Stylized Facts of Industrial Dynamics: The Distribution of Firms’ Size
Economists interested in business cycle and growth theory have long been trained to the use of stylized facts as a practical guide in implementing their research agenda, as the pioneering accounts of Burns and Mitchell (1946) and Kaldor (1963) testify. The advent of the neo-classical counter-revolution in the late 1960s, rooted in what Robert Solow dubbed the holy trinity of Rationality, Equilibrium and Greed, has somehow inverted the logic of scientific discovery in economics. The first step in nowadays orthodox macroeconomics consists in building a model of microeconomic behavior based on axiomatic descriptions of preferences and technology. Afterwards, the model is solved via the representative agent and taken to the data. Alas, as shown inter alia in Caballero (1992) and Kirman (1992), the falsifiability of the model may be fatally prevented due to a fallacy of composition, that is the presumption that what is true of each single part of a whole is necessarily true of the whole as well. In particular, the straight application of a microeconomic rationale to aggregate data can be seriously misleading whenever the probabilistic forces at work as the number of entities grow large, i.e. the Central Limit Theorem, the Law of Large Number or any of their extensions, are not properly taken into account.
KeywordsBusiness Cycle Total Asset Stylize Fact Pareto Distribution Total Sale
Unable to display preview. Download preview PDF.
- 3.See, for example, Takayasu and Okuyama (1998), Voit (2000) and Knudsen (2001).Google Scholar
- 6.Prom a technical viewpoint, the HP is a low-pass filter. Hence, the cyclical component is obtained by subtracting from the raw series the filtered one. The smoothing parameter A has been tuned at the value 100 for annual data. See Hodrick and Prescott (1997).Google Scholar
- 9.According to the business cycle chronology calculated by the Economic Cycle Research Institute, Italy experienced a peak in February 1992 and a trough in October 1993. Gallegati and Stanca (1998), using annual data, calculate turning points to be in 1990 (peak) and 1993 (trough). It is generally accepted that the following expansion has lengthen at least until the first quarter of 2001.Google Scholar
- 10.The reason lies in the fact that in Axtell’s calculation, the minimum size s0 has been assumed fixed and equal to 1. From the argument reported in Blank and Solomon (2000), who discuss a paper by Gabaix (1999) who makes use of the same assumption, it emerges that the formula (4) in Axtell (2001) implicitly returns the Pareto exponent α if and only if the minimum size is assumed to be a constant fraction c of the current average of firms’ size 〈s〉, so that one should posit s0 = c〈s〉(t), which clearly varies in time.Google Scholar
- 11.Mizuno et al. (2002) prove this last result only by means of visual inspection: no calculations of the scaling exponents are explicitly reported.Google Scholar