Behavox uses big data to monitor rogue traders

Peter Lee
Published on:

Fintech start up aims to spot market rigging, using clues in how traders talk, not what they say

After all the market rigging scandals in which traders from different banks have colluded to fix Li and every other conceivable bor, the forex Fix, gold and who yet knows what other benchmarks, traders can be under no doubt that their employers are now keeping a very close watch on them. Every individual with which they communicate by phone or email has to be identified and approved; banks are increasingly preventing traders from bringing personal mobile devices into work to take conversations offline and banning them from chat rooms, while traders know that every conversation is recorded and monitored for red flag words that might indicate wrong doing. Ask your pal at another bank to help you fix that new TV at home and those people suddenly appearing at your desk will be compliance suggesting you come with them for a chat.

Behavox, a year-old fintech company, founded and led by Erkin Adylov, a former Goldman Sachs analyst and GLG fund manager, and now a part of the Level39 accelerator in Canary Wharf and a member of Innovate Finance, aims to take this surveillance to a new level with sophisticated behavioural analysis. There’s a limit to what compliance officers can detect just from reviewing what traders say to each other – they often speak in code anyway. As important can be how traders talk to each other and also new departures in the way they talk.

Erkin Adylov
We’re looking for changes in behaviour and anomalies in how people communicate

Erkin Adylov

"We’re looking for changes in behaviour and anomalies in how people communicate," Adylov tells Euromoney. "So for example, an important thing to notice may be how often a trader laughs on the phone. Our system establishes a true picture of the relationship between a trader and his or her counterparts. So, if a trader laughs more often in conversation with a person, uses more slang in emails to them or swears more, that is a good indication of how close they are. As well as being a designated counterparty, that person – maybe working at another firm – is clearly the trader’s mate. The trader may also be good mates with his risk manager. Does that represent a conflict of interest to bear in mind?" The difficulty for any trading-desk manager is that each trader may have 400 to 500 approved contacts. If he’s running a desk with 20 traders, that’s communications with 10,000 contacts the manager has to keep an eye on. "You need a compliance system to do that," Adylov says.

Behavox doesn’t just analyse undisclosed friendships. Its systems also quickly identify who are the most important contacts for a trader as revealed, for example, in how prepared a trader is to take calls even at the very busiest times of day and whose emails a trader never leaves unanswered for long, even if communications are relatively infrequent.

It uses 50 different metrics to gauge closeness and 50 to gauge importance of contacts for voice and the same again for email and chat. The key is to monitor deviations from the average way in which a trader communicates with all counterparties and changes to the way a trader communicates with particular counterparts.

"Why is this trader suddenly whispering to this individual that is not even on his approved contact list? Oh, the trader is married and this is his new mistress. That doesn’t matter," Adylov says. "But why is he no longer speaking to this contact that used to be so important to him? Human relationships don’t usually change nature very quickly. Or why has that mate whom he used to communicate with only infrequently suddenly become the most important contact whose messages get returned instantly? Are they friends who are suddenly passing inside information? Is there a new pattern to the trader’s use of his security pass? Is he going outside the building to take calls from this person on his own mobile?" It can be revealing to monitor changes in communication and then overlay that with analysis of trades put on before and after talking to a certain contact, especially if those trades show a pattern of getting in or out of a security before it falls or rises.

Like many good ideas Behavox came together by accident. The founders had done four and a half years of research and development into big data analysis of human conversations they thought was pretty interesting with no clear sense of its commercial value until Adylov saw the potential application to trader surveillance.

For the system to work, Behavox has to be able to apply its analysis across phone networks, office email, Gmail accounts accessed from work, Bloomberg chat and then overlay that with individual trading records. It has approached three banks to take the system, and all three have signed up for pilot studies, typically back-testing huge data dumps of internal records requested by regulators on previous market-rigging investigations. So far, Adylov claims, it works through the data much faster than banks own systems, reveals more suspicious episodes worthy of investigation but also fewer false positives.