A Spatial Agent-Based Covid-19 Model
Using an agent-based Covid-19 model, we simulated the spread of the virus in Cape Town.
How effective are ‘lockdown’ measures and other policy interventions to curb the spread of Covid-19 in emerging market cities that are characterized by large heterogeneity and high levels of informality? The most commonly used models to predict the spread of Covid-19 are SEIR models which lack the spatial resolution necessary to answer this question. We develop an agent-based model of social interactions in which the distribution of agents across wards, as well as their travel and interactions are calibrated to real data for Cape Town, South Africa. We characterize the elasticity of various policy interventions including increased likelihood to self-isolate, travel restrictions, assembly bans, and behavioural interventions like washing hands or wearing masks. Even in an informal setting, where agents’ ability to self-isolate is compromised, a lockdown remains an effective intervention. In our model, the lockdown enacted in South Africa reduced expected fatalities in Cape Town by 26% and the expected demand for intensive care beds by 46%. However, our best calibration predicts a substantially higher case load, demand for ICU beds, and expected number of deaths than the current best estimate published for Cape Town.
This working paper was