Agent-based modelling in economics

11 Mar, 2021 at 11:19 | Posted in Economics | Comments Off on Agent-based modelling in economics

I DSGE sono i modelli più diffusi in accademia e nei dipartimenti di ricerca delle banche centrali. Anche prima della Grande Recessione, dovuta, secondo la vulgata, alle banche. Il modello è spesso rimproverato per non aver previsto la crisi, ma il problema è diverso. Come vedremo nel prossimo capitolo, e già si intuisce dal problema delle nonlinearità, la disciplina economica non può prevedere: il vero problema è che non essendoci il tempo non possono esserci né le banche, né il debito e il credito, non sono contemplate le crisi finanziarie. La moneta e il credito sono “aggiunte” ai sistemi di equilibrio generale walrasiano “reali” (baratto perfetto) come i RBC. I DSGE sono modelli di baratto mascherati da modelli monetari … Se prima della crisi finanziaria globale, il settore finanziario non aveva alcun ruolo nei modelli DSGE, la Grande Recessione ha messo in luce tale limitazione e numerosi aspetti del settore finanziario sono stati incorporati nei modelli DSGE di seconda e successiva generazione. Sfortunatamente, questi innesti sono sbagliati perché non affrontano il difetto fondamentale di questi modelli – e questo “errore” è ripetuto in tutte le generazioni successive di modelli DSGE … In modo analogo, il sistema deferente/epiciclo era in grado di modellare alcune delle orbite dei corpi del sistema solare attorno alla Terra senza abbandonare il sistema di Tolomeo.

In Mauro Gallegati’s Il mercato rende liberi (LUISS University Press, 2021) the reader is presented with a forceful critique of mainstream economics in general, and more specifically, of the kind of assumptions that  mainstream macroeconomic DSGE modelling (New Classical, ‘New Keynesian’, RBC, etc.) is built on. For most heterodox critics of mainstream economics the themes in this well-written and highly accessible book are probably well-known, and I will here only discuss what Gallegati at the end of it presents as the alternative to mainstream orthodoxy — agent-based modelling.

Agent-based models are formal models usually constructed using mathematical programming and performing simulations and ‘artificial experiments’ with the intention of being able to (more explicitly than in conventional mainstream game theory) describe aggregate effects and dynamics of interacting individuals and socio-economic structures without standardly having to assume equilibria, non-emergence, Walrasian auctioneers, representative agents, rational expectations, etc., etc..

Agent-based models come in different degrees of realism and are usually conceptualised as different kinds of self-organising complex systems. But one thing they all have in common is reliance on mathematical formalism. In essence the agent-based modelling endeavour in macroeconomics is an attempt at providing new alternative mathematical models where many of the bizarre and ridiculous known-to-be ‘unrealistic’ assumptions in standard DSGE models are replaced with other less ‘unrealistic’ assumptions. But the idea that mathematical modelling as such is always appropriate to apply is never seriously questioned. And that’s where I find it hard to follow. One set of mathematical tractability assumptions are substituted for another. But what if the mathematical modelling in itself is the problem? What if the use of mathematical-formalistic modelling in itself biases your research efforts in specific directions? If it is the mathematical-formalistic approach in itself that is the problem, we only end up with different models based on the same unquestioned mathematical modelling strategy. From my own critical-realist perspective I can’t see that mathematical modelling is the self-evidently appropriate way to perform analyses of societies and economies. The kind of ‘closures’ demanded of the target systems for warranting the analyses, I would argue, simply often aren’t there.

As a critique of mainstream economics, yours truly fully appreciates the work done by people lika Mauro Gallegati and his mentor Alan Kirman. But although their alternative agent-based models in many ways are superior to the more traditional mainstream ‘Walt Disney’ kind of models, I am not convinced that their unquestioned attachment to mathematical-formalist modelling is the right way to move forward in making economics a more realist and relevant science.