Models whose assumptions are at odds with real markets are likely to produce poor output.
Also, because so much of the modern financial world relies on such models so heavily, widespread application of unsuitable models can create unstable feedback loops and lead to the emergence of system-wide instabilities. For example, the severe market crash in 1987 is often attributed to the prevalence of models that assumed independent Gaussian price returns, and thereby severely underestimated the probability of large price changes.
Bizarrely, financial crises can be induced by the very models designed to prevent them.
Market crashes serve as a wake-up call to reject idealistic simplifications and to move towards a more realistic framework that encompasses the real statistical properties of price changes observable in empirical data. Despite considerable recent effort in this direction, this goal remains elusive, due partly to the fact that many of the statistical properties of real price series are highly non-trivial and sometimes counter-intuitive.
