ORGANIZATIONAL INERTIA AND DYNAMIC INCENTIVES. Marcel BOYER Jacques ROBERT

Transcription

1 ORGANIZATIONAL INERTIA AND DYNAMIC INCENTIVES by Marcel BOYER Jacques ROBERT We would like to tank Bentley Macleod, Micel Poitevin, Jean-Pierre Ponssard, Bernard Salanié, seminar participants at te University of Florida, University of Soutern California, Catolic University of Milan, ETAPE Seminar in Industrial Economics (Paris), University of Toronto, Université de Toulouse, Queen s University, University of Maryland, UBC, and participants in te Econometric Society meetings of Pasadena and Hong Kong for teir comments on earlier drafts of tis paper. Financial support from SSHRC (Canada) and FCAR (Québec) is gratefully acknowledged. We remain, of course, solely responsible for te content of tis paper. Bell Canada Professor of Industrial Economics, CIRANO, and CIREQ, Université demontréal. Department of Information Tecnology and CIRANO, HEC Montréal.

2 ABSTRACT Wy would an organization give to internal interest groups te incentives and power to block canges tat migt be beneficial to te overall organization? How will an organization coose to allocate rents and decision power? Wy and in wat sense does suc an allocation generate inertia? Te dynamic principal-agent model presented in tis paper is meant to address tose questions in a formal way. We model te level of inertia as an endogenous rational coice made by te organization (principal). Our results are tree-fold: first, te efficient organizational response to te presence of private information on te value of cange will in general be to bias te decision rule towards te status quo; second, te compensation of te agent differs (an upfront payment versus a distribution of stock options) according to weter te information on te value of cange is private to te principal or te agent; finally, te efficient distribution of real autority for recommending cange in an organization need not always be profitably retained by te principal. Keywords: Inertia, Flexibility, Principal-Agent, Dynamic Incentives J.E.L. D3, D8, G30, J33.

3 1. INTRODUCTION Figting inertia, favoring continuous cange and building flexible companies or agile corporations ave become popular buzzwords in te management literature. Inertia and flexibility are for all practical purposes antonyms in te literature on organizations. Wit few exceptions, flexibility as a positive tune: more flexibility is always better and more efficient. 1 If flexibility is so valuable, wy ten is it te case tat so many organizations (including, not te least, public bureaucracies) appear to fail to meet te callenge of cange? Indeed, many large and powerful companies ave reacted too slowly to te need of cange and eiter went bankrupt or were brougt to te brink of bankruptcy and obsolescence before adapting. Similarly, te political cost of canging obsolete socioeconomic policies as been blamed for te growing burden of government welfare programs. Inertia is a pervasive problem tat organizations face in spite of frequent calls for cange and flexibility by different stakeolders. Wy is inertia so prevalent? Rumelt (1995) claims tat te most crucial problem facing top level management is not product-market strategy but organizational cange: If managers are to commit energy, careers, time, and attention to a program of cange, tere must be trust tat te direction cosen will not be ligtly altered. Here we touc te central paradox tat cange may require te promise of future inertia. Today s inertia may be te result of a commitment necessary to implement cange at some earlier date: to raise te power of incentives, some organizations end up promising future rents to teir members. Over time, tese rents, disseminated across te organization, inibit cange as different groups in te organization want to protect teir rents. Sclerosis and inertia set in until te very survival of te organization and of te rents temselves 1 A report from Business International (1991) stresses te need for companies to be flexible given te important canges in te way competition is likely to operate in te future: markets are becoming more volatile. On te basis of a large number of case studies, Business International claims tat flexibility is te all-inclusive concept integrating a wole set of recent management teories, and moreover tat... collaboration inside and outside te company is te way flexibility is acieved. Tere are few formal definitions of flexibility in te literature. George Stigler (1939) pioneered te analysis of cost flexibility by stating tat firms ave in general to make a coice among different equipments or tecnologies giving rise to different cost configurations, some more flexible tan oters. More formal definitions of flexibility were given by Marsak and Nelson (196) and Jones and Ostroy (1984) in decision teoretic contexts. Tose definitions are reviewed and discussed in Boyer and Moreaux (1989). Harrigan (1985) defines strategic flexibility as... [Te] firms abilities to reposition temselves in a market, cange teir game plans, or dismantle teir current strategies wen customers tey serve are no longer as attractive as tey once were. Valuing flexibility in investments and in organization in general is te central teme of te real options literature; see Boyer et al. (003, 004). 1

4 are in danger. Even ten, te organization may be unable to orcestrate cange. Inertia in organization may take many forms or come from many sources and its analysis is present under different names in many strands of te literature in economics, psycology and management. It underlies te analysis of te rational suppression of potentially valuable information in organizations [Crémer (1995), Friebel and Rait (003)], of te inflexible rules in constitutions [Boyer and Laffont (1999)], of te relative value of commitment and flexibility [Boyer and Moreaux (1997)], of te evolutionary teory of surviving organizations [Hannan and Freeman (1984), Boone and van Witteloostuijn (1997), Carroll and Hannan (000)]. Boyer (004) examines te beavior of corporations toward teir purcase of D&O (directors and officers) liability insurance. Using an original data set, e presents evidence of significant inertia in te structure of suc insurance contracts (policy limit and deductible) wic tends to remain constant even in te presence of canges in firm risk and risk exposure. To set te stage for te approac to organizational inertia we will take in tis paper, let us briefly considerdifferent settings, typical of common situations in organizations. A first setting relates to te fact tat career possibilities, bonuses and promotions, are commonly linked to te successful completion of projects. If tat is so, one may expect tat better informed agents will tend to pursue a project even if tey know tat an alternative mutually exclusive project now represents a more profitable opportunity for te firm. Abandoning te initial project in favor of te alternative project will be detrimental to te agent s career. To counter suc opportunism, it will in general be necessary to jointly determine in a proper career evaluation process te rewards accruing to te agent in te two mutually exclusive projects. In an interview wit Te Economist ( ), Livio DeSimone, Cairman and CEO of 3M, stressed tat employees become less innovative if teir job security is treatened and terefore, it is a policy of 3M to give suc job security to its labor force. In order to avoid too muc inertia, e as imposed toug innovation goals (30% of annual sales must come from products less tan four years old; 10% from products introduced during te year) and very demanding organizational caracteristics (marketing folks ave direct contacts wit scientists; R&D staff are directly involved in product strategy; cross-functional teams abound). It may even be necessary in suc contexts to value and reward a recommendation to abandon a project coming from tose wo were responsible to provide te necessary efforts to acieve its

5 successful completion! Suc career concerns may of course take different forms. Prendergast and Stole (1996) develop a model of individual decision making wen decisions are a signal of people s ability to learn. Considering te manager of an investment project, tey sow tat, to appear as a fast learner, te manager exaggerates is own information and becomes unwilling to cange is investment program as new information arrives (inertia), a result tat tey relate to te existing psycology literature concerning cognitive dissonance reduction. Inertia appears also in te political cover-up of unfavorable information by agents. Suc situations occur because te efforts sunk by te agent in defending an initial position cannot be transferred to te alternative position. Te new information, on te increased benefits associated wit te alternative position and/or on te reduced benefits associated wit te initial position, may be idden or manipulated by te agent to make it appear less favorable to te alternative tan it really is. It may again be necessary, from an organizational performance viewpoint, to value and reward te failure in making te initial position a success. In a paper dedicated to advocacy, Dewatripont and Tirole (1999) argue tat competition among enfrancised advocates of special interests may improve performance (flexibility in our context) by relying on different viewpoints, eac being proposed by specific interested parties rater tan by non-partisan representatives wo may potentially be impaired by conflicting incentives. In a different context, inertia may ave positive value in situations were an independent appraisal concludes tat a partially completed project sould be abandoned because its completion will involve additional costs wic cannot be recuperated from te total future benefits to be generated by te project. Systematically applying te textbook principle bygones are bygones may lead to reduced ex ante efforts to make te initial project profitable. Te principal may find necessary, and profitable, to commit ex ante to pursue suc projects even if information, unfavorable to pursuing te project, is revealed to er. Agents wo are asked to switc to task/project B before completing task/project A (and seeing te outcomes of teir efforts) and later to switc to C before completing B may end up investing no effort in raising te probability of success of te task/project under te expectation tat it will not be completed. In a population ecology perspective, Hannan and Freeman (1984) develop an evolutionary model of inertia. Tey observe tat organizations, wen compared wit ad oc collectives, are caracterized by iger levels of reliability and accountability. Reliability refers to te 3

6 organization s capacity to produce quality products in a consistent way, wile accountability refers to te capacity to document processes in order to take corrective actions if necessary. Hig levels of reliability and accountability will be attained troug reproducible and stable structures built on tree major pillars, namely institutionalization, standardization and routines. Tose pillars also turns out to be factors of inertia. In Hannan and Freeman s model, structural inertia appears as a by-product of reliability and accountability. In a similar vein, Carroll and Hannan (00) write (p.6): Researc in corporate demograpy suggests tat an organization s istory strongly constrains its subsequent possibilities. In particular, researc sows tat te social and economic conditions at te time of an organization s founding ave a lasting effect upon its structure and operation... Overcoming tis inertia is muc more difficult tan te literature on management implies, especially for core features of an organization suc as its mission and its form of autority... [A]mple teory and evidence suggest tat te process of cange itself migt be so disruptive tat attempting radical cange elevates te risk of mortality. One may criticize tis evolutionary model of inertia insofar as reliability and accountability may very well require sufficient flexibility to adapt at proper times to canging conditions, eiter internal or external. Hence, it is possible and maybe equally likely tat organizations tat ave ig level of reliability and accountability are in fact igly flexible ones, especially in contexts were production and/or marketing conditions cange frequently or were tose conditions are igly volatile and require frequent canges in te mix of products, inputs and/or in te relative pace of development of investment projects. A striking example of suc a situation can be found in te telecommunication industry were tecnological cange and market or competition volatility require, for increased reliability and accountability, increasing levels of flexibility. For suc reasons, we feel tat te teory of inertia must be developed on firmer grounds. Suc a teory must be able to make explicit ow inertia enters into a value maximization or a survival problem rater tan as a by-product or unforeseen consequence of organizational design. In tat vein, we develop a model of inertia based on te rational maximization of te organization value. In suc a rational value maximization context, inertia appears as a way to reduce informational rents some members may be able to capture. More precisely, we consider te allocation of rents and power to initiate, bring or block cange as te result of te organizational design itself. Te relatively simple model presented in tis paper is meant to address in a formal way te 4

7 following questions: Wy would an organization give to interest groups witin te organization te incentives and power to block canges tat migt be beneficial to te overall organization? How will an organization coose to allocate rents and decision power? Wy and in wat sense does suc an allocation generate inertia? We model te endogenous determination of te level of inertia as a direct, explicit and rational coice made by te organization. We represent an organization as composed of a principal (te owner/manager/supervisor), wo is te residual claimant, and an agent (te executive/worker/supervised). Te agent is asked to exert unobservable, specific and sunk effort to increase te probability of success of an initial project. Later on, new information (a signal) is obtained on te profitability of an alternative project. Te projects being mutually exclusive, te organization must decide weter or not to abandon te initial project in favor of te alternative one. If te organization decides to switc to te alternative project, te agent is again asked to exert unobservable, specific and sunk effort to increase te probability of success of te alternative project. Finally, te outcome for te selected project is observed and sared according to te contract between te principal and te agent. Our objective is to better understand te unavoidable arbitrage between incentives and flexibility in dynamic contexts of asymmetric information and to caracterize te general features of an appropriate response to tis callenge. More flexibility to abandon te initial project to pursue te alternative project will in general be detrimental to te level of specific efforts tat te agent will be willing to exert to increase te probability of success of te initial project: ence te fundamental trade-off between ex ante incentives and ex post flexibility. Moreover, te existence of informational rents will generate distortions in te coice between project 1 and project. We consider tree different informational structures or organizational forms: te signal may be observed by bot te principal and te agent, by te principal only, or by te agent only. In eac setting, te agent s efforts, bot for te initial and te alternative projects, are unobservable by te principal. In order to induce te agent to provide te proper level of effort, te principal must reward te agent if te project undertaken is successful wile limited liability prevents te principal from financing tis reward troug a penalty in case of failure. Payments to te agent must be non-negative in all cases. Te principal must select and commit (credibly) to a compensation profile and a switcing decision rule providing te necessary incentives for 5

8 te provision of effort and te trutful revelation of te signal observed. We will sow tat te efficient organizational response to te non-congruence of te principal s and agent s self-interest will in general be to bias te decision rule towards te status quo. We terefore confirm in a formal way Rumelt s (1995) assertion of te paradox tat successful cange may require te promise of future inertia and Dewatripont and Tirole s (1996) statement tat ex ante efficiency may require a commitment to ex post inefficiency. We compare te results (incentive compatible payment profile and switcing rule) obtained in te different frameworks. We also discuss te efficient distribution of autority in an organization in te following sense. If te signal on te profitability of te alternative project can or sould, for some tecnical or economic reasons, be observed only by eiter te agent or te principal, wo sould be responsible for observing te signal and recommending cange? As we will see, te effective or real autority regarding cange need not always be retained by te principal even if se is as efficient as te agent in observing te signal. We present te formal model in section. We analyze in section 3 te bencmark case were te signal is common knowledge, in section 4 te case were te signal is observed only by te principal, and in section 5 te case were te signal is a private information of te agent. We compare te results in section 6 and we derive implications regarding te coice of organizational form, more precisely te efficient distribution or delegation of autority in an organization. Te Appendix contains te detailed proofs of te propositions and corollaries. Our results are tree-fold. First, te efficient organizational response to te presence of private information on te value of cange will in general be to rationally bias te decision rule towards te status quo. Second, te compensation profile of te agent differs according to weter te information on te value of cange is private to te principal or te agent: an upfront payment in te first case versus a distribution of stock options in te second case. Finally, te efficient distribution of real autority for recommending cange in an organization need not always be profitably retained by te principal: te agent is more likely to be endowed wit te real autority to initiate cange te smaller te expected profitability of te status quo is, te larger te expected profitability of cange to te alternative is wen te signal is bad, and te iger te cost of effort to raise te probability of success of te alternative project is. 6

9 . THE MODEL Te model we develop builds on a bare-bones abstract representation of an organization. Te results are not specific to one set or type of organizations. Te factors of inertia we identify are terefore general and likely to be present in all organizations were some features of incomplete information are present. We derive predictions regarding te level of inertia in organizations as a function of teir respective socioeconomic context. Te organization is represented by a principal andan agent,bot risk neutral. After investing in an initial project, te organization will observe a signal θ on te profitability of an alternative project. Based on te observed value of θ, te organization may coose eiter to pursue te original project 1 or to abandon it in favor of te alternative project. Te timing of observations and decisions is as follows. First, te agent invests an unobservable level of effort e wic may be low (`) orig() in te initial project 1, at a cost V1 ` and V 1 respectively. Tis investment in effort determines te probability of success p e 1 of tat project, wit p 1 >p`1.effort is specific to te project and considered as sunk. Second, te signal θ about te probability of success of an alternative project is observed: it may be good or favorable (g) wit probability ρ and bad or unfavorable (b) wit probability (1 ρ). Te projects being mutually exclusive, te organization must decide eiter to abandon te initial project in favor of te alternative one or to maintain te initial project (te status quo). If project is selected, ten te agent must again provide some unobservable level of effort e wic is eiter low (`) orig(), at a cost of V ` and V respectively. Te probability of success of project depends on effort e {`, } andonte value of te signal θ {g, b} and is given by p g, p`g, pb or failure, net profits) of te project cosen is observed and distributed. or p`b. Finally, te outcome (success Te outcomes of te projects are random. Te expected level of net profits depends on te project pursued, on te level of effort invested by te agent and on te value of θ. LetR1 e be te expected return from project 1 wen effort e as been exerted by te agent and let R eθ be te expected return of project given e and θ. Finally, let ψ denote te cost of effort per unit of efficiency in raising te probability of success, namely V 1 p, ψ g 1 p`1 V, ψ p g p`g b V. p b p`b An incentive system takes te general form of a compensation profile w specifyingapayment contingent on te project pursued (1 or ), on weter it is a success s or a failure f, andon 7

10 weter te announced value of θ is g or b: w = {w1 s,wf 1,wsg,wfg,wsb,wfb }. Limited liability requires tat w 0. A switcing rule, wic specifies wen project 1 will be abandoned in favor of project, is a pair r =(r g,r b ), were r g [r b ] denotes te probability tat te alternative project is cosen wen te value of θ observed or announced is g [b]. We wis to limit our attention to cases were bot te effort and te signal are meaningful in te following sense: wen te signal θ is common knowledge, te principal always prefers to elicit a ig level of effort for bot project 1 and project, e i =, andaswitctoproject occurs if and only if te signal is favorable, θ = g. To concentrate on tose (more interesting) cases, we restrict our attention to exogenous parameters [R1 e, Reθ, ρ, pe 1, peθ, V 1 e, V e ] suc tat:3 A1-a: p 1 >p`1, p g >p`g, pb >p`b =0, V1 ` = V ` =0. A : It is better for te Principal to induce ig effort in all projects : R1 p 1 1 ρ >R`1, R θ p θ ψ >R`θ for θ {g, b}. A3 : It is better to switc to te alternative project wen te signal is favorable : R g pg ψg >R 1. A4 : It is better to stick to te original project wen te signal is unfavorable : R1 p 1 1 ρ >Rb p b ψ. b Our objective is to explore in tis context ow information asymmetries affect te structure of incentives and te decision to undertake project. We will consider tree alternative information structures. In te first case (bencmark case), te signal θ is jointly observable by te principal and te agent; in te second case, it is observable only by te principal and in te tird case, it is observable only by te agent. 4 3 Te assumptions p`b =0andV1 ` = V ` =0aremadewitoutlossofgenerality. 4 We do not model te process by wic a new project is discovered. One possible way to model tis process in te first context is to suppose tat effort can be extended eiter to raise te probability of success p c(e c)ofte current project 1 or to raise te probability p n (e n ) of discovering a new and better project. Designing efficient scemes for total effort provision e c + e n = e at cost ψ(e) andforallocatingtateffort between te two objectives is clearly a major concern of organizational design. See for example Sinclair-Desgagné and Gabel (1997) wo claim tat a properly designed auditing procedure could induce a iger effort level e c and aigereffort level e n : te procedure would call for an audit of e c ifanewandbetterprojectisfoundwitapaymentmadeincase e c is found to be ig. Moreover, te value of e could depend on market structure as one can infer from Tirole (1988, cap. 4). 8

11 3. THE SIGNAL θ IS OBSERVED BY THE PRINCIPAL AND THE AGENT. Te effort level exerted by te agent is always a private information of te agent and terefore, te principal must offer a compensation profile suc tat it is privately beneficial for te agent to provide a ig level of effort. To acieve tis, te principal must create a wedge for eac project between te payments made in case of success and failure. Tis, togeter wit limited liability, generates an informational rent for te agent. Assuming tat te agent s reservation wage is 0, tose effort incentive conditions are: 5 w sg wfg V p g p`g w sb w fb V p b p`b ψ g (1) ψ b. () For project 1, te wedge (w1 s wf 1 ) must take into account te fact tat te project may be abandoned in favor of project once te effort cost is sunk. From te switcing rule (r g,r b ), project 1 is pursued wit probability ρ(1 r g )+(1 ρ)(1 r b ) and so te wedge must satisfy (w1 s w f 1 ) ρ(1 r g )+(1 ρ)(1 r b ) > V 1 p 1, (3) p`1 wic sows tat greater flexibility must be compensated by a iger w1 s. 5 If project is cosen wen θ = g, te incentive condition is p g wsg +(1 pg )wfg V p`g wsg +(1 p`g )wfg tat is te wedge must satisfy (1). Similarly, if θ = b, te wedge must satisfy (). Te limited liability assumption implies w fg 0, and terefore w sg ψg. Hence, we obtain tat te net payment received by te agent is no less tan p g ψg + wfg V p`g V /(p g p`g )+wfg > 0, and terefore exceeds te agent s reservation utility and te agent receives an informational rent. In te case of project 1, te agent knows tat a switc will occur to project wit probability ρr g +(1 ρ)r b. If tere is suc a switc, ten te agent will obtain a rent of p θ ψ V from te compensation profile relevant for project. But given tat ρ is independent of weter te effort put into project 1 is ig or low, te value of te appropriate rent is added on bot sides of te relevant incentive constraint for e 1. Terefore, te effort inducing wedge for project 1 depends on te probability tat a switc will occur but is independent of te rent itself accruing to te agent from te realization of project. Hence, te agent will put a ig level of effort in project 1, e 1 =, wen [ρ(1 r g )+(1 ρ)(1 r b )][p 1 (w s 1 w f 1 )+wf 1 ] V 1 [ρ(1 r g )+(1 ρ)(1 r b )][p`1(w s 1 w f 1 )+wf 1 ] tat is wen te wedge satisfies (3). Ex ante, te agent receives from project 1 an expected payment p 1 [ρ(1 r g )+(1 ρ)(1 r b )] [ρ(1 r g)+(1 ρ)(1 r b )] V 1 + w f 1 = p 1 V1 /[p 1 p`1]+w f 1 wic is also te ex post rent from project 1 if te decision to pursue project 1 is taken. 9

12 Wen θ is observable and contractible, te optimal organizational design will maximize te principal s expected profits subject to te limited liability constraints (w 0) and te ig effort incentive constraints (1), () and (3). Proposition 1 : Under A1-a, A, A3 and A4, wen te signal θ is observed by bot te principal and te agent, ten te principal cooses a compensation profile w wic induces te agenttoexertaiglevelofeffort for projects 1 and, tat is, w f 1 = wfb = wfg =0; w s 1 = /(1 ρ); w sb = ψ b, w sg = ψg and a switcing rule r =(r b,r g )=(0, 1) suctataswitctoprojectoccursifandonlyif θ = g. 6 (All proofs are given in te Appendix) We use te above as a bencmark for te following sections. 7 We will sow tat wen te signal is private information of eiter te principal or te agent, switcing to project may not always occur wen θ = g, indicating te existence of inertia. Moreover, we will sow tat, if te signal cannot be observed by bot te principal and te agent, te principal will sometimes be better off observing te signal erself and sometimes be better off letting te signal be observed by te agent only (empowerment of te agent). We will derive te condition under wic eiter case occurs. 6 Tis result olds also in te case were te signal is observed by bot te principal and te agent but is not contractible. Te principal, as te residual claimant, implements te optimal compensation profile wit r b =0 and r g = 1. Te principal as no incentive to misreport θ. If θ = g, weaver g pg 1 ψg >R 1 >R1 p 1 1 ρ and tus te principal will prefer to recommend cange. If θ = b, te principal knows tat if se recommends cange, te agent will coose a low level of effort unless (w s w f ) ψb. Since by assumption R1 p 1 1 ρ > max{r`b V `, R b p b ψ}, b it is not in er interest to recommend cange. 7 Note tat from a social welfare point of view, a switc to te alternative project sould occur ex post wen θ = g (sould not occur wen θ = b) if and only if te expected net total benefits from project, assuming tat te agent exert a ig level of effort in all cases, are larger (smaller) tan te expected gross total benefits from te original project, tat is, if and only if ( <R g V if θ = g R 1 >R b V if θ = b Under te assumptions of te model, te switcing rule r b =0andr g = 1 is also te socially optimal rule. Indeed we ave: R1 <R g pg ψg <Rg V and R1 >R1 p 1 1 ρ >Rb p b ψ b = R b V. 10

13 4. THE SIGNAL θ IS OBSERVED ONLY BY THE PRINCIPAL. We assume, in tis section, tat te signal on te profitability of te alternative project is observed only by te principal. Te principal, wo cannot commit not to use opportunistically er private information on θ, must coose a compensation profile w and a switcing rule r suc tat te agent exerts te ig level of effort expecting rationally tat te principal will reveal trutfully te observed signal and apply te announced switcing rule, given te principal s interests in telling te trut about te signal. 8 To better see te principal s problem of credibility, let us first consider te full information solution payment profile w and switcing rule r caracterized in Proposition 1. Te principal can deviate from telling te trut by manipulating er message in two ways: se can pretend tat te alternative project is bad (se reveals θ = b and project 1 is pursued) wen it is good (se observes θ = g), or alternatively, se can pretend tat te alternative project is good wen se observes θ = b. Under te assumptions of te model, te former is never profitable. Indeed, from Assumption A3, we ave: R g pg ψg > R 1 p 1 1 ρ. Te principal as no interest in pursuing te original project by claiming tat a good alternative project is bad even if er announcement is taken as trutful by te agent. On te oter and, pretending tat te alternative project is good wen se observes θ = b could be profitable. Telling te trut is less profitable tan lying wen θ = b (terefore always switcing to project ), wenever By assumption A4, we ave R 1 p 1 1 ρ R b p b ψ g >R 1 p 1 1 ρ (4) >Rb pb ψb. Terefore, if ψg ψb,tatis,if inducing ig effort is more costly for a good project tan for a bad project, condition (4) never olds and te principal as no interest in misreporting θ even if te agent believes te announcement. However, if (4) olds, tat is, if ψ g is sufficiently smaller tan ψb,wemust impose an incentive compatibility constraint on te principal s announcement in order to make 8 See Laffont and Martimort (00) for a discussion of Informed Principals. Tey write: Teliteratureon informed principals is relatively tin but complex (p. 360). We like to tink tat our paper is an interesting contribution to tis literature. 11

14 it credible. To concentrate on te more interesting case, we will assume tat (4) olds, wic implies tat: A1-b: p g p`g >pb p`b, tat is, p (e, θ) is supermodular. If inducing ig effort is less costly for a good project tan for a bad project (Assumption A1-b), te contractual relationsip between te agent and te principal must be modified to constrain te opportunistic beavior of te principal. Tis will lead to additional agency costs. More generally, given a compensation profile w and a switcing rule r, trut-telling conditions are respectively given in te Appendix by (17), wic guarantees tat te principal will not pretend tat project (te signal θ) is bad wen it is good, and (18), wic guarantees tat se will not pretend tat project is good wen it is bad. Te principal s problem is now to maximize er profit subject to te effort incentive conditions and te two additional constraints (17), wic will turn out to be non-binding, and (18), wic may be binding and can be rewritten as: ³ ³ [R1 p 1(w1 s w f 1 )+wf 1 ] [R b p b (w sg wfg )+wfg ] (5) Te way to insure tat 5 is satisfied at te lowest cost to te principal is to set w fg above 0, i.e. give te agent a fixed lump sum payment wenever Project is undertaken. Tis acts as a golden paracute, a compensation given to te agent wenever Project 1 is abandoned. In order to satisfy (4), we must set: w fg =max{0, (R b p b ψ g ) (R 1 p 1 1 ρr g )}. (6) Te implication of tis is tat (i) te principal must bear extra agency costs, (ii) since tese costs are increasing and convex in r g,itmay be optimal to coose r g < 1, representing a distortion (inertia) from te complete information switcing rule caracterized in Proposition 1, tat is, a bias towards te status quo. We can illustrate tose results as follows (see Figure 1). Te principal s expected profit Π(r g )canbewrittenas were C P (r g ) is te expected payment to te agent, Π(r g )=(1 ρr g )R 1 + ρr g R g CP (r g ) (7) C P (r g ) = p 1 + ρr g p g ψg (8) 1

15 +ρr g max ( (R b p b ψ g ) (R 1 p 1 1 ρr g ); 0 Let us define r g P as te value of r g for wic (R b pb ψg )=(R 1 p 1 1 ρr g ). 9 One can verify tat te function Π(r g ) is increasing and linear for r g r g P but concave for r g r g P,wit akinkat r g P. Te difference between C P (r g ) and its linear part p 1 + ρr g p g ψg is te extra agency cost necessary to credibly convey tat te principal will not misreport θ. 10 Te optimal level of flexibility is unique and always lie in te interval [ r P g, 1]. 11 Te optimum r P g may be eiter at te kink r g P (wic may be at 0) of te expected labor cost curve, or at te tangency point between te convex portion of te labor cost curve and an isoprofit curve, or at te end point 1. One can verify tat r P g < 1 if and only if te following two conditions are met. R b p b ψ g R1 p 1 R b p b ψ g ³ R g pg ψg (1 ρ) p 1 (1 ρ), ). and tat complete inertia, tat is r P g = 0, is optimal if and only if More generally, we obtain te following: (R b p b ψ g )+p 1 R g pg ψg 0. Proposition : Wen te signal θ on te profitability of te alternative project is observed only by te principal, we ave te same incentive sceme except tat w fg is given by (6). Because te extra agency costs so generated are increasing and convex in te probability of undertaking te alternative project, it may be optimal for te organization (te Principal) to sometimes stick to te original project (inertia) even if te signal of profitability of te alternative project is good. Te optimal flexibility level is maximal if te principal finds no value in misreporting te value of te signal θ even if se could do it witout cost (and w fg = 0). Wen se as to bear extra costs to make er announcement credible, se may coose rg P < 1 in order to save on te 9 Let r g P =0wenR b p b ψ g >R 1 p 1 and r g P =1wenR b p b ψ g <R 1 p ρ Trut telling condition (18), under wic te principal reveals trutfully θ wen θ = b, is binding if rg P > r g P. 11 As sown above, te slope of te isoprofit curvesisρ(r g R 1 ). Te slope of te expected labor cost function to te left of r g P is ρp g ψg. By A3, te latter is smaller. 13

16 informational rent. In some cases, te best te principal can do is to never abandon project 1 (maximal inertia): te incentives for te principal to always pretend tat project is good wen θ = b are so strong tat it becomes too costly to credibly convey tat project is good. Complete inertia is ten better for te principal and is implemented in te organization. 1 Finally, we ave: Corollary.1 : Te level of inertia in an organization (wen te principal is te only one being informed of te value of te alternative project) is positively related to p 1 p 1 (p g pb )ψg pg p g of inertia pb p`g V and negatively related to te difference (Rg Rb V1 p 1 p`1 and to ). Tat is, te level increases wit V1 and V, te cost of effort, wit Rb, te profitability of project wen effortisigbuttesignalisunfavorable (tat is because a larger R b increases te incentive to lie wen θ = b and terefore makes it more difficult and costly for te principal to credibly convey tat project is good); decreases wit (p 1 p`1 ),teefficiency of effort in increasing te probability of success of project 1, wit (p b p`b ),teefficiency of effort in increasing te probability of success of project wen te signal θ is unfavorable, wit (p g p`g ),teefficiency of effort in increasing te probability of success of project wen te signal θ is favorable if p b <p`g, wit R g and R1, respectively te profitability of project wen effort is ig and te signal is favorable and te profitability of project wen effort is ig (tat is because tey bot make it easier for te principal to credibly claim tat project is good). 1 If one interprets te original project as a bold cange from a previously pursued strategy, te complete inertia result may be understood as necessary to induce te optimal level of effort by te agent to raise te probability of success of tis original cange in strategy: no more cange will be made watever te new information (θ) to be observed or gatered in te future. 14

17 5. THE SIGNAL θ IS OBSERVABLE ONLY BY THE AGENT. We consider now te case were te signal on te profitability of te alternative project is observed only by te agent. Under te common knowledge (of θ) incentive sceme, te agent may pretend tat te alternative project (te signal θ) is bad wen it is in fact good. Tis would be profitable for te agent if always sticking to te original project (lying wen θ = g) is better for im tan telling te trut and engaging in te new project wen θ = g, tatis,if and only if: p g ψg V <p 1 (9) 1 ρ He can also pretend tat te alternative project is good wen it is bad (θ = b) and, since te original project would never be completed, perform no effort on te initial project. Tis would be profitable if and only if: and p 1 1 ρ V 1 (1 ρ) max e {`,} (peb ψ g V e ). (10) Under our assumptions, bot (9) and (10) may be satisfied under te complete information compensation sceme and switcing rule (Proposition 1): te rent obtained from pursuing a good project may not be ig enoug to induce te agent to abandon project 1 wen e as invested (sunk cost) a ig level of effort in it, wile at te same time be ig enoug for te agent to prefer exerting no effort in project 1 and always recommending a project cange. In order to guarantee trutful revelation, te principal must provide te incentives necessary to prevent opportunism by te agent. Again, tis will generate extra agency costs. In order to identify tose, we consider te general principal-agent problem. As usual, we must impose trut telling incentive conditions to guarantee tat te agent will not always claim tat θ = b, tereby generating too muc inertia, or always claim tat θ = g, tereby generating too muc flexibility at te expense of too little effort invested in project 1. We sow in te Appendix tat, under A1-a, A1-b, A, A3 and A4, we must ave w f 1 = wfg =0andr b =0. Furtermore, te incentive constraints can ten be written as: p g wsg V p 1w1 s (11) p 1w1 s max{p b w sg V, 0} + max{v 1 (1 r g)(p 1 p`1 )ws 1, 0}. (1) (1 ρ)r g 15

18 It follows tat te solution to te principal s problem can take tree different forms (cases) depending on weter te above trut telling conditions are binding or not. Proposition 3A : If p 1 ψ1 1 ρ <pg ψg V, neiter (11) nor (1) are binding. Hence, te principal can do as well as if te signal θ were observable. We ave w f 1 = wfg =0, w1 s = /(1 ρ), w sg = ψg, r b =0, r g =1. Condition 11 is satisfied by construction. Since te incentive are not sufficient to induce ig effort if project is bad, we ave max{p b wsg V, 0} = 0 and V 1 /(1 ρ) p 1 ws 1 = V1 /(1 ρ)(p 1 p`1 ). It follows tat condition 1 is also satisfied. Te intuition is tat te rent accruing to te agent wen te alternative project is good (θ = g) is ig enoug tat e will not lie about θ. In tis case, tere are no agency costs associated wit te fact tat θ is observable only by te agent. Proposition 3B : If p g ψg V < p 1 ψ1 1 ρ <pg (1) is not. We ave w f 1 = wfg =0, w1 s = 1 ρr g, p g wsg ψ b + p 1 ψ1 V p b,ten(11) is binding but i p = 1 1 ρr g + V, r b =0. In order to prevent te agent from always claiming tat project is bad, te principal must offer a w sg iger tan te level oterwise necessary to induce effort in project. As long as max{p b wsg V, 0} remains small enoug, condition (1) remains non-binding. Te condition of Proposition 3B guarantees tis. Te extra rent necessary to elicit trutful beavior from te agent is ten given by ( ) ρr g p g (wsg ψg )=ρr g max 0,p 1 p g 1 ρr ψg + V. (13) g It is increasing and convex in r g, tereby implying a bias towards te status quo. Teoptimal r A g solves: ³ max (1 ρr g)r1 + ρr g R g V p 1 r g [0,1] ³ In particular, inertia (rg A < 1) appears if R g V <R1 + p 1 1 ρr g. (14). (1 ρ) 16

19 Proposition 3C : If p g values w sg and w s 1 equality. V <p 1 ψ1 1 ρ, ten bot (11) and (1) are binding. Te ψ b + p 1 ψ1 p b are ten cosen suc tat te bot constraints (11) and (1) oldwitstrict Te organizational design must now prevent te agent from exerting no effort in te initial project and always pretending tat te alternative is good, and at te same time prevent im from exerting ig effort in te initial project and always pretending tat te alternative project is bad. Te former is acieved by increasing w1 s (11) and (1), we obtain: w1 s = max, 1 ρr g = p 1 ws 1 + V w sg ψ g and te latter by increasing wsg. Solving for ψ g ³ V 1 V p g pb (1 ρ)r g (p g pb )p 1 (1 ρ)r g +(1 r g )(p 1 p`1 )pg Te principal s problem is in tis case more complex to analyze. One can sow tat te above extra agency costs are increasing but not necessarily convex. We will not develop furter tis case. In order to give te agent te necessary incentives to report trutfully te value of te signal θ, it may be necessary to increase te payment w sg above ψ g (Case ): te agent gets a better deal wen te alternative project is a success and since w sg wfg is also increased, te agent is overinduced to provide a ig level of effort. Te increase in w sg is similar to issuing, in addition to te provision of incentives for effort, extra stock options on te success of project. However, tese adjustments may imply tat te agent as now an interest in reporting tat project is good wen in fact it is bad. To make sure tat tis does not appen, it may be necessary to increase bot w1 s and wsg (Case 3): te agent is ten properly induced to reveal te observed θ (extra stock options are issued for bot projects 1 and ) and is overinduced to exert a ig level of effort in bot projects 1 and. Hence, te incentive intensity for effort is stronger wen te agent is responsible for observing te signal θ and recommending cange. Tis result is reminiscent of Milgrom and Roberts (199, cap. 1) discussion of te complementarity between discretion and incentives. 17

20 6. CHOOSING THE ORGANIZATIONAL FORM: DELEGATING AUTHORITY OR NOT We ave seen tat te optimal incentive system depends on te information structure caracterizing te organization: observing te signal on te value of te alternative project or course of action provides te observer wit te power to recommend or initiate cange. Te compensation package differs significantly in te tree cases considered: (i) bot te principal and te agent observe te value of te alternative project, (ii) only te principal does it, or (iii) only te agent does it. It will typically be te case tat te agent s compensation profile w is caracterized by te minimal effort-incentive compatible pay in case (i), by an additional upfront compensation payment if cange occurs in case (ii), and by an additional stock options package in case (iii). Moreover, te level of inertia as measured by (1 r g )willdiffer: r g isequalto1incase(i),to r P g in case (ii), and to r A g in case (iii), wit r P g and r A g typically less tan 1. In tis section, we raise te following question. If te principal could decide on te informational structure, at some differential cost, wic one would se coose? Clearly, if te cost is te same for all structures, ten te principal would coose te organizational form under wic bot te agent and er observe te signal θ in order to minimize total agency costs by avoiding te extra payments te oter structures imply. For te remaining of tis section, we will assume, for matter of simplification only, tat te cost of aving bot observe te signal is proibitively ig compared to te cost of te oter two information structures, wic we will simply assume to be equal to zero. Hence te coice we want to caracterize is between an informed agent and an informed principal, 13 te informed party being ten responsible for recommending te pursuit of project 1 or a switc to project. Sould te principal (te legal residual claimant) exercise er formal autority to decide on cange or sould te real autority be delegated to te agent? We already sowed tat: Proposition 4 : If p g p`g pb p`b i, te principal as no interest in misreporting te value of θ and terefore sould retain te responsibility to become informed and te power to recommend weter to switc or not to te alternative project. 13 Te level of effort always remains a private information of te agent. 18

21 However, if p(e, θ) is supermodular (i.e. p g p`g >pb p`b : assumption A1-b), ten bot te retention of autority by te principal and its delegation to te agent present problems. Te agent as vested interests in te pursuit of project 1 and tere is no reason to believe tat is interests coincide wit tat of te principal or te organization as a wole. On te oter and, te principal as residual claimant may beave opportunistically in order not to pay te rent promised to te agent were project 1 pursued and succeeded and/or in order to fool te agent in putting ig effort in an alternative bad project. In bot cases, agency costs may ave to be incurred to control suc opportunistic beavior. Because agency costs are increasing at an increasing rate wit te level of flexibility wit wic te organization can switc ex post to te alternative project, te principal may want to embed some inertia in er organization in order to reduce tose agency costs. Comparing te agency costs in te different contexts, we obtain: Proposition 5 : Wen p g p`g >pb p`b i, te principal finds it profitabletogivete autority to te agent if and only if, watever te probability of switcing r g cosen, te agency cost incurred are smaller: (R b p b ψ g ) (R 1 p 1 ) >p 1 1 ρr g 1 ρr g ³ p g ψg V (15) tat is, ³ (R b p b ψ g ) R 1 > p g ψg V. (16) Te left-side of te (15) corresponds to te extra benefit te principal gets wen project 1 is abandoned in favor of a bad project wile te agent believes tat project is good; te rigt-side corresponds to te extra rents te agent gets wen project 1 is pursued rater tan a good project. If te former is larger, it is preferable to delegate autority to te agent. Tis proposition 5 gives a nice and simple caracterization of te desirability for te principal of empowering te agent wit te autority to recommend cange. It solves in a sense te stage 0 of organizational design, tat is, te stage wen information structures are determined or cosen. Altoug in most if not all principal-agent model, te (asymmetric) information structureisconsideredasgiven,itneednotbeinmanyrealcases. Weliketotinktat 19

22 proposition 5 is a step in tackling te task of better understanding asymmetric information structures as te outcome of some rational efficiency calculus. Using te expression for ψ g, we can rewrite (16) as and terefore, R 1 R b + pb p`g p g p`g V < 0 Corollary 5.1 : Te agent is more likely to be endowed wit te responsibility of observing te signal θ and wit te real autority of recommending cange, (i) te smaller R 1 ;(ii)telarger R b; (iii) te smaller ψg, tat is, te smaller V raising te probability of success of a good project,(p g and/or te larger te efficiency of effort in p`g ), if and only if pb >p`g. Furtermore, one can verify from (6) and (7) in te Appendix tat CP (r g ) r g CA (r g ) r g only if (15) olds. Terefore: if and Corollary 5. : Wenever it is preferable for te principal to delegate te real autority to te agent [retain suc autority], we ave rg P [ ] rg A : te autority is granted to te party wo ultimately allows a more flexible organization. Te result of Proposition 5 is illustrated in Figure 1 were te C A (r g ) function (6) is sown togeter wit te C P (r g ) function (7) and te isoprofit curvesofπ(r g,c). In Figure 1, we suppose tat condition (15) is satisfied and it is terefore preferable to empower te agent wit te responsibility of observing θ and te real autority to recommend cange. In a recent paper, Agion and Tirole (1997) sow tat te allocation of formal autority in organizations, tat is, te allocation of rigts to decide, may differ significantly from te allocation of real autority, tat is, te allocation of effective control on decisions. Tey consider different ways to credibly increase te subordinate s or agent s real autority in a formally integrated structure wit te supervisor or principal keeping te legal rigts to decide: te work overload of supervisors, te design of lenient discipline rules for deviant beavior by te agent, te timing of background studies leading to an urgency of decision, te repeated 0

23 interactions leading to te principal s reputation for non-intervention, an improved performance measurement, and finally te splitting of decision rigts between multiple superiors. 7. CONCLUSION Using a simple model, we sowed tat a principal may find profitable to limit er flexibility to initiate cange by giving te agents some power to block canges tat se would like to undertake. We sowed tat inertia (bias towards te status quo) can be optimal from an ex ante point of view as a means to reduce informational rents. We ave sown tat, wen te principal retains te autority to initiate cange, it will typically be te case tat te agent s compensation profile calls for an upfront compensation package if a cange does occur wile, wen te principal empowers te agent wit tat autority, it calls for a stock options package. Wen te information structure is itself endogenously determined, te principal may sometimes be better off coosing an organizational form under wic se becomes te informed party and retains te autority to recommend or implement cange and sometimes be better off coosing an alternative organizational form under wic te agent becomes te informed party and is empowered wit initiating cange. Even if te principal is te informed party, te agent can capture an informational rent because te principal needs to make er announcement credible in order to elicit effort from te agent at minimal cost. Inertia emerges endogenously in bot cases as a way to reduce te agent s rents. Te current discussion and arguments for flexibility in production, uman capital, financial structure and contracts, and more generally in organizations, neglect te fundamental dynamic trade-off wic we caracterized between flexibility and incentives and wic is likely to be present in many situations. 1