The Relationship Between Firm Size and Firm Growth in the U.S. Manufacturing Sector

This paper investigates the dynamics of firm growth in the U. S.
manufacturing sector in the recent past. I use panel data on the
publicly traded firms in the U. S. manufacturing sector: from a
universe of approximately 1800 firms in 1976, I am able to follow most
of them for at least three years, and over half of them from 1972 until
1983. I consider several problems, both econometric and substantive,
which exist in analyzing this kind of data: the choice of size measure,
the role of measurement error, and the effect of selection (attrition)
on estimates obtained from this sample.
Using time series methods, suitably modified for panel data (where
the number of time periods per observational unit is small), I analyze
the behavior of employment over time and find that most of the change in
employment in any given year is permanent in the sense that there is no
tendency to return to the previous level. Year-to-year growth rates are
largely uncorrelated and there is almost no role for measurement error.
I find that Gibrat's Law is weakly rejected for the smaller firms in my
sample and accepted for the larger firms; Other measures of size
produce essentially the same results.
Correction for attrition from the sample changes the results
somewhat: I use a simple model in which firms leave the sample because
they are small and/or undervalued (since many exits are acquisitions)
and find that Tobin's Q, the raio of market valuation to the value of
the underlying assets of the firm, is a much better predictor of exit
probability than size alone (firms with low Q are more likely to exit
the sample). When I use this estimate of the probability of exit to
control for selection bias, Gibrat's Law is weakly rejected for firms of
all sizes and there are significant positive effects on firm growth from
both investment in physical capital and R&D expenditures, with R&D
having a somewhat higher net effect.