre-law period, since the changes in securities lawwere then being debated in Parliament. The coefcients ofprincipal interest are the interaction terms between thepostlaw dummy and the delisting and dilution

8 An alternate denition of Tobins q, also commonly used in

governance research, is q=(market value of equity+book value of long-

term debt)/(book value of assets). This denition is not appropriate for

Bulgaria because long-term debt markets are virtually nonexistent pre-

2002, so for nancing rms rely on equity and short-term debt (often

accounts payable). Perhaps not coincidentally, since 2002, the Bulgarian

long-term debt market has developed rapidly.9 An event study is not feasible. As is often the case for legal reforms,

it is difcult to pinpoint an exact event date (Bhagat and Romano, 2002).

In addition, most Bulgarian rms trade once a month or less, so we

cannot reliably estimate abnormal returns over a limited time period.

ARTICLE IN PRESS

that participated in mass privatization and were then listed on the Bulgarian

ons of quarterly Tobins q, estimated as (book value of assets+market value of

e during 2002 and 2003, zero during 2000 and 2001. We drop observations for

ore or after the law change. Delisting and dilution propensities are estimated

ationswithout and with controls for ROA, leverage, and ownership. ROA is

d in Table 7. All regressions include quarter dummies. t-Statistics, using Whites

e at 10%, 5%, and 1% levels, respectively. Signicant results, at 5% level or better,

th italics.

Tobins q

l (1)

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173 169Table 9Tunneling propensities and changes in rm valuation.

Beginning with a database that includes all 1,040 Bulgarian companies

Stock Exchange (BSE) in 1998, this table shows rm-xed effects regressi

equitybook value of equity)/(book value of assets). Postlaw dummy=onthe fourth quarter of 2001, and drop rms with no observed q either bef

using Models 1 and 2 from Table 7. For each model we run two specic

dened in Table 8; leverage, Shapley values, and Shapley wedge are dene

robust standard errors, are in parentheses. *, **, And *** denote signicanc

are in boldface. Marginally signicant results (po0.10) are indicated wi

Dependent variable:

Propensities based on: Mode

(1)

Postlaw*delisting propensity 0.355***(2.65)

Postlaw*dilution propensity 0.643***(3.05)

ROA

Leverage

Shapley private

Shapley state

Shapley wedge

Postlaw dummy 0.234***(2.36)

Mean (median) pre-law ratiopropensities. Assuming that the post-law relative declinein the protability of high-equity-tunneling-risk rmsreects greater cash-ow tunneling, then regressionswhich omit the ROA control provide an estimate of theunconditional value effects of the law change, whileregressions with this control provide an estimate holdingcash-ow tunneling constant.

We nd positive, economically large, and statisticallysignicant coefcients on the interaction variables forboth propensity measures across all specications. Thecoefcients on the interaction variables are somewhatlarger if we control for protability, as expected given theresults in Table 8. To give a sense for economic magnitude,with propensity Model 1 and no controls, a low-to-highchange in propensity to delist (dilute) predicts an increaseof 0.33 (0.43) increase in Tobins q, while a one-standard-deviation increase predicts a 0.08 (0.10) increase. Theseare substantial compared to the pre-law mean Tobins q of0.51. The low pre-law mean, compared to US levels, isconsistent with equity tunneling risk suppressing pre-lawprices.

In unreported robustness checks, we obtain resultssimilar to those in Table 9 if we: (i) use price/sales, price/earnings or market/book as the dependent variable; (ii)include state-controlled rms in the sample; (iii) excluderms that change ownership from majority-state owner-ship to non-majority state ownership, or from no majorityprivate owner to a majority private owner; or (iv) use acombined measure of propensity to either delist or dilute.

Number of rms (observations)Model (2)

(2) (3) (4)

0.372*** 0.243* 0.271**(2.77) (1.84) (2.03)

0.702*** 0.610*** 0.691***(3.35) (2.48) (2.88)0.012 0.006

(0.09) (0.04)

0.031 0.084(0.20) (0.53)0.174*** 0.184***(2.69) (2.70)0.237** 0.203*(2.17) (1.84)0.055 0.033

(1.18) (0.68)

0.256*** 0.178* 0.204**(2.62) (1.79) (2.12)

0.51 (0.44)We also estimate a two-stage model, to address possibleendogeneity between Tobins q and ROA. In the rst stage,we regress Tobins q on ROA using pre-law data. In thesecond stage, we use this model to predict Tobins q acrossthe whole sample, and use the difference between actualand predicted q as the dependent variable in regressionsotherwise similar to Table 9. The results from the two-stage estimation are qualitatively similar to Table 9.10

As an additional robustness check on our results, weuse a difference-in-differences approach, in which wedivide rms into high- and low-equity-tunneling-riskgroups and measure post-law changes in the valuationof high-risk rms, relative to low-risk rms. We use twoalternative control groups. The rst uses ownership by thelargest holder to proxy for equity tunneling risk. Firmswith no 20% blockholder form the low-equity-tunneling-risk group. In contrast, if a rm has a large private

126 (1,084)

10 To further conrm robustness, we run a variety of additional

specications. First, instead of logit propensity regressions, we use linear

probability ordinary least squares or probit. Second, we estimate the

rm value regressions with Tobins q and propensities winsorized at

1%/99% or 5%/95%, and with both Tobins q and continuous independent

variables in logs. Third, we change the pre- and post-law estimation

windowsinclude 2001Q4 in the prelaw period, or omit 2002Q1 from

the post-law period. Fourth, we use operating margin instead of ROA to

control for protability or include a battery of control variables

(ln(sales), sales growth, and sales/assets; interactions between the

postlaw dummy and the control variables. Results from all variations are

consistent with those reported in Table 10.

ARTICLE IN PRESS

valua

that

regre

dum

ed q e

umm

r abov

p. Pro

nersh

nd qu

specti

ership

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173170Table 10Difference-in-differences regressions: after-minus-before changes in rm

Beginning with a database that includes all 1,040 Bulgarian companies

Stock Exchange (BSE) in 1998, this table shows difference-in-differences

value of equity book value of equity)/(book value of assets). Postlaw

observations for the fourth quarter of 2001, and drop rms with no observ

is rms with no 20% blockholder. Private50%+dummy (Private2050% d

blockholder). The propensity-based control group is rms that have eithe

other rms are included in the high tunneling propensity treatment grou

two specicationswithout and with controls for ROA, leverage, and ow

wedge are dened in Table 7. All regressions include rm-xed effects a

parentheses. *, **, And *** denote signicance at 10%, 5%, and 1% levels, re

signicant results (po0.10) are indicated with italics.

Dependent variable:

control group based on: Own

(1)

Postlaw*Private50%+ 0.766***(3.28)

Postlaw*Private2050% 0.769***(3.29)

Postlaw*High tunneling propensity

ROA

Leverageblockholder, this holder often has both the ability and theincentive to engage in equity tunneling; hence, theserms should have a high probability of tunneling. Wecreate two treatment groupsrms with a 50% privateblockholder, who clearly has the ability to cause tunnelingto occur, and rms with a 2050% blockholder, who willoften have this ability.

The second approach is based on the propensityanalysis. The treatment group is rms which haveabove-median propensity to either dilute or to delistusing propensity Model 1 from Table 7. The control groupis rms which are at below-median risk for both delistingand dilution.

Table 10 reports rm-xed effects DiD results for bothsets of control and treatment groups. The variables ofinterest are the interaction terms between the postlawdummy and the treatment group dummy variables, whichcapture the post-law change in Tobins q for these rms,relative to the control group. Tobins q rises signicantlypost-law for the ownership-based treatment groups (50%control and 2050% control), as well as for the propensity-based treatment group, both with and without controllingfor protability.

Last, we present visually the changes in q around the2002 legal changes for groups of rms at different risk of

Shapley private

Shapley state

Shapley wedge

Postlaw dummy 0.711***(3.03)

Number of rms 142

Number of observations 1,214tion.

participated in mass privatization and were then listed on the Bulgarian

ssions of quarterly Tobins q, estimated as (book value of assets+market

my=one during 2002 and 2003, zero during 2000 and 2001. We drop

ither before or after the law change. The ownership-based control group

y)=one for rms with a private majority owner (a 20 to 50% private

e-median propensity to dilute or above-median propensity to delist. All

pensities are calculated using Model 1 in Table 7. For each model we run

ip. ROA is measured as in Table 8; leverage, Shapley values, and Shapley

arter dummies. t-Statistics, using Whites robust standard errors, are in

vely. Signicant results, at 5% level or better, are in boldface. Marginally

Tobins q

Tunneling propensities

(2) (3) (4)

0.779***(3.34)

0.764***(3.29)

0.250*** 0.264***(3.55) (3.80)

0.039 0.007(0.40) (0.05)0.152 0.123(1.02) (0.78)tunneling. Sample size is limited, because we use abalanced panel of rms with non-missing Tobins q ratiosfor all eight quarters in the 20012002 period. Weseparate the sample into high-tunneling-risk rms versuslow-risk rms, using the same ownership and tunnelingpropensity approaches as in Table 10. For each group, wecompute mean Tobins q for each quarter in 20012002.We then compute percentage changes in these meanratios, using the rst quarter of 2001 as the base period.Fig. 1 shows the percentage change for high-equity-tunneling-risk rms, minus the percentage change forlow-equity-tunneling-risk rms. The mean Tobins q forhigh-equity-tunneling-risk rms jumps relative to low-risk rms in Q1 2002 by 40% with the ownership-basedcontrol group, and 80% with the propensity-based controlgroup. Thus, the rise in valuation ratios for high-equity-tunneling-risk rms is concentrated in the quarter whenthe legal reforms become effective.

A non-causal story which explains our results is hardto come by. One would have to posit that one set of rm-level characteristics is correlated with delisting risk whileanother quite different set is correlated with dilution risk(the two sets load quite differently on factors such as size,protability, and ownership in propensity logit regres-sions). Both sets of rm characteristics have to predict

0.169*** 0.147**(2.82) (2.31)0.105 0.210**(1.21) (2.08)0.010 0.064(0.19) (1.36)

0.713*** 0.168* 0.177**(3.07) (2.19) (2.39)

126

1,084

ARTICLE IN PRESS

Q4Qu

opensi

s rela

ation a

ns q,

isk r

risk

uity-t

V. Atanasov et al. / Journal of Financial Economics 96 (2010) 155173 171Fig. 1. Percentage change in Tobins q for high-equity-tunneling-risk rmincludes all 1,040 Bulgarian companies that participated in mass privatiz

gure shows quarterly time series of the percentage change in mean Tobi

tunneling risk minus the percentage change in mean q for low-tunneling-r

for all eight quarters of 20012002. For solid line, low-equity-tunneling-

dilution, estimated using logit Model 1 of Table 7. Other rms are high-eq

with no majority owner. Other rms are high-equity-tunneling-risk.-