Navigating through Economic Complexity: Phase Diagrams & Parameter Sloppiness

We argue that establishing the phase diagram of Agent Based Models (ABM) is a crucial first step, together with a qualitative understanding of how collective phenomena come about, before any calibration or more quantitative predictions are attempted. Computer-aided *gedanken* experiments are by themselves of genuine value: if we are not able to make sense of emergent phenomena in a world in which we set all the rules, how can we expect to be successful in the real world? ABMs indeed often reveal the existence of Black Swans/Dark Corners i.e. discontinuity lines beyond which runaway instabilities appear, whereas most classical economic/finance models are blind to such scenarii. Testing for the overall robustness of the phase diagram against changes in heuristic rules is a way to ascertain the plausibility of such scenarii. Furthermore, exploring the phase diagrams of ABM in high dimensions should benefit enormously from the identification of ``stiff'' and ``sloppy'' directions in parameter space.