XNRNALOF Journal of Economic Behavior & Organisation Vol. 30 (1996) 2543
ELSEVIER
Competition
and the evolution of efficiency’
Tomas Sjiistriim and Martin L. Weitzman Depatiment of Economics, Harvard University, Cambridge, MA 02138, USA Received
10 September
1995; revised 9 December
1995
Abstract This paper presents an evolutionary model of the relationship between interfirm competition and intrafirm organizational or Xefficiency. We model Xinefficiency within the firm as a prisoner’s dilemma effortmonitoring problem, whose evolution is influenced by external competitive pressure from other firms. A closed form stochastic equilibrium displaying “survival of the fittest” dynastic cycles is derived and analyzed. The main result is that there exists a well defined sense in which
competition
is a surprisingly
powerful force for efficiency.
JEL classification: Cl2; LlO Keywords:
Competition;
Evolution;
Group selection; Prisoner’s dilemma;
Xinefftciency
1. Introduction There are several possible reasons to favor competition. The argument most frequently encountered in “practical” policy debates, business discussions, and media reports is that exposure to competition compels firms to exert greater effort at improving their efficiency. Without competition, employees of a firm take their customers for granted, do “business as usual” or “work to rule”, and generally lack the proper incentives to increase productivity, improve quality, develop new products, and so forth. Competition improves social welfare because it ensures that only the most efficient and innovative f”ums survive. According to this view, the Japanese invasion of the international automobile market improved global welfare not because automobiles were previously being produced from the wrong factor proportions, or because automobile prices were not close enough to marginal costs, or because the worldwide scale of automobile production ‘We are grateful to two anonymous
referees for helpful comments.
01672681/96/$15.00 0 1996 Elsevier Science B.V. All rights reserved I’ll: SO1672681(96)008402
26
T. Sjiistrtim, ML. Weitzmard J. of Economic Behavior & Org. 30 (1996) 2543
had been insufficiently large to take advantage of increasing returns. Rather, the improvement came because the Japanese invaders forced home manufacturers everywhere to make a better product at lower cost or else face bankruptcy and extinction. Such a story could easily be repeated for a large number of important actual situations. In this paper we study the balance between the forces of withinfirm Xinefficiency and betweenfirm competitive replacement. We model Xinefficiency as a prisoner’s dilemma problem caused by shirking among workers whose individual effort contributions cannot be perfectly monitored. The implied “free rider drift” toward a minimal level of Xefficiency is influenced by the betweenfirm force of competitive replacement, which itself is determined by the distribution of Xefficiency in the population of firms. We investigate the existence and stability of an equilibrium distribution of Xinefficiency among the firms. It is not our intention to present a particularly deep or innovative story about Xinefficiency or competitive takeovers per se. Rather, we simply assume that these forces are at work, and study the Schumpeterian cycles of creative destruction that result. The focus of the paper is on the statistical properties of equilibria, particularly the limiting distribution of efficiency. By assuming a continuum of firms, the resulting model becomes simple enough to allow us to calculate explicitly the equilibrium distribution of efficiency and to investigate how it depends on the parameters of the model. Two very different strands of literature relate to these issues. First, there is the standard principalagent theory with incomplete information. Hart (1983) Hermalin (1992), Horn et al. (1990), Scharfstein (1988, 1988). In this basically static approach, Xinefficiency is treated as maximizing behavior subject to incomplete contracts. This approach has provided some important insights. In particular, Scharfstein (1988a) and Hermalin (1992) show that more competition does not necessarily lead to more efficiency. However, our paper does not focus on the intricacies of the agency problem, and the mechanism leading to Xinefficiency is handled in much more of a bare bones manner than in the cited papers (basically, we simply assume the imitation of coworkers). Our paper is much more closely related to the evolutionary strand. At a sufficiently high level of abstraction, there is an isomorphism between the themes of this paper and biological models of group selection (Wilson, 1983). In these models, a population is divided into locally isolated groups, corresponding to our firms. Within each group, there is the usual individual selection, but there is also selection among groups. A group with a low average fitness may become extinct, and its site occupied by the members of some more successful group. (Alternatively, groups may contribute genes to a common “mating pool”, with successful groups contributing more genes to the pool.) The basic finding is that group selection may override the effects of individual selection. Independently of our research, a model like this has been developed by VegaRedondo (1993). In his model, workers play a coordination game with two Nash equilibria: an efficient “cooperative” equilibrium and an inefficient “uncooperative” equilibrium. Within each firm, all workers coordinate on some equilibrium, but firms where the workers coordinate on the inefficient equilibrium have an exogenously given lower chance of survival than firms where the workers play cooperatively. The long run outcome is that the whole population of firms end up at the same equilibrium, with the same level of efficiency.
T Sjtisttim,
ML
Weitzmanl J. of Economic Behavior & Org. 30 (1996) 2543
21
The behavior of our model is totally different. We have a prisoner’s dilemma within the firm and a stable (nondegenerate) equilibrium distribution of efkiency among firms, where each firm’s survival probability is endogenously determined. In contrast to VegaRedondo (1993), it is not an equilibrium for all firms to be maximally efficient. A firm’s chances of survival are endogenously determined by its capacity to withstand challenges from other firms. If a firm is more efficient than the competition, it will not be invaded, but (because the prisoner’s dilemma has a unique Nash equilibrium) it will eventually succumb to “freerider drift”. The more Xinefficient the firm gets, the more likely it will be replaced by a competitor. But only relative efficiency matters. In “partial equilibrium”, where one firm is subjected to an exogenously given distribution of challengers, there is a unique steady state distribution of efficiency for this firm. But in general equilibrium, where each firm lives in the same environment, there are many long run steady state distributions of efficiency. It turns out that they all correspond to different maximal efficiency levels. If there are no highly efficient firms present in the population, an inefficient firm is likely to survive for a long time, which supports a stable equilibrium with a low average level of efficiency. But if there are many efficient firms present, inefficient firms have a short life, which supports a stable “bootstrap” equilibrium with a high average level of efficiency. For a given maximal efficiency level, there exists a unique stable nondegenerate steady state distribution of efficiency. Thus, although the analysis is in the same spirit, the economic intuition behind our model is significantly different from VegaRedondo (1993). In a sense, it is not surprising that in a model like ours where more efficient firms replace less efficient firms there is some tendency to resist or slow down any reversions to inefficiency. However, once we consider the general equilibrium formulation it is not obvious that competitive challenges will provide anything but temporary relief from freerider drift. Since each firm’s chances of survival are endogenously determined by its capacity to withstand challenges from other firms, might it not be that all firms will lose efficiency together, so that the whole system will degenerate? But this does not happen. Thus, the novelty of this paper is to show rigorously that competitive challenges represent a surprisingly strong force for efficiency. Competitive challenges perform a “magic trick” by permanently maintaining, and even creating, efficiency in a system that otherwise would be running down over time. That the long run survival, and even increase, of efficient firms is actually quite “paradoxical” can be seen by looking at general equilibrium in a model with only a finite number of firms. With a large finite number of firms, the system will have similar properties to those described above, with one significant difference. Eventually, a coincidental bad string of almost simultaneous downward mutations will lower the maximal efficiency level of the finite system, and indeed over time this will cause average efficiency to ratchet down toward zero. Since there are no exogenous upward mutations, competitive challenges provide only temporary relief from freerider drift, and the whole population will eventually converge to zero. The finite system degenerates in the long run with probability one. In contrast, in the continuum case the equilibrium distribution is nondegenerate, and indeed will approach maximum Xefficiency if competitive challenges are sufficiently frequent. Of course, the continuum assumption is only a mathematical abstraction, an
28
I: Sjii,ttim,
M.L. We&mad
.I. of Economic
Behavior & Org. 30 (1996) 2543
idealization of the case where the number of firms is sufficiently large to make simultaneous downward mutations incredibly unlikely. But we think the important economic intuition, that a favorable distribution of efficiency can maintain itself by its own bootstraps, is best captured by our general equilibrium continuum model. Moreover, under reasonable assumptions there is no discontinuity in the limit as the number of firms grows large. Competitive challenges can maintain high efficiency in a large finite system if there is an arbitrarily small infusion of “new” high efficiency firms. We show in the appendix that, as long as there is an infinitesimal probability of exogenous “upward” mutations, a large finite world will have a unique steady state distribution that approximates the equilibrium for the continuum case.
2. Xinefficiency
as a prisoner’s
dilemma
problem
There is a large amount of evidence that internal, organizational, or Xefficiency is an empirically significant phenomenon (see Frantz, 1988). Following Leibenstein (1987), we understand Xinefficiency to be a prisoner’s dilemma problem about the level of “effort” exerted by a firm’s employees. Consider a world populated by infinitely many identicallysized, symmetrical “island firms”. Each island firm has the same fixed large number of employeeinhabitants. Let the variable x stand for an average employee’s effort level. (Throughout the paper, all variables are normalized per employee.) If every employee is working at effort level x, this is also the firm’s level of Xefficiency. A problem of observability and monitoring exists, since the firm does not know the contribution of any individual employee, although its aggregate level of Xefficiency is known to it.Let R(x) be the net revenue per employee
R’ > 0,” < 0
when each employee’s C(x)
c’ > 0,” < 0
(1) effort level is x. Let (2)
be the moneyequivalent effort cost or disutility that each employee incurs when working at effort level x. The “group optimal” level of effort for the firm and its employees is the value x* that maximizes R(x)  C(x), satisfying the first order condition: R/(x*) = C’(x*). The variable x could stand for many things. Essentially, x symbolizes “generalized effort.” Depending on the context, x might stand for: working harder, working smarter, not working to rule, not doing business as usual, taking initiative, managerial tautness, having an enthusiastic attitude toward new methods or equipment or potential customers, being innovative, thinking up new and better ways of doing things, cooperating, being altruistic, having selfless genes, and so forth. In a dynamic interpretation of the model, x could stand for the potential growth rate of the organization, which derives, say, from being more innovative or generally “more fit” for growth than competing organizations. With such an interpretation, a decreased x does
T. .Sj6sfriim, ML. We&mad
J. of Economic Behavior & Org. 30 (1996) 2543
29
not imply decreased efficiency in any absolute sense, but rather a failure to improve fast enough to keep up with the competition. (In this case R(x) would be something like the present discounted net revenue value of growth at rate x, while C(x) is the present discounted individual effortcost of an employee being sufficiently innovative to attain level x). Let lJ be the (fixed) reservation utility level for all islandfirm employeeinhabitants, i.e. U is what an island inhabitant could get in the best alternative to working for the firm (for example, by being selfemployed). A firm at Xefficiency level x pays its employees the smallest amount needed to retain them, namely
W(x) = Then long run profit (per employee)
v + C(x)
for the firm at Xefficiency
II(x) z R(x)  W(x) = R(x)  C(x) 
u
level x is2
(5)
II symbolizes “generalized performance” or, ultimately, relative fitness. Under the assumptions made so far, the Xefficiency level that would yield highest profits or greatest fitness to the firm is the level of worker effort x** that maximizes Eq. (5), satisfying the first order condition: II’@**) = 0
(6)
Comparing Eq. (6) with Eq. (5) and Eq. (3), we see that x* = x**. Thus the profitmaximizing level of Xefficiency is the groupoptimal level of individual effort. We will henceforth refer to x* as “the” optimal level of x. If x > x*, then a decrease in x increases efficiency. To avoid confusion on this point, in what follows we restrict attention to the interval [0,x*], where an increase in x indeed means higher efficiency. For any isolated islandfirm there is a tendency for x to drift down over time. The basic story behind this “freerider drift” could be told as follows. Suppose the firm’s employees are arrayed like discrete points around a circle. Every worker observes the two workerneighbors on either side, but the firm cannot observe any workers directly. Let every employee except for one of my neighborworkers be working at effort level x. Since there are a large number of workers, the Xefficiency of the firm is essentially x and each employee’s pay is W(x). Suppose my neighborworker is working at effort level y, observable to me. Then my behavior visavis my neighbor is essentially reactive. If my neighborworker has a lower utility level than mine, I act out of inertia and do not change my effort level. However, if my neighborworker is enjoying more utility than me, then I imitate the effort behavior that leads to such higher utility. My new effort level will be x’ = min{x, y}. In this environment any stochastic mutation causing a random change in individual employee effort will induce a downward ratcheting of Xefficiency over time. Unless the downwardratcheting spiral of free rider drift is countered by some other force, the long run steady state equilibrium of any isolated islandfirm is zero individual effort, which translates into zero Xefficiency. The group fitness of any isolated collection of
‘Without significant loss of generality, in what follows we assume away the issue of breakeven inequality conditions. (The analysis here of breakeven corner solutions seems neither interesting nor insightful.)
30
individual genes.
Z Sj6sttim, M.L. We&man/J. of Economic Behavior & Org. 30 (19%) 2543
units has a tendency
over time to deteriorate
3. Free rider drift vs. competitive
from what are essentially
selfish
pressure
This section presents the partial equilibrium model. We are analyzing the characteristics of a single islandfirm, called the “home” islandfirm, embedded in a large archipelago of potentially invasive islandfirms, called “challenger” islandfirms, having exogenously given characteristics. The next section will extend the analysis to a general equilibrium framework within which the characteristics of the challenger islandfirms are themselves endogenously determined. Free rider drift within the home islandfirm is caused by mutations, which are modeled as a homogenous Poisson process. With Poisson intensity ~1,a mutation occurs in a single employeeinhabitant of the home islandfirm. Formally, the probability of one mutation occurring in a given firm within any time interval [t, t + 61 is (7)
p6 + 46) and the probability
of two or more mutations lim
6O
occurring is of the order o(S). By definition,
00 =
0
6
.
The mutated employee works at effort level y, where y is a random variable drawn from an exogenously given cumulative probability distribution function G(y) with support [0,x*]. If the common effort level of the islandfirm had been x, and an employeeinhabitant mutates to level y, then the new common effort level is (as discussed in Section 2) x’ = min [x,y]. Thus, we assume this adjustment is sufficiently rapid relative to the mutation rate so that, in effect, the firm goes instantaneously to its new effort level x’. A similar assumption is made by VegaRedondo (1993), but (as pointed out by our referee) it is very strong. If this assumption is not made, there will be a distribution of effort levels not only among firms, but even within each firm. This would make the model richer. However, our assumption that imitation is fast is interesting because it creates the most unfavorable climate for efficiency. Any other assumption would lead to a higher average effort level. More generally, we could consider I independent mutations at intensity rates /Li= 1,2,...,1 and corresponding
probability
distribution
(10)
functions
Gi(y,) = 1,2,...,1.
(11)
It can readily be confirmed that the analysis goes through for this case ‘as if’ there were a single type of mutation at intensity rate
p&p, i=l
(14
7: Sjiisttim,
with a corresponding
M.L. Weitrnad
.I. of Economic Behavior & Org. 30 (1996) 2543
‘as if’ probability
distribution
31
function (13)
The aggregate mutation frequency parameter p might be influenced by a variety of internal cultural norms or sanctions within the home islandfirm. But however small it might be, so long as p is positive the mutation process described above, if unopposed, will induce free rider drift to a long run steady state level x=0. Let the proportion of challenger islandfirms having Xefficiency performance level no greater than z be H(z). The exogenous specification of H(.) constitutes the partial equilibrium assumption. The support of H(.) is contained in the interval [0,x*]. Depending on the context, the home islandfirm bumps into, interacts with, is exposed to, or is invaded by a randomly selected challenger islandfirm. Again this is modeled as a Poisson process. The probability of one competitive challenge occurring in any time interval [t, t + 61 is
and the probability of two or more challenges is of the order o(S). Since challenges are uncorrelated with mutations the probability of both a challenge and a mutation occurring in the same timeinterval [t, t + S] is of the order o(6). The coefficient X, called here the “challenge rate,” parameterizes the degree of interaction, accessibility, competitiveness, openness, trade liberalization, ease of takeover, or the like. Conversely, l/X is a measure of the degree of isolation, remoteness, insularity, or protection of the home firmisland. The following description would appear to represent the simplest possible reduced form model of the “takeover” process that occurs after a challenger islandfirm invades the home islandfirm. Suppose the Xefficiency of the home islandfirm had been x before the invasion incident, with corresponding fitness level II(x). Let the Xefficiency of the challenger islandfirm be z, with corresponding fitness level If(z). Then the challenger in effect “takes over” the home islandfirm and induces its own performance level if and only if
‘w > W)
(‘5)
z > x.
(16)
i.e. if and only if
The result of a competitive challenge to the home firm is “survival of the fittest” between invader and defender. The new efficiency level of the home firm (or whatever takes its place on the island) is x”, where x”  max [x, z].
(17)
We believe Eq. (17) is the simplest way to model “survival of the fittest.” It is perhaps easiest to think of the invading firm with a higher performance level displacing the existing firm by taking over its niche, in effect buying it out because the challenger is more profitable or more competitive or more fit, and therefore better able to outgrow the
32
7I Sjtisttim,
M.L. Weitzmad
J.
of Economic Behavior & Org. 30 (19%) 2543
home firm on its original territory. Alternatively, a story could be told about the home firm rising to the challenge of a more efficient invader by somehow imposing a corporate culture that increases the common degree of effort to the challenger level. In any case the outcome is ‘as if’ the fittest or more profitable firm survives the competitive encounter. Analogously to IO13), the case of J independently challenging invader subpopulations can be aggregated ‘as if’ there were a single invader population with ‘as if’ challenge rate
and ‘as if’ cumulative
population
distribution
function
Theorem I. The distribution of home islandfirm eficiency converges globally to the unique steady state probability distribution function given by
F(x) =pG(x) + A(1 ifH(x)
< 1, and F(x) = 1 ifH(x)
(20)
H(x))
= 1.
Prooj Let X, be efficiency in the home islandfirm at time t. Let XOE[O,X*]be a given initial level of efficiency. (More generally, we could consider a given initial distribution of efficiency.) Fix x E [0,x*]. If H(x) = 1 then no challenger is ever more efficient than X. Thus once the home firm falls below X, it can never climb above X. Since on the other hand a mutation must eventually cause it to fall below x, the probability that the home firm is less efficient than x converges to 1. Now suppose H(x) < 1. The relevant conditional probabilities are Pr(x,+h 5 x 1 x/ > x) = bG(x)S + o(S)
(21)
Pr(xt+b 5 x 1x, 5 x) = 1  X( 1  H(x))6 + o(S).
(22)
and
Let p(t) s Pr(x, 5 x) be the unconditional Eq. (22), we get P(r + 6) = p(tW(x,+a
that xt < x. Using Eq. (21) and
5 ,x I xt 5 x) + (1  p(t))Pr(x,+b
= p(t)( 1  X( 1  H(x))@ Rearranging
probability
+ (I  p(t))pG(x)S
and letting S&O, we obtain
dp(t) = dt
a(x)@(t)
 F(x))
I x I 4 > x) + o(6).
(23)
T Sjiisttim, M.L. W&man/J.
ofEconomicBehavior & Org. 30 (1996) 2543
33
where the constant a(x) is defined as a(x) 5 jJG(x) + X(1  H(x)). The linear differential U(X) > 0.
Eq. (24) has p(t) converging
exponentially
(25) to F(x) at damping rate
QED
As XIp is higher, the steady state distribution for the home firm is strictly improved in the sense of stochastic dominance. As X/p 4 0, then F(x) + 1 for any x E (0, n*], so that all probability mass goes to x = 0. As X/h ) 00, then F(x) + 0 for any x E IO,_?), so that all probability mass becomes concentrated at x = x*. The global stability is intuitively obvious. The internal force of free rider drift pulls down efficiency over time, while the external force of competitive pressure acts to push up fitness. For x close to x*, the home islandfirm is extremely vulnerable to free rider drift, since any mutation will pull down x. But for x close to zero, the home islandfirm is extremely vulnerable to challenge, since any invader can take over. These two opposing forces push the islandfirm away from extremes and move it toward a middle ground of Xefficiency. 4. General equilibrium
analysis
In the last section we treated the cumulative population distribution of the challenger islandfirms as a given function H(.), proceeding from there to derive the limiting distribution of homeisland efficiency F(.). We now endogenize H(.) in a general equilibrium setting with symmetric islandfirms. In effect, the general equilibrium model is the same as the partial equilibrium model except for the added restriction that the home islandfirm should share identical statistical properties with the other islandfirms. Crudely speaking, the general equilibrium model of this section is the partial equilibrium model of last section with H(x) chosen to satisfy the additional restriction H(x) E F(x), a substitution which essentially turns Eq. (20) into either F(x) = pG(x)/X or F(x) = 1. (The latter condition for x = 0 means that F(0) = 1 is a stationary distribution, since there are no upward mutations.) There is an infinitely large number of perfectly symmetric islandfirms. Let @(x; t) be the fraction of island firms having Xefficiency less than or equal to x at time t. There is no exogenous inflow of new islandfirms into the system, nor are there any “upward” mutations in existing firms. Hence the maximal efficiency at any time t > 0 can be no greater than the maximal efficiency at time 0. Therefore, the support of the initial distribution of efficiency, @(.;O), plays a significant role in determining the limiting distribution. Without loss of generality we suppose the greatest element in the support of a(.; 0) is x*. If the maximal initial efficiency is x’ < x*, efficiency levels higher than x’ are irrelevant, and we can replace x* by x’ throughout with identical results. We make no other assumptions about the initial distribution of firms a(.; 0). As before, with exogenously given intensity p a mutated employee appears who works at effort level y, where y is a random variable drawn from the given distribution G(y). We suppose G(x*) > 0 so that harmful mutations can occur. If an islandfirm is at Xefficiency level x and an internal employeemutation occurs to level .Y, the new
34
T. Sjiistriim, M.L. Weirman/ J.
of Economic Behavior & Org. 30 (1996) 2543
Xefficiency of the firm will be x’ = min{x, y}. Invasions by a challenger firm occur at intensity X. The invader firm is drawn randomly from the population of islandfirms. At time t the probability that a challenger has an efficiency below x is @(x; t). The reduced form outcome of the invasion of a home firm at efficiency level x by a challenger firm at efficiency level z is a “survival of the fittest” efficiency level x” = max{n, z}. Challenges occur independently, and are uncorrelated with mutations. We now investigate what happens to @(x; t) as t + 00. Consider the population distribution function Q(x) defined as follows: Case I: X > p. Then a(x) s Case 2:
*
if x = x* ifx
l
ifx>x ifx
{ fG(x)
X < ~1. Then Q(x)
E
where X is defined to be the (smallest)
{
fG(x)
value of x satisfying ;G(n)
Case 3:
(27)
= 1.
(28)
Q(x) = G(x).
(29)
X = p. Then
Our basic result in the general equilibrium formulation is that, for all x and from any initial distribution, @(x; t) converges to Q(x). The intuition behind this result is simple. Suppose only a few firms have efficiency greater than x, i.e. @(x; t) is close to (but not equal to) one. More precisely, suppose pG(x)/X < @(x; t) < 1. These more efficient firms essentially win every contest and will be very successful in replacing other lirms. In fact such replacements will occur at a rate X(1  @(x; t))Q(x; t), which is the probability that a firm with efficiency greater than x challenges a firm with efficiency less than x. Firms with efficiency greater than x are mutated down below x at rate p( 1  @(x; t))G(x). The first effect dominates, implying a falling @(x; t), if and only if X(1  @(x;t))@(x;t) > pG(x)(l  Q(x; t)). But this holds by the assumption pG(x)/X < Q(x; t) < 1. Thus, the replacement of less efficient firms by more efficient firms dominates the downward drift of the more efficient krns. On the other hand, free rider drift will dominate, implying an increasing Q(x; t), if @(x; t) < ,uG(x)/X. In a dynamic equilibrium, @(x; t) = pG(x)/A w h enever @(x; t) < l.We now provide a formal statement and proof of the theorem. Theorem 2.
For all x E [0,x*], lim,,,+(x;
t) = (a(x).
Prooj Fix x E [0,x*] and define Q(t) 3 1  @(x; t). It is obvious that 9 is continuous. If x = x*, then Q(t) = 0 for all t, and the conclusion of the theorem holds for this x. So suppose 0 5 x < x*. By assumption, the support of Q(x; 0) contains x* > x, so 9(O) = 1  @(x;O) > 0. We need to show Q(t) ) 1  @j(x), Consider a firm with efficiency less than (or equal to) x at time t. Challenges to this firm occur with Poisson intensity X, and a challenge at time t will raise the challenged
T SjSsttim,
ML. WeitzmadJ.
of Economic Behavior & Org. 30 (1996) 2543
35
firm’s efficiency level strictly above x if and only if the challenger’s efficiency is strictly higher than x, which it is with probability q(t). As Q(r) is approximately constant in a short interval [t, t + S], the probability that a firm with efficiency less than x at time t has efficiency strictly greater than x at time t + 6 is Xmqt) + o(S).
(30)
As the proportion of firms with efficiency less than x at time t is 1  Q(t), the measure of firms that rise above x in the interval [t, t + S] is [email protected](t)(l
 Q(t)) + o(S).
(31)
The proportion of firms with efficiency strictly exceeding x at time t is Q(t), and the probability that such a firm has efficiency no greater than x at time t + 6 is pG(x)S + o(S).
(32)
Thus we obtain 1  Q(t + 6) = (1  Q(t))  MQ(t)(l Rearranging,
dividing
 Q(t)) + ‘I’(t)pG(x)G + o(S).
by 6, and taking limits as S + 0 yields the differential
F = Q(t)(X There are two subcases
(33)
equation
 pG(x)  A*(t)).
to consider:
SubCase 1: X  pG(x) 5 0. This condition implies x 2 X (x 2 x was defined Eq. (28) ). Since the right hand side of Eq. (34) is strictly negative so long @(t) > 0, a(t) + 0 = 1  a(x) for this subcase. SubCase 2: X  pG(x) > 0. This condition implies rewrite Eq. (34) as a logistic differential equation’
in as
x < x and hence (a(x) < 1. We
where r(x)  X  pG(x) > 0
(36)
K(x) z 1  (a(x) > 0.
(37)
and
It follows that Q(t) increases whenever 0 < Q(t) < K(x), and decreases Q(t) > K(x). Since Q(O) = 1  @(x;O) > O,@(t) f 1  $(x). QED
whenever
The proof of Theorem 2 shows that the behavior of 1  @(x;t) for @(x; t) > (p/X)G(x) is logistic growth. Thus, if @(x; t) is close to one, so that almost all firms have efficiency below X, then the increase in the number of firms with efficiency greater than x is very fast, essentially growing exponentially at rate X  pG(x). The few existing firms with efficiency above x are tremendously successful at “invading” other firms before the invaders themselves succumb to freerider drift. On the other hand, when
36
Z Sjktrtim,
M.L. W&mad
J.
ofEconomic Behavior & Org. 30 (1996) 2543
@(x; t) is close to zero, meaning almost all firms are more efficient than x, then the decline in the number of such efficient firms is basically exponential, at decay rate pG(x). If @(x; t) is close to a(x), the change in (a(~; t) is very slow. The ratio X/,U plays a critical role, as it does in the partial equilibrium case. An increase in A/p unambiguously improves efficiency, in the sense of stochastic dominance. When X > p, challenges occur sufficiently frequently, relative to freerider drift, that in the long run the fraction of islandfirms massed together at the optimal effkiency level x* is 1  p/X, a positive number strictly increasing in X/p. The fraction of islandfirms at suboptimal levels x < x* is approximately /*./A, so that inefficiency disappears altogether as (X/p) + cc. When A < p, challenges occur so infrequently, or free rider drift is sufficiently strong, that all of the most efficient firms are eliminated in the long run. The maximal efficiency level that survives is X < x*, as defined by Eq. (28). Note that x declines strictly in X/p, approaching total Xinefficiency as (A/p) + 0. We assumed above that x* is in the support of a(.; 0). However, if instead x’ < x* is the greatest efficiency level initially present, then Theorem 2 is still true if x* is replaced by x’ throughout and 2629) are modified in the obvious way. In this case, if x 5 min{x’,x}, then (a(~; t) + pG(x)/X. Otherwise, cP(x; t) + 1. As a special case, if x’ = 0 is the greatest efficiency level initially present, then the population of islandfirms remains forever at the degenerate steady state with all probability mass concentrated on x = 0. This is because there is no mechanism by which more efficient firms are introduced into the system if they are not present initially. But if the initial distribution is not degenerate, we do not converge to the degenerate distribution. The reason is the bootstraps provided by highefficiency firms “invading” lowefficiency firms. As, was noted in the introduction, the assumption of an infinite number of islandfirms is important because it guarantees an adequate supply of efficient firms. In the finite case, a coincidental bad string of almost simultaneous downward mutations may lower the maximal efficiency level, and indeed over time this will cause average efficiency to ratchet down toward zero. In the long run the finite system degenerates with probability one.3 Fortunately there are various ways of obtaining limit results for finite systems. It turns out that under reasonable assumptions there is no discontinuity in the limit. In the appendix we show that a large finite world will have a unique steady state distribution that approximates a(.) if there is but an infinitesimally small probability of exogenous “upward” mutations. Suppose upward jumps in efficiency occur at a Poisson rate E > 0. and suppose the number of islandfirms is n < 00. In this case there exists a unique stationary distribution of efficiency. Theorem 3 in the appendix establishes that this distribution approximates a(.) for small E and large n.
5. Conclusion No economist will be surprised by the idea that competition is a force for efficiency. Furthermore, it is hardly surprising that in a model where more efficient firms replace less ‘In a loose way the finite case exhibits characteristics roughly analogous to genetic drift in biology, whereas the infinite case has some rough analogies to HardyWeinberg equilibrium.
Z Sjiisttim, M.L. Weitznud J. of Economic Behavior & Org. 30 (1996) 2543
31
efficient firms there is some tendency to resist or slow down any reversions to inefficiency. However, in our model competitive challenges represent a surprisingly strong force for efficiency. Competitive challenges perform a “magic trick” by maintaining, and even creating, efficiency in a system that otherwise would be running down over time. These results are derived for the case of an infinite number of firms. Competitive challenges can also maintain high efficiency in a large finite system if there is an arbitrarily small infusion of “new” high efficiency firms. Appendix A In this appendix we consider the case of a finite number of islandfirms. We show that if there is a small probability of “upward” mutation and a large but finite number of firms, the unique stationary distribution approximates arbitrarily closely G(x) (as defined by Eqs. (2629)). Let n < cc be the number of firms. Fix x < x* and let i 5 n be the number of firms with efficiency strictly greater than x at time t. Under the assumptions made previously, the probability that one of the i firms with efficiency strictly above x is brought down below x during the interval [t, t + S] is pi,il(6)
= &SG(x) + o(S) = piS + o(S)
(38)
where pi s ipG(x).
(39)
The probability that more than one firm is brought down is o(6). Consider now one of the ni firms with efficiency below x at time t. Referring to the arguments developed earlier, the probability that a challenge raises efficiency in this firm strictly above x during [t, t + 61 is
x6;+o(6)
(40)
since i/n is the proportion of the firms with efficiency strictly greater than x at time t. As before, the probability of multiple challenges is of the order o(6). Now we make an important new assumption. With an exogenously given Poisson intensity E(X), an “upward” mutation raises a firm’s efficiency above x, even in the absence of any challenge. We suppose E(X) > 0 for x < x* and E(X) = 0 for x 2 x*. (In contrast to the earlier analysis, we do not need to assume that firms with efficiency x* are present initially.) Under this assumption, the probability that one of the n  i firms with efficiency below x E (0, x*) is brought up above x is
Pi,i+l(S)=(ni)
(Ai+i EX) ( ) S+0(6)=XiS+O(S)
(41)
where Ai_(.i)(AA+E(X)).
The probability
that more than one firm is brought up is o(6).
(42)
38
7: Sjiisttim,
ML
J. of Economic Behavior & Org. 30 (1996) 2543
Weitzmd
In what follows, we write E(X) = E for convenience Define P&i(S)
=
l Pi,iI
(n will be fixed throughout),
(6) Pi,i+l(@.
(43)
Let pi(t) be the unconditional probability that exactly i firms have efficiency greater than x. Using Eq. (38) and Eq. (41), =Pil(r)Pil,i(Q
Pi([email protected]
+Pi+l(t)Pi+l,i(G)
+pi(r)pi,i(h)
strictly
+0(S) (4)
=Pil(f)XilS+pi+l(t)/h+16+pi(t)(l
Rearranging
Ai6/JiS)
+0(b).
and taking limits as 6 + 0, we get dPi(t) =Pil(t)Xil dt
+pj+l(t)pj+l
pj(t)(Xi
+ pi).
In steady state Eq. (45) has to equal zero for i = 0, 1, . . . , n. This leads to the system of equations PiIX;1
Pi(A
+Pi+lPi+l
+ Pi)
=
0
(46)
where pi is the steadystate probability that precisely i firms have efficiency strictly greater than x. This system can be solved recursively. For i = 0 we get (using the fact ~i(t) = 0 for all t)  Po(Xo + ho) = 0
PIPI
(47)
so that Pl
Continuing
recursively,
x0 + PO
= .PO Pl
Pl
we obtain steady state probabilities Pi =
fori=O,l,...,
riPO
(49)
n,where 7Ti =
Aa . Pl
fori=
= xopo
Xl ’ . . . Xi1 . P2
. ...
Pi
(50)
1,2,...,nand 7ra = 1
Since the probabilities
epi i=O
Combining
(51)
must add to one, =p(je7r, i=O
= 1.
(52)
Eq. (49) and Eq. (52) we see that pi = xco..
(53)
To indicate that most variables defined so far depend on n and E, write pi = pi(n, E) for the ith steady state probability, ri = ni(n, E) for the ith coefficient defined by 50 and 5 l),
T. Sjijstriim, ML. Weitzmad J. of Economic Behavior & Org. 30 (1996) 2543
etc. For j > 0, Eq. (50) can be rewritten
q(v)
=
as
J$ni)(Ai+&) &(n, E) n Pi+1 (n,E) = (i + l)~G(x)
jl
i=o
i=o
39
=
jl
x
i
_ (54) j II CLG(X) n+llc(x) ’
0 ( n
&
i=o
We want to prove that the finite world approximates the infinite world if E(X) is small for each x and n is large. Let a)(x) be defined by 2629). The number of firms with fitness greater than x is a birthdeath process with steadystate probabilities given by Eq. (54). Theorem 3 shows that as the world grows, in this unique steadystate all probability mass is concentrated in a neighborhood of 1  (a(x). This neighborhood can be made arbitrarily small by choosing E(X) sufficiently small. Theorem 3. have
For any S > 0, there exists E* > 0 such that if E < E*, then as n + cw we
(55) where the sum is taken over the set {i : 1  Q(x)  6 < i/n < 1  Q(x) + 6). Proo$ Fix x and define q* = 1  a(x), where Q(x) is given by 2629). If q* = 1 then G(x) = 0 so that there is no possibility of harmful mutations: in this case the proof is trivial. The case q* = 0 needs a few straightforward modifications of our argument, which we leave to the reader. Hence, suppose 0 < q* < 1. Throughout the proof, q will represent the proportion of islands with efficiency greater than x, so that the formulas will make sense only under the implicit assumption that qn is an integer between zero and n. By Stirling’s approximation (Courant, 1988, p.361),
(1 +_L) < ( ,!,!,);) l
;
for all n and q E (0,l). By a straightforward log
qq1
computation,

q)(lq&lP(l q)7ei2 .I,&
(56)
40
T. Sjtisftim,
ML. WeirzmadJ.
of Economic Behavior & Org. 30 (1996) 2543
Similarly,
(58) Using Eq. (56) and Eq. (57) in Eq. (54), we have an upper bound O
qy
1+ & (
 &T~)“2
>
x ((s)‘(
=
1  4)(14)n(2q(l
for nqn(n, &> for
1 + :)“Aexp[q))’
(2q(l  q)m“2(fyq,E))“:
1+$ (
(59)
>
where R(q,&) z qq1
 q)(lq)
(s)q(
1
+gi’*erp(q). (60)
Using Eq. (56) and Eq. (58) in Eq. (54), we get (for 0 < i < n) the lower bound
T(V)
L
(q+, 4)”
(61)
(2TrP$‘2K(i, n, E) ’
where K(i,n+)
= ( 1 + &)(I
+$)(l+g)($
i))“‘.
(62)
7: Sjiisrtim, M.L. We&man/J.
41
of Economic Behavior & Org. 30 (1996) 2543
From Eq. (57) and Eq. (58) it follows that
(&)(n(l,r))” i T”hE)
=
f&$+&J
I P(l,&>)“,
(63)
where by definition fl(l,&)
= (S)
(64)
(1 +;)l’*exp(l).
We also define C!(O, E) = 1 and note that 7r&, E) = 1 = 1” = (G(O,E))“.
(65)
With these definitions a(.,~) is continuous on [0, 11. We will find q that maximizes Q(q,&) for given E > 0. For q E (0,l) we have log R(q, E) = qlogq
 (1  q)log( 1  q) + qlog (S)
+;1og(+I)
4
and
alogwq,E) =
1
logq1+1og(1q)+l+log
as = Moreover,
log(?S)
(67)
for q E (0, l),
a2hmh E) =
E +
xq2
 (1  q)q(Xq + E) < O.
aq2
(68)
Setting Eq. (67) equal to zero yields Iqxq+E =. q PW)
1 (69)
Eq. (69) has a unique positive root q = q(E) E (0,l) which by Eq. (68) is the unique maximizer of R(q, E) in the interval [O,l]. It is easy to check that q(E) + q* as E ) 0. It can also be verified that E_fiT_q, fl(q,E) Fix 6 > 0. By Eq. (68), R is uniformly of (E,q) = (O,q*), i.e.
= exp (q*)(l
> 1.
(70)
strictly concave in q in some small neighborhood
a2wqq, El as2
 q*)l
1 < 2(1
4’)
(71)
in some such neighborhood. By Eq. (71) and the fact that q(E) + q* we can find E’ > 0 andA>OsuchthatifeS,tben
q&E) 5 fl(cJ’,&) 2A.
(72)
42
Z Sjtisrtitn,
M. L. Weitnnad
J. of Economic Behavior & Org. 30 (19%) 2543
But Eq. (70) implies, by definition of double limit, that there exists TJ> 0 and E” > 0 such that if 1q’  q* (< u and E < E”, then R(q’,&) Let E* = min{&‘,8’}.
Combining
> n(q*, E)  A.
Eq. (72) and Eq. (73) yields WI’, E) > Q(q,E)
+ A
whenever 1q’  q* (< u, ( q  q* I> 6, and E < E*. For each n > l/u we can select a positive integer i(n) such that 1i(n)/n From Eq. (74), if ( i/n  q* (> 6 and E < E”, then
Since fl(i(n)/n)
(73)
(74)  q* I< v.
is bounded as n f cc, Eq. (75) implies that there exists B < 1 such that (76)
whenever I i/n  q* I> 6. Note that this includes the cases i = 0 and i = n. The final step is to compute, for E < E* and n > l/v,
c
PiCn,&)
{ i:l(i/n)q’)>6}
< K(i(n),n,E)
+ KCiCnj
(2nn)1/2(n(q,E))n
(77)
n El C~i:O6~(l
> ,
+ t> (f (’  ;i)2nn)Y1’2(n(f~E)ln
(27rn)“2(fl(+),,))fi < 2K(i(n),n, ~)((27rn)“~ + n2)B” In Eq. (77), the first equality uses Eq. (53), the first inequality is obvious, the second inequality uses Eq. (59) and Eq. (61) the third inequality uses 63,65 and 76). But K(i(n), n, E) is bounded as n + co, and B < 1. Therefore, the final expression in Eq. (77) converges to zero. Therefore, Eq. (55) holds. QED References Courant, R., 1988, Differential and Integral Calculus, Vol. I, (Wiley, New York). Fran& R.S., 1988, XEfficiency: Theory, Evidence and Applications, (Kluwer, Dordrecht). Hart, O.D., 1983, The market mechanism as an incentive scheme, Bell Journal of Economics, 14, 366382. Hermalin, B., 1992, The effects of competition on executive behavior, Rand Journal of Economics, 23, 350365.
T Sjiistnh, M.L. We&man/J.
of Economic Behavior & Org. 30 (1996) 2543
43
Horn, H., H. Lang and S. Lundgren, 1990, XInefficiency and International Competition, IIES Seminar Paper No 480, Stockholm University. Leibenstein, H., 1987, Xefficiency theory, The New Palgrave: A Dictionary of Economics, (MacMillan, New York). Scbarfstein, D., 1988, Product market competition and managerial slack, Rand Journal of Economics, 19, 147155. Scbarfstein, D., 1988, Tbe disciplinary role of takeovers, Review of Economic Studies, 55, 185199. VegaRedondo, F., 1993, Competition and culture in an evolutionary process of equilibrium selection: A simple example, Games and Economic Behavior, 5, 618631. Wilson, D., 1983, Tbe group selection controversy, Ann. Rev. Ecol. Syst., 14, 159187.