Everything Fails at Once: Correlation in Household Finance

This desk has now arrived at the same structure from four separate directions without naming it. Employer-linked credit fails at job loss. A guarantor fails alongside the borrower. Home equity is hardest to access when property values fall. Credit lines contract precisely when people need them. Each of these was noted as a feature of one arrangement. They're the same finding: nearly every mechanism a household relies on to absorb a shock is tied to the conditions that produce the shock. Individually each looks sound, which is how they get built. Together they fail on the same day.

The inventory

Arrangement	Fails when	Which is
Employer-linked credit	You lose the job	The shock itself
Employer-provided insurance	You lose the job	The shock itself
A guarantor or cosigner	Their circumstances worsen too	Frequently the same shock
Home equity access	Property values fall	Correlated with downturns
Credit lines	Your assessment worsens	Exactly when you need them
Retirement account balances	Markets fall	Correlated with recessions
A second earner	Same household, same economy	Partially correlated
Selling an asset	Everyone is selling	Correlated

Read the second column as a list and one date covers most of it. A recession that costs someone their job also reduces their partner's hours, lowers their home's value, depresses their retirement balance, tightens every lender's criteria, and puts their guarantor under the same pressure.

Per our employer analysis, the first two rows are the sharpest case — an arrangement that is cheapest precisely because it's tied to the relationship whose loss is the event it would need to absorb.

Read the failure conditions as a list. One date covers most of them.

Why it keeps happening

Not carelessness. Correlation is a direct consequence of what makes each arrangement cheap or available in the first place.

• Employer linkage is cheap because the employer knows you and controls your pay — per our employer analysis, that's three costs eliminated at once, and it's the same relationship whose loss is the shock.
• A guarantor is available because they're close to you, per our guarantee analysis — and closeness is what produces the correlation.
• Home equity is cheap because the asset secures it, and the asset's value moves with the cycle.
• A credit line is available because you look creditworthy, and it's withdrawn when you stop — per our shock analysis, that's the defining property of credit as a substitute for insurance.

In each case the property that makes the arrangement work is the property that makes it correlate. Which means this isn't a defect to engineer out of existing products — it's a trade-off inherent in them, and the only escape is a differently structured arrangement rather than a better version of the same one.

And it explains why the correlated options are the affordable ones. Uncorrelated protection — cash, committed facilities, insurance — costs more precisely because it has to hold when everything else doesn't. So households facing the tightest budgets systematically hold the most correlated protections, which is the distributional edge of this finding.

Why assessment misses it

The measurement reason, and it's structural.

Assessment happens at the level of the individual borrower and the individual arrangement. A model predicting whether a particular applicant repays can be perfectly accurate while saying nothing about whether many borrowers fail together.

• Correlation affects the tail, not the expected value — and expected value is what most assessment produces, per our policy analysis.
• A guarantee is assessed on the guarantor's position today, not on whether it will hold when called.
• Development samples frequently span benign periods, per our validation analysis — a model fitted on years without a severe downturn has never observed correlated failure and cannot have learned it.
• Households don't model it either. Someone with a job, a card, home equity, and a relative who could help has four options on paper and frequently one scenario in which they all disappear.

The third point deserves weight. Most credit models in use have been fitted on data containing no severe correlated event, so the relationship they encode is the benign-period one. That isn't a criticism of the modelling — it's an observation about what the data could possibly contain — and it means stress behaviour has to be reasoned about structurally rather than estimated.

The stacking problem

The dynamic that turns correlation into a sequence, and it's the household analogue of what our time preference analysis describes with price.

A household meeting a shock draws on protections in order:

1. Savings, if any.
2. Credit lines — which per our shock analysis may already have been reduced.
3. Family — who may be under the same pressure.
4. Assets — worth less, and slower to sell.
5. Retirement funds — depleted, and expensive to access per our liquidity analysis.
6. High-cost credit — which is where the sequence in our time preference analysis begins.

Each step failing pushes toward the next, and the steps are correlated with each other as well as with the shock. Which means the ordering isn't a menu of independent options — it's a cascade in which the failure of one raises the probability that the next has also failed.

And the household experiences this as a series of separate disappointments rather than as one event. The line was reduced, then the relative couldn't help, then the house wouldn't sell. Each looks like bad luck; together they're one correlated structure doing exactly what it was always going to do.

What isn't correlated

The short list, and its shortness is the point.

Arrangement	Correlated?	Why
Cash in an accessible account	Barely	Nothing withdraws it
A committed facility	No, by construction	The decision was made in advance
Insurance on a defined event	No	Pays on the event, not on assessment
A guarantor in a different economy	Less	Different shock exposure
A second income in an unrelated sector	Less	Different cycle

The common feature of the top three: the decision to provide support was made in advance and isn't revisited at the moment of need. That single property is what distinguishes protection that holds from protection that evaporates.

Which reframes the emergency fund. Per our savings analysis it's usually presented as prudence; per our structural analysis it's the only concentration mitigation available. The correlation framing adds the strongest version: it's the only protection that doesn't depend on someone's assessment of you at the moment you need it — and that independence, not the amount, is what makes a modest buffer outperform much larger contingent access.

And it's the argument for the instrument our structural analysis identified as missing. A committed facility arranged in advance is uncorrelated by construction, which is exactly why firms use them and why households mostly don't have one.

The lender's version

The same structure from the other side, and it's underpriced.

A lender holding many correlated exposures has more tail risk than the sum of its individual assessments implies. Where it shows up:

• Guaranteed portfolios, per our guarantee analysis — the second source of repayment fails alongside the first.
• Employer-channel lending, where a single employer's difficulties hit many borrowers at once — single-name exposure disguised as a consumer portfolio.
• Geographic concentration, which correlates borrowers through a local economy.
• Sector concentration, same mechanism.
• Collateral-backed lending, where values fall as defaults rise.
• Small business portfolios, per our small business analysis, where concentration compounds.

The second is the one growing fastest and examined least. Per our distribution work, the payroll channel is among the few genuine cost reductions available — and a book built through it carries employer-level concentration that individual credit assessment cannot see, because each borrower's file looks unrelated to every other.

What a lender should do about it:

• Measure concentration by employer, geography, and sector, not only by risk grade.
• Stress the correlated scenario rather than the average one.
• Discount guarantees for correlation rather than valuing them at face.
• Treat a channel's concentration as a risk attribute, per our channel analysis.

Designing against it

What follows for products, and the principle is single: availability should not depend on an assessment made at the moment of need.

• Committed rather than revocable facilities. The expensive property and the whole value.
• Arrangements triggered by events rather than by assessment — insurance logic rather than credit logic.
• Hardship terms agreed in advance, so per our arrangements guide the option exists before the difficulty rather than being requested during it.
• Savings features built into credit products, so a buffer accumulates without requiring a separate decision.
• Deliberate decorrelation — not linking a household's credit, insurance, and income to one counterparty.

The third is the cheapest and least used. A lender that defines in advance what happens on job loss — a payment holiday, a term extension — has created an uncorrelated protection at almost no cost, because per our arrangements guide these are options lenders already grant and simply grant late, discretionarily, and to whoever asks.

And the honest tension: committed facilities cost more to provide, because the lender can't withdraw when risk rises. That's not a market failure — it's the price of the property being bought. The current arrangement is cheaper and worth less, and neither side of the transaction describes it that way.

The strongest objections

"Correlation is well known and priced." True in wholesale markets and institutional risk management, where correlated exposure is a standard concept. The claim here is narrower: it's not priced or communicated at the household level, and consumer products are sold as independent protections when they aren't. A household with four correlated options is not told they're correlated by anyone.

"You're describing ordinary macroeconomic risk." Partly, and the distinction is that these correlations are built into product structures rather than being facts about the economy. An employer-linked loan is correlated by design — someone chose that structure because it lowered cost — and that's a different thing from a recession affecting everyone.

"You've measured none of it." Correct, and the central limitation. The correlations here are argued from structure rather than estimated, and their magnitudes matter enormously for whether this is a serious problem or a modest one. Data on how household protections actually co-fail is not generally available, in part because — as our coverage analysis notes — many of the arrangements aren't recorded anywhere.

Testable implications

1. Employer-channel portfolios should show employer-level clustering of defaults beyond what individual risk explains — testable from any lender's own book.
2. Guarantee call rates should spike in downturns disproportionately to individual default rates.
3. Households should exhaust multiple protections in rapid sequence rather than one at a time, observable in cash flow data.
4. Credit line reductions should concentrate on borrowers approaching difficulty, which is the shock analysis prediction and the clearest case of assessment-dependent withdrawal.
5. Households with a buffer should show better outcomes than households with larger contingent access at equivalent income.
6. Models fitted only on benign periods should underpredict losses in stress by a margin reflecting unobserved correlation.

The first is the most immediately actionable and the one a lender can run this quarter. If defaults in an employer-channel book cluster by employer beyond what borrower characteristics explain, the portfolio has concentration that no individual assessment captured — and it's measurable with data every such lender already holds.

The conclusion we'd hold: household financial protection is assembled from arrangements that are individually reasonable and jointly fragile, because the properties making each one cheap are the properties making it correlate. The only protections that hold are the ones where somebody decided in advance and can't change their mind — which is a short list, and it's the expensive end of it.

Frequently asked questions

---

This report presents an analytical framework and the authors' interpretation; it is not financial or investment advice, and nothing here is a recommendation about any individual's savings, insurance, or borrowing arrangements. The correlations described are argued from the structure of the arrangements rather than estimated from data, and their magnitudes are not established here; the implications identified as testable are hypotheses.