Using panel data for five capital goods industries, this paper estimat
es dynamic externalities. In contrast to previous studies, panel data
allow separation of externalities from fixed effects and identificatio
n of a lag structure. I find strong evidence of Marshall-Arrow-Romer (
MAR) (own industry, or localization) externalities. For Jacobs (urbani
zation) externalities effects are smaller. In terms of lag structure,
for MAR externalities the biggest effects are typically from several y
ears ago, but die out after six years. For urbanization phenomena, eff
ects persist to the end of the time horizon of the data-eight or nine
years back. (C) 1997 Academic Press.