Neural Network:An Alternative Statistical Model for Predicting Financial Time Series Data

This study presents the predictive performance of Neural Network Model. Nigerian Naira (NGN) exchange rates against Ghana Cedi (GHC) using daily exchange rates values were used. The data set were divided into three in the ratio of 3:1:1 for training (parameter estimation), test and validation respectively. Neural Network (NN) Model with back propagation training algorithm using descent gradient minimisation technique and logistic activation function was developed. The architecture for NN Model was determined through Automatic Network Search (ANS). The tuning parameters considered for the training of NN Model are the learning rate and momentum with the values (0.1 and 0.3) and (0.5 and 0.5) for model A and B respectively. The performance metrics considered for the evaluation of the Models is Mean Square Error (MSE) and Mean Absolute Error (MAE). The results show that the performances of NN increase with increase in parameter values, indicating that higher learning rate and momentum values facilitate better convergence.