Financial forecasting is challenging due to small sample sizes, high noise, non-stationarity, and non-linearity. Statistical and probabilistic methods have difficulty with small sample size, high noise data due to correlations between input and output variables which are caused by noise and are random in nature. We present a noisy time series prediction method which is based on the notion that short term dynamical evolution predictability is possible but is obscured by observed correlations due to noise. The task considered is the prediction of daily foreign exchange rates, specifically, the prediction of whether the exchange rate will increase or decrease at the close of business for the next day.
展开▼
