The maths models used in economics are rarely checked against real-world data, but when they are they often turn out to be wrong.
A major instance of this problem comes from models used for trading, which often assume Gaussian distributions and Brownian-motion behaviors. According to those, extreme moves such as the ones occurring during economic crises such as the one of 2007 are several sigmas away from the mean -- making them supposed to happen perhaps once every 10^10 years. Yet they happen every 20 years. Taleb's book "Black Swan" is an interesting resource on this topic, for those interested.