When Model Risk becomes Model Fraud

[AZPalta]

Been following this story for a long while given that Barr Rosenberg was a major contributor to Quant Asset Management as an industry and a field of thought. (Capital Ideas by Peter L. Bernstein for those interested).

http://www.pionline.com/article/20110922/DAILYREG/110929952

Punishment seems justified, given the crime was hiding the mistake not just using a bad model .

 

[Andy Nguyen]

The SEC complaint against Barr Rosenberg:

http://www.sec.gov/litigation/admin/2011/ia-3285.pdf

 

[alain]

Been following this story for a long while given that Barr Rosenberg was a major contributor to Quant Asset Management as an industry and a field of thought. (Capital Ideas by Peter L. Bernstein for those interested).

http://www.pionline.com/article/20110922/DAILYREG/110929952

Punishment seems justified, given the crime was hiding the mistake not just using a bad model .

Barr Rosenberg was the founder of BARRA.

 

[Andy Nguyen]

MCSI Barra

 

[alain]

MCSI Barra

I think it's called just MSCI nowadays.

 

[Sergei P]

Actually, SQA has put up a program (coming up in NYC on April 26, 2012) on Barr Rosenberg and repercussions of this SEC case on the quant industry. Jeremy Baskin, CEO of AXA Rosenberg, is one of the panelists.
http://www.sqa-us.org

 

[DominiConnor]

One way this case is interesting is that he appears to have been caught by someone in his own firm, something that is quite common for low-mid level offences and he did something that was pretty flagrant and had a paper trial of him trying to cover up. From that point he was a dead man walking.

It's really hard to see how a regulator could catch him, firstly because they don't often examine models, but also because they can't, because they are very rarely smart enough.

Model design and choice is a matter of judgement and unless you're dumb enough to put in an email "don't tell anyone about the hole", it's hard to see you being caught. Often for simplicity one chooses volatility to be a constant determined by option prices, sometimes you add jumps, what distribution do you choose for these ? None have universal answers, that allow a 3rd party to say "you chose that because it gave the result you wanted"