Determinants of Airline Carrier Conduct, 8 International Review of Law & Economics, 187
(1988).
Return to Professor Ayres' Homepage
Return to Professor Ayres' Resume
Since Joe Bain's seminal work in 1959, industrial organization research has
focused on structure, conduct, and performance. Bain thought that, aside from feedbacks
of second order, structural variables determined industry conduct and that, in turn,
structure and conduct determined industry performance. Subsequent scholars seized
upon the structure-conduct relationship to suggest that structural theories could be used
in antitrust analysis. If structural variables (such as concentration) influence the
likelihood of collusive conduct, an understanding of structural characteristics could then
be used to (1) focus investigative resources on markets in which collusion is likely, (2)
detect actual instances of collusion, and (3) reduce the likelihood of collusion by
changing the structure itself.
Empirical tests of the ways in which structural variables influence collusion,
however, have only been indirect. Dozens of econometric articles have examined the
structure-performance relationship by regressing structural variables (such as
concentration) on performance variables (such as profits). But these "structure on
profits" regressions have only indirectly tested the relationship between structure and
collusion - by assuming, often implicitly, that abnormal profits stem from collusion.
Demsetz noted that often such indirect tests of conduct are unidentified, because firm-
specific efficiency as well as collusion could induce a positive correlation between
profits and concentration.
This article represents a first attempt to overcome the problems inherent in
inferring conduct from performance by regressing estimates of conduct itself on the
structural variables that theory suggests induce collusive behavior. Directly examining
the relationship between conduct and structure can test not only traditional structural
theories - such as Stigler's hypothesis that the number of sellers influences the degree
of collusive behavior and Posner's hypothesis that collusion is likely to take place in
markets in which the gains from collusion are the greatest - but also newer structural
theories that have been explicitly advanced in the airline context, such as the
hypothesis that airline routes are "contestable markets" that will be competitive
regardless of concentration.
The tests in this article have important policy implications in detecting and
deterring collusion. The analysis not only indicates where collusion has taken or will
take place, but it also quantifies the extent to which structure influences behavior. While
many theories predict that more competitors will induce more competition, theory has
little to say about the size of the competitive gain. Thus, empirical analysis not only can
confirm existing theories, but also can provide policymakers with information about the
possible benefits of changing structure.
Calculating quantitative estimates of behavior is the crucial starting point.
Following Iwata, marginal cost and elasticity estimates were combined with price and
market share data to estimate conjectural variations in the airline industry both before
and after deregulation. As derived below the conjectural variation measures how
competitively a firm reacts to changes in the output of its rivals: Firms that reduce their
output in response to output reductions of their rivals are acting collusively; firms that
increase their output to offset output reductions of their rivals are acting competitively.
The conjectural variation approach to modeling airline carrier conduct has both
important strengths and weaknesses as an analytic tool. As elaborated below,
collapsing an airline's behavior into a scalar strategy variable places restrictive
assumptions on the model. Studying the airline industry, however, allowed the
estimation of over two thousand conjectural variations (by carrier and route) using
appealing measure of market conduct because it not only represents the firm's
expectations of other firms' conduct, but also, be feeding back into its own reaction
function, determines the firm's own conduct (its non-cooperative strategy).
Section II of the paper describes the calculation of the conjectural variations and
tests whether carriers displayed competitive, collusive, or Cournot behavior. Section III
forms the central part of the paper: There I describe the structure/conduct regressions
and test whether specific structural variables influence conduct. In section IV slopes of
the firms' reaction curves are estimated. Tests for the equality of the conjectured and
actual reaction curve slopes are made. Such tests are shown to test not only for the
presence of Bresnahan consistency but also for a generalized form of Stackelberg
leadership.
Return to top of the Page