Stock tracking : a new multi-dimensional stock forecasting approach
Huifeng Shen, Congfu Xu, Xuemei Han, Yunhe Pan · 2005
This paper proposes a new approach-stock tracking that can forecast multi-dimensional stock information simultaneously. In this approach, a vector is used to represent stock information in point and a stock model is constructed to describe the stock fluctuation. A finite stock trend set is defined for simplifying the stock model. When the stock keeps the same trend, the model can be degenerated to a linear one and each trend has its own unique model parameters. The approach devises a new way to compute the transition matrix of the stock model and employs it for forecasting stock prices, volumes and indices etc. As for the stock trend changing, the discrete Markov process is adopted for stock forecasting. The experiments demonstrate the effectiveness of this approach. Furthermore, our approach can be used to solve those multi-dimensional financial forecasting problems where the state and observation space are the same Hilbert space, the trend set is a finite set, and each state corresponds to one observation.