Researchers are developing a software program that will correlate stock price movements with news story headlines as a way to detect insider trading.
The system is nicknamed Cassandra, or Computerized Analysis of Stocks and Shares for Novelty Detection of Radical Activities. It's being designed by the University of Sunderland in northeastern England.
The project recently received
There are a variety of software systems offered by private companies that are used by regulators to try to detect suspicious trades, but none look at news headlines, which often move markets, Addison said. However, false positives remain a problem, where the software thinks it's found indicators of insider trading when there are none, he said.
Insider trading is defined as trading stocks on information that's not in the public domain. Indicators can be when a major stockholder of a company sells or buys shares out of context or out of proportion to how the company's stock is normally traded or out of sync with news events.
Cassandra will look at a "time series," or a compilation of events for a particular company, such as internal changes or regulatory changes, Addison said. It will also look at share data along with news headlines published by wire services such as Bloomberg, Reuters and the Associated Press.
Cassandra's strength will be its ability to digest large amounts of data and then detect things that seem strange for further analysis.
"If we don't find significant news stories to explain away the change in the time series, the conclusion is that someone could be insider trading," Addison said.
Addison said researchers are about a third of the way through the project and have spent much time figuring out what the financial market would need from such a system. By June, they hope to have a working prototype that will be focused on data and headline analysis.
"We are keeping it as straightforward as we possibly can at the moment because we have a limited amount of time and a small budget," Addison said.
Cassandra will use modeling techniques to find out what is normal activity for a particular company. It will also use adaptive learning algorithms that can learn about a time series and then make generalizations of what should happen when new data is introduced. The algorithms will also remember over time what they've learned before, Addison said.
He doesn't want to reveal too much detail on Cassandra, as there are many talented computer programmers in the finance field who may try to think of countermeasures to hide their activity.
Cassandra has to have three traits: be simple to use, cheap to buy and offer good technical support, Addison said. Cassandra could take a couple of forms. One would be as a DLL (Dynamic Link Library) that companies could integrate into their existing financial analysis software tools, Addison said. Cassandra could also work well as a subscription Web service, he said.
The work now will focus on the prototype, and once it's illustrated that the concept works, Addison said they'll seek more funding. Plans also call for continuing to do more market research and talking to interested customer groups.
Once a market need is solidified, a more sophisticated development project can be undertaken, Addison said. However, it could be as long as three to five years to fully develop the system.