Stock trading using fuzzy neural networks

• Main Author: He, Guangxu
• Other Authors: Wang Lipo
• Format: Final Year Project
• Language: English
• Published: 2016
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: http://hdl.handle.net/10356/67491
• Institution: Nanyang Technological University

Stock trading can be generally divided into two types – fundamental analysis and technical analysis. Fundamental analysis is based on the financial position and the prospect of the company and the industry, while technical analysis uses the price movement and characteristics of a security to predict its future price movements. Neural networks and its applications in the financial market have been popular due to their capabilities in handling nonlinear relationships. Therefore, many types of research have been done to apply the neural network techniques into the financial market to propose different neural network forecasting models based on technical analysis of stock trading. This final year project (FYP) presents a neural network and fuzzy system based methodology to forecast index/stock close price and trend. Firstly, the results of one published paper, An Adaptive Network-Based Fuzzy Inference System (ANFIS) for the Forecasting – The Case of Close Price Indices by Ilija Svalina, Vjekoslav Galzina, Roberto Lujic ́, and Goran Šimunovic ́[1], were reproduced. In this published paper, the adaptive neuro-fuzzy inference systems (ANFIS) was used to test on CROBEX® index close price. Secondly, some modifications were made to improve the results of the published paper to get a 21% improvement. Thirdly, a newly designed algorithm was created by using ANFIS (genfis1: Grid Partition), ANFIS (genfis2: Subtractive Clustering), ANFIS (genfis3: FCM Clustering), Radial Basis Network (RBN), and Nonlinear Autoregressive Neural Network (NARXNET). One day ahead and multiple days ahead predictions can be achieved by using the newly designed algorithm, in addition, seven indexes and six stocks were used to prove the accuracy and reliability of the newly designed algorithm.