Money Flows Between Securities: Network Analysis in a Stock Market

Application of network science to study various phenomena has increased in the recent years as for example networks have been used to model the effects of age to the spread of COVID-19, how the World Trade Web changed during the financial crisis of 2008 and to rank web pages. Networks have been used to study financial markets although research has focused mainly on interbank lending. The financial crisis of 2008 has been shown to have happened partly because of the financial industry, and the interbank lending networks showed early-warning signals of the crisis. In 2008 stock markets crashed and investors changed their allocations to different assets based on their views of the future. Therefore, money flowed between securities which can be modelled using networks, and network science offers tools to study the topological features. The goals of the thesis were to define money flow networks and to study possible changes in the networks during the financial crisis of 2008 in Finland. Money flows to securities have been used before as a technical indicator but the sources of the flows have not been incorporated to the analyses. Thus, a method for approximating money flows between securities from transaction data is defined which forms a network of money flows between securities. To compare the money flows, the absolute money flows are scaled using the largest money flow during the last 90 days between two securities. The networks of financial institutions and households are analyzed by ranking the securities based on their centralities and by analyzing changes in the z-scores of subgraph abundancies. The ranking of securities based on their centralities is used to find out if some securities are favoured or neglected during the crisis, and different centrality measures are used with statistical testing to ensure the results. The z-scores of subgraph abundancies are used to find general changes in the structure of the networks during the crisis. Since the subgraph counts are affected by the number of links and degrees of nodes, two different random graph models are used in calculating the deviations from the expected subgraph counts. The centrality rankings showed that large companies are more often in the top percentiles of the rankings while the bottom percentiles mostly do not have certain securities in them more often than expected. Finance institutes had more random ranking than households as in the daily networks households did not invest in smaller companies as often. The z-scores of subgraph abundancies had multiple observations and interpretations, but further analyses are needed to understand the changes better. Household networks had more complex structure than finance institute networks which may mean households had less similar opinions about the securities. Based on the analyses, the centrality rankings failed in finding securities with unusual money flows, even though differences between investing of finance institutes and households were found, and there were changes in the structure of the networks during the crisis. In the end, money flow networks should be studied further, and future analyses should incorporate additional market data