Isn't the difference between the AAA held by Berkshire Hathaway and the AAA assigned to a CDO tranche that very simple erroneous assumptions can lead you to assign low risks to large numbers of credit events that are subtly but decisively correlated? Finding uncorrelated investments in large groups of standardized assets is actually as much a tricky social science project as it is a green-eyeshades math problem, right?
When you AAA a product built out of supposedly uncorrelated risks that turns out to be correlated, you're fucked.
So, to extend your analogy: suppose you designed instruments backed by $5 debts from college students. And suppose as time goes on, Ivy League students used it less and less, while at UMich-Flint the program spread like wildfire. At some point, without even changing the explicit structure of your instrument, basic market forces might up converging on highly correlated (and very risky) buckets of risk, mispriced at AAA (or whatever).
And all the while, every incentive is set up to get people to throw away unfavorable results and cherry pick the most lucrative interpretations.
My point (you are by this point I am sure exasperated with me) is that ratings agencies and product designers are trying to build models to assign risks to quickly moving targets, by taking a spot measurement of the current correlation coefficients and extrapolating them.
Whereas on the other hand the credit history and business model backing BRK is (relatively) straightforward.
Your comment makes total sense [1] : I started looking at these collateralized debt obligations (i.e. CDOs) back in 2002. That was around the time of the first 'crash' in these structures (CBOs : structures backed by High Yield Bonds).
The basic error with CBOs is that the rating agencies allowed the structures additional diversity 'points' for investing in different sectors (a decent idea), but then segmented the market so that many, many different telecom-related businesses got put in different sectors. The telecoms bubble then proceeded to wipe out about half the contents of each CBO. Why didn't anyone point out the problem? Because the investment banks, having been given the published specification of how the Rating Agencies would model any hypothetical structure, went about optimizing the portfolios of bonds to maximize the 'ratings arbitrage' available.
Your point about migration of the underlying risks is spot-on. The rating agencies looked at the historical performance of each of the classes of risk. For CBOs, once people saw how the rating agencies would give them points for particular sorts of High Yield bonds, suddenly telecoms (& fibre, & cable, & satellite, & etc..) companies found it very easy to raise what were known in the market as 'CBO bonds' : bonds that no rational investor would want, but would be very appealing for a structure to buy (since it would help its ratios).
Similarly, once there were evening classes in how to improve your FICO score, the whole history of sub-prime borrower repayment statistics became irrelevant. The manipulation of the fundamental inputs to the models by 'good hardworking Americans' was rampant... But the Rating Agencies were being paid well to continue to rate structures (at a crazy pace), and the Investment Banks were in no hurry to point out that the models should be harsher.
Part of the whole problem, though, stems from banking & insurance regulations that mean that its very cheap (from an equity capital point of view) to leverage up AAA paper. That's what drove everyone to demand AAA paper in the first place. And the AAA designations is/was handed out by rating agencies that now claim it was 'free speech' and they're not liable for anything.
As for the moving target aspect : Investment Banks are continually trying to innovate, since that's where it's less competitive, and the margins haven't been competed away. They would ask Rating Agencies to look at new products all the time. If all the Ratings Agencies were too conservative (or cautious), then the product wouldn't work, and would be abandoned. However, if one of them could be persuaded to come up with an exploitable methodology, then they would get all the business...
And before anyone says : Ahh, Wall Street is just about the exploitation of loopholes, think about examples from hacker-space : SEO comes to mind...
[1] except for the example of BRK : That's a bit of a special business. Reinsurance is a tricky thing, and it's possible to look very smart until you discover you're an idiot. AXA looked pretty smart, until... Better examples would be AAA industrial businesses, or a hard-asset business, for instance.
When you AAA a product built out of supposedly uncorrelated risks that turns out to be correlated, you're fucked.
So, to extend your analogy: suppose you designed instruments backed by $5 debts from college students. And suppose as time goes on, Ivy League students used it less and less, while at UMich-Flint the program spread like wildfire. At some point, without even changing the explicit structure of your instrument, basic market forces might up converging on highly correlated (and very risky) buckets of risk, mispriced at AAA (or whatever).
And all the while, every incentive is set up to get people to throw away unfavorable results and cherry pick the most lucrative interpretations.
My point (you are by this point I am sure exasperated with me) is that ratings agencies and product designers are trying to build models to assign risks to quickly moving targets, by taking a spot measurement of the current correlation coefficients and extrapolating them.
Whereas on the other hand the credit history and business model backing BRK is (relatively) straightforward.