-
ABM never took off in economics because it was already a “model-oriented” discipline. ABM is inherently empirical which means that:
- It is hard to validate the created model;
- It is hard to prove that these models are indeed theoretically sound;
- There are a lot of bad ABM models for economics compared to good ones since it is easy to do; and
- Data Analytics had more ground compared to Agent Based Modeling
-
1 gives a review on ABM with regards to modeling economic behavior
-
In the context of computational social science, ABM can be characterized as having the following components
- Agents have psychological traits (heuristics) and socio-demographic attributes. These traits could be based on real world data.
- Agents operate on an environment where they interact with others.
- Agents operate under rules of interactions and decision making processes are deduced from social and psychological theories.
- Macro-level structure that emergent as a consequence of micro-level behavior. Tuning the micro helps understand the macro.
-
Data driven ABMS are initialized and validated using data from surveys, digital media, social network data, crowd sourced data, digital sensors and information networks, existing databases, census data, and urban data.
-
Heuristics can be used to model human behavior. Agents in the system can either be “automatic” (i.e., have no cognitive function) or learn from the environment.
- Behavioral heuristics underutilize the possibilities offered by expectation formation theory — that is, heuristics based on the expectation of a variable (such as price).
- Heuristics may provide a more accurate and robust tool for modeling action also within an uncertain environment than sophisticated techniques.
-
Besides modeling the behavior of agents we have to model realistic networks of interactions where these are relevant for the dynamics of the system
- Large scale networks could be used to study cascades via cascade models.
- ABM can be used to study the extent to which network structure influences macroscopic outcomes. However, they do not answer whether the proposed structures are found in reality, how they formed, or how they might develop
- Modeling could be done using data or analyzing the dependencies of real world markets.
-
ABMs can be used to explain the stability of a financial system
- Early work has focused on micro-structures such as transactions and trading
- They can help in identifying mechanisms that lead to instabilities, and to evaluate policies to mitigate them.
- They can also be used for risk analysis and to the extent that cascades in the system are affected by the network’s structure.
- ABM could be used to test the effectiveness of micro and macroprudential policies to counteract banking crises.
- ABM can be used to generate a dataset to evaluate the consequences of policy.
-
ABMs have so far mostly been used to generate insights and qualitative descriptions of scenario that may occur rather than quantitative forecasts.
-
ABM can be extended to apply to heterogeneous agents — that is, agents that behave differently and not necessarily rationally .
-
ABM can be used to explain experimental results.
-
ABMs can be used to generate priors to produce scenarios for machine learning algorithms in a semi-supervised manner to reduce errors and prevent the amplification of distortions
-
Links
Footnotes
-
Steinbacher et al. (2021) Advances in the Agent-Based Modeling of economic and social behavior ↩