Latent submarket dynamics and industry evolution: lessons from the US laser industry

1. Introduction

It is common to talk about niches within industries. Until recently, however, the idea of niches, or submarkets in today’s economic parlance, has not been prominent in economic modeling. When firms are modeled as producing differentiated products, it is commonly assumed that each firm produces a single product variety from among a continuum of close substitutes. In contrast, submarkets are defined as islands of activity that are insulated from the rest of an industry on both the demand and supply side. Firms can belong to more than one submarket, and submarkets may come and go over time, giving rise to entry, exit, firm growth, and decline. Such dynamics have been implicated as being important in a number of industries (Christensen, 1993; Christensen, Suarez and Utterback, 1998; Malerba et al., 1999; Bottazzi et al., 2001; de Figueiredo and Silverman, 2007). Models featuring submarkets have been developed to provide competitive accounts of disruptive technologies (Adner, 2002; Adner and Zemsky, 2005) and to explain a rich array of regularities that have been accumulating about firm-size distributions (de Juan, 2003; Sutton, 1998; Buzzacchi and Valletti, 2006), firm innovation and patenting rates (Klette and Kortum, 2004), and firm growth and exit rates (Bottazzi and Secchi, 2006; Klepper and Thompson, 2006).

In this article, we explore the power of submarket dynamics to help explain the determinants of an industry’s market structure and the character of innovation. Klepper and Thompson (2006) discuss various submarket developments such as obsolescence of technologies, decline in geographic areas, and changes in regulation that could alter an industry’s market structure.¹ We consider a novel way this could occur. Submarkets often begin as independent but have the potential of becoming substitutes if one is improved sufficiently, which can be triggered by an unexpected technological advance. This can contribute to an escalation of R&D within the improved submarket that leads to the emergence of an integrative submarket with far-reaching implications.² We develop a model along these lines that fuses the stochastic submarket model of industry evolution developed by Klepper and Thompson (2006) with the model of shakeouts in Klepper (1996, 2002).

We apply the model to the US laser industry. Lasers come in many forms that historically defined independent submarkets. For many years this led to a rise in the number of laser producers as reflected in Figure 1, which presents the annual number of entrants, exits, producers, and industry (real) sales and output from the inception of the US laser industry in 1961 through 2007. In the late 1980s, a major technological advance in one laser submarket fundamentally altered incentives to improve that submarket’s product. We contend this led to the emergence of an integrative submarket and to the shakeout of laser producers reflected in Figure 1 that has been ongoing since 1996. Seldom, if ever, has an industry experienced a pronounced shakeout after >30 years of growth in output and the number of producers (Gort and Klepper, 1982). We show that our model can readily explain these two eras and has numerous other implications regarding innovation, entry, exit, and firm survival that are supported in the laser industry.

The laser industry reflects the power of submarket dynamics to explain a sudden change in the evolution of a new industry. The notion of an integrative submarket is related to the idea of a dominant design, which has also been used to explain shakeouts (Utterback and Suárez, 1993). We discuss the overlap in the two theories and also how they differ, with our theory able to explain a rich array of empirical patterns in the laser industry not addressed by the dominant design theory. We also consider other theories of shakeouts, noting where they fall short in explaining the wide array of empirical patterns characterizing the shakeout in lasers. Other innovative industries that experienced shakeouts such as hard disk drives and automobiles are also considered, and we discuss how our model can be generalized to explain the historical evolution of their market structure.

The article is organized as follows. In Section 2, we present a brief history of the laser industry. In Section 3, we lay out our model and derive various implications from it. In Section 4, we test the implications. In Section 5, we discuss the consistency of our findings with other models of shakeouts and reflect on the role that submarket dynamics played in the evolutionary paths followed by various innovative new industries.

2. Overview of the laser industry

The laser is based on the idea that if an electron in an excited state is bombarded by a photon of proper energy, another photon of identical energy is emitted. Materials in lasers are excited by a suitable energy source and photons of identical energy are built up in a cavity and then released as a tightly collimated beam of light of a single wavelength with all waves in phase. These qualities along with the great potential intensity of laser light make lasers useful for a wide range of applications.

Numerous materials have been made to lase. The material determines the wavelength of the laser light, which in turn dictates the applications serviced by a laser. Following Klepper and Sleeper (2005), we distinguish nine broad categories of lasers: solid state (crystals), semiconductor, chemical dye, and six gas lasers, helium neon (HeNe), carbon dioxide (CO2), ion, excimer, helium cadmium (HeCd), and a catchall category of other gas lasers, which correspond largely to nonoverlapping clusters of wavelengths. Solid-state and semiconductor lasers can be further broken down in terms of the crystals and semiconductor materials used. Distinctions also can be made in each of the nine laser categories according to such factors as power (intensity), continuous versus pulsed operation, longevity, ruggedness, and cost of operation.

These characteristics along with the wavelength of the laser and its overall efficiency dictate its uses. For example, CO2 lasers emit infrared light at 10,600 nm that can cut thick sheets of metal, excimer lasers emit ultraviolet light that is useful in semiconductor manufacturing and treating the human eye (193 nm), HeNe lasers emit red light that is useful in barcode scanning and alignment tools (633 nm), and semiconductor lasers are small and inexpensive, making them useful in large-scale applications such as CD players and laser printers. At times, existing lasers within a category have been challenged by new ones in the same category, such as when Ruby crystal lasers were displaced by lasers based on neodymium-doped yttrium aluminum garnet crystals (called Nd:YAG). On rarer occasions an existing laser was challenged by a new one in another category, such as when semiconductor lasers that emit red light were developed to challenge HeNe lasers in barcode scanning applications. But for the most part, the development of new lasers has brought new users into the industry without affecting the demand for other lasers. While occasionally lasers are successfully challenged by other types of light sources or by nonlight sources in certain applications, such as in surgery, over time the industry has grown greatly through the development of new types of lasers to meet new user needs.

We will conceptualize this process of growth in terms of the creation and in rarer instances destruction of independent product submarkets. It is an exaggeration to claim no competition or substitution between different types of lasers, as the case of HeNe and semiconductor lasers illustrates. But for the most part differences in the wavelengths and other properties of lasers have tended to differentiate the submarkets they serve and limited the extent of substitutability among them on the demand side. It is also an exaggeration to claim no scope economies across the lasers produced in different submarkets. However, historically the majority of laser firms has specialized in only one or two types of lasers and has coexisted with a small cadre of firms producing a wide range of lasers,³ which is characteristic of industries characterized by limited scope economies across products.

A novel development beginning around 1988, though, changed in a far-reaching way submarket dynamics in the industry. Lasers need to be excited or “pumped” to initiate and sustain laser activity. Different energy sources are used to pump different types of lasers. Historically, solid-state lasers were pumped by flash and arc lamps, but it has long been known that semiconductor lasers are potentially more efficient pumps for solid-state lasers. After a long stream of research in universities and government labs combined with improvements in semiconductor lasers, around 1988 practical diode (semiconductor) pumps for solid-state lasers emerged. These diode pumps are up to 20 times smaller than lamp pumps of similar technical specifications (DeShazer, 1994). They opened up a rich trajectory of improvements in the efficiency of solid-state lasers (Byer, 1988) that has greatly expanded their use in two significant ways.

First, steady improvements in diode pumps resulted in decreases in the prices of solid-state lasers that made them competitive with other lasers, particularly CO2 lasers, at their natural wavelengths.⁴ Diode pumping also expanded the range of solid-state materials that could be used as lasers, leading to the creation of new solid-state lasers servicing a wider range of applications (Koechner, 2006). Among these new solid-state lasers are a whole new category called fiber lasers. These lasers are composed of special fiber-optic cables that guide and amplify laser light emitted by diode pumps. They are particularly useful for medical and industrial applications where laser light needs to be delivered to hard-to-reach places.

Second, diode-pumped solid-state (DPSS) lasers improved the efficiency of solid-state lasers sufficiently that it became feasible to use nonlinear optical devices to alter the wavelength of the emitted light to produce harmonic wavelengths. Such devices can be used to double the frequency of light, which halves the wavelength, and can be applied repeatedly and in combinations to generate a series of different wavelengths of light from a particular laser. Energy is lost in the process, which along with the cost of the nonlinear devices (and in the case of gas lasers, packaging problems) limited the historical use of wavelength conversion. But with the advent of DPSS lasers, it became feasible to use wavelength conversion to create new solid-state lasers emitting wavelengths in the visible range, making them competitive with HeNe, ion, dye, and other types of gas lasers.⁵ We compiled a list from Laser Focus of all the distinct wavelengths supported by DPSS lasers that produce an uninterrupted laser beam. Figure 2 illustrates the broadening of wavelengths offered by solid-state lasers with a range of natural wavelengths of 1000–1550 nm and harmonic wavelengths spanning 250–800 nm.

The modern evolution of solid-state lasers has been described as a transformation from producing a ray to a rainbow. Our theoretical model focuses on how this ushered in a new era of competitive dynamics in the industry.

3. The model

The model fuses the stochastic submarket model in Klepper and Thompson (2006) with the model of shakeouts in Klepper (1996, 2002) to explain how the advent of an innovation such as DPSS lasers could trigger a shakeout in an industry previously composed of independent submarkets.

3.1 Specification of the model

Klepper and Thompson (2006) cast their model in general terms to accommodate a wide range of underlying processes, and we retain that approach to keep the model flexible. We cast the model in discrete time. Let t = 0, 1, 2, … denote the period, where period 0 marks the start of the industry when its first submarket is created. In all subsequent periods, there is a probability λ of a new submarket being created. At the start of the industry and in all subsequent periods, there are K potential entrants into each submarket. Each has a probability of θ of entering a new submarket, independent of the other submarkets in which it is active. If a firm enters a submarket, on entry it randomly draws an output Q in the submarket from a distribution F(Q). There is a constant hazard γ of a submarket being destroyed in each period. If a submarket is destroyed, firms lose their entire output in the submarket; otherwise a firm’s output in a submarket remains constant over time.

A firm’s overall size in period t is the sum of its output in each submarket in which it is active. By definition, a firm enters the industry when it enters its first submarket and exits the industry when the last submarket it is in gets destroyed. All growth and decline in firm size are driven by the creation and destruction of submarkets. Klepper and Thompson (2006) show that this model can explain a number of regularities in firm growth and exit and also various patterns in the evolution of the laser industry through 1994 (before its shakeout commenced).

We broaden the model as follows. First, we allow for latent substitutability between submarkets on the demand side. Let the total quantity demanded of the product of a submarket be f(P.), where P. is the price of the submarket’s product and f’ < 0, and let c denote the initial average cost of producers of the submarket’s product. If P. falls below a cutoff P.* < c, some buyers switch their purchases from other submarkets to this lower-priced submarket. Let the probability of a switch for any buyer be denoted as s(P.), where s’ < 0 for P. < P.*. This allows for buyers to be heterogeneous (only some switch at any given price). It also implicitly assumes that the composition of buyers from each firm is the same in that each buyer has the same probability of switching to a submarket in which P. falls below P.*. For simplicity, we assume that if the price of the product in any submarket falls below P.*, new submarkets will no longer be economically viable and so no new submarket will be developed.

Second, instead of firms randomly drawing an output initially in a new submarket, we specify entry into a new submarket as a two-step process. This allows us to distinguish between early and late entrants. We assume that when a new submarket is created, firms have to conduct product R&D to enter the market. All have the same probability of this effort being successful. Only a few succeed, though, and they play a Cournot-Nash game, charging a price P. > c and dividing up the total quantity demanded in the submarket equally. One period later the successful firms can be costlessly imitated by all other potential entrants. Bertrand competition then causes the price P. to be driven down to c. This increases the quantity demanded in the new submarket, enabling some firms to expand and others to enter. We model this by assuming that the increase in the quantity demanded is randomly distributed across the K firms, with some firms getting none of the increased output. If these firms were not initial innovators, then by definition they do not enter the submarket. This two-stage entry process implies that earlier entrants will be expected to be larger than later entrants.

Third, following Klepper (1996, 2002) we allow for the possibility of firms engaging in R&D to lower the average cost of production of a submarket’s product. Let c_t denote the common average cost of production of all producers in a submarket at the start of period t. If producer i spends r_it on process R&D in period t, its average cost of production declines to c_t − g(r_it), where g’ > 0 and g’’ < 0 to reflect diminishing returns. We assume that g’(0) is bounded and g’(r_it) → 0 as r_it → ∞. Furthermore, process innovation is assumed to be costlessly imitated by all producers in a submarket one period later, which is captured by specifying that c_t+1 = c_t− max_i(r_it).

Let P_t denote the price of a submarket in period t. All firms are assumed to be price takers, with the price in every period such that the quantity demanded equals the quantity supplied. If c_t− g(r_it) < P_t then a producer can expand its output subject to a cost of growth. If the total quantity demanded rises in period t due to firms expanding, causing P_t to be less than P_{t −1}, it is assumed that all firms that remain in a submarket share in the expansion of the market in proportion to their total output (i.e., their market share remains the same). If firm i wants to increase its output further by ΔQ_it, it must take away market share from rivals, which involves a cost of growth of m(ΔQ_it), where m’(0) = 0, m’’ > 0, and m(ΔQ_it) → ∞ as ΔQ_it → ∞. If a producer’s maximum profit in a submarket in period t (given its profit-maximizing choices of r_it and ΔQ_it) is negative then it exits the submarket and if it is zero it is indifferent between exiting and staying in the submarket. For simplicity, all exiters are assumed to be atomistic and hence their exit has no effect on the market-clearing price.⁶ We assume that the future potential of improvements in a submarket is unknown and in each period t firms choose r_it and ΔQ_it to maximize their profits in period t.

Last, all submarkets are assumed to be too small initially for R&D to be profitable for any firm, but then this changes in one submarket. This is specified as follows. Let Q* denote the minimum firm output in each submarket for which process R&D is profitable, where Q* is defined by g’(0)Q* = 1. We assume that up to period T, the output of all firms in each submarket is less than Q*. In period T, an exogenous technological change in one submarket, denoted as submarket d, increases the total demand for the submarket’s product. The growth in demand is randomly allocated across the K potential producers. Those that get a positive output change that were not previously in the submarket enter the submarket for the first time. We assume that in period T some firms have an output larger than Q*.

3.2 Industry evolution up to period T

We first consider the implications of the model for the evolution of the industry up to period T. For all periods t < T, the model is effectively the same as in Klepper and Thompson (2006). All firms are equally likely to enter any new submarket, the output of entrants into a submarket (after imitation of the innovators) is randomly determined, and submarkets have a constant hazard of destruction in every period. Furthermore, no process R&D is conducted in any submarket. Consequently, each submarket’s price after imitation equals c and there is no substitution on the demand side among submarkets.

Klepper and Thompson (2006) establish that under these conditions there is a steady-state distribution for the number of submarkets, firms, and industry output. This implies that initially the number of submarkets and firms will rise toward their steady-state values and then vary randomly about them. Furthermore, in any given period, entry and exit will be concentrated in a limited number of submarkets. Summarizing,

Result 1: After the start of the industry, the number of firms is expected to rise over time toward its steady-state value and then vary about it and in any given period entry and exit are concentrated in a limited number of submarkets.

Up to period T, firms exit submarkets only if they are destroyed. The probability of a firm exiting the industry in some time interval equals the probability of all of its submarkets being destroyed (and no new ones being created that it enters). This is clearly a decreasing function of the number of submarkets in which it produces at the start of the interval. The expected number of submarkets in which a firm produces t periods after its entry is (1 − γ)^t + θλ(1 − γ)^j, where the first term is the probability that the first submarket it entered is still active after t periods and the second term is the expected number of additional, new submarkets it entered that are still active after t periods.⁷ Assuming θλ is sufficiently large relative to γ^,,⁸ the expected number of submarkets rises with t (the age of the firm) and is independent of when the firm entered the industry. Therefore,

Result 2: The hazard of exit from the industry falls with a firm’s age and is independent of its overall size and time of entry, whereas the hazard of a firm exiting a submarket is independent of its age, size, and time of entry in the submarket and industry.

3.3 Evolution beginning with period T

Equation (2) implies that the larger the output of the firm in submarket d, the greater the effort it devotes to process R&D. Intuitively, the profit from lowering average cost is proportional to the firm’s output, so the larger a firm’s output then the more it spends on process R&D. For firms to cover the cost of their R&D, P_t − c_t + g(r_it) must be positive, in which case equation (3) implies that ΔQ_it > 0 for firms that remain in submarket d. Consequently, for submarket d to clear, some firms must exit in each period, which requires P_t to be nonincreasing over time.⁹ With larger firms earning greater profits from R&D, the least profitable firms in submarket d are the smallest firms. These firms are expected to have entered submarket d latest, hence the latest entrants in submarket d will be expected to exit first. Eventually, P_t falls below P.* and buyers switch from other submarkets to submarket d. When a firm in another submarket loses all of its buyers it exits the submarket, and the industry too if it was only producing in that submarket. Given that all buyers in a submarket have the same probability of switching to another submarket when its price falls sufficiently, the smallest firms in other submarkets are expected to exit first. These firms are more likely to be the latest entrants in their submarket.

Thus, beginning with period T, the industry experiences the following evolution. Initially some firms enter submarket d and the number of firms in the industry rises as long as no submarket is destroyed in period T. Subsequently, some firms engage in process R&D in submarket d and the price of submarket d declines. The total quantity demanded and hence output in submarket d (and the industry) rises and firms exit from submarket d, giving rise to a shakeout in submarket d. At the industry level, in addition to exiters from submarket d, some firms eventually exit other submarkets as well, and the number of producers and output in these submarkets both fall. Some of these exiters will also exit the industry, and if no new submarkets are created then the number of firms in the industry declines, giving rise to a shakeout at the industry level. Summarizing,

Result 3: After period T, the number of firms in submarket d first rises and then steadily falls as the output of submarket d rises, giving rise to a shakeout in submarket d. In other affected submarkets, the number of firms and output eventually both fall together and the industry experiences an overall shakeout.

When submarket d is created firms engage in product R&D to enter the submarket. Subsequently, firms do no further R&D until period T + 1, at which point some firms begin to engage in process R&D. Therefore,

Result 4: Beginning in period T + 1, R&D in submarket d rises and stays higher than it was in any period after the start of submarket d.

In period T, the probability of a firm exiting the industry equals the probability of all its submarkets being destroyed (and no new ones created that it enters). In period T + 1 and later periods, firms could also exit from submarket d and other affected submarkets without them being destroyed. The firms that exit submarket d after period T are more likely to be the later entrants into submarket d. Similarly, the firms that exit the affected submarkets first are more likely to be those that entered the submarkets later. Hence,

Result 5: After period T, the hazard of exit from the industry rises and remains higher than in period T. The hazard of exit in submarket d and other affected submarkets is greater for later entrants into the submarkets and hence the hazard of exit from the industry is greater for later entrants.

If no new submarket is developed after period T, entry initially is concentrated in submarket d and then ceases. Exit is also initially concentrated in submarket d and any destroyed submarkets and then occurs in the other affected submarkets as well as any that are destroyed. Therefore,

Result 6: Beginning with period T, entry and exit are initially concentrated in submarket d and then entry is not (concentrated) in any submarket and exit is dispersed over multiple submarkets.

3.4 Testable implications

Ideally, to test our implications, we would define submarkets based on a rich set of product attributes but we base our definition on wavelengths or clusters of wavelengths due to data limitations. However, our focus on the upstream laser industry (and not the end consumers who consider a much larger bundle of product attributes that downstream system manufacturers will have added along the value chain) mitigates definitional concerns.¹⁰ We have data on firms, output, sales, and patents at the level of the nine broad categories of lasers, which we condense to eight categories by including HeCd, which was small, in the category of other gas lasers. All of the categories contain at least one meaningful submarket if not more. Submarkets are conceptual and compartmentalizing the many different types of lasers that have been developed into meaningful submarkets is not feasible. Therefore, we adapt the results above to accommodate the existence of multiple possible submarkets within each of the broad types of lasers. Klepper and Thompson (2006) show that the predictions of the model pertain to these broader laser categories as well as the underlying submarkets under the assumption that submarkets are randomly distributed among these laser categories.¹¹ In empirical analyses, we control for broad laser categories to address the potential nonrandom distribution of submarkets.

We identify period T with the advent of DPSS lasers in 1988. As the price of these lasers decreased, they made in-roads into markets serviced by CO2 lasers. Through wavelength conversion, DPSS lasers also eventually became competitive with primarily HeNe, ion, dye, and other gas lasers. Based on these assumptions, Results 1–6 directly imply the following testable hypotheses concerning the laser industry:

Hypothesis 1: Entry into the industry initially is positive and output and the number of producers rises. After 1988 output and the number of producers continues to rise at first but eventually the number of producers falls even as the output of the industry continues to rise, giving rise to a shakeout in the industry.

Hypothesis 2: The output and number of producers of solid-state lasers rise over time, particularly after the advent of DPSS lasers in 1988, and then at a later point the number of producers declines even as the output of solid-state lasers continues to rise.

Hypothesis 3: For CO2, HeNe, ion, dye, and other gas lasers, initially entry is positive and output and the number of producers rises. Some time after 1988, entry declines and both output and the number of producers fall in tandem.

Hypothesis 4: Annual R&D expenditures related to solid-state lasers rise around 1988 and remain at a higher level in subsequent periods, which gives rise to a greater number of patents and innovations per year related to solid-state lasers after 1988. The leading patenters/innovators are the largest producers of solid-state lasers, which are more likely to be from the ranks of the earliest producers of solid-state lasers.

Hypothesis 5: From the beginning of the industry up to 1988, entry and exit in successive time intervals are concentrated in a limited number of laser types, with the types of laser differing across time intervals (and for entrants and exiters within time intervals). After 1988, at first entry and exit are concentrated in solid-state lasers. In subsequent time intervals neither entry nor exit is concentrated in any one type of laser.

Hypothesis 6: At some time after 1988, the hazard of exit from the industry rises and remains at a higher level.

Hypothesis 7: For all entrants, the hazard of exit declines with age. At each age, it is higher for entrants after 1988 than earlier entrants, but among pre-1988 entrants it is unrelated to time of entry (for ages corresponding to years before 1988).

Hypothesis 8: After 1988, the hazard of exit from solid-state, CO2, HeNe, ion, dye, and other gas lasers is greater for entrants after 1988 than earlier entrants, whereas before 1988 it is unrelated to time of entry (assuming each type of laser is composed of a single submarket¹²).

4. Data analysis

Data on the US laser producers are compiled using Laser Focus World’s annual Buyer’s Guide, which is based on an annual survey of laser producers. The data, originally collected by Sleeper (1998), are extended to cover the whole period from 1961 to 2007. Many different categories of lasers are listed, which we aggregated to the eight broad laser types, enabling us to identify which of the eight types of lasers firms produced each year.¹³ Annual data on global sales and output (since 1986) are collected from the same source.

4.1 Shakeouts and trends in output and the number of producers

As noted in the introduction, Figure 1 indicates that the number of US laser producers increased steadily through 1996 to a peak of 172 and then declined steadily to 87 in 2007.¹⁴ Over this same period global (real) sales of lasers rose steadily,¹⁵ with the average annual rate of growth of (real) sales increasing from 16.8% in 1974–1995 to 20% in 1996–2007. Thus, consistent with hypothesis 1, the laser industry has been undergoing a pronounced shakeout of producers that began after the advent of DPSS lasers.

Figure 3 is the analog of Figure 1 for solid-state lasers. Apart from a modest decline in the number of producers from 35 in 1967 to 21 in 1978, the number of producers rose over time to a peak of 78 in 1997 and then declined to 50 by 2007.¹⁶ Global production of solid-state lasers grew sharply throughout this period, causing solid-state lasers to grow from 2% of all nondiode lasers produced in 1994 to 40% in 2007. Nevertheless, consistent with hypothesis 2, the number of producers of solid-state lasers went through a pronounced shakeout that is still on-going. These patterns parallel those at the industry level, as predicted.

Figure 4 presents the analogous patterns for CO2, HeNe, ion, other gas, and dye lasers. While each is subject to idiosyncratic factors, consistent with hypothesis 3, all of them reflect considerable growth in output and the number of producers from the start of the industry and then a decline in the number of producers since 1997 (if not earlier). In contrast to the overall market where total output grew sharply during its shakeout, in all but CO2 lasers the decline in the number of producers was accompanied by a fall in output, as predicted in hypothesis 3.

It may be argued that these patterns are driven by the way we define submarkets. To alleviate the concern, we investigate patterns of entry and exit at the most disaggregated-level our data permit and observe similar patterns. We do not report graphs at the disaggregated-level for brevity’s sake but describe the pattern in the largest submarket for solid-state lasers, Nd:YAG (1064 nm). The number of Nd:YAG producers increased from 2 in 1962 to 53 in 1998 and fell to 26 in 2007; those using diode-pumping among them increased from 5 in 1989 to 38 in 1999 and fell to 19 in 2007.

4.2 Technological change related to solid-state lasers

Hypothesis 4 predicts a rise in patenting related to solid-state lasers beginning in the late 1980s. We consider four types of patents related to solid-state lasers. Patent class 372 refers to coherent light generators, which includes all types of lasers. Subclasses 40–50 relate to solid-state and diode laser materials while subclasses 51–65 relate to gas and dye laser materials. Figure 5a presents the annual number of patents in these subclasses from 1961 to 2007.¹⁷ It reflects a rise in patenting in solid-state and diode laser materials in comparison with patenting in gas and dye laser materials beginning in the 1980s, consistent with hypothesis 4.

The rise in patenting should be related to diode-pumping of solid-state lasers. Subclasses 70–80 in class 372 refer to the pumping of solid-state lasers using either a conventional source or a laser and subclasses 81–90 refer to pumping gas lasers using electrical or chemical energy sources. Figure 5b presents the annual number of patents in these two sets of subclasses. Consistent with hypothesis 4, in the 1980s there is a sharp increase in the number of patents related to the pumping of solid-state lasers, whereas in the same period patents related to the pumping of gas and dye lasers decline.

The improvements in the pumping of solid-state lasers made it feasible to use wavelength conversion to develop solid-state lasers generating a wider range of wavelengths. Figure 5c presents the annual number of patents in subclasses 21 and 22 in class 372 pertaining to wavelength conversion. Consistent with hypothesis 4, there is little patenting in these classes until the late 1980s, after which the annual number of patents increases sharply. Last, diode pumping enabled the development of fiber lasers, which have attracted a sizable niche market. Subclass 6 in class 372 pertains to solid-state lasers that use diode pumping and fiber channels. Figure 5d presents the annual number of patents in this subclass. As expected, beginning in the late 1980s there were a substantial number of patents in this subclass.

It is also predicted that the leading innovators in solid-state lasers will be the largest producers of solid-state lasers, which are expected to come from the earliest entrants. Perhaps the purest indicator of patenting induced by diode pumping is patents related to wavelength conversion. Table 1 lists the top five US laser patenters in wavelength conversion since 1988. It also lists when they began producing solid-state lasers and the number of years they produced solid-state lasers after 1988.

The top two patenters after 1988 were Spectra Physics and Coherent. They were not only the largest producers of lasers overall but among the 257 producers of solid-state lasers in 1989 and later years, they were the 6th and 11th (tied) earliest entrants. Amoco entered later in 1988, but was still the 40th earliest entrant among the 257 solid-state producers in 1989 and later. Furthermore, it is a spinoff from Amoco Oil’s corporate research lab and its founder recognized the potential of DPSS lasers early on (Feder, 1988).¹⁸ Photonics Industries and Uniphase were later solid-state entrants in 1994, but were still relatively early entrants that were tied for the 99th earliest entrants among the 257 solid-state producers in 1989 and later years. Other than Amoco, all the firms were still producing solid-state lasers at the end of our data period in 2007.¹⁹ Thus, consistent with hypothesis 4, patenting in DPSS-related areas was led by early and long-lived producers of solid-state lasers.

Another measure of innovation pertaining to solid-state lasers is the rate of creation of new types of solid-state lasers. The number of new categories of solid-state lasers averaged 0.60 per year from 1961 to 1987 and then increased to 1.05 per year from 1988 to 2007, consistent with hypothesis 4. Many of these new laser types were made possible by diode pumping and thus shared a common technology. Consequently, they tended to be produced in tandem, especially by the leading solid-state producers. For example, Coherent and Spectra Physics, the two leading patenters in wavelength conversion, produced 17 and 15 different types of solid-state lasers, respectively. Both were also heavily involved in producing semiconductor diode arrays for diode pumping. Only two other firms, Schwartz Electro Optics and Excel Technology, produced a comparable number of types of solid-state lasers as Coherent and Spectra Physics, but judging from the list of semiconductor producers neither seems to have been as involved as Coherent and Spectra Physics in diode arrays. Consistent with the model, Schwartz (tied for 36th) and Excel (tied for 56th) were also relatively early entrants among the 257 firms that produced solid-state lasers in 1989 and later years.

Another reflection of innovation related to DPSS lasers is the proliferation of new wavelengths of solid-state lasers reflected in Figure 2. Before the advent of DPSS lasers, solid-state lasers generated wavelengths only in the range of 1064 nm. Diode pumping led to the creation of new DPSS lasers that generated natural wavelengths in the 1000–1600 nm range and through wavelength conversion harmonic wavelengths in the visible spectrum spanning 250–800 nm. Not surprisingly, Coherent and Spectra Physics were in the vanguard of the new DPSS lasers. Among the 49 firms in 2006 listed as producing continuous wave DPSS lasers, Coherent and Spectra Physics ranked two and three with 32 and 28 lasers with distinct wavelengths, respectively, in the 2006 specification tables of Laser Focus (pp. 98–106). Only two other firms had lasers with >10 wavelengths, including Melles Griot with 35 and Lee Laser with 26. Consistent with the model, both were relatively early solid-state entrants and continued producing until the end of our data period—Melles Griot entered in 1990 and Lee entered in 1986, qualifying them as the 51st (tied) and 24th (tied) earliest entrants, respectively, among the 257 solid-state producers in 1989 and later.

4.3 Entry and exit in laser types

Hypothesis 5 predicts that until fairly recently, entry and exit in short time intervals should be concentrated in just a few types of lasers, with these laser types changing over time. In more recent years, it is expected that entry and exit initially would be concentrated in solid-state lasers and then not concentrated in any one laser type.

To test these predictions, following Klepper and Thompson (2006) the history of the industry is divided into 5-year intervals, which provide a sufficiently large sample for each period to carry out statistical tests. The analysis begins in 1970 when all but excimer lasers had been developed, with excimer lasers excluded from the analysis. For entrants and exiters, the laser type(s) produced in the first and the last year, respectively, are reported in Tables 2 and 3 for each 5-year interval and for all periods combined.²⁰ As the tables indicate, the total number of entrants and exiters producing each laser type over the entire period varies greatly by laser type, reflecting possibly different number of submarkets within each laser type and different sizes of submarkets.

If hypothesis 5 is correct, then the numbers in each cell of Tables 2 and 3 should be close to zero or the total number of entrants (exiters) in the respective time interval, which is denoted as Tot_t. The alternative, null hypothesis for entry (exit is analogous) is that the numbers in each cell should equal Tot_t times the fraction of entrants in all periods that initially produced the respective laser type, denoted as f_i. For example, entrants collectively produced 892 lasers in their first year and the fraction of these that were CO2 lasers was 124/892 = 0.139. Hence, in the period 1980–1984 when entrants collectively produced 103 lasers in their first year, under the null hypothesis the expected number of entrants producing CO2 lasers is 0.139 × 103 = 14.32.

We can test the null hypothesis using the test statistic Σ_i,t [(E_it − Tot_t*f_i)/Tot_t*f_i]², where E_it is the actual number of entrants (exits) in Table 2 (3) in laser type i in time interval t and Tot_t*f_i is the expected number under the null hypothesis. It is computed for the first four 5-year intervals corresponding to the pre-DPSS era of the industry for entrants and exiters. Ignoring that some firms produced multiple lasers in their first or last year, the test statistic has a distribution with 18 degrees of freedom. For entrants the test statistic equals 52.96, which is significantly different from 0 at the .001 level, and for exiters it equals 32.47, which is significantly different from 0 at the .02 level. Hence, the null hypothesis of no significant clustering of entry and exit in the early era is rejected, consistent with hypothesis 5.

It is also predicted that the submarkets where entry and exit are concentrated changes during the early era. Following Klepper and Thompson (2006), laser types where the number of entrants producing a laser type in a time period exceeds the expected number plus 1.65 standard deviations are identified and shown in bold face in Tables 2 and 3.²¹ Early on entry and exit are concentrated in HeNe lasers but then shift to other types of lasers, consistent with hypothesis 5.²²

In the later era we distinguish three periods, 1990–1994, 1995–1999, and 2000–2007. At some point after 1988 entry was expected to be concentrated in solid-state lasers, followed by a concentration of exit in solid-state lasers, after which entry and exit were not expected to be concentrated in any one laser. We see that by the 1995–1999 period, entry was concentrated in solid-state lasers and then again solid-state lasers and semiconductor lasers in 2000–2007. Exit was not concentrated in any one type of laser in the 1995–1999 period but was concentrated in solid-state and semiconductor lasers in the 2000–2007 period. Thus, we do see the predicted concentration of entry and then exit in solid-state lasers, but it appears that sufficient time has not yet elapsed for the solid-state market to have settled down.

4.4 Hazard of industry and submarket exit

Hypotheses 6, 7, and 8 predict various changes in firm hazard patterns with the advent of DPSS lasers after 1988.

Hypothesis 6 predicts that the overall hazard rate from the industry will be greater after 1988. Figure 6 reports the annual number of exits expressed as a fraction of firms in the prior year for the period 1965–2007. The fraction of firms exiting is high initially, rising briefly >20% in 1974, and then declines to a low of ∼4% in 1982 before recovering to 10% in 1987. Thereafter the exit rate is markedly higher, varying between 10% and 22% per year. Thus, consistent with hypothesis 6, around 1988 there appears to be a marked increase in the overall industry hazard rate that does not abate over time.

Hypothesis 7 allows for the hazard rate to vary with age and predicts that at comparable ages the hazard of exit should be unrelated to time of entry for entrants before 1988 but should be greater for entrants after 1988. To control for age, Kaplan–Meier (K-M) survival curves were estimated for three cohorts of entrants corresponding to the years 1961–1974, 1975–1988, and 1989–2007. These curves, which are presented in Figure 7, estimate the probability of entrants surviving to each age. The vertical axis is scaled logarithmically so that the negative of the slope of a K-M curve at a given age is the hazard of exit of the respective cohort at that age. Up to age 15, nearly all exits in the earliest cohort of entrants and most in the middle cohort pertain to the pre-1988 era. Consistent with hypothesis 7, the K-M curves for the two cohorts do not differ much and are not statistically different at the .10 level (P = 0.7103) based on the log rank test. In contrast, all the exits for the post-1988 entrants correspond to years after 1988. Consistent with hypothesis 7, the K-M curve for the post-1988 entrants is steeper than the other two, reflecting a higher hazard across all ages, and is significantly different from the other two at the .005 level based on the log rank test.

Hypothesis 8 predicts that after 1988, the hazard of exit for solid-state producers and for CO2, HeNe, ion, dye, and other gas laser producers should be lower for earlier entrants.²³ Consider first the 42 solid-state producers that survived until 1988. Only seven of them entered before 1975, so they are grouped into a single category of early entrants and their survival experience is compared with the 215 solid-state producers that entered after 1988. Figure 8 presents K-M survival curves for the two cohorts, where age for the earlier cohort refers to years since 1988 (and age for the second cohort is their chronological age). The survival rate is consistently lower at each age for the later entrants and the curve for the later entrants is significantly different from the one for the earlier entrants at the 0.01 level based on the log rank test.

A similar analysis is performed for the producers of CO2, HeNe, ion, dye, and other gas lasers; individual observations for CO2, HeNe, dye, ion, and other gas lasers are pooled and analogous K-M curves are estimated for entrants before and after 1988. Figure 9 indicates that the K-M curve for the later cohort of entrants lies below that for the earlier cohort of entrants at every age, with the two curves significantly different at the 0.001 level based on the log rank test. Thus, consistent with hypothesis 8, after 1988 the hazard rate of producers of both solid-state lasers and CO2, HeNe, ion, dye, and other gas lasers is greater for the post-1988 entrants.

4.5 Regression analyses of the hazard of industry and submarket exit

We complement these analyses with regression analyses for the hazard of industry exit and submarket exit. Following Franco et al. (2009), we use a firm–year observation structure with complementary log–log specification, which allows for recovering continuous-time hazard rates from our annual data. This formulation, Franco et al. (2009) note, allows for easier incorporation of time-varying covariates and controls for unobserved heterogeneity using random-effects specification. We do not use fixed-effects models as our key explanatory variable, order of entry, is time-invariant.²⁴

First, we investigate the hazard of industry exit. The dependent variable is set to 1 if the firm exits the industry at the end of the year. The explanatory variable, Entry Order, is 1 for the earliest cohort of entrants in 1961 and increases for subsequent cohorts of entrants. Post-1988 is a dummy variable set to 1 if the year is greater than 1988 and 0 otherwise. Number of Submarkets measures the number of submarkets a firm services in a given year. The results of the estimation are shown in Table 4. The coefficient estimate of Entry Order in specification (1) is negative but small in magnitude and statistically insignificant. However, the coefficient estimate of the interaction of Entry Order and Post-1988 is positive and statistically significant at the 0.05 level, reflecting the pattern depicted in Figure 8 that later entry is associated with a higher hazard of exit after the advent of DPSS lasers in 1988. The results are robust to the inclusion of number of submarkets in specification (2) and submarket-specific dummies in specification (3). Consistent with our theory, firms present in more submarkets have a lower probability of exit from the industry as reflected by the negative and significant coefficient estimate of Number of Submarkets.

It may be argued that (a) several submarkets exist within the solid-state laser type; (b) that the advent of DPSS lasers led to a growth in economies of scope; and (c) that the lower hazard of industry exit is explained not be economies of scale in R&D but economies of scope. To test this, we interact Number of Submarkets with Post-1988 in column (4) to Table 4. The coefficient estimate of the interaction term is negative but not significant, reflecting that the survival advantage associated with firm scope did not change after 1988 compared with before. Our broader results remain qualitatively similar in column (4).

Our definition of submarkets may also be criticized for its seemingly technological, rather than demand-side, basis. In our theory, the supply-side independence derives merely from a lack of economies of scope in R&D while inherent technological differences across submarkets are not assumed. A lack of economies of scope in R&D may derive from the submarket/use-specificity of R&D. Furthermore, independent submarket may be present within a technological trajectory.²⁵

We argue that an improper specification of submarkets (in the sense that their technology-based definition does not ensure supply-side independence), would only attenuate the patterns our theory predicts. That is, if there is considerable overlap among solid-state lasers and gas lasers in terms of the wavelengths they service even before the advent of DPSS lasers until 1988, then the effect on gas laser producers’ survival would be attenuated because of the advent of DPSS lasers relative to the case where the definition of submarkets is more precise and captures only latent substitutability. Yet, we find robust patterns of displacement of other laser types after 1988.

To further establish the robustness of our results to alternative definitions of submarkets, we investigate the hazard of submarket exit at the most disaggregated level our data permit, composed of 139 different laser types spanning gases, dyes, solid-state crystals, and semiconductors. Our theory predicts that the adoption of DPSS technology per se does not lower the hazard of submarket exit. Only larger firms with DPSS technology, with a larger incentive to invest in R&D escalation process, have a lower hazard of submarket exit. To test this, we set the dependent variable to 1 if the firm exits the submarket at the end of the year. The explanatory variable DPSS is set to 1 for firms producing laser types that use DPSS technology in a given year and 0 otherwise. Number of Submarkets measures the number of submarkets a firm services in a given year at the disaggregate level of laser types. Consequently, the number of firm–year laser type observations present at this disaggregated level is 11,570 compared with 4465 observations present at the broader submarket-level in previous analyses. Industry Clock controls for industry age effects on firm survival. The results of the estimation are shown in Table 5.

The coefficient estimate of DPSS Producer in column (1) is negative and significant, reflecting the lower hazard of submarket exit for firms with the knowledge of DPSS laser technology, consistent with our theory. The coefficient estimate of Number of Submarkets is negative and significant reflecting the lower hazard of submarket exit for firms in more submarkets. In column (2), we interact Number of Submarkets with DPSS Producer. The coefficient estimates of DPSS Producer and Number of Submarkets turn positive and neither is statistically significant. In contrast, the interaction term is negative and significant, reflecting a lower hazard of submarket exit for larger firms with DPSS technology, which is a distinctive prediction of our theory.

5. Discussion

For >30 years, the laser industry was characterized by steady growth in output and the number of producers. Entirely new types of lasers and variants of existing lasers were created that brought new users into the industry. Judging by entry patterns and the rise in the number of producers, these new lasers enabled new firms to enter and compete in the industry. The advent of DPSS lasers around 1988 fundamentally changed these dynamics. A rich trajectory of technological opportunities led to steady improvements in DPSS lasers, fueling a sharp rise in their use both absolutely and as a share of all nondiode lasers produced. Lasers that had always serviced distinct groups of buyers began to experience competition from DPSS lasers, leading producers of these lasers to exit the industry. At the same time, producers of solid-state lasers experienced a shakeout that decreased their ranks. The result was a sharp shakeout of producers from the industry that is still on-going.

Before the advent of DPSS lasers, there was little sign of earlier entrants possessing any competitive advantages. But with the onset of DPSS lasers, earlier entrants fared better than later entrants in terms of survival. The earliest entrants into solid-state lasers were in the vanguard of the surge in patenting and the creation of new solid-state lasers, led by Spectra Physics and Coherent, the long-time leaders of the industry. Our model attributes their leadership to their greater size, which enabled them to earn a greater return from DPSS-related R&D or equivalently to amortize the costs of such R&D over a greater level of output. John Ambroseo, the CEO of Coherent, remarked in 2004 (Kincade and Anderson, 2004) that as the laser industry moved into more mainstream applications, firms needed critical mass to drive engineering programs, which necessitated consolidation.²⁶

Apart from our particular interpretation of events in the laser industry, it seems clear that the shakeout in the industry was related to technological change. A number of theories portray shakeouts as a reflection of developments related to demand and not technological change (e.g., Horvath, Shivardi, and Woywode, 2001). For example, Wang (2008) explains the differential timing of shakeouts across countries based on a demand-side theory. Wang shows that higher consumer income and larger market size lead to faster demand diffusion and earlier shakeout. To explain the common finding that earlier entrants into shakeout industries tend to survive longer (Klepper, 2002), such theories posit that superior firms enter the industry first, when demand conditions are the least favorable, and thus exit last during the shakeout (Braguinsky, Gabdrakhmanov, and Ohyama, 2007; Wang, 2008). However, the fact that the hazard of exit was unrelated to time of entry for the first 30 or so years of the evolution of the laser industry suggests that earlier entrants were not superior firms. It was not until the technological developments brought on by DPSS lasers that advantages of early entry emerged, suggesting these advantages and the shakeout that ensued were fundamentally related to technological change.

The developments in lasers also seem revealing about how technological change contributes to shakeouts. A popular idea is that as industries evolve, demanders experiment with different variants of an industry’s product and eventually coalesce around a particular conception of the product, dubbed a dominant design. This in turn focuses competition around the dominant design, contributing to a shakeout (Utterback and Suárez, 1993). The idea of a dominant design clearly overlaps with that of an integrative submarket. The emergence of an integrative submarket certainly contributes to a convergence toward a single technology, although it need not result exclusively from demanders experimenting and coalescing around a particular design of the product. In the laser industry the emergence of an integrative submarket resulted from scientists figuring out how to harness laser light to stimulate solid-state crystals rather than from any action by demanders. Furthermore, this advance led to a proliferation of new types of solid-state lasers rather than coalescence around any particular solid-state laser design. In addition to providing a broader theory of how an industry coalesces around a particular technology, the integrative submarket theory also provides an explicit mechanism for a shakeout to emerge at both the industry and (integrative) submarket level. This makes it possible to explain a wide range of dynamic empirical patterns characterizing the evolution of the laser industry that cannot be addressed by the dominant design theory of shakeouts.

More broadly, however, an integrative submarket can transpire for several reasons, including technological, regulatory, and demand-side changes. Consequently, the validity of the concept of integrative submarket depends not on a technology-based definition of submarkets but on whether in industries composed of submarkets with weak linkages, a submarket supports an R&D escalation process à la Sutton (1998) or Klepper (1996, 2002). Given that technology need not be the organizing principle for submarkets and that independent submarkets can exist within a technological trajectory, the submarket dynamics we study are broader than technology dynamics (cf. endnote 25).

Another theory of shakeouts developed by Jovanovic and MacDonald (1994) features the role of technological developments in increasing production scale economies and the minimum efficient firm size. Dominant designs are expected to have a similar influence on the scale of production and thus to operate similarly in limiting the number of producers. It is always difficult to assess the nature of the firm average cost curve and the minimum efficient sized firm. But it is noteworthy that diode pumping of solid-state lasers led to a proliferation of different types of DPSS lasers, with the leading firms in the vanguard of this process. The key to their success seems to have been based on innovation and development of a wide range of DPSS lasers rather than the production of any one type of laser at a sufficient volume to exploit potential scale economies.

Recently Tong (2009) developed a model of shakeouts based on independent submarkets, which has a number of features in common with our theory. In his model, shakeouts occur within each submarket at different times (related to when the submarkets are created). The effect of these shakeouts on the total number of producers in the industry is at first offset by the creation of new submarkets, and an overall industry shakeout does not occur until the rate of creation of new submarkets slows down. Figure 4 indicates, however, that among gas and dye lasers, all but HeNe experienced a steady rise in the number of producers into the 1990s, and the decline in the number of producers that ensued in each laser type other than CO2 was not accompanied by a rise in output, which is required for a shakeout (and is implied by his model under general conditions). Rather, solid-state lasers took over an increasing share of the industry’s output as they began to compete with dye and various types of gas lasers, suggesting that the shakeout was part of a process in which submarkets ceased being independent as maintained in his model.

The two eras of the laser industry provide a stark contrast that illuminates how the emergence of an integrative submarket can influence an industry’s dynamics and contribute to a shakeout and change in the character of innovation. These dynamics would appear to be relevant to other industries as well, with different twists that can readily be accommodated within the framework of the proposed model. For example, consider the historical evolution of the hard disk drive industry. At a certain point, new smaller disk drives were developed that appealed to new users and thus defined new submarkets. The leaders were slow to enter these new submarkets, no doubt, in part, because they seemed to have limited size. But unexpectedly, the smaller drives were improved sufficiently that they appealed to users of the older, larger drives (Christensen, 1993) and dominated the industry, contributing to a pronounced shakeout (Christensen, Suárez, and Utterback, 1998). A notable difference from lasers is that new entrants displaced the industry leaders (Christensen, Suárez, and Utterback, 1998), reflecting that the smaller disk drives were pioneered by entrants. This can be readily accommodated in the model by allowing submarket d to be a new submarket in which early on the successful developers of the submarket’s product capture a sufficiently large output for cost-reducing R&D to be profitable.²⁷

Another industry where submarket dynamics also seem to have been important in contributing to a shakeout was the US automobile industry. The industry is generally dated as beginning around 1895, and at first many different types of automobiles of varying size and types of engines (steam, electric, and internal combustion) with appeal to different types of consumers were produced (Wells, 2007). Then Ford Motor Co. introduced the Model T, which was both light weight and powerful, allowing it to appeal to both urban and rural buyers. The result was an integrated submarket of much larger size, which enabled Ford to invest heavily in cost-reducing R&D, leading to the development of the moving assembly line and the Fordist system of mass production. Perhaps not surprisingly, the shakeout in the automobile industry began in 1909, 1 year after the introduction of the Model T (Klepper, 2002), which also marked a pronounced rise in process relative to product innovation (Klepper and Simons, 1997). This case is similar to DPSS lasers in that the Model T was made possible by vanadium steel, a new alloy instrumental in the wide appeal of the Model T that Ford exploited but was developed elsewhere (Wells, 2007). Buenstorf and Klepper (2010) argue that similar dynamics influenced the onset and severity of the shakeout in the US tire industry.

Sometimes submarket dynamics appear to have been pertinent by their absence. For example, both TV receivers and penicillin experienced pronounced shakeouts that started soon after these industries began (Klepper, 2002). Unlike lasers, disk drives, and autos both products do not appear to have been composed of meaningful independent submarkets. Rather, to the extent that there were different variants of each product that appealed to different types of buyers, they were not that different on the production side and were widely produced by most firms. Without true independence on the supply side, producers in different submarkets could not coexist, and it did not take long for shakeouts to start. This could be accommodated in the model by assuming that cost-reducing R&D can be applied to all of a firm’s products, so that a firm’s overall size conditions its incentive to engage in cost-reducing R&D. Under these circumstances, all producers would compete technologically from the outset of an industry, hastening the onset of a shakeout.

When independent submarkets are prominent from the outset, asymmetries in their size and innovative potential will militate toward the eventual emergence of an integrative submarket, paving the way for a shakeout. But there is no technological imperative that rules out innovations that cause an industry to become more fragmented at any point in time (see, also, Windrum, 1998). For example, after World War II, a single type of camera emerged to service both serious hobbyists and snapshooters, but subsequently two new types of cameras were developed to service each segment of the market, in effect fissuring an integrated submarket into two independent submarkets. Not surprisingly, the result was renewed entry and product innovation, reversing the typical life cycle pattern in which initial entry gives way to a shakeout and R&D shifts from product to process innovation (Windrum, 2005).²⁸

We are accustomed to thinking of innovative industries coming from a particular mold regarding the character of innovation and market structure. But the laser industry indicates how quickly an industry can go from one extreme to another. Technological developments can unleash major changes in the nature of an industry’s submarkets, with far reaching implications. Sutton (1998), Bottazzi and Secchi (2006), and Klepper and Thompson (2006) show how a wide range of regularities regarding entry, exit, and firm growth could be explained by submarket dynamics. We can now add to the list shakeouts, first-mover advantages, and the nature of technological change.

Acknowledgements

Klepper gratefully acknowledges support from the Economics Program of the National Science Foundation, Grant No. SES-0111429. We thank two anonymous referees, Ashish Arora, Serguey Braguinsky, David Greenstreet, Peter Thompson, and Roberto Weber for helpful comments.

References