The Application of SHESN on Financial Time Series Prediction

Echo State Network (ESN) has been widely applied to all kinds of time series prediction tasks because of their advantages in modeling nonlinear systems and strong memory capacity. The improved scale-free highly clustered echo state network (SHESN), whose reservoir has both small-world and scale-free characteristics, has better echo state property and can achieve more accurate prediction results compared with the conventional ESN. In this paper, the extended application of the scale-free highly clustered echo state network model on financial time series prediction tasks is researched. The two widely used financial series S&P500 index and US/Euro exchange rate are chosen here. The experimental results show that the SHESN model has better prediction performance and demonstrate the scale-free highly clustered topology has stronger computational capability.