Empirical analysis of the dependence structure in traffic data using copula function

Abstract

Statistical distributions of link and path travel times are important inputs to reliability-sensitive transportation models. This paper proposes to use copula functions for specifying and calibrating a tractable dependence structure for link travel times. Our empirical study indicates that travel time data tend to have a quasi-tail dependence structure, which is inconsistent with those embedded in commonly used copulas. This means that comparing with traditional fields using copula function, such as financial, traffic data demonstrate special features. A new copula function needs to be proposed to model this newly discovered feature.