Nasrin Shabani - Tamadon Investment Bank, Tehran, Iran
Ahmad Kuchaki - Tamadon Investment Bank, Tehran, Iran

Stock market prediction is one of the most important challenges that data analysts face in the finance sector. Therefore, a plethora of research projects have been carried out to facilitate forecasting the future of stock market prices. Recently, Hidden Markov Model (HMM) has been successfully utilized for this purpose. In this paper, we elaborate on combining HMM and Term Frequency-Inverse Document Frequency (TF-IDF) term weights using online political news to predict next day’s stock prices for a few selected companies in Iranian stock market. The HMM we use is based on Maximum a Posteriori (MAP) estimation instead of common Maximum Likelihood Estimation (MLE) approach. Ourresults show that some surprisingly huge changes in stock prices could well be forecasted by our model. We compare the results of this study to other well-known machine learning algorithm, Artificial Neural Network (ANN), by using Mean Absolute Percentage Error (MAPE).

forecasting, text analysis, hidden markov model, MAP estimation, stock market, TF-IDF