Intentional introductions of nonindigenous species: a principal-agent model and protocol for revocable decisions

Ecological Economics 34 (2000) 333–345

METHODS

Intentional introductions of nonindigenous species: aprincipal-agent model and protocol for revocable decisions

Michael H. Thomas a,*, Alan Randall b

a Agribusiness Program, College of Engineering Sciences, Technology and Agriculture,Florida Agricultural and Mechanical Uni6ersity, 302 Perry Paige South, Tallahassee, FL 32307, USA

b Department of Agricultural, En6ironmental and De6elopment Economics, The Ohio State Uni6ersity, 2120 Fyffe Road,Columbus, OH 43210, USA

Received 8 October 1999; received in revised form 14 February 2000; accepted 11 April 2000

Abstract

Alien or genetically altered species, purposefully released to generate various benefits, may contribute to unantici-pated damage to the delicate balance of an existing ecosystem. In an ideal world, harm can be avoided in either oftwo ways: (1) perfect ex ante information would allow the choice of only beneficial releases; and (2) perfectrevocability would allow ex post revocation of any release that turned out to be harmful. Currently, standard decisionprotocols regulating releases depend heavily upon ex ante information, which is often costly and uncertain due tohighly complex ecosystems. We propose a more balanced approach that combines imperfect ex ante information withimperfect revocability. A principal-agent model is used to address moral hazards affecting purposeful releases. Amodel protocol is sketched to implement the concepts developed in this paper, paying particular attention toincentives that encourage releasing agents, to provide the optimal degree of revocability. © 2000 Elsevier Science B.V.All rights reserved.

Keywords: Invasive species; Intentional introductions; Principal-agent model; Decision theory; Revocability; Release protocol

www.elsevier.com/locate/ecolecon

1. Introduction

While managing complex ecosystems, deliberateactions taken (or permitted) by public agencies orrent seeking agents may pose a risk of creating

potentially large and uncompensated economiclosses to unwitting third parties. An examplemight be the decision to introduce anonindigenous1 (exotic, alien or genetically al-

1 While commonly termed exotics in many ecological jour-nal, the term nonindigenous has gained prevalence followingthe Nonindigenous Aquatic Nuisance Species Control Act of1990.

* Corresponding author. Fax: +1-850-5612441.E-mail addresses: [email protected] (M.H.

Thomas), [email protected] (A. Randall).

0921-8009/00/$ - see front matter © 2000 Elsevier Science B.V. All rights reserved.

PII: S 0921 -8009 (00 )00189 -0

M.H. Thomas, A. Randall / Ecological Economics 34 (2000) 333–345334

tered) species in order to enjoy various expectedbenefits. Yet, such an action may disrupt thedelicate balance of an existing ecosystem, causingunanticipated harm to third parties and, in theextreme, ecological devastation. Economic agents/managers facing this type of difficult decision willoften try to gather all available information aboutthe host ecosystem and the nonindigenous species,to predict the effects of the introduction andminimize exposure to possible bad outcomes.

As an aid in the decision process, professionalorganizations, such as the American Fisheries So-ciety, have developed several protocols over thepast 25 years to guide release efforts (Courtenayet al., 1984; Kohler and Stanley, 1984; Kohler andCourtenay, 1986). These protocols share the prin-cipal characteristic of relying heavily upon thecollection of information useful for predictinglikely outcomes prior to a release. This approachto information collection might mean studyinghow the nonindigenous species fits within its orig-inating ecosystem with the purpose of applyingthis knowledge to predict how the potential hostecosystem will function with the newly addedspecies. With complex ecosystems, however, thispre-release or ex ante information can face twoburdens: (1) it can be costly to gather; and (2) itmay turn out to be unreliable for predictive pur-poses. The poor record of existing protocols inpreventing harmful introductions may be partiallydue to their heavy reliance on ex ante informa-tion. The burden of fully understanding and pre-dicting the likelihood of outcomes prior to releasehas led to reluctance by many decision-makers tofollow any release protocol. These agents may befrustrated for several reasons, not least of which isthe difficulty of deciding when there is enoughinformation to proceed responsibly with a release.

In principle, the avoidance of bad outcomes canbe accomplished in either of two ways: (1) assem-ble perfect ex ante information so that all strate-gies with bad outcomes can be avoided; or (2)only take actions with revocable outcomes, per-mitting the decision-maker to look at outcomesafter-the-fact and reverse any action that leads toharm. While the process of revoking a decision isex post, it still depends upon the ex ante selectionof only those actions that can be reversed and

outcomes revoked. In the real world, neither per-fect ex ante information nor perfect revocability islikely to be attainable. Instead, one might expectincreasing costs and diminishing returns from thepursuit of perfection with either approach. Wepropose to use this insight to develop an im-proved release protocol that balances the stress onex ante information, evident in current protocols,with more attention to revocability.

We proceed by introducing the revocabilityprinciple and comparing it to the decision strategyof depending upon perfect ex ante information.Following this discussion, we examine the princi-ple of revocability within the framework of aprincipal-agent model and analyze the incentivesfor optimal provision of revocability under vari-ous assumptions. The results focus attention on exante assignment of liability and the use of assur-ance bonds to provide effective insurance againstharmful outcomes. Finally, we apply these princi-ples to design an improved protocol for purpose-ful releases. For a detailed narration of theprotocol, see Appendix A.

2. Analysis

2.1. Re6ocable decisions and ex ante information

One common decision heuristic is a strategyknown as irreversible hill-climbing (Glover, 1989,1990b). This strategy is based upon pursuing anobjective outcome — much like a climber wouldapproach a hill — only move up towards theobjective outcome (hill-top). Unfortunately, asmany climbers will attest, following such a sim-plistic strategy can readily result in a dead endridge and a sub-optimal outcome. The only-move-up strategy prohibits backtracking to seek a dif-ferent path with potentially different outcomes forimprovement, making it unlikely to consider all ofthe potential outcomes.

A slight generalization of this approach isknown as reversible hill-climbing, which relaxesthe search strategy of only-move-up and allowsdecisions and their outcomes, originally passedby, to be saved as future alternatives (Glover,1990a). Continuing the metaphor of hill-climbing,

M.H. Thomas, A. Randall / Ecological Economics 34 (2000) 333–345 335

while remaining uninformed of the surroundingenvironment, the decision-maker is allowed tomap a course ex post without regard to cost andbacktrack to earlier decisions when it appears tobenefit the quest for a superior outcome. Whenallowed to reverse any previous decision, eventu-ally all possible outcomes are considered, assuringan eventual success. The hill climber will alwaysreach the goal outcome. This approach may betermed the revocation of outcomes, or the revo-cability principle.

To this point, it has been assumed the decision-maker can clearly identify all potential actions. Inboth cases collecting ex ante information couldhelp elucidate either the likelihood of an outcomeor subsequent decisions, alternatives and out-comes. A fully informed ex ante decision wouldrequire the complete understanding of all decisionalternatives and their ex post outcomes: perfectprescience.

Any search approach would become more effi-cient by allowing the collection of ex ante infor-mation about the remaining outcomes. In the caseof reversible hill-climbing (revocability principle),the use of ex ante information would allow for amore efficient trip to the hilltop by providing atleast a rough road map allowing the decision-maker to avoid backtracking to some obviousdead-end ridges. With irreversible hill-climbing,the only means to obtain the single optimum(hilltop) with certainty is to collect all ex anteinformation and become fully informed aboutevery remaining alternative and outcome, devel-oping a refined and detailed road map to thehighest hill. Unlike the reversible hill-climbingapproach, where actions can be revoked, the fullyinformed approach requires ex ante evaluation ofex post outcomes before actions are taken.

Based upon these two hill-climbing approaches,ex ante fully informed and ex post revocability, toresolve the question of potentially damaging orcatastrophic outcomes, the decision-maker canview the act of collecting information in twodistinctively different ways. The obvious first ap-proach is ex ante to learn incrementally all there isabout the alternative actions, and when all ex postoutcomes are known with certainty, then beginthe process toward the desired outcome with due

deliberation. If the goal is to improve utility whileavoiding a catastrophe, then an acceptable resultmight be any outcome that avoids the catastro-phe. With fully informed decisions, however, thedecision-maker could choose the global optimumand thereby do better than simply avoiding thecatastrophe.

The second approach is that of taking onlyactions that have revocable outcomes. If a deci-sion-maker can look at the outcomes after thefact and revoke those actions that led tocatastrophic outcomes before they come tofruition, the catastrophe is avoided. While theprocess of revoking a decision is ex post, it singu-larly depends upon the ex ante selection of onlythose actions that can be reversed and outcomesrevoked.

The key difference between these two ap-proaches then becomes revocability of outcomes.Within the revocability framework, the process ofex post evaluation entails a sequential movementthrough a series of actions, with the assurances ofgood record keeping and revocation to allowbacktracking when needed. The pathway of ac-tions can always be reversed with a return toeither a more desirable previous outcome or theavoidance of a pending catastrophe resulting fromthe most recent action.

2.2. A complex ecosystem

When releasing a nonindigenous species into alittle understood and enormously complex ecosys-tem, outcomes are uncertain and difficult to pre-dict. One way to view these uncertain outcomes exante is to employ the rules of probability. To theoutside observer of this complex system, the exante outcome set that could result from selectingaction i would be the n array of possible outcomesYi, j, j=1, n, and their associated probabilities p̂i, j

where Sp̂i =1. After an action is taken by thedecision-maker, an unknown environmentalmechanism selects an outcome from the outcomeset. This ex post outcome is a single vector ofevents with probability of pi,k=1, if outcome koccurs ex post, with the other n-1 outcomes notoccurring.


2.3. Moral hazard and a principal-agent modelfor social optimum

The second major issue regarding the release ofnonindigenous species concerns the distribution ofpotential losses resulting from harmful outcomes.If the releasing agent is able to capture many ofthe benefits while avoiding most of the costs of‘releases that go bad,’ public harm may, in theworst cases, be widespread and large. The poten-tial for moral hazard exists when a releasing agent(whether a private agent or a public agency) isable to avoid responsibility for harmful outcomesfollowing a release. The avoidance of negativeconsequences provides little incentive for releasingagents to exercise responsibility in their conduct,such as taking only revocable actions. It becomesa useful exercise to employ a principal-agentmodel to investigate the problem of moral hazardin the provision of revocable actions by a privateagent (Pauly, 1968; Arrow, 1970). This approachpermits a static analysis of marginal conditionsand a review of incentive compatible behavior.2

Begin by defining Ui(Y) as the decision maker’sutility resulting from the action of selecting alter-native i from an m array of choices. Alternatively,if action k is selected, its outcome will generateutility Uk(Y). If one assumes well-behaved prefer-ences, i.e. completeness, reflexivity, continuity,strong monotonicity, local nonsatiation and strictconvexity, then the outcomes from different ac-tions can be compared and ranked ordered bypreference.

Next, define the principal as a risk-neutral pub-lic agency with oversight authority concerning therelease of non-indigenous species. As the decision-maker, the principal has determined that the actof species introduction is likely to be beneficial,yet could potentially lead to a damaging orcatastrophic outcome for society. The principal is

benevolent and determines that a revocable deci-sion rule would be beneficial (i.e. improve ex-pected net social welfare). The principal contractswith a risk-neutral agent to provide the benefits ofspecies introductions, but also protect societyfrom large social losses by insisting that the agentretain the option for a socially optimal level ofrevocability. Furthermore, assume that the optionof revoking the outcome has positive costs andthat reversing a decision would result in eliminat-ing the undesirable outcome and the loss of anypotential benefits that would have accrued if theaction were allowed to stand. Unless the act ofrevocation and the foregone opportunity entailedare truly costless, it is inappropriate to treat revo-cability as a free good.

Assume an agent is considering the release of anonindigenous species. This release can generate aprivate benefit X for the agent and possibly alarge (yet reversible) social loss S where S\X ;the probability of S, P(S)=u. Let R be defined asan option to revoke the release of this nonindige-nous species. Assume that success in revoking theoutcome is random with probability P(Y)=r.

Assume a cost, c(r, u), for both retaining theoption to revoke and for actually revoking anoutcome. This cost will vary positively with r andu. The first component includes expenditures andtransaction costs necessary to assure the success-ful revocation of an outcome prior to a release.The second component includes expenditures andtransaction costs necessary to revoke the out-come, assuming the revocation option is used.Assume the agent may earn X independent of theoccurrence of S, and forfeits X and avoids S withcertainty only by both choosing to revoke theinitial outcome and observing a successful revoca-tion following the occurrence of S, with probabil-ity P(X=0)=ur. The agent then observes X withprobability P(X)= (1−u)+u−ur, which sim-plifies to P(X)=1−ur. Furthermore, observethat the social loss S cannot exceed the cost ofcertain revocability of a certain harmful outcome,or S5c(r=1, u), because certain revocabilityprevents S with certainty.

For the initial model, assume that both princi-pal and agent share the same benevolent desire tomaximize X and avoid S by using revocability

2 We acknowledge the anthropocentric stance of a principal-agent model which explicitly seeks an anthropo-optimal out-come rather than an outcome respecting both anthropocentricand ecocentric values. However, we believe that the insightgenerated from this particular approach is significant andtherefore, warrants consideration as a valid approach to theimportant ecological issue addressed herein.


only when necessary. There is no assumption ofmoral hazard. Imposing the appropriate probabil-ities of revocability, social loss and private gain,the expected objective function and resulting firstorder condition become,

Maxwrt r

E( U) = (1 − u r) X − c(r, u)

− u (1 − r) S, (1)

and

c %(r*, u) = u(S − X)\ 0, (2)

respectively, with second order condition,

− c¦(r*, u)B0 or c¦(r*, u)\0. (3)

The socially optimal level of revocability r* oc-curs where the marginal cost of revocability isequal to the expected net loss from the release.With positive marginal costs, revocability is not afree good.

Differentiating the first order condition withrespect to u results in,

dr*du

=S − X

c¦(r*, u)\0. (4)

As the probability of damage or catastrophe in-creases, the level of r* will likewise increase.

Moving to a comparison of S and r*, differenti-ating the first order condition with respect to Sresults in,

dr*dS

=u

c¦(r*, u)\0. (5)

When the size of social damage increases, therewill be increased efforts to raise the probability ofrevocable outcomes. Together, as u and S in-crease, r* will likewise increase to provide thesocially optimal level of revocability.

Now, consider private action with moral haz-ard. To accomplish this, permit the agent to ap-propriate all of the benefit X and choose the levelof revocability r̃.

Furthermore, assume that with probability u

(as before) an agent i could lose si. Initially, definethe private loss as less than the social loss of theaction; siBS. From Eq. (2) and Eq. (4) when,

siBS, then r̃Br*, (6)

and when,

X\si, then r̃=0 (7)

To address the problem of moral hazard, allowthe principal to assign ex post liability on releas-ing agents. Parties injured by an action are al-lowed to make claims against responsiblereleasing agents after the injury. Assuming zerotransaction costs under litigation, wealthy agentscould lose considerably more than their directprivate loss si, potentially up to their current netwealth wi. From Eq. (5), as

wi�S, then r̃�r*, (8)

and when

wi]S, then r̃]r*. (9)

However, for some agents net wealth may besignificantly smaller than the social loss. In suchcases, bad outcomes could bankrupt the agentand leave society with the loss largely uncompen-sated. The agent may feel that with little privatewealth to lose, it is worth a gamble of possiblylosing si for the certain benefit X. From Eq. (7),when

X\wi+si, then r̃=0, (10)

and when

XBwi+si, then r̃\0. (11)

With the ex post assignment of liability, thiswould leave agents with small net wealth eitherunder- or uninsured and under-providing revo-cability. If the loss S should occur, society wouldbe left with a large uninsured loss. With ex postassignment of liability, there is no guarantee thatagents of modest wealth will not gamble withlarge social losses. The avoidance of this poten-tially large loss to society is the motivation for theprincipal’s enforcement of ex ante liability. Theprincipal is motivated to require the private agentto post an ex ante assurance bond at least as largeas the worst-case loss to society. The avoidance ofthis potential loss and likely transaction costsassociated with large losses will likely motivatethe agent to be careful in its actions and producethe socially acceptable level of revocability, r*.


2.4. A principal-agent model with assurancebonding

Now consider the possibility where an agent isallowed to choose the level of revocability r̃, but isrequired to guarantee compensation B in theevent the potentially large social loss S occurs,e.g. insure society against losses up to B byproviding an assurance bond. The use of assur-ance bonding as a market incentive for privateagents to provide an efficient level of care in theiractions has seen wide application in the literature(Solow, 1971; Mill, 1972; Costanza and Perrings,1990; Macauley et al., 1992).

Assurance bonding is imposed on the releasingagent to insure it can indemnify other agents forany losses that may accrue to them as a result ofthe release. To implement assurance bonding, theprincipal requires the releasing agent to post abond that covers worst-case losses caused by theaction. The principal could require the introduc-ing agent to post a bond to cover all perceivablecosts necessary to return the host environmentand affected parties to their pre-introductionstatus, should that be necessary. This could meansetting bonds equal to the cost of full revocabilityc(r=1, u), if environmental restoration is re-quired, or the lesser of the loss S and c(r=1, u),if simple compensation of the affected parties isallowed. The firm providing the bond could beexpected to use all available ex ante informationabout S, u and c, plus whatever is known aboutthe agent’s previous performance in similar enter-prises, to determine the coverage availability andcollateral necessary for the bonding services.

Where information is readily available and con-sequences of actions are easily predicted, thebonding process amounts to an exercise of deter-mining the potential losses and allowing the riskmarket to work. With uncertain and potentiallycatastrophic consequences, however, a principalcould play a more prominent role in the decisionprocess. This role could extend from research anddissemination of important information, in orderto improve the functioning of risk markets, todetermining the amount of the bond, distributingthe risk and equity within society and making thedecision to permit introduction or not (Costanza

and Perrings, 1990). As these actions may entailsignificant transaction costs, the principal couldrecover this expense through one or more av-enues. For example, the principal could levy anapplication fee on all agents proposing anintroduction.

Returning to the initial principal-agent model,redefine the principal as an oversight agency thatwill permit or deny the agent to introduce non-indigenous species and accumulate X the benefitsof the release. The agent will provide the optionof revocability R, a random variable with proba-bility r and with positive cost c(r, u). The proba-bility of a large social loss S is u. Again, S isassumed to be significantly larger than the ex-pected value of X. Should S occur, the agent mustpay the minimum of S or the assurance bond Bthat it provides as collateral against large losses.Furthermore, assume that the bonding agent canobserve the level of revocability provided by theagent, as measured by r, and avoid the problem ofmoral hazard in the provision of revocability.

The contract then requires that the principalchoose the level of the bond B and the profitmaximizing agent then chooses the level of revo-cability r̃. As before, neither X nor S will berealized if an action is revoked. A decision torelease invokes three possible outcomes for theagent:1. Revocable damages: −c(r, u), with probabil-

ity ur.2. Irrevocable damages: X−c(r, u)−min[S, B ],

with probability u(1−r).3. No damages: X−c(r, u), with probability

(1−u).The expected profit function for the agent

becomes

Maxwrt r

E(p) = (1 − r)

u {X − c(r, u) − min[S, B ]}

+ (1 − u)(X − c(r, u))

+ u r [−c(r, u)], (12)

the resulting first order condition is

c %(r̃, u) = u(min[S, B ] − X)\ 0 (13)

and the second order condition is

− c¦(r̃, u)B 0, or c¦(r̃, u)\ 0. (14)


This establishes that the firm’s optimal level ofrevocability r̃ occurs where the marginal expectedcost of revocability is equal to the expected netsocial loss of the release if S\B or the netexpected private loss after forfeiting the bondwhen B\S.

As before, the static conditions establish posi-tive relationships between revocability and theprobability of damaging social losses and the sizeof those expected losses. Differentiating c %(r̃, u)with respect to u results in,

dr̃du

=min[S, B ] − X

c¦(r̃, u)\0. (15)

When the smaller of S or B is larger than X,which is assumed to be the case, as the probabilityof damage or catastrophe increases, the level of r̃will likewise increase.

Moving to a comparison of S and r̃, differenti-ating c %(r̃, u) with respect to S results in,

dr̃dS

=u

c¦(r̃, u)\ 0. (16)

As before, when the size of S increases, there willbe increasing efforts to provide revocability r.There are no incentives to provide revocability tocover potential damages past the level of B. To-gether, as u and min(S, B) increase, the profitmaximizing level of revocability, r̃, will likewiseincrease.

It is noteworthy that when the bond B equalsor exceeds the social loss S, the positive first ordercondition becomes the same as the first modelresult and the agent will select the same level ofrevocability that is socially optimal. Or moreformally,

when B]S, then c %(r*, u) = c %(r̃, u) and r*

= r̃. (17)

When an agent is required to place a bond atleast as large as the worst-case social loss, he willprovide the socially acceptable level of revocationand take great care in his actions. If fair insuranceor bonding is available, the agent will be indiffer-ent between purchasing this insurance or self-in-suring against the loss.

If the principal permits an agent to self-insureits actions with its wealth and the worst-case

social loss is larger than this bonded wealth, a badoutcome would bankrupt the agent leaving societyto cover the shortfall. This result, similar to Eq.(6), would lead to the firm under-producing revo-cability and the following holds,

BBS, then c %(r̃, u)B c %(r*, u) and r̃Br*.(18)

The principal will need to either insist on B\S orimpose r* on the agent to obtain the sociallyoptimal level of revocability. Imposing r* willrequire the principal be able to observe the agent’sprovision of revocation and may reintroducemoral hazard should the agent be able to hidesome of its actions.

Now, consider the situation where the worst-case social loss is uninsurably large, S. The worst-case uninsured loss to society can be measured asthe difference,

D= (S−B( i), (19)

where B( i is the maximum bond or insurance cov-erage available. The size of D could be reason forthe agency to simply deny the release; if theprincipal allows the release, it is assuming respon-sibility for the potential uninsurable losses onsociety’s behalf.

The results of Eq. (18) hold for B( iBS. Aprincipal permitting a release under these condi-tions will be motivated to require the agent toinsure up to B( i and to impose a level of revocabil-ity as close as feasible to the social optimum.

When the releasing agent is a public servantworking for a public agency, two additionalsources of moral hazard come into play. First,while government (a large entity with a vast arrayof diverse enterprises) can self-insure — unless Sis really too large even for government to cover— government may not be liable for the full arrayof potential private losses. This concern is exacer-bated when potential impacts extend beyond thejurisdictional limits of the public agency involved.For example, an upstream state fish and gameagency planning the release of an nonindigenousfish into the Mississippi River drainage could notbe held liable for the potentially large damages indownstream jurisdictions.


Second, the decision-makers in public agencies,the public servants, personally face relativelyweak incentives to minimize social costs. Withweak personal incentives, the public servant islikely to respond to pressures from private partiesfavoring or opposing the proposed release. Insome instances, the public servant may be closelyaligned with private constituency groups andvalue X, the private benefit resulting from theaction, more than the large worst-case social lossS.

There seems no assurance that a public agencyproposing a purposeful introduction will make thesocially optimal decision. This is a strong argu-ment for treating all publicly sponsored releaseswith the same scrutiny as those proposed byprivate agents. In particular, the principal may besceptical of a public servant proposing that gov-ernment self-insure a risky release. Requiring thereleasing agency purchase the socially optimalinsurance or bond would encourage optimal deci-sions. As a means to fund the purchase of thebond, the releasing agency could require the pri-vate potential beneficiaries to pay via user-specifictaxes or fees.

If B( iBS, the socially responsible level of revo-cation cannot or will not be provided, leavingsociety to bear the uninsured risk. It is inappro-priate to allow public servants in releasing agen-cies to undertake intolerably large risks on behalfof society. The ‘principal’ of our models should beinstitutionalized as an oversight agency indepen-dent not only of private agents, but also of gov-ernmental agencies desiring to releasenonindigenous species.

3. Conclusion

Ecologists have long warned society to take theintroduction of nonindigenous species seriouslyand have developed many protocols to help guidedecisions concerning intentional releases. Theseprotocols typically follow the ex ante approach torelease and seek to fully elucidate the issue beforeproceeding. They argue against introductions un-til the full spectrum of implications are under-stood via the collection of more and better ex ante

information, with no guarantees that enough in-formation will ever be collected. Because this in-formation is most often costly, this seeminglyendless effort to gain ever more information hasproved burdensome to releasing agents with theunintended consequence of protocols beinglargely ignored. However, there is a second ap-proach to avoiding bad outcomes, that of allow-ing only revocable decisions. Assuming,reasonably, that marginal costs of damage avoid-ance are increasing for both approaches, ex antefull information and revocability, we propose toimprove current procedures by paying a little lessattention to ex ante full information and a littlemore to revocability.

By combining the concepts of revocable actionsand incentive compatible behavior, we present animproved protocol (see Appendix A). This proto-col starts by first identifying the potentially af-fected parties and implementing Coasian liabilityprinciples when the affected parties are knownand property rights clearly established (Coase,1960). When the affected parties are large innumber and/or dispersed, the protocol suggests alimited role for an independent oversight author-ity to act on the behalf of the affected party. Theauthority would deny permits to releasing agentsthat fail to post bonds sufficient to compensatefor worst-case damage. One can imagine caseswhere there are large expected benefits from arelease, but a very small chance of uninsurablylarge damage. The oversight authority may decideto permit a methodical step-by-step process ofcontrolled releases, designed to make maximumfeasible use of revocability and learning-by-doing.Starting with a very small revocable release intightly controlled circumstances and with thor-ough review of the results, each subsequent stepwould involve larger releases, less rigid controlsand a lower level of revocability. The processwould be terminated as soon as the prospect of asufficiently harmful outcome emerged and, withhigh probability, the harmful outcome avoided. Ifall goes well at each step in the process, a benefi-cial release is completed and harm avoided.

While ex ante information remains important inthis alternative protocol, its success is dependentless on reliable prediction of the consequences of


a release and more on precommitment to avoidirrevocable actions. The key difference betweenthis alternative protocol and its predecessors is theattention it pays to revocability of outcomes. Ad-ditionally, moral hazard is avoided by the estab-lishment of an independent oversight authority tomake permitting decisions and the ex ante assign-ment of liability to releasing agents.

We implement a principal-agent model to deter-mine the socially optimal level of revocabilityprovision, the effects of moral hazard on revo-cability provided and the capacity of legal liabilityand assurance bonding to eliminate moral hazard.The principal-agent model generates the followingcontributions:� Eq. (4) and Eq. (5) demonstrate the existence

of a socially optimal level of revocability, r*.� When private rent-seeking agents have less to

lose from their actions than society, they willunderprovide revocability Eq. (6). When theprivate benefit exceeds their own worst-caseloss, agents will choose zero revocability Eq.(7).

� When the principal (oversight authority) allowsthe agent to act freely, subject to ex postliability for any damage to affected parties, thefollowing holds:� If the agent’s wealth exceeds the worst-case

social loss, the agent chooses at least thesocially optimal level of revocation (Eqs. (8)and (9)).

� If the agent’s wealth is less than the worst-case social loss, bad outcomes will bankruptthe agent and leave the society to absorb theexcess loss.

� If the expected benefit is large, but theagent’s wealth is small compared to theworst-case social loss, the agent may gamblefor the benefit and under-insure againstlosses (Eqs. (10) and (11)). These resultsprovide the rationale for imposing ex anteliability on the agent.

� When ex ante permission from the principal isnecessary before the release is allowed and theagent is required to post an ex ante bond tocover worst-case losses, the following applies:� When the bond is at least as large as the

worst-case loss, the agent will select the so-cially optimal level of revocability (Eq. (17)).

� When the size of the worst-case loss exceedsthe bond, the agent under-produces revo-cability and the principal can choose to denypermission to release, or to permit the re-lease, impose the highest feasible (but neces-sarily socially suboptimal) level ofrevocability (Eq. (18)) and absorb the risk ofuninsured excess loss D, on behalf of thepublic.

� Releases planned by public servants workingfor governmental agencies should be treated nodifferently than those planned by privateagents. An independent oversight authority(the principal) should require all parties, publicand/or private, proposing the release of non-indigenous species to post bonds and imple-ment the maximum feasible level ofrevocability if the largest bond obtainable fallsshort of the worst-case social loss.Finally, implementing a decision-making frame-

work that attends to both ex ante informationand revocability of bad outcomes, a new protocolcan be established that would prohibit or greatlyreduce the likelihood of catastrophic releases.This new protocol is presented in the followingappendix.

Acknowledgements

This research was funded in part by the OhioSea Grant College Program (project R/ZM-14)from the National Oceanic and Atmospheric Ad-ministration (NOAA grant NA90AA-D-SG496),US Department of Commerce.

Appendix A. A decision protocol based onrevocable actions

Start with the assumption that ex ante liabilityis established and a benign permitting authority isproviding oversight to help avoid moral hazard.This permitting authority will be termed the over-sight authority or OA. The revocable action deci-sion protocol begins with an initial request by anagent desiring to release a nonindigenous species.The OA’s initial action will involve an ex ante


review of the possible outcome set, in particularthe size and probability of loss, and number andidentity of affected agents.

A.1. Stage I: Initial ex ante re6iew of effects andaffected parties

A.1.1. Step 1: Does a large social loss exist inthe outcome set?

The OA will require the agent to perform aninitial ex ante review of the nonindigenous speciesusing all available literature, expert opinion andsimulation. This study will determine the potentialsize and scope of loss. The public will be allowedto review and comment on the agent’s initialfindings. The OA, using a panel of technical ex-perts, will determine the facts of the review. Ifthere is no potential of loss in the outcome set,then the release is permitted unconditionally. Ifthe possibility of harmful outcomes exists, the OAmoves to the second step.

A.1.2. Step 2: Identify parties affected by releaseWith the existence of harmful outcomes iden-

tified, the OA next requires the agent to determinethe number and identity of the adversely affectedparties. If there are only a few easily identifiableaffected agents, move to step three, which in-volves private contracts between releasing andpotentially affected parties.

When the potentially affected parties are un-known or large in number, optimal private con-tracts among acting and affected parties are moreproblematic. With numerous and/or unknown af-fected agents, proceed to the first step of Stage IIin the protocol.

A.1.3. Step 3: Affected parties are few andknown

With few and well identified affected parties,the agent notifies them of the pending release andreports this notification to the OA. The OA an-nounces it will permit the release only with theunanimous consent from all parties listed. Thisallows the affected parties to enter into privatecontracts with the agent wishing to release thenonindigenous species. With clear beneficiaries,risk-bearers and property rights, the individual

parties will work out an efficient outcome. Ac-cording to Coase (1960), with private benefits andrisks and unattenuated property rights, marketswill efficiently minimize net social costs.

Furthermore, it is reasonable to assume that theaffected agents are motivated to learn ex antetheir own potential loss sj, for all j affected partiesand require the agent to insure with their ownindividual bond, Bj]sj. With this requirement,according to Eq. (17), the agent will take due careand follow the optimal level of revocability. Whenthe agent can prove to the OA that all affectedparties have agreed to the release, the OA permitsthe agent to proceed with the introduction; other-wise the release is denied.

In this Coasian solution, the OA may provideindependent research and information concerningthe nature and extent of ex ante losses, butCoasian bargains are based on the affectedparties’ estimates of ex ante losses.

A.2. Stage II: Insure against worst-case socialloss

If Stage I results in a positive probability of asocial loss and the potentially affected parties aremany and/or difficult to identify ex ante, the OAshould implement ex ante liability by requiring asa condition for granting a release permit that theagent post an assurance bond at least as great asthe worst-case social loss. To make informed deci-sions on the size of assurance bonds, the OA willbe interested in gaining additional information onthe size of the worst-case social loss, S and therevocability cost function, c(r, u).

A.2.1. Step 1: Ex ante refinement of worst-casesocial loss and re6ocability

As a logical extension of the first step in StageI, the process of collecting information continues,to refine the estimate of the worst-case social lossresulting from the release. In an effort to identifyS and c(r, u) more accurately, the OA shouldexpect the agent to utilize only information al-ready in existence and conduct no new research atthis step. It is not the intent of this initial reviewto budget large research projects to determine thefull extent of the worst-case social loss. Simply


put, the task is to see how much useful informa-tion exists and to size up the situation.

When information is scarce and conclusionsvague, the OA will want to err on the conserva-tive side. While the costs and burden of researchshould be the responsibility of the releasing agent,the OA could serve as an independent provider ofinitial information to help set limits for perfor-mance bonds developed in the proceeding step. Asalways the OA serves also as reviewer and ulti-mate judge of the credibility of the informationgathered and conclusions drawn by the agent.

A.2.2. Step 2: Co6erage of worst-case social lossUsing the refined estimate of the worst-case

social loss and c(r, u) from Step 1, the OA re-quires the releasing agent to insure against thissocial loss before allowing the release. The mostdirect method is to require the agent to post anassurance bond. To avoid the problem of moralhazard, assume the bondsman is able to observethe level of revocability attained by the agent.Then the bondsman sets fees conditional on S, u

and the level of revocability r. The agent opti-mizes by minimizing own-costs (the cost of revo-cability plus the bonding fee), thus increasing theexpected net benefit resulting from the release,and protecting the bondsman by assuring thatrisk markets reflect true potential losses.

The OA determines S and the cost of fullrevocability (equivalent to reversing the socialloss) and requires the agent to insure or post abond up to this amount, such that B]min[S, c(r=1, u)]. Earlier it was shown that whenB]S, the agent will provide the socially optimallevel of revocability r*. With the worst-case socialloss thus insured, the OA can permit the introduc-tion. If the releasing agent cannot meet this condi-tion but nevertheless desires to proceed with therelease, the agent is asking society to bear anuninsured potential loss. Proceed to the nextstage.

A.3. Stage III: Conditions for release withuninsured potential loss

When worst-case social loss is larger than theagent’s ability to bond against its occurrence, the

OA may then choose between denying the releaseor permitting it conditionally. If the agent isallowed to proceed unconditionally, Eq. (18)shows that the agent will not exercise due cautionand under-provide the level of revocation.

A.3.1. Step 1: Gathering additional resources forinsurance

One case in which the OA may consider permit-ting a release when the agent is unable to post thebond against the worst-case social loss, ariseswhen other private agents agree to assume theuninsured portion of liability. When this occurs,the result reverts to B\S and the expanded agentgroup should be allowed to release as before inStage II, Step 2; otherwise move to the next step.

A.3.2. Step 2: Expected benefit much larger thanexpected loss

The OA may be tempted to permit a releasewith an uninsurable worst-case loss if the expectedbenefit of the release dwarfs the expected socialloss, or E(X)�E(S). (Note that to determineE(S), the OA needs to know both S and u ; inStage II, to determine the size of the bond, theOA needs only to know S ; the bonding firm,however, needs to know S and u). However, evenwith large expected benefits, when the size of theworst-case loss from taking action i is potentiallylarge enough to place a major burden on society,UiBUmin, the OA will be motivated to take arisk-averse stance. If the agent can convince theOA that the expected net benefit is too large todismiss, the OA will likely still demand a carefulapplication of revocability. Otherwise, societycould be left with the large social loss, as demon-strated by the positive static condition of Eq. (16).

To justify proceeding, the expected benefits ofthe release would need to outweigh any expectedloss by a large margin. A simple benefit-costcriterion where a proposal is accepted if expectednet present value is positive would be inappropri-ate, given the potential for large uninsured socialcosts. An appropriate criterion may be that ex-pected benefits exceed expected losses by at leastan order-of-magnitude.

If the condition of disproportionate large ex-pected benefits is met, move to the next step;otherwise deny the release.


A.3.3. Step 3: Stepwise release: experiments withre6ocation

Even with disproportionately large expectedbenefits, the large worst-case social loss dictatesthat extreme care must be taken before proceed-ing with an uncontrolled release.

This extreme care should be in the form of astepwise or experimental revocation. Here the ac-tion of release reverts to a process of many smallsteps or controlled experimental releases, withdiminishing degrees of revocability or increasingc(r, u) before ending with the eventual uncon-trolled release. This permits the controlled processto be stopped, and its bad outcome revoked, withdecreasing probability over the sequence of stepsor experiments, allowing for more deliberation bythe OA and the agent. Furthermore, this stepwiseprocess will serve as a learning experience forboth the OA and the agent. By keeping each stepsmall with initial steps fully revocable, new infor-mation on S, u and c(r, u) can be gained to allowfor more informed steps or experiments in thefuture.

A.4. An example stepwise release

The procedure starts with an agent requesting apermit for the first step in the stepwise process.This step, and all subsequent steps, must be per-mitted by the OA, or the release is terminated.

A typical step in this procedure will start withfeasibility studies performed by the agent to gainestimates of S, u and c(r, u). The studies are thenpresented for public review and comment and OAdetermination of fact. Following the review andcomments, the OA sets a bond that equals orexceeds the estimated worst-case loss resultingfrom this step, B]S. . Note that the agent willconsider the potential benefits from the wholestep-wise release process when deciding whetherto post the bond for the next step, but the OA willset the level of the bond considering only theworst-case damage from this particular step. Astepwise release program thus enhances insurabil-ity because the relevant damage to insure is onlythe damage from the next step.

If the agent agrees to post the bond, the OApermits the specified step. Following the release,

the OA requires the agent to fully report theoutcome of this specific experimental release. Thisreport is open to public comment and review andOA determination of fact. If the report is favor-able for continued stepwise release, the processmoves to the next step. The post-release studies ofthe just-completed step will provide basic infor-mation for the feasibility study for the next step.

The interest in revocation within the stepwiseprocess will allow the research effort by the agentto narrow in scope. Emphasis will focus on infor-mation useful to eliminating the species from theproposed host ecosystem, estimates of r and itsassociated cost function c(r, u). New researchshould be directed into exploring effective meth-ods for controlling and/or eradicating the species.

It is important to develop procedures that allowfor the carefully controlled experimental releasesin the early steps. For example, early researchmight focus on the development of completelyeffective physical barriers and/or eradication tech-niques. Early steps might use sterile release stockwith its minimal cost of revocation. For latersteps or an uncontrolled release, simply usingsterile release stock would prevent a completeevaluation of the species’ true effect on the newecosystem. At some point, the ability of the spe-cies to spread within the new ecosystem must beevaluated. More complete revocation effortsmight look at developing ‘genetic bombs’ thatcould be used to allow eradication of a successfulbut harmful released species.

As the steps progress, the information the OAgains on c(r, u) and u will allow it to continuouslyadjust the size of the necessary performance bondfor each experimental step. With the social lossinsured at each step, the agent will revert to thesocially optimal level of revocability. It is possiblethat the information gathered in sequential re-lease-steps may lead to the estimate of u becomingcloser to 0 or 1, making release approval or denialless problematic. A more likely outcome is in-creased information on the process of revocability(control and removal of the species). This couldlead the agent to a control program with afford-able cost c(r=1, u), even significantly less thanthe worst-case social loss and small enough forthe agent to bond a full release.


Nevertheless, a different scenario could result.The subsequent experimental steps may involveincreasing c(r, u) as the species’ release is lesscontrolled. Here the cost of revocation couldeventually become boundless with a complete anduncontrolled release. At some step in the processthe agent will reach the point where the social lossexceeds the agent’s ability to insure the experi-mental release. When S\B, again, there is anuninsured social loss. The OA must make thedecision to either allow the final uncontrolledrelease, or stop the process and deny the release.If the OA gambles and allows the release, thiswould imply that the socially optimal level ofrevocation is less than certainty and the OAmakes the final determination to take the chanceof S in order to enjoy the large expected benefitE(X).

References

Arrow, K.J., 1970. Alternative approaches to the theory ofchoice in risk-taking situations. In: Essays in the Theory ofRisk Bearing. North Holland, Amsterdam.

Coase, R., 1960. The problem of social cost. J. Law Econ. 3,1–44.

Costanza, R., Perrings, C., 1990. A flexible assurance bondingsystem for improved environmental management. Ecol.Econ. 2, 57–75.

Courtenay, W.R. Jr, Hensley, D.A., Taylor, J., McCann, J.A.,1984. Distribution of exotic fishes in the continental UnitedStates. In: Courtenay, W.R. Jr, Stauffer, J.F. Jr (Eds.),Distribution, Biology and Management of Exotic Fishes.Johns Hopkins University Press, Baltimore, MD.

Glover, F., 1989. Tabu search-part I. ORSA J. Comput. 1 (3),190–206.

Glover, F., 1990a. Artificial intelligence, heuristic frameworksand tabu search. Manage. Decis. Econ. 11, 365–375.

Glover, F., 1990b. Tabu search-part II. ORSA J. Comput. 2(1), 4–32.

Kohler, C.C., Courtenay, W.R. Jr, 1986. Regulating intro-duced aquatic species: A review of past initiatives. Fish-eries 11, 34–38.

Kohler, C.C., Stanley, J.G., 1984. A suggested protocol forevaluating proposed exotic fish introductions in the UnitedStates. In: Courtenay, W.R. Jr, Stauffer, J.F. Jr (Eds.),Distribution, Biology and Management of Exotic Fishes.Johns Hopkins University Press, Baltimore, ML, pp. 387–406.

Macauley, M.K., Bowes, M.D., Palmer, K.L., 1992. UsingEconomic Incentives to Regulate Toxic Substances. Re-sources for the Future, Washington, DC.

Mill, E.S., 1972. Urban Economics. Scott Foresman, Glen-view, IL.

Pauly, M.V., 1968. The economics of moral hazard. Am.Econ. Rev. 58, 531–537.

Solow, R.M., 1971. The economist’s approach to pollutioncontrol. Science 173, 498–503.

.

Documents

Intentional introductions of nonindigenous species: a principal-agent model and protocol for revocable decisions