Modelos de previsão de preços aplicados aos contratos futuros de boi gordo na BM&F

Abstract

This paper studies the applicability of time series models as a decision tool of buy and sell orders of live cattle futures contracts in the Brazilian Futures Market (BM&F), on dates close to expiration. The models considered are: ARIMA, Neural Networks and Dynamic Linear Models – DLM (this in the classic and bayesian approach). Weekly data, of the spot and futures markets, from 1996 to 1999, are used to calculate the forecasts. The main purpose is to calculate the returns, in buy/sell orders of live cattle futures between 1998 and 1999, in order to show the potentials or limitations of each model. The results show positive returns in almost all contracts analyzed, indicating the potential of the models as a decision tool in operating with futures contracts close to expiration date, with distinction on the performance of the Classic DLM and ARIMA models, although some differences in forecasting accuracy.