Strategic Decisions in Agent-Based Freight Transportation Models: Methods and Data

Freight transportation has significant impacts on energy, emissions, and the economy. High-fidelity, agent-based modeling tools, which accommodate the rich heterogeneity of the business environment, allow government agencies and researchers to study the impacts of freight transportation and its sensitivity to policies. However, no existing agent-based models address business strategy. This is a major gap, since strategy is well known to drive the behavior of firms as they make and align various decisions. With this in mind, my core thesis is that strategy unifies decisions that individual actors make, and that this consistency can be achieved in agent-based models through carefully constructed framework and model design. First, I present an innovative, theoretical foundation for freight transportation forecasting that uses firm strategy to unify agent behavior in upstream and downstream model components. I devise a novel mathematical system that jointly considers firm strategy and numerous strategic decisions using a Seemingly Unrelated Regression (SUR) system, then extend the SUR model by introducing latent variables, which are strategies. I solve the system using Gibbs Sampling. My concept is proven by modeling Logistical Sophistication and Customer Service strategies jointly with eight strategic decisions that involve private fleet ownership and distribution center (DC) control. The strategies are shown to impact the strategic decisions, and the connection of firm strategy to downstream decisions are demonstrated. Furthermore, I develop two innovative methods to generate measurements of strategy using natural text, which is a complete gap in the literature. The Simple Scaled Bag-of-Words (SS-BOW) algorithm creates measurements of firm strategy based on relative frequency of word usage, while the word2vec-Principal Components Analysis (w2vPCA) algorithm creates measurements based on quantifying differences in word use. The concept is proven using large-scale text data, generated by my Attitudinal Data Development Engine (ADDE). A battery of evidence proves the value of the methods. The measurements are also input to the SUR model, which produces intuitive results. Finally, a global sustainability analysis, including a case study of automobile production strategy, demonstrates that modeling agent strategies is extremely relevant to predicting the impacts of business activity on transportation flows, energy use and emissions.