Where to get your ABM fix

Mini-summary (tl;dr)

• I collect all of the literature on Agent-Based Modelling from 2015 onwards and identify some key features
• Tools: EconLit (subscription) and Citationgecko (free)
• Resulting list of key links are on the resources page
• Common journals: J.o. Economic Dynamics and Control, Computational Economics & J.o. Economic Behaviour and Organisation
• Common authors: Dosi, Roventini, Caiani, Salle & Seppecher, Grazzini

Where to start?

The first steps in research are usually getting "an idea of the state of the literature" that has come beforehand. In effect we want to define our genre by assessing: what has been done before, what is the state of the art, where are the open questions, where is the tension, what are the debates, who are the actors, where are the playing fields, to name but few variants of the question we want to answer.

Since my PhD will focus in part on macroeconomic Agent-based modelling (ABM) my starting questions are along this vein and aim at two things: (1) understanding the streams of thought, the questions that are being answered, the families of models posited to do so, how these are validated empirically, and (2) how can I most effectively keep up to speed with the ABM community (who, where, what...).

• What are the different questions that are asked? What are the different modelling streams?
• What is typically done in each paper: model proposal and how much validation? What are the expected outcomes of the model? How is it analyzed?
• Who are the main contributors to the field of macroeconomic ABM research? Where do they publish? Where do they pre-publish?
• What are the main forums of exchange: conferences, reading groups, forums? How can I subscribe to stay up to date?

In this post I will cover the first step toward answering the above questions:

Getting all the sources in one place

Thankfully I discovered that my institution benefits from a subscription to EconLit, the American Economic Association's database of indexed articles, theses, working papers and co from 1980 onwards. Using this I searched generically for Agent based model, yielding ~3200 results. I refined as follows:

1. Date range 2015 - Present (1755 remaining results)
2. Peer-reviewed (1131 results)
3. Reduce the list of journals to only economic ones (sorry Journal of Transportation Research...)

First observations - most represented journals

Looking at the results overall, these are some of the findings I made from the raw EconLit output. Firstly, I really wanted to know whom I should follow and which journals I should subscribe to.

The result is somewhat unsurprising really - subscribe to the Journal of Economic Dynamics and Control yeh who is tired of DSGE and would like to read about complexity in economics.

Second observations - who writes the most?

From the author list it seems that Italy has a strong lead on Agent-Based modelling (read Dosi & Roventini, 2017 for more on that). Given that this is about co-authorships this list is unsurprising as a lot of them work closely together. Overall - it is more interesting to look at whether there are distinct clusters of authors.

Last but not least - are there distinct streams?

Since I am pretty much a nerd, I wanted to know how all of these articles fit together. To do so without reading each, I had to make a network visualisation. At this stage, it is only co-authorships (getting references done by each is a harder due to data). Anyway, a quick python script using networkx and then visualising with Gephi has led me to the following graph:

There are two main clusters:

1. Dosi & company (lower cluster) - These authors are a rather large group that have published together very regularly. Last in 2020. Their baseline appears to the Keynes+Schumpeter model, with forays into macroeconomics, financial instability, consumer behavior
2. Caiani, Riccetti, Russo & company (upper cluster) - their baseline model is from 2016 (fun fact: Stiglitz co-authored, though who knows how involved he was....). They focused in their creation on stock-flow consistency
3. Further clusters include: Salle & Seppecher / Jakob Grazzini & co. / Oscar Claveira / Huanhuan Zheng / Frank Westerhoff / Thomas Lux

The most prominent cluster is certainly that of Giovanni Dosi and Andrea Roventini, and looks like a promising model class from which to start since it has been investigated in depth and thus any meta-modelling, fitting or other validation and reduction techniques I want to explore can be compared and contrasted to the existing work.

While this is the ABM literature in general, I will next focus on Macroeonomic ABMs but more so on the literature about empirical validation and estimation for ABMs.