Dynamic modelling of household automobile transactions within a microsimulation framework

The primary objective of this dissertation was to develop a dynamic model of household automobile transactions within an integrated land-use, transportation, and environment (ILUTE) modelling system framework. A market-based decision making process and a transaction approach were applied for this project due to their consistency with the actual processes followed by a decision maker in a real world. The proposed framework has a nested tree with transaction choices (adding new vehicle to fleet, disposing one vehicle, trading one of the vehicles in fleet, or do-nothing decision) in the upper level and class and vintage choices in the lower levels. Different approaches to operationalize such a model were employed including hedonic price, regression, multinomial logit, nested logit, artificial neural networks, and random parameter logit models.; The vehicle class and vintage choices, given that a transaction has occurred, are modelled using a well-developed form of discrete choice modelling technique, the nested logit model. The same decision process has been also modelled in a multi-layer perceptron artificial neural network in order to investigate the applicability of this technique to the household vehicle choice problem. The comparison of the results of both techniques confirmed the robustness of machine learning algorithms and showed that these types of models can generate better prediction results than the traditional discrete choice methods.; Understanding and predicting the complex activity of individuals' behaviour to become active in the market was another aspect of this investigation in which the dynamics of transaction behaviour were modelled. A limited-information nested logit model estimation approach was used to incorporate the vehicle type choice model into the main dynamic transaction choice model.; An exponentially smoothed weighted average of past transactions choices was introduced to account for state-dependence and a random parameter logit model was employed to account for heterogeneity across decision makers. The result of random parameter logit model estimation indicated that heterogeneity is not a significant factor in the dynamic model developed here.