Mean speed prediction with endogenous volume and spatial autocorrelation: A Swiss case study
Axhausen, Kay W.
- Working Paper
Rights / licenseIn Copyright - Non-Commercial Use Permitted
In the present paper a modeling approach to address the issue of mean speed prediction on a large scale network is presented. Methodologically, we exploit the family of spatial regression models to treat both for the spatial dependence and the endogeneity aspect between speed and volume. The estimation of the model takes place by means of a 2-step instrumental variables- generalized method of moments estimator, allowing us to obtain consistent and unbiased parameter estimates. An empirical case study is designed and conducted in order to model mean speed values on a national planning network in order to check the applicability and the predictive performance of the proposed modeling approach. A particular focus is given on the instrumentation of demand on truly exogenous and capable of capturing the interregional demand patterns variables. Moreover, different spatial weight matrices are tested thoroughly to conclude on a matrix identification based on free-flow travel time. Our findings suggest that the proposed model has the ability to provide accurate estimates, outperforming a much more complex and data demanding transport planning model, even though the superiority of such models is taken for granted in many cases. Last, the developed modeling approach coupled with the implied volume regression model can form a coherent direct demand modeling approach, suitable both for prediction and forecasting applications Show more
External linksFull text via SFX
Journal / seriesArbeitsberichte Verkehrs- und Raumplanung
PublisherIVT ETH Zurich
SubjectAverage speed; Spatial regression; Instrumental variables; Endogeneity; AADT
Organisational unit03521 - Axhausen, Kay W.
02226 - NSL - Netzwerk Stadt und Landschaft / NSL - Network City and Landscape
MoreShow all metadata