The 2008 Subprime and Housing Collapse

Between 2000 and 2006, US house prices roughly doubled. Mortgage lenders, especially those making "subprime" loans to borrowers with weak credit, kept lending on the assumption that if a borrower defaulted, the house could simply be sold for more than the loan was worth. That assumption depended entirely on prices continuing to rise. Nationally, in the modern data, they never had fallen. So almost nobody had built a model, or a bond, that assumed they could.

From a mortgage to a security

A subprime loan did not sit on a bank's balance sheet. It was sold into a pool with thousands of others, and the pool's cash flows were sliced into tranches — see Mortgage-Backed Securities. The senior tranches got paid first and were rated AAA; the lowest "equity" tranche absorbed the first losses and was rated, if at all, as junk. Banks then took the mezzanine tranches that were hard to sell, pooled those together, and re-tranched them again into collateralized debt obligations (CDOs), manufacturing a fresh batch of AAA paper out of BBB material. The ratings agencies modeled default correlation across mortgages using a handful of years of data from a period when house prices only went up — see ABS CDOs and the 2008 Correlation Failure.

Layer	What it was	What happened
Subprime mortgage	Loan to a weak-credit borrower, often adjustable-rate	Defaults began rising in 2006-07 as low "teaser" rates reset higher
MBS tranche	Claim on a pool of thousands of such mortgages	Losses concentrated in lower tranches, then reached "AAA" ones once assumptions broke
CDO / CDO-squared	A bond made of the middle tranches of other mortgage bonds	Correlated losses across the whole housing market wiped out tranches assumed to be nearly independent
Bank balance sheet	Held CDOs directly, or insured them via credit default swaps	Losses that were meant to be dispersed across the system turned out to be concentrated in a few large, leveraged institutions

The mechanism: leverage meets a correlated shock

The structures were built to survive idiosyncratic mortgage defaults — one borrower losing a job, one region having a bad year. They were not built to survive a national house-price decline, because the historical data used to calibrate them contained no such event. When prices turned down in 2007, defaults rose everywhere at once, and losses that the models had treated as diversified turned out to be highly correlated.

Two amplifiers turned a housing correction into a systemic crisis:

• Leverage. Investment banks were funding long-term, illiquid mortgage assets with short-term borrowing, much of it overnight Repo and Reverse Repo. When the collateral's value became uncertain, lenders stopped rolling that funding, forcing fire sales into an already falling market.
• Interconnection. Losses did not stay with whoever originated the loan. They passed through securitization to pension funds, money-market funds (see Money Market Funds and Breaking the Buck), and insurers like AIG, which had sold credit protection on CDOs without holding capital against the risk of a system-wide loss.

Bear Stearns was rescued in March 2008. Lehman Brothers was not, and its bankruptcy on 15 September 2008 froze short-term credit markets globally within days — money-market funds, banks, and companies that relied on commercial paper for routine funding suddenly couldn't get it. The Federal Reserve and Treasury responded with emergency lending facilities, the $700bn Troubled Asset Relief Program, and eventually years of near-zero rates and quantitative easing (see Quantitative Easing and Central Bank Balance Sheets).

The lesson

A model calibrated on a period with no examples of the event you're worried about will confidently tell you that event is nearly impossible. That was true of default correlation in 2008, and it recurs anywhere a rare regime hasn't yet appeared in the sample — see Russia 1998 and the LTCM Contagion for the same failure mode a decade earlier, with a different asset class. The second, more durable lesson is about funding: an asset can be fundamentally solvent and still be destroyed if it is financed overnight and its lenders lose confidence in the collateral, because the fire sale itself pushes the price down further, triggering more sales.