Agent-Based Models: A Game-Changer in Real Estate Economics

In the realm of economics, understanding the ways individuals’ preferences and decisions affect markets has always been a fundamental challenge. Traditional economic models often rely on simplified assumptions, “closed-form solutions” (which is just another way of saying simple equations), and heavily aggregated data, overlooking the complexity of individual behaviors and interactions. However, the emergence of agent-based models (ABMs) has revolutionized the way social scientists perceive and analyze social phenomena.

Agent-based modeling is a computational method that simulates the actions and interactions of autonomous agents to understand how complex systems evolve. These agents could represent individuals, households, firms, or any other entities relevant to the economic context under study. In short, you model preferences and create rules regarding how the sides will strategize, trade, and interact and you watch what happens. Unlike traditional models, ABMs allow for heterogeneity, adaptability, and emergent properties, making them particularly well-suited for studying real-world markets characterized by diverse agents and nonlinear dynamics: the real estate market in particular.

In the context of real estate, ABMs offer several distinct advantages:

Micro-Level Insights

ABMs enable researchers to model the behavior of individual agents, such as homebuyers, sellers, developers, and investors, each with their unique preferences, strategies, and constraints. By capturing the heterogeneity among agents, ABMs provide micro-level insights into decision-making processes that drive macro-level market outcomes. Furthermore, polling of home builders and other important market actors can be built directly into the model. This sort of data, despite its vital importance to real estate analytics, is much harder to incorporate into a traditional, econometric model.

Spatial Considerations

Real estate markets are inherently spatial, with location playing a crucial role in determining property values, demand patterns, and neighborhood dynamics. ABMs can incorporate spatial elements, such as geographic features, land use patterns, and transportation networks, allowing for more realistic representations of local housing markets and urban environments.

Dynamic Interactions

Traditional equilibrium models often assume static market conditions, overlooking the dynamic nature of real estate markets characterized by feedback loops, network effects, and path dependencies. ABMs capture the dynamic interactions between agents and their environment, revealing how local shocks or policy interventions propagate through the system and lead to emergent phenomena such as housing bubbles or urban sprawl. 

While there are interesting metrics that can help to detect bubbles, such as looking at rent-buy ratios (i.e. the cost of renting vs. buying) these are complicated by the fact that anticipation of rent increases can push up home prices in anticipation and vice versa. ABMs might help us detect bubbles that have been difficult to track with standard metrics. 

Policy Analysis

ABMs serve as valuable tools for policy analysis and scenario planning, allowing policymakers to assess the potential impacts of different interventions or regulatory changes on real estate markets. By experimenting with alternative scenarios within a simulated environment, policymakers can anticipate unintended consequences and design more effective policy interventions to address housing affordability, urban development, or other socio-economic challenges. 

This makes them much more proactive than the traditional approach to answering such questions, difference-in-difference models (otherwise called a controlled before and after study), which require a similar policy change to have been enacted elsewhere in the past. 

Forecasting and Risk Assessment

ABMs can be used for forecasting future trends in real estate markets and assessing potential risks associated with different scenarios. By integrating historical data, economic indicators, and demographic trends, ABMs can generate probabilistic forecasts that account for uncertainty and variability, helping investors, developers, and policymakers make informed decisions in an uncertain environment.

Models Utilizing Demographic Data

It is not always clear how one can build demographic assumptions, such as rates of family formation, retirement, etc. into a traditional economic model even though such things have substantial effects on the housing market. However, with agent-based models, building models that account for these sorts of factors is relatively straightforward.

Despite their advantages, ABMs also pose several challenges, including data limitations, computational complexity, and model validation. Constructing an ABM requires careful calibration and validation against empirical data to ensure its relevance and reliability as a decision-making tool (Of course, so does regular modeling even if, in the case of home price indices and the like this has not always been done). Most importantly, it requires regression testing—an idea separate from the notion of statistical regression: Namely, if I feed it data from the past, but leave a gap between the recent past and the present, how well does it predict what happened in that gap—i.e. how well does it predict the recent past using only data further back.

Again, however, this is true of any complicated model framework that has the potential to overfit the data: Machine learning is conceivably worse in this regard in that it models relationships in the data directly instead of working bottom-up using economic theory. That said, many traditional statistical methods can be taken “as is” without nearly as much risk. The more advanced your methods become, the more important testing on out-of-sample, or even out of time, data becomes. 

Hostility towards ABMs?

There is still some widespread hostility towards ABMs among traditional econometricians. This can be explained through various factors, most notably a fear of learning new techniques and an over-reliance on rules of thumb when evaluating journal articles (What are your p-values, for example). That said, several European central banks use ABMs as part of their risk modelling (a task of no small importance), and the Bank of England famously relies on one to understand England’s real estate market—indeed, the results of this model are kept secret out of fears that endogeneity, or people’s knowledge of the model’s predictions, might invalidate them. 

To understand how serious endogeneity is, consider a model that predicts a stock market crash in two months. If people believed this model, the stock market crash would happen as soon as its results were published—invalidating its prediction. Despite its clear successes not only in economics but other crucial fields like epidemiology, you can witness the controversy for yourself here: What is the popular opinion on agent based modelling (ABM) in our field? (Be forewarned, the above thread may reduce your opinion of economists.)

The role of Kukun and ABMs

Kukun is one of the few real estate analytics firms with experience in building ABMs. Our forecasts, specifically the newest version of our Kukun Investment Outlook, hybridize traditional time series methods, machine learning, and ABMs to produce a truly revolutionary forecast. 

In conclusion, agent-based models represent a powerful paradigm shift in economic analysis, offering a more nuanced and realistic approach to understanding complex systems such as real estate markets. By capturing the heterogeneity, spatial dynamics, and emergent properties of economic interactions, ABMs provide valuable insights for policymakers, researchers, and industry stakeholders seeking to navigate the complexities of real-world markets and address pressing socio-economic issues. As computational resources and data availability continue to improve, the potential for ABMs to inform evidence-based decision-making in real estate economics will only grow stronger in the years to come.

If you would like to learn more about ABMs and how they work, consider reading any of the following articles and books:

1. Agent-based modeling: Methods and techniques for simulating human systems

2. Agent-Based Modeling in the Philosophy of Science (Stanford Encyclopedia of Philosophy)

3. Modeling Social Behavior: Mathematical and Agent-Based Models of Social Dynamics and Cultural Evolution

4. Agent-Based Models (Quantitative Applications in the Social Sciences)

5. Growing Artificial Societies: Social Science From the Bottom Up (Complex Adaptive Systems) 

Franklin Carroll

Franklin Carroll is Kukun's VP of Modelling and Analytics; After finishing his studies at the University of Chicago, he worked at Fannie Mae where he helped build Collateral Underwriter, roll out the Value Verify initiative, and specialized in collateral related and credit related policy.

Published February 28, 2024