Stock Price Forecast using Wavelet Transformations in Multiple Time Windows and Neural Networks

This paper presents a highly reliable and accurate stock-price prediction model. We aim to anticipate the stock price with respect to multiple patterns in different time scales. The stock price time-series are decomposed, using discrete wavelet transform (DWT), into temporal resolution of varying scales. Then, each subseries is used to predict the stock price using two types of neural network (NN) models with one and two hidden layers. Results show that having multiple time windows in input datasets together with DWT decrease the RMSE of NN models below 10%.