The Forbearance Calculus: Why Lenders Systematically Collect Too Hard

Our previous report argued that consumer credit contains an option held by the borrower. This one concerns the option held by the lender: the right to wait rather than enforce. It has value for the same reason any option does — enforcement is irreversible, waiting preserves choice, and the underlying situation may improve. Yet lenders exercise it far less than the arithmetic supports, and the reason is not that they've calculated and disagreed. It's that one side of the calculation is measured and the other is invisible by construction. Legal fees and charge-offs appear in reporting. The recoveries forgone by enforcing on a borrower who would have recovered appear nowhere, because that outcome was foreclosed by the decision.

The lender's option

At any point after default, a lender chooses between enforcing and waiting. Enforcing produces a certain, small, immediate recovery. Waiting produces an uncertain, potentially larger, delayed one — and preserves the ability to enforce later.

That structure is an option, and it has three sources of value:

• Time value. The borrower's circumstances may improve. Employment resumes, the seasonal trough passes, the medical episode ends.
• Information value. Waiting reveals which kind of borrower this is. A borrower who resumes partial payment has told you something no model would have.
• Preservation value. Enforcement destroys the productive relationship between borrower and collateral — the vehicle that generates the income, the business that services the debt. Selling a distressed asset realizes a fraction of its value to the borrower, and that gap is destroyed rather than transferred.

Against these, waiting has real costs: further deterioration, additional accrual that may be uncollectible, carrying cost, and the possibility that collateral value falls or disappears.

Our claim is not that waiting is always right. It's that the option has substantial value, that value is routinely set to zero in practice, and the reason is structural rather than analytical. A lender who has never calculated the forbearance option isn't disagreeing with this argument; they haven't encountered it.

Enforcement realizes a fraction of what the asset is worth to the borrower. That gap isn't transferred to the lender — it's destroyed, and both parties pay for it.

Working the arithmetic

An unsecured balance of $8,400, ninety days delinquent. The lender chooses.

Path A — enforce now. Charge off and place with an agency at a typical contingency rate. Assume 14% gross recovery on aged paper, less a 30% commission:

• Gross recovery: $1,176
• Net of commission: $823
• Timing: spread over 12–18 months

Path B — forbear three months, then a 24-month plan. Assume, from the distribution discussed below, that 45% of borrowers offered structured relief complete or substantially complete, 20% partially perform before failing, and 35% fail immediately and land in Path A anyway.

• 45% complete: 0.45 × $8,400 = $3,780
• 20% partial, recovering roughly 35% before failing: 0.20 × $2,940 = $588
• 35% fail to Path A, at a discount for the delay — say $700 net: 0.35 × $700 = $245
• Less servicing cost of the plan across the population, roughly $110 per account
• Expected net: approximately $4,503

The difference is more than five to one. Even if you halve the completion assumption to 22%, Path B produces roughly $2,400 — still nearly three times Path A.

Now find the break-even. Path B falls to Path A's $823 only when the completion rate drops to about 4%. That is: structured relief has to fail more than 96% of the time before immediate charge-off is the better decision. No plausible estimate of plan completion is that low — the figures in our plan design guide are an order of magnitude better.

The arithmetic isn't close, which raises the obvious question: if it isn't close, why is the practice so widespread?

The measurement asymmetry

Here is the answer, and it's the analytical core of this report.

Enforcement costs are observed. Forbearance costs are observed. The cost of enforcing when you shouldn't have is unobservable in principle.

Consider what a lender learns from each decision:

Decision	Outcome	Observable?
Forbear	Borrower recovers, pays in full	Yes — recorded as recovery
Forbear	Borrower fails anyway, later charge-off	Yes — recorded as a cost of forbearing
Enforce	Borrower would have failed anyway	Yes — recorded as loss avoided
Enforce	Borrower would have recovered	No — never observed

Three of four cells are measured. The fourth — the borrower who would have paid in full had they been given three months — generates no data at all, because enforcement destroyed the counterfactual. It doesn't show up as a loss. It doesn't show up as anything.

This is the same censoring problem as in our reject inference analysis, appearing at the other end of the credit lifecycle. There, a lender never observes how declined applicants would have performed. Here, a lender never observes how enforced-upon borrowers would have recovered. In both cases the institution is systematically blind in exactly one direction, and in both cases the blindness pushes toward the conservative choice.

The consequence compounds through management reporting. A forbearance program that fails on 35% of accounts generates a visible 35% failure rate, which looks like a problem. An enforcement policy that destroys value on 45% of accounts generates no failure rate at all, because those accounts closed as expected. The measured program looks worse than the unmeasured one regardless of which performs better, and a manager who responds rationally to their reporting will shrink the forbearance program.

Four structural biases toward enforcement

Beyond measurement, four institutional features push the same direction.

Incentives reward speed. Collections staff are frequently measured on resolution rates and accounts cleared. A forbearance keeps the account open and unresolved; a charge-off closes it. The individual making the decision is rewarded for the action that destroys value, which is a straightforward principal-agent problem and almost never addressed in metric design.

Accounting treatment penalizes modification. Modified loans frequently attract additional disclosure, classification, or provisioning requirements. A charged-off loan attracts none of that — the loss is taken and the matter closes. The accounting cost of forbearance is immediate and visible; the accounting benefit is diffuse and deferred. This is a real constraint rather than an imagined one, and it argues for policy structures that capture the economics without triggering the treatment where possible.

Institutional discomfort with appearing lenient. Widespread forbearance invites the concern that borrowers will learn to expect it — the moral hazard argument, addressed below. The concern is legitimate in principle and is applied far more aggressively than the evidence supports, partly because "we were too soft" is a career-ending finding and "we destroyed recoverable value" is invisible.

Separation of decision rights from economic exposure. Where servicing is separated from ownership — the arrangement our servicing analysis describes — the party deciding whether to forbear is frequently compensated on a fee basis rather than on recovery. A servicer paid per action has weak incentive to run a patient workout, and the owner bearing the loss isn't making the call. This is the sharpest version of the problem and the most structurally entrenched.

What the mass forbearance episode showed

The pandemic-era programs are the closest thing to a natural experiment this question has, and they're worth reading carefully rather than triumphantly.

What made them analytically valuable: relief was extended broadly rather than only to borrowers individually assessed as likely to recover. Normally forbearance is granted selectively, so outcomes are contaminated by selection — of course the borrowers a workout officer judged likely to recover recovered. Broad programs relax that selection substantially, and the resulting outcome distribution is closer to what the untreated population would have produced.

The headline finding across mortgage and consumer programs: the substantial majority of participants exited into performing status rather than into loss. Borrowers resumed payment, caught up through deferral or modification, and continued.

Three caveats we'd insist on, because the result is frequently over-claimed:

• The macro environment was unusually favorable. Income support was large, employment recovered fast, and asset prices rose. Forbearance during a period when borrowers' circumstances improve independently will outperform forbearance in a prolonged downturn.
• Housing equity did a lot of work. Mortgage borrowers with substantial equity had options — sale, refinance, extraction — that unsecured borrowers don't, per our equity analysis. Generalizing mortgage results to unsecured credit overstates the case.
• Participation was self-selected among those who applied, so some selection remains.

What survives the caveats, and it's the important part: the share of distress that turned out to be temporary was considerably larger than lender practice had implicitly assumed. That's not a claim about the exact rate. It's a claim about the direction of the prior error — and the direction is the thing the measurement asymmetry predicts.

The distinction that decides it

Forbearance is right for temporary distress and wrong for structural distress, and almost all the value in a workout function comes from telling them apart.

	Temporary	Structural
Cause	Job gap, medical event, seasonal trough, one-time expense	Permanent income decline, obligation exceeds capacity
Has an end date?	Usually identifiable	No
Affordable after?	Yes	No
Right instrument	Deferral or short forbearance — bridge the gap	Modification changing the payment, or disposition
What forbearance does	Solves it	Defers it and adds accrual

The operational point that follows is unglamorous and consequential: this distinction is usually determinable by asking. Not modeling — asking. "What happened, and when do you expect it to change?" separates the two categories in most cases, and the answer is frequently specific and verifiable.

Which means the highest-return investment in a workout function is contact quality, not decision sophistication. A collections operation that reaches borrowers and has a real conversation will out-decide one with better models and worse contact — because the determining variable is something the borrower knows and the model doesn't. That reframes the contact-rate constraint in our contact data analysis from an efficiency problem into the binding input on decision quality.

And it explains a failure mode worth naming: applying forbearance to structural distress is genuinely harmful. A borrower whose income has permanently fallen, given three months of deferral, arrives at month four owing more with no improved capacity. That's not leniency — it's deferral of an inevitable outcome at the borrower's expense, and it is the legitimate core of the case against forbearance.

The redefault objection

The strongest counterargument, stated properly: modified loans redefault at high rates, so forbearance mostly delays losses while adding cost.

Redefault rates on modifications are genuinely substantial and were particularly poor in the early post-crisis modification programs. Any honest treatment has to engage that. Three responses:

The comparison is wrong. Redefault rate is compared against zero, as though the alternative were a performing loan. The alternative is a charge-off recovering cents on the dollar. A modification that redefaults at 40% still recovers from the 60% that don't, plus partial recovery from those that do — and our arithmetic above showed the break-even sits near 4% completion, not 60%.

Modification design drove much of the early failure. Modifications that capitalized arrears and left payments at or above the original level were solving the lender's accounting problem rather than the borrower's affordability problem. Programs that meaningfully reduced payments performed substantially better. The finding "modifications redefault" is partly a finding about badly designed modifications, and it was used to justify not attempting well-designed ones.

Selection cuts the other way than assumed. Modifications are granted to borrowers already in serious distress — the worst of the book. Comparing their redefault rate to the portfolio average and concluding modifications don't work confuses the treatment with the population.

Where the objection retains force: on structural distress, and on repeat modifications. A second or third modification for the same borrower is strong evidence the original diagnosis was wrong, and continuing is deferral rather than workout. A policy that forbears once and diagnoses properly is different from one that forbears indefinitely.

The moral hazard version deserves the same treatment. Does available forbearance induce strategic non-payment? In principle yes, and the option framework in our companion report says the effect should be strongest where exercise cost is lowest. In practice the frictions are large — applying, disclosing hardship, credit consequences, uncertainty about approval — and the evidence for widespread strategic use of hardship programs is thin. Our reading: the moral hazard concern is real, small in most consumer contexts, and invoked at a magnitude far exceeding its demonstrated size because it is the argument that justifies the choice the measurement asymmetry already favors.

Designing a policy that captures the value

1. Measure the counterfactual you can measure. You can't observe what an enforced borrower would have done — but you can randomize. Offer relief to a random subset of eligible accounts and compare. This is the same holdout logic as reject inference and it converts an unobservable into a measurable one.
2. Separate the diagnosis from the decision. Establish temporary versus structural first, then choose the instrument. Most operations choose the instrument based on delinquency stage, which is a proxy for neither.
3. Match instrument to diagnosis. Deferral for a bridging gap; payment-reducing modification for reduced capacity; disposition where the obligation cannot be supported at all.
4. Fix the incentives. If staff are measured on accounts resolved, they will resolve accounts. Measure on recovery per account and the behavior changes.
5. Require a first payment. The strongest completion predictor, and it screens temporary from structural at low cost.
6. Limit repetition. One workout with a proper diagnosis; a second is evidence the diagnosis failed.
7. Track forgone recovery explicitly as a reported category, even estimated. A number that appears in reporting, however imperfect, competes with the numbers already there.
8. Align servicer compensation with recovery where servicing is outsourced, or the structural bias persists regardless of policy.

Testable implications

Our house view, stated so it can be wrong:

1. Randomized relief offers should outperform enforcement on net recovery for accounts in early-stage delinquency with identifiable temporary causes. This is directly testable and we'd expect the effect to be large.
2. Lenders who service their own loans should forbear more than those using third-party servicers compensated per action, holding portfolio quality constant.
3. Operations measured on recovery per account should forbear more than those measured on resolution speed — a within-firm test available to anyone with two teams on different metrics.
4. Payment-reducing modifications should redefault materially less than arrears-capitalizing ones at equivalent borrower distress.
5. Contact quality should predict workout outcomes more strongly than any model score, because the determining variable is borrower-held information.
6. Strategic use of hardship programs should remain low even where they are well publicized and easy to access. If a lender publicizes relief and sees a large jump in requests from borrowers who could pay, the moral hazard objection is stronger than we've credited.

The first is the one worth running, and the reason is worth stating plainly. Every lender already runs an experiment on this question — they enforce, and they never look at what would have happened otherwise. Adding a randomized relief arm costs the difference in recovery on a small sample and answers a question that currently gets settled by institutional habit.

The broader point is that this is not primarily an argument about being kind to borrowers, though it happens to point that way. It's an argument that an unmeasured cost is being treated as zero, and that the resulting decisions are leaving money on the table at a ratio the arithmetic suggests is not close.

Frequently asked questions

---

This report presents an analytical framework and the authors' interpretation; it is not legal, accounting, or financial advice. Worked figures are stylized illustrations chosen to demonstrate the mechanism rather than benchmarks. Loss mitigation practices, modification accounting treatment, and consumer protection obligations are governed by regulation that varies by product and institution; consult qualified counsel and your accounting advisors.