Modeling of Brazilian financial series: Forecasting prices of some stocks

The goal of this paper is modeling Brazilian financial time series as non-linear and non-stationary stochastic processes. The random walk hypothesis is tested. We model the time series as random walks and also using price trend models, to determine how well each model explains the observed series behaviour. Three daily series are studied: the closing value of the São Paulo stock exchange index and the closing prices of the US dollar and of gold, from 1986 to 1998. We conclude that the price trend models have a good performance only in the beginning of the period covered by the data and the random walk model explains better than price trend the rest of the time.