Hybrid models for intraday stock price forecasting based on artificial neural networks and metaheuristic algorithms

摘要

Stock market prediction is one of the critical issues in fiscal market. It is important issue for the traders and investors. Artificial Neural Networks (ANNs) associated with nature inspired algorithms are playing an increasingly vital role in many areas including medical field, security systems and stock market. Several prediction models have been developed by researchers to forecast stock market trend. However, few studies have focused on improving stock market prediction accuracy especially when utilizing artificial neural networks to perform the analysis. This paper proposed nine new integrated models for forecasting intraday stock price based on the potential of three ANNs, Back Propagation Neural Network (BPNN), Radial Basis Function Neural Network (RBFNN), Time Delay Neural Network (TDNN) and nature inspired algorithms such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and Artificial Bee Colony (ABC).The developed models were named as GA-BPNN, PSO-BPNN, ABC-BPNN, GA-RBFNN, PSORBFNN, ABC-RBFNN, GA-TDNN, PSO-TDNN and ABC-TDNN. Nature inspired algorithms are employed for optimizing the parameters of ANNs. Technical indicators calculated from historical data are fed as input to developed models. Proposed hybrid models validated on four datasets representing different sectors in NSE. Four statistical metrics, Root Mean Square Error (RMSE), Hit Rate (HR), Error Rate (ER) and prediction accuracy were utilized to gauge the performance of the developed models. Results proved that the PSO-BPNN model yielded the highest prediction accuracy in estimating intraday stock price. The other models, GA-BPNN, ABC-BPNN, GA-RBFNN, PSO-RBFNN, ABC-RBFNN, GA-TDNN, PSO-TDNN and ABC-TDNN produced lower performance with mean prediction accuracy of 97.24%, 98.37%, 84.01%, 85.15%, 84.01%, 83.87%, 89.95% and 78.61% respectively.(c) 2021 Elsevier B.V. All rights reserved.