FINANCIAL TIME SERIES FORECASTING – A MACHINE LEARNING APPROACH

The Stock Market is known for its volatile and unstable nature. A particular stock could be thriving in one period and declining in the next. Stock traders make money from buying equity when they are at their lowest and selling when they are at their highest. The logical question would be: "What Causes Stock Prices To Change?". At the most fundamental level, the answer to this would be the demand and supply. In reality, there are many theories as to why stock prices fluctuate, but there is no generic theory that explains all, simply because not all stocks are identical, and one theory that may apply for today, may not necessarily apply for tomorrow. This paper covers various approaches taken to attempt to predict the stock market without extensive prior knowledge or experience in the subject area, highlighting the advantages and limitations of the different techniques such as regression and classification. We formulate both short term and long term predictions. Through experimentation we achieve 81% accuracy for future trend direction using classification, 0.0117 RMSE for next day price and 0.0613 RMSE for next day change in price using regression techniques. The results obtained in this paper are achieved using only historic prices and technical indicators. Various methods, tools and evaluation techniques will be assessed throughout the course of this paper, the result of this contributes as to which techniques will be selected and enhanced in the final artefact of a stock prediction model. Further work will be conducted utilising deep learning techniques to approach the problem. This paper will serve as a preliminary guide to researchers wishing to expose themselves to this area