SocGen, Facebook, and historical data analysis
Back in my formative years, during the tail end of the S&L crisis, I spent quite a bit of time working on financal software, including asset-liability management, risk management, and pool selection for the securitization of assets.
Although I’m a ‘data’ guy, for a layman I have a fairly good understanding of risk management in the financial industry. A side effect of that understanding is that I’m been following the ongoing risk management fiasco with more than a passing interest. (The media might call it a ‘credit crunch’ or a ‘subprime contagion’, but as far as I can tell, it all boils down to a lack of attention to risk.) It’s also turning out to be a good source of entertainment.
Last week, SocGen opened the latest chapter in the saga. While preparing to announce profitable results, despite subprime related writedowns of € 1.1 billion, they discovered a â?¬ 50 billion (US $73 billion) position in the European futures market. A position that was somewhat larger than the capitalization of the bank. The apparent culprit is a ‘rogue trader’ , and the results of unwinding the positions result in a trading loss of â?¬ 4.9 billion (US $7.2 billion.) Not a fun conference call for them.
Usually, the next steps in a story like this follow a predictable pattern of complaining aobut the rating agencies, and trying to figure out what went wrong with the risk managment process. And lots of data crunching.
Of course, theres also a lot of interest in the accused trader, Jerome Kerviel, so the media start digging to find out more about this person, who almost brought down an entire bank single handed. Somewhere in the blitz of articles, this Guardian story caught my attention:
When Jerome Kerviel in Facebook’s Banque Société Générale group was named in the media as the man who has cost French bank € 4.9bn (US $3.7bn) he still had 11 friends.
One hour after his name was published on news sites, four friends had already deserted the 31-year-old.
At the time of writing, the list has shrunk further to just four friends – providing more fodder to the Facebook refuseniks who question whether someone linked to a social networking page is really worthy of the name “friend”.
Now, as as data guy, this is starting to get interesting. Tracking facebook friends is a new angle. Apart from Jerome’s shallow, fleeing Facebook friends, I have to wonder – is there any value in historical data from systems like Facebook?
Facebook stores history, in the form of activities that appear as a news feed. There is some filtering for type of activity, and which friends I see. But within Facebook, it’s like a stock ticker – I look, something changed, ok, it’s gone. I don’t know how much history is stored within the system; I see only the latest changes.
As Tim O’Reilly keeps reminding us, ‘data is the next Intel Inside.’ If I had historical information on, say, Jerome’s count of Facebook friends, what would I see ? Are there trends that are relevant? Are there tops and bottoms? Analytic indicators comparable to overbought and oversold?
Can we detect behavioral changes based on activity in a social networking system like Facebook or MySpace – a sudden change in number of friends ? Or LinkedIn – a recent blitz of recommendations correlated with an employers bad quarter?
It would take a large data set to come up with any reliable analysis, and the owners of these systems have that data in the aggregate already. But with API’s and tools like SnapLogic, it’s also possible for anyone to start collecting and analyzing this type of data, and to see what they come up with.