Abstract:
Concurrent scheduling of agents presents a challenge for researchers who wish to develop
scalable agent-based models (ABMs) without sacrificing intelligibility or fine control
over model elements. In this thesis, I advance our understanding of the requirements and
challenges of concurrent scheduling by investigating the problem outside of existing ABM
modeling frameworks. I examine the possibility space of agent activation regimes, considering
as axes: parallelization, selection order, updating regime, endogenous or exogenous
access to model state, uniformity of activation, and reproducibility.
This analysis informs a systematic review of ABMs on a popular repository of ABM
source code to determine how researchers are currently addressing agent activation issues.
The review suggests that there is currently widespread homogeneity of modeling practices
regarding agent activation.
I also expand an existing ABM of economic exchange to demonstrate the effects of
varying activation regime on model results and model runtime, extending the analysis to a
parallel computing context. This work also extends previous work on agent activation by
applying the examination on a more complex model. Varying the activation regime
produces significant differences in behavior and model outcomes in this more complex
model.
This research contributes to the existing literature on the implementation of agent-based
models and may be of use for further advances in ABM library development. The results of
the case study may also be of interest to researchers of the foundations of economic theory.