Support Vector Machines and Artificial Neural Networks: Assessing the Validity of Using Technical Features for Security Forecasting

James Di Padua, Dublin Institute of Technology

Document Type Dissertation

A dissertation submitted in partial fulfilment of the requirements of Dublin Institute of Technology for the degree of M.Sc. in Computing (Data Analytics) Sept. 2016.


Stock forecasting is an enticing and wellstudied problem in both finance and machine learning literature with linearbased models such as ARIMA and ARCH to nonlinear Artificial Neural Networks (ANN) and Support Vector Machines (SVM). However, these forecasting techniques also use very different input features, some of which are seen by economists as irrational and theoretically unjustified. In this comparative study using ANNs and SVMs for 12 publicly traded companies, derivative price “technicals” are evaluated against macroand microeconomic fundamentals to evaluate the efficacy of model performance. Despite the efficient market hypothesis positing the illsuitability of technicals as model inputs, this study finds technical indicators to be nearly as performant as fundamentals at forecasting the future prices of a security. Additionally, all model predictions were fed into an automated trading machine and evaluated against a simple BuyandHold, finding model performance at par with the passive BuyandHold investment strategy.