Skip to main content

Vikram Pandit Wants To Crowdsource Bank Risk Management

  • Author:
  • Updated:

A thing you might want is for investors to be able to understand the financial situation of the companies they invest in. Traditionally, that is a thing that many people want, anyway.* Much of our system of corporate finance is dedicated to that and it mostly works okay.

A place where it breaks down a bit is in financial institutions. Because big financial institutions more or less take shareholder money, leverage it 10 or 30 times, and invest it all in a large and ever-changing mix of mark-to-market assets, some of which they mark themselves. Then they tell you things like "our assets have a current expected value of around X, with a daily variance of around Y" and since they're sporting they also give you some sort of rough breakdown of what classes those assets fall into and stuff. This does not give you precise confidence about what those assets are worth today or what they'll be worth in a week. And you can't really find out much granular detail about the assets, because disclosing them all would be a competitive problem and/or just take too long / make your eyes glaze over. If you're lucky maybe the banks disclose in some useful form actionable information about whatever you're currently worried about, but you're probably worried about the wrong things anyway.

So you do the best you can, and rely on external sources, like ratings agencies, who might know more than you, maybe, sometimes, or like Warren Buffett. Or you rely on government oversight to keep your financial institutions more or less solvent. But regulators, too, need some sort of heuristic for figuring out what assets are risky and how risky they are. After all, a big part of their job is regulating those risks, by doing things like setting capital requirements. It turns out that this is hard. So they sometimes outsource that job to ratings agencies. That doesn't always work. Then they get all "we're going to stop outsourcing risk regulation to ratings agencies." That doesn't always work either.

Vikram Pandit has his own idea and it's pretty neat:

Regulators should create a “benchmark” portfolio and require all financial institutions, not just banks, to measure risk against that. The benchmark portfolio would not actually exist on the balance sheet of any one institution. Rather, it would be a collection of real investments that stand in for the kinds of assets that most financial institutions actually hold at the time. What is more, its contents would be 100 per cent public.

Institutions would be required to produce, on a quarterly basis for that benchmark portfolio, a hypothetical loan/loss reserve level, value at risk, stress-test results and risk-weighted assets. Right now these measures are run only against an institution’s actual portfolio and only a limited number of the results are disclosed. Worse, those results have no common frame of reference. The benchmark portfolio would supply that needed frame of reference.

So, yeah, let's do this. As a way to help investors understand bank valuation and risk tolerance it's - well, all the commentary on this idea has been along the lines of "it couldn't hurt," and that seems about right. Like, why not do it? It would at least have some potential entertainment value when someone like gets a sign wrong and screws up their benchmarking. It'll probably be Citi.

Of course like any set of specific rules, it could be gamed. Here the game is of the form:
1. Develop models that (a) make the benchmark portfolio look very risky but (b) make portfolio P of profitable but very risky things look very safe.
2. Buy portfolio P.

Right? Adding the specification "make the benchmark portfolio look riskier than it should" to your current specification of "make our portfolio look less risky than it should" probably doesn't up the degree of difficulty all that significantly.

Still, there may be a broader methodological takeaway here. If you're a regulator looking to figure out how to risk-weight assets, say, you have some choices:

(1) Use credit ratings. This is pre-Dodd-Frank U.S. practice. This has fallen into disfavor.
(2) Let each bank decide a risk-weighting for itself. This is Basel. Bold prediction: This will fall into disfavor.
(3) Use some sort of market determinant - like, use spread cutoffs to sort of ratings-bucket bonds, or use market implied vols to calculate VaRs, etc. This either devolves into (2) (if the banks do the math) or requires regulators to be better at it than bankers which is trubs.
(4) Other (what?).

Pandit's idea gestures in the direction of a useful answer for option (4), which is to use a more crowdsourced set of bank determinations to decide things like risk weightings. If every bank risk-weights Asset X because they all have some greater or lesser exposure to it, and most of them weight it as a BBB-looking asset, then use that as a benchmark in setting capital requirements for that thing - and increase capital requirements for anyone who's weighting it as AAA. Sure each bank has incentives to make its own assets look less risky (and reduce its capital requirements), but it also has incentives to make its' competitors assets look more risky (and increase their capital requirements). In the aggregate, these things might wash out. The trick, as in so many things, is to have a market where some people have opposite incentives. Pandit's proposal is can be gamed, and is probably imperfect, but it gets that basic instinct right. So, y'know. It can't hurt.

Apples v apples – a new way to measure risk [FT]

* But not always! It's fun to read how investing, or whatever you'd call it, worked in, like, the 1890s, when people were all "hey, I have a new venture, perhaps you would like to invest in it," and you did, and asking for audited financials first was viewed as very uncool. That would never happen now. Further back in the day, the most awesome blind pool ever was "a company for carrying on an undertaking of great advantage, but nobody to know what it is." And now you complain about Groupon! Anyway.