An improvement of hidden Markov model for stock market predictions

This paper predicts Tehran Exchange Dividend and Price Index (TEDPIX) by finding a pattern in TEDPIX through settled transactions and open orders volume effects. To do so, we improve an autoregressive hidden Markov model (AR-HMM) by adding a more hidden layer. Then, we utilised a genetic algorithm for long term daily trend predictions. By exploiting the obtained information of predicted five days using the genetic algorithm, we update the parameters of improved AR-HMM. This stepwise prediction-updating process continues until all desired number of future days stock exchange indices get predicted. Comparing our new scheme with other studied Markov family models shows that the added features lead to achieve more accuracy and less prediction errors. Experimental results show that mean absolute percentage error of all predictions by our improved AR-HMM approach are less than 5% which indicates far better performance of our method against Markov and Hidden Markov Models. Copyright © 2022 Inderscience Enterprises Ltd.