Smart technology predicts better

Click to follow
The Independent Online
LEADING City financial institutions are responding with enthusiasm to a new computer technology for making predictions that is based on trading histories in areas such as credit evaluation, investment portfolio management, stock market prediction and direct marketing.

Neural Networks, one of the new generation of Intelligent Systems programming techniques, is being developed by Professor Philip Treleaven, head of the computer science department at University College, London.

Its advantage over traditional computers is that it continues to learn on the job, interpreting the results of previous information, so that whenever new data is added, it adapts its strategy and comes up with the best course of action in each set of circumstances.

Proximity to the City means UCL is well placed for working with financial institutions, such as BZW, NatWest Markets and the Stock Exchange. The new system, Professor Treleavan says, is achieving 'a lot of success' in predicting the movements of prices and currencies.

According to Universe, a quarterly journal on innovation within the college, Neural Networks operates on the basis that patterns can be identified 'in what is seemingly a chaotic melee' of fluctuations. 'Once these patterns have been established, they can be used for market prediction.'

Professor Treleaven wants to see business needs give impetus to new technology. He and his 25 staff are working on Neural Networks in three ways. One is to provide firms with the free services of a pre-graduate or MSc student during their two- month project. This should be 'something interesting'.

Another method is to work with large clients on a one-to- one basis. The department builds the system and hands it over to the company, trains their staff to use it and provides follow-up. For BZW, it built a demonstrator system to track the movement of the FT-SE 100 index. The result, says Professor Treleaven, was between 20 and 30 per cent better than using the previous technology.

The third Neural Networks activity is through industry clubs. Each company pays from pounds 10,000 to pounds 15,000 a year in return for regular visits to the department, plus information on relevant new technology. It can then identify developments appropriate to its business.

The company either gets its staff to build a system or hires 'rocket scientists' to do so, working with UCL's computer science department.

Neural Networks is not restricted to financial services. Retail companies such as Thorn EMI and Radio Rentals use it for customer profiling. The next step is to target advertising outlets with greater precision.

There are also hopes of reducing direct mail. 'Companies know it's unsatisfactory and are trying to focus more on the customer,' Professor Treleaven says.

His department also co-operates on projects with European industry, including Philips Electronics, Siemens and Thomson. This seems appropriate as Professor Treleaven is also the Tory candidate for south-west London in next year's MEP elections.