Stock prediction using support vector regression

• Main Author: Manasa, Abburu.
• Other Authors: Wang Lipo
• Format: Final Year Project
• Language: English
• Published: 2013
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; DRNTU::Engineering::Computer science and engineering::Computing methodologies::Pattern recognition; DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/54528
• Institution: Nanyang Technological University

This report depicts the work done by me, Abburu Manasa, as a contribution towards my Final Year Project (FYP) on Stock Prediction Using Support Vector Regression, under the supervision of Associate Professor Wang Libo. The success of any portfolio depends on the stocks selected, which in turn requires a robust and accurate system to anticipate the future performance of stocks. Several attempts have been made time and again to predict future stock prices; ranging from conventional, linear statistical models to latest developments in Artificial Neural Networks (ANN) like Back Propagation (BP) that can handle noise, instability and non-linearity. Recent developments in the field of computational finance have led to the proposal of a novel technique called Support Vector Machine (SVM) put forth by Vapnik and his co-workers. SVMs can be used for both classification and regression purposes and for our research we mainly focus on regression. Support Vector Machine for Regression (SVM-R) has been proven to be more advantageous than BP when it comes to handling financial data. It needs fewer parameters to learn and has the potential to handle noise, randomness and chaos as will be discussed in this paper. For experimentation purpose, two data sets representing TAIEX (Taiwan Stock Exchange Market Weighted Index) and RELIANCE (Reliance Industries Limited) data have been used. The technical indicators for this data were calculated manually. This data was then put into WEKA software to select the best attributes for evaluation purposes using attribute evaluators. The data set was then put through SVM for regression algorithm and back propagation algorithm. Through trial and error method the parameters were selected. Using the best suitable attributes and parameters, results were generated. These results were compared based on a few error estimating parameters and the optimal conditions were suggested. The results were compared to results generated using BP and an inference was drawn as to why SVM-R is a better performing alternative to financial market data prediction as compared to BP. The software studied include: WEKA, LIBSVM, DTREG, O-SVR etc. The various parameters include open price, close price, high price, low price, adj.close, volume, Moving Average (MA) for a 15 day period, Exponential Moving Average for a 15-day period (EMA), MACD (Moving Average Convergence Divergence), Rate of Change for a 10 day period (ROC10), Relative Strength Index (RSI), Bollinger bands (BB), Volatility, On Balance Volume (OBV), Relative Difference in Percentage (RDP) and Chaikin Money Flow (CMF). The performance evaluators include Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Relative Absolute Error (RAE), Root Relative Absolute Error (RRAE), Directional Symmetry (DS) and Weighted Directional Symmetry (WDS). A literature review was conducted to understand previous work performed in this area and use this work as a basis for our experimentation. Gaps to be filled were identified and suggestions for future improvements have been made.