Improving artificial neural network based stock forecasting using fourier de-noising and Hodrick-Prescott Filter

Accuracy in financial forecasting is a key determinant of profits in the financial markets. This paper proposes improvements to existing Artificial Neural Network based forecasting approaches using de-noising in frequency domain and the Hodrick-Prescott Filter. Traditionally used technical indicators are replaced with open, close, high, and low prices only. Forecasts achieved via these improvements are seen to outperform existing results. 8 stocks from the Dow Jones Industrial Average, were considered over a period of 6 years, between 2000 and 2005. The best and worst directional accuracy achieved were 90% and 79% respectively.