Abstract:
Stochastic processes are frequently used in the assessment and follow-up of activities depending on restricted and unstable data. Various kinds of activities, including prediction, speech and gestures recognition and involvement even in social interaction can be simulated using stochastical processes. The reasons for using Hidden Markov Model to predict stocks are HMM have a solid statistical background and can easily handle highly volatile and time-invariant data. The proposed mechanism is supposed to be a Markov process with numerous hidden states. This study refers HMM to time series data for economic purposes to investigate the parameters which the model can predict. Most of the machine learning techniques typically suffer two types of problems. First, these techniques are too complex to handle and secondly, these techniques could not adjust their initial parameters to get the desired results. To overcome these problems, HMM is used in which model train its data through the Baum-Welch Algorithm and Viterbi Algorithm adjust the initial parameters of the model to obtain the desired output. HMM has been using four visible states (Open, High, Low and Close) to predict the next day’s stock prices. The predicted stocks prices of the next day are used as the present-day parameters for the second forecast in the range of predictions. Stocks of different companies are predicted and the Mean Absolute Percentage Error is calculated by using actual and predicted prices. These results are compared with Artificial Neural Networks (ANN), Auto-Regressive Integrated Moving Average (ARIMA) and Simple Vector Machine (SVM) to evaluate the performance of the model. This analysis provides stock broker to decide based on the probability and transition pattern of each hidden state that cannot be identified from business information. In other words, HMM is the future model for stock trading as it reflects the stock price trends well.
