"Hey market what are you up to?"
-
I view the source of the nonsense as the idea that just because you can do math to show two sets of debts are “equivalent” that means it’s wise to treat them like they really are identical. But, in the end, behind every loan contract there are people and there is no math to make people equivalent.
@futurebird @cshlan @PizzaDemon
That normally doesn’t matter. The estimation of the probability that someone will default has error margins, but most of the time they’re independent variables and so large enough sample sets mean they don’t matter. There are two cases where it typically goes wrong, neither is particularly to do with the humans on the customer side.
The first, which caused the 2008 crash, Enron, and so on (including, I suspect, the AI Bubble Crash) is that it’s possible to just lie. When a mortgage customer lies, it doesn’t matter that much because a load of other customers don’t and the ones that do are just another kind of outlier that’s averaged out. But when the bank lies and shuffles paper trails enough that it looks like their loans are lower risk than they are, and then they sell them on that basis, it causes problems. Similarly, if the loans are smaller numbers and are to massive companies reporting ‘revenue’ and not telling you that they are getting that revenue only because the loan is ‘invested’ in companies that then use that money to buy their products, then it’s a problem.
You couldn’t convince banks that a bundle of loans backed by NFTs are AAA rated, but you possibly could if you mixed them in with a load of mortgages to the lowest-risk customers and gradually diluted the mortgage ones. Or if you’re actually loaning money to a company that is selling NFTs and is reporting revenue that exceeds the loan amount, while quietly moving things from the capex column to the revenue column by investing in their own customers.
The second, which is more interesting (to me, at least. I don’t find lying that interesting) is that we remain very bad at reasoning about correlated risk. Prior to Katrina, a load of insurance companies did reasoning like ‘these two businesses are in completely different markets in different towns, so the risk of them both needing to claim on insurance at the same time is low’. Only it turns out that they both depended on the same electricity substation, or the same water treatment plant. When the hurricane took out their common dependency, both claimed at once. Suddenly a load of those independent variables turned out not to be independent and that caused, as I recall, six insurance companies to go out of business. This is still a big problem with things like cybersecurity. How do you find two things to insure that are not both more likely to claim in the case of a critical Windows vulnerability, for example? It’s also a problem now, because ‘the country elects a president who actively attacks the economy’ was the kind of thing that everyone knew was a common risk for most businesses and individuals (most of whom get their income via employment at businesses), but not something people estimating insurance claim rates or loan defaults thought was high enough probability to bother modelling.
-
@futurebird @cshlan @PizzaDemon
That normally doesn’t matter. The estimation of the probability that someone will default has error margins, but most of the time they’re independent variables and so large enough sample sets mean they don’t matter. There are two cases where it typically goes wrong, neither is particularly to do with the humans on the customer side.
The first, which caused the 2008 crash, Enron, and so on (including, I suspect, the AI Bubble Crash) is that it’s possible to just lie. When a mortgage customer lies, it doesn’t matter that much because a load of other customers don’t and the ones that do are just another kind of outlier that’s averaged out. But when the bank lies and shuffles paper trails enough that it looks like their loans are lower risk than they are, and then they sell them on that basis, it causes problems. Similarly, if the loans are smaller numbers and are to massive companies reporting ‘revenue’ and not telling you that they are getting that revenue only because the loan is ‘invested’ in companies that then use that money to buy their products, then it’s a problem.
You couldn’t convince banks that a bundle of loans backed by NFTs are AAA rated, but you possibly could if you mixed them in with a load of mortgages to the lowest-risk customers and gradually diluted the mortgage ones. Or if you’re actually loaning money to a company that is selling NFTs and is reporting revenue that exceeds the loan amount, while quietly moving things from the capex column to the revenue column by investing in their own customers.
The second, which is more interesting (to me, at least. I don’t find lying that interesting) is that we remain very bad at reasoning about correlated risk. Prior to Katrina, a load of insurance companies did reasoning like ‘these two businesses are in completely different markets in different towns, so the risk of them both needing to claim on insurance at the same time is low’. Only it turns out that they both depended on the same electricity substation, or the same water treatment plant. When the hurricane took out their common dependency, both claimed at once. Suddenly a load of those independent variables turned out not to be independent and that caused, as I recall, six insurance companies to go out of business. This is still a big problem with things like cybersecurity. How do you find two things to insure that are not both more likely to claim in the case of a critical Windows vulnerability, for example? It’s also a problem now, because ‘the country elects a president who actively attacks the economy’ was the kind of thing that everyone knew was a common risk for most businesses and individuals (most of whom get their income via employment at businesses), but not something people estimating insurance claim rates or loan defaults thought was high enough probability to bother modelling.
@david_chisnall @cshlan @PizzaDemon
I think you are making much more sophisticated points. And risk evaluation is fascinating.
Frankly I just have a trauma response to the word "bundle" in the context of debt trading and that's just as someone who likes to read about markets who didn't even lose anything. Just seeing it happen was enough.
So I was shocked to find out that it's still going on and people are getting burned again.
