Is Failure Good

Strategic Management Journal Strat. Mgmt. J., 26: 617–641 (2005) Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/smj.470

IS FAILURE GOOD? ANNE MARIE KNOTT1 * and HART E. POSEN2 1 Robert H. Smith School of Business, University of Maryland, College Park, Maryland, U.S.A. 2 The Wharton School, University of Pennsylvania, Philadelphia, Pennsylvania, U.S.A.

Approximately 80–90 percent of new firms ultimately fail. The tendency is to think of this failure as wasteful. We, however, examine whether there are economic benefits to offset the waste. We characterize three potential mechanisms through which excess entry affects market structure, firm behavior, and efficiency, then test them in the banking industry. Results indicate that failed firms generate externalities that significantly and substantially reduce industry cost. On average these benefits exceed the private costs of the entrants. Thus failure appears to be good for the economy. Copyright  2005 John Wiley & Sons, Ltd.

INTRODUCTION The entrepreneur who seeks private benefit in the face of long odds and succeeds is a sung hero in the innovation literature. When successful, the entrepreneur’s innovation generates economic benefits above and beyond her private benefits. By filling unmet needs or satisfying old needs in new ways, the entrepreneur fuels a process of creative destruction (Schumpeter, 1942). This process increases social welfare either by displacing less effective incumbents or by forcing incumbents to match the entrant’s innovation. But what about entrepreneurs who fail ? While they suffer private losses, does their failure generate any economic benefits to offset those losses? It seems possible that the same creative destruction process fueled by successful ventures may also be fueled by unsuccessful ventures. In short, the process may be blind to the outcome of a particular Keywords: failure; entry; efficiency; spillovers; selection; competition

∗ Correspondence to: Anne Marie Knott, Robert H. Smith School of Business, University of Maryland, 4515 Van Munching Hall, College Park, MD 20742, U.S.A. E-mail: [email protected]

Copyright  2005 John Wiley & Sons, Ltd.

venture. If so, then failed entrepreneurs may be as heroic as successful entrepreneurs. In fact, there is some preliminary evidence to that effect. In a longitudinal study of Texas sales tax receipts, Hicks (1993) found that employment growth and wages increased as the half-life (the length of time to achieve a 50% decrease in the population) of firms decreased. Thus, at first glance, it appears that failure may be beneficial. However, while the benefits are intriguing, the direction of causality is unclear. Even if failure ‘causes’ growth, the causal mechanism is not obvious. A policy of arbitrarily killing off firms to achieve the benefits is most certainly ill advised. The primary goal of this paper is to determine if the economic benefits are real. Does excess entry generate these benefits or are they merely artifacts of a process that jointly produces entry, failure and growth? If the benefits are real, the secondary goal is to understand the causal mechanisms producing them. This understanding may lead to policies that stimulate ‘beneficial failure’. To examine the broad question, we focus on three potential mechanisms through which failure (excess entry) might affect market structure and thereby efficiency growth. The first mechanism is

Received 7 April 2004 Final revision received 25 January 2005

618

A. M. Knott and H. E. Posen

a selection effect—firms surviving from a larger pool (more excess entry) ought to perform better on average than the same number of firms surviving from a smaller pool (Demsetz, 1973; Jovanovic, 1982). The second mechanism is a competition effect—the more firms in a market, the greater the stimulus to innovation (Barnett and Hansen, 1996; Peretto, 1999; Aghion et al., 2001; Knott, 2003). The third mechanism is a spillover effect—the knowledge produced by excess entrants while ‘wasted’, in that it is no longer appropriable by the failed firm, may be captured by survivor firms through spillovers. The effects of interest here are the durable effects of excess entry, i.e., the effects due to firms that enter, but then fail. This paper characterizes the potential impact of each of the three mechanisms and then conducts an empirical test of the relative power of each mechanism in explaining the economic impact of failed firms. This test is conducted in the banking industry post deregulation. The industry was chosen because it is fragmented, has a substantial knowledge component, and is marked by substantial failure. Fragmentation is important because it allows us to compare differences across markets while controlling for technology. The knowledge component is important because the failure mechanisms rely on efficiency improvements. These improvements become more likely as knowledge and human capital come to dominate physical capital. We examine banking efficiency for all 50 states plus the District of Columbia over the period 1984–97. Analysis is done in two stages. In the first stage, we characterize the annual efficiency of each firm relative to a non-moving frontier. In the second stage, we model firm efficiency in each year as a function of the three mechanisms plus a set of controls. The approach we take is similar to that taken in studies examining the micro-level components of aggregate productivity growth (see Bartelsman and Doms, 2000, for a review). The paper proceeds as follows. First we review the failure literature to extract hypotheses about the economic impact of failure. This review yields three mechanisms. We characterize the impact of each of these mechanisms on cost dynamics. Following that we discuss the banking industry and the empirical model that appends the mechanisms to traditional models of banking efficiency. We conclude with results and discussion. Copyright  2005 John Wiley & Sons, Ltd.

EXCESS ENTRY AND FAILURE Approximately 10 percent of all firms in the United States fail each year (U.S. Small Business Administration, 1999). However, the rate of new firm creation is roughly 11 percent of existing firms, so the stock of firms grows at a net rate of 1 percent. These aggregate numbers appear rather tame—suggesting that each firm faces a 10 percent hazard rate, and also suggesting that there is a healthy rate of churn—old entrenched firms being replaced by vibrant young firms. However, these aggregate numbers mask interesting dynamics. Failure is disproportionately drawn from new firms. Data from the U.S. Census Business Information Tracking Series (Headd, 2003) indicate that 51 percent of firms exit within their first 4 years. Thus failure risk looms large for entrants, and is relatively non-existent for established firms. Moreover, the phenomenon looks less like churn than competition for the right to produce in perpetuity. The literature on failure examines the causal mechanisms for failure: Who are the 10 percent and why do they fail? We examine a very different question: Is 10 percent the right number, or would the economy (producers and consumers jointly) be better off with either higher or lower failure rates? Perhaps the failure rate is too low—more failure will produce greater economic growth by creating more industries and unseating less effective firms in established industries. Alternatively perhaps the failure rate is too high, because investments in failed new ventures produce no economic benefits and thus are merely a misallocation of resources. If the failure rate is too low, perhaps we should subsidize entry. If the failure rate is too high, perhaps we should tax entry, so that only firms assured of success will enter. Both the amount question and the mechanism question have public policy implications—the amount question determines whether intervention is warranted; the mechanism question determines the form intervention should take. We are far from answering the policy question of how to influence failure rates. However, we believe we can begin the debate by tackling the question of whether failure produces any economic benefits. If so, what are these benefits, and how substantial are they? Before we begin, we want to frame the discussion and our empiricism, by reviewing the literature on failure. Strat. Mgmt. J., 26: 617–641 (2005)

Is Failure Good? Failure mechanisms Probably the best approach to reviewing the literature on failure is to work across disciplines, but within level of analysis. There are two relevant levels: market (population) and firm (organization). They are related in that market phenomena determine when failure is most likely to occur, while firm phenomena determine who is most likely to fail. Industry-level studies Studies of failure tend to find that industries have life cycles: an emergent stage where demand is uncertain and potential firms are cautious, a growth stage where demand exceeds supply and many firms enter, and a shake-out where demand stabilizes. The shake-out occurs because market capacity grows in anticipation of continued demand growth. However, when growth subsides, capacity exceeds realized demand and firms compete for slots in the market. In evolutionary economics (Nelson and Winter, 1982; Klepper, 1996) firms exit when their cost exceeds the decreasing market price. The firms surviving the shake-outs are those with superior technical capability who throughout the industry’s life invested in process innovation to drive down cost. The firms making these investments expand output to fully capture the benefits of lower unit costs. The continuously shifting cost curve combined with the continuously expanding scale of these firms hastens the exit of later entrants or less R&D-intensive rivals. Thus evolutionary economics offers an endogenous model of industry evolution that simultaneously predicts industry concentration and mortality rates. What evolutionary economics fails to explain, however, is the fact that in most industries entry and exit continue in steady state. This steady-state churning is the phenomenon of interest here. Population ecology takes equilibrium industry size as exogenous, but arrives at similar conclusions to evolutionary economics regarding mortality. These are embodied in two density hypotheses. The first hypothesis pertains to contemporaneous density. The more firms competing for a given set of resources, the higher the mortality rate (Hannan and Freeman, 1989; Carroll and Hannan, 1989). A second hypothesis pertains to the density at founding. This hypothesis holds that firms founded during adverse environmental conditions suffer an Copyright  2005 John Wiley & Sons, Ltd.

619

initial period of high mortality, but firms surviving the period have lower subsequent mortality than those founded in more munificent periods (Baum and Mezias, 1992; Swaminathan, 1996; Delacroix and Rao, 1994; Bamford, Dean, and McDougall, 2000). This is because firms surviving a competitive environment are more fit than those chosen randomly. What these industry-level studies share is an assumption of excess entry and a selection hypothesis that high-performing firms will drive out lower-performing firms. An intriguing but distinct question is why there is any excess entry. One very interesting study (Camerer and Lovallo, 1999) proposes that excess entry arises from hubris regarding entrepreneurs’ ranking in a capability distribution, and then demonstrates the phenomenon experimentally. These experimental findings are corroborated by empirical data (Wu and Knott, 2005). We ignore issues of cause and degree of excess entry, and focus exclusively on whether there are economic benefits from the excess. Firm-level studies Firm-level studies of failure hold that there are firm life cycles and that these affect mortality as do industry life cycles. Typically young firms have the highest mortality.1 In organization theory, this observation is attributed to Stinchcombe (1965). His ‘liability of newness’ hypothesis holds that new firms have higher failure rates than older firms due to: (a) the cost of learning new roles and tasks; (b) constrained resources; (c) lack of informal communications structures; and (d) lack of formal connections with customers and suppliers. There is substantial empirical support for the decrease in mortality associated with age. However, it is unclear whether age or size (which is highly correlated with age) is the driving factor (Hannan, 1998). An alternative interpretation of the same observation is that firms have heterogeneous but unknown capability, and that they enter an industry as long as the expected value of profits exceeds the entry costs (Jovanovic, 1982; Lippman and Rumelt, 1982). Once firms enter they learn how 1 Young firms include ‘adolescent firms’ with adequate resources to sustain themselves through the first few years (Fichman and Levinthal, 1991).

Strat. Mgmt. J., 26: 617–641 (2005)

620

A. M. Knott and H. E. Posen

capable they are, and the firms with inferior capability ultimately exit. The converse of new firm liability is established firm advantage. Established firms accumulate knowledge over time, allowing them to become more efficient (Nelson and Winter, 1982; Levinthal, 1991; Klepper, 1996) and more consistent (inertial) (Hannan and Freeman, 1989) as routines become standardized (Nelson and Winter, 1982). Alternatively this knowledge allows firms to better understand their relative fitness (Jovanovic, 1982) such that they can make informed exit decisions. The first three studies suggest firms update their capability; the latter suggests firms merely update estimates about their capability. There is an extensive learning literature demonstrating that firms update their capability. What is less clear from this literature is whether the updating is autonomous, or whether it requires deliberate efforts and investment to reduce costs. Early literature (see Yelle, 1979, for a review) showing the relationship between cumulative manufacturing experience and reduction in unit cost was interpreted to mean that learning was a costless byproduct of manufacturing experience. More recent work (Sinclair, Klepper, and Cohen, 2000; Knott, 2002) has begun to show that the effects of cumulative manufacturing experience disappear when R&D activities are incorporated properly. In other words, while knowledge may accumulate as a byproduct of manufacturing activity, the exploitation of that knowledge requires overt activity to incorporate it in new products or processes. If learning is autonomous then it should depend only on the level of cumulative experience. The corresponding economic good hypothesis is learningby-doing or spillovers. If learning is overt, then we would expect it to increase with the level of competitive pressure. The implicit hypothesis is competition. If only estimates of capability are updated, then the implicit hypothesis is selection. The general perspective shared by these literatures is that there is a pool of entrants with randomly distributed capability endowments. More firms enter than the market can accommodate (excess entry), and inevitably those with inferior capability will be forced to exit. There are three important questions inherent in this phenomenon: (1) Why do firms make these entry mistakes? (2) How many such mistakes will there be? and Copyright  2005 John Wiley & Sons, Ltd.

(3) What is the economic impact of these mistakes? Camerer and Lovallo (1999) tackle the first question. Economic Census data provide empirical answers to the second question. We tackle the third question of economic impact. We do so from the perspective of those who survive. Thus our work is a complement to the prior literature discussed above that considers failure from the standpoint of those who fail.

MECHANISMS The literature thus suggests three hypotheses through which failure might generate economic benefits. The first hypothesis is a selection effect: firms surviving from a large population (more excess entry) ought to be better performers on average than firms surviving from a smaller population (Demsetz, 1973; Jovanovic, 1982). This occurs because the performance distribution of a large population is denser than a small population. Hence if we draw the best n firms from two populations with identical distributions, the nth firm from the larger population will be further to the right (higher performance) than the nth firm from the small population. Excess entry thus simultaneously produces both high failure rates and superior firms. This hypothesis suggests that failure is merely a byproduct of a phenomenon (excess entry) that yields superior firms. Accordingly, the selection effect postulates a passive role of excess entry and failure. The excess entry does nothing to change the behavior of surviving firms. It merely determines the distribution of firms that remain. The second hypothesis is a competition effect. Unlike the selection effect, this hypothesis holds that excess entry affects the behavior of survivors. In particular, recent work has shown analytically (Peretto, 1999; Aghion et al., 2001; Knott, 2003) and empirically (Geroski and Pomroy, 1990; Blundell, Griffith, and Van Reenan, 1995; Barnett and Hansen, 1996; Nickell, 1996; Hou and Robinson, 2003) that the rate of innovation in a market increases with the number of firms.2,3 Increasing 2, Competition has been characterized both as cross-price elasticity and as the number of rivals. We define it here to be the number of rivals. However, the constructs are equivalent. Given a fixed distribution of consumer tastes, increasing the number of equidistant firms will also increase cross-price elasticity. 3 An earlier received view in industrial organization economics (see Reinganum, 1989) held that innovation is increasing then

Strat. Mgmt. J., 26: 617–641 (2005)

Is Failure Good? the number of firms increases the likelihood that a firm’s market position will be eroded, and this threat (associated with both declining profitability and increased risk of failure) stimulates innovation (Barnett and Hansen, 1996). From this perspective, excess entry increases the competitive pressure in the market because it temporarily inflates the number of firms. This excess competition leads to costreducing investment that yields permanent benefits in the efficiency of survivors. This effect is consistent with the ‘density at founding’ hypothesis (Hannan and Freeman, 1989; Carroll and Hannan, 1989) discussed earlier. Firms surviving a period of intense competition have higher fitness than firms that have been insulated from competition. This effect is implicitly the hypothesis held by Porter (1990) in the Competitive Advantage of Nations. A highly competitive region forces innovation that makes surviving firms more fit for global competition. The effects discussed above are permanent benefits from competition. In addition to these permanent benefits, we anticipate temporary responses to competition. The temporary effects involve behaviors to maximize profits in the current period, such as removing slack and reducing prices (Vickers, 1995). These temporary effects and the associated impact on profits are what trigger cost-reducing investment to better position the firm for future competition (permanent effects). The third hypothesis is a spillover effect. While excess entry yields ‘wasted investment’ in that these investments are no longer appropriable by the failed firm, the value of the investments may be captured by survivor firms through ‘spillovers.’ These spillovers take on many forms: advertising expenditures that expand demand for the product class, improvements in the technology supplied to the industry, and training of industry employees. While these spillovers occur even in the absence of failure, the excess entry may create a larger base of output over which the cumulative learning occurs. We examine the separate impact of each of these hypotheses. If only the selection effect holds, then public policies to stimulate entry are misplaced. decreasing in the level of competition—without any competition there is no incentive to invest, but thereafter expected returns decrease with the number of competitors. While there was some inter-industry evidence that innovation increases then decreases with market concentration, those results disappeared in studies using panel data along with controls for technology (See Cohen and Levin, 1989, for a review). Copyright  2005 John Wiley & Sons, Ltd.

621

Rather, policies should focus on increasing the cost of entry, such that only firms assured of having high efficiency will enter. Under the competition and spillover effects, however, entry by weaker firms alters the behavior and capability of survivors. If either effect holds, public policy to stimulate entry may be welfare enhancing to the extent the public benefits exceed the private costs of the failed firms’ investments.

MODEL We model each of the three mechanisms to construct a test of their contributions to the economic good effect. Before distinguishing between the three mechanisms, we outline some assumptions and definitions that will be common across them. The first thing we define is a measure for the economic benefits. Hicks (1993) used employment growth and wages. Employment rates and wages are a function both of industry characteristics (labor demand) and economy characteristics (labor supply). We prefer a measure that restricts attention to firm characteristics. Accordingly, we look at firm efficiency. The advantage of an efficiency measure is that it captures innovation, so long as innovation manifests itself either as lower cost for same demand, or same cost for enhanced quality/demand. We make some simplifying assumptions in characterizing the impact of excess entry, which we relax in the empirics. First we assume that the populations of potential buyers and potential firms are fixed, and that all innovation is process innovation. This allows us to work from a single inverse market demand function, P (q), where P  (q) < 0 for all q. While we allow increases in demand, we assume these arise from downward shifts in the aggregate cost function C(q), where C(0) = 0, C  (q)
0 and C  (q)
0 for all q. Second, we assume that firms operate at the same scale. Accordingly, we can define the carrying capacity of the market in terms of the number of equalsized firms needed to satisfy demand. We designate this market capacity as k. Note that k, the carrying capacity, can vary across markets and over time. For example, the New York market supports more firms than does the Wyoming market, and the Arizona market in 2003 supports more firms than it did in 1980. We define the number of firms who actually enter the market as n. Thus excess entry Strat. Mgmt. J., 26: 617–641 (2005)

622

A. M. Knott and H. E. Posen

is defined as n − k firms. Finally, we assume that the endowment of firms (their initial unit cost) is normally distributed, and we order firms i = 1 to n from best performing (lowest cost) to highest cost. The cost of the worst-performing, surviving firm is designated ck , the cost threshold for the market. This basic framework forms the basis for comparing the three mechanisms. We now discuss the logical consequences of this set of assumptions for each of the mechanisms.

the failed firms.
q2

P (q)dq − 0

k min

ci (q)

i=1

 q1 
k  ci (q) − F (n − k) (1) −  P (q)dq − i=1

0

where F is the set-up cost, n is the number of total entrants, and k is the number of surviving entrants. Thus excess entry shifts the aggregate cost curve down and to the right. The extent of this shift is a function of the distribution of the potential entrant pool and the probability of entry. This shift increases both output and buyer surplus by displacing less efficient producers. The net effect on producer surplus is ambiguous.

Selection The selection hypothesis holds that as the number of entrants, n, increases, the efficient threshold, ck , determining which k firms survive is increased. This effect is illustrated in Figure 1. In the baseline (Figure 1a), four firms enter the market. Arranging their unit cost from lowest to highest (left to right) creates the aggregate cost curve, C1 . This cost curve intersects demand, D, at q1 , generating market price, p1 . If we allow excess entry (Figure 1b), then as additional firms enter they compete on cost to determine survivors. As in Figure 1(a), firms are ordered from lowest to highest unit cost, but here Firms 6 and 8 have replaced Firm 4 because they had lower cost. Two other firms, 5 and 7, attempted to enter, but found their cost to be higher than Firm 3. The new set of firms defines a new aggregate cost curve, C2 , with shallower slope. This reflects the fact that with competitive selection survivor firms are increasingly drawn from a smaller portion of the right tail of cost distributions. Because C2 is below C1 , output is increased (from q1 to q2 ), and price is decreased (from p1 to p2 ). The welfare contribution from excess entry under this hypothesis is expressed as the present value of the annual surplus increase minus the investments of

Competition The competition hypothesis holds that entrants affect the strategic behavior of surviving firms. In particular, the intensified competition stimulates strategic investment to reduce cost and thereby escape competition (Barnett and Hansen, 1996; Peretto, 1999; Aghion et al., 2001; Knott, 2003). What distinguishes this hypothesis from the prior one is that under the selection effect the unit costs of survivor firms remain at their initial endowments. Aggregate cost decreases only because firms with less efficient endowments are replaced with more efficient ones. Here, under the competition effect, the aggregate cost curve shifts because the costs of existing firms decrease below their initial endowments. Competitive pressure to remain in the market, and the eroding market position from shares stolen by the entrants

P

P C1 p1

C2

p2

f1

f3

f2

D

D

f4

f1 q1 Baseline (k draws)

Figure 1. Copyright  2005 John Wiley & Sons, Ltd.

Q

f2

f8

f6

f3 q2 q3

Q

Excess Entry (k+e draws)

Welfare increase from selection effect (Hypothesis 1) Strat. Mgmt. J., 26: 617–641 (2005)

Is Failure Good? P

623

P C1 p1

p1

C3

p3

f1

f2

f3

f4

f4

Q

P (q)dq − 0

k 

cit (q)

i=1

 q1 
k  ci (q) − F (n − k) (2) −  P (q)dq − 0

i=1

Thus competition shifts the aggregate cost curve down and to the right due to decreases in the cost of each survivor firm. Because investment decisions are typically made annually in response to the competition faced in the preceding period, the extent of these cost improvements is a function of the cumulative number of excess entrants and the length of time they remain in the market. As in the selection effect, this shift increases both output and buyer surplus. Copyright  2005 John Wiley & Sons, Ltd.

f2 q1 q2 Excess Participation

Q

Welfare increases from competition effect (Hypothesis 2)

(Mankiw and Whinston, 1986), stimulate costreducing investment. This competition effect is characterized in Figure 2. Here the original four entrants remain in the market but their costs are now lower due to cost-reducing investments to better position them for future competition. The baseline in Figure 2 (Figure 2a) is the same as that in Figure 1(a). If we allow excess entry (Figure 2b), then firms are under additional competitive pressure from these excess entrants. This pressure stimulates investments to reduce cost. If we ignore selection effects, then the same four firms supply market output, but the aggregate cost curve shifts down because each firm has lower cost than it would in the absence of the additional competitive pressure. Again, because C2 (q) is below C1 (q), output is increased (from q1 to q2 ), and price is decreased (from p1 to p2 ). The welfare contribution from excess entry under the competition hypothesis is expressed as follows:
q3

D

f3 f1

q1 Baseline (k draws)

Figure 2.

D

Spillovers The spillover hypothesis holds that the knowledge generated by failed firms is expropriated by surviving firms. This expropriation takes place in the same manner as spillovers between surviving firms—through employee turnover across firms, interactions with suppliers and customers, publications in the trade literature, and patents (Hoetker and Agarwal, 2003). It also becomes imbedded in the products of failed firms and can become imbedded in inputs from suppliers. Following convention, we assume that spillovers are a function of the cumulative activity of all firms in the market (Spence, 1981; Lieberman, 1987).4 This spillover effect is like the competition effect in that excess entry decreases the costs of surviving firms. The distinction between the two hypotheses is subtle. Under competition firms are driven to invest in innovation to maintain their competitive position—thus they actively reduce costs. Under the spillover effect firms passively benefit from the cumulative activity of other firms in the market. This occurs because the maximum size of the market determines the aggregate spending on marketing and R&D, the innovative activity of suppliers, and the number of employees with industry-specific skills (the sources of spillovers). 4 Whereas spillovers are typically measured as the pool of industry R&D weighted by technological and geographic proximity (Jaffe, 1986, 1988), our measure of spillovers (cumulative output) takes a form similar to that for learning curve effects (Lieberman, 1987). We wish to make two points regarding our measure. First, given industry R&D intensity is fairly stable over time, relative pooled output will match relative pooled R&D. We prefer output to R&D investment because in many instances innovation expenditures will not appear in an R&D line item, e.g., adopting new manufacturing technology from suppliers. Second, we use the term spillovers rather than learning curves to distinguish between intra-firm investments and rival investments.

Strat. Mgmt. J., 26: 617–641 (2005)

624

A. M. Knott and H. E. Posen

Thus spillovers shift the aggregate cost curve down and to the right due to passive decreases in the cost of each survivor firm. The extent of these cost decreases is a function of the cumulative output of all entrants. The resulting shift in aggregate cost increases both output and buyer surplus such that:
q4

P (q)dq − 0

k 

cit (q)

i=1

 q1 
k  ci (q) − F (n − k) (3) −  P (q)dq − 0

i=1

In addition to the market-level effects on cost that we model here, there may be industry-wide effects. These will be treated as exogenous shocks. In fact they too may be endogenous to aggregate industry activity.

EMPIRICAL APPROACH All three mechanisms predict lower aggregate cost with excess entry for a given market. They differ only in the means through which lower cost is achieved. Accordingly, our empirical strategy compares firm efficiency across markets and over time within the same industry. This approach allows us to examine the effects of different levels of entry and exit while controlling for industry demand and technology. The empirical approach we take is similar to that in studies examining the micro-level components of aggregate productivity growth (see Bartelsman and Doms, 2000, for a review of studies in the manufacturing sector; see Foster, Haltiwanger, and Krizan, 2002, for a recent study in the retail sector). We extend those studies along four dimensions. First, we examine a new sector—financial services. Financial services are important in that they are a large ($2.1 trillion in 2001) and growing (real annual rate of 6.5% over the past 10 years) sector of the economy. Moreover, the fact that they are more labor intensive than other sectors suggests that the patterns of productivity growth may differ from those in sectors where vintage capital effects are important. Second, we introduce market effects. Previous studies dealt with manufacturing industries where all firms share a common market, Copyright  2005 John Wiley & Sons, Ltd.

or retail, where the heterogeneity in retail offerings limit market comparison. Third, we examine annual data whereas prior studies have used census data and thus have looked at changes over the 5-year period between censuses. Finally, and most importantly, we introduce a new factor—excess entry. When prior studies have examined exit and entry they have done so in the context of reallocation effects—displacement of old inefficient establishments with more efficient entrants. While reallocation is one component of the excess entry phenomenon, it considers only the selection effects of successful entrants and failed incumbents, whereas we also consider the externality on industry cost. The empiricism in these micro-level studies is typically done in two stages. In an initial stage, researchers develop estimates of establishmentlevel efficiency in each period. In a subsequent stage they recompose changes in aggregate productivity growth using contributions from entrants, exits, and continuing establishments. In general the studies have found (1) that there is considerable heterogeneity in establishment productivity, (2) that relative productivity is durable—establishments with high relative productivity in one period are likely to exhibit high relative productivity in subsequent periods, (3) that a significant portion of productivity improvement (approximately 25%) arises from reallocation effects—the displacement of low productivity establishments with higher productivity entrants, and (4) the most significant source of productivity improvement is expansion of existing plants. We begin our empiricism by replicating the approaches in the prior studies. We then extend the empiricism to examine the impact of entry and exit on market structure and the impact of structure on firm efficiency. Industry We conduct our tests in the banking industry following deregulation. The industry was chosen because it is fragmented with localized competition, has a substantial knowledge component (such that cost improvement is feasible), and is marked by significant failure. Moreover, failure in the industry is due to inefficiency rather than market selection over differences in product offerings (Berger and Humphrey, 1992). Thus industry gains from failure, if any, are likely to match the mechanisms we have modeled. Strat. Mgmt. J., 26: 617–641 (2005)

Is Failure Good? Fragmentation is important because it allows us to compare discrete markets within the same industry. Since the markets are in the same industry we assume that each market faces the same inverse demand function and shares the same technology. Thus we can compare the changes in market structure (arising from excess entry) while controlling for other factors affecting cost across distinct industries. We can also control for differences in level of demand through differences in economic conditions across markets. Furthermore, because banking is regulated, we can obtain firm-level panel data. We confine our analysis to the period following deregulation because we can plausibly argue that knowledge obtained prior to deregulation is heavily depreciated, and thus can be ignored. The high failure and merger rates post deregulation support this contention (see Figure 3). Our operational definition of market in the analysis is a state. In part this definition arises from a data limitation. The unit of observation in the FDIC data is an insurance certificate. A separate certificate is required for each state in which a bank operates,5 but covers all branches for that bank operating within the state. Ignoring for a moment the data limitation, there are two discrete definitions of market: the state, representing certificate/headquarters level competition; or municipality, representing branch-level competition. A reasonable argument for not doing branchlevel analysis, even if data were available, is that it

625

is difficult to determine a relevant radius for competition. Consumers might choose a branch close to their home or one close to their office, but they may also choose a bank based on the fact that it had branches near both, suggesting aggregation to a metropolitan area. Continuing that logic, a state is merely further aggregation, representing on average 7.1 Metropolitan Statistical Areas (MSA), 1.3 Primary Metropolitan Statistical Areas (PMSA), or 0.4 Consolidated Metropolitan Statistical Areas (CMSA). Given the difficulty in choosing a level of aggregation for branch-level competition, and given the fact that the state captures headquarters competition, we define a market as a state.6 Bank failure Exits from the banking industry take the form of failures and mergers. Studies of failure in the banking industry during the period indicate that the firms which failed are those deemed inefficient with respect to the new production function (Berger and Humphrey, 1992). It is important to note that the FDIC rarely allows banks to fail outright (their term is ‘paid outs’—banks close their doors, leaving the FDIC to compensate depositors). The majority of failures are resolved through forced mergers between the ‘failed’ institution and a healthier institution. Thus it is less 6 As an additional test of reasonableness, Petersen and Rajan (2002) examine the distance between small firms and the bank branch they use most frequently. They find that the distance capturing 75 percent of firms in 1990–93 is 68 miles, and growing rapidly due to information technology. This implies a circumscribed area of 14,524 miles, which is greater than the land area of 10 states, and equal to 26.3 percent of the mean land area of all states excluding Texas and Alaska. Thus state is a meaningful market boundary even for branch-level competition.

5 Interstate branching was forbidden prior to June 1997 unless state legislation expressly approved it. The first state to allow interstate branching was New York in 1992, under conditions of reciprocity. As of 1993 only four states allowed interstate branching. Results are robust to excluding data post 1993.

Banking Entry & Exit 1984-1997 700 600 Count

500

enter

400

merge

300

fail

200 100 1997

1996

1995

1994

1993

1992

1991

1990

1989

1988

1987

1986

1985

1984

0

Year

Figure 3. Copyright  2005 John Wiley & Sons, Ltd.

Banking failures following deregulation Strat. Mgmt. J., 26: 617–641 (2005)

626

A. M. Knott and H. E. Posen

likely that knowledge from the failed firms will be broadly expropriated by surviving firms. Exits may also take the form of unforced mergers which act as another mechanism through which lowerefficiency banks exit the industry. In the case of mergers, both forced and unforced, the appropriable assets remain private, rather than being dispersed. Figure 3 quantifies exits by each of the types.

The basic translog cost function mimics the objective function for firm i in year t that is minimizing total cost, cit , by choice of output levels, yit , taking input prices, wit , as given:   β1j yitj + β2k witk cit = β0 + j

+ 1/2

j

+ 1/2

Empirical model Analysis proceeds in two stages. In the first stage we model an industry cost frontier to collect measures of efficiency for each firm in each year. In the second stage, we model firm efficiency (from stage 1) as a function of the three failure benefit mechanisms. Stage 1: Firm efficiency We follow convention in studies of bank efficiency by modeling a stochastic cost frontier using a translog cost function (Cebenoyan, Papaioannou, and Travlos, 1992; Berger, Hancock, and Humphrey, 1993; Mester, 1993; Hermalin and Wallace, 1994). Stochastic frontier analysis, developed by Aigner, Lovell, and Schmidt (1977), is based on the econometric specification of a cost frontier. The stochastic frontier model assumes that the log of firm i’s cost in year t, cit , differs from the cost frontier, cmin , by an amount that consists of two distinct components: a standard normally distributed error term eit , and a cost inefficiency term modeled as a non-negative random variable uit —which we assume to take the form of a truncated normal distribution.7 We use the translog cost function to accommodate the complex array of bank inputs and outputs. In addition, the translog form accommodates tradeoffs in both market strategies (product mixes and prices), and operational strategies (input mixes).8 7 Other distributional assumptions are also possible, the most common of which are the half normal and exponential distributions. All results are robust to these alternative distributions. 8 Note that the translog model used in panel studies of the banking literature does not include a fixed effect. The objective of their studies, and ours, is to capture between-firm differences in efficiency over time. The inclusion of a fixed effect would remove mean firm efficiency differences and thus only capture variation within firms over time. We remove mean firm differences in the second stage of our analysis in which we directly model the drivers of heterogeneous firm efficiency.

Copyright  2005 John Wiley & Sons, Ltd.

 

 j

β y y +

j

k

+

k j j 3jj it it

β4kk witk witk

k

β5j k yitj witk + uit + eit

(4)

k

where: cit = log observed firm cost; yitj = vector of log output levels—j indexes output elements; witk = vector of log input prices—k indexes input elements; uit = cost inefficiency with truncated normal distribution; eit = error term with normal distribution. We pool data for all firms over 14 years using the stochastic frontier model to capture firm–year measures of cost inefficiency relative to a global cost frontier. We collect the estimates of the expected value of firm–year cost inefficiency in Stage 1, E(uit |eit ), which for convenience we continue to label as uit , and use the estimates as the dependent variable in Stage 2 to test the failure mechanisms. Stage 2: Test of mechanisms In a second stage, we model firm cost inefficiency, uit , as a function of excess entry in a given market where, as per the discussion above, a market is operationalized as a state. We utilize different models for the selection effect versus the competition and spillover effects. Selection We test the selection effect by characterizing a time-varying cost threshold modeled as the addit  tive effect of cumulative entry, As,t−1 , the total 1

number of firms that have entered state s from 1984 through year t, and firm inefficiency, uit . We Strat. Mgmt. J., 26: 617–641 (2005)

Is Failure Good? use a logit model to estimate firm failure hazard as a function of the cost threshold, and vectors of controls for state effects, firm-level effects, and year effects:  t  As,t−1 + α2 uit P (exit)it = α0 + α1

627

and significant. Note that if potential entrants discipline the market through limit-pricing (discouraging entry by charging a low price) (Bain, 1949; Milgrom and Roberts, 1982), then there should be no response to entry. In that case both β1 and β2 should be insignificant.

1

+ γ1 StateCtrlst + γ2 F irmCtrlsit

Spillovers

+ Y eart + eit

Finally, we test the spillover effect by examining the cumulative output of all firms over each year a firm is in a market (permanent plus ultimate t   failures), ysit . The spillover effect is distin-

(5)

If there are selection effects, such that increasing the number of entrants improves the pool of survivors, we expect the coefficients α1 and α2 to be positive and significant. This would indicate that cumulative entry drives exit, and that the mechanism for determining who exits is firm efficiency.9 Competition We test the two remaining mechanisms jointly in a single equation for firm cost inefficiency, uit : uit = β0 + β1 Est + β2

t 

Est + β3

1

t   1

ysit−1

i

+ γ1 StateCtrlsst + γ2 F irmCtrlsit + Y eart + λi + eit

(6)

We test the competitive effects by looking separately at current period excess competition, Est , and cumulative excess competition (competition), t  Est , for market s in year t. Current period i

excess competition examines the immediate effects of increased competition. This allows us to characterize the competitive pressure triggering the innovation response (Levitt and March, 1988; Barnett and Hansen, 1996). If incumbent firms respond tactically to entry by reducing slack, we expect β1 to be negative. If they respond by reducing price, we expect β1 to be positive, reflecting lower price for a given level of resource costs (equivalently higher cost to price ratio). Cumulative excess competition examines the strategic effects of cost-reducing investment to better position the firm for future competition. If firms respond strategically to entry through costreducing investment, we expect β2 to be negative 9 We later compare α1 and α2 to determine the marginal impact of an excess entrant on the industry cost threshold.

Copyright  2005 John Wiley & Sons, Ltd.

1

i

guished from the competition effect by the fact that spillovers are assumed to accumulate passively from learning by doing. In contrast, the competition effect presupposes an overt response to excess rivalry. If surviving firms benefit from spillovers we expect β3 to be negative and significant. Note that the spillover effect includes learning from incumbents as well as entrants, since there is no reliable means to distinguish between them. Accordingly, we assess the entrants’ contributions mechanically by applying β3 to the entrants’ output. Equation 6 tests the competition and spillover effects discussed above, while controlling for timevarying market and firm characteristics, year effects, and firm fixed effects, λi . The year effects will capture industry-wide shocks such as new technology. The firm fixed effects will capture fixed differences between firms in addition to capturing fixed differences between markets, such as regulation and taxation. Data The data for the study come from the FDIC Research Database, which contains quarterly financial data for all banks filing the ‘Report of Condition and Income’ (Call Report). Upon entry into the market, each bank is allocated a unique certificate number by the FDIC—and we take the bank (certificate number) as our fundamental unit of analysis. The FDIC classifies and compiles data on two distinct types of banking entities: (a) Commercial banks, which include national banks and state chartered banks (excluding Thrifts) insured by the FDIC; and (b) Savings banks, which operate under state or federal banking codes related to Thrift institutions. Commercial banks, which Strat. Mgmt. J., 26: 617–641 (2005)

628

A. M. Knott and H. E. Posen

are the focus of this paper, differ from Savings banks in that Saving banks have traditionally been limited in both the types of deposits they could accept and the types of loans they could provide. Given the well-known irregularities in the Thrift industry during the 1980s, we confine our analysis to commercial banks. We examine each of the 50 states plus the District of Columbia for the period 1984–97. This initial data set contains 694,587 firm–quarter observations. Following convention in the banking literature we aggregate to annual data by averaging the quarterly data (Mester, 1993). The final first-stage data set comprises 170,859 firm–year observations. While there is considerable debate as to the choice of inputs and outputs in the banking sector, a review of the literature suggests that there is some convergence around a version of Equation 4 that sees physical capital, financial capital, and labor as inputs to the production process and various forms of loans (mortgages, other loans and securities) as outputs (Wheelock and Wilson, 1995). We collect data to construct seven variables related to banking efficiency in log thousands of constant 1996 dollars. The dependent variable is total cost, cit —total interest and non-interest expenses. The six independent variables are divided between input prices, wit , and output quantities, yit . Input prices are: (a) labor price—salary divided by the number of fulltime equivalent employees; (b) physical capital price—occupancy and other non-interest expenses divided by the value of physical premises and equipment; (c) capital price—total interest expense divided by the sum of total deposits, other borrowed funds, subordinated notes and other liabilities. Output quantities are stocks ($1000) of: (d) mortgage loans; (e) non-mortgage loans; and (f) investment securities. In order to test the hypotheses in the secondstage model, we gather aggregate state–year data on entry and exit from the FDIC database to create the cumulative entry (additions), excess entry, cumulative excess entry, and cumulative output variables. We define entry as a new commercial banking institution that comes into existence either by way of a new charter or the conversion of an existing charter. Likewise, we define exit as an event that leads to the termination of a bank (certificate) either in the form of a forced failure or merger, or an unforced merger. We define excess entry as the number of institutions in excess of the Copyright  2005 John Wiley & Sons, Ltd.

carrying capacity of the market s in year t. Using the logic of efficient markets, we measure this variable simply as the number of exits from market s in year t, where exits include both failure and mergers.10 To the extent that optimistic firms remain in the industry during periods of losses, this measure understates the excess. Cumulative excess is therefore the sum of the exits over all years since 1984. In addition, in order to create the cumulative output for market s in year t, we aggregate firm-level output at the state level as the sum of mortgage, non-mortgage loans, and investment securities of all firms located in the state. These were the y1 , y2 and y3 variables in the Stage 1 analysis. A number of firm-level control variables are included in the model. In order to control for branch scale effects not already accounted for in the first-stage estimation, we include three variables: (a) branch count—number of branches operated by the bank; (b) branch size—the average size of a branch measured in terms of total output in thousands of constant 1996 dollars; (c) cum output bank—the sum of bank output over all branches and years in thousands of constant 1996 dollars. Approximately one-third of banks are owned by a bank holding company that controls more than one bank (certificate). For such banks, we include a dummy variable, holding company, as well as a number of measures of the size of the holding company: (c) hc certificates—the number of additional banks (certificates) held by the holding company; (d) hc branches—the number of additional branches in the bank holding company system beyond the number of branches in the observation certificate; (e) hc states—the number of additional states in which the holding company operates banks. We control for economic differences across markets and over time using annual data on population and housing starts. Permanent differences across states, e.g., regulatory conditions, are subsumed by the firm fixed effects.

10 The important consideration here is the extent to which exits remove operational capacity from the market. Descriptive data suggest that all exits, be they failures or mergers (FDA forced as well as unforced), remove branch capacity, and thus we consider both types of exits as constituting excess entry.

Strat. Mgmt. J., 26: 617–641 (2005)

Is Failure Good? RESULTS Stage 1: Firm cost efficiency Table 1 provides variable descriptions and summary statistics of the data used in Stage 1. We estimate the Stage 1 stochastic frontier model assuming a truncated normal distribution for the inefficiency term and a normally distributed error term. Estimation is conducted using maximum likelihood techniques. Results from the Stage 1 analysis using Equation 4 are given in Table 2. The objective of the Stage 1 analysis is to provide the firm–year cost inefficiencies that serve as the dependent variable in Stage 2. While a discussion of the estimated coefficients of the frontier model is outside the scope of this paper, the coefficient estimates are consistent with expectations as: (a) total costs appear to rise with output and increases in the price of capital; and (b) firms substitute labor and physical capital in response to changing prices for these inputs. The more important result of the Stage 1 frontier estimation is the expected value of the inefficiency terms, uit . The distribution of the cost inefficiency is given in Figure 4 and the mean value over time is depicted in Figure 5. The mean uit over the entire period is 0.171, which indicates that the total cost for the mean firm is 18.6 percent above that of a firm

629

on the cost frontier. The data also indicate that while the mean cost inefficiency changes over time in response to changing technologies and demand conditions, the general trend is toward increasing efficiency (decreasing cost). Comparison with longitudinal micro-data (LMD) studies To provide more insight into the dynamics of banking sector efficiency we mimic LMD studies and present a transition matrix of firm efficiency (Table 3). The transition matrix characterizes how a firm’s efficiency in one period is related to its efficiency in the subsequent period. The table decomposes firm efficiency into quintiles, then depicts year-to-year movements across quintiles by comparing rows and columns. As an example, take the row labeled 1. This represents the highestperforming (lowest cost-inefficiency) firms in a given year. The table indicates that the majority of firms (78%) remain in the top (lowest cost) quintile in the subsequent year, 15 percent drop one quintile, 2 percent drop two quintiles, less than 1 percent drop to each of the bottom two quintiles, 2.7 percent merge with other firms and 0.5 percent fail outright. The table indicates three patterns of interest. First, relative firm efficiencies are fairly stable.

Table 1. Stage 1 data summary Variable c w1 w2 w3 y1 y2 y3

Description

Obs.

Mean

S.D.

Min.

Max.

cost price labor price physical capital price capital mortgage other loans securities

174,869 174,673 173,999 173,789 172,503 173,373 173,828

7.98 2.97 −1.64 −3.54 9.47 9.63 9.58

1.27 0.28 0.73 0.41 1.58 1.40 1.47

−0.82 −4.85 −9.93 −12.58 −1.37 −0.70 −1.27

16.76 9.39 5.63 3.32 17.86 18.55 17.51

Units: ln(thousand, 1996 dollars).

Variable c w1 w2 w3 y1 y2 y3 ∗

c

w1

w2

w3

y1

y2

y3

1 0.1929∗ −0.0418∗ 0.1464∗ 0.8640∗ 0.9146∗ 0.7603∗

1 0.2792∗ 0.0921∗ 0.1151∗ 0.1493∗ 0.0821∗

1 0.0452∗ −0.1322∗ −0.0594∗ −0.0922∗

1 −0.0463∗ 0.1201∗ 0.0259∗

1 0.7711∗ 0.6883∗

1 0.6934∗

1

Significant at 5%

Copyright  2005 John Wiley & Sons, Ltd.

Strat. Mgmt. J., 26: 617–641 (2005)

630

A. M. Knott and H. E. Posen Table 2. Results from Stage 1 regression Dependent variable: ln(cost) 170,859 observations

w1 w2 w3 y1 y2 y3 1 /2y1y1 y1y2 y1y3 1 /2y2y2 y2y3 1 /2y3y3 1 /2w1w1 w1w2

Coeff.

S.E.

−8.691e − 01∗∗ −2.085e − 01∗∗ 2.078e + 00∗∗ 1.942e − 02∗ 4.262e − 01∗∗ 2.784e − 01∗∗ 9.331e − 02∗∗ −6.028e − 02∗∗ −2.049e − 02∗∗ 1.225e − 01∗∗ −5.541e − 02∗∗ 8.827e − 02∗∗ 2.011e − 01∗∗ 3.016e − 02∗∗

(30.644) (17.544) (85.535) (2.073) (41.009) (30.382) (165.733) (120.751) (39.425) (165.107) (95.188) (190.369) (43.998) (16.267)

w1w3 /2w2w2 w2w3 1 /2w3w3 y1w1 y1w2 y1w3 y2w1 y2w2 y2w3 y3w1 y3w2 y3w3 Constant E(uit |eit ) 1

Coeff.

S.E.

−2.312e − 01∗∗ −4.964e − 03∗∗ −2.519e − 02∗∗ 2.564e − 01∗∗ 3.230e − 02∗∗ −7.015e − 04 −3.159e − 02∗∗ −2.019e − 02∗∗ −9.330e − 03∗∗ 2.952e − 02∗∗ −3.038e − 02∗∗ 1.418e − 02∗∗ 1.620e − 02∗∗ 5.320e + 00∗∗ 1.707e − 01∗∗

(61.988) (4.675) (15.625) (72.979) (22.174) (0.969) (24.569) (12.595) (10.312) (21.413) (21.670) (19.267) (12.903) (54.442) 0.172

Absolute value of t-statistics in parentheses. ∗ significant at 5%; ∗∗ significant at 1%

0

2

Density 4

6

8

Cost Inefficiency Density Function

0

.5

1 cost_inefficiency

1.5

2

Figure 4. Histogram of firm-year efficiency metrics

The highest percentages are along the diagonal, meaning that in general firms remain in the same cost quintile. Second, failures and mergers increase with firm cost. The highest percentage of mergers and failures are coming from the highest-cost firms. Thus the table provides preliminary evidence of a selection effect as the lowestperforming firms are being driven from the market. These patterns (persistent heterogeneity and Copyright  2005 John Wiley & Sons, Ltd.

selection effects) are similar to those found in LMD studies of other sectors. Third, new firms tend to enter at the industry extremes, 27 percent enter the lowest cost quintile, while 60 percent enter the highest cost quintile. Again, this pattern is shared with LMD studies in other sectors. To complete the parity with prior LMD studies, we conduct an additional test of the reallocation effects. We created distributed lag models for Strat. Mgmt. J., 26: 617–641 (2005)

Is Failure Good?

631

0

.05

mean of u_tn_pl .1 .15

.2

Mean of Cost Inefficiency

1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997

Figure 5. Table 3.

Mean cost inefficiency

Year-to-year transition matrix of firm efficiency Next year quintile Low cost 1

High cost 2

3

4

5

Merge

Fail

Total

Enter

670 26.64

63 2.50

96 3.82

179 7.12

1507 59.92

0 0.00

0 0.00

2515 100

1

31,759 77.93

6,054 14.86

905 2.22

358 0.88

379 0.93

1,097 2.69

201 0.49

40,753 100

2

5,362 16.75

17,327 54.14

6,797 21.24

1,178 3.68

246 0.77

1009 3.15

86 0.27

32,005 100

3

702 2.23

6,447 20.48

15,857 50.38

6,374 20.25

770 2.45

1,187 3.77

135 0.43

31,472 100

4

254 0.84

923 3.05

6,074 20.10

16,164 53.50

5,141 17.02

1,414 4.68

244 0.81

30,214 100

5

394 1.26

346 1.10

817 2.61

4,736 15.12

22,282 71.11

1,968 6.28

790 2.52

31,333 100

39,141 23.26

31,160 18.52

30,546 18.15

28,990 17.23

30,325 18.02

6,675 3.97

1,456 0.87

168,293 100

Low cost

This year quintile

High cost Total

Includes firms in all states in the industry at any time from 1984 to 1997. Entrants are those entering as new certificates, but do not include conversions.

entering and exiting firms to track their efficiency following entry and preceding failure. Figures 6 and 7 present graphical results of this analysis.11 11

Coefficient estimates are available from the authors.

Copyright  2005 John Wiley & Sons, Ltd.

Figure 6 presents the results for exiting firms. The results indicate that lower-performing firms exit the industry, confirming the preliminary results in the transition matrix. The new insight is that failing firms not only have higher cost than the Strat. Mgmt. J., 26: 617–641 (2005)

632

A. M. Knott and H. E. Posen Exiting Firm Cost Inefficiency Cost Inefficiency: Relative to firms more than 5 years from exit 0.20 0.18 0.16 0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00

fail merge

4

3

2

1

0

Years to Exit

Figure 6. Exiting firm cost inefficiency histories Entering Firm Cost Inefficiency Cost Inefficiency: Relative to incumbent firms and entrants more that five years post entry 0.60 0.55 0.50 0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 - 0.05

enter_survive enter_merge enter_fail

0

Figure 7.

1

2 Years from Entry

4

Entering firm cost inefficiency history

industry overall, but their cost deteriorates as they approach their exit year. While these effects hold for both failures and mergers, they are more pronounced for failures. In fact, failing firms have cost inefficiencies that are twice the industry average, resulting in a 26 percent increase in overall cost.12 Figure 7 presents the results for entering firms. The results indicate that entering firms tend to have higher costs than incumbent firms, and that their cost inefficiencies are twice the industry average in the entry year (0). We further disaggregate the data into firms that survive more than 5 years, and those that fail or merge within 5 years of 12 The cost inefficiency (in ln form) increases from the mean of 0.18 by approximately 0.18 to 0.35. This corresponds to a 26 percent increase in costs.

Copyright  2005 John Wiley & Sons, Ltd.

3

entry. For surviving entrants, performance nearly matches that of incumbents by the third full year. Entering firms that fail match surviving entrants’ performance for the first 3 years, then show a dramatic reduction in efficiency, mimicking the results in Figure 6. Interestingly, firms that merge within 5 years of entry are the poorest-performing entrants. In general, the results in Figure 7 indicate that those firms who entered in the highest cost quintile in Table 3 are unlikely to recover. Overall the micro-level components of banking efficiency seem to match the patterns found in other sectors. There is considerable heterogeneity in banking efficiency (Figure 4), and the relative efficiencies are durable (Table 3). Strat. Mgmt. J., 26: 617–641 (2005)

Is Failure Good? Stage 2: Test of mechanisms Having generated firm cost inefficiencies and showed comparability between the micro patterns of aggregate efficiency improvement within banking and those in other sectors, we now proceed with the formal tests of the excess entry mechanisms. We look first at tests of selection effects and then at tests of competition and spillovers effects. Summary statistics for the data supporting these tests are presented in Table 4. Selection Results for test of the selection effects using a pooled logit estimation of Equation 5 are given in Table 5. The equation creates a measure of an evolving cost threshold by examining market-level effects (how much exit there will be) and firm-level effects (which firm is most likely to exit). The main results, presented in Model 4, indicate that both cumulative additions and cost inefficiency are positive and significant. That is, an increase in either cumulative additions or cost inefficiency leads to an increase in the probability of exit. An important question is that of the effect of excess entry on the cost threshold for survival. To assess the economic significance of the selection effect, we equate the marginal effects of cost inefficiency and cumulative additions at their means. This comparison indicates that each additional entrant into a market lowers the cost threshold for survival, ck , (increases the exit hazard for a given level of firm cost efficiency) in that market by 1.35 percent. The control variables also suggest interesting patterns. The state-level economic controls for population growth and new housing growth are significant only in a model without the main variables. The firm-level variables indicate that the least efficient firms are the ones most likely to exit. This matches intuition as well as preliminary results from the transition matrix. They also indicate that firms with riskier portfolios (higher percentage of real estate loans) have higher exit probabilities. Larger scale (branch size and number of branches) decreases the probability of exiting, but ownership by a bank holding company significantly increases the probability of exiting. This probability increases further with the scale of the holding company (number of bank certificates, branches, and states of operation). This Copyright  2005 John Wiley & Sons, Ltd.

633

is a rather counter-intuitive result. However, further analysis (available from the authors) indicates that the results pertain to unforced mergers (consolidating weak banks) rather than failures. Competition and spillovers Results for the test of competition effects using Equation 6 are given in Table 6. The equation considers both the immediate, tactical effects of current competition and the strategic effects of cumulative exposure to excess competition. We test the tactical effects through the variable excess entry and test the strategic effects through the variable cumulative excess entry. We examine the impact of each variable on firm cost inefficiency (negative coefficients decrease cost). Model 2 indicates that excess entry is positive and significant, indicating that margins fall (costs rise) with the number of current excess competitors. This suggests the immediate effect of excess competition is to reduce price (same cost for lower output) rather than to remove slack. This result is also consistent with the stealing share effect (Mankiw and Whinston, 1986), where fixed costs are allocated over lower firm output. While our primary interest is the strategic impact of cumulative excess competition on firms’ innovative behavior (Model 3), we show the results for current competition first to demonstrate that (1) excess entry hurts performance, and (2) the degradation in performance appears to stimulate an immediate response. These results are consistent with those of Barnett and Hansen (1996) and Barnett and Sorenson (2002), who show that firms respond to recent competition. Turning next to the permanent effects, Model 3 indicates that cumulative excess entry leads to cost-reducing investment as the coefficient is negative and significant. We evaluated the marginal contribution of an additional excess entrant at the mean value of the independent variables. This analysis indicates that each excess entrant in a market yields a permanent 0.004 percent reduction in the mean cost of each incumbent firm. Given the mean terminal value (1997) for cumulative excess entry in a market was 336 firms, the implied cost reduction in each surviving firm was 1.34 percent. Finally, we examine the spillover effect of passive contributions arising from the activity of failed Strat. Mgmt. J., 26: 617–641 (2005)

Copyright  2005 John Wiley & Sons, Ltd.

a

cost inefficiency—ln($000) 1.55E − 01 1.80E + 01 excess entry 1.05E + 02 cum excess entryt−1 4.05E + 01 cum additionst−1 5.90E + 08 cum output ($000) branches 5.73E + 00 1.64E + 05 branch scale ($000) 4.88E − 01 real estate pet 1.73E + 06 cum output bank 3.71E + 00 hc certificates 3.93E + 01 hc offices 4.10E − 01 hc states population 6.64E + 06 6.59E + 04 delta population permit 3.47E + 04 −4.37E + 02 delta permit

Mean 1.45E − 01 1.51E + 01 9.88E + 01 4.93E + 01 8.31E + 08 2.61E + 01 1.12E + 06 1.94E − 01 1.93E + 07 1.05E + 01 1.57E + 02 1.51E + 00 6.07E + 06 1.19E + 05 4.26E + 04 9.13E + 03

S.D.

1 1.00 0.00 −0.06 0.19 0.06 −0.01 0.03 −0.02 0.00 0.07 0.06 0.08 0.15 0.18 0.16 −0.03

Data summary Stage 2a Observations: 136,759

Units in counts or dollars unless otherwise stated.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Table 4. 3

4

5

6

7

8

9

10

11

12

13

14

15

16

1.00 0.51 1.00 0.26 0.51 1.00 0.26 0.40 0.43 1.00 −0.05 −0.02 0.04 0.10 1.00 −0.01 −0.01 0.00 0.07 0.01 1.00 0.08 0.26 0.25 0.23 −0.01 −0.07 1.00 −0.02 0.01 0.03 0.15 0.61 0.23 −0.04 1.00 0.06 0.02 0.03 −0.05 0.07 0.02 0.01 0.04 1.00 0.00 0.04 0.05 0.01 0.22 0.07 0.01 0.12 0.62 1.00 −0.01 0.03 0.02 0.01 0.23 0.09 −0.01 0.16 0.72 0.78 1.00 0.26 0.23 0.55 0.74 0.09 0.02 0.21 0.08 −0.04 0.01 −0.03 1.00 0.14 0.22 0.67 0.40 0.07 0.00 0.18 0.04 −0.01 0.04 0.00 0.76 1.00 0.19 0.15 0.54 0.34 0.07 0.00 0.14 0.03 0.00 0.04 −0.01 0.76 0.90 1.00 −0.01 0.01 −0.18 −0.10 −0.01 0.00 −0.03 −0.01 0.01 0.00 0.01 −0.17 −0.26 −0.09 1.00

2

634 A. M. Knott and H. E. Posen

Strat. Mgmt. J., 26: 617–641 (2005)

Is Failure Good? Table 5.

635

Test of Exit Hazard Logistic regression results Dependent variable: probability(exitit ) (1)

(2) 2.032e − 03∗∗ (5.673)

cum additionst−1 cost inefficiencyt−1 branch-count branch scale real estate pct holding company hc certificates hc branches hc states delta population delta permit year dummies Constant Observations Log L Chi2 Prob>Chi2 Pseudo R sq

−2.912e − 03∗∗ (5.214) −1.793e − 07∗∗ (3.839) 4.933e − 01∗∗ (6.387) 1.042e + 00∗∗ (33.428) 4.082e − 03∗∗ (3.222) 9.209e − 04∗∗ (12.708) 6.557e − 02∗∗ (6.746) 7.143e − 07∗∗ (5.906) −1.893e − 06 (1.122) sig. −3.618e + 00∗∗ (53.869) 136,478 −23169.138 3849.524 0.000 0.077

−2.908e − 03∗∗ (5.241) −1.871e − 07∗∗ (3.922) 4.538e − 01∗∗ (5.857) 1.046e + 00∗∗ (33.546) 3.224e − 03∗ (2.529) 9.206e − 04∗∗ (12.749) 6.954e − 02∗∗ (7.179) 9.023e − 08 (0.529) −2.489e − 06 (1.422) sig. −3.689e + 00∗∗ (54.022) 136,478 −23153.303 3881.195 0.000 0.077

(3)

(4)

7.588e − 01∗∗ (13.100) −2.603e − 03∗∗ (4.689) −1.750e − 07∗∗ (3.944) 5.230e − 01∗∗ (6.826) 1.041e + 00∗∗ (33.360) 4.423e − 03∗∗ (3.495) 9.506e − 04∗∗ (13.101) 5.571e − 02∗∗ (5.734) 4.741e − 07∗∗ (3.832) −2.447e − 06 (1.440) sig. −3.721e + 00∗∗ (55.066) 136,478 −23096.873 3994.054 0.000 0.080

1.563e − 03∗∗ (4.346) 7.300e − 01∗∗ (12.479) −2.611e − 03∗∗ (4.730) −1.807e − 07∗∗ (3.997) 4.908e − 01∗∗ (6.380) 1.044e + 00∗∗ (33.448) 3.720e − 03∗∗ (2.917) 9.506e − 04∗∗ (13.134) 5.915e − 02∗∗ (6.094) 3.205e − 09 (0.019) −2.902e − 06+ (1.660) sig. −3.770e + 00∗∗ (55.047) 136,478 −23087.544 4012.713 0.000 0.080

Absolute value of z statistics in parentheses + significant at 10%; ∗ significant at 5%; ∗∗ significant at 1%

entrants (Model 4). The coefficient on cumulative output market is negative and significant, indicating that each dollar of market output significantly reduces the costs of surviving firms. Again, we examined the marginal effects of an additional entrant who ultimately exits, taking into account its mean output. On average, each such entrant generated an output of $128 million for 7.4 years prior to exiting—the result of which was a permanent reduction in the cost of each surviving firm by 0.0007 percent. In 1997, in each market, the mean value of excess entry was 336 firms. Accordingly, the corresponding cost reduction in surviving firms was 0.24 percent. This learning from failed firms is in addition to the learning from surviving firms. One issue with interpretation is whether the excess entrants expanded Copyright  2005 John Wiley & Sons, Ltd.

the market, thereby increasing cumulative experience, or whether they merely stole share. If the latter is true, then there is no net spillover benefit from the excess entrants. However, our finding that excess competition stimulates pricecutting indicates that output was expanded by the excess entrants. This suggests their experience does increase the spillover pool. A brief discussion of the control variables is also warranted. In all models we include controls for market economic conditions and time-varying firm attributes that might affect bank productivity. Looking first at firm attributes, the coefficients on the branch count (number of branches) and branch size (branch scale) are negative and significant. In contrast, the coefficient on cum output bank is positive. If we evaluate each variable Strat. Mgmt. J., 26: 617–641 (2005)

636

A. M. Knott and H. E. Posen

Table 6.

Results from Stage 2 regression Regression results Dependent variable: cost inefficiency Fixed effects (1)

(2) 1.488e − 04∗∗ (6.863)

excess entry cum excess entryt−1

(3)

−4.671e − 05∗∗ (8.418)

cum output mrkt branch count branch size cum output bank holding company hc certificates hc branches hc states population permit year dummy Constant Observations R2

−1.319e − 04∗∗ (6.053) −1.246e − 08∗∗ (23.610) 7.273e − 11∗∗ (2.721) −6.492e − 03∗∗ (6.141) −5.841e − 05 (0.933) 3.305e − 05∗∗ (9.323) 2.643e − 03∗∗ (5.862) −7.339e − 09∗∗ (6.888) −2.717e − 07∗∗ (11.440) sig. 2.081e − 01∗∗ (27.462) 136,759 0.731

−1.312e − 04∗∗ (6.023) −1.245e − 08∗∗ (23.591) 7.295e − 11∗∗ (2.730) −6.603e − 03∗∗ (6.246) −6.777e − 05 (1.083) 3.295e − 05∗∗ (9.295) 2.657e − 03∗∗ (5.893) −8.048e − 09∗∗ (7.520) −2.843e − 07∗∗ (11.937) sig. 2.111e − 01∗∗ (27.820) 136,759 0.731

−1.350e − 04∗∗ (6.196) −1.256e − 08∗∗ (23.789) 6.556e − 11∗ (2.452) −6.578e − 03∗∗ (6.224) −5.232e − 05 (0.836) 3.226e − 05∗∗ (9.100) 2.678e − 03∗∗ (5.940) −5.062e − 09∗∗ (4.606) −2.333e − 07∗∗ (9.650) sig. 1.850e − 01∗∗ (22.176) 136,759 0.731

(4)

(5)

−7.033e − 12∗∗ (8.258) −1.331e − 04∗∗ (6.111) −1.244e − 08∗∗ (23.569) 1.042e − 10∗∗ (3.860) −6.653e − 03∗∗ (6.294) −4.421e − 05 (0.706) 3.301e − 05∗∗ (9.314) 2.521e − 03∗∗ (5.589) −4.637e − 09∗∗ (4.161) −2.863e − 07∗∗ (12.025) sig. 1.811e − 01∗∗ (21.557) 136,759 0.731

1.952e − 04∗∗ (8.844) −3.953e − 05∗∗ (6.877) −6.889e − 12∗∗ (7.703) −1.349e − 04∗∗ (6.192) −1.250e − 08∗∗ (23.695) 9.773e − 11∗∗ (3.613) −6.867e − 03∗∗ (6.499) −5.164e − 05 (0.825) 3.221e − 05∗∗ (9.088) 2.570e − 03∗∗ (5.700) −3.696e − 09∗∗ (3.268) −2.700e − 07∗∗ (11.049) sig. 1.787e − 01∗∗ (21.242) 136,759 0.731

Absolute value of t statistics in parentheses ∗ significant at 5%; ∗∗ significant at 1%

at its mean value in 1997, the net effect is negative, suggesting that larger firms have lower costs. At the holding company level, the results are ambiguous, highlighting the complex role of the multi-unit structure—in the provision of scale advantages, in influencing learning between banks within a holding company, and perhaps in buffering individual banks from the learning associated with competitive interaction (Barnett, Greve, and Park, 1994). In particular, the holding company dummy is negative and significant, indicating a cost reduction associated with holding company ownership. The holding company scale variables coefficients are mixed. While both hc branches (number of branches) and hc states (number of Copyright  2005 John Wiley & Sons, Ltd.

states) are positive and significant, hc certificates (holding company certificate count) is negative and significant. Nevertheless, the net effect of holding company, evaluated at mean 1997 levels, is negative—indicating that ownership by a holding company is on average associated with lower cost. Looking finally at market economic conditions, results indicate that increases in both population and the number of housing starts decrease cost. The main conclusion, however, is that the competition and spillover effects persist in the presence of these controls. In sum, our results indicate support for all three mechanisms. First, we find that selection effects are at work as additional entry increases Strat. Mgmt. J., 26: 617–641 (2005)

Is Failure Good? the probability of exit for inefficient incumbents. Second, we find that competition from excess entrants decreases price margins in the year of the entry—which in turn leads to a long-term effort on the part of incumbents to reduce costs over subsequent years. Third, excess entrants improve the efficiency of incumbents by adding to the pool of spillovers. Robustness We have conducted a number of robustness checks on the results and in each case the results are robust to alternative specifications. First, the specification of the stochastic cost frontier model relied on the use of a truncated normal distribution of the cost inefficiency term. We tested the sensitivity of the results to different distributional specifications. We ran models assuming both a half-normal distribution and exponential distribution and the empirical results are robust to all such specifications. Second, a significant assumption of the cost frontier model specified in the first stage is that all firms are drawn from the same cost inefficiency distribution. We relaxed this assumption and allowed for heterogeneity in the cost inefficiency distribution. In particular, we modeled the stochastic cost inefficiency term using a halfnormal distribution where the variance of the distribution was assumed to be a function of the scale of the firm. The results are robust to this respecification. Third, an alternative explanation for the efficiency gains from excess entry is that all such gains are driven by very efficient entrants displacing incumbents. While the descriptive results suggest this is not the case—as entrants on average do not match the performance of incumbents (see Figure 7), we ran our models on a restricted sample of incumbents—firms that were in the sample over the entire 14-year period. Our results are robust to this sample splitting, suggesting that performance improvements are driven by the hypothesized mechanisms rather than reallocation effects. Finally, in our analysis, we are concerned with the effect of excess entry on surviving firms’ costs. An alternative question is that of the effect of excess entry on the cost of all firms—including those that exit (had they not done so). While this latter question is not our central concern, we estimated a sample selection model to account Copyright  2005 John Wiley & Sons, Ltd.

637

for the fact that low-efficiency firms are more likely to exit. From the estimation of Equation 5, we extracted the inverse mills ratio and included this as an additional variable in estimation of Equation 6. The results are robust to this alternative specification.

PRIVATE COSTS AND SOCIAL WELFARE IMPLICATIONS Our empirical results indicate that failed entrants generate externalities that substantially reduce industry cost. Failed entrants improve the quality of the survivor pool through the selection, competition and spillover effects characterized in Figures 1 and 2. Accordingly, policies to subsidize entry may enhance social welfare. One important question, however, is whether the social benefits from excess entry exceed the private costs of the failed entrants. While formal welfare analysis is beyond the scope of this paper, we provide some informal analysis to anticipate likely results. The informal analysis compares the private costs of the excess entrants to the externalities they generate. We estimate these private costs by summing firm operating profits (losses) over the tenure of each firm. This approach assumes that all up-front and non-recoverable investments are either amortized across accounting periods or are expensed in the final year. We collect separate estimates for three categories of firms: (1) firms that enter and exit during the observation period, (2) firms whose founding predates the observation period, but who fail during the observation period, and (3) mergers within a holding company. Table 7 summarizes the mean private costs for firms in each category. The table indicates that on average failed entrants (those who enter and exit during the observation period) incur no private losses. In fact they actually exhibit net profits averaging Table 7. Private costs of exiting firms Firm category Firms who enter and exit Incumbents who fail Incumbents who merge

Number of firms

Mean profits

790 153 637

$15.14 million −$1.30 million $19.09 million

Strat. Mgmt. J., 26: 617–641 (2005)

638

A. M. Knott and H. E. Posen

$15.14 million over their tenure. This suggests that the hit and run opportunities proposed by contestability theory (Baumol, 1982) exist in this setting. The private gains plus the positive externality on survivor costs indicate that the failed entrants enhance social welfare. Similar conclusions are drawn for firms who merge. They exhibit net profits averaging $19.09 million over their tenure. Results for the displaced incumbents (firms who enter before the observation period, but later fail) are less conclusive. The data indicate that these firms incur private losses averaging $1.3 million over the observation period. Since the entry of these firms predates the observation period, and since Figure 6 indicates that firm costs increase in the years immediately preceding exit, these firms may in truth enjoy private gains. Nevertheless we take the conservative stance that these losses are net losses. We then compare the incumbent losses to the externality generated by their displacement. To do so, we sum the economic value of the externality for the average entrant across the three mechanisms. The externality for the selection effect was a 1.34 percent reduction in surviving firm cost; the externality for the competition effect was 0.004 percent reduction in surviving firm cost; and the externality for the spillover effect was 0.0007 percent in surviving firm cost. Thus the total externality is a permanent decrease in industry cost of 1.3447 percent. At the mean firm cost of $2.9 million, and the mean number of firms per market of 195, this corresponds to an aggregate cost reduction from an excess entrant of $7.6 million. Since this reduction is actually an annuity, the externality is the present value of that annuity stream. Given that even the singleyear aggregate savings of $7.6 million exceeds the cumulative losses ($1.3 million) of the failed incumbent, it appears that displacing entrants also enhance welfare.

DISCUSSION Approximately 10 percent of all firms in the United States fail each year. Our question is whether the efforts of these failed entrepreneurs are in vain. Do their efforts merely represent private losses, or are there public gains to offset these losses? Copyright  2005 John Wiley & Sons, Ltd.

We proposed three alternative mechanisms through which failure of excess entrants might benefit consumers and surviving producers. These mechanisms (selection effects, competition effects, and spillover effects) were derived from failure theories in organization theory and evolutionary economics. We tested all three mechanisms plus the effects of current excess competition in the banking industry following deregulation. We found significant and substantial support for each effect. Thus we can say that the economic benefits are real. Excess entry and subsequent failure increase aggregate industry efficiency. Moreover these social benefits exceed the private costs of failed firms (indeed many exiting firms have net gains). Thus, in this setting, excess entrants appear to enhance social welfare. Our results were obtained in the banking industry. We chose this setting because the fragmentation of banking markets allowed us to compare differences in market structure (arising from failure) while controlling for other factors affecting market structures across industry. Because there are few fragmented industries with firm-level panel data, it will be difficult to replicate this study in other settings. Nevertheless it is worth speculating whether our results are unique to banking. Banking is a service industry with high human capital (knowledge) intensity and low physical asset intensity. Low asset intensity implies there are fewer sunk investments to inhibit adoption of more efficient practices. Our results may be less pronounced in manufacturing industries where vintage capital inhibits adoption of innovations. Fortunately, the newer industries, the ones accounting for the greatest excess entry and failure, tend to be more like banking than manufacturing. Thus failure should increasingly provide the benefits seen here. The specification employed in this paper is not without caveats. Two are of note. First, our main means of identification come from variation in levels of entry and exit across states. Our specification assumes that entry choice is exogenous in the sense that entrants are randomly assigned to states. Given the fact that 94.3 percent of entry is by local entrepreneurs, this assumption seems plausible. Second, our measure of excess competition is exits. To the extent that there are lags in the exit process, it is likely that our measure understates the actual level of excess entry at any given Strat. Mgmt. J., 26: 617–641 (2005)

Is Failure Good? time and thus underestimates the magnitude of the failure benefits. Within the field of strategy, researchers have only recently begun to consider the joint effects of competition-driven innovation and selection. While progress has been made (Barnett and Hansen, 1996), prior research has sought to address these issues using high-level measures of performance (e.g., profits or survival). However, competition-induced innovation by one rival is often met with innovation by other rivals. Such innovation typically affects cost or price (quality improvement), but when matched by rivals may have no net effect on profits. An efficiency-based measure allows us to capture both cost-reducing and quality-improving innovation even when there is no net impact on profits or survival. With this study, we hope to reinforce research in strategy that utilizes operational measures of performance. The frontier approach that we employ may be useful to strategy scholars in re-examining prior conclusions regarding competition-driven innovation, selection, and learning from studies that have used higher-level performance measures. The conventional view of creative destruction, which hails the entrepreneur, is one of macro-level selection in which entrepreneurs need to be successful to generate economic benefits—the nimble entrant displaces the ossified incumbent. We introduced a behavioral view of creative destruction whereby entrepreneurial entry applies pressure to incumbent firms which triggers search and selection at a micro-level inside the firm. The search and selection results in displacement of inefficient practices within incumbents. In this view even the failed entrepreneur is heroic in that she prevents successful incumbents from resting on their laurels.

ACKNOWLEDGEMENTS The authors would like to thank the Mack Center for Technological Innovation for financial support. We would also like to thank Bill Barnett, John de Figueredo, Loretta Mester, Philip Strahan, Sid Winter, participants in the Strategy Research Forum, the Fuqua strategy seminar, the HBS entrepreneurial management seminar, the HBS Strategy Conference, the NYU Global Research Initiative seminar, Reginald H. Jones Center Brown Copyright  2005 John Wiley & Sons, Ltd.

639

Bag seminar, Dan Schendel, and two anonymous reviewers for helpful comments.

REFERENCES Aghion P, Harris C, Howitt P, Vickers J. 2001. Competition, imitation and growth with step-by-step innovation. Review of Economic Studies 68: 467–492. Aigner D, Lovell AK, Schmidt P. 1977. Formulation and estimation of stochastic frontier production function models. Journal of Econometrics 6: 21–37. Bain J. 1949. A note on pricing in monopoly and oligopoly. American Economic Review 39: 448–464. Bamford C, Dean T, McDougall P. 2000. An examination of the impact of initial founding conditions and decisions upon the performance of new bank start-ups. Journal of Business Venturing. 15: 253–277. Barnett WP, Greve HR, Park DY. 1994. An evolutionary model of organizational performance. Strategic Management Journal , Winter Special Issue 15: 11–28. Barnett WP, Hansen MT. 1996. The Red Queen in organizational evolution. Strategic Management Journal , Summer Special Issue 17: 139–157. Barnett WP, Sorenson O. 2002. The Red Queen in organizational creation and development. Industrial and Corporate Change 11: 289–325. Bartelsman EJ, Doms M. 2000. Understanding productivity: lessons from longitudinal microdata. Journal of Economic Literature 38: 569–594. Baum J, Mezias S. 1992. Localized competition and organizational failure in the Manhattan hotel industry, 1898–1990. Administrative Science Quarterly 37: 580–604. Baumol W. 1982. Contestable markets: an uprising in the theory of industry structure. American Economic Review 72: 1–15. Berger AN, Humphrey D. 1992. Measurement and efficiency in commercial banking. In Output Measurement in the Service Sectors, Griliches Z, Berndt ER, Bresnahan TF, Manser M (eds). University of Chicago Press: Chicago, IL; 245–302. Berger AN, Hancock D, Humphrey D. 1993. Bank efficiency derived from the profit function. Journal of Banking and Finance 17: 317–347. Blundell R, Griffith R, Van Reenan J. 1995. Dynamic count data models of technological innovation. Economic Journal 105: 333–344. Camerer C, Lovallo D. 1999. Overconfidence and excess entry: an experimental approach. American Economic Review 89: 306–318. Carroll G, Hannan M. 1989. Density dependence in the evolution of populations of newspaper organizations. American Sociological Review 54: 524–541. Cebenoyan A, Papaioannou G, Travlos N. 1992. Foreign takeover activity in the United States and wealth effects for target firm shareholders. Financial Management 21: 58–68. Cohen W, Levin R. 1989. Innovation and market structure. In Handbook of Industrial Organization, Strat. Mgmt. J., 26: 617–641 (2005)

640

A. M. Knott and H. E. Posen

Schmalensee R, Willig R (eds). Elsevier Science: Amsterdam; 1059–1108. Delacroix J, Rao H. 1994. Externalities and ecological theory: unbundling density dependence. In Evolutionary Dynamics of Organizations, Singh JV, Baum JAC (eds). Oxford University Press: Oxford; 255–268. Demsetz H. 1973. Industry structure, market rivalry and public policy. Journal of Law and Economics 16: 1–10. Fichman M, Levinthal D. 1991. Honeymoons and the liability of adolescence: a new perspective on duration dependence in social and organizational relationships. Academy of Management Review 16: 442–468. Foster L, Haltiwanger J, Krizan CJ. 2002. The Link between aggregate and micro productivity growth: evidence from retail trade. NBER working paper 9120. Geroski P, Pomroy R. 1990. Innovation and the evolution of market structure. Journal of Industrial Economics 38: 299–314. Hannan M. 1998. Rethinking age dependence in organizational mortality: logical formalizations. American Journal of Sociology 104: 126–164. Hannan M, Freeman J. 1989. Organizational Ecology. Harvard University Press: Cambridge, MA. Headd B. 2003. Redefining business success: distinguishing between closure and failure. Journal of Small Business Economics 21: 51–61. Hermalin B, Wallace N. 1994. The determinants of efficiency and solvency in savings and loans. Rand Journal of Economics 25: 361–381. Hicks DA. 1993. A nanoeconomic perspective on the growth and development of the Texas manufacturing base, 1970–1991. A report prepared for the Office of the Comptroller, State of Texas. Hoetker G, Agarwal R. 2003. Death hurts, but it isn’t fatal: the post-exit diffusion of knowledge created by innovative companies. Presentation at Academy of Management. Hou K, Robinson D. 2003. Industry concentration and average stock returns. SSRN working paper. Jaffe A. 1986. Technological opportunity and spillovers of R&D. American Economic Review 76: 984–1001. Jaffe A. 1988. Demand and supply influences in R&D intensity and productivity growth. Review of Economics and Statistics 70: 431–437. Jovanovic B. 1982. Selection and the evolution of industry. Econometrica 50: 649–670. Klepper S. 1996. Exit, entry, growth and innovation over the product life cycle. American Economic Review 86: 562–583. Knott A. 2002. Exploration and exploitation as complements. In The Strategic Management of Intellectual Capital and Organizational Knowledge, Choo C, Bontis N (eds). Oxford University Press: New York; 339–358. Knott AM. 2003. Persistent heterogeneity and sustainable innovation. Strategic Management Journal 24(8): 687–705. Copyright  2005 John Wiley & Sons, Ltd.

Levinthal D. 1991. Random-walks and organizational mortality. Administrative Science Quarterly 36: 397–420. Levitt B, March JG. 1988. Organizational learning. Annual Review of Sociology 14: 319–340. Lieberman M. 1987. The learning-curve, diffusion, and competitive strategy. Strategic Management Journal 8(5): 441–452. Lippman SS, Rumelt RP. 1982. Uncertain imitability: an analysis of interfirm differences in efficiency under competition. Bell Journal of Economics 13: 418–438. Mankiw NG, Whinston MD. 1986. Free entry and social inefficiency. RAND Journal of Economics 17: 48–58. Mester L. 1993. Efficiency in the savings and loan industry. Journal of Banking and Finance 17: 267–286. Milgrom P, Roberts J. 1982. Limit pricing and entry under incomplete information: an equilibrium analysis. Econometrica 50: 443–460. Nelson R, Winter S. 1982. An Evolutionary Theory of Economic Change. Belknap Press of Harvard University Press: Cambridge, MA. Nickell S. 1996. Competition and corporate performance. Journal of Political Economy 104: 724–746. Peretto PF. 1999. Cost reduction, entry, and the interdependence of market structure and economic growth. Journal of Monetary Economics 43: 173–195. Petersen M, Rajan R. 2002. Does distance still matter? The information revolution in small business lending. Journal of Finance 57: 2533–2570. Porter M. 1990. The Competitive Advantage of Nations. Free Press: New York. Reinganum JF. 1989. The timing of innovation: research, development, and diffusion. In: Handbook of Industrial Organization, Vol. 1, Schmalensee R, Willig RD (eds). North-Holland: Amsterdam; 849–908. Schumpeter J. 1942. Capitalism, Socialism and Democracy. Harper: New York. Sinclair G, Klepper S, Cohen W. 2000. What’s experience got to do with it? Sources of cost reduction in a large specialty chemicals producer. Management Science 46: 28–45. Spence M. 1981. The learning curve and competition. Bell Journal of Economics 12: 49–70. Stinchcombe A. 1965. Social structure and organizations. In Handbook of Organizations, March JG (ed). McGraw-Hill: New York; 142–169. Swaminathan A. 1996. Environmental conditions at founding and organizational mortality: a trial-byfire model. Academy of Management Journal 39: 1350–1377. U.S. Small Business Administration. 1999. The State of Small Business: A Report of the President, 1998 . United States Government Printing Office Business Administration: Washington, DC. Vickers J. 1995. Concepts of competition. Oxford Economic Papers 47: 1–23. Wheelock D, Wilson P. 1995. Evaluating the efficiency of commercial banks: does our view of what banks Strat. Mgmt. J., 26: 617–641 (2005)

Is Failure Good? do matter? Review—Federal Reserve Bank of St Louis 77: 39–52. Wu X, Knott A. 2005. Entrepreneurial risk and market entry. Working paper, SBA Office of Economic Research.

Copyright  2005 John Wiley & Sons, Ltd.

641

Yelle L. 1979. Learning curve, historical review. Decision Science 10: 302–327.

Strat. Mgmt. J., 26: 617–641 (2005)