ENTRY AND EXIT DECISIONS IN FLEXIBLE TEAMS

For the Journal of International Business Studies

Abstract The present study identifies a major limitation of current research on multinational corporations (MNCs). Joint

decision making in a distributed setting is of critical importance in practice, but has received little attention in

our theories. To address this gap in the knowledge, we examine the effects of flexible decision teams when

MNCs assess turbulent markets. Remarkably, flexible teams comprised of fallible evaluators can outperform

what is usually thought of as an optimal decision. Our main result supports the claims advanced in recent

empirical studies. Structural flexibility can help MNCs achieve high levels of performance even in conditions of

turbulence.

Keywords: Organization Theory, Decision-Taking Structures, Industrial Organization, Entry and Exit

Decisions, Market Turmoil, Flexible Teams

Acknowledgements: This paper received the best paper award at the 3rd annual JIBS/AIB/CIBER invitational conference, September

2005. Comments from the editor Arie Y. Lewin, two anonymous reviewers and Bill McKelvey, Paola Perez-Aleman, Phanish Puranam,

Nils Stieglitz, Kannan Srikanth, and Henk W. Volberda are gratefully acknowledged. Comments from participants at the 3rd annual

JIBS/AIB/CIBER invitational conference and support from the Strategic Organization Design Unit (http://www.sdu.dk/sod), University

of Southern Denmark is also acknowledged.

*Corresponding author. Email: tok@sam.sdu.dk, Tel: +45-65501000, Fax: +45-66155129.

Thorbjrn Knudsen*

Strategic Organization Design Unit (SOD),

Department of Marketing & Management

University of Southern Denmark

Campusvej 55,

DK-5230 Odense M,

Denmark

Phone: +45 - 6550 1000

Fax: +45 - 6615 5129

Email: tok@sam.sdu.dk

Michael Christensen

Strategic Organization Design Unit (SOD),

Department of Marketing & Management

University of Southern Denmark

Campusvej 55,

DK-5230 Odense M,

Denmark

Phone: +45 - 6550 1000

Fax: +45 - 6615 5129

Email: chm@sam.sdu.dk

1

ENTRY AND EXIT DECISIONS IN FLEXIBLE TEAMS

1. INTRODUCTION

The present article is concerned with organizational flexibility in multinational corporations (MNCs), i.e. large

firms that operate in multiple national markets. While the notion of organizational flexibility includes many

dimensions, we choose to focus on flexible decision teams. Our teams are flexible because they switch from

centralized to decentralized modes of making decisions. Contrasting prior research on entry modes (e.g. joint

ventures, greenfield investments, acquisitions), the present article examines the way such flexible decision teams

can influence entry and exit decisions.

Recent empirical studies on new organizational forms broadly support the claim that MNCs

experiment with flexible organizational structures in response to increased levels of turbulence in international

market (Ilinitch, DAveni, & Lewin, 1996; Lewin & Volberda, 1999). However, these advances in description of

flexible organization structures have been largely ignored in our theories of decision making in MNCs (Delios &

Henisz, 2003; Kogut & Singh, 1988; Reuer, 2000). It appears to be a stylized fact that flexibility is beneficial in

turbulent markets, but we do not really know why. To address this gap in our knowledge, we examine whether

flexible decision teams can be designed to reduce mistakes and thus increase profits, as MNCs decide to enter

and exit turbulent markets.

A critical issue is the problem of entry and exit in turbulent international markets, whether via

joint ventures, greenfield investments or acquisitions (Dixit & Pindyck, 1994; Kogut & Singh, 1988; Reuer,

2000). Sometimes costly mistakes get made,i for instance, when MNCs enter markets that result in losses and

ignore markets that would have become the sources of gains. Similarly, deciding when to exit can also be

problematic. Under turbulent conditions, a promising market opportunity can turn sour. Sometimes, MNCs exit

markets when their continued presence would have led to gains, or fail to exit markets where continued presence

can only lead to further losses.

2

Recent studies highlight the challenge of making entry and exit decisions in turbulent

international markets (Delios & Henisz, 2003; Kogut & Singh, 1988; Reuer, 2000). Delios & Henisz (2003)

found that changes in investment sequences occur because firms shift from an emphasis on developing

knowledge about international markets and consumers in stable environments, to an international expansion

strategy in more turbulent policy environments. Increased market turbulence tends to exacerbate the frequency

and severity of bad decisions. Thus, the evaluation of market opportunities in turbulent policy environments is a

critical determinant of entry and exit. While research into the efficacy of alternative modes of entry and exit into

international markets has a long pedigree, there has been little research into the ways that evaluations of market

opportunities influence entry and exit in international markets (exceptions are Delios & Henisz, 2003; Reuer,

2000). The present article offers a framework that enables us to consider these issues in a systematic way.

We develop a modeling structure that allows comparison of industry entry and exit under

alternative assumptions of managerial ability, and different levels of market turbulence. Three types of decision

maker the optimizer, the local searcher, and the fallible evaluator are compared in a generic entry-exit model

where turbulent conditions in an international market are modeled as fluctuating, short-run profits. Two extreme

forms of decision teams the centralized hierarchy and the decentralized polyarchy (the flat team) are

considered. The aim of the model is to examine whether teams of fallible evaluators can benefit from being

located in flexible decision teams, e.g. from shifting between hierarchical and polyarchical modes of

organization.

Our most important finding was that flexible decision teams comprising simple fallible agents

were superior to infallible agents able to solve a complicated dynamic programming problem. This result is quite

remarkable and also very encouraging from a behavioral perspective since it highlights the power of realistic

assumptions in our theories.

The article is organized as follows. Section 2 reviews the literature on new organizational forms.

Section 3 develops a modeling structure that allows comparison of industry entry and exit under alternative

assumptions of managerial ability, and different levels of market turbulence. We also examine how flexible

3

teams can improve entry and exit decisions in turbulent markets and why they have such remarkable properties.

Section 4 considers the implications of this study and concludes the article.

2. LITERATURE REVIEW

The last decades research on organizational forms has made headway, in converging upon a few particularly

important causes of the new forms that have been observed. Reflecting the wide agreement in the organization

literature, Child and McGrath (2001) in their introduction to a collection of nine articles in a special research

forum on new organizational forms, published in the Academy of Management Journal, point to increased

information intensity and internationalization as the main reasons for the emergence of these new forms.

The challenges referred to can be grouped into four broad categories: (1) increased

interdependence in interactions among organizations; (2) the possibility of performance disembodied from asset

ownership; (3) higher velocity characterizing almost all aspects of organizational functioning; and (4) changes in

power, in terms of a shift from a power base of tangible assets and inputs to power derived from the possession

of knowledge and information. The widely observed new organizational forms, it is argued, have emerged in

response to these challenges. According to Child and McGrath (2001), the objective of the new forms is to

delegate decision rights to where the relevant knowledge and information reside, and then to use information

technology (IT) for support.

In contract economics, this move of decision rights to be closer to those with information, has

been referred to as the organization redesign solution, as opposed to the traditional management information

systems (MIS) solution, according to which it is information that is moved closer to the decision maker

(Brynjolfsson & Mendelson, 1993; Nault, 1998). Fundamental advances in IT and measurement technologies

have facilitated a number of experiments with organizational forms (Nault, 1998; Zenger & Hesterly 1997) that

are often referred to by the notion of new organizational form (Daft & Lewin 1993). A common characteristic

of these experiments is the use of IT in hierarchies to achieve a decentralization of decision rights (Child &

4

McGrath, 2001; Nault, 1998). Whereas advances in IT are commonly viewed as the facilitator of new forms,

hypercompetition is seen as the primary cause of their emergence since traditional forms are considered

maladaptive when high turbulence is the norm (Child & McGrath, 2001; Daft & Lewin, 1993; Volberda, 1996).

Hypercompetition refers to the shift in the rules of competition that was observed through the

1990s (Ilinitch, DAveni & Lewin, 1996; Volberda, 1996). Technological change, the shortening of product life

cycles, and the increasing aggressiveness of competitors have led to shorter and shorter periods of competitive

advantage, punctuated by frequent disruptions. That is, hypercompetition and market turbulence go hand in

hand. This development associates turbulence with flexible organizational arrangements as in Volberdas

proposition: In a fundamentally unpredictable environment which may also be dynamic and complex

(hypercompetitive), the optimal form employs a broad flexibility mix dominated by structural and strategic

flexibility and has a nonroutine technology, an organic structure, and an innovative culture (1996: 366-367).

Volberda (1996) nicely sums up the argument for flexibility that has been forwarded in much of the literature on

new organizational forms: In the new mode of hypercompetition rents do not derive from specialized routines

but from adaptive capability. The reason is that, with hypercompetition, competitive change cannot be predicted

but only responded to more or less efficiently, ex post (1996: 360).

The identified imperative of firm-level flexibility, in turn, stresses the need to maintain some level

of organizational consistency. Thus, it has been emphasized that organizations must respond to the twin

pressures of continuity and change. That is, organizations need to exhibit increasing flexibility while maintaining

consistency and reliability (Bartlett & Ghoshal, 1998; Volberda, 1996, 1998).

A broad empirical literature supports the argument that flexibility outperforms rigidity under

conditions of turbulence.ii If we examine the observed changes in organizational characteristics in more detail, an

impressionistic picture with marked features appears, including downsizing (Bowman, Singh, Useem &

Bhadury, 1999), a trend toward more collaborative business relationships (Cannon & Homburg, 2001; Nault &

Tyagi, 2001), network organizations (Achrol, 1997), flexibility achieved by more organic and temporary work

arrangements (Bigley & Roberts, 2001; Child, 1997; Heydebrand, 1989; Ilinitch, DAveni, & Lewin, 1996;

5

Lewin & Volberda, 1999; Zahra & ONeill, 1998), more nimble governance structures (Cecil, Ciccotello &

Terry, 1995; Kole & Lehn, 1997), decentralization of decision rights (Moller & Rajala, 1999), and new

employment contracts characterized by less commitment between employer and employee (Charness & Levine,

2002). While flexibility spans a number of meanings in these studies, it is generally thought to be beneficial in

turbulent markets. However, we do not really know why.

To address this neglect, the present study aims to direct attention to the role of flexible decision

teams in MNCs evaluation of market opportunities. To stimulate further research on this important topic, we

develop a general modeling structure from which testable propositions can be drawn. This effort adds to prior

research on entry and exit decisions in MNCs, including studies on the evaluation of foreign opportunities

(Cavusgil, Kiyak & Yeniyurt 2004; Schooler, 1974). By examining flexible evaluation structures, we also add to

prior studies that examine the effect of voting rules in stable evaluation structures (Birnberg, Pondy & Davis,

1970).

The present article examines the effects of flexible decision teams when MNCs are making entry

and exit decisions in turbulent international markets. A critical issue is the assessment of market opportunities in

the face of irrecoverable entry and exit costs. Using the common analogy from statistical inference, MNCs make

errors of omission (Type I errors) and commission (Type II errors). MNCs enter markets that offer losses

(commission errors) and steer clear of markets that would have been sources of gains (omission errors). MNCs

face the poker players problem of deciding when to enter the game and when it is time to exit. Under turbulent

conditions, a promising market opportunity can turn sour. Mistakes occur when MNCs leave an international

market too early, or too late. Such mistakes can be costly, and increased market turbulence can lead to both more

frequent and even more severe mistakes. We examine whether flexible decision teams can somehow be designed

to reduce the frequency of such mistakes and thus increase profits.

6

3. THE MODEL

We develop a modeling structure that allows comparison of industry entry (exit) decisions under alternative

assumptions of managerial ability and different levels of market turbulence. Three types of decision makers the

optimizer, the local searcher, and the fallible evaluator are compared in an entry-exit model where turbulent

market conditions are modeled as a fluctuating short-run profit.

Individual evaluators: optimizer, local searcher, fallible evaluator. The MNC operates in a

turbulent international market. We invoke the following standard assumptions from dynamic industry entry-exit

models. The MNC observes the next periods potential operating profit (t+1), earned in case that the MNC is

active (an inactive business unit earns an operating profit of zero). Based on this information, the decision maker

estimates critical levels of operating profit at which an inactive business unit should enter, and an active business

unit should exit. The optimizer uses a dynamic programming approach to extract the optimal critical levels of

operating profit from the Bellman equation.iii Individual agents capable of using such methods can identify and

use a proposed optimal profit level to trigger industry entry and exit. These would be very close to perfection,

but real-world decision makers are unlikely to achieve such high standards. The optimizer is an unrealistic, but

useful theoretical comparison point against which the performance of fallible decision makers can be assessed.

The second type of decision maker is the local searcher, familiar in behavioral theories of the firm

(Cyert & March, 1963; Knudsen & Levinthal, 2007; Levinthal & March, 1981; Nelson and Winter, 1982). The

set of alternative actions is not presumed to be laid out in its entirety ex-ante, but must be discovered or searched

for. In the present context, the local searcher enters a new market if the operating profit is at least equal to zero

(or some other trigger point) and exits if the short-run profit becomes negative. In this way, such agents search

for new markets that satisfy some minimum performance criteria (e.g. potential profit should be at least zero).

However, alternatives, once identified, must also be evaluated (Knudsen & Levinthal, 2007). While the local

searcher is usually portrayed as being capable of perfect evaluation, the fallible evaluator offers a more realistic

characterization of decision makers than does the local searcher.

7

The third type of decision maker, the fallible evaluator, has imperfect discriminating ability. This

type of decision maker behaves like the local searcher, but is not capable of making sharp evaluations of short-

run profit for the reasons proffered in the literature on new organizational forms and elsewhere (ambiguity,

complex interactions, limited information, limited computation power, etc.). That is, our fallible evaluator

behaves in accordance with the stylized description of real-world decision makers offered in much of the

organization and management literature (Knudsen & Levinthal, 2007).

The discriminating ability of the fallible evaluator is modeled as a linear screening function.iv

Figure 1 presents five examples of the fallible agent. These include the agent who is virtually unable to

distinguish whether the currently observed profit is a good indicator of future industry profits (= 0.01) as well

as the agent who has a much better idea about the relation between observed profit and post-entry performance

(= 0.50). The fallible agent becomes more able as the inclination of the screening function increases. Suppose

the ability of an agent is captured by a screening function with slope = 0.05. If the currently observed profit is

5, we can read the probability on the y-axis that an entry decision will be made. As shown in Figure 1, the agent

in our example would enter the market with probability 0.75.

It is apparent from Figure 1 that a steeper slope (higher value of ) centers the zone of uncertainty

on a more narrow range of values for observed profit. By contrast, a lower value of spans a wider range of

uncertainty where observed profit maps onto a probabilistic entry decision. The slope of the screening function

captures fallibility, and fallibility translates into a wider zone of uncertainty. The lower the slope, the more

fallible is the decision maker. At the extreme, the agent with = 0 is a coin flipper who would enter (or exit) the

market with probability 0.5 no matter the level of observed profit. A more fallible agent also has a wider zone of

uncertainty within which costly reversals of prior decisions can occur.

FIGURE 1

8

Figure 1 also depicts the local searcher ( ). In our example, the local searcher would enter

with probability 1 for any non-negative level of observed profit and this agent would exit with probability 1 as

soon as negative profits were observed. The local searcher is unable to shift between multiple decision criteria

even though such an agent is infallible as regards quality discrimination. In our case, the observed levels of profit

that lead to successful decisions are far apart when firms pay an irrecoverable fee to enter and exit. Costly entry

and exit decisions introduce multiple decision criteria because they require consideration of two trigger points

for observed profit. In a sense there is a trade-off between treatment of multiple decision criteria and certain

quality discrimination. The local searcher can make extremely sharp judgments about quality but this virtue goes

hand in hand with a zone of uncertainty that shrinks to a single point, i.e. the local searcher is unable to consider

more than one trigger point.

It is disturbing that the characterization of the local searcher in most of our formal models does

not match the empirical reality in entry and exit decisions. To illustrate this point, Figure 1 provides an example

of optimal entry and exit points computed by dynamic programming. Only if observed profit is at least 9.89

would the firm enter the market and become active. Once active, it would only exit if observed profit fell below -

9.90. Our fallible evaluator, by contrast, has a zone of uncertainty that can be made wide enough to span the two

optimal points for entry and exit. As the two optimal points for entry and exit become farther removed from each

other, however, the slope of the screening function must decline. In Figure 1, the fallible agent with a slope of

about = 0.05 spans the two optimal points for entry and exit. Unfortunately, a lower slope implies that more

decisions will be reversed. The role of the organizational structure in this case, is to remedy the unfortunate and

costly tendency to reverse decisions while maintaining the span of difference between trigger points for entry

and exit. As sunk costs increase, the individual agent is less able to handle the conflicting demands of multiple

decision criteria and certain quality discrimination. The way out is to structure costly entry decisions in decision

teams.v

Decision teams. We use Christensen & Knudsens (2007) recent extension of the Sah & Stiglitz

(1985, 1986, 1988) characterization of organizational architectures to model evaluation in teams. This modeling

9

approach was recently extended to the study of imperfect evaluation within the context of NK models (Knudsen

& Levinthal, 2007). The problem we are considering in the present article, however, is different from those

problems for which the NK model is best suited. The NK model has become the standard tool for organizational

analysis of complex spaces where alternatives are hard to locate even though the payoff for each particular

configuration remains fixed, once and for all. In contrast, the problem we are addressing relates to fluctuating

payoffs.

The intuition in what follows is that fallible evaluators can (always) benefit from being placed in

flexible teams whereas local searchers are beyond help because they discriminate perfectly. Ironically, imperfect

discrimination of fallible agents is a source of flexibility that can be utilized by designing appropriate decision

teams. As we shall see, this gives rise to the surprising result that flexible teams with relatively few fallible

decision makers can outdo the so-called optimizer. Notably, the fallible decision makers in our model are

modeled as local searchers and thus, in accordance with much of the literature on organizations and

management, are quite realistic.

Local searchers are limited in their ability to search for new alternatives, but once identified, they

are capable of perfect discrimination. Within the present context, local search approaches optimality when the

critical levels of operating profit become a single point say, optimal entry (exit) for non-negative (negative)

operating profit. This happens when entry and exit costs go to zero.

In contrast to local searchers, fallible agents are not capable of perfect discrimination. It can be

shown, however, that fallible evaluators can be placed in teams such that their joint screening function comes

arbitrarily close to perfect discrimination (Christensen & Knudsen, 2007). Then, under the assumption of (very)

low costs of entry and exit, fallible evaluators could always gain from being located in fixed organization

structures.

However, we are interested here in the much more realistic case of significant exit and entry costs.

In a turbulent market with significant entry (and exit) costs, the optimal points of entry and exit are located far

apart (as illustrated in Figure 1), i.e. firms only enter if operating profits are (very) high and they only exit if

10

operating profits are (very) low. To help fallible agents overcome the inbuilt rigidity that narrows their focus to a

single point of operating profit, we are led to consider the possible advantage of using flexible evaluation

structures. In particular, we consider two extreme forms of evaluation structure. One is the hierarchy, in which a

proposal to enter (or exit) a market is validated at successively higher levels in the team. Only if the proposal is

accepted at each level, will the MNC enter a new market. The second form is a polyarchy, a flat, decentralized

structure in which acceptance by any one actor is sufficient for the proposal to be approved. A flexible decision

team is modeled by shifting between the two forms of organization.

The effect of locating fallible evaluators (e.g. = 0.05) in a hierarchical form is shown in Figure

2. In what we term a hierarchy, the short-run profit is initially considered by a member of the decision team. If

the proposal is rejected by that team member, it is eliminated from further consideration and the business unit

discontinues its activities (or remains inactive). Alternatively, if the proposal is approved by that decision maker,

then it is passed on to the next decision maker in the chain of command. A proposal is acted upon only if it has

been positively vetted by all of the evaluators in the team. Formally, sequential decision making in a hierarchy

corresponds to joint decision making in a team (or committee) where a proposal of entry is only accepted if each

member accepts it.

Figure 2 compares the effect of locating fallible agents in hierarchical decision teams with two

(H2), five (H5) or ten (H10) members, and also compares these structures to the optimizer and the local searcher.

The single fallible agent is represented by a dashed line and we use a linear screening function with a slope of =

0.05 for the purpose of illustration. H5 is a hierarchy employing five such agents. We can read off the probability

of entry for the hierarchical teams on the y-axis and compare these to the single fallible agent, the optimizer and

the local searcher. Suppose the observed profit is 5. In that case, the local searcher enters with probability p=1,

the single fallible agent enters with probability p=0.75, and H5 enters with a probability of p=0.24. The

hierarchy is a skeptical organizational form that tends to tone down a positive vetting made by a single fallible

team member (reducing the entry probability from 0.75 to 0.24). The effect is to reinforce the status quo unless a

very promising observation is made. As the size of the hierarchy increases, its screening function approaches a

11

vertical line. Figure 2 illustrates this effect by mapping out the change in screening functions from H2 over H5 to

H10.

FIGURE 2

The effect of locating fallible evaluators (e.g. = 0.05) in a flat team, also known as a polyarchy,

is exactly the mirror image of Figure 2 and therefore not shown here. In the flat team, a proposed alternative can

be adopted by any of the members of the decision team. Only if all decision makers in succession reject an

alternative is it dismissed. Formally, sequential decision making in a polyarchy corresponds to joint decision

making in a team (or committee) where a proposal of entry is accepted if only a single member accepts it. As the

mirror image of the hierarchy, the polyarchy is an optimistic organizational form that tends to promote change.

Flexible decision teams. The aim of our model is to examine whether teams of fallible evaluators

can benefit from being located in flexible organizations, e.g., from shifting between hierarchical and flat,

polyarchical modes of organization. This case could also be viewed as the consistent use of a hierarchical form

with shifting targets (accepting entry versus accepting exit). In joint decision making, this is equivalent to

shifting from a rule requiring that each member of an decision team must accept entry to a new rule, requiring

that only one member accepts continued presence, after entry.

Note how the local searcher in Figure 2 is represented by a single vertical line ( ) while the

optimizer is represented by two vertical lines computed by dynamic programming. Costly entry and exit

decisions require consideration of two trigger points for observed profit. These two points represent the

challenge of balancing two kinds of error. One is the error of staying in the market despite a downturn in the

business cycle and the second is the error of missing an upturn in the business cycle by remaining inactive.

A flexible use of teams that shifts between entry decisions in hierarchical teams and exit decisions

in flat, polyarchical teams will effectively mimic the optimizers use of two decision criteria. For example, using

H5 in Figure 2 in the case of an entry decision and then delegating the exit decision to a flat, polyarchical, team

12

would shift between the H5 curve and its mirror image, a curve for a 5-member polyarchy (not shown here). As

reported below, this leads to the surprising result that simple flexible decision structures comprising 4-5

imperfect people can outdo the entirely superhuman performance of the "optimal" agent (an encouraging result

as regards realism).

It also turns out that the organizational structure that best helps the myopic decision maker in our

model is a flexible team of quite realistic size (4-5 members). The reason that larger teams, such as H10 in

Figure 2, are inferior is that they span a narrower zone of uncertainty. In Figure 2, H10 is closer to the straight

line of the perfect agent than H5. More generally, as more agents are included in the decision team, the hierarchy

(and polyarchy) approaches a straight line. However, our results show that in turbulent markets, some zone of

uncertainty, as represented by a non-linear trigger point in Figure 2, is superior to perfect discrimination

represented by a straight line.

The nature of the environment. The local environment for the MNC subsidiary is more or less

turbulent and the MNC considers whether its subsidiary should operate. More precisely, short-run profit is a

random walk, modeled as an exogenous Markov process.vi The MNC observes the short-run profit of the next

period and decides whether it should enter the market. The next periods potential profit is the exogenous

Markov process:

t+1 = h(t , t+1) = AVG + (t AVG) + t+1 (1)

where the random shocks are i.i.d. and normally distributed, N(, 2), with mean = 0 and variance

2. The

parameter is the autoregressive coefficient, which determines the rate of reversal towards the mean AVG. The

following analyses set AVG = 0 and = 0.90, but vary the volatility parameter in order to examine alternative

levels of turbulence in international markets. Higher volatility increases turbulence within the environmental

13

conditions defined by . This is shown in Figure 3, where the lower panel illustrates a highly turbulent task

environment (=10), while the upper panel illustrates a less turbulent environment (=4).

FIGURE 3

Figure 3 shows the optimal entry and exit points computed by dynamic programming (these

points are identical to the examples from Figures 1 and 2). Only if observed profit is at least 9.89 does the

optimizer enter the market and become active. Once active, the optimizer exits only if observed profit falls

below -9.90. The actual entry and exit decisions for the optimizer are shown in the panels of Figure 3. The upper

panel of Figure 3 also shows a period where H5 has comparative advantage over the optimizer (from t=13 to

t=36). The underlying reasoning is subtle. As depicted in Figure 2, the optimal evaluator uses a single linear

trigger point, but the organization comprising fallible agents has a curvilinear set of trigger points. This

curvilinear set of trigger points can effectively work as a confidence interval around the exit and entry points

used by the optimizer. Supposing the observed profit is positive, H5 enters with a positive probability. On

average, therefore, H5 would be active during some of the periods between t=13 and t=36 (upper panel of Figure

3). During this time, the net profit is positive. Therefore, H5 earns an expected positive profit. By contrast, the

optimizer earns a profit of zero during the same period because it remains inactive (until t=36).

But why is H5 superior to the fallible agent? Again, it is useful to refer back to Figure 2. If the

observed profit is 5, the single fallible agent enters with probability p=0.75 and H5 enters with a probability of

p=0.24. The single fallible agent is quick to enter and once active in the market, it is also quick to reverse the

decision, and exit. As profits fluctuate the single fallible agent will make many reversals of prior decisions.

When exit and entry are costly, the net result is diminished profits when compared to the steadier course of H5.

Some restraint is required before the firm enters and that is exactly what H5 provides. If more agents were

included, however, the restraint would become excessive. The move from H5 to H10 in Figure 2 illustrates this

effect.

14

Our results generally indicate that H5 is superior to larger or smaller decision teams as well as the

optimizer (see Figure 5, below). However, the advantage shrinks as volatility increases from =4 to =10. The

lower panel of Figure 3 illustrates a task environment with =10. In such a highly volatile environment, the

shifts from low profits to high profit peaks are much sharper. As volatility increases to a level of =10, the shifts

that trigger entry and exit will therefore tend to be the same for the optimizer and H5. In consequence, the

advantage of H5 over the optimizer shrinks (as shown in Figure 5).

The decision to enter and become active. The state variables are the short-run profit, , and the

current operational status, o, of the MNC in the international market.vii

If the MNC is currently operating in the

international market, o is 1. If the MNC does not operate, o is zero. At the beginning of each period, the MNC

makes an operating decision, a, for the next period. If the MNC operates in the next period, a is 1. If the MNC

does not operate in the next period, a is zero.

That is, the MNC observes the next periods potential profit () and its current operational status

(o). It then takes an action (a), about whether it should operate in the next period. In consequence, the MNC

earns a reward f(, o, a) that depends on the current state of the economic system, the current operational state of

the MNC, and the action taken. In addition, an unknown exogenous shock influences the next periods potential

profit. The state transition function is (see e.g. Miranda & Fackler, 2002):

g(t+1, o, a, ) = (h(, ), a) (2)

If the MNC is not operating and decides to enter the international market under consideration, it pays a fixed

entry cost, KEntry. Should the MNC be operating in an international market and then decide to exit, there is a

shutdown cost of KExit. In all simulations, KEntry was set at 10 and KExit was set at 10 (our results hold for a broad

range of these values). The reward function is:

15

f(, o, a ) = a KEntry (1 o)a KExit (1 a)o (3)

Estimation of critical levels of short-run profit. The optimal value of the business unit, V(, o),

given an observed short-run profit and operational status o, is estimated by the Bellman equation. We use

numerical methods provided by Miranda & Fackler (2002) to approximate the value function V(, o):

V(, o) = max{ f(, o, a) + E V(h(, ), a)} (4)

The functions, f() and h() are given by eq. 2 and 3, is the discount rate.viii

Without loss of

generality was set at 0.10 (additional analyses show that alternative values provide qualitatively similar

results). We extract, by numerical methods, the critical levels of operating profit that maximize the value

function (eq. 4). These provide simple criteria for optimal entry and exit. In addition, note that the two

components of the value function express a trade-off between immediate rewards f(, o, a) and the expected

future gains following the immediate action E V(h(,),a).

The unrealistic dynamic programming approach described so far is only available to the

optimizer. The local searcher and the fallible agents simply compare short run to a criterion for satisficing

performance. If short run profits are too low, it will stay out (or exit in the case that it is active). The fallible

decision maker behaves exactly like the local searcher. The only difference is that the fallible decision maker

sometimes makes a wrong decision, i.e. exits when the firm should have remained active according to the

satisficing criterion.

The critical profit levels extracted for a particular example of = 4, and = 0.90 are shown in

Figures 1 and 2. Figure 1 compares the critical profits levels to alternative levels of discriminating ability in a

fallible evaluator, and Figure 2 shows the effect of placing a fallible evaluator in a hierarchy (using a

polyarchical structure gives the mirror image of Figure 2). Figure 4 shows the results obtained for this particular

16

example, with = 4, and = 0.90. The results for a broad range of parameter values are reported in Figure 5.

The configuration space included a full range of parameter values: = 0.01, 0.05, 0.10, 0.50, = 1, ..., 10, and

= 90 (additional simulations were conducted to assess robustness). We set entry costs, KEntry, at 10 and shutdown

costs, KExit, at 10. A random draw determines the initial operating status, o, of the business unit.

A business unit optimally operates during the next period if the short run profit for that period is

above some level, H, and a business unit optimally shuts down if the profit is lower than L. Given entry and

exit costs, L < 0 < H. Once the MNC has accepted that one of its units enters a new market, it is often optimal

to continue operating despite periods of negative operating profits.

The estimation of optimal actions provides a useful comparison point for assessment of the

performance of decision makers with much less information and computational power than required to solve eq.

4. As previously explained, the performances of three types of decision makers are examined. The first type is

the optimizer who is capable of finding the optimal decision criteria in terms of unique trigger points (as shown

in Figure 1-2). Given information about the systems state transition function, the optimizer solves the dynamic

programming problem implied by the corresponding Bellman equation. The second type of decision maker is the

local searcher who enters if the short-run profit is at least zero and exits if short-run profit is below zero (other

trigger points were examined to assess robustness). The third type of decision maker is the imperfect evaluator

(we examine different levels of evaluation capability as shown in Figure 2).

The MNC uses a flexible organizational form to evaluate market opportunities. Prior to entry, the

MNC uses the skeptical hierarchical form that allows entry only after validation at a number of successive

levels. After entry, decisions are delegated; the decision team shifts to an optimistic polyarchical mode where

approval of the market conditions by any one actor is sufficient to stay in the market. Only if everybody in the

decision team disapproves of the market conditions, will the MNC exit. Our choice of organization structures is

kept simple for illustrative purposes. The results are not sensitive to this choice. In the general case, similar

results are obtained when entry decisions take place in a structure that is more centralized than the exit decision

(which is often the case as decision power is usually delegated to a subsidiary).

17

Results

We compare the performances of the three types of decision makers in a horse-race over 1000 periods. On the

basis of 1000 samples, we find that optimizers always outperform local searchers and single fallible evaluators.ix

Generally, local searchers, who are perfect evaluators using a wrong decision rule (critical profit level of = 0,

in the case of both entry and exit), also outperform single fallible evaluators (as shown in Figures 4 and 5). The

single fallible evaluator does not perform well, tending to promote frequent (and costly) reversals of prior

decisions. Since entry and exit, in each case, are associated with a non-recoverable cost, the many mistakes

impose a considerable cumulative cost.

The disadvantage of the local searcher and the single fallible evaluator is ameliorated in a

turbulent environment (Figure 5). With higher turbulence, the range between the values of short run profit that

are optimal for entry and exit widens (holding entry and exit costs constant). Thus, a high results in

pronounced, long periods of negative and positive profit that are visible to the more simple-minded local

searcher.x The performance of the local searcher shown in Figure 4 is markedly higher than the performance of

the single fallible decision maker and even a flexible team comprising two fallible decision makers. As more

decision makers are added to the team, the quality of the decisions further improves until it comprises five

members. As further members are added, however, there would be a marginal decline in performance.xi

FIGURE 4

The critical values of short run profit used in the simulation reported in Figure 4 are the same as

those in Figures 1 and 2. From Figure 2, it can be seen that the effect of adding members to the hierarchical

decision team is to push a portion of the screening function to the south-east of the critical entry value. Adding

members implies that the screening function of the decision team begins to approximate the optimal entry level.

At some point, however, the screening function is pushed too far to the east, beyond the optimal level. The exit

18

case is perfectly symmetrical to the case of entry and therefore not shown here. By switching between the

hierarchy when entry opportunities are evaluated and the flat, polyarchy when exit options are considered, a

team of fallible evaluators can approximate the two optimal values of entry and exit.

Rather surprisingly, and consistent with the conjectures in the literature (Bartlett & Ghoshal,

1998; Child & McGrath, 2001; Lewin & Volberda, 1999; Volberda, 1996, 1998), it turns out that flexible

evaluation structures can dramatically improve the performance of the fallible evaluator. We show that, rather

remarkably, MNCs employing flexible teams of fallible evaluators can obtain higher levels of profits than the

levels obtained by the optimizer (Figure 5). This result is obtained if the MNC uses a centralized decision

making structure (a hierarchy) prior to entry and then switches to a flat (polyarchical), decentralized structure

after its business unit has entered the international market.

FIGURE 5

Considering the results depicted in Figure 5, it is clear that a team of fallible agents generally

outperforms a single fallible evaluator. Moreover, the gain in performance occurs for very moderate team sizes

of about five members. Using teams of five in the way described here, rather than relying on single evaluators,

will generally lead to improved decisions. The most surprising result reproduced in Figure 5 is that fallible

decision makers who use flexible decision making structures are generally superior to the so-called optimal

decisions based on a dynamic programming approach.xii

Figure 2 shows the cause of this surprising result: the

optimal evaluator uses a single linear trigger point, but the organization comprising fallible agents has a

curvilinear set of trigger points. The set of curvilinear trigger points effectively works as a confidence interval

around the optimizers exit and entry point. A team of fallible decision makers would exit before the operating

profit plunges to the negative level required by an optimizer.

So, it is the non-linearity induced by the joint effect of individual evaluators that produces

curvilinear trigger points. This non-linearity gives our small teams an edge. Thus, flexible teams of fallible

19

agents become a source of competitive advantage in a turbulent environment.xiii

The effect is most pronounced in

moderately volatile task environments. In a highly volatile environment, the shifts from low profits to high profit

peaks are much sharper (see Figure 3) and the shifts that trigger entry and exit will therefore tend to be the same

for the optimizer and H5. In consequence, the advantage of H5 over the optimizer shrinks as volatility increases

from =4 to =10. The only exception to the surprising advantage of H5 over the optimizer is a very low

volatility of =1. In that case, the optimizer will never enter (since entry and exit costs are substantial at KExit=10

and KEntry=10). By contrast, H5 will on average earn negative profits because it occasionally will enter and then

discover that the operating profits cannot cover the entry cost. However, our flexible teams of simple fallible

agents are generally superior to the so-called optimizer when task environments become more volatile (>1). In

a world with increasingly turbulent international markets, this result points to the advantage of small flexible

decision teams.

4. CONCLUSION

This study investigates entry and exit decisions in turbulent markets with fluctuating short-run profits. It

complements prior efforts by examining the viability of flexible teams, e.g. shifts between centralized and

decentralized modes of decision making. We found that fairly small teams generally outperform a single

decision maker. This result reflects the properties of a centralized, hierarchical structure, which reduces

commission errors (accepting a bad proposition) whereas as a decentralized, polyarchical structure reduces

omission errors (rejecting a good proposition). As a firm uses a centralized structure when considering entry into

a new international market and then leaves the choice of continued operation to a decentralized local team, it

actually mimics an optimal choice. Indeed, we found that fallible decision makers located in flexible teams were

even able to outperform what is commonly thought to be an optimal decision. That is, our teams of simple

fallible agents were superior to infallible agents who are able to solve a complicated dynamic programming

problem, given all the relevant statistical information and the systems state transition function. This result is

20

quite remarkable and also very encouraging from a behavioral perspective since it highlights the power of

realistic assumptions in our theories.

Our model is a behavioral extension of classical entry and exit models. The work most closely

related to ours is Dixits (1989) analysis of entry and exit decisions under uncertainty. While this work points to

the benefit of fuzzy trigger points in turbulent markets, our model uncovers the remarkable result that flexible

decision teams are a realistic way for simple-minded managers to realize and obtain the benefits of such fuzzy

trigger points. A number of limitations should be considered when our results are extrapolated to real-world

cases. First, we abstract from a detailed analysis of market structure and industry dynamics and thereby limit our

analysis to established international markets. Second, we only examine shifts between extreme hierarchical and

flat organizational structures. Third, we abstract from multiple goals and assume that profits are a sufficient

summary measure of market attractiveness. Fourth, we abstracted from two of the main aggregation forces in an

organization incentives and coordination. While the first two limitations add important nuances to our

analyses, they do not alter the prediction that fairly small teams, comprising a handful of agents, generally

outperform other modes of decision making when considering costly entry and exit. The third and fourth

limitation, however, define boundary conditions of our model.

These limitations to our study can be overcome in a number of ways. First, the generation of new

markets is to some extent captured in the notion of irrecoverable entry and exit costs. However, we abstract from

other issues relating to a detailed understanding of market structure and industry dynamics. Future research on

these important issues could usefully draw on the extension of the NK-modeling framework developed by

Lenox, Rockart & Lewin (2006, 2007). Two features of our model should be integrated into the analysis. We

were addressing a problem relating to random walks in payoffs rather than ruggedness stemming from

complexity within a context where payoffs are fixed, once and for all. In order to overcome this limitation,

fluctuating payoffs should be included in the NK model. Another limitation of most prior NK models is that they

do not extend to analyses of teams. Knudsen and Levinthal (2007) recently pointed the way to such an extension.

Second, the limitation relating to the range of organizational structures being studied can be overcome by the

21

methods offered by Christensen and Knudsen (2007) and applied within the NK-modeling framework by

Knudsen and Levinthal (2007). While adding nuances, such analysis would not alter our main results because we

have used the organizational structures with the most extreme screening properties (Christensen & Knudsen,

2007). Third, as regards a more detailed analysis of multiple goals, recent work on the NK model seems

promising (Ethiraj and Levinthal, 2007). Finally, the consideration of incentives and coordination adds layers of

complexity to the analysis, which are likely to uncover interesting contingencies, albeit none that we expect to

contradict the advantage of flexible teams when considering entry and exit decision. Overall, agent based

approaches seem promising in reducing some of the limitations of the present study, albeit at the cost of a more

involved model. As researchers begin to consider joint decision making, the need will become more pressing to

include wider issues pertaining to incentive alignment and coordination of activities in a distributed setting.

Our contribution is to provide a simple model of flexible teams composed of decision makers

whose behavioral and cognitive abilities appear quite realistic. Because our decision makers are fallible, they can

gain much from the structural support of the organization within which they are located. Our theory is also in

accordance with realistic assumptions about the size of management teams as it predicts that teams with four to

five members are superior both to smaller and larger teams. Briefly, the intuition for our claim is as follows.

Very small teams of fallible decision makers will make erroneous estimates and therefore experience many

periods with negative profits. In contrast, too large teams will tend to wait too long before they leap and thereby

miss many opportunities to gain from periods of mainly positive profits. In essence, our results highlight that the

challenge presented by turbulent market conditions can be met by fairly small flexible teams. With increases in

environmental turbulence from global competition we would expect future empirical studies to highlight the

prevalence and efficacy of flexible decision making structures.

In order to verify our claims, we need more data on the relation between organization structures

(or voting rules) and the quality of the decisions that are made. Why do MNCs pursue international expansion in

turbulent policy environments where opportunities are notoriously hard to evaluate? Why do they enter a

turbulent market in the first place? Our theory highlights that it would be important to understand how decisions

22

are actually made in such circumstances. Are decisions about entry and exit made by individual managers or by

teams? To the extent that teams are involved, what are the characteristics of the decision making structures (size,

flexibility, distribution of authority)? How do these characteristics impact on the quality of the decisions taken?

Despite much prior research on entry and exit decisions in MNCs, we lack evidence on the way joint decisions

are structured. The many studies of MNCs, including those on entry and exit decisions, have given scant

attention to the role of decision making structures. This is surprising since many decisions are made in teams

(such as boards, top management teams, ad hoc committees). An early experiment by Birnberg, Pondy & Davis

(1970) examined the effect of voting rules in stable teams and found that a simple majority rule outperformed

unanimity (equivalent to hierarchy) and veto (equivalent to polyarchy). However, we have very little knowledge

about the actual structuring of joint decision making in MNCs. More generally, the study of evaluation in

(business) organizations is an important but rather neglected topic (Knudsen & Levinthal, 2007).

The present study has identified a major limitation for current research on MNCs. Joint decision

making is of critical importance in practice, but has not received much attention in our theories. Our modeling

structure begins to fill this gap as it can be used to understand why alternative ways of structuring joint decisions

have an impact on performance. While our model begins to unravel how individuals decide in contexts defined

by other individuals on whom they are interdependent, there remains a huge gap in theories of organization

design. When MNCs and other organizations pursue strategic objectives, they do so by virtue of the way

individuals jointly generate, evaluate, and implement alternatives. Yet, theories of organization design seem to

take little account of the micro-foundations: of how individuals decide, act, or learn in a context defined by other

individuals with whom they have interdependence. A theory that predicts how individual choices are aggregated

and become stable would be a major contribution to the literature on international business and organization

theory more generally.

23

REFERENCES

Achrol, R.S. 1997. Changes in the theory of interorganizational relations in marketing: Toward a network

paradigm. Academy of Marketing Science, 25(1): 56-71.

Allen, L. & Pantzalis, C. 1996. Valuation of the operating flexibility of multinational corporations. Journal of

International Business Studies, 27(4): 633-653.

Bartlett, C.A., & Ghoshal, S. 1998. Managing across borders: The multinational solution (2nd ed.). Harvard

Business School Press: Boston.

Becker, M.C., & Knudsen, T. 2004. The role of routines in reducing pervasive uncertainty. Journal of Business

Research, 58(6): 746-757.

Bellman, R. E. 1957. Dynamic programming. Princeton University Press: Princeton, NJ.

Bigley, G.A. & Roberts, K.H. 2001. The incident command system: High-reliability organizing for complex and

volatile task environments. Academy of Management Journal, 44(6): 1281-1299.

Birnberg, J.G., Pondy, L.R. & Davis, C.L. 1970. Effect of three voting rules on resource allocation decisions.

Management Science, 16(6): B356-B372.

Bowman, E.H., Singh, H., Useem, M., & Bhadury, R. 1999. When does restructuring improve economic

performance? California Management Review, 41(2): 33-54.

24

Brynjolfsson, E. & Mendelson, H. 1993. Information systems and the organization of modern enterprise. Journal

of Organizational Computing, 3(4): 245-255.

Cannon, J.P. & Homburg, C. 2001. Buyers-supplier relationships and customer firm costs. Journal of Marketing,

65(1): 29-43.

Cavusgil, S.T., Kiyak, T., & Yeniyurt, S. 2004. Complementary approaches to preliminary foreign market

opportunity assessment: Country clustering and country ranking. Industrial marketing management, 33(7): 607-

617.

Cecil, H.W., Ciccotello, C.S., Grant, C.T. 1995. The choice of organizational form. Journal of Accountancy,

180(6): 45.

Charness, G. & Levine, D.I. 2002. Changes in the employment contract? Evidence from a quasi-experiment.

Journal of Economic Behavior & Organization, 47(4): 391-405.

Child, J. 1997. Strategic choice in the analysis of action, structure, organizations and environment: Retrospect

and prospect. Organization Studies, 18(1): 43-76.

Child, J. & McGrath, R.G. 2001. Organizations unfettered: Organizational form in an information-intensive

economy. Academy of Management Journal, 44(6): 1135-1148.

Christensen, M. & Knudsen, T. 2007. The human version of Moore-Shannon's Theorem: The design of reliable

economic systems. Available at SSRN: http://ssrn.com/abstract=996311. Accessed 24 June 2007.

25

Cyert, R.M. & March, J.G. 1963. A behavioral theory of the firm. Prentice Hall: Englewood Cliffs, NJ.

Daft, R. & Lewin, A.Y. 1993. Where are the theories for the new organizational forms? Organization Science,

4(4): i-vi.

Delios, A. & Henisz, W.J. 2003. Policy uncertainty and the sequence of entry by Japanese firms, 1980-1998.

Journal of International Business Studies, 34(3): 227-241.

Denrell, J. 2005. Why most people disapprove of me: Experience sampling in impression formation.

Psychological Review, 112(4): 951978.

Denrell, J., & March, J. G. 2001. Adaptation as information restriction: The hot stove effect. Organization

Science, 12(5): 523538.

Dixit, A. 1989. Entry and exit decisions under uncertainty. The Journal of Political Economy, 97(3): 620-638

Dixit, A. & Pindyck, R. S. 1994. Investment under uncertainty. Princeton University Press: Princeton.

Ethiraj, S.K. & Levinthal, D.A. 2007). Hoping for A to Z while rewarding only A: Complex organizations and

multiple goals, Working paper.

Heydebrand, W.V. 1989) New organizational forms. Work and Occupations, 16: 323-357.

Howard, R. 1960), Dynamic programming and Markov Processes. MIT Press, Cambridge, MA.

26

Ilinitch, A.Y., DAveni, R.A., & Lewin, A.Y. 1996. New organizational forms and strategies for managing in

hypercompetitive environments. Organization Science, 7(3): 211-220.

Knudsen, T. and D. A. Levinthal 2007. Two faces of search: Alternative generation and alternative evaluation.

Organization Science, 18(1): 39-54.

Knudsen, T. & Winter, S.G. 2007. An evolutionary model of spatial competition. Paper prepared for the DRUID

conference 2007.

Kogut, B. & Singh, H. 1988. The effect of national culture on the choice of entry mode. Journal of International

Business Studies, 19(3): 411-432.

Kole, S. & Lehn, K. 1997. Deregulation, the evolution of corporate governance structure, and survival. The

American Economic Review, 87(2): 421- 42

Lenox, M.J., S.F. Rockart and A.Y. Lewin 2006. Interdependency, competition, and the distribution of firm and

industry profits. Management Science, 52(5): 757-772.

Lenox, M.J., S.F. Rockart and A.Y. Lewin 2007. Interdependency, competition, and industry dynamics.

Management Science, 53(4): 599615.

Levinthal, D.A. & March, J. G. 1981. A model of adaptive organizational search. Journal of Economic Behavior

& Organization, 2: 307-333.

27

Lewin, A.Y., & Volberda, H.W. 1999. Prolegomena on coevolution: A framework for research on strategy and

new organizational forms. Organization Science, 10(5): 519-534.

Miranda, M.J. & Fackler, P.L. 2002. Applied computational economics and finance. The MIT Press: Cambridge,

MA.

Moller, K. & Rajala, A. 1999. Organizing marketing in industrial high-tech firms: The role of internal marketing

relationships. Industrial Marketing Management, 28(5): 521-535.

Nault, B.R. 1998. Information technology and organization design: Locating decisions and information.

Management Science, 44(10): 1321-1335.

Nault, B.R. & Tyagi, R.K. 2001. Implementable mechanisms to coordinate horizontal alliances. Management

Science, 47(6): 787-799.

Nelson, R.R. & Winter, S.G. 1982. An evolutionary theory of economic change. Harvard University Press:

Cambridge, MA.

Reuer, J.J. 2000. Parent firm performance across international joint venture life-cycle stages. Journal of

International Business Studies, 31(1): 1-20.

Rivkin, J. W. 2000. Imitation of complex strategies. Management Science, 46(6): 824844.

Schooler, R.D. 1974. The determinants of foreign opportunity attractiveness. Management International Review,

14(4/5): 115-127.

28

Sah, R. & Stiglitz, J. 1985. The theory of economic systems: Human fallibility and economic organization.

American Economic Review, Papers and Proceedings, 75(2): 292-297.

Sah, R. & Stiglitz, J. 1986. The architecture of economic systems: Hierarchies and polyarchies. American

Economic Review, 76(4): 716-727.

Sah, R. & Stiglitz, J. 1988. Committees, hierarchies and polyarchies. Economic Journal, 98(391): 451-470.

Volberda, H.W. 1996. Toward the flexible form: How to remain vital in hypercompetitive environments.

Organization Science, 7(4): 359-374.

Volberda, H.W. 1998. Building the flexible firm. Oxford University Press: Oxford, England.

Zahra, S.A. & ONeill, H.M. 1998. Charting the landscape of global competition: Reflections on emerging

organizational challenges and their implications for senior executives. Academy of Management Executive,

12(4): 13-21.

Zenger, T.R. & Hesterly, W.S. 1997. The disaggregation of corporations: Selective intervention, high-powered

incentives, and molecular units. Organization Science, 8(3): 209-222.

29

Figures

-10 -5 0 5 100

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Observed profit

Pro

ba

bili

ty o

f e

ntr

y

= 0.01

= 0

= 0.05

= 0.10 = 0.50

Optimal exit

Optimal entry

Local searcher

Figure 1: Levels of ability of fallible evaluator (=0.00, 0.01, 0.05, 0.10, 0.50). The five examples include an

agent who is virtually unable to distinguish whether observed profit should lead to entry (= 0.01), as well as an

agent who is much more certain about the relation between observed profit and the promise of entry (= 0.50).

The fallible agent becomes more able as the slope of the screening function increases. Figure 1 also shows the

local searcher ( ) and the two optimal points for entry and exit computed by dynamic programming.

30

-10 -5 0 5 100

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Observed profit

Pro

ba

bili

ty o

f e

ntr

y

H2 H5 H10 Optimizer

Single fallible:p(entry) = 0.75

H5:p(entry) = 0.24

Local searcher

Singlefallible

Local searcher:p(entry) = 1

Optimizer:p(entry) = 0

Figure 2: Levels of ability for hierarchical decision teams with two (H2), five (H5) or ten (H10) fallible

members compared to the optimizer and the local searcher. The teams employ fallible agents who are

represented with a dashed line in our example (=0.05). H5 is a hierarchy with five such employees. We can

read off the probability of entry at the y-axis for hierarchical teams and then compare these to the single

fallible agent, the optimizer and the local searcher. For example, the optimal point for entry is shown when

= 4, = 0.90, KExit=10, and KEntry=10. If observed profit is 5, then the optimizer will not enter the market. By

contrast, the single fallible agent would enter with p=0.24, H5 with p= 0.75 and the local searcher with p=1.

31

0 10 20 30 40 50 60 70 80 90 100

-50

0

50

Volatility, =4

Op

era

ting

Pro

fit

0 10 20 30 40 50 60 70 80 90 100

-50

0

50

Volatility, =10

Op

era

ting

Pro

fit

Optimizer entry

Optimizer exitOptimizer exit

Optimizer exit

Optimizer entryOptimizer entry

Optimizer entry Optimizer entry

Advantage for H5

Figure 3: Examples of profit distributions with low (= 4) and high volatility (= 10). In both cases, the

autoregression, is 0.90. Optimal entry points for KExit=10, KEntry=10 are also shown. The entry and exit

decisions for the optimizer are shown in both panels. The upper panel also shows a period where H5 has a

comparative advantage over the optimizer. In this period, H5 would earn a positive profit whereas the optimizer

is inactive an earns a profit of 0.

32

0 100 200 300 400 500 600 700 800 900 1000-6000

-4000

-2000

0

2000

5

Time steps

Cu

mu

lative

We

alth

2

1

L

O*

Figure 4: Performance of the optimizer, the local searcher and a single fallible evaluator (= 0.10), compared

with the performance of flexible decision teams with two or five fallible members. Example for = 4, = 0.90,

KExit=10, KEntry=10.

33

1 2 3 4 5 6 7 8 9 10-10

0

10

Volatility,

Ave

rag

e p

rofit

5

2

1

L

Optimizer

Figure 5: Performance of the optimizer, the local searcher and a single fallible evaluator (= 0.10), compared

with the performance of flexible decision teams with two or five fallible members. Example for = 110, =

0.90, KExit=10, KEntry=10.

34

ENDNOTES

i By turbulence we mean the frequency of events and their magnitude. A highly turbulent environment has rapid variation in

profit levels as well as high volatility of profit levels. ii An exception is Becker and Knudsen (2004) who find that routinization can help managers focus on issues that can be

solved, as opposed to wasting resources in the pursuit of problems that are beyond solution. iii

The Bellman equation is a common numerical method used to extract information on optimal expected rewards from a

Markov decision process such as the one considered here. This method, also known as dynamic programming, was

developed by Bellman (1957) and Howard (1960). iv

We use an approach that is similar to Knudsen & Levinthal (2007). v More generally, the structuring of decision flows among multiple agents becomes a critical issue when imperfect agents

face a difficult trade-off between treatment of multiple decision criteria and certain quality discrimination. A more

comprehensive guide to designing decision structures is provided in Christensen & Knudsen (2007). vi

In the Behavioral Theory of the Firm tradition, random walk models of the type used here are commonly used to capture

stochastic payoffs (Cyert & March, 1963; Denrell, 2005; Denrell & March, 2001; Levinthal & March, 1981; Nelson and

Winter, 1982). In order to provide a common reference point, the specification of the industry entry-exit model follows

closely the simple generic model provided by Miranda and Fackler (2002), ch. 8 and 9. The classical treatment in

economics is Dixit (1989) who offers a more advanced treatment that foreshadows our results. vii

Time subscripts should be self-explanatory and are therefore omitted in the remainder of the text in order to simplify the

exposition. viii

In solving for the Bellman equation, the expectation E, is represented by weighted discrete shocks with std. dev. . ix

As described below in the text, all results reported here were confirmed through comprehensive additional tests. x We analyse a situation with symmetrical entry and exit costs. If they were asymmetrical (e.g. lower exit costs), the trigger

points for entry and exit would also be asymmetric in which case the local searcher would be at more disadvantage. xi

Results are not shown here, but available upon request. xii

The so-called optimal decisions based on a dynamic programming approach is superior only if there is very modest

turbulence or very low entry/ exit costs. xiii

We use the hierarchy and the flat structure to simplify the analysis, but the results can be obtained for any flexible

decision structure that changes from more to less centralized. This is shown in additional robustness checks available

upon request. This proposition is valid within the modelling framework suggested here. It is due to the general property that

centralized decision structures are more conservative and decentralized decision structures more optimistic.