Review of Industrial Organization 16: 313–318, 2000.
© 2000 Kluwer Academic Publishers. Printed in the Netherlands.
Correlation Tests for Competitive and Cournot
Department of Economics and Finance, University of Wyoming, P.O. Box 3985, Laramie, WY
Abstract. This note derives simple correlation tests for either competitive or Cournot conduct.
Robustness of the procedure is theoretically explored and a time-series correction for inﬂation or
market growth is derived.
Key words: Oligopoly, conduct, correlation, conjectural variation.
Much progress has been made in recent years toward measuring competition di-
rectly through behavioral characteristics. One approach uses conjectural variations,
or the change in rivals’ output anticipated by a ﬁrm in response to a change in its
own output, typically estimated from the ﬁrst-order conditions for proﬁt maximiza-
tion (Iwata, 1974; Bresnahan, 1982; Tirole, 1988, p. 245). Ashenfelter and Sullivan
(1987) propose a nonparametric measure related to the conjectural variation. Their
approach is formulated in terms of monopoly behavior and generalizes to the num-
bers equivalent of a symmetric n-ﬁrm Cournot oligopoly. It is one-sided in the
sense that competition cannot be directly inferred; the procedure can only reject the
particular oligopoly or monopoly hypothesis in favor of more competitive conduct
This note derives a different test of conduct based on empirical correlations,
either parametric or nonparametric. Our test is related to the conjectural variation,
but is one-sided in the opposite direction to that of Ashenfelter and Sullivan. It al-
lows the competitive hypothesis to be rejected in favor of less competitive conduct,
and therefore ﬁlls a gap in the empirical methodology of conjectural variations.
A variant of the test applies to Cournot oligopoly. A time-series correction for
inﬂation or market growth is also derived.
The usefulness of this test stems from several properties. First, it requires less
data than many other “new empirical IO” tests, using only output price, own quan-
tity, market quantity, and marginal cost. These ﬁgures are commonly available
except marginal cost, which can be omitted if constant. Second, the nonparametric
version of the test is more robust with respect to distributional assumptions than