Title: 2-Gaussian distribution modeling of financial data
Authors: Cesar Berci - FEEC - UNICAMP (Brazil) [presenting]
Celso Pascoli Bottura - UNICAMP (Brazil)
Abstract: The inadequacy of the normal distribution for representing the empirical probability distribution model of financial data, due to heavy tails and the leptokurtosis of the assets returns, among other reasons, is a challenging problem. The use of a novel stable distribution function that can model heavy tails and leptokurtosis is usual for probabilistic modeling of financial data. Some authors suggest that the $\alpha$-stable is the stable distribution that has the best fit for financial data. However, this kind of distribution has no finite variance or higher moments or even an analytical closed-form, limiting its uses. For instance, it is inadequate for applications based on volatility analysis. An alternative, the 2-Gaussian distribution, a linear combination of Gaussians curves, inspired on Radial Basis Neural Networks, simpler than $\alpha$-stable distribution and more flexible than a Normal distribution, was applied to model assets returns, reaching superior results than the others distributions, including the $\alpha$-stable.