Non-Linear Forecast of Returns at Bovespa: Trading Volume in a Smooth Transition Auto Regressive Model (Previsao Nao-Linear De Retornos Na Bovespa: Volume Negociado Em Um Modelo Auto-Regressivo De Transicao Suave)

Abstract:     
In this study, the predictive power of a logistic smooth transition auto regression model (LSTAR) in generating statistically significant returns is evaluated when the transition variable is trading volume and the lagged return itself, for the Sao Paulo Stock Exchange's Ibovespa Index, with the analysis based on daily data between 1996 and 2006. The reason for the inclusion of trading volume is found in some market characteristics and behavioral finance results, which indicate the existence of a negative relationship between trading volume and future returns. The model shows a good adjustment to the data, although it does not have the ability to generate additional profits if the transaction costs are of 0.5% per trade. For lower costs there is some predictive power, though lower than a AR(1) model and an buy and hold strategy. Considering the risk, for transaction costs of 0.035% per trade, the autoregressive model allowed a Sharpe index 20% bigger than the buy and hold strategy.

Keywords: Return forecast, Non linear models, Ibovespa index, Trading volume, Brazil

JEL Classifications: C32, C53, G10

Accepted Paper Series