Now we turn to a quite different application of our approach and models: the rate at which firms leave the semiconductor industry. Opportunities in this industry, emanating from changing technology and markets, are seized by venture capitalists and entrepreneurs in ways that have attracted much publicity and created a few large personal fortunes. The popular press often depicts the scientist-entrepreneur as a romantic figure, a modern explorer taking great risks for high stakes. There does appear to have been great risk. But in the semiconductor business a failed venture appears to be less a black mark than a sign of experience. Venture capitalists seem willing to provide funding for a new venture on the heels of a recent failure. T. J. Rogers, for example, persuaded his employer, American Microsystems, Inc., to spend (and lose) $25 million on a technology called VMOS. After changing jobs, Rogers succeeded in acquiring venture capital to start Cypress Semiconductor only a few years later, in 1983. Three years after founding, the company’s stock was valued at
$300 million (Fortune 1987). The same instabilities in the market and the technology that generate opportunity impose these risks on entrepreneurs. Moreover, as we indicated in the previous chapter, there is a substantial liability of newness in the semiconductor business.
From an ecological point of view, the issue is how the probabilities of failure are influenced by the dynamics of the population of firms and the conditions under which they operate. In particular, competition within and between populations has obviously affected the life chances of firms in this industry. One of the major goals of this chapter is to show that such competitive processes can be understood and analyzed empirically with the same approach that we used to study such nonmarket organizations as labor unions. So the question is: Does market competition, for which the semiconductor industry is justly renowned, yield similar processes of orga- nizational competition to those we just reported for unions, which do not operate in such a clear market context?
We recorded histories of participation of 1,197 firms in the u.s. semi- conductor industry between 1946 and 1984. Of these, 302 firms were still operating in 1984, which means that we treat them as censored on the right. The remaining 895 firms exited (either went bankrupt, were absorbed by other firms, or left the industry). Yearly counts of exits rise and fall over time, as Figure 11.3 shows. If anything, the variability in exit rates over time is stronger than for entries. In the peak year, 1972, there were 110 exits, fully one-eighth of the total. Comparing fluctuations in exits in Figure 11.3 with variations in density in Figure 9.6 suggests that the effect of density on the exit rate was non- monotonic. The exit rate peaks at the same time that density peaks. We examine later in this section whether this pattern holds under controls for environmental conditions.
We analyze the rate of exiting using models of the form in equation (11. D- Since the exact dates of semiconductor firm exits are not available, we do not try to use parametric models of time dependence. As in the previous chapter, we use partial likelihood estimators which treat the effects of time in the industry nonparametrically. As in the analysis of unions, we broke each spell between the firm’s first and last appearance into yearly spells.71 Spells not ending in an exiting event are censored on the right. This procedure generated 6,856 spells for analysis. Values of several time-varying covariates were associated with each spell, updated at the beginning of each year. These include time since entry, counts of firms (density), or counts of life events such as number of exits. Variables describing economic conditions were measured in the year covered by the spell.
Figure 11.3 Semiconductor firm exits by year
For example, a firm might first have appeared in the Electronics Buyer’s Guide in 1958. Suppose its last year of appearance was 1961. Its starting and ending times would be something like 1957.53 and 1960.48. The first spell would be 1957.53 to 1958; the second, 1958 to 1959; the third, 1959 to 1960. Each of these spells would be right-censored. The last spell would be 1960 to 1960.48 and would not be censored. A series of independent variables would be attached to each spell, as described in the previous paragraph. For the first spell, the number of firms observed in the year 1957 would be entered (53), and the total sales for 1958 would be added ($571 million). Time since entry for this spell is zero, because at the start of the year, the firm had not yet entered the population.
Table 11.4 presents the results of our analysis, proceeding from simple to complex as before, with models to the right adding regressors so that likelihood ratio tests can be performed. We begin by asking whether density has the predicted non-monotonic effect on the exit rate of semiconductor firms. As before, we expect that the coefficients representing the effects of density and its square in equation (11.1) will satisfy the inequalities β1 < 0 and β2 < 0. In each column of Table 11.4, the first-order and second- order effects of density are statistically significant, and their signs are in the predicted direction.
The next step is to see whether the estimated effect of density has the predicted non-monotonic form over the range of variation in our data. Figure 11.4 plots the effects of density on the rate of exiting. As expected, the multiplier drops as density increases at the low end of its range, but then rises. At its minimum (when N = 191), the exit rate is only 3 percent as large as a zero density. As density increases above 191 to its maximum value of 335, the multiplier rises to 0.22. Although the exit rate atN = 335 is only about a quarter as large as at zero density, the multiplier has still risen seven-fold from the minimum at N = 191.
We conclude that there is strong evidence of density dependence in exit rates in the semiconductor industry. The predicted non-monotonic pattern of this dependence is stronger in the best-fitting model than it is in the simpler specifications. This pattern differs from the pattern for labor unions in Figure 11.2 in that the semiconductor exit rates drop much more sharply at low levels of density. This is surprising because one would expect that legitimacy problems would be much more salient for labor unions than they are for semiconductor companies. On the other hand, when semiconductor firms were rare, the technology was new. For this population, low levels of density occur early in historical time. Our controls for periods may not be fine enough to remove the historical time dependency from the analysis. If this is true, the steep decline in the rate of exiting reflects the vulnerability of organizations without well-established routines for dealing with a radically new technology. Such a perspective fits the arguments of Tushman and Anderson (1987) that fundamental technical innovations often destroy organizational competence.
Figure 11.4 Effect of density on exit rate of semiconductor firms (estimates from model 4 in Table 11.4)
The second research question is whether experience in the industry also affected the failure rates of these firms. We do not have data on the ages of these organizations. Rather, we observe the time since entry into the semi- conductor industry. Following the arguments and results of Chapter 10, we expect that the time since entry of the firm, “experience,” will have a negative effect on the rate of exiting. We estimated effects for the logarithm of time since entry and found strong evidence supporting our expectation of experience dependence in rates of exiting from the semiconductor industry. Each model in Table 11.4 includes the log of time since entry. All estimates of this effect are negative, as we would expect if a liability of newness is operating. These coefficients are always many times their standard errors. Apparently the risks of failure that such new high-technology ventures suffer drop off quite quickly as the firms develop experience.
The third question concerns the effects of environmental conditions. We expect industry sales to have a negative effect on the rate of exiting but the interest rate of corporate bonds to have a positive effect on the rate, indicating that when the cost of capital is high, the highly capitalized semiconductor firms suffer. Both of these predictions are supported by the analyses reported in Table 11.4.
Finally, we expected that growth in applications for semiconductor tech- nology would lower the exiting rate. Technical innovations generate op- portunities for firms manifesting various subforms, and these should make survival easier than when the technology only provides a few viable niches. We try to control for these changes by including period effects in the model. A glance at Figure 11.3 would suggest that the 1960-1969 period would show higher exit rates than the 1946-1969 period, and that the 1970-1983 period would be higher still. In fact, the antilog of the coefficient for the second period in column 4 of Table 11.4 is 2.78, which means that the exit rate in the second period is 178 percent higher than that in the prior period. In the third period, the rate rises by a further 55 percent. So proliferation of applications does not appear to have improved life chances in this industry. As the industry has developed, the exit rate has risen sharply.
Column 2 in Table 11.4 adds effects of rate dependence. We want to see if the pattern of change in exit rates has the wavelike pattern previously observed for the founding and entry rates of unions and semiconductor firms but not for the disbanding rates of labor unions. We expect that number of exits in the previous year would show a non-monotonic pattern like that found for density dependence. The function should turn down at high prior rates of exiting as the negative signaling begins to wear off. Another way of looking at this is to note that if the prior rate has a positive monotone effect on current rate of exit, the rate would continue to acceler-ate. Unless the entry rate accelerated at the same time or the effect were offset by environmental changes, the population would vanish.
Figure 11.5 Effect of prior exits on exit rate of semiconductor firms (estimates from model 4 in Table 11.4)
The coefficients in column 2 of Table 11.4 have the signs that we ex- pected. The estimated effects differ significantly from zero, and the model improves significantly over the model that excludes the effect of prior exits.73 Figure 11.5 shows how the multiplier of the rate varies with the number of exits in the previous year. It shows that the effect of recent exits has been a strong one. At its peak, when roughly 50 firms had exited in the previous year, the multiplier is about 2, which means that a firm’s rate of exit in such a year is double the rate that would hold if no firms had exited in the previous year. The multiplier diminishes as the number of prior exits ranges higher and eventually falls below unity.
An additional issue we explored with these data is the relationship between waves of prior entries and the exit rate. One way to view this is that when entry rates are high, a larger number of firms enter the industry with fatal flaws in their design, in business strategy, or in managerial expertise. If such a defect results in a higher exit rate, it should produce two results in our model: it should increase apparent experience dependence, and it should also produce an effect of prior entries with the quadratic function we observe. However, the two are difficult to untangle with our data because we have imprecise information on timing of entry and exit. A more elaborate specification of experience dependence might, in fact, wipe out the effect of entries. Consequently, these effects should be interpreted with caution.
Column 3 of Table 11.4 shows estimates of the model with the addition of number of entries and its square. The increase in the log-likelihood of 8.8 is significant at the .05 level. Waves of exits do seem to follow waves of foundings.
A fourth question is whether exit rates differ by form. The last column in Table 11.4 adds a binary variable that indicates whether the firm is a subsidiary of a larger corporation or whether it is an independent firm. We expected to find that the exit rate of subsidiary firms would be lower because these firms are buffered from many of the contingencies that vary so dramatically over the short run in the semiconductor industry. Furthermore, the superior access to resources afforded by the parent firm makes exiting due to scarce resources less likely. On the other hand, subsidiary firms tend to be much smaller organizations, as we indicated in Chapter 7. Adding the distinction between subsidiaries and independents to the model improves the fit significantly. The coefficient for SUBSIDIARY is negative and significant, as expected. The point estimate implies that the exit rate of subsidiary firms was only about half as large as that of independent firms.
The last set of issues concerns interactions between subpopulations of subsidiary and independent firms. We have already shown that the exit rates of subsidiary firms are lower than those of independent firms. But is this protected status achieved at the expense of the independent firms? When we explored this question in the context of entry rates, we found that the dynamics of the two populations were quite different. In fact, most of the variables that had significant effects on entry rates of the independent firms did not have significant effects for the subsidiary firms. This same situation obtains for exits, albeit less strongly.
Table 11.5 reports the results on competition between subpopulations. It gives two sets of results, one for each subpopulation. The models are based on the one reported in column 3 of Table 11.4. Table 11.5 excludes the effect of SUBSIDIARY, of course; and it adds a cross-effect of the other subpopulation’s density.
This model predicts the exit rate of independent firms much better than that of subsidiary firms. All coefficients in the independent-firm column are significant, except for the effect of the second period. In the subsidiary- firm model, most coefficients are not significant. There is no apparent density dependence in this subpopulation, nor is there any rate dependence. We do see evidence that effects of time in the industry and of the measures of business conditions are similar in the two subpopulations. We tried various simplified versions of the model on this subpopulation, with no better success.
We are most interested in examining the competitive relationship between these two subpopulations. When we analyzed unions we found an asymmetric pattern: the number of craft unions had a positive effect on the disbanding rate of industrial unions, but not the reverse. Table 11.5 shows a very similar pattern. There is a powerful effect of the number of subsidiary firms on the exit rate of independent firms, but not the reverse. However, unlike the results for labor unions, the cross-effect is log-quadratic. Both the first-order and second- order cross-effects differ significantly from zero.
Figure 11.6 plots the implied cross-effect of the density of subsidiary firms on the exit rate of independent firms. Over most of the range of density of subsidiary firms, the cross-effect is positive. That is, a high density of subsidiary firms increased the exit rate of independent firms. At its peak, when Ns = 66, the multiplier is 38. This means that the implied exit rate of independent firms is almost forty times higher the rate that would hold if there were no subsidiary firms in the industry. Eventually, when the subsidiary firm population approaches its maximum, the function turns down. But even at the maximum observed level of Ns, the multiplier is still about 20.
Given that the density of subsidiary firms has such a powerful deleterious effect on the life chances of independent firms, why does the effect diminish at high density? Perhaps competitive pressures have their greatest effect when there are many marginal firms in the population, but the vulnerable firms exit as density of the competing form rises. Perhaps rising competition from subsidiary firms leads the independent firms to develop competitive responses. Such companies often develop alliances (joint ven-tures, second sourcing arrangements, and exchanges of technical information) with members of the competing population to gain some of the advantages of the subsidiary form without actually adopting it. Finally, they may develop alliances with firms in the captive side of the business, or with important customers, which provide the buffering and factor market presence whose absence threatens their viability. “Strategic alliances are a partial substitute for deep pockets and vertical integration. They entail exchanges of precious masks—the photographic negatives of a chip’s layout—and even more precious secrets of design and production technology. For example, in 1984 Advanced Micro Devices agreed to give LSI Logic Corp. some of its own chip designs in exchange for access to LSI Logic’s software for designing chips” (Fortune 1986).
Figure 11.6 Effect of subsidiary firm density on exit rate of independent firms (estimates from Table 11.5)
A notable example of such strategic moves among independent firms is the alliance between IBM and Intel. Intel is a major supplier of microcomputer microprocessor chips to IBM. When Intel was suffering from cash shortages during the 1983 recession in semiconductors, IBM invested $250 million in Intel with an explicit agreement not to acquire a controlling interest, and Intel weathered the storm without becoming one of IBM’s divisions. The change in form that absorption would imply was widely described as a likely death knell for Intel, and perhaps for the independent form itself ( S a n Jose Mercury 1984).
Whatever the source of this drop in the multiplier due to cross-effects of the density of subsidiary firms, we should not make too much of it. In fact, the decline in the cross-effect at high density of subsidiary firms is small in substantive terms since the exit rate remains roughly twenty times higher than the rate at low densities. It is interesting to contrast this very powerful interpopulation competitive effect with the modest difference in the life chances of individual members of the two subpopulations noted earlier. Recall that the exit rate of subsidiary firms is about half as large as that for independent firms. But here we see that the exit rate of independent firms rises almost forty-fold when the density of subsidiary firms grows large.
Source: Hannan Michael T., Freeman John (1993), Organizational Ecology, Harvard University Press; Reprint edition.