Financial time series forecasting using optimized multistage wavelet regression approach

This paper presents financial time series forecasting with multistage wavelet transform (WT). First, the time series data is processed through WT with different mother wavelet functions to extract high frequency and low frequency coefficients. Later, standard particle swarm optimization (PSO) algorithm is utilized to find optimal regression models in order to predict future samples. Mean square error (MSE) is opted as cost function for PSO to find optimal coefficients of the regression model. This study further extended to various mother wavelet functions and their decomposition levels to investigate their impacts on time series prediction. These investigations help to data scientists for selection of process parameters and variables. Further, the impact of control parameters of PSO is also discussed to show the importance in the search mechanism especially in regression problems.