Publication Detail
Long Term Forecasting with Dynamic Microsimulation
|
UCD-ITS-RR-91-16 Research Report Alumni Theses and Dissertations |
Suggested Citation:
Goulias, Konstadinos G. (1991) Long Term Forecasting with Dynamic Microsimulation. Institute of Transportation Studies, University of California, Davis, Research Report UCD-ITS-RR-91-16
The use of cross-sectional models in travel demand forecasting involves some fundamental shortcomings. First, it is based on the pressumption that cross-sectionally observed variations in travel behavior can be used as valid indicators of behavioral changes over time. Second, the input variables — household socioeconomics and demographics — needed by the disaggregate travel demand models are provided at an aggregate level externally. Approximate disaggregate inputs are then obtained using inaccurate allocation and postprocessing techniques. The result is a questionable basis on which travel demand forecasts are made.
An alternative travel demand forecasting system is described in this thesis. The system consists of two components: a microsimulator of household socioeconomics and demographics, and a dynamic model system of household car ownership and mobility. Each component comprises interlinked models at the household level. Replicated in the socioeconomic and demographic microsimulator, are interactions and causal paths that underlie lifecycle evolution of individuals and households. Simulation units evolve from year to year, experiencing marriages, divorces, births, deaths, and so forth. Employment, income, driver's license holding, education level, and household size and composition, are among the variables that are internally generated in the simulation. Most model parameters can be modified to represent different future growth paths.
Output from the microsimulator is then used to predict household car ownership and mobility. Thus many explanatory variables, exogenous in other forecasting models, are endogenous in this model system. The mobility component is made up of a household car ownership model, trip generation models, a modal split model, and travel distance models. Most models contain lagged dependent variables and serially correlated errors. Future travel demand is simulated for each simulation year.
The parameters of the model system have been estimated using observations from 5 waves of the Dutch National Mobility Panel data set, covering a period of four years between April 1984 through April 1988. Other sources of information were also used to estimate key parameters. The model structure, data requirements, estimation methods, and assumptions are presented in this thesis. Examples of forecasting, illustrating the system's predictive capability and limitations, are performed for the year 2010.
An alternative travel demand forecasting system is described in this thesis. The system consists of two components: a microsimulator of household socioeconomics and demographics, and a dynamic model system of household car ownership and mobility. Each component comprises interlinked models at the household level. Replicated in the socioeconomic and demographic microsimulator, are interactions and causal paths that underlie lifecycle evolution of individuals and households. Simulation units evolve from year to year, experiencing marriages, divorces, births, deaths, and so forth. Employment, income, driver's license holding, education level, and household size and composition, are among the variables that are internally generated in the simulation. Most model parameters can be modified to represent different future growth paths.
Output from the microsimulator is then used to predict household car ownership and mobility. Thus many explanatory variables, exogenous in other forecasting models, are endogenous in this model system. The mobility component is made up of a household car ownership model, trip generation models, a modal split model, and travel distance models. Most models contain lagged dependent variables and serially correlated errors. Future travel demand is simulated for each simulation year.
The parameters of the model system have been estimated using observations from 5 waves of the Dutch National Mobility Panel data set, covering a period of four years between April 1984 through April 1988. Other sources of information were also used to estimate key parameters. The model structure, data requirements, estimation methods, and assumptions are presented in this thesis. Examples of forecasting, illustrating the system's predictive capability and limitations, are performed for the year 2010.
Ph.D. Dissertation