US Housing & Mortgage Report 2008 — Where the Break Began

The 2008 crisis is usually narrated through the institutions that failed. The Global Investment Outlook 2008 argues that those failures were a funding phenomenon rather than a credit one. This report goes underneath both accounts, to the collateral, and asks a narrower question: what had to be true about US housing for the rest of it to happen?

The answer is a single assumption. Nationwide house prices had not fallen year-on-year in the modern era, and the structures built on mortgages treated that as a property of the world rather than an observation about a particular period. Everything else — the underwriting, the securitisation, the ratings, the leverage — was downstream of it.

This matters because it reframes what actually went wrong. Poor-quality lending was real, and it was not sufficient. A pool of weak loans spread across many regions is survivable if the regions are genuinely independent, because losses in one are offset by performance in another. The pools were not built to withstand every region moving together, and they did not have to be, so long as the assumption held.

The assumption was load-bearing in a second, less obvious way. A borrower who cannot afford a payment does not necessarily default, provided they can refinance or sell. Both remedies require rising or at least stable prices. When prices stopped rising, a large population of loans that had been performing on the expectation of refinancing lost their exit simultaneously. Affordability had been substituted with appreciation, and the substitution was invisible while appreciation continued.

Three consequences followed, and each recurs later in this archive.

Origination separated from ownership. Once a loan was written to be sold, the party choosing whether to make it was not the party who would suffer if it failed. That is a genuine incentive problem, but it is not fraud, and treating it as fraud obscures how ordinary the behaviour looked at the time.

Complexity destroyed observability. Tranched claims on pools of claims made it difficult for any holder to know their exposure, which is what turned a credit problem into the counterparty freeze described in the Global Investment Outlook 2008.

Foreclosure became self-reinforcing. Repossessed homes sold into falling markets set the comparable prices that determined the next round of valuations — the same mark-to-market spiral, operating on physical collateral.

The assumption underneath everything

It is worth stating the central assumption precisely, because it was rarely written down as an assumption at all.

The claim. House prices in the United States might fall in a given city or region, but a simultaneous nationwide decline had no precedent in the available data. Regional markets were driven by regional economies — local employment, local supply, local demographics — and those drivers were not synchronised.

Why it was reasonable. It was consistent with the historical record, and the mechanism behind it was plausible. Housing is local. A downturn in one metropolitan area genuinely had, in the past, coincided with strength elsewhere.

Why it failed, and this is the important part: the assumption described a period in which the dominant driver of house prices was local. It stopped being true once the dominant driver became national credit conditions. When the availability and price of mortgage credit is what sets house prices, and mortgage credit is set nationally, then every regional market is exposed to the same variable. The diversification did not weaken gradually. It was removed by the very growth in mortgage credit that the structures were built to intermediate.

The pools were diversified against the risk that had historically mattered and concentrated in the one that came to matter. Diversification is a claim about which factor dominates, and that claim expires.

This is the archive's common-factor problem in its clearest form. Pooling removes idiosyncratic risk — one borrower losing a job. It does nothing about a shared exposure. A pool of thousands of mortgages across forty states was one position with respect to national credit conditions, and no amount of geographic spread changed that.

How a mortgage stopped being a loan

The second structural change was to what a mortgage was, institutionally.

The traditional arrangement. A lender wrote a loan, held it, and collected payments for its life. The lender's return depended entirely on repayment, which meant the decision to lend and the consequence of lending sat in the same place.

The arrangement that replaced it. A loan was written in order to be sold — into a pool, which issued securities, which were bought by investors, some of whom repackaged them again. The originator's return came from the act of origination.

What that does to incentives is not mysterious:

• Volume becomes the objective, because revenue is a function of loans written rather than loans repaid.
• Underwriting standards become a competitive cost. A firm applying more caution writes fewer loans and loses share to one applying less, and there is no near-term penalty for the looser firm.
• Information stops travelling. The originator knows things about the borrower that the eventual holder cannot see — and has no mechanism, and limited incentive, to transmit them.

The standard account stops there, and it is incomplete in a way worth correcting. Originators frequently did retain exposure, through representations and warranties, through residual tranches, and through the simple fact that many were part of institutions holding the securities. The problem was not that nobody had skin in the game. It was that the retained exposure was priced off the same assumption as everything else. A firm holding the riskiest slice of its own pool was not indifferent to losses; it had modelled those losses on a world where prices did not fall nationally.

That distinction matters for what you conclude. If the failure was purely misaligned incentives, aligning them fixes it. If the failure was a shared assumption held by aligned and misaligned parties alike, alignment is insufficient — and the 2023 episode described in the Global Investment Outlook 2023, where institutions holding impeccable collateral failed anyway, suggests the second reading is the durable one.

Why prices mattered more than defaults

The intuitive model of mortgage loss runs: borrower stops paying, lender loses money. The actual mechanism has two stages, and the second dominates.

Stage one: the borrower stops paying. This is driven by affordability — income, employment, payment resets. It is what underwriting is meant to assess.

Stage two: the lender recovers what the collateral fetches. This is driven entirely by house prices. A default on a property worth more than the loan produces little or no loss; the property is sold and the debt repaid. The same default on a property worth less than the loan produces a loss equal to the shortfall, plus costs.

So the loss function is a price function, and this has three consequences that were underappreciated:

• Equity is the buffer, and it evaporates first. A borrower with meaningful equity has both the means and the motive to sell rather than default. Falling prices remove that buffer across the whole pool simultaneously.
• Refinancing is an exit that closes exactly when it is needed. Loans underwritten on the expectation that the borrower would refinance before a payment reset depend on the borrower having equity and on credit being available. Both fail together.
• Default becomes rational rather than merely unfortunate. A borrower substantially underwater faces a different calculation from one with equity, independent of whether they can pay.

Underwriting asked whether the borrower could pay. The loss depended on what the house was worth. Those are different questions, and only the first was being assessed carefully.

The compounding effect is the one that produced the tail. Falling prices increase both the probability of default and the loss given default, at the same time, for the same reason. Models that treated those as independent inputs understated the tail severely — not by a modest margin, but structurally, because the error grows precisely in the scenario the tail is meant to describe.

The ratings problem was a correlation problem

Credit ratings on structured mortgage products have been treated as the crisis's clearest villain. The mechanism deserves more precision than that, because the useful lesson is not about dishonesty.

What a rating on a structured product represents. Not an opinion about the underlying loans individually, but a modelled estimate of how a tranche performs given assumptions about how the pool behaves collectively. The senior tranche is protected by the subordinate ones; the question is how much loss must occur before that protection is exhausted.

The input that determines the answer is correlation — the degree to which losses arrive together. It is more important than the average quality of the loans:

• With low correlation, losses arrive steadily and predictably. Subordination absorbs them and the senior tranche is genuinely safe, even over a pool of weak loans.
• With high correlation, losses arrive in a cluster. Subordination is consumed at once and the senior tranche takes losses it was modelled never to take.

The correlation assumptions were drawn from a period in which the assumption of the first section held. They were calibrated on data from a world where regional markets moved independently, and were then applied to a world where they did not.

Two further mechanics made this worse:

• The structures were re-securitised. Pooling tranches of pools and tranching again produces instruments whose sensitivity to the correlation assumption is far greater than the original securities — an error in the input is amplified at each stage.
• The rating was a single ordinal label. A rating compresses a distribution into one symbol, and a senior structured tranche and a government bond carrying the same label had very different sensitivities to being wrong. The label conveyed the central estimate and discarded the model risk, and the model risk was the entire exposure.

The conflict-of-interest critique is real and is not the mechanism. Issuer-pays creates pressure, and it does not explain why sophisticated buyers with no such conflict reached the same conclusions. They shared the input.

The workout machinery that did not exist

By 2008 a large population of loans was heading for default in a market where foreclosure destroyed value for everyone. Renegotiation should have been the obvious answer. It largely did not happen, and the reasons are structural rather than a failure of will.

Servicing had been built for administration, not negotiation. A servicer's function was to collect payments, remit them, and handle a low, steady volume of delinquencies. It was staffed and systematised for that. It was not equipped to assess thousands of individual borrowers' circumstances and restructure terms.

Securitisation had removed the person who could say yes. A held loan can be renegotiated by the lender, who can compare a modification against a foreclosure and choose the better outcome. A securitised loan has no such party. The servicer acts for a dispersed set of holders with conflicting interests — a modification that helps a senior tranche can harm a subordinate one — and its contractual duties and liability exposure both pointed toward the mechanical path.

The tranche conflict deserves emphasis, because it made a collectively rational outcome unreachable:

• Reducing principal reduces losses overall but reallocates who bears them.
• Extending a term changes the timing of cash flows, which different tranches value differently.
• Any modification benefiting one class could be challenged by another, so the defensible action was frequently the one that followed the contract mechanically.

The result was a foreclosure wave that was collectively self-defeating. Each repossessed property sold into a falling market established a comparable price. Those comparables set appraisals for surrounding homes, which reduced equity for neighbouring borrowers, which increased their probability of default. The mark-to-market spiral described in the Global Investment Outlook 2008 operating on physical collateral, with a foreclosure timeline instead of a margin call.

What an allocator could act on

Ask what has to remain true, and check whether the structure created its own exception. Mortgage pools depended on regional independence, and the growth of national mortgage credit was what removed it. A diversification assumption undermined by the very activity it enables is the most dangerous kind, because the erosion is invisible from inside.

Distinguish the trigger from the loss driver. Defaults triggered mortgage losses; house prices determined them. When those are correlated — and here they were driven by the same variable — a model treating them as separate inputs understates the tail exactly where it matters.

Treat a rating as a central estimate with the model risk removed. Two instruments with the same label can have entirely different sensitivities to their assumptions being wrong. The question worth asking is not what the rating is but which input it is most sensitive to.

Ask who can renegotiate. Distress is survivable when someone has both the authority and the incentive to restructure. If ownership is dispersed across classes with conflicting claims, the mechanical path — foreclosure, liquidation, enforcement — becomes the default even when everyone would prefer otherwise. The existence of a decision-maker is a risk factor.

Watch for appreciation substituting for affordability. A loan that works only if the borrower can refinance is a bet on prices, not on income. The same substitution appears wherever a position depends on an exit that requires favourable conditions — a pattern the Private Credit Report 2024 examines in a different market.

Assume the diversifying factor and the loss factor converge under stress. Geographic spread, sector spread and asset-class spread all rest on a claim about which factor dominates. In a severe episode the dominant factor is usually funding, and funding is common to everything.

What 2008 established

• A diversification assumption can be destroyed by the activity it makes possible — national mortgage credit removed the regional independence that mortgage pools were built on.
• Originating to distribute separates the decision to lend from the consequence of lending, though the deeper failure was a shared assumption held by exposed and unexposed parties alike.
• Mortgage loss is a house-price function, not a default function, and default probability and loss severity move together because they share a driver.
• Structured ratings are correlation estimates, and re-securitisation amplifies sensitivity to that single input.
• Dispersed ownership removes the ability to renegotiate, which makes foreclosure the default path and turns it into a price-destroying feedback loop.