Abstract:Economic agent-based models (ABMs) are becoming more and more data-driven, establishing themselves as increasingly valuable tools for economic research and policymaking. We propose to classify the extent to which an ABM is data-driven based on whether agent-level quantities are initialized from real-world micro-data and whether the ABM's dynamics track empirical time series. This paper discusses how making ABMs data-driven helps overcome limitations of traditional ABMs and makes ABMs a stronger alternative to equilibrium models. We review state-of-the-art methods in parameter calibration, initialization, and data assimilation, and then present successful applications that have generated new scientific knowledge and informed policy decisions. This paper serves as a manifesto for data-driven ABMs, introducing a definition and classification and outlining the state of the field, and as a guide for those new to the field.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the limitations of traditional economic models (such as equilibrium models) in simulating and predicting real - world economic phenomena. Specifically: 1. **Simplifying Assumptions**: Traditional equilibrium models rely on a series of simplifying assumptions. For example, they use a representative agent to represent the behavior of all individuals, which limits the ability to capture diverse behaviors. 2. **Ignoring Heterogeneity and Human Behavior**: These models usually assume that individuals are perfectly rational, ignoring the heuristic decision - making processes and other behavioral characteristics that are common in reality. 3. **Lack of Flexibility**: Since equilibrium models are based on equation - solving, they may not be able to flexibly reflect complex real - world dynamics, especially when dealing with phenomena such as sudden crises, inequality, and bounded rationality. To solve these problems, the author proposes a new method - Data - Driven Economic Agent - Based Models (ABMs). This model overcomes the limitations of traditional ABMs in the following ways: - **Initialization with Micro - data**: Use actual micro - data to initialize the individual attributes in the model, reducing the number of arbitrary assumptions. - **Construction of Synthetic Populations**: By reconstructing population characteristics consistent with reality, the model can more accurately reflect the attributes of individuals and households. - **Time - series Tracking**: Enable ABM to track and fit actual time - series data, thereby improving the quantitative reliability and predictive ability of the model. ### Specific Problems The paper mainly explores the following specific problems: 1. **How to Define and Classify Data - Driven ABMs**: Propose a technical definition and classify ABMs according to whether the model obtains individual - level attributes and variables from data and whether it tracks empirical time - series. 2. **New Methods for Parameter Calibration, Initialization, and Data Assimilation**: - **Parameter Calibration**: Introduce methods for deriving parameter values from data, including obtaining them directly from data, relying on previous estimates or expert knowledge, and making inferences through a four - step method (sampling, selecting summary statistics, defining a loss function, and selecting parameter criteria). - **Initialization**: Discuss how to initialize individual attributes and initial conditions from data, especially generating synthetic populations, reconstructing networks, and ensuring consistency with national accounts. - **Data Assimilation**: Explore how to estimate the time - series of unobserved latent variables, with a focus on methods such as the Kalman filter, particle filter, and probabilistic graphical model. ### Goal The goal of the paper is to promote data - driven ABMs as a powerful tool for economic research and policy - making, providing more reliable policy recommendations by more accurately simulating complex economic phenomena.

Data-Driven Economic Agent-Based Models