Data Mining On The Burst Behavior Of Stock Market

By employing the basic theories and metrics of statistical physics,non-linear complex systems and complex networks,econophysics investigates the scientific issues in the financial systems,such as the fluctuations of the stocks' return,the correlations among stocks,etc.Data-mining on the burst behavior of stock market is one of the most important research topics of econophysics.This thesis proposes an algorithm to detect the burst behavior based on the candlestick graph.It also analyzes the features and dynamical evolutions of burst behaviors in the worldwide nine major stock markets.The main results are as follows:1.The features and general patterns of burst behaviors in the worldwide nine major st.ock markets have been analyzed.Based on the candlestick graph,an al-gorithm for detecting the burst behavior of financial time series is proposed.It is used to detect,the burst behavior of all constit,uent stocks in the nine major stock markets.Results show that,during the burst behavior period,the stock price series exhibit,three trends.The time interval of all adjacent,burst behaviors is statistically analyzed,it indicates that the burst behavior is related to the periodicity,such as the monthly effects.By comparing the cumulative distributions of the burst behavior duration time of the nine major stock markets,it is found that the stock market in developed countries is more stable than the emerging stock market.The price distri-bution of the burst,behavior int.erval approximately obeys the power law distribution,while the price return distribution shows a sharp fat-tail distribution.By counting the number of constituent stocks whose price is higher than the average price of the constituent stocks during the burst behavior,it is found that the higher the price of the composite index,the larger number the const.ituent,stocks whose price will rise.2.Based on the occurrence time sequence of t.he burst behavior of the con-stituent stocks and composite index,the constituent st,ocks can be divided into three categories:Pioneer stocks,Synchronizer stocks and Follower stocks.It is found that over a relative longer period of time,some pioneer stocks'burst behaviors always oc-cur earlier than those of the comprehensive index.The occurrence of burst behaviors of these pioneer stocks provides certain hints on the prediction of the behavior of the composite index.Some follower stocks' burst behavior always occur later than those of the compositeindex.Thus the occurrence of burst behavior of the comprehensive index provides certain hints on the prediction of the behavior of some follower stocks.3.A PMFG complex network of stock markets is constructed,the evolution of structure and dynamics of the stock markets are analyzed.It is shown that during the burst behavior interval,different markets exhibit different trends,the network structure of the stock market in developed countries shows a partial contraction and a global expansion,then the trend of local expansion and global contraction.The network structures of stock markets in emerging countries has shown a trend of partial expansion and global contraction.By generating six edge weight sequences based on the categories of the constituent stocks,it shows that the edge weight sequence has anegative correlation.By analyzing the connections of the entire burstbehavior interval network,it shows that most of the edges are very fragile However,there are still a few links that are stable,they constitutes the backbone of the network,and there is a tendency of gathering among similar stocks.