Sloppy Models

Agent-based Models (ABMs), and non-linear macroeconomic models more generally (e.g. SFC models), face criticism for being perceived as &quote;black boxes&quote; - complex systems where the underlying mechanisms are unclear, the possible outcomes seem limitless, and the large number of parameters makes empirical validation challenging. This project addresses these concerns by demonstrating that ABMs may actually have a finite (even narrow) set of distinct behavioral patterns, and these patterns depend on just a few key parameters. This finding makes these models more manageable and understandable than previously thought

In particular, we introduce an innovative approach from biophysics called "sloppiness". This method reveals that high-dimensional models are typically sensitive to only a small number of well-defined parameter combinations, while many other parameters can vary significantly without affecting the model's behavior. This insight helps us:

• Identify the most influential parameters in the model
• Understand how these parameters affect the model's behavior
• Efficiently explore the model's possible outcomes

By focusing on these well-constrained parameters, we can better understand and predict the model's behavior, making ABMs more practical and reliable tools for research and analysis.