Network-based Financial Data Analysis And Mining

Econophysics is an interdisciplinary research field applying theories and methods originally developed by physicists to solve problems in economics and finance. It has attracted considerable interest to investigate economic processes and financial market. At the same time, complex network as an effective and powerful tool can describe a wide range of real-life systems. However, exploring financial market with complex network is recently at an initial stage, and there even are no enough empirical studies. This thesis based on complex network analyses topological characteristics of futures trading networks, trailes network evolution, establishes model, explores super traders’ behaviors and attack resilience, and with data mining examines collusive behaviors and price transmission. We hope to provide a help in understanding financial market and its development law, and proposing a thought for analysis and policy decision of market regulators to prevent extreme market and systematic risk.1. Construction of futures trading network is introduced, and its topological characteristics are comprehensively analyzed. First, we use real trading records to construct futures trading networks, in which nodes are traders, and two nodes have an edge if two corresponding investors make deals with each other. Then, we conduct a comprehensive statistical analysis on the constructed networks. Empirical results show that the networks exhibit features such as scale-free behavior with interesting odd-even-degree divergence in low-degree regions, small-world effect, hierarchical organization, power-law betweenness distribution, disassortative mixing, and shrinkage of both the average path length and the diameter as network size increases.2. Behavior characteristics of super traders in futures markets are analyzed, and attack resilience of evolving futures trading networks is investigated. We explore empirically behavior characteristics of super traders, rich-club phenomenon and the response to targeted attacks of futures trading networks. The results first show that super traders present distinct features from common traders in trading activities, and play a dominant role in some statistical behaviors. Then the networks exhibit rich-club structure clearly that super traders intensely interconnect each other. Finally we find that the networks show robustness against the removal of super traders under static attacks, but for targeted attack of super traders absent from trading activities, three networks present different responses:extreme vulnerability for copper network, no influence for rubber network, and slight impact for the whole market network.3. Evolution of futures trading network is completely trailed; network link dynamics is revealed and its evolutional model is proposed. With real trading records, we study evolutionary route of futures trading network. We find that the network progressively reaches a steady state after an initial volatile stage, and the giant component dominates the network’s development throughout. During evolution process, the network continuously exhibits scale-free behaviors in degree and strength distributions, small-world effect and disassortative mixing. We disclose link dynamics at the microscope level as the driving forces of network evolution. Depending on the fact that more active investors make more transactions, the activeness model is proposed successfully to reproduce the results of the empirical observations.4. A method to detect potential collusive cliques in financial markets is proposed. A method is proposed to detect the potential collusive cliques involved in an instrument of future markets by first calculating the correlation coefficient between any two eligible unified aggregated time series of signed order volume, and then combining the connected components from multiple sparse weighted graphs constructed by using the correlation matrices where each correlation coefficient is over a user-specified threshold. Experiments conducted on real order data show that the proposed method can effectively detect suspect collusive cliques, which have been verified by senior futures experts.5. Two methods including price transmission graph and trend influence are proposed, and with them, price transmission and interaction among the world’s three leading copper futures markets are investigated. With correlation function between any two price series of three exchanges, a price transmission graph is established. The graph illustrates that LME always leads the global copper futures market, following by COMEX and strongly affecting SHFE. Although SHFE in price deviates from LME, there is not a leader-follower relationship in intraday session. With trend influence method to examine price interaction between them based on the high frequency price series, the results uncover that there is an apparent interaction of intraday price changes, and a relatively strong influence on SHFE from LME. An additional discussion on overnight price changes and trading session suggests that too short trading session of SHFE weakens its expected position in the global market.