Stock Market Forecasting Using Continuous Wavelet Transform and Long Short-Term Memory Neural Networks

Abstract

The analysis and exploitation of complex and large-volume data requires new approaches, and modeling it in time series is a very successful technique. A characteristic time series is the one that defines the dynamic financial market and its asset prices. This research presents a novel forecasting methodology, which uses the Continuous Wavelet Transform for the definition of representative elements that define a time series, and a recurrent neural network architecture for the forecast of prices of financial stocks related by the item of income in the short and medium time term. The proposed model, inspired by the Continuous Wavelet Transform and Neural Networks of the "Long short-term memory" type, uses the most representative coefficients of the Wavelet transform based on the time series in the time domain, for the prediction of future prices of stocks in short prospective periods. The results show a very successful projection using this methodology. Future research will analyze the interrelationship presented by the price time series of the same stock market section, in the domain of Wavelets, and how it affects the stock market forecast.