The goal of this paper is to evaluate the predictability of a stock market using the one-step-ahead forecasts of the returns as a signal for automatic trading. The paper presents a proposal of a multiple regime model ...The goal of this paper is to evaluate the predictability of a stock market using the one-step-ahead forecasts of the returns as a signal for automatic trading. The paper presents a proposal of a multiple regime model that combines aspects from STAR models, and decision trees. The resulting model, so-called STARX-Tree, is a regression tree with smooth transition and linear ARX models fitted in the terminal nodes. The methodology was tested on 23 stocks of the U.S. stock market. The forecasting model is evaluated through statistical and financial measures and compared to the random walk model, the naive approach, the neural networks and the linear ARX model. The results pointed out that the STARX-Tree model outperforms the comparative models under the fmancial criterion.展开更多
摘要The goal of this paper is to evaluate the predictability of a stock market using the one-step-ahead forecasts of the returns as a signal for automatic trading. The paper presents a proposal of a multiple regime model that combines aspects from STAR models, and decision trees. The resulting model, so-called STARX-Tree, is a regression tree with smooth transition and linear ARX models fitted in the terminal nodes. The methodology was tested on 23 stocks of the U.S. stock market. The forecasting model is evaluated through statistical and financial measures and compared to the random walk model, the naive approach, the neural networks and the linear ARX model. The results pointed out that the STARX-Tree model outperforms the comparative models under the fmancial criterion.