Essays on financial intermediaries

The rstessay studies a causal eect of institutional ownership IO on bank loanpricing, using the Russell Index 1000 inclusion/exclusion as the disconti-nuity design setting.. Chapter 1How

ESSAYS ON FINANCIAL INTERMEDIARIES

RUICHANG LU

NATIONAL UNIVERSITY OF SINGAPORE

2014

ESSAYS ON FINANCIAL

INTERMEDIARIES

RUICHANG LU BBA, Shanghai University of Finance and

Economics, 2006 M.Econ., Shanghai University of Finance and

Economics, 2008

A THESIS SUBMITTED

FOR THE DEGREE OF DOCTOR OF

PHILOSOPHY DEPARTMENT OF FINANCE

NATIONAL UNIVERSITY OF SINGAPORE

2014

Foremost, I would like to express my sincere gratitude to my advisor,

Dr Anand Srinivasan, for the continuous support of my Ph.D study andresearch, for his patience, motivation, enthusiasm, and immense knowledge.His guidance helped me in all the time of research and writing of this thesis.Besides my advisor, I would like to thank my thesis committee members,

Dr David Reeb and Dr Nagpurnanand Prabhala, for their encouragement,insightful comments, and hard questions

My sincere thanks also go to many of my seniors in the Ph.D program,

Dr Shen Jianfeng, Dr Li Yan and Dr Lin Chunmei, etc, who have oeredkind help to me in various occasions I also thank my fellow Ph.D class-mates, including Cheng Si, Hu Rong, Wang Tao, whose companionshipsmade the Ph.D journey much more delightful

I would like to thank National University of Singapore, Humanities andSocial Sciences Research Fund R-315-000-104-646, for nancial support forthe second chapter of this thesis

Last but not the least, I would also like to thank my family for thesupport they provided me through my entire life and in particular, I mustacknowledge, especially my wife, Zhang Xiaojun, and my son, Lu Shangyin,without whose love and encouragement, I would not have nished thisthesis

May 30, 2014

1 How Does Institutional Ownership Aect Bank Loan Pric-ing: Evidence from a Regression Discontinuity Design 1

1.1 Introduction 1

1.2 Data and Empirical Strategy 6

1.3 Empirical results 10

1.3.1 Discontinuity Test for Institutional Ownership 10

1.3.2 Discontinuity Test for Bank Loan Contract Terms 12

1.4 Potential channel 15

1.4.1 Liquidity increase 15

1.4.2 Monitoring eort by institutional investors 16

1.5 Robustness Check 17

1.5.1 Public oat adjustment by Russell 17

1.5.2 Manipulation 18

1.5.3 Nonparametric form, bandwidth choice, and placebo tests 19

1.6 Conclusion 19

2 Subjective or Objective? Nonparametric Estimation Of Misreporting and Mis-assessment in Corporate Credit Rat-ing 35 2.1 Introduction 35

2.2 A closed-form identication and estimation 40

2.2.1 Model setup 40

2.2.2 Assumptions and identication 40

2.3 Data and sample description 52

2.4 Empirical result 54

2.4.1 Misreporting behavior in overall sample 54

2.4.2 Misreporting behavior in overall sample 54

2.4.3 Misreporting behavior across industries 55

2.4.4 Misreporting behavior across business cycle 57

2.5 Robustness checks 58

2.5.1 Evaluation of the assumptions 58

2.5.2 Exclusion of recent nancial crisis period 59

2.6 Conclusion 60

Appendices 70 1 Simulation for testing the consistency of estimator 73

This thesis consists of two essays on nancial intermediaries The rstessay studies a causal eect of institutional ownership (IO) on bank loanpricing, using the Russell Index 1000 inclusion/exclusion as the disconti-nuity design setting Specically, I nd that an exogenous positive shock

in institutional ownership appears to only aect the pricing term of bankloans but not the non-pricing terms On average, a 35 % increase in IO willlead to a 29 bps lower loan spread which is about 1/5 of the average spread.However, the non-pricing terms such as collateral, maturity, and covenants

do not change with the increase in IO The reduction in loan spread is ported by the evidence that rms with high IO will have lower credit riskmeasured by expected default frequency using Merton model Also, thiseect is weaker for the family rms Further investigation reveals that in-crease in liquidity and direct monitoring from institutional investors could

sup-be the channels through which institutional ownership aects bank loanspricing Moreover, although the cost of bank loan is lower for rms withhigher institutional ownership, these rms do not borrow more frequentlythan those with lower institutional ownership

The second essay investigates the misreporting and mis-assessment ofcorporate credit ratings by credit rating agencies (CRAs) We distinguishbetween "mis-assessment", which is the noise from the unobservable truerating to the rating perceived by CRAs (the internal rating), and "misre-porting", which is the dierence between perceived and reported rating byCRAs Using a sample of corporate credit ratings during 1986-2011, we

nd that the mis-assessment in credit rating is very small and statisticallyinsignicant Also, there is a U-shaped relationship between true credit rat-ing and misreporting probability Specically, CRAs misreport the creditratings for high-grade rms with a probability of 3%, for middle-grade rmswith a probability of 0, and for low-grade rms with a probability of 6%.Second, the misreporting behavior of CRAs dier signicantly across theindustries The nancial industry has the highest misreporting probabil-ity (35% in the lowest-grade rms) and the largest misreporting magnitude(rating grade jump between true and reported grade) The energy industryhas the lowest misreporting probability Last, when economic conditionsare bad, the credit rating agent is more likely to deate the rating

Keywords : Institutional Ownership, Bank Loan Pricing,

Re-gression Discontinuity Design, Credit Rating, reporting, Mis-assessment, Nonparametric

Mis-List of Tables

1.1 Summary statistics for rms around 1000th rank 29

1.2 Discontinuity test for bank loan contract terms 30

1.3 Regression results using switchers 31

1.4 Cross section results 32

1.5 Potential channels 33

1.6 Robustness check 34

2.1 Yearly distribution of the ratings 64

2.2 One year transition probability matrix for credit ratings entire sample 65

2.3 Misreporting probability for each rating group  entire sample 66 2.4 Misreporting behavior dierence across industries 67

2.5 Misreporting behavior across business cycle 68

2.6 Misreporting probability for nancial rms (exclude current nancial crisis) 69

List of Figures

1.1 Stocks around the Russell 1000 Inclusion Threshold 24

1.2 Discontinuity Test for Institutional Ownership Around Rus-sell 1000 Inclusion Threshold 25

1.3 Discontinuity Test for Dierent Legal Type of Institutional Ownership 26

1.4 Discontinuity Test for Bank Loan Contract Terms 27

1.5 Density Test for Bank Loan Borrowing 28

2.1 Model setup 63

Chapter 1

How Does Institutional

Ownership Aect Bank Loan

Pricing: Evidence from a

Regression Discontinuity Design

1.1 Introduction

The syndicated loan market has become the most important source ofglobal corporate nancing over the past 20 years In year 2009, the size ofinternational syndicated loan market reached a record high of $1.8 trillion,which is even larger than the international bond markets with a size of

$1.5 trillion (Chui, Domanski, Kugler, and Shek, 2010) Therefore, it is

of immense economic signicance to understand the factors that aect thecost of bank loan given the size of the syndicate loan market

In this paper, I focus on how institutional ownership (IO hereafter) ofstocks of a borrowing rm inuences its bank loan pricing Institutionalinvestors own a signicant proportion of public equity in the US stock mar-ket The institutional equity ownership increases dramatically in the past

decades, from 10% in the 1970s to more than 60% nowadays This matic change in institutional ownership is believed to have great impact onthe corporate governance structure of company As suggested in Shleiferand Vishny (1997), corporate governance is an important channel throughwhich suppliers of capital to corporation assure themselves of getting a re-turn on their investment While the impact of governance on cost of debthas received recent attention in several papers (Bhojraj and Sengupta,2003; Klock, Mansi, and Maxwell, 2005; Anderson, Mansi, and Reeb, 2003;Cremers, Nair, and Wei, 2007; Chava, Livdan, and Purnanandam, 2009;Roberts and Yuan, 2010), few papers have examined the impact of insti-tutional ownership on bank loan contract terms This paper studies thistopic, specically focusing on the causal eect of institutional ownership

dra-on bank loan cdra-ontract terms

Theoretically, the net impact of institutional ownership on ers is unclear On one hand, the involvement of institutional investors inmonitoring has the potential to reduce agency problems, which in turn willincrease shareholders' value and benet debtholders Institutional investorsmay also discipline managers through shareholder activism or the threat

debthold-of exit (Gillan and Starks, 2007; Edmans, 2009; Admati and Peiderer,2009; Edmans and Manso, 2011) Also, ownership by institutions may re-duce coordination costs (Grossman and Hart, 1980; Shleifer and Vishny,1986) and can lower agency costs through economies of scale in delegatedmonitoring On the other hand, debtholders' concern of asset substitutionmight be heightened with higher IO (Jensen and Meckling, 1976) Strongershareholder control better aligns management and shareholders, which maylead to wealth transfer between shareholders and debtholders (e.g throughmore dividend payout) Therefore, it is ultimately an empirical issue to testthe eect of IO on bank loan pricing

Empirically, it is a challenging task to establish the causal relation

be-tween IO and bank loan pricing While institutional ownership may causedierences in bank loan pricing, the institutional investors may also choosestocks because of some unobservable rm characteristics that drive thebank loan pricing as well.

My empirical strategy to test the causal eect of institutional ownership

on bank loan pricing utilizes a regression discontinuity design around theRussell 1000 and Russell 2000 index cut-o.1 Specically, all the eligible se-curities are ranked based on their market capitalization on the last tradingday in May each year The rst 1000 largest stocks will be included in theRussell 1000 Index and stocks with rank from 1001th to 3000th will be in-cluded in Russell 2000 index The breakpoint of Russell Index 1000/2000 isthe rank of 1000th Therefore, mechanically, those just-included stocks andjust-excluded stocks in Russell 1000 Index are very similar in terms of themarket capitalization However, since both Russell 1000 and 2000 index aremarket value weighted index, the stocks just-included and just-excluded inRussell 1000 will get quite dierent weight in the index respectively Inter-estingly, stocks with smaller market capitalization will be included at thetop of Russell 2000 index and have a large weight, because they are com-pared to other smaller stocks in Russell 2000 index In contrast, the stocks

at the bottom of Russell 1000 index will have a small weight, because theyare compared to other large stocks in Russell 1000 index Combining withthe fact that Russell 2000 index is much more popular in the mutual fundindustry, there is a signicant jump in institutional ownership at the cutopoint (i.e., 1000th rank) Therefore, I can employ a regression discontinuityapproach to investigate the impact of the jump in institutional ownership

on bank loan pricing To the extent that the exclusion restriction is valid,

I can investigate how other variables of interest such a loan spread,

matu-1 Chang and Hong (2012) are the rst to exploit this discontinuity and nd that the smaller rms that are just included in the more popular Russell 2000 index experience higher returns right after the reconstitution of the index, which the authors attribute to price pressure due to higher institutional demand for the Russell 2000 stocks.

rity, collateral, and covenants behave around the cut-o point (i.e., 1000thrank) I then can make causal inferences and calculate how the dependentvariables of interest respond to a given shock in institutional ownership.2

My main nding is that the exogenous positive shock in institutionalownership of a stock caused by the index inclusion/exclusion aects thepricing of bank loans in a signicant manner On average, a 35% increase

in IO will lead to 29 bps lower loan spread which is about 1/5 of theaverage spread However, non-pricing terms such as collateral, maturity,and covenants do not change with the increase in IO This evidence issupported by the fact that rms with higher IO have a lower credit riskwhich is measured by expected default frequency using Merton model

I also examine additional channels by which an increase in IO mayaect bank loan pricing One potential channel is an increase in liquidity

I nd that the liquidity, using Amihud measure, is 10% higher for the rmsjust-excluded in the Russell 1000 An increase in liquidity may have twoeects on credit risk One eect is that it can facilitate the exercise ofcorporate control because it allows large shareholders to emerge to correctmanagerial failure (Maug, 1998) It may also increase the liquidity fading

by these institutional investors and discipline the managers by threat toexit" or treat of governance" (Edmans, 2009) Therefore, the liquidityimprovement will add value to the rm and this benet is shared withdebtholders The other eect is that increase in liquidity may also lowerthe expected return of the rms and further directly lower the credit risk

of rms given that the rm's fundamentals stay the same

Another potential channel through which IO may aect bank loan ing is the actual monitoring eort by institutional investors The proxy-voting participation for just-excluded rms is higher by 45 percentagepoints than just-included rms Therefore, rms with higher IO do have

pric-2 More detail will be discussed in section 1.2.

higher participation rate than the rms with lower IO Another supportingevidence is that there are jumps in holding for most types of institutionalinvestors such as public pension funds, bank trust Large institutionalshareholders (notably CalPERs and other public pension funds) are knownfor their involvement in governance-related activities Therefore, increase

in public pension funds' holding may mitigate the agency problem Takentogether, these evidences demonstrate that rms with higher institutionalownership could be monitored better by institutional investors The benetfrom better alignment between the manager and shareholders may spillover

to the debtholders

I further investigate the cross-sectional dierence of the eect of IO onbank loan pricing across family and non-family rms I nd that the resultsare weaker for the family rms, suggesting that the benets of additionalmonitoring are lower when a controlling shareholder is already present inthe rm

I also nd that, although the cost of bank loan is lower for the rms excluded from Russell 1000, they do not borrow more frequently relative

just-to the just-included rms

Last, I conduct several robustness tests to address the concern of domness and to check whether our results are sensitive to methodologicalchoices My results are robust to alternative methodologies such as dier-ent nonparametric form or bandwidth choice I also do the placebo tests bypicking 600th, 800th, 1200th, and 1400th ranks as the cuto points Theresults show that there is no signicant eect at these random thresholds,which suggests that the main results in this paper are not picking up arandom pattern in the sample

ran-Focusing on how institutional ownership causally inuences the bankloan pricing, this paper contributes to the emerging literature on the role ofcorporate governance on cost of debt Chava et al (2009) nd that lenders

demand a premium from borrowers with shareholder-friendly managers.While their paper focuses on the eect of takeover risk as the mechanism

of corporate governance on cost of bank loan, my paper focuses on theinstitutional ownership as the mechanism of corporate governance Thispaper also complements the results in Cremers et al (2007) and Bhojrajand Sengupta (2003) While both papers focus on the bond return, thispaper studies the causal eect of institutional ownership on bank loan.Loan market is dierent from bond market in the following aspects First,bondholders do not have incentive to monitor the rms due to the free-rider problem In contrast, banks are generally regarded as the insider ofthe rms Therefore, although higher institutional ownership would lead

to lower bond yield, it is unclear whether the increase in shareholder itoring will lower the cost of bank loan given that banks already exerttheir own monitoring eort Second, bank loans are informationally moreecient than publicly traded bonds, because they are priced by the expe-rienced loan ocers with in-depth knowledge of the rms (e.g (Altman,Gande, and Saunders, 2010)) Therefore, bank loan market could be a bet-ter setting to investigate the eect of institutional ownership on the cost ofdebt

mon-The rest of paper is organized as follow: Section 1.2 discusses the dataand empirical strategies Section 1.3 shows the main empirical results.Section 1.4 explores the channels through which the IO aects bank loanpricing Section 1.5 presents the robustness checks Section 1.6 concludes

1.2 Data and Empirical Strategy

Russell U.S index captures 99% of the U.S equity market and 100%

of the investable U.S market The indexed stocks need to be traded on amajor U.S exchange, with its headquarter in U.S or asset primarily in US,

or revenues from US The membership is then determined by the marketcapitalization at the last trading day in each May Common stock, non-restricted exchangeable shares and partnership units/membership interests(in certain cases) are used to calculate a company's total market capital-ization If multiple share classes of common stock exist, they are combinedtogether In cases where the common stock share classes are independentlyfrom each other (e.g., tracking stocks), each class is considered for inclusionseparately.

One important characteristic of the Russell indices is that these indicesare transparent and easy for managers to construct by themselves, in con-trast to the black box approach of the S&P 500 index This transparencyhas resulted in its popularity among a signicant fraction of mutual fundmanagers During annual reconstitution, the closing price on the last trad-ing day in May on the primary exchange is used to determine market capi-talization If a security does not trade on its primary exchange, the lowestprice from another major US exchange is used In the case where multi-ple share classes exist, a primary trading vehicle is determined, and theprice of that primary trading vehicle" (usually the most liquid) is used inthe calculation The impact of rebalance of Russell index is huge given itspopularity For example, according to Nasdaq, approximately $687.9 mil-lion shares representing $9.5 billion in value were traded in the closing 1.15seconds on last trading day of June across the nearly 2,200 Nasdaq-listedstocks in 2012

I obtain the annual constituents list for the Russell 1000 and Russell

2000 from Russell Investments for the sample period of 1990 to 2006 Thesample period stops at 2006 because after that Russell Company imposes a

exible band policy Specically, rms may stay in the prior year index if itsmarket value is close to the cut-o point market capitalization Therefore, Ionly use the sample before 2007 in order to obtain a clean setting However,

my result is qualitative the same if I include the observations after year2006.

The validity of regression discontinuity design relies on the ness of the index membership assignment around the cuto points Inthis setting, the just-included and just-excluded from Russell 1000 index

random-is random, which leads to a jump in institutional ownership According

to Russell U.S Equity Indexes Construction and Methodology" 3 , all theeligible securities are ranked by their total market capitalization on thelast trading day in May each year The largest 1000 stocks are included

in Russell 1000 Index and the 1001th to 3000th largest stocks are included

in the Russell 2000 Index The breakpoint of Russell Index 1000/2000

is the rank of 1000th Therefore, mechanically, those just-included andjust-excluded stocks in Russell 1000 Index are very similar in terms of themarket capitalization and so the assignment to the left or right of the indexthreshold is essentially random Stocks with smaller market capitalizationwill be included at the top of Russell 2000 index and have large weights,because they are compared to other smaller stocks in Russell 2000 index

In contrast, the stocks at the bottom of Russell 1000 index will have smallweights, because they are compared to other large stocks in Russell 1000index Figure 1.1(a) demonstrates the continuity in market capitalizationand Figure 1.1(b) demonstrates the discontinuity in weight Combiningwith the fact that Russell 2000 index is much more popular in the mutualfund industry, there is a signicant jump in institutional ownership at thecuto point (1000th rank)

Therefore, I can employ a regression discontinuity approach to gate the impact of the jump in institutional ownership on bank loan pric-ing To the extent that the exclusion restriction is valid, I can investigatehow other variables of interest such a loan spread, maturity, collateral,

investi-3 More detail can be found at http://www.russell.com/Indexes/

and covenants behave around the cut-o point (i.e., 1000th rank) I thencan make causal inferences and calculate how the dependent variables ofinterest respond to a given shock in institutional ownership.

Indeed, I show that the just-excluded stocks in Russell 1000 indexhave discontinuously higher institutional ownership compared to the just-included stocks Discontinuity plots with some data smoothing and breaktests proposed by Lee and Lemieux (2010) are shown in Figure 1.2(a).The plots of institutional ownership after the reconstitution month of Juneshow the dramatic discontinuity The dierence in institutional ownership,

a proxy for demand by institutions between the just-included versus thejust-excluded stocks is around 35% The mean institutional ownership per-centage in the sample is 60% So the dierence is about one-half of thismean, which is a sizable increase This nding veries the premise of theexperiment that there is a signicant dierence in demand for stocks at thebottom of the Russell 1000 and for stocks at the top of Russell 2000 index

The empirical strategy I employ here is to t the linear function forthe stocks around the cut-o point Specically, I run the linear regressionaround the cuto rank 1000th:

Y = α + β1∗ R2000 + β2∗ |Rank| + β3∗ |Rank| ∗ R2000 + Y eari+ , (1.1)

where for |Rank| < Bandwidth, R2000 is a dummy variable that equals

1 if the stocks are in Russell 2000 index, Rank is the relative rank fromthe 1000th rank with negative denoting that stocks in Russell 1000 indexand positive number denoting stocks in Russell 2000 index, and Y eari arethe year dummies Bandwidth is the number of rms in each side of cutopoints Year dummies are included in all regressions

I choose the bandwidth Bandwidth = 100 for most of my tests tially, the choice of bandwidth faces a trade-o between testing power and

Essen-accuracy Larger bandwidth will have higher testing power but lower curacy for including observations that have low predicting power I choose

ac-100 as the bandwidth for the following reasons: rst, the choice of width has great impact on the estimation result Rule-of-thumb (ROT)bandwidths for dierent interest variables are about 80-200 for each side

band-I conservatively use the same bandwidth for all variables of interest as ourbaseline results and put more results on dierent choice of bandwidths inthe robustness check Second, about a half of rms in the (-100, 100) band-width have loans information in our sample Therefore, 100 rms on eachside of the cuto point is reasonable to get enough testing power; In therobustness checks, I also try the Rule-of-thumb (ROT) bandwidth and try

to t the function using local polynomial with couples of variation Theresults are qualitatively the same See section 1.5 for detail discussion

1.3 Empirical results

In this section, I report the empirical results of discontinuity tests in IOand loan contract terms

1.3.1 Discontinuity Test for Institutional Ownership

In this subsection, I test whether there is a discontinuity in institutionalownership

As suggested in Lee and Lemieux (2010), I plot the discontinuity ininstitutional ownership with some data smoothing Results are shown inFigure 1.2 In Figure 1.2(a), I plot average institutional ownership (in 10rank bins" for smoothness) relative to the Russell 1000/2000 threshold.The X-axis represents the distance from the Russell 1000/2000 thresholdwhere 0 represents the smallest rms in the Russell 1000, negative numbersrepresent larger rms away from the last Russell 1000 rank while positive

numbers represent smaller rms just away from the rst Russell 2000 indexrank.

I further decompose the IO into groups based on their expected ment horizon using the classication method developed by Bushee (1998)

invest-I classies institutions into three groups (i.e., dedicated, quasi-indexer, andtransient), based on their past investment patterns in the areas of portfo-lio turnover, diversication, and momentum trading Transient" investorshave high portfolio turnover and highly diversied portfolio holding Ded-icated" investors have large average investments in portfolio rms and ex-tremely low turnover Quasi-indexers" investors have diversied holdingsand low turnover Figure 1.2(b) to Figure 1.2(d) show that both the tran-sient IO and Quasi-indexer" IO have signicant jumps for just-includedRussell 2000 stocks The jump is about 10% for transient IO and 25%for the quasi-indexer IO Meanwhile, there is no signicant jump for thededicated IO This suggests that the dierence in demand between the just-excluded and just-included stocks mainly come from the indexer which isobvious because they benchmark to the index and the active traders

I also decompose the IO based on the types of institutional investorsfrom CDA/ Spectrum database, following Bushee (1998) I combine theCDA type 3 (investment company) and type 4 (independent investment ad-visor) into one group In addition, I dig deeper to distinguish the ESOPS,university and foundations endowments, and private/public pension funds.Earlier research on shareholder activism (e.g., Guercio and Hawkins (1999))shows that public pension funds pursue a highly active role in the gover-nance of companies principally through the submission of shareholder pro-posals Figure 1.3(a) to Figure 1.3(f) show that there is a signicant jumpfor public pension fund holding for just-excluded Russell 1000 stocks Thejump in institutional ownership is from around 0.8% to 2.8% I also observejumps for investment company holding and bank trust holding However,

I do not observe any signicant jumps for corporate pension fund holding,insurance company holding, or university and foundation fund holding.Overall, I observe a signicant jump in IO at cuto point and thisjump is concentrated in quasi-indexer (3/4 of overall jump) and transientinvestors (1/4 of overall jump) but not dedicated investors I also nd asignicant jump in public pension fund holding, bank trust holding andinvestment company holding.

1.3.2 Discontinuity Test for Bank Loan Contract Terms

In this subsection, I test whether there is any discontinuity in bank loancontract terms I investigate the bank loans borrowed by the rms within 1year after index membership assignment (i.e., from July to next year June).Since the membership assignment is mechanical and creates the exogenousshock to the institutional demand, the dierence of loan contracts for thejust-excluded and just-included stocks could be attributed to the shock ofthe institutional ownership This enables me to identify the causal eect

of institutional ownership on bank loan contract Loan spread is the drawn spread Collateral is a dummy variable that equals one if the loanfacility has collateral requirement Maturity is the length of loan lending inmonth Covenants is the number of distinct nancial and general covenants

all-in-I rst plot the mean loan spread, maturity, collateral, and covenants acrossall years over 10 rank intervals for 100 bins to the left of the threshold andfor 200 bins to the right of the threshold The X-axis represents the distancefrom the Russell 1000/2000 threshold where 0 represents the smallest rms

in the Russell 1000, negative numbers represent larger rms away fromthe last Russell 1000 rank while positive numbers represent smaller rmsjust away from the rst Russell 2000 index rank The graph shows a cleardiscontinuity in loan spread at the threshold but no signicant discontinuityfor collateral, maturity, or covenants

Table 1.2 reports the formal discontinuity tests for IO and bank loancontract terms using Equation 1.1 I use the OLS for IO, loan spread, andmaturity regression, Logit regression for collateral, and Poisson regressionfor covenant Columns 1 reports the dierence of IO is about 35% at thecuto point and is statistically signicant at the 1% level Column 2 showsthat the dierence in loan spread at the discontinuity is equal to 29 bps and

is statistically signicant at the 1% level Combining with the fact that theaverage spread for loans around the cut-o point is 150bp, this represents

a 20% reduction in spread, which is economically signicant Columns3-5 show the result for collateral, maturity, and covenants None of theabove variables are statistically signicant Combining together, the resultsuggests that the institutional ownership only aects the pricing term ofloan contract but not the non-pricing terms Column 5 shows the resultfor discontinuity test for credit risk Essentially, the loan pricing reectsthe credit risk of the rms Following Merton distance to default model,

I use the expect default frequency (EDF) as my measure of credit risk.For each rm, I calculate the monthly average EDF during year T July toyear T+1 May I nd that rms just-included in Russell 2000 have loweraverage EDF and it is statistically signicant at 1% level This evidencesupports the previous nding that rms just included in Russell 2000 willenjoy lower loan spread compared to rms just excluded in Russell 2000

As the main robustness test, I only use the sample of rms switchingbetween Russell 1000 and Russell 2000 Only the switching year and theyear before switching will be included in the sample For example, rm Awas in Russell 1000 during 1990-1995, and switched to Russell 2000 during1996-2006 I only use loans issued to rm A in year 1995 and 1996 for

my analysis Also, I require that the switcher must stay in the (-100, 100)band both before and after the switching This is a conservative sample toasure there is no signicant change in rm fundamentals when rms switch

between indices I compare the loan contract terms for the same rms inthese 2 years Specically, I run the rm xed-eect regression for thissubsample and control for the year xed eect.

Y = α + β1 ∗ R2000 + Y eari+ F irmi+ , (1.2)

where R2000 is a dummy variable that equals 1 if the stocks are in Russell

2000 index, F irmi are the rm dummies, and Y eari are the year dummies

I use the OLS for spread and maturity regression, Logit regression forcollateral, and Poisson regression for covenant The result is robust to theincludsion/excludsion of the year dummies, loan purpose dummies (e.g.working capital/general purpose, etc.), and/or loan type dummies (e.g.term loan/credit line etc.) Overall, there are 220 switchers (189 unique

rms) switching within the (-100, 100) widows in our sample 98 unique

rms get loans, among which 56 unique rms get loans both before andafter the switch Table 1.3 shows the regression results First ve columnsreport the OLS regression results for loan spread, collateral, maturity, andcovenants respectively Spread reduction is about 40 bps for the rmsswitched from Russell 1000 to Russell 2000 Compared to the regressiondiscontinuity design test, the estimated sign is the same and the magnitude

is larger (29 bps for RDD) for loan spread regression This could be due tosmall sample of the switchers Moreover, the switch does not aect othercontract terms (collateral, maturity, and covenants) and none of them isstatistically signicant This is consistent with the results in RDD test.Next, I investigate the cross-sectional dierence of the eect of IO onloan pricing Table 1.4 reports this set of tests First column investigateswhether this eect is stronger for the rms close to distress I sort rmsbased on distress risk measured by the expected default frequency (EDF)

I divide the sample into high and low distress risk sub-sample using themedian of EDF at the end of May across all years I nd that rms with

high distress risk enjoy lower loan spread after getting an exogenous shock

in IO

In second column of Table 1.4, I investigate whether the family rmswould benet less from the increased institutional ownership Family rmsare regarded as having incentive structures that result in fewer agencyconicts between equity and debt claimants (Anderson et al 2003) I ndthat the eect is weaker for the family rms, suggesting that the benets ofadditional monitoring are lower when a controlling shareholder is alreadypresent in the rm

Last, I test whether rms with higher IO are more likely to borrowfrom banks since they enjoy lower cost of debt I nd that the likelihood

of borrowing is quite similar The result is reported in Figure 1.5 and thistest is suggested in McCrary (2008) There is no evidence that the rmswith higher IO will borrow more frequently

Taken together, these results point to a causal eect of institutionalownership on the pricing of bank loans but no eect on non-pricing terms

of loan Firms with higher IO will have lower cost of bank loan This issupported by the nding that rms with high institutional ownership willhave lower credit risk In the next section, I explore the channels throughwhich institutional investors aect pricing of bank loan

owner-two eects on credit risk One eect is that it can facilitate the exercise ofcorporate control because it allows large shareholders to emerge to correctmanagerial failure (Maug, 1998) It may also increase the liquidity fading

by these institutional investors and discipline the managers by threat toexit" or treat of governance" (Edmans, 2009) Therefore, the liquidityimprovement will add value to the rm and this benet is shared withdebtholders The other eect is that increase in liquidity may also lowerthe expected return of the rms and further directly lower the credit risk

of rms given that the rm's fundamentals stay the same

Generally, more involving institutional investors will increase the uidity of the stocks and enable more and faster information incorporatedinto the stock price This is supported by the analysis of subtypes of insti-tutional ownership, I nd that only the transient institutional shareholdersand quasi-indexers (relaxing short sale constraint) will increase the hold-ing in Russell 2000, but not the dedicated investors Using the RDD, I

liq-nd that the liquidity is 10 percentage higher for the rms that are justincluded in the Russell 2000 The result is presented in Table 1.5 column

1 This result is consistent with the literature suggesting that institutionalinvestors increase the stock liquidities

1.4.2 Monitoring eort by institutional investors

In this subsection, I test whether, besides the theoretical argument,

rms with higher institutional ownership will be monitored more by stitutional investors Generally, the predictions stem from the idea thatmonitoring by institutional investors may have spillover eect from share-holders to debtholders Therefore, higher institutional ownership rms willenjoy the lower loan spread

in-Following Crane, Michenaud, and Weston (2012), I collect data from ISSRisk Metrics Shareholders Proposal and Vote Results database I measure

proxy-voting participation at the rm level in the scal year following theindex inclusion The results are presented in Table 1.5 column 2 Theproxy-voting participation for just-excluded rms is higher by 45 percentagepoints than just-included rms Therefore, rms with higher IO do havehigher participation rate than the rms with lower IO.

I admit that the proxy-voting participation rate is not a perfect measurefor the institutional investors' monitor eort, for institutional investorscould outsourcing shareholder voting to proxy advisory rms such as ISSand Glass Lewis Therefore, the increase in the voting rate is increasingmechanically In this sense, the monitoring work is fullled by the proxyadvisory rms but, at least, the higher institutional ownership make thecoordination to vote for/against the proposal easier Moreover, the increase

in the voting rate is higher than the increase in IO, which suggests thatthe expanded voting pool includes not only new institutional investors butalso the pre-existing institutional investors

1.5 Robustness Check

In this section, I conduct several robustness tests to address the concern

of randomness and to check whether our results are sensitive to ological choices

method-1.5.1 Public oat adjustment by Russell

After the membership of the stocks are determined, the actual indexweights are adjusted by Russell company based on the investable shares.The investible shares data are considered proprietary by Russell and notavailable to the public This adjustment may reduce the randomness of themembership assignment For example, large market capitalization rmswith small investiable shares would stay at the bottom of Russell 1000

These rms are not comparable to the rms on the top of Russell 2000with similar investiable market capitalization but much smaller marketcapitalization To eliminate the rms with large adjustment made by Rus-sell, following Crane et al (2012), I calculate the percent dierence betweenthe unadjusted weight using CRSP market capitalization and the adjustedweight reported by Russell I drop observations in the top 5% of squaredpercent dierence In this way, I remove the stocks that have large weightadjustment from the sample The results are qualitatively the same for thesample of excluding the 5% of observations with large adjustments.

1.5.2 Manipulation

Another concern with this design is that some rms may have tives to manipulate their index membership for the expected reduction in

incen-nancing cost Such manipulation would lead to self-selection and aect

my causal inferences I argue that this is unlikely for the following tworeasons Firstly, since the smaller rms will be included in Russell 2000index and enjoy the reduction in spread, rms need to short sell stocks

to push down the stock price However, the stock price will go up afterthe rms are included in Russell 2000 and the short position will suer

a loss, which may reduce the incentive to manipulate the index inclusion.Secondly, the ranking is only decided by the closing market capitalization

at the last trading day in May Since dierence in size for rms aroundthe threshold would be small, it is dicult to precisely control their rank-ing relative to other rms in the dynamic trading market Therefore, it isunlikely that rms could self-select on one side of the threshold Even inthe presence of manipulation, Lee (2008) formally shows that discontinuitydesign is still valid as long as rms do not have precise control over theirassignment I further test the manipulation using density test suggested inMcCrary (2008) If the rms expect the benet of lower cost of loans and

self-select into the Russell 2000 index, we should observe that rms at thetop of Russell 2000 index will borrow more loans The result of density test

is presented in gure 1.5 I do not nd any signicant dierence in terms

of borrowing frequency between rms at the bottom of Russell 1000 indexand those at the top of Russell 2000 index

1.5.3 Nonparametric form, bandwidth choice, and

placebo tests

In this section, I test whether the results using linear function formare robust to dierent nonparametric form or bandwidth choice Overall,results are qualitatively the same and suggest that IO only aects the loanspread but not the non-pricing contract terms Results are shown in Table1.6 Panel A presents the results using local polynomial specication and

a third-degree polynomial with an Epanechnikov Kernel with a Rule ofThumb (ROT) bandwidth suggested in Fan and Gijbels (1996) We alsotest the results using 50% and 200% of the ROT bandwidth

To demonstrate the signicance of the threshold of 1000th rank in thisRDD, in the falsication tests, the same estimation technique is applied

to the 600th, 800th, 1200th, and 1400th ranks as thresholds Results arereported in Panel B of Table 1.6 These results demonstrate that there is nosignicant eect at these random thresholds, which suggests that the mainresults in this paper are not picking up a random pattern in the sample

1.6 Conclusion

Over the past two decades, the syndicated loan market has become themost important source of global corporate nancing Factors that inuencethe cost of bank loan are therefore of immense economic signicance

In this paper, I explore an exogenous discontinuity in institutional

own-ership and investigate the causal impact of institutional ownown-ership on thebank loan pricing I nd that higher institutional ownership causes a de-crease in loan spread but not in other non-pricing contract terms such ascollateral, maturity, or covenants Moreover, this impact is weaker for fam-ily rms Further investigation reveals two potential channels of the casualimpact First, the benet from monitoring eort by institutional investors

is shared with debt holders Second, the liquidity increase due to moreinstitutional investors involvement leads to the improvement in the corpo-rate governance because the threat to monitor/exit" more reliable, and/orlower credit risk Overall, this paper suggests that institutional ownershiphas a large causal impact on the cost of debt and the random inclusion in

a stock market index could have a signicant impact on the cost of bankloan nancing

Admati, Anat R, and Paul Peiderer, 2009, The "wall street walk" andshareholder activism: Exit as a form of voice, The Review of FinancialStudies 22, 24452485

Altman, Edward I, Amar Gande, and Anthony Saunders, 2010, Bank debtversus bond debt: Evidence from secondary market prices, Journal ofMoney, Credit and Banking 42, 755767

Anderson, Ronald C, Sattar A Mansi, and David M Reeb, 2003, ing family ownership and the agency cost of debt, Journal of FinancialEconomics 68, 263285

Found-Bhojraj, Sanjeev, and Partha Sengupta, 2003, Eect of corporate nance on bond ratings and yields: The role of institutional investors andoutside directors, The Journal of Business 76, 455475

gover-Bushee, Brian J, 1998, Investors on myopic r&d investment behavior, TheAccounting Eeview

Chang, Yen-Cheng, and Harrison Hong, 2012, Rules and regression tinuities in asset markets, Working Paper

discon-Chava, Sudheer, Dmitry Livdan, and Amiyatosh Purnanandam, 2009, Doshareholder rights aect the cost of bank loans?, Review of FinancialStudies 22, 29733004

Chui, Michael KF, Dietrich Domanski, Peter Kugler, and Jimmy Shek,

2010, The collapse of international bank nance during the crisis: dence from syndicated loan markets, BIS Quarterly Review

Evi-Crane, Alan, Sebastien Michenaud, and James Weston, 2012, Large holders as monitors: Is there a trade o between liquidity and control?,Working Paper

share-Cremers, KJ Martijn, Vinay B Nair, and Chenyang Wei, 2007, Governancemechanisms and bond prices, The Review of Financial Studies 20, 1359.Edmans, Alex, 2009, Blockholder trading, market eciency, and manage-rial myopia, The Journal of Finance 64.

Edmans, Alex, and Gustavo Manso, 2011, Governance through trading andintervention: A theory of multiple blockholders, The Review of FinancialStudies 24

Fan, J, and I Gijbels, 1996, Local polynomial modelling and its tions, Chapman, Hall, London

applica-Gillan, Stuart L, and Laura T Starks, 2007, The evolution of shareholderactivism in the united states, Journal of Applied Corporate Finance 19,5573

Grossman, Sanford J, and Oliver D Hart, 1980, Takeover bids, the free-riderproblem, and the theory of the corporation, Bell Journal of Economics

11, 42

Guercio, Diane Del, and Jennifer Hawkins, 1999, The motivation and pact of pension fund activism, Journal of Financial Economics 52, 293340

im-Jensen, Michael C, and William H Meckling, 1976, Theory of the rm:Managerial behavior, agency costs and ownership structure, Journal of

nancial economics 3, 305360

Klock, Mark S, Sattar A Mansi, and William F Maxwell, 2005, Does porate governance matter to bondholders?, Journal of Financial andQuantitative Analysis 40, 693719

cor-Lee, David S, 2008, Randomized experiments from non-random selection

in us house elections, Journal of Econometrics 142, 675697

Lee, David S, and Thomas Lemieux, 2010, Regression discontinuity designs

in economics, Journal of Economic Literature 48, 281355

Maug, Ernst, 1998, Large shareholders as monitors: Is there a trade obetween liquidity and control?, The Journal of Finance 53, 6598.McCrary, Justin, 2008, Manipulation of the running variable in the regres-sion discontinuity design: A density test, Journal of Econometrics 142,698714

Roberts, Gordon, and Lianzeng Edward Yuan, 2010, Does institutionalownership aect the cost of bank borrowing?, Journal of Economics andBusiness 62, 604626

Shleifer, Andrei, and Robert W Vishny, 1986, Large shareholders and porate control, The Journal of Political Economy 461488

cor-Shleifer, Andrei, and Robert W Vishny, 1997, A survey of corporate ernance, The Journal of Finance 52, 737783

gov-(a) Market Cap (b) Weight

Figure 1.1: Stocks around the Russell 1000 Inclusion Threshold

This gure shows the average stock market capitalization and index weights forthe rms around the Russell 1000 inclusion threshold at the end of June Firmsare assigned to the Russell 1000 or 2000 based on the market capitalization of

rms at the end of each May Index weights are determined by using a oatadjusted market capitalization within each index at the end of each June

(a) Institutional Ownership (b) Institutional OwnershipTransient

(c) Institutional OwnershipDedicated (d) Institutional OwnershipQuasi-indexer

Figure 1.2: Discontinuity Test for Institutional Ownership Around Russell

1000 Inclusion Threshold

This gure plots the dierent types of institutional ownership based on the ing pattern(e.g transient, dedicated, and quasi-indexer) against Russell sizerankings at the end of June across all years The X axis represents the relativedistance from Russell 1000 inclusion threshold, with 0 represents the last rm inRussell 1000 Each dot represents the average IO over 10 ranks

trad-(a) Public Pension Fund (b) Bank Trust (c) Investment Company

(d) Insuance Company (e) Corporate Pension Fund (f) University and Foundation

(a) Loan Spread (b) Collateral (c) Maturity

(d) Financial Covenants (e) General Covenants (f) Total Covenants

Figure 1.4: Discontinuity Test for Bank Loan Contract Terms

This gure plots the dierent contract terms of bank loans against Russell sizerankings at the end of June across all years The X axis represents the relativedistance from Russell 1000 inclusion threshold, with 0 represents the last rm inRussell 1000 Each dot represents the average contract terms over 10 ranks

Figure 1.5: Density Test for Bank Loan Borrowing

This gure plots the frequency of bank loan borrowing against Russell size ings at the end of June across all years The X axis represents the relativedistance from Russell 1000 inclusion threshold, with 0 represents the last rm inRussell 1000