A study of aggregate bi-modal urban travel supply, demand and network behavior using simultaneous equations with autoregressive residuals

This paper emphasizes the existence of a difference among demand functions, which describe how consumers react, supply functions, which analyze the behavior of suppliers, and cost functions, which specify how prices and levels of service on a link or in a network vary with vehicle flows, ridership flows and other factors such as technology, Four aggregate demand, supply and cost models are formulated: each one regroups a subset of demand, supply and cost functions for two modes in Montreal. A significant part of the analysis pertains to the study of a regulated transit supplier. The parameters of all models are estimated by at least two of four limited-information and full-information estimation techniques. All of these procedures are designed to take proper account of equation-specific autocorrelation schemes of the residuals: (i) LSGAU, a least-squares generalized autoregressive estimator; (ii) SSGAU, a two-stage least-squares generalized autoregressive estimator; (iii) ISUGAU, an iterated version of Parks' seemingly unrelated procedure generalized to multiple autocorrelation cases; and (iv) IFIVER, an iterated version of Fair's full-information instrumental variables efficient estimator. A sample comparison of results obtained by IFIVER and other full-information estimators is also provided: the latter are MIFIDA, a modified iterated version of Dhrymes' full-information dynamic autoregressive estimator, and FIML, the maximum likelihood estimator. It is shown that removing simultaneous equation biases has a significant impact on demand, supply and congestion function estimates: notably, it modifies the relative importance of waiting time and in-vehicle time elasticities of demand and the measures of the value of time associated with them.