Z/Yen, in co-operation with Sun Microsystems, the London Stock Exchange and four leading brokers, has just completed a large research study to trial PropheZy, a commercial application of a support vector machine, as an anomalous trade detector. Using three months of data comprising over 190,000 trades with a value of over Â£54bn in order to predict a fourth month, the objective was to see if PropheZy could predict the likely price range of a trade (specifically one of twenty price movement bands, based on a logarithmic scale).
The project proved that PropheZy successfully predicted price movement bands. Setting the level of acceptable accuracy at âwithin 0 to 4 bandsâ out of 20, PropheZy was able to predict over 50% of the trades price bands acceptably. Using these predictions, it was possible to set a level for best execution anomalies using price band prediction differences. If trades executed at âbestâ price (or better) are excluded from the anomalous trades, the number of anomalies is reduced to approximately 1%. Participating brokers concluded that the system was providing trades worthy of investigation:
âThis system would be a great way of seeing a small number of âodd lookingâ trades that we could check - the fact that the same principles could be applied to fixed income and other instruments makes it particularly interestingâ.
âI was fascinated to see the selection of trades that this system identified â there were good reasons why all of them traded at the prices shown but they were just the sort of trades that we should have been looking atâ.
The instruments covered by this system could be extended, with further testing, to derivatives, commodities, foreign exchange and other markets. Exchanges might wish to provide an automated, centralised compliance service for their members. During the course of the project the team built a prototype âCompliance Workstationâ combining a number of tools (PropheZy, VizZy, FractalIntelligence and Decisionality within an Excel framework) that spots anomalies and visualises them, allowing a compliance officer to âdrill downâ for detailed examination with an appropriate âaudit trailâ.
Professor Michael Mainelli, Chairman of Z/Yen Limited, said:
âThis projectâs results are very exciting because they show that the automated sifting of trades can identify anomalies, thus reducing the costs of complying with MiFID while simultaneously increasing the effectiveness of the compliance functionâ.
The research is due to be published in June 2006 in the Journal of Risk Finance in two parts:
â¢ âBest Execution Compliance - Need For New Techniques For Managing Compliance Riskâ
â¢ âBest Execution Compliance Automation - Towards An Equities Compliance Workstationâ