Please use this identifier to cite or link to this item:
https://scholarbank.nus.edu.sg/handle/10635/36528
Title: | New Modeling Methods and Applications to Financial Markets | Authors: | QIN QIN | Keywords: | Stock Market, Modeling, Linear regression, Neural network, Random Forest, Feature Selection | Issue Date: | 24-Aug-2012 | Citation: | QIN QIN (2012-08-24). New Modeling Methods and Applications to Financial Markets. ScholarBank@NUS Repository. | Abstract: | In this thesis, new modeling methods and applications to financial markets are studied. Five topics are explored: 1. Multiple-indicator model The return and the volatility of Shanghai Composite Index are predicted based on regression model and neural network model using the daily and weekly data. 2. Multiple-time model Multiple-time models are proposed based on the combinations of different time frames data. The Shanghai Composite Index (SH000001) as well as 3 representative stocks (Petrochina Co., Ltd. (SH601857), Industrial And Commercial Bank Of China Limited (SH601398) and China Vanke Co.,Ltd. (SZ000002)) are tested in the models. Some useful findings are obtained. 3. Multiple decision trees We combine the technique of gradient boosting and the technique of random forest to propose a new method: Gradient Boosted Random Forest. It is tested based on the Singapore stock market. The results are compared with the ones produced by linear model and random forest technique. 4. Multiple neural networks with randomized algorithms A new method ? multiple neural networks with randomized algorithms ? is proposed by combining the technique of random forest and the technique of neural network. Illustrative example as well as the popular examples and practical examples are tested in the new method. The results show that the proposed method can work well on the noised data. 5. Multiple-input selection Four input selection methods are firstly investigated. Besides, we propose a new input selection method, which is an improvement on the spearman?s rank correlation coefficient method and correlation-based feature selection method. The data sets obtained from artificial linear system and artificial nonlinear system are employed to test these methods. The simulation results indicate that different input selection methods differ in the power of feature acception and feature rejection. | URI: | http://scholarbank.nus.edu.sg/handle/10635/36528 |
Appears in Collections: | Ph.D Theses (Open) |
Show full item record
Files in This Item:
File | Description | Size | Format | Access Settings | Version | |
---|---|---|---|---|---|---|
Qin Qin.pdf | 2 MB | Adobe PDF | OPEN | None | View/Download |
Google ScholarTM
Check
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.