Agents with and without principles

L

-3i

DEWEY

z)

Massachusetts

Institute

of

Technology

Department

of

Economics

Working Paper

Series

AGENTS

WITH

AND

WITHOUT

PRINCIPALS

Marianne

Bertrand,

UChicago

&

NBER

Sendhii Mullainathan,

MIT

&

NBER

Working Paper

00-31

January

2000

Room

E52-251

50

Memorial

Drive

Cambridge,

MA

02142

This

paper can

be downloaded

without charge from the

Social

Science Research

Network

Paper

Collection at

http://www.ssrn.com

MASSACHUSETfsliSTITTJTr OF

TECHNOLOGY

NOV

8

₂₀₀₀

Agents with

and

without

Principals

Marianne

Bertrand

Sendhil

MuUainathan

*

January

14,

2000

Who

sets

CEO

pay?

Our

standardanswer to this question has been shaped by principal agent theory:

shareholdersset

CEO

pay.

They

use pay to limit the moral hazard problem caused by the low ownership

states of

CEOs.

Through

bonuses, options, or long termcontracts, shareholders can motivate the

CEO

to

maximize

firm wealth. In other words, shareholders use

pay

toprovideincentives, a view

we

referto as the

contracting view.

An

alternative view,

championed

by practitioners such as Crystal (1991), argues that

CEOs

set their

own

pay.

They

manipulate the compensationcommittee

and

hencethepayprocess itselfto

pay

themselves

what

they can.

The

only constraints they face

may

bethe availabilityoffunds or

more

general fears, such

as not wantingtobe singled outin the Wall Street Journalas beingoverpaid.

We

refer to this second view

as the

skimming

view. Inthis paper,

we

investigatethe relevance of thesetwoviews.

I

The

Effect

of

Takeover Threats

on

CEO

Pay

We

begin with

some

illustrativefindings from an earlier paper (Bertrand

and

MuUainathan, 1999a). In the

mid

1980s, several

US

states passed legislation

making

hostile takeovers

more

difficult. These laws very

likelyreduced the

power

ofan important disciplining

mechanism,

the threat ofbeingtaken overin caseof 'Bertrand: Department ofEconomics, PrincetonUniversity,

CEPR

and

NBER;

MuUainathan: Department ofEconomics, Massachusetts Institute ofTechnologyand

NBER.

e-mail: mbertran@princeton.eduand mullain@mit.edu. Address: Marianne Bertrand, A-17-J-1, Industrial Relations Section, Firestone Library, Princeton University, Princeton, NJ 08540, US.A.;Sendhil MuUainathan, E52-380a, Department of Economics, MIT, 50 Memorial Drive Cambridge,

MA

02142, USA.

We

thank our discussant, George Baker,formany helpfulcomments.

poor

management.

How

do

we

expect

CEO

pay to respondtothis change in thelegal environment? Inthe

contractingmodel, whereshareholdersset

CEO

pay,the

main

effect ofthe anti-takeover laws shouldbe on

payforperformance. Shareholders, seeing the weakeningofone discipliningmechanism, should respond by

strengthening another, pay for performance. In the

skimming

model, on the other hand, the

main

effect

should be on

mean

pay.

CEOs

facing areduced threat ofa hostile takeover can

now

skim

more

resources

fromtheir firm.

InBertrand

and

Mullainathan (1999a),using panel data on about 600firmsbetween 1984

and

1991,

we

studythe impact ofthe legislativechanges on the level of

CEO

pay

and

its sensitivity toperformance.

We

focus on the adoption by states of Business Combination Statutes. These statutes impose a

moratorium

period (3 to 5 years) on specified transactions between the target and araider holdinga certain threshold

percentageofstock unlessthe boardvotes otherwise.

They

were adopted by several states at differenttimes

throughthe 1980s

and

early 1990s.

The

staggering ofthe laws overtime allowsus to identify theeffect of

thelaws after controllingforyear

and

firmfixed effects.

Forthefull sample,

we

found

an

increasein

mean

pay and

an increaseinpayforperformance (especially

foraccountingmeasures ofperformance). Theseresults are intriguingbecause they aie consistentwith both

views of

CEO

pay.

Mean

pay rises as the

skimming model would

havepredicted but

pay

forperformance

also rises asthe contracting

model

would

havepredicted.

To

help resolvethisambiguity,

we

look

more

closelyatwhichfirmsexperienced the increasesin

mean

pay

and

in payfor performance.

We

focus

on

how

firms with different corporategovernance are affected bythe

new

laws.

More

specifically,

we

separate the firmsinour sampleintotwo groups based on whetherthe firmhas

a large shareholder present or not. Largeshareholders are very oftenthought to bean effectivegovernance

mechanism and

are an easy

way

to

measure

corporate governance given the available data (Shleifer and

Vishny, 1986).

We

definealargeshareholderasan

owner

who

has a block ofat leastfivepercent of

common

shares in thesample base year. Blocks that are

owned

by the

CEO

are, ofcourse,excluded.

Table 1 reports our findings for

mean

pay.

The

dependent Vciriable is the logarithm of total

CEO

Digitized

by

the

Internet

Archive

in

2011

with

funding

from

Boston

Library

Consortium

IVIember

Libraries

Combination Statute, year fixed effects, firm fixed effects, a quadratic in

CEO

age, a quadratic in

CEO

tenure, the logarithmoftotal assets

and

thelogarithm oftotalemployment.

Column

(1) focuses on firmswith a largeshareholder. In thesefirms,

we

see astatistically insignificcint

increasein

mean

payofonly

2%

followingthelaws.

Column

(2) focusesonfirmswithout alargeshareholder.

For thesefirms, in contrast,

CEO

pay grew by (this time a statistically significant) 7.5%.

The

increasein

mean

pay was actually quite heterogeneous. While firms without a large shareholder experienced a large

increase, firms withalarge shareholderexperienced almost

no

increase.

In Table 2,

we

similarly break apart the pay for performanceresults. Here

we

find the oppositeeffect.

The

increaseinthesensitivityofpaytoaccountingperformanceisconcentrated

among

the firms with alarge

shareholder. Firms without one

show no

increasein payforperformance.

We

have just

shown

that in response to the passage of anti-takeover legislation, firms with a large

shareholder increased

pay

for performance, while firms without a large shareholder increased

mean

pay.

Thissuggest that the

two

models of

CEO

pay neednotbe contrasted. Instead, they

may

both betrue

and

indeed

may

be quitecomplementary. This intuitionis reinforced inthe additional tests thatfollow.

II

Further

Evidence

II.l

Are

CEOs

Rewarded

for

Luck?

InBertrand

and

Mullainathan {1999b),

we

find

two

further pieces of evidence. In thefirst test,

we

examine

whether

CEOs

are rewarded for observable luck.

By

luck

we

mean

changes in firm performance that are

beyond

the

CEO's

control. Insimple agencymodels,pay should notrespond tolucksinceby definition the

CEO

cannotinfluence luck. Tying

pay

to luck doesn't provide betterincentives (the

CEO

can'tchangeluck),

but merely addsrisk to the contract (Holmstrom, 1979).

Under

the

skimming

view, onthe other hand,pay

will becorrelated with lucksincethe

CEO

can use luckydollarsto pay herselfmore.

To

empirically

examine

the responsiveness ofpay to luck,

we

use three different measuresofluck. First,

performance on a regular basis. Second,

we

use changes in industry-specific exchange ratefor firms in the

traded goodssector. Third,

we

use year-to-yeardifferences in

mean

industry performanceto proxyforthe

overall economic fortunes of a sector. For all three measures,

we

find that

CEO

pay responds to luck. In fact

we

find that, for allthree luck measures,

CEO

pay is as sensitiveto a "lucky dollar" as toa "genered

dollar."

Most

importantly,

we

find that

CEO

pay

responds less to luckin the better governed firms. Similarly

toour takeoverresults,

we

find that thepresence ofalargeshareholder reduces the

amount

ofpayfor luck. Qualitativelyequivalent resultsholdfor othergovernancemeasures such as thelevel of

CEO

entrenchment

(measured as

CEO

tenure interacted with the presence of a large shareholder)

and

board size. Again,

improved governance leads to greater concordance with the contracting view, while

weakened

governance

leadstogreaterconcordance with the

skimming

view.

II.2

Are

Stock

Options

Grants

Gifts?

The

secondtest presentedin Bertrand

and

Mullainathan (1999b) focuses on the granting of stockoptions.

Contract theory predictsthat

when

stock options are granted, other

components

ofpay should be adjusted

down

so that

CEOs

areleftindifferent betweenthepay packagecontaining options

and

theonecontaining

no

options. Supportersofthe

skimming

view, ontheotherhand,

would

highlightthe factthat stock optionsdo

notappear on balancesheets. Becauseofaccountingrules,firmsdonot chargetheirearningsforthe options they grant.

CEOs

can therefore pay themselves through option grants without affecting the company's

bottom

line. Ifshareholdersmainlylookat this

bottom

line,options grants areaneasy

way

toskim without

drawing

unwanted

attention. Thus, the

CEO

who

gives herselfoptions would not need to lower the other

components

of pay at all. Thus, while contracting predicts a charge foroptions,

skimming

predicts little

charge.

In the empirical test,

we

focus

on

the question of

how

the strength ofgovernanceinthe firm affects the chargeforoptions. Usingthe

same

governance measuresasinthe previoustest,

we

findthatpoorlygoverned

on the board, the

CEO

is charged less for each dollar's worth ofoptions granted. Again,

we

find greater

resemblanceto

skimming

in thepoorly governedfirms.

Ill

An

Independent

Test

The

abovefindingspointtowardsthe coexistence of

skimming and

agencymodels. Inthis section,

we

provide

anotherindependenttest of thisidea

by

revisitingthe evidencein

Aggarwal

and Samwick

(1999). Aggarwal

and

Samwick's paperstartswiththe followingimportantprediction ofthe contractingmodel: thesensitivity

ofpay toperformance should decrease as the riskiness orvariance ofperformanceincreases. In support of

thatprediction,

Aggarwal and

Samwick

find thatthesensitivity ofpaytoperformanceislarger infirmswith

lessvolatile stockprices.

In the context of

what

we

have

shown

before, a natural question arises:

Does

the tradeoflf between

performance volatility

and

pay-performance sensitivity appear strongerin the better governed firms?

We

testthishypothesis using a

CEO

compensation datathat covers792diflferentcorporations over the 1984-1991

period, provided to us by

David

Yermack.

Compensation

data

was

collected fi-om the corporations'

SEC

Proxy, 10-K,

and 8-K

filings. Other data

was

transcribed from the Forbesmagazine annual survey of

CEO

compensation as well as from

SEC

Registration statements, firms'

Annual

Reports, direct correspondence

with firms, press reports of

CEO

hires

and

departures,

and

stock prices published by Standard

&

Poor's.

Firms wereselected intothe sample onthebasis of their Forbes rankings. Forbesmagazinepublishesannual

rankingsofthe top500 firms on four dimensions: sales, profits, assets

and

market value.

To

qualify for the

sample acorporation

must

appearin oneof these Forbes 500rankings at leastfourtimes between 1984

and

1991. In addition, the corporation

must

have been publicly traded forfour consecutive years between 1984

and

1991.

While

this data set covers a smaller set ofcompanies than the

Execucomp

database (used by

Aggarwal

and

Samwick), it does contain

some

informationon the structure ofcorporate ownership, which

isnot available in

Execucomp.

compensa-tion

and

the value of optionsgrantedinthatyear. Itisa

measure

offlow compensation. Unlike

Execucomp,

Yermack's data doesnot containinformationonthe value of stock options

and

equityshares heldby

CEOs.

In practice,

we

use the logarithm of total compensation.

We

use the real rate of return to shareholders (percentagechangeinthe real value of shareholder wealth, including dividend payments) as our

measure

of

performance.

Our

risk

measure

is based onthe sample varianceofdailystock returnsforthelast120 daysofthe fiscal

year. Following

Aggarwal and

Samwick,

we

usethe cumulative distribution function

(CDF)

ofthe variance ofreturnsinthe sample asourriskmeasure.

The

smallestobservedvarianceinthesample has a

CDF

value of0; thelargestobservedvariancehas a

CDF

valueof I.

We

followBertrand

and

Mullainathan (1999a)

and

split ouroriginalsampleinto

two

subsamplesoffirms

based on whetherornot theyhave at leastoneblockof at least fivepercentof

common

shsiresinthesample base yecir (1984), whether the block holder is or is not a director.

As

before,

we

exclude blocks that are

owned by

the

CEO.

The

results forthe full sample,not reported here,

match

the findings in

Aggarwal

and

Samwick

(1999).

CEO

pay becomes

less sensitive to the rate of return to shareholders as the volatility of stock returns

increases. InTable 3,

we show

how

the presenceofalargeshareholdermediatesthesefindings. Regressions

areestimated by

OLS. Each

regressioncontainsasindependent variablestheperformance measure, the risk

measure

and

the interaction ofperformance withrisk. Thisinteraction termis

what

cJlows us tounderstand

the effectofrisk

on

the pay toperformancesensitivity. Inaddition,

we

include firm fixedeffects, yeaxfixed

effects, aquadraticin

CEO

age

and

aquadraticin

CEO

tenure (but do not report these coefficients in the Table).

Column

(1) focuses on firms with alarge shareholderpresent. Here

we

find support forthe contracting

model. Highervariance

means

lowerpayforperformancesensitivity.

Column

(2) focuses onthe firmswhere

there is no large shareholder present. For this group, there is

no

relationship: the pay for performance

sensitivity does not

depend

on the riskiness of the stock.

Hence

the existing finding that the variance of

firmsin theScimple.

IV

Synthesizing the

Empirical Findings

We

have so far laid out aset ofempirical facts that support the general claim that better governed firms

behave according to the contracting

model

while worse governed ones behave according to the

skimming

model.

A

key to

moving

forward will betodevelop a

model

that isconsistent withall thesefacts.

A

veryfirstconcern thatneedstobe addressed iswhetherourfindingsrepresentaspuriousrelationship.

The

apparentcorrelationbetweenthe use ofoptimal

CEO

compensation contractsandthe presenceoflarge

shareholdersmight not reflect a truedirect relationship. Instead,

skimming and

weak

governance might be

relatedto eachotherthrough

some

thirdfactorthat

we

aie not observing or are not adequatelycontrolling

for.

But what

could thisthird factor be?

Firm

sizemight be one example.

Owning

5 percent ofthe shares ofalargeflrmis

more

expensive so that large firmstypicallyhavelesslargeshareholders.

We

can, ofcourse, appropriately control forsizein the abovetests

and

when

we

did so theresults did not change.

The

deeper

question, however, is

why

a third factor (such as size) would consistently lead tothe pattern ofresponses

that

we

see. For example,

why

would

larger firms (which

would

be correlated with poorer governance)

respondto takeoverlegislation with greater

mean

pay

and

lesspay forperformanceincreases, rewardtheir

CEOs

more

forluck, chargetheir

CEOs

lessforoptions

and

notaccountforvarianceinchoosingthepayfor

performancesensitivity? Ifit isnot becauseofpoorergovernance,then

why?

Itis always apossibilitythat

some

unobserved factor drives our results.

We

simply find it hard to point to any such factor that would

intuitively

match

the consistent patterns that

we

observe.

Assuming

that these resultsareindeed about governance, the simplest

way

toexplain

them

seemstobe

through a bargainingmodel.

Suppose

shareholders

and

the

CEO

bargain over the pay package.

Skimming

could then be

modeled

as the case

where

the

CEO

has

much

or all ofthe bargaining power. Contracting

would bethe casewhereshareholdershave

much

orallofthe power. Sucha bargaining

model

wouldalsohave

While

intuitivelyappealing, this

model

cannot

match

the results above.

To

seewhy, notethat the Coase

Theorem

appliesinthismodel.

The

bargaining

power

ofthe

CEO

willonlydetermine

how much

averagepay

shegets,not the structure ofher contract.

A CEO

with

more

bargainingpowerwill not chooseto get

more

pay forluck. She will alsohaveluck shocks

removed

from her pay but will simply expect ahigher average

compensation. Similarly, there is

no

reason that she

would

want

to be chargedless foroptions grants. She

will

want

to facethe

same

chargeforoptions but merelytake a bigger compensation package overall.

More

generally,the optimalcontractwill alwaysbe chosen withbargaining

power

simplydeterminingthe division ofrentsbetween the shareholders

and

the

CEO.

An

alternative modeling approach could be to focus on the superior monitoring technology of large

shareholders.

Our

findingscould betheresult ofan optimalcontracting processwhereprincipalsalwaysset

pay, whether governanceis

weak

or strong, but face different signal to noise ratios

when

evaluating

CEO

effort. In firms with large shareholders, principals can

more

easily separate

CEOs'

effort from other noisy

movements

in firm performance. Such a view could potentially explain

why

there is

more

pay for luck in

firms without large shareholders.

One

would, however, need to

assume

that observing

movements

in oil

prices or in aggregateindustry shocks requires superior monitoring technology

and

cannot be done ecisily,

for

example

by openingthe business section of the daily newspaper.

The

monitoring

model

cannot at all explain ourfindingson the impactoftakeoverthreats.

Why

would

principals infirmswithoutlargeshareholders decide to give higherpay once

CEOs

areprotectedfromhostile raiders? Their monitoring technology

may

be weaker but they still

know

the laws have been passed and

should react tothem. Similarly,thetrade-offbetweenincentivepay

and

varianceofpayareequallypuzzhng.

Ifanything, it istheprincipals ofthe well governed

who

should careless aboutstockmarketprice volatility since they have access to better signals of effort.

As

a whole, it is hard to imagine

how

differences in

monitoringalone could explain the array ofresults.

Thus

our findingssuggest that governanceis notjust about increased bargaining

power

or better

mon-itoring. Instead, they suggest that governance is about

who

has effective control. In contracting models,

we

always

assume

that

some

metaphorical principal controls the pay process.

Even

when

governance is

weak,this principalstill setspay (perhaps takingintoaccount a worse monitoringtechnology).

Our

results

suggest thatabetter

model

ofgovernance

would

be onethat recognizes that

good

governanceis

what

allows shareholderstomaintaineffective controlof, forexample, the pay process.

To

be concrete, consider the details of the pay process. In practice,

CEO

pay is usually set via the

compensation committee. This

committee

may

cater to the interests of the

CEO

or ofthe shareholders.

When

governanceisgood, suchas

when

thereisalarge shareholderpresent, thiscommittee

may

make

sure

that the pay package looks optimalfromthe shareholders' perspective.

The

committee willrespond tothe

passage of takeover legislation, or will be

more

reluctant to radse the

CEO's

bonusjust because oil prices

rose andso on.

When

governanceisweak, however, this committee

may

be

much

more

willingto cater tothe

CEO.

How

does the committee set pay then?

Even

though the

CEO

has de facto control of this committee, she still

facesconstraints.

The

committee

may

be quite reluctant toattractthe attention of shareholdersor ofother

important constituencies, such as labor unions or the business press. This places constraints not just on

how

much

can be

skimmed

but alsoon

how

skimming

will takeplace. For example,

more

can be

skimmed

when

firms' performance is high as shareholders

may

be paying even less attention to the firm.

Pay

for

luckthen naturally arises. Also, ifshareholders mostly pay attention to their company's

bottom

line, the

compensation committee will grant relatively

more

stock options as they are not charged directly against

earnings.

V

Conclusion

Thisdiscussion highlights asetofopen questionscis

we

move

forwardfromthe empirical regularitiesabove. First,

we

need to better understand

what

happens

when

the

CEO

has gained de facto control ofthe pay

process.

What

are the real constraints on pay setting? This will require a

more

rigorous formalization of the

skimming

view. Second,

we

need to reinterpret

what

corporategovernance actually does.

We

need to reconceptualize governance asthe transfer ofdefacto control ofimportant decisions from the

CEO

to the

shareholders. Such a reconceptualization will haveapplications

beyond

executive compensation.

Take

for

example

the decisionbyfirmstoadopttakeover protectionsuchaspoisonpills. Should

we

thinkthisdecision

is

made

with the interests ofshareholders or

management

in

mind?

This question is

somewhat

analogous

to whether

we

think

CEO

payis the result ofoptimalcontracting or skimming. Perhaps governanceplays

a central role in thisapplication too, with well governed firms using poison pills to raise bargaining

power

during takeover attempts

and

poorly governedfirmsusing poison pills toentrench

management.

This final

example

highlights thebroadervalue ofa reconceptualizationof

good

governanceasbeing

what

gives

CEOs

principals.

References

Aggarwal, Rajesh and

Samwick,

Andrew.

"The

Other Side ofthe Trade-off:

The

Impact

ofRiskon

Executive Compensation." JournalofPolitical

Economy

, February 1999, 107(1), pp. 65-105.

Bertrand,

Marianne

and

MuUainathan,

Sendhil. "Corporate Governance and E.xecutive

Compensa-tion? Evidence from TakeoverLegislation."

Mimeo,

PrincetonUniversity, 1999a.

Bertrand,

Marianne

and MuUainathan,

Sendhil.

"Do

CEOs

Set Their

Own

Pay?

The

Ones Without

PrincipalsDo."

Mimeo,

Princeton University, 1999b.

Crystal, Graef. In Search ofexcess:

The

Overcompensation of

American

Executives.

New

York:

W.W.

Norton Co., 1991.

Holmstrom,

Bengt.

"Moral

Hazard

and

Observability." Bell Journalof

Economics

. Spring 1979, 10(1),

pp. 74-91.

Shleifer,

Andrei

and

Vishny,

Robert.

"LargeShareholdersand CorporateControl." Journalof Political

Economy, June

1986,94(3), pp. 461-488.

Table

1

—

The

Impact

Anti-

Takeover

Legislation

on

CEO

Pay:

The

Role

of

Large

Shareholders"

DependentVariable: LogofTotal

CEO

Compensation

(1) (2)

Large

Shareholder?

Yes

No

Anti-Takeover

Law

Adopted .026 .075" (.040) (.031) Adjusted R- .633 .768

Sample Size 2281 2268

"Notes:

1. "Large Shareholder" is a

dummy

variable thatequals 1 ("Yes") ifthe firm hasa strictly positivenumberofblocks of atleast fivepercent of

common

sharesinthebase year(1984), whether theblock holder isor is not adirector. Blocks ofat least five percent that are owned by

CEOs

areexcluded.

2. "Anti-TakeoverLawAdopted" isa

dummy

variablethatequals 1afterthe adoptionofan anti-takeoverlaw (BusinessCombinationStatute) bythe state the firmisincorporatedin. 3. Each regression includes as controls yearfixed effects, firm fixed effects, a quadratric in

CEO

age, a quadraticin

CEO

tenure, thelogarithm oftotal assets and the logarithm of totalemployment.

4. '* denotessignificanceat the5%.

Table

2

—

_The

_Impact

of

Anti-Takeover

Legislation

on

Pay

for

Performance:

The

Role

of

Large

Shareholders"

Dependent Vaxiable: LogTotaJ

CEO

Compensation

(1) (2)

Large

Shareholder?

Yes

No

Anti-Takeover

Law

Adopted .017 .085'" (.040) (.030) Anti-Takeover

Law

Adopted* 1.126** .316

Ace. RateofReturn (.582) (.584)

Ace. RateofReturn 2.352** 2.533***

(1.116) (1.08)

Adjustedi?^ .641 .782

SampleSize 2281 2268

"Notes:

1. "Large Shareholder" isdefined asinTable 1.

2. "Anti-TakeoverLawAdopted" isdefined as inTable 1.

3. AccountingRateofReturnistheratioofNet IncomeoverTotalAssets. Ithasbeendemeaned.

4. Eachregression includes as controls year fixed effects, accounting rate of return interacted with year fixed effects, firm fixed effects, aquadratric in

CEO

age, a quadratic in

CEO

tenure, thelogarithm oftotal assets andthelogarithmoftotalemployment.

5. '" denotessignificance at the 5%;'*" at the 1%.

Table

3

—

_The

_Impact

of

Risk

on

CEO

Pay:

The

Role

of

Large

Shareholders"

Dependent Variable: Log of Total

CEO

Compensation

Large Shareholdei (1) •?

_Yes

(2)

No

Performance

.003—

(.000) .001-— (.000) Performance*

CDF

ofVariance -.004—* (.001) .000 (.001)

CDF

ofVariance -.07 (.06) -.17'" (.05) SampleSize 2301 2025 "Notes:

1. "Large Shareholder" isdefined asinTable1.

2. "Performance" isdefined as the realgrowthrate ofshareholderwealth (includingdividend payments) and ismeasuredinpercentagepoints.

3. Eachregressionincludes as controlsyearfixedeffects,firm fixedeffects,a quadraticin

CEO

ageandaquadraticin

CEO

tenure.

4. *"* denotessignificanceat the1%;'*" atthe .1%.

Date

DueDEC

2000