HawkDove.jl simulates a spatial Hawk-Dove game in which players differ in their attitude toward risk. Agents play against their neighbours, updating their strategy each round and, periodically, adapting their risk attitude toward that of more successful neighbours. The long term emergent behavioir of the agents depends strongly on the amount of information agents have about the past decision of other agents. A small-world rewiring parameter lets the observation network diverge from the spatial grid, which turns out to matter: for some parameter ranges the population converges to being predominantly risk-averse, but that convergence is sensitive to how much rewiring is introduced. The model is a Julia rewrite and substantial extension, including the cooperative Stag hunt game, built on the high-performance computing Vahana.jl framework, of a Python/Mesa model from Princeton’s SimulatingRisk. Running on a strong workstation the system can calculate the long term behavior of one million agents over the weekend.
This is unpublished, ongoing work with no accompanying paper — documented here anyway, on the principle that research software is worth a public record even before (or absent) a formal publication.
Explore the code
- Author: Malte Vogl and Gesine Steudle
- Status: unpublished, work in progress
- Repository: gitlab.gwdg.de/mpigea/dt/hawkdove.jl