Predictive Analytics: JPMorgan rolls out a program to identify rogue employees before they go astray.
A really interesting article on Bloomberg about how JPMorgan are using predictive analytics to identify outliers in their trading organisation:
It looks like they’re taking many multiple inputs, comparing it to known patterns of rogue activity by back-testing to determine what activity is significant and then using that to predict future behaviour. At idax we’re working with a number of the big banks and I know of at least two other organisations that are doing something very similar, so no surprises there given the size of the fines.
However this raises a couple of questions in my mind. Firstly, to what degree are these tests successful. We all know that the problem tends to be false positives – you can find the outliers, but you find ten times as many other people too. Which begs the question: Is the major benefit of these exercises PR, or can you really find the bad guys. I guess we’ll find out.
But secondly, I’m guessing that they’re using a learning algorithm where you improve over time by adding more data and more business intelligence, referred to as “supervised” learning. It’s very effective but tends to be quite high maintenance. What we use at idax is, “unsupervised” learning, where you need no business knowledge and no back testing data. The advantages are: It’s much quicker to set up – hours rather than weeks; you get results straight away; there’s loads of high quality actionable information; and for access control, it doesn’t cost the earth.
NB: Bloomberg get a 9/10 from me on the article. Minus 1 for mentioning “Minority Report”. Can we please move beyond Tom Cruise.