FS: Was the risk of a credit crunch quantifiable?
If yes: did anybody quantify it accurately, and can the rest of the global financial system learn from what they did, or from what such a hypothetical person should have done?
If no: are there in any case lessons that can be learned from this episode -- lessons which might reduce systemic risk in future? Or should we all just learn to live with uncertainty?
RR: I hate to be evasive, but it does depend on what we mean by 'quantifiable'. If we mean 'Could the magnitude and probability of the event have been predicted?', the answer is realistically 'No'. If one thinks about it for a moment, one can see that it is not even obvious what the term 'probability' means in this context: for exceptional events of this nature, we 'live only once', that is, we (thankfully) do not have the statistical luxury of many repetitions of these extraordinary events under identical conditions. Therefore a 'frequentist' view of probability (probability = frequency of occurrence) does not really make sense. (By the way, unfortunately this is the only type of probability that the current financial risk management thinking seems to know about, but this is another story).
Having said this, if the question about quantifiability has a more limited meaning, something like: "Could a financial storm of similar magnitude to the one we are observing have been predicted?", the answer is probably positive. After all, many observers have been saying for a while that the conditions of extreme monetary liquidity (too much money chasing too few investment opportunities) of the last few years created the perfect environment for a 'bubble' to form first, and to pop then. Risk managers around the world have known very well for quite some time that these conditions of loose money were a financial disaster waiting to happen. Guessing precisely which disaster would materialize is, however,unfortunately no easier than predicting from which of the many cracks in a dam the water will eventually burst out.
In a way there is a sort of 'uncertainty principle' at play in risk management: we don't seem to be able to specify both the probability and the nature of a large adverse event at the same time. If we can point to a specific series of severely adverse events, it is extremely difficult to associate with them a meaningful probability. Conversely, if we believe that we can specify a (low) probability of loss, it becomes almost impossible to tell what series of events are associated with this loss.
FS: Do you feel that you understand how a spike in US subprime default rates seems to have been responsible for all manner of ills, including very large spreads globally between risk-free overnight rates and overnight Libor? Or was it less that subprime *caused* the rest of it, and more that it just happened to be the first shoe to drop, and that the other things would have happened anyway?
RR: After the event, I think I can understand it very well. The problem is that many causes can have the same effect. Any kind of major disturbance that puts into question the health of financial institutions, for instance, will have a widening effect on swap spreads (witness what happened in 1998), or on the cost of protection on mono-line insurers. The conditions of opaqueness and asymmetric information that give rise to this reaction are common to many stress events. Whenever investors feel that 'something nasty and big may be lurking down there', the risk premium widens in predictable ways. So, 'subprime', in my opinion, did cause many of the contagion events, but other shocks could have had similar consequences.
Admittedly, some events, like the difficulty in re-financing SIVs and CP conduits, have been more severe because of the specific nature of this particular crisis. But, in situations of severe distress, the phenomenon of 'flight to quality' (that I prefer to call 'flight to liquidity') is almost universal.
FS: Do you have a handle on how something like mortgage default rates can go from irrelevant to all-important seemingly overnight? It was really very recently that mortgage-bond analysts were still telling me that the only thing you need to worry about is prepayment rates, not default rates. If people don't know what is or will be important, how can they position themselves to not get sandbagged by something like a credit crunch?
RR: Until recently, assessing the riskiness of the payments from a pool of mortgages on the basis of diversification (eg, regional) made a lot of sense. A security could be constructed (and meaningfully rated) in such a way that only systematic risk (eg, interest-rate and house price risk) would be left in it. Statistical analyses, however, rely on the present looking like the past. If, all of a sudden, the lending standards are relaxed, and no-documentation NINJA (No Income, No Job or Assets) mortgages are granted, all past relationships go out of the window. If fraud is at play, it does not really matter, and it does not add to diversification, whether the mortgage is made to a borrower in Nebraska or in California. There is now a new variable that links (correlates) borrowers much more strongly than it was possible under the previous regime.
A simple example can perhaps explain this better. Suppose that I want to create a pool of subjects with the characteristic 'fair hair' appearing in the same ratio as in the population at large. According to my model blond people tend to be taller than the average person. Therefore I require for acceptance into my pool the requirement that all heights should be present in the same ratio as in the population at large. This is all good and well, and I may think that I have 'diversified'. If I fail to test for blue eyes, however, and there is now an incentive for blue-eyed people to apply for inclusion in the pool, my previous diversification criterion becomes next to useless. The feature 'blue eyes' has a very strong correlation with blond hair, and all of a sudden my pool is very undiversified.
As the lending standards of the 06 and 07 mortgage vintages have proven to be so dramatically different than the credit criteria for earlier pools, models that rely on past diversification experience cannot be expected to give a reliable guidance to new NINJA world we live in.