This paper aims to infer about the distribution of extremes values of a continuous random variable, represented as the severe daily losses in financial markets investments. The Extreme Value Theory (EVT) plays a fundamental role in modeling rare events associated with great losses and very small probabilities of occurrence. One of the great concerns in risk management is to develop analytic techniques to foresee those exceptions. In that way, the tails of the rare losses' probability density function (pdf) are of great importance in evaluating that kind of risk, turning EVT into a valuable tool for an accurate evaluation of high loss risks. The estimations of expected maximum losses in financial series are investigated by means of: i) traditional methods, which used all sample data in fitting the random variable pdf; ii) the Extreme Value methodology, particularly the Generalized Extreme Value distribution (GEV), which only used a set of maximum values detected in the sample data in estimating the pdf of expected maximum losses. The findings indicate, firstly, an important underestimation of extreme losses with the traditional methods, mainly in the pdf lower tail limits, and, secondly, that the GEV distribution proved to be more efficient in forecasting extreme losses in the analyzed series: Ibovespa, Merval, Dow Jones.
Financial Investment Risk; Distribution of Extreme Values; Value-at-Risk (VaR); Extreme Losses
