Financial crises, as others phenomena of the nature, have characteristics of inherent unpredictability, resulting largely from non-linearity of the dynamics. But in financial markets (and in general in the study of social phenomena) we could defend another issue: the limited validity the formulation of the model derives from mathematical models of it. In this article, I argument that, in the study of financial markets, the main problem is the assumptions rather than the mathematical modeling. I suggest to introduce in the analytic models heterogeneous agents to study the interactions and the emergence of macro structures. I guess that micro foundations of the modeling of financial markets is the key to an understanding of the capital markets and to propose a some kind of policy against cyclical crises.