Authors

Jinwen Sun

Type

Text

Type

Dissertation

Advisor

Stoyanov, Stoyan | Glimm, James | Kim, Young Shin (Aaron) | Xiao, Keli

Date

2017-12-01

Keywords

Applied mathematics

Department

Department of Applied Mathematics and Statistics

Language

en_US

Source

This work is sponsored by the Stony Brook University Graduate School in compliance with the requirements for completion of degree.

Identifier

http://hdl.handle.net/11401/78217

Publisher

The Graduate School, Stony Brook University: Stony Brook, NY.

Format

application/pdf

Abstract

Exploring the possibility of market shocks forecasting is a significant topic for both academia and practice in finance. Measured by innovations generated from conventional time series models, market shocks are being assumed to follow specific distributions in the extensive literature. However, inconsistency occurs all the time in the real-world data. In this thesis, we propose and then apply a mutual information-based ARMA-GARCH-Artificial Neural Network framework to predict the direction of innovations under a high-frequency scenario. We leverage on the strength of neural networks in addressing complex pattern recognition problems. We study performances of two variable/feature selection techniques based on mutual information. Moreover, we conduct a series of comprehensive tests based on U.S. stock market high-frequency data to validate the effectiveness of our framework. | 103 pages

Share

COinS