BEHAVIORAL CONSEQUENCES OE CORPORATE...

MANAGEMENT SCIENCEVol. 38. No. 9, September 1992

Primed in U.S.A

BEHAVIORAL CONSEQUENCES OE CORPORATEINCENTIVES AND LONG-TERM BONUSES:

AN EXPERIMENTAL STUDY*

ANDREW SCHOTTER AND KEITH WEIGELTDepartment ofEeonomics, 269 Mercer Street, New York University,

New York. New York 10003Department of Management, The Wharton School. Sleinberg-Dietrich Hall,

University of Pennsylvania, Philadelphia. Pennsylvania 19104

This paper examines whether long-term managerial bonus schemes change the allocative behaviorof subjects in a laboratory' setting. Using four different compensation schemes, we show that anecessary condition for reconciling divergent lime preferences between principals and agents isa compensation scheme that induces behavior consistent with lower discount rates. Within subjectresults show that subjais recognize changes across compensation schemes and change their behavioras predicted by formal theor>\ Results also suggest that subjects become more myopic in theirinvestment decisions if compensation contracts are incorrectly structured.(EXPERIMENTAL HCONOMICS. MANAGERIAL COMPENSATION. AGENCY THEORY.LEARNING)

1. Introduction

Principal-agent models play a pivotal role in the effort to develop an empirically viabletheory ofthe firm. The models address the basic question of whether managers takeactions to serve their own self-interests or the interests of shareholders. Agency theoryrecognizes the divergent interests of managers and shareholders, and attempts to alignthe divergence via contracting schemes. Interests of managers and shareholders can divergefor several reasons.' This paper focuses on divergent discount rates, a divergence thatcauses stockholders and managers to value current and future profits differently. Becausemanagers often tend to discount the future at a higher rate, they may take actions thatare too myopic from the shareholder's perspective. Hence, managers may choose actionsthat yield flashy short-run profits today, at the expense of long-run profits tomorrow.

Theorists, who believe that managerial behavior is motivated by the structure of in-centives, specify the use of compensation contracts to align divergent interests (Dukesetal. 1981, Horwitzand Kolodny 1981. Lambert and Larcker 1985). Firms have heededthis advice, and try to align discount rates with long-term bonus schemes. These schemestie an agent's compensation to long-term (usually 3-4 years) organizational performance.Long-term bonus schemes are used by 99% ofthe top 200 industrial firms and 95% ofthe top 20 banks (Sibson & Company 1989). 73% of these long-term compensationcontracts were instituted after 1980, and their usage has steadily increased (O'Dell 1986;Sloan 1991).

Implicit in normative models, and the empirical use of long-term bonus schemes, isthe crucial (and basic) assumption that agent behavior is governed by the structure ofcompensation contracts. A few empirical studies have tested this assumption using naturaldata, but they have not been totally successful. The studies did not differentiate betweenchanges in behavior due to contract changes, and changes in behavior due to uncontrolledvariables {Lambert and Larcker 1987). Also, data for key parameters (e.g., utility

• Accepted by William T. Ziemba: received July 17, 1987. This paper has been with the authors 10 monthsfor I revision.

' Divergent interests include direct agency costs, risk attitudes, and decision-making time horizons (see Ross1973: Shavell 1979; Lambert and Urcker 1985).

1280O025-19O9/92/38O9/l280$0l.25

Copyright fe iy92. The Insliiutc of Management Sciences

CORPORATE INCENTIVES AND LONG-TERM BONUSES 1281

functions, payoffs) were not available. Hence, how changes in compensation contractsinfluence decision-making behavior is still a largely unanswered empiiical question.

Despite these problems, testing the association between compensation contracts andbehavior seems useful. At best, normative theories only prescribe optimal contract design;the processes of decision-makers recognizing changes and rationally modifying behaviorare simply assumed to follow. Given the limitations of natural data, we tested the asso-ciation in a laboratory setting. In so doing, we are able to provide answers for severalimplementation issues: Do individuals recognize changes in incentives and modify theirbehavior to maximize monetary payofls? Behaviorally, are individuals actually calculatingthe monetary maximizing solution of a compensation contract or do simple heuristicsgovern their actions? If heuristics are used, do decision rules approach the monetarymaximizing solution? Finally, can we recognize compensation contracts that do notreconcile the divergent interests of managers and shareholders?

Our laboratory experiment used students at the Wharton School of Business. Subjectswere asked to make investment decisions with scarce resources, a decision environmentanalogous to that faced by managers. Laboratory compensation contracts were varied tosee if changes in contracts induced the predicted changes in decision-making behavior.Our baseline compensation contract was a stylized version ofthe compensation contractof a Fortune 100 firm (Corporation X). Corporation X had recently instituted a long-term bonus scheme whose mechanics were described to us.' The firm hoped the schemewould induce managers to more carefully consider the long-term implications of theirdecisions. We believed the scheme's design would not change decision-making behaviorbecause it did not affect the discount rate of managers. To show that the discount ratedivergence problem is not irreparable, we then devised and tested alternative long-termcompensation contracts that induced behavior consistent with lower discount rates.

In summary, our results show that a necessary condition for reconciling the divergenttime preferences of shareholders and managers is a compensation contract that effectivelyinduces behavior consistent with lower discount rates. Individuals do recognize changesin compensation contracts, and their behavioral change approximates that predicted bynormative models. Few individuals calculate monetary maximizing solutions; insteadthey rely on simplifying heuristics.

We proceed as follows: §2 discusses Corporation X's incentive scheme and the tradeoffsit offers agents. §3 describes the experimental design. Behavioral predictions are presentedin §4 by deriving the monetary maximizing behavior of laboratory subjects under theincentive schemes shown in §2, Hypotheses and resuhs are presented in §5, and conclu-sions in §6.

2. Corporation X's Managerial Compensation Contract

The design of Corporation X's compensation contract is similar to that of other firms.A manager's salary is derived from three sources: a base salary, a short-term bonus schemewhere bonuses are an increasing function of current business unit performance (call thisprofit), and a long-term bonus scheme where bonuses are a function ofthe arithmeticmean of profit over the past four periods.

Denote a manager's base salary as B. For expositional simplicity, assume in the short-term bonus scheme that if T, measures business unit performance in period /. then theperiod's short-term bonus is S, = CVTT,. where a is a constant of proportionality (i.e.,period /'s short-term bonus is proportional to the business unit's performance in period/) . Corporation X's long-term bonus in period / is L, ^ ^({TT, + -K,..] + x,- ^ + T, . I ) / 4 ) ,

^ Managers at Corporation X told us that their bonus scheme was modelled on the scheme of another Fortune100 firm. Therefore, we do not believe the sct.eme was lirm specific.

1282 ANDREW SCHOTTER AND KEITH WEIGELT

where 0 is also a constant of proportionality.-' Thus, total managerial compensation inperiod / equals:

C, = B, + S, + L, = B,+ a x , + ^((7r,+ 7r,_i + 7r,_2 + 7r,_3)/4).

Finally, let r ' represent the firm's discount factor, as stated by shareholders, and let T^'be the discount factor ofthe manager.** If r ' > T^', then the manager discounts the futureat a higher rate than shareholders.^

Assume that managerial decisions made in period / and corporate performance inperiods / + n are correlated. Then given T'^' and a preference to maximize discountedmonetary gains over corporate tenure, a manager chooses a decision set that defines apath of expected business unit profits (whose realizations depend on chance). If T'^'> T'\ then this decision set yields an expected profit path which, when viewed by share-holders, is too high in the present and too low in the future.

Does Corporation X's long-term bonus scheme rectify this situation, i.e., does it inducemanagers to modify behavior? As we formally see in §4, the answer is no. At best, thepresent scheme is ineffective because depending upon how the manager perceives theplan, the monetary maximizing manager chooses actions that are either identical to onesmade under the short-term plan or are even more myopic (see §4). The reason is simple.Ignoring start-up and terminal problems, any decision set which maximizes the manager'sshort-term bonus will also maximize the long-term bonus since the latter is simply aweighted average ofthe former. Corporation X's long-term bonus scheme fails to revisetradeoffs between current and future performance beyond that already reflected in theshort-term bonus scheme and the manager's own discount rate.

The Purpose of Long-Term Incentive Programs

Theoretically, a successful long-term bonus scheme should alter the manager's per-ceptions ofthe future, either by revising objective tradeoffs or the subjective discountrate. Corporation X's current long-term bonus scheme does neither. We show this bydeveloping an alternative long-term scheme which does alter perceptions ofthe future,and thus behavior. The rationale and description of our plan is as follows: Assume amanager discounts the future at rate /;. The manager will value future profits less thanshareholders if the firm's discount rate is less than p. Firms can reduce p by magnifyingor inflating compensation in later periods. For example, firms can make compensationin period ! (t years after joining the firm) a function ofthe discounted present value ofprofits, discounted back to the initial period at a predetermined rate (T (o does not nec-essarily have to equal r). For an appropriately chosen range of a. this makes profitsearned in later periods relatively more valuable, thus altering the manager's perceptionsof future periods. In the manager's mind the tradeoff between short- and long-term per-formance has changed. Managers now view profits earned today less favorably. Hence,managers should give greater consideration to the long-term consequences of investmentand production decisions.

3. Experimental Design

Laboratory experiments are accepted as a useful and legitimate way to test economictheory {see Smith 1976, 1982). The control ofthe laboratory allows the testing of theories

' Of course, profits are not the only index of corporate performance to which bonuses can be linked. Theproper index to which to link bonuses depends on the stated goals of each business unit,

''Since Corporation X's compensation contract is only offered to managers whose decisions substantiallyaffect performance, assume units have one manager whose decisions (along with elements of chance) uniquelydetermine performanee. Of course, this simplifying assumption neglects any organizational decision dynamics.

'The discount "factors" ofthe corporation and its managers are x' = 1/(1 + r) and - '" ^ 1/(1 + p).respectively, where r and p are the discount ""rates" tbey use.


TABLE 1

E.xperimental Design

Session

1

2

3

4

Session

Length

10 rounds

10 rounds

10 rounds

10 rounds

Number ofTen-Round

Trials

20

20

20

20

Endownfientper TenRounds

400

400

400

400

Compensation

Scheme

Corporation X'sshort-term

X's long- andshort-term

Alternative planff = 0.20

Alternative plana = 0.08

Number ofSubjects

18

24

15

26

for which natural tests are difficult: researchers can design environments that enable themto examine relevant attributes of subjects, specify parameters, and control for individualdifferences on relevant dimensions (Roth 1985). As previously noted, little, if any, em-pirical research sbows that compensation contracts induce individuals to lengthen theirdecision-making horizon (Lambert and Larcker 1985). Our experimental design con-trolled parameters across subjects which enabled us to test the theory's descriptive validity.

Overview ofthe Task

An experiment using four different conditions was run to test the behavioral impli-cations of changes in compensation contracts. Sixty-seven students participated in theexperiment. During recruitment, and again before the experiments began, subjects weretold that earnings were a function of their decisions. On average, an experiment lastedabout 75 minutes and subjects earned $21.84.'' The decision task and procedure wereidentical across experimental sessions, although payoff structures differed since each con-dition mimicked a different compensation contract. Subjects were randomly assigned toan experimental session, so any significant difference in behavior across conditions shouldbe due to payoff structures. The experimental design is shown in Table 1.

Subjects reported to the computer lab and were given written instructions describingthe experiment and their decision task (see Appendix A). ' An administrator read theinstructions aloud and answered any questions; then the session began. Page I oftheinstructions shows the initial computer screen of each subject. The decision task wasfairly simple: Subjects allocated 400 "units" over a ten-round decision problem. In round1 they decided how many units to allocate. This number appeared on the screen in theround 1 decision no. column. They then had 400 minus their round I number to allocateover the 9 remaining rounds. After making their round 1 allocation decision, a subjectwould press any letter on the keyboard to generate a random number from a uniformdistribution whose range was - 2 to +2. This number appeared in the round I randomno. column. The computer then calculated the square root ofthe decision number, added

'' We chose to use graduate business students since this is a population of potential corporate managers. Wefelt that inference made from these subjects would have greater external validity than those made using under-graduates. Most of our subjects had held corporate positions.

^ Experimenters generally use bland labels. They don't use value-laden terms because these can affect subjectbehavior through nonmonetar\ preferences.


the random number to it, and entered this amount in the round 1 output column.^ Thisoutput was then transformed into a dollar payoff (the transformation was dependent onthe contract being tested). Subjects then moved to round 2 where the procedure was thesame. Any units remaining after the ninth round were automatically allocated in round10. After completing ten rounds, subjects were shown their payoffs, and then the computerreturned to the initial screen shown in the instructions. Each subject completed 20 it-erations ofthe lO-round allocation task. Before the session began, subjects knew howmany units they had to allocate, how many rounds they had to allocate, the range oftherandom numbers, and the value ofthe multiplier.

Payoffs

The four experimental sessions differed only with respect to their payoff structure. Inall sessions, the output (i.e., the square root ofthe decision number plus the randomnumber) was first multiplied by a number called the "multiplier." Multiplier values wereidentical across sessions and decreased across rounds. Values are shown in the instructions.

Consider the 400 "units" as an investment input which if spent today will yield payoffstoday but if spent tomorrow will yield payoffs tomorrow. The tradeoff for subjects betweenoutput today (requiring investment today) and output tomorrow is represented by themultiplier. The multiplier endowed all subjects with an implicit discount rate of 20%.Hence, one unit of output earned in round 1 is worth 5.15 units if compounded overthe remaining nine periods, while a unit of output earned in round 5 is worth only 2.48units compounded over the remaining 5 periods.'' Random elements in the environmentare represented by the effect ofthe random number on output. As in the coiporate world,investment decisions are made before seeing the realization of the random variable.Thus, asking subjects to allocate scarce units over the ten decision rounds is similar toasking managers to make investment decisions which yield returns over their tenurehorizon.

Session I mimicked salient characteristics of Corporation X's short-term bonus scheme.Earnings in each decision round were simply equal to the total number multiplied bythe multiplier. (If output was negative then its value was not multiplied.) Hence, a subject'stotal earnings were simply the earnings over the ten rounds multiplied by a pre-specifiedconversion rate.

Session 2 was a laboratory version of Corporation X's existing short- and long-termbonus scheme. Earnings consisted of two components: current round earnings and earn-ings based on the moving average of output ofthe past 4 rounds.'" The sum of thesecomponents was then multiplied by the multiplier for the round.

As previously noted, we hypothesized that subjects' behavior would not change acrossthe two sessions because Corporation X's long-term bonus scheme (Session 2) failed toaffect the discount rate of subjects. Sessions 3 and 4 tested alternative incentive schemeswhich did affect discount rates of subjects, and thus we expected behavior to change. Inthese sessions we induced behavior consistent with a lower discount rate by using anumber called the "magnifier." Output earned in any round was first multiplied by themagnifier, and then by the multiplier. Multipliers were identical to those used in Sessions1 and 2, and induced a discount rate of 20%. The magnifiers induced behavior consistentW\xh the firm's discount rate. These numbers magnify output earned in later rounds at

" The square root ofthe output was used so tbat the output function would be concave and thus exhibit adiminishing marginal product which bounds the optimal amount of units to allocate in any period. If thesubject's total number was negative in any round then output was equal to the negative ofthe square root ofthe absolute value of that negative number. However, at the equilibrium, negative outputs were not possiblesince at least 4 units were allocated in every period.

' This discounting mechanism was explicitly explained to subjects in the instructions (see Appendix A).'" During the first three periods these means were taken over one. two. and three periods, respectively.


TABLE 2

Optimal Unit Attocation for Session 1

Round

123456789

10

Allocation

125.4987.1460.5242.0229.1820.2614.079.776.784.71

a discount rate of a back to the session's first round. To illustrate how this works, assumethe firm does not discount the future at all, thus output earned in any period is equivalentto that earned in any other period. If the manager discounted the future at 20% (as ourmultiplier induces him to do), then by setting a ^ 0.20 (Session 3). we can inducesubjects to value output equally across all rounds." In Session 4 we set o ^ 0.08. Herethe magnifier does not completely offset the subject's discount rate, thus we expect subjectsto allocate units differently relative to Session 3.

4. Kxperimental Predictions

In this section we present theoretical earnings maximizing solutions to our experiment,which generate behavioral predictions.

Session 1

Risk neutral subjects in Session I faced the following problem:

10

max y \E(xy^^) + E(e,)]{\.20)''-'1 = ]

lU

S.t.

where e, is an i.i.d. random variable distributed uniformly over the interval [—2, +2] .Since the random term is additive and has a mean of zero, this yields the familiar inter-temporal first-order condition:

Subjects should allocate units so their expected discounted marginal product is equalacross rounds. Behaviorally, this yields the optimal allocation scheme shown in Table 2.

Session 2

In Session 2, the perceived maximization problem depends upon how subjects interpretstart-up conditions. The long-term bonus scheme equals a four-period moving averageof previous outputs. Rules governing the start-up period (years 1-3) can significantlyimpact investment decisions, especially if the horizon is not long and payoffs are dis-

" Note that, in all rounds of Experiment 3, the product of the multiplier and magnifier is constant and equalto 5.15. Hence, output is equally valued across all rounds.


counted. Corporation X used a simple rule: the long-term bonus of a new manager inyear 1 equals the short-term bonus for that year. The long-term bonus in year 2 is themean of the short-term bonus in the first two years, etc. If subjects perceive this start-upcondition they will maximize (remembering that t', is additive with a mean of zero):

V - 2xyH\.20f + [xi'' + (̂-vl̂ ' + .vV-)l( 1.20)' '

/=4

10

S.t. 2 Ai = 400, £"(<?/) = 0.

The first-order conditions can be written as functions of Xi, and satisfy the following setoften equations (including the budget constraint).

x\'^ = 0.625^1'-; xV^ = 0.462x|'' and

xy^ - I . 2 0 ' - ' A 7 ' 0 . 6 3 6 , r = 4, . . . , 10,

111

2 A-, - 400.( = 1

If start-up considerations are not perceived, the problem is transformed into a fictitious"steady state" form and formulated as if ii started in period 4 and continued for tenmore periods. More precisely, subjects will maximize

V= 2{A-y' + £-(c,)+ i[xy'--^Eie,)1 = 4

14

s.t. 2 A, - 400.(=4

These first-order conditions are identical to those in Session 1 and yield an identicalsolution.

Hence, at best, the laboratory version of Corporation X's long-term bonus schemeshould cause behavior identical to that observed in Session 1. Furthermore, if subjectsperceive the program in its complete form they will act even more myopically. Table 3shows the behavioral predictions.

Sessions 3 and 4

In Sessions 3 and 4, subjects should maximize the following objective function:

max A^- + E{e,)){ 1 + aVU 1 + p)'-',

s.t. 2 A; - 400,( = 1

where p is the induced discount rate, and a is the rate used to alter the discount rate of


TABLE 3

Optimal Unit .illocation for Session 2

Complete Problem

Allocation

198.8011.1642.5226.9418.7112.589.026.264.353.02

Perceived as:

Round

123456789

10

Steady State

Allocation

125.4987.1460.5242.0229.1820.2614.079.776.784.71

subjects, and hence, change their behavior. First-order conditions are a generalization ofthose in Session I. They take the form:

Note that when p = a (Session 3). an allocation of 40 per round is optimal: When a^ 0, the first-order conditions are identical to those in Session I. In Session 4, we set a= 0.08. Table 4 shows the behavioral predictions for Sessions 3 and 4.

Summary of Predictions

Predictions of subjects" investment behavior across the four conditions are shown inTables 2-4 and Figures 1-4.

5. Hypotheses and Results

a. Hypotheses

The predictions described in §4 provide testable hypotheses. When testing economictheory with experiments, one can analyze the data on either the individual or aggregatelevel. Hypotheses testing with individual level data is a stronger test since it requires thatall subjects act in strict accordance with the theory. Like most experimental studies, we

TABLE 4

Opiiinal Vni! .Utocalion for Sessions 3 ami 4

Session 3(ff = 0.20)Allocation

40.040.040.040.040.040.040.040.040.040.0

Round

123456789

10

Session 4(o = 0.08)Allocation

86.570.156.846.037.230.224.4I9.g16.013.0


40

20

00

80

60

40

20

n

1

\

-

2

1 1 1

-

4 6 8 1

Rounds

FIGURE I. Session I—Mean ObservedAllocations (19th and 20th Iterations).

180

Rounds

FiGtJRE 2. Session 2—Mean ObservedAllocations (I9th and 20th Iterations).

FIGURE 3. Session 3—Mean ObservedAllocations (!9th and 20th Iterations).

FiGiiRt 4. Session 4—Mean ObservedAllocations (19th and 20th Iterations).

Predicted Allocation: Observed Mean Allocation; Predicted Complete Probtem Al-location.

reject the individual level hypotheses, but find support for aggregate level hypotheses.On average, subjects do behave as predicted by economic theory.

Hypothesis 1 tests the goodness of fit between observed and predicted behavior.

Hypothesis 1. In Sessions 1-4. the observed mean allocation path in the \9th and2O//7 iterations does not differ from that predicted by the theory.

Hypothesis 2 tests whether the laboratory version of Corporation X's long-term schemeinduces subjects to alter their allocation behavior relative to Session 1. The theory predictsthat behavior in Session 2 will be identical to that in Session 1 (assuming that subjectsignore start-up and terminal conditions).

Hypothesis 2. Observed investment behavior in Sessions 1 and 2 is identical.

Hypotheses 3a and 3b test the behavioral effect of alternative long-term bonus schemes.Theoretically, the alternative compensation contract offered in Session 3 should changebehavior relative to that in Sessions I and 2. The Session 3 contract should induce a zerorate of time preference (the predicted investment path is a constant allocation of 40 units


per round). Consequently, behavior in Session 3 should be less myopie than that observedin Session 1: subjects in Session 3 will allocate fewer units in rounds 1-3. more units inrounds 5-10, and a similar number in period 4. See Figures I and 3. (A more formaldefinition of myopia is offered later.)

Hypothesis 3a. Ifx,{ 1) andx,{3) are the mean allocations observed in the t"' roundof the \9th and 20th iteration of Sessions 1 and 3. respectively, then

.v,( l)> A-;(3) for / < 3 .

,v,(l) = A,(3) for / = 4,

.v,(l)<.x,(3) for t>4.

The contract in Session 4 increased a to 0.08. Predicted behavior in Session 4 is thefollowing: Subjects should behave more myopically than in Session 3 but less so than inSession 1. Observed investment allocations in rounds 1-3 should be: Session 1 > Session4 > Session 3. In round 4 investment allocations should be: Session 4 > Session 3> Session 1. Allocations in rounds 5-10 should be: Session 3 > Session 4 > Session 1.(See Figures I. 3, and 4.)

Hypothesis 3b. lfx,{\), x,{3). and x,i4) are the mean investment allocations inperiod t of Sessions 1. 3. and 4, then

and

b. Results

X,

X,

Xt

( 4 ) ^

(3 ) =

> . v , ( 4 ) ;

• A , ( 3 ) ?

> A , ( 4 ) .

> A-,(3)

- A,( 1 )

>.v.(l)

jar

for

for

t

t

t

< 3.

= 4,

> 5.

While we summarize results of the 1 st. 10th. and 20th iterations in Table 5, our statisticaltests use pooled data from the 19th and 20th iterations of each session. We do this simplyfor expositional purposes. The raw data are available from the authors upon request.

Hypothesis I. Because allocations across rounds are not independent (due to the 400unit constraint), we cannot apply standard goodness-of-fit tests to the data. However, ifone considers the predicted investment allocations as a hypothesized distribution functionand the observed allocations as an empirical distribution function, we can get some ideafor the goodness of fit by applying the Kolmogorov-Smirnov test. Basically, this testcompares the cumulative distribution functions of the hypothesized and empirical pop-ulations. Results show that in all four sessions there are no significant differences betweenthe hypothesized and empirical allocation patterns.'" Visual inspection confirms thisfinding. Figures 1-4 plot the predicted and mean observed investment allocations in the19th and 20th iterations of each session. Overall, the pattern of observed allocations isvery similar to that predicted by the theory. For example, across the ten rounds, themean deviation between predicted and observed allocations is 2.9 units in Session 1; 5.0units in Session 2; 1.8 units in Session 3: and 2.6 units in Session 4. Subjects in Session2 appeared to perceive the task as a steady state problem: they ignored start-up andterminal conditions.

Although mean observed behavior closely follows predicted, there is considerable vari-ance in individual behavior and a tendency for median observed behavior to fall below

'̂ In fact, the Kolmogorov-Smirnov test statistic which measures the maximum deviation between the hy-pothesized and empirical distributions is very tow in all sessions—0.034 in Session 1. 0.061 in Session 2, 0.022in Session 3. and 0.030 in Session 4.


TABLE 5

Summary Rotiml-hv-Round Statistics in Iterations 1.10. and 20

IterationRound

median1 meanvariance





median6 meanvanance



median9 meanvanance


1

8386

2.789

5063683

5360599

4048633

2828217

3027243

2524198

2124404

1920275

1621753

Session 1

10

8890

1.903

7488

2.966

5362631

4746511

3633478

2024568

1018221

1016

221

1013158

510152

20

100113

2.343

8390

1.575

5856560

4639344

3230316

2521160

1717198

1013179

9II155

59

122

1

6565

1.087

6566

1.477

5559388

464594

4038136

3534339

3032399

2224223

2022298

1516218

Session 2

10

95112

4.866

8079579

6359404

5045327

3531302

2025211

1318214

101267

6985

410

265

20

UK)101

1.081

8079365

6365693

4648276

3034272

2023188

1014185

812141

5993

414

1,006

Session 3

1

4062

2.689

4043454

404053!

4038467

4033170

4033371

4029226

4040529

4039579

4044

1.827

10

4040185

4042151

404097

404572

4045280

4045259

4043349

403K117

4031214

4031296

20

40418

404115

404348

4044124

404465

403827

403828

403756

403839

403688

1

6367904

5660738

5052347

4244210

4044415

3332353

3029201

2927272

2022284

1423613

Session

10

8084

1.642

7069585

5358556

5052478

4038499

3531465

2221212

2021286

1015

213

511

215

4

20

7287

2,514

5661482

5055590

4547118

4040215

3634344

2624230

2019168

1015168

1018

698

mean observed behavior. Variance is considerable and does not decrease in later iterations(except in Session 3). Table 5 presents "snapshots" by showing the mean, median andvariance of subject behavior in the first, tenth, and twentieth iteration.

This variance around the predicted allocation path may imply that even simple com-pensation contracts carry an "institutional risk" which must be borne by firms. Evenwith identical compensation contracts, managerial behavior may vary from firm to firm,or even from plant to plant.

Hypothesis 2. Hypothesis 2 predicts that investment behavior will be the same inSessions 1 and 2. Figures I and 2 show that differences in round-by-round allocationsappear small. To determine whether differences are significant, we used a Mann-Whitney-L'test on the round-by-round data ofthe 19th and 20th iterations (Conover 1985). Thistest measures whether two populations have locational differences. In the absence of


TABLE 6

Observed Mean Allocations and Results of Mann-WhitneyTest for Session I vs. 2 (for \9tb and 20th Iteration)

Round

1234

67S9

10

Session1

120.680.558.443.030. t21.316.612.810.08,5

2

105.583,265.846.634.623.213.111.79.17.6

r-value

0.341.250.950.750.460.460.870.780.310.31

locational differences, it is assumed that the populations are equivalent. Table 6 presentsthe results. The table clearly shows no significant difference in behavior between anyround of the two sessions. Only i z-valuc (round 2) is even above one. The laboratoryversion of Corporation X's long-term bonus scheme failed to alter investment behaviorin Session 2 relative to Session 1. This result was predicted by the theory since the firm'slong-term bonus scheme did not fulfill a necessary condition, namely induce behaviorconsistent with a lower discount rate.

Hypothesis 3a. Hypothesis 3a tests whether a compensation contract that theoreticallyinduces subjects to act as if they have a lower discount rate actually changes behavior.In Session 3 the contract theoretically induces a zero rate of time preference. Hypothesis! has established that observed behavior was similar to predicted in Sessions 1 and 3. Asshown in Table 7, a Mann-Whitney-C' test on the round-by-round data of the 19th and20th iterations of Sessions I and 3 confirms that behavior in the two sessions was sig-nificantly different. The table clearly shows that observed mean investment allocationsin every round are highly significantly different, except round 4 where the theory predictssimilar allocations (42.02 in Session 1 and 40 in Session 3). Relative to Session 1, Session

TABLE 7

Observed Mean Allocations and Re.sults of Mann-WhitneyTest for Session 1 vs. "i {for \9th and 2Qth Iteration)

Round

1234567%9

10

Session1

120.680.558.443.030.121.316.612.810.08.5

3

40,04L842,742.342.239.039.0337.737.937.4

z-value

4.66***4.36***2 .00"0.561,55*4,15***3.87***4.48***4.51 *••4,02***

*•• / K 0,001,** p< 0,05.*p<O.IO.


TABLE 8

Kru.skat-Wattis Test /or .Significant Diffirencesin Behavior Between Sessions 1. 3, and 4

Round

123456789

10

Test Statistic

6.6**0.2

16.1***23.4***29.0***23.6***24,3'**2 6 . 7 ' "37.i***37.0***

*** />< 0,001.** p< 0.05.

3's compensation contract did significantly alter the investment decisions of subjects andthey did act "as if" they had a zero rate of time preference.

Hypothesis 3b. Hypothesis 3b predicts that altering the compensation contract bychanging o to 0.08 will cause subjects to change their investment behavior in Session 4.As shown by Figures 1, 3, and 4. the rank order ofthe observed mean allocation pathsis as predicted. We analyzed the round-by-round data using two statistical tests: A Kruskal-Wallistest (Table 8). and a set of pair-wise Mann-Whitney tests between Sessions I and3, I and 4, and 3 and 4 (Table 9).

The Kruskal-Wallis test determines whether data from different samples can be con-sidered as being selected from a single (combined) sample. Table 8 shows that we eanreject this hypothesis for all rounds except 2. Behavior in Sessions 1. 3. and 4 is significantlydifferent. The pair-wise tests indicate in what rounds these differences appear. Across thethree sessions, there is significant differences in observed mean investment strategiesexcept in those rounds where the theory predicts similar allocation strategies. In Sessions1 and 3, behavior is significantly different in every round except 4. Predicted allocationbehavior in round 4 is 42.02 for Session 1 and 40 for Session 3. In Sessions 1 and 4,

TABLE 9

Results of Mann-Whitney Test for Pair-Wise Compari.son Using Mean Attoeat ions

Round

123456789

10

1

137.57.•1.9

54.439.026.618.816.815.49.48.0

Session3

46.443.340.041.639.639.340.334.034.540.7

4

73.052.048.840. i30.535.129.728.831.330.0

1 vs. 3

4.66***4.36"*2.00'*0.561.55*4.15***3.87***4.48***4.51"*4.02*'*

z-value1 vs, 4

2.10**2.11**0.561.121.43*L89**L36*1.142.06**1.12

3 vs. 4

3.66***2.45**L90**0.882.61**4.50***4.61***4 .71"*3.24***

*** /)< 0.001.** ;7 < 0.05.* P<0.\0.


allocation behavior is similar in rounds 3, 4 and 10. Similar behavior in rounds 3 and 4is expected since predicted allocation in these rounds is 60.52 and 42.02 for Session 1and 56.8 and 46.0 for Session 4. Similar allocation bebavior in round 10 was not expected.Allocation behavior is significantly different in all rounds of Sessions 3 and 4 exceptround 5. This was expected since the predicted allocation behavior is 40 in Session 3 and37.2 in Session 4.

Hence, differences in compensation contracts across the 3 sessions did apparentlychange the investment behavior of subjects. The change in bebavior was as predicted bytheory. Subjects in the 3 sessions did act as if their discount rates were inherently different.

Within Subject Behavior. The analyses above bave examined bebavior between sub-jects. We have not examined whether a subject will change her behavior when given adifferent compensation contract. To test within subject behavior, we randomly selected8 subjects from Sessions 2 and 4 and asked them to perform the allocation task again.We kept them in the laboratory, and publicly read tbem tbe instructions again, Tbe onlydifference in the instructions was tbe payofl'structure. So 8 subjects wbo participated inSession 2 were then given tbe instructions (and payoff structure) for Session 4: and 8subjects who participated in Session 4 were given the instructions for Session 2.

There was no significant difference in tbe investment behavior of the 16 subjects wboonly participated in Session 2 and the 8 subjects wbo first participated in Session 4, andthen in Session 2. The same was true for tbe bebavior of only Session 4 subjects and the8 subjects who first participated in Session 2. However, tbe mean allocations of bothgroups of 8 were significantly different when compared across experimental conditions.Subjects did change tbeir behavior when they were given a different compensation con-tract. And their behavior changed as predicted by the theory: Subjects wbo first participatedin Session 2 and then in Session 4 allocated fewer units in early rounds and more in laterrounds in the latter session. Subjects who first participated in Session 4 and then inSession 2 did tbe opposite.

Myopic Behavior. One objective of our experiment was to determine whether changesin compensation contracts would alter the myopic behavior of subjects, and if so by howmuch. To do this, we define myopia, in the context of our experiment as follows:

DEFINHION. Subject A is more myopic tban subject B if:I^.dt)> Fad), for round t ^ 1 ,2 10 where:Flit) and F B ( / ) are the cumulative amount of units (cumulative mean amount of

units) allocated by subjects/i and B, respectively, by round /.Hence, if one subject or group of subjects has a cumulative distribution of units which

is everywhere above tbe cumulative distribution of another subject or group of subjects,we say tbe former is more myopic.

Call M, our myopia measure for Session /, i = I. 2, 3. 4. Then given our bonusschemes, the predicted myopia ordering is MI --- M2> M4> M^. We test this by plottingthe cumulative distributions of mean investment allocations in the 19th and 20th iterationsof Sessions I, 2, 3, 4 (Figure 5).

Observed myopia orderings are as predicted: Subjects in Session 1 are the most myopic,followed closely by subjects in Session 2, and then by subjects in Sessions 4 and 3.

Learning. One interesting question that remains is whether subjects accumulateknowledge during the session. To investigate this, we examine the observed mean cu-mulative allocative distributions across sessions in the 1st, 10th and 20th iterations (Fig-ures 6-9).

In Sessions 1 and 2, subjects appear to become more myopic over time. That is. inbotb sessions, the cumulative unit distributions of the 1st iteration is everywhere belowthat of the 10th iteration, which is below (or equal) that of the 20th iteration. Hence, bydefinition, compensation contracts in Sessions 1 and 2 motivated subjects to becomemore myopic. This is not surprising since both contracts theoretically reward myopicbehavior.


400

8 105 6 "

Round

EiCiiJRF 5. Cumulative Mean Allocation (19th and 20th Iterations).

LEGEND: O Session 1; V Session 2: D Session 3; 0 Session 4.

In Sessions 3 and 4 compensation contracts do not reward myopia, and myopic behaviordoes not increase. While subjects may have an initial level of myopia, this myopia tendsto slightly increase by iteration 10 and then decreases by iteration 20. These resultssuggest that if reward structures in compensation contracts encourage myopic behavior,then individuals do become more myopic. Firms that institute contracts which rewardmyopic behavior (i.e.. Corporation X) will exacerbate potential agency problems, andmay induce lower corporate performance.

6. Discussion and Conclusions

The development ofan empirically valid theory ofthe firm requires an understandingof compensation contracts and their effect on managerial decision-making. Normativeincentive models assume that conditions in contracts influence investment and productionbehavior; however this hypothesis has virtually no empirical support. Our findings supportthe hypothesis: subject behavior was significantly modified when the theory predicted itwould be. In our sessions, changes in compensation contracts caused subjects to lengthentheir decision-making horizons. These results suggest that compensation contracts canalign divergent time preferences when bonus schemes induce managerial behavior thatis consistent with lower discount rates. Because managers maximize their discountedearnings through corporate actions, correctly constructed compensation contracts caninduce managers to maximize preferences of shareholders.

Admittedly, principal-agent contracting settings are complex and many factors affectbehavior. Our laboratory design was a simple abstraction ofthe corporate decision-makingenvironment. External validity questions remain unanswered and we view results assuggestive. The sessions were designed to test the descriptive validity of the underlying


400

300 -

^ 200 -

100

Round

FIGURE 6. Cumulative Mean Allocations{1st. 10th and 20th Iteration).

400

300 -

^ 200 -

100

FIGURE 7. Cumulative Mean Allocations(1st. lOth and 20th Iteration).

400

FIGURE 8. Cumulative Mean Allocations(1st. 10th and 20th Iteration).

400

300 -

5 200

ElGURE 9, Cumulative Mean Allocations(1st. 10th and 20th Iteration).

LEGEND: O 1st Iteration; V 10th Iteration: • 20th Iteration.

theory. They suggest how individuals would react to changes in compensation contracts.Some field work does support our results. Larcker (1983) examined whether changes incompensation schemes were associated with changes in managerial behavior. Specifically,he looked at the association between the adoption of long-term compensation schemesand increases in corporate capital investment. Results indicate that, after firms adoptedsuch schemes, managers significantly increased capital investment expenditures.

The sessions also allowed us to more closely examine how individuals adapt to changesin compensation contracts. While subjects behaved as predicted, few, if any, were makingthe necessary calculations to maximize monetary gain.'^ Yet, on average, subjects actedas //they were making the calculations. We surveyed subjects at the conclusion of eachsession. One question asked them to describe their investment allocation strategy. Manyreplies mirrored the intuition underlying the mathematics. Subjects chose decision heu-ristics which behaviorally approximated the normative solution. Some examples were asfollows:

"I'd allocate units using a tapered strategy—with more allocated at the beginning."

Although, subjects had the required information and were provided with calculators.


"Fd allocate a slightly decreasing number of units each round. I'd start at 90 units andend with 5 in round 10."'Td start by designating a high number of units in round 1 (usually 115) and thengradually decrease the number of units over the remaining rounds. I'd try to leave 10units for the last round."(a subject in Session 3) "40 units in each round seemed to be the most effectivestrategy, although 1 did try others."Subjects appeared to experiment with various heuristics and then chose the one that

worked best {i.e.. maximized monetary earnings).Learning did occur. When we asked subjects whether their investment allocation strat-

egy was the same in iteration 20 as in iteration 1, 94% said their strategy changed. Thisresult reinforces the idea that individuals do not calculate the maximizing monetarysolution, but instead experiment with various strategies. They arrive at an optimal strategythrough trial and error. Hence, it seems reasonable to expect that in scltings with longlearning periods (like corporations) individuals will adapt to compensation changes andchange their behavior accordingly.

Finally, laboratory experiments may prove a practical way for firms to test compensationcontracts before their implementation. Presently, few contracts are pre-tested becausetesting methods are not available. A badly designed contract can be very costly, both interms of firm performance and managerial salaries. Conducting a "test" contract withinthe firm is also costly in terms ofthe required testing period and implementation issues.The relative small cost and time involved in laboratory testing seem cost effective.'"

'•"This project was funded in part by grants from the New York University Fund to develop University/Industry Linkages, and the Reginald Jones Center for Management Policy. Strategy, and Organization. Otherresearch support was offered by the C. V. Starr Center for Applied Economics at New York University andDavid Hauck. We would like lo thank Clivc Bull for advice in early stages of this project's development andDavid Larcker and two anonymous referees for their comments. Our thanks to Ionnis Loumakis for his pro-gramming assistance and Ken Rogoza for his research assistance.

Appendix A

Subject No.

Instructions

This is an experiment in decision making. Various research institutions have provided funds for this research.The instructions are simple, and if you follow them carefully and make good decisions, you could earn aconsiderable amount of money which will be paid to you in cash.

Experimental Procedure

Your Computer Screen:

As you sit down at your computer terminal, the screen will contain the following table.You and the computer will fill in this table as the experiment proceeds.

Round Decis. # Random Output Multiplier Earnings

1 5.15

3 3.5«

5 2.4^6 1217 1.72

9 1.210 ]


Your Task

In the experiment, you will perform a simple task. You have been endowed with 400 units, and must allocateall 400 units over the 10 rounds of the experiment. The number of units that you choose to allocate for anyspecific round will be called your decision number for that round. For example, in round 1 you can choose toallocate any number of units between 0 and 400. Say you decide to allocate 20 units in round I. Your decisionnumber for round I is then 20. In round 2, you can now allocate any number of units between 0 and 380 {yourbeginning endowment of 400 minus the units allocated in round 1). Note that all 400 units must be allocatedover the 10 rounds. To choose a decision number for round I. the computer will ask you on a prompt linebelow this table to type in a decision number. Do so by typing any number between 0 and 400 and then hittingIhe return key.

The computer will then ask you to verify your decision number by stating: Is — your decision number(y/n)? If it is, type y; if not, type n and then enter another number and hit return again. You will tben be askedto verify your corrected decision number. Once your number is verified, it will appear in the Decis# column,row I of the table.

Random Number

After you have chosen a decision number for round I, the computer will help you generate a random numberfor that round by asking you to hit any key on the keyboard. By hitting such a key you will cause the computerto display a random number between - 2 and +2 in the random column. This number is generated randomlyin such a way that each integer between - 2 and +2 has an equally likely chance of occurring.

Look at the attached worksheet. This sheet shows you the square root of integers between 0 and 400. Thecomputer will calculate the square root of your decision number and add it to your random number. The sumof tbese numbers (the random number and the square root of your decision number) is then displayed in theoutput column.

Note that your total number may be negative. This will occur whenever your random number is negative,and greater than the square root of your decision number. For instance, if you allocale 2 units in a round, anddraw a random number of - 2 , then your total number in that round will be -0.6 ( 1 , 4 - 2 ) , and your outputwill be -0.6.

Finally, the computer will subtract your decision number (Decis# column) from your initial endowment of400. It will then tell you how many remaining units you have at the bottom of the screen. Round I is nowover, and you can move on to round 2.

Round 2 proceeds in a similar manner. Your remaining endowment is given at the bottom of the screen.Your decision number for round 2 can be any amount of units between 0 and the number shown at tbe bottomof the screen. Choose another decision number and generate a random number. Tbe computer will calculatetbe square root of your decision number, add it to your random number and record the sum in row 2, tbeoutput column. Your remaining number of units will then be displayed at the bottom of the screen. We willdo tbis lor !() rounds. If at any round you bave allocated all your 400 units, then you must consider yourdecision number to be 0. The computer will automatically place 0"s in tbe remaining rounds.

Payoffs

The determination of your payoffs is quite slraigbtforward. At the end of eacb round your output (outputcolumn) is placed in a riskless asset earning 20% interest, compounded at the end of each remaining round ofthe experiment. For instance, output earned in round 1 will be compounded at 20% interest for 9 rounds, whileoutput earned in round 5 will be compounded at 2(Y?o interest for 5 periods. To illustrate how mucb compoundingis wortb look at the Multiplier column. This column tells you bow many times your units multiply if you letthem compound over various periods of time. We will call tbese numbers your multiplier factor. Notice tbatthe effects of compounding can be dramatic. For instance, 10 units of output earned in round I is worth 5L5wben compounded at 20% over the 9 remaining rounds. The same number ol" units earned m round 5 is worthonly 24,8 units, or 2.48 times its initial value. Note tbat the multiplier factor for round 7 drops to 1.72. For 10units earned in round 7 the compounded value would be only 17.2 units.

Hence, units allocated in early rounds contribute more to your earnings. However, there is a tradeoff. Tbeattached worksheet shows tbat as your decision number increases, its square root increases at a decrea.sing rate.For example, while the square root of 100 is 10. the square root of 400 is only 20. Because of the diminishingrate of increase, there is a tradeoff in allocating your units. The more you allocate in an early round, the greateris the multiplier effect. However, because of the diminishing rate of increase, the less eacb additional unitallocated increases your output in that round.

At ibe end of each round the computer will multJDly your output for that round (output column) by therelevant multiplier (multiplier column) and record this product in the Earnings column. If in any round youroutput (output column) is negative, tben tbe computer will record this amount as your total earnings (earningscolumn), i.e., losses are not compounded.


You will be paid in the following manner. First, the computer will add up the sums of your compoundedearnings. It will divide this total by 300 and multiply this sum by $1.00. You will be paid this amount whenyou leave. We will perform this 10-round task 20 times in succession. Your final payulTwill be the sum of yourpayoffs in each ofthe 20 10-round experiments. At the end of each 10-round session, the computer will tellyou your earnings for that session, and your total earnings so far. You are free to make as much money aspossible. Are there any questions?

References

CONOVER, W . . Practical Nonparametric Statistics, John Wiley and Company, New York, 1985.DUKKS, R.. T. DYCKMAN AND J. ELLIOT. "Accounting for Research and Development Costs: The Impact on

Research and Development Expenditures," J. Accounting Res.. Supp. 18 (1981), 1-26.HORWiTZ, B. AND R. KOLODNY, "The Economic Effects of Involuntary Uniformity in the Financial Reporting

of R&D Expenditures," J. Accounting Res.. Supp. 18 (1981), 37-74.LAMBERT, R. AND D. LARCKER. "Executive Compensation. Corporate Decision-Making and Shareholder Wealth:

A Review ofthe Evidence," Midland Corporate Finance J.. 2 (Winter 1985), 6-22.AND^ •, "An Analysis ofthe Use of Accounting and Market Measures of Performance in Executive

Compensation Contracts." J, Accounting Res,. 25 (1987 ). 85-125.LARCKER. D., "The Association Between Performance Plan Adoption and Corporate Capital Investment," J.

Accounting and Economics, 5 (1983), 3-30.O'DHLL, C , Major Findings from People, Performance, and Pay, The American Productivity Center. Chicago.

1986.ROCK, M., Handbook of Wage and Salary Administration. (2nd Ed.), McGraw-Hill Inc.. New York. 1984.Ross, S.. "The Economic Theory of Agency," Amer. Econotnic Rev., 63 (1973), 134-139,ROTH, A.. "Laboratory Experimentation in Economics," .advances in Econiimic Theory Bewley, T. (Ed.).

Cambridge University Press, Cambridge, United Kingdom, 1985.SHAVI:LL, S., "Risk Sharing and incentives in the Principal and Agent Relationship," Hell J. Economics, 10

(Spring 1979), 55-74.SlBSON & COMPANY, Fxecutivc Compensation Annual Report. 1989.SlOAN, RICHARD, "An Empirical Examination ofthe Role of Accounting Earnings in Top Executive Com-

pensation Contracts," Unpublished dissertation. University of Rochester, January 1991.SMIIH. V,, "Experimental Etonomics: Induced Value Theory," Amer. t-.conomic Rev.. 66 (1976), 274-279.

, "Microeconomic Systems as an Experimental Science," Amer. Economic Rn., 72 (1982), 923-955.

BEHAVIORAL CONSEQUENCES OE CORPORATE...

Documents

Transcript of BEHAVIORAL CONSEQUENCES OE CORPORATE...