Building an Agent-Based Model to Explain Gentrification in European Cities
This thesis develops a computational Agent-Based Model (ABM) to simulate and explain the socio-spatial dynamics of gentrification across European cities. By modeling individual household decisions, landlord strategies, and neighborhood evolution, the model reveals emergent patterns that match observed gentrification trajectories.
Institution: Bocconi University, M.Sc. Data Science & Business Analytics (2023-2025)
Research Question
Can an Agent-Based Model, grounded in heterogeneous agent behaviors and spatial interactions, replicate and explain observed gentrification patterns in European cities? And what are the key mechanisms driving neighborhood-level socioeconomic change?
Methodology
Built using the Mesa framework in Python, the model simulates thousands of heterogeneous agents (households, landlords, businesses) interacting on a spatial grid derived from real urban data. Calibrated against empirical datasets from multiple European cities.
Model Architecture
Household Agents
Heterogeneous agents with income levels, housing preferences, and mobility decisions. They evaluate neighborhoods based on amenities, affordability, and social composition.
Landlord Agents
Property owners adjusting rents based on local demand, neighborhood desirability, and investment strategies. Their decisions directly drive displacement dynamics.
Spatial Environment
Grid-based city representation calibrated from real GIS data with neighborhood characteristics, transit accessibility, and proximity to amenities.
Key Mechanisms
Rent Gap Theory
The model operationalizes Neil Smith's rent gap theory, tracking the difference between current land value and potential value under 'higher use', triggering investment when the gap widens.
Neighborhood Tipping Points
Emergent tipping behavior where a critical mass of higher-income residents triggers accelerating change in neighborhood composition, amenity attraction, and rent increases.
Displacement Cascades
Lower-income households displaced from gentrifying areas relocate to adjacent neighborhoods, potentially initiating new waves of change — a spatial cascade effect observed in real cities.
Key Findings
8/8
Thesis Grade
5+
Cities Modeled
1000s
Agents Simulated
Mesa
ABM Framework
The ABM successfully reproduces spatial clustering patterns observed in empirical gentrification data across European cities.
Neighborhood tipping points emerge naturally from individual agent interactions, without being explicitly programmed.
The rent gap mechanism is the strongest predictor of where gentrification initiates, while amenity proximity determines its speed.
Displacement cascades create predictable spatial patterns that can inform urban planning policy interventions.
Sensitivity analysis reveals that income inequality and housing supply elasticity are the most impactful macro-level parameters.
Technical Stack
Simulation
- •Python + Mesa ABM framework
- •NumPy/Pandas for data processing
- •Spatial grid with real GIS calibration
Analysis
- •Spatial econometrics validation
- •Monte Carlo sensitivity analysis
- •Matplotlib/Seaborn visualization
Geospatial
- •GeoPandas for spatial data
- •OpenStreetMap amenity data
- •Real census tract boundaries
Data Sources
- •Eurostat urban audit data
- •National housing price indices
- •Census demographic data