A naive reason has been explained by Nassim Nicholas Taleb in his book titled Black Swan.
In a deeper look, one should be aware that no historical data analysis can truly estimate the real tail risk of financial markets. By the same token, standard deviation, max drawdown, expected shortfall, VaR, Conditional Var... No single or combination of such metrics can truly estimate the tail risk. Major News, market shocks and dislocations, randomness, etc can all lead to extreme behaviors and heavily tailed marginals.
I once wrote a thesis on Extreme Value Theory, and my work is basically concerned on modeling the occurrence of large-impact and low-probability extreme events on various domains. And our finding is that almost all extreme events will converge to a Generalized Pareto Distribution.