USE OF ARTIFICIAL NEURAL NETWORKS FOR ESTIMATIONS OF RATES ON BOVESPA PETROBRAS PN

Abstract

The increasing sophistication of some operations in the finance market had increased a lot the exposure of the risk from some operations. So, it requests the use of some advanced technologies to model and to esteem series of asset prices. One of the approaches that are receiving importance in the modeling of prices and volatility is the Artificial Neural Networks. The main purpose of this article is to esteem, through ANN, the prices of Petrobrás PN using a series of daily prices between January 2, 2001 and May 9, 2008. This series represents 1821 daily observations. After the ANN was modeled in the software Matlab 7.6.0. R2008a, it was calculated the RMSE, estimated the Petrobrás PN prices and compared the estimated and real prices of the selected asset. The conclusion was that, comparing the estimated prices to the real prices of Petrobrás in the analyzed period, there is a high degree of adherence of the model to ANN in short term. However these studies could have other tests where they will consider different windows of time, including moments with huge stress in the prices and others with fewer adherences in the market, aiming to increase the reliability level of the model.