Studying economic complexity with agent-based models: advances, challenges and future perspectives

Abstract Agent-based computational economics has considerable achievements. However, it has gone too quickly into a direction similar to the one of models based on solely analytical—as opposed to algorithmic—dynamic systems of difference equations. An increasingly large focus has been put on matching moments of real-world time series of data, a set of stylised facts, or on estimation. Reasons why this is not desirable are discussed. Firstly, both estimation and inference from models will be biased, unless they represent the real data-generating processes. Secondly, surrendering the attempt to incorporate realistic microfoundations is not only against the original ACE agenda, but also is subject to a form of Lucas critique. Thirdly, characteristics of complex systems, especially differences between feedback loops and emergent phenomena that characterise systems of various levels of complexity, undermine the justification of building structurally simplistic models. That is, an attempt at reducing the interaction of many different sectors, populated with agents using various decision rules will yield information loss, i.e., some phenomena by definition are possible to emerge only in systems of higher levels of complexity. A different research agenda is proposed, with the aim of systematically analysing and uncovering the mechanisms, feedback loops and impact channels of complex multi-sectoral economic and financial systems.