برآورد از مدل های تکاملی به عنوان ابزاری برای پژوهش در سازمان های صنعتی

This paper develops a structural evolutionary microeconomic model where the forces of chance and selection are at work and matches this model to data. As a concrete example we explore the process of industry concentration by modeling bottom-up starting with profit maximizing firms and introducing stochastic elements at various levels of the market. An estimation procedure is developed to connect the model to data of the U.S. household laundry equipment industry. The results for the structural model are then contrasted to a version of Gibrat's model estimated with the same approach. It turns out that the structural model provides a more accurate account of the historical data. This indicates that capturing links between firms operative through the market mechanism promises a more accurate assessment of the future course of concentration of an industry.

The process of economic change is the outcome of both chance as well as of the internal structure (i.e., the mechanisms or processes) of the entity under consideration such as the firm, the industry or the economy as a whole. Nowhere has this become clearer than in the field of industrial economics. Specifically, the analysis of the concentration process of an industry is the subject of a large literature (see Curry and George, 1983, for a survey). Chance as the primary determinant of the concentration process has been explored in a stochastic tradition starting with Gibrat (1931). Sutton (1997) offers a good survey of this research. This first strand of the literature dealing with concentration starts with measures (or assumptions) regarding the distribution of firm growth and investigates the likely outcome of this process over time. Typically, these studies do not explicitly model market mechanisms.1 Mechanisms and processes are the key elements in another strand of research on industry concentration. Analyzing the determinants of market share instability and the role of innovation in the concentration process Nelson and Winter (1977), Silverberg et al. (1988) and Mazzucato, 1998 and Mazzucato, 2000 have modeled interdependencies of firms. These contributions to dynamic economics (typically labeled evolutionary economics) strongly depend on the tool of computer simulation. The inclusion of chance elements threatens to make the interpretation of evolutionary models even more difficult given that simulations already offer insights of less generality when compared to the study of analytic models (as, e.g., noted by Nelson and Winter, 1977). The present study proposes a way of stochastic evolutionary modeling that avoids these problems. We explore market functioning by modeling bottom-up starting with profit maximizing firms and introducing stochastic elements at various levels of the market. Then, instead of an ad hoc specification of the relative magnitudes of these random influences we propose an estimation procedure that allows us to quantify these stochastic influences based on historical data.2 With this procedure stochastic simulation of evolutionary models is made into a tool for the empirical analysis of market phenomena. As an application of this procedure we study the historical development of a particular U.S. industry. Finally, we empirically compare the results of this structural evolutionary analysis to a model in the Gibrat tradition. Stochastically simulated paths based on Gibrat's model have been analyzed previously (see, e.g., Scherer and Ross, 1990, pp. 141–143; McLoughan, 1995). Scherer et al. (2000) extend the Gibrat model to make simulated industry concentration paths conform better to empirical observations. Contrary to these simulations in the literature we apply the same estimation approach to the Gibrat model as that proposed for the structural model. The comparison of models favors the structural evolutionary variant over the Gibrat variant.

We develop a model of the change in industry concentration that integrates the elements of market mechanism and chance. The model is stochastically simulated and connected to real world data by searching the parameterization that minimizes deviations from simulated paths and historical industry data. Thus, we overcome the problem of loss of generality in simulations and turn evolutionary modeling into a tool for the empirical analysis of the industrial concentration process. This offers a new methodological alternative to the list of research methods for the study of market evolution described by Geroski and Mata (2001). For the particular industry studied here (the U.S. household laundry equipment industry) the estimates of the structural evolutionary model fit the empirical concentration data better than an estimated version of Gibrat's model. Both the structural evolutionary model and the Gibrat model build on elements of chance as determinants of industry development. Hence, when it comes to assessing the likely future course of an industry the range of possible trajectories of concentration implied by a model (and its estimated parameters) becomes particularly important. The analysis presented here shows that the structural evolutionary model leads to a narrower range of possible trajectories of concentration measures than is the case with the parameterized version of Gibrat's model. Hence, both industry analysts wishing to assess future profits (possibly rising with concentration) and competition regulators wishing to prevent developments disadvantageous to consumers may gain from such a structural evolutionary analysis. It should be noted that specific applications of the described approach would not be limited to industries with the kind of historical development (a clear tendency toward higher concentration) shown by the household laundry equipment industry. In fact, industries that are at present less concentrated are particularly interesting candidates for this type of forward looking investigation. However, the analysis of such industries is likely to require a modeling of the processes of innovation and market entry. Our exemplary analysis does not include such complications.