Assessing Input Uncertainty in Commodity-Based Freight Demand Models

Abstract

Freight demand models are a set of tools utilized for the forecasting, planning, analysis, and/or optimization of the movement of commodities, such as the billions-of-dollars-worth of goods and services that are moved annually in Canada, and they contain uncertainty. There are two types of uncertainty that affect freight demand models: input and model uncertainty. Input uncertainty is concerned with the fact that there is error in the data used as inputs to model transportation demand such as biased surveys, incomplete datasets, varying commodity and industry classifications, etc. Model uncertainty is concerned with the fact that the model specification and calibration/estimation may contain error such as omitted variables, inappropriate assumptions, simplifications, etc.

There is a lack of understanding surrounding the uncertainty of freight demand models. Regardless, these models are widely researched, developed, and applied without characterizing the uncertainty of typical data sources used as inputs. The contributions of the variation present in different inputs to the model results are unknown, making it impossible to know the robustness of the model outputs or how the results might be improved. The literature review revealed that the most common freight model classification system is based on the unit of demand generation, the most used freight demand models in the North American practice are commodity-based, and input uncertainty has a greater effect on transportation demand models. Thus, this thesis proposes a formal five-step framework (i.e., uncertainty source identification, distribution of source identification, simulation, estimation of output distributions, and analysis of results) to analyze the effects and propagation of input uncertainty on the uncertainty of the outputs in commodity-based freight demand models.

The framework is applied to an Aggregate-Disaggregate-Aggregate version of a strictly empirical commodity-based freight demand model used to analyze the effects the Comprehensive and Progressive Trans-Pacific Partnership on Canada’s trade infrastructure. Essentially, uncertainty for three inputs is introduced and a set of outputs is simulated through repeated simulation. The three inputs are high level supply chain characteristics, value-weight ratios, and domestic mode shares – each being an input to one sub-model of the freight demand model. Dispersion, confidence intervals, and performance against the outputs of an illustrative base case are explored. In general, the case study model generates consistent results to the base case when looking at the conclusions of aggregated outputs, despite the tendency to high variance of the disaggregated outputs and the poor results of the confidence interval analyses. Implementation of the framework generated insight on the accuracy of the case study model, and it highlighted the specific instances where the modeler needs to be more cautious of the results when using only point data, as in the illustrative base case.