Cross-market behavior modeling

Publication Type:
Issue Date:
Full metadata record
Files in This Item:
Filename Description Size
01front.pdf104.52 kB
Adobe PDF
Thumbnail02whole.pdf1.92 MB
Adobe PDF
During the 2007 global financial crisis which was triggered by subprime borrowers in the US mortgage markets, strong market linkages were observed between different financial markets. The sharp fluctuations in the global stock market, commodity market and interest market illustrate some of the coupled behaviors that exist between various markets, namely the crisis effect is passed from one market to another through couplings. Here coupled behaviors refer to the activities (e.g. changes of market indexes) of financial markets which are associated with each other in terms of particular relationships. Therefore, a good understanding of coupled behaviors is of great importance in cross-market applications such as crisis detection and market trend forecasting. For instance, if the coupled behaviors are properly understood and modeled, investors can predict financial crisis and avoid the big loss, by detecting the changes of coupled relations between financial crisis period and non-crisis period. However, understanding and modeling coupled behaviors is quite challenging for following reasons: (1) The various coupled structures across financial markets (e.g. coupled relations between different types of markets, and coupled relations between the same type of market in different countries) bring challenges in terms of understanding and modeling them. (2) Various types of couplings. The typical forms are intra-coupling, inter-coupling and temporal-coupling. (3) The complex interactions between markets are driven by hidden features which cannot be observed directly from observation/data. (4) Different applications in cross-market analysis lead to the consideration of input factors/variables selection. All of these challenges the existing methods for cross-market analysis, which can be roughly categorized into two types: time series analysis represented by Logistic regression, Autoregressive Integrated Moving Average (ARIMA) and Generalized AutoRegressive Conditional Heteroscedasticity (GARCH) models. Model-based methods explore machine models such as Artificial Neural Networks (ANN) and Hidden Markov Models (HMM). The main limitations lie in their deficiencies: (1) Existing approaches are usually focused on the simple correlations of the cross-market, rather than coupled behaviors between markets. (2) State-of-the-art research work is usually built directly from the observation/data. Hidden features behind the observation/data are often ignored or only weakly addressed. (3) Some approaches follow assumptions that are too strong to match real financial markets. Based on the above research limitations and challenges, this thesis reports state-of-the-art advances and our research innovations in understanding and modeling complex coupled behaviors for the purpose of cross-market analysis. Chapter 3 presents a new approach, called Coupled Market Behavior Analysis (CMBA) for financial crisis detection. This caters for nonlinear couplings between major indicators selected from different markets, and it detects different coupled market behaviors at crisis and non-crisis periods. Chapter 4 seeks to overcome the limitations of most current methods which conduct financial crisis forecasting directly through observation and overlook the hidden interactions between markets. In this chapter, Coupled Market State Analysis (CMSA) is presented to build forecasters based on coupled market states instead of observation. Chapter 5 reports a new approach for market trend forecasting by analyzing its hidden coupling relationships with different types of related financial markets. Chapter 6 proposes Hierarchical Cross-market Behavior Analysis (HCBA) to forecast a stock market’s movements, by exploring the complex coupling relationships between variables of markets from a country (Layer-1 coupling) and couplings between markets from various countries (Layer-2 coupling). In addition, Chapter 7 designs a Coupled Temporal Deep Belief Network (CTDBN) which accommodates three different types of couplings across financial markets: interactions between homogeneous markets from various countries (intra-market coupling), interactions between heterogeneous markets (inter-market coupling) and interactions between current and past market behaviors (temporal coupling). With a deep-architecture model to capture the high-level coupled features, the proposed approach can infer market trends. In terms of cross-market applications (i.e. financial crisis detection and market trend forecasting), our proposed approaches and frameworks for modeling coupled behaviors across financial markets outperform state-of-the-art methods from both technical and business perspectives. All of these outcomes provide insightful knowledge for investors who naturally seek to make profits and avoid losses. Accordingly, cross-market behavior modeling is a promising research topic with lots of potential for further exploration and development.
Please use this identifier to cite or link to this item: