Morgan Stanley: VaR Datapoint of the Day

Morgan Stanley (MS) has some of the most sophisticated risk-management systems on planet earth. And yet, somehow:

The company disclosed today that daily trading losses during the quarter exceeded the firm's trading value-at- risk calculation on six days during the quarter.

Needless to say, this is something which is not meant to happen. "Value at risk" means, basically, the amount of money you could conceivably lose in one day. To lose more than that in a day is a sign that your models might well be broken. To lose more than that six days in one quarter is a sign of utter cluelessness.

Update: jck has the rather misleading graphic, from Morgan Stanley's Q3. It shows four trading days with more than $125 million in losses, and three days with losses of somewhere between $50 million and $125 million, which means that Morgan Stanley's VaR was probably somewhere around $75 million or so.

What it doesn't show is just how skewed the distribution really is: on its worst day, Morgan Stanley lost $390 million. Which is more than four times its VaR. Fat tail, much?

Comments

[Michael Bommarito]

Hey Felix, just a few caveats...
i) The probability of exceeding a given loss has nothing to with the amount that loss will be exceeded by.
ii) If jck's statement of their 95% VaR level is correct, then you would precisely expect 1 out of 20 days to exceed such a loss. Over a 56-day quarter, this translates into 2.8 expected days of such loss.
iii) If the market factors that caused these MS losses were predictable, then tails in the market itself should have been predictable, and no one should have really lost money. Tails are what make financial markets so profitable and so risky, and without them, no one would care too much about markets at all.

2007 Oct 11 08:19 AM