Developing an automated stock recommendation system by incorporating technical indicators and textual data

The likelihood of forecasting the price of financial assets such as stocks, options ETFs etc. by means of technical indicators is one of the most important challenges for researchers, individual investors and companies linked to the financial market [1]. There is a lot of excitement and zeal when it comes to the topic of stock prediction; currently there are many problems faced by the stock prediction community, like the systems may lack accuracy, systems may not have the ability to predict each and every significant event, and most of the modern day systems focus on high frequency trading. The goal of the thesis is to develop a stock recommendation system by incorporating patterns extracted from technical indicators and features extracted from the textual data. We use a three-step method comprising of: (a) extracting sentimental analysis features from textual data i.e. positive words and negative words, (b) extracting features from time series data, and (c) combining features and using machine learning for prediction. We tested the presented works on seven popular stocks from S&P; 500, by combining five different combinations of features. The best results were obtained when technical indicators were used with a mean accuracy of 70.97%, while some stocks like General Electric (GE) giving accuracy as good as 85.71%. The second to best results are obtained when a part of textual features are used along with technical indicators to get a mean accuracy of 67.44%.