The CMCRC Data Mining research program is leveraging the extensive financial transaction data bases made available to the CRC by it’s industry partners to research optimised trading strategies for exchange traded securities and automated classification systems to identify anomalous trading behaviour for security market surveillance. The databases we are working with include ASX SEATS comprising full intra day order book data and Reuters data for over 200 world markets at the trade and quote level. These transaction databases are augmented by announcement databases such as ASX Signal G to explore the relationship between announcements and market reaction.

Our research on trading addresses intra-day trading strategies which take into account intra-day liquidity and volatility patterns and the influence trading has on other market participants, otherwise known as market impact. The objective of this research is to create an infrastructure whereby trading performance can be continuously measured and optimised taking into account prevailing market conditions. We are collaborating with our partners Credit Swiss First Boston Equities and ABN Amro to achieve this goal. Our research on surveillance addresses the reduction of classification error rates for detection of abnormal liquidity or volatility and unusual trading patterns amongst market participants. This research is being performed in collaboration with Smarts Pty. Ltd., a provider of stock exchange surveillance and broking compliance software.

Data mining is a process of discovering useful summaries of information from large amounts of data stored in databases, data warehouses, or other information repositories. These summaries may consist of patterns, associations, changes or anomalies occurring in both categorical and numerical data. Data mining aims to automate the process of discovering such relationships in the data and involves the tasks of:

• Data pre-processing: including data collection, data cleaning, data selection, and data transformation
• Data mining: A variety of statistical, combinatorial search and optimisation techniques for knowledge discovery which target large repositories of high dimensional data
• Post data mining: including pattern evaluation, model deployment, knowledge maintenance, and visualisation

The data mining research program facilitates a multi-disciplinary approach to financial data analysis combining the expertise of data mining specialists, software engineers, traders and market regulators.