Date of Award
Fall 1997
Document Type
Thesis
Degree Name
Master of Science in Electrical Engineering (MSEE)
Department
Department of Electrical and Computer Engineering
Abstract
The stock markets represent a complex system of inter-related factors that are difficult to model using conventional statistical techniques. Forecasting stock market trends is a challenging and rewarding task. This has led to extensive research into stock market behavior and the development of analytical and artificial intelligence tools for stock trading. The most promising application of artificial intelligence in finance is the use of neural networks. This thesis is concerned with the application of neural networks for the generation of stock market timing signals.
We have examined different analytical approaches to stock market trading. For the generation of trading signals, it is necessary to recognize price trends and forecast impending trend reversals. Relevant performance metrics for an effective trading system are presented. The moving average technical analysis is performed on actual stock market data using a simulated one-point trading strategy and results of profitability analysis are studied. We have explored the application of neural networks using the corner classification technique for the development of daily trading systems. Results of the neural network trading strategy on actual stock market data show that the neural network approach clearly outperforms the moving averages technique.
The stock market data is extremely noisy. We have analyzed the effect of preprocessing of input data on the profitability of the system. A modular neural network system for learning different stock market dynamics is proposed and results are analyzed. A decision rule base, to combine neural network predictions and results of technical analysis, is proposed to include a measure of strength in trading signal generation. We also describe a system to model intra-day price fluctuations. Results show that an intelligent stock trading system can be developed by combining neural network based forecasting with an expert decision rule base.
Recommended Citation
Mehta, Asheesh, "Prediction of Timing Signals in Financial Series using Neural Networks" (1997). LSU Historical Dissertations and Theses. 8412.
https://repository.lsu.edu/gradschool_disstheses/8412