Using Big Data to Combat Big Risk

Everyone is talking about the use of "big data" these days and so now is a good time to reflect on the potential uses of big data by different industries and policy makers to solve some of their long standing issues. Here we look at how banks and regulators can use the principles of "big data" to solve some problems -- like how to identify traders who are taking undue risks, or investment salesmen who are fronting a Ponzi scheme.

First, banks should leverage data to expose the objective reality of different traders’ performance. Coaches in baseball, memorably portrayed in book and film in Michael Lewis’s Moneyball have mined data to expose when long-held views about the value of certain types of players do not coincide with the reality. Mining trades’ data could similarly be done to challenge similarly held views about the value and consistency of different traders. It would be very useful to counteract with actual data the halo effect on occasion bestowed on certain traders by past heroic trading exploits. Such heroism, for example, achieved by successfully taking high levels of risk in a tough market, is often rewarded by supervisors with latitude to take greater risks. In certain cases, as seems to have been the case in the London Whale episode, such latitude can be disastrous.

A disciplined data driven approach would serve to assess traders performance on a more objective basis relative, for example, to contextual factors such as: amount of risk taken relative to reward, performance of market benchmarks, volatility of returns over a longer run period. Such data by providing a more objective basis for performance assessment would enable better calibration of pay, risk limits, and trader mandates and would lay bare the reality behind a trader's reputation which may or may not have been fairly earned. Solid data analysis of ongoing performance should help to separate out myth from reality and help to prevent encouragement of excessive risk taking.

Objective data analysis of the type that might enable managers to identify the truly high and consistent performers is hard to do, however, when the data upon which it is based is bad. How to evaluate the true performance of Lance Armstrong when we now know he artificially inflated his performance? This takes us to the second use of big data. Can we identify manipulative or cheating patterns of behavior? Now this is a field of great promise because, underlying many of banks' top risks, are patterns of behavior that are hard to detect but that can lead to disaster. Identifying the hallmarks of such patterns would be a major advance. Let’s look at three areas: rogue traders, insider traders and fraudulent investment schemes.

The rogue trader of course, as has been demonstrated several times, can work at the margins for several years. He typically starts off by taking relatively small unauthorized risks, and generates profits at first that are set aside in a non-active account to smooth out, through inter-account transfers, emerging losses in the main trading account. As the losses grow, the trader is forced to put on riskier, larger positions, all unbeknownst to management and supervisors. His patterns of behavior include: certain cash transfers between accounts; trades with certain possibly fictitious counter-parties; trades with unusual settlement periods; large numbers of cancelled trades; and failure to take vacation. These patterns, however, appear as isolated data points in a sea of daily processed data that includes thousands of other, benign, data points. Developing an ability to draw out the patterns, to connect the dots between these different behaviors, can help filter out the risky behaviors from the benign.

[Finding Alpha in the New Normal ]

Second, insider traders work across a network of contacts, activities, to execute trades in anticipation of an event that will give rise to a lift or a dent in the stock price. Any types of trades that fall outside the normal domain or expertise of a trader that are executed ahead of a market announcement should be fertile ground for analysis.

Third, big data could be part of an enhanced tool set to catch the people behind the next Ponzi scheme. The Commodities Futures Trading Commissioner, after the failure to identify the fraudster behind the Peregrine Ponzi scheme, has talked about mining the data of futures brokers – this could include patterns of asset transfers for example - to become more effective in policing segregated customer assets. There are other tools that can be put into the hands of risk managers, investors and those who conduct due diligence on traders and investment managers. Analysis of speech patterns by James Pennebaker of Texas University ("the Secret Life of Pronouns"), for example, showed that liars tend to use more upbeat words like "pal" and "friend" but fewer excluding words like "but," "except" and "without."

Can access to this type of knowledge help those who are seeking to identify potential fraudsters and rogue traders? Perhaps. These are just three areas that could be advanced for risk management purposes with effective data mining tools and techniques. Will banks and regulators succeed in making such advances or will we be hearing about the next Madoff, Adoboli or Iksil in 12 months’ time? Only time will tell.