This Quebec pension fund lost some $25 billion due to non-bank asset-backed commercial paper (ABCP). Their Value-at-Risk (VaR) model did not take into account liquidity risk. As usual, the quants got the blame. But can someone tell me a better way to value risk than to run historical simulations? Can we really build risk models on disasters we have not seen before and cannot imagine will happen?
(Hat tip: Ray)
nicholas nassim taleb ,"fooled by randomness" is a must read. simply we dont live in a gaussian world .
The NY Times recently quoted Taleb as saying, "VaR is like an airbag that works all the time, except when you have an accident." I think that's a perfect characterization.
Can we prepare for what we have not seen? The folks in the insurance business have faced this problem for centuries. Some actuaries use Extreme Value Theory, and I've often wondered if the finance world needs to look more closely at that.
Are the quants to blame for VaR's short-comings? Sort of. I ran the VaR reports for previous employer -- who got wiped out. In retrospect, I should have been telling everyone, "These numbers do not mean what you think they mean." That was my error.
Soros has a great point here:
He says that our statistic models fail to regard the role of uncertainty.
Quarterback, you're right: that's much more under the sun that the gaussian way to measure things...
Mr. Chan, I'm starting an English version of my brazilian blog, could I refer to your blog in my blogroll?
Thanks for your comments. Sure you can link to my blog from your blogroll!
Taleb makes two points, and it is often confusing which one he is arguing for.
1. The distribution of risks is not gaussian, but there may be a distribution out there that we can reconstruct from historical data.
2. For some types of risk, the historical data does not give rise to a distribution with underlying process.
It would be more helpful to the discussion if Taleb would focus his prescriptions on whether it was 1 or 2 he was arguing for, since they are drastically different claims.
I believe 2) is true. Even if you use a non-Gaussian distribution, as long as it is based on historical data, there will be some risks where you could not have predicted.
Taleb has been working with Daniel Goldstein. Goldstein is one of a number of theorists working on fast and frugal rules, which if I understand the claims correctly, aim to retrodict less of the past, but predict more of the future.
The downside of these class of rules, some of which measure risk, is whether their lack of precision.
Its interesting reading these posts, I think people should re-visit the original father of market neutral and quant funds "ed thorp" even Taleb and Bill Gross of Pimco site ED as big influences, to take a quote from Ed written years ago ,
"Two fallacies of which we were well aware were that previous historical limits on financial variables should not be expected to necessarily hold in the future, and that the mathematically convenient lognormal model for stock prices "SUBTANTIALLY UNDERESTIMATES" the probabilities of extreme moves."
IN ED's models he built in for what would happen if the market dropped 25% in a day twice the worst day ever recorded .
I think if you read some of his articles you will see just how close to the mark he saw this coming and that everybody was just far to leveraged and in the same boat together , at some point the boat was bound to tip over ...
Yes, people should consider scenarios that they've never seen. I had never seen a market environment like 2008 but survived it pretty well by preparing for something of its magnitude. I remember a client once asking me, "...come on, what's the probability that all your stocks will decline by 70% in six months?..." I told him it was 100%, eventually.
Awesome points. Glad to hear you did well in the stock market in 2008, wish I could say the same.... Thankfully, I was pretty heavy and long into gold.
Hedge against inflation? How about a hedge against "God knows what"!
Post a Comment