The most credible explanation of why risk management based on state-of-the-art statistical models can perform so poorly is that the underlying data used to estimate a model’s structure are drawn generally from both periods of euphoria and periods of fear, that is, from regimes with importantly different dynamics.
-
-
