Extreme Value Theory to Zero inflated data
Inflated Zeros Data, Extreme Events, Bayesian Approach.
Extreme events are usually responsible for producing big gains or big losses to society. For decades already existing a specific distribution, to predict and prevent such events, known as Generalized Extremes Value (GEV) Distribution. However, in many situations with extreme data, there is the presence of excess zeros in the database, making the analysis and decreasing the accuracy in the estimation. Zeros Inflated distribution (ZID) is recommended to model these data have inflated zeros. It is the aim of this work to create a specific distribution to model extreme events with inflated data zeros. Therefore, it was be done mixing the GEV distributions and ZID also a Bayesian approach in the search for a better fit, and applied in extreme data with excessive amounts of zeros, being chosen for analysis daily precipitation of rain on the city of Natal in Rio Grande do Norte state and the cities of Paulistana, Peaks, St. John’s Piaui and Teresina of Piaui state. It was also used the Standard GEV distribution to model these same data collected for comparison, and thus be able to check the quality setting found by the new extreme value distribution with inflated data zeros. Therefore, it was found that both the model and the algorithm were well developed, indicating a better fit for extreme inflated data zeros, and the GEV pattern could not find the equilibrium distribution data when the data have many zeros. And when the extremes data has inflated zeros, the new model converges to the Standard GEV, identifying the absence of zeros