The headline finding in the New York Fed's latest work on homeowner's insurance is that moral hazard costs insurers relatively little. That is the wrong number to take comfort in. Moral hazard is cheap for carriers because deductibles and coverage limits push the cost onto policyholders, who are left holding 29 percent of expected losses even after they buy coverage. The bill has not shrunk. It has been reassigned, and it lands most heavily on lower-credit households and on homes in the places most exposed to severe disasters.
Hyeyoon Jung of the New York Fed and Jaehoon (Kyle) Jung of NYU Stern reach this result by matching millions of homeowner's insurance contracts with property-level exposure and disaster risk, then estimating a Holmström-style contract model. Their October 5, 2026 Liberty Street Economics post follows an April piece that established households are "on the hook" despite being insured. The new contribution is a price for that arrangement: what it costs, how much risk households keep, and who keeps it.
Why a Small Moral Hazard Cost Is Not Good News
The usual intuition says moral hazard is an insurer's problem. A fully covered homeowner skips the roof repair, lets the gutters clog, and the carrier pays for the neglect. On that view, a low estimated cost of moral hazard means the market handles the problem well.
The model shows why that reading misses the transfer. Insurers cannot see how well a home is maintained, so they write contracts that make the owner absorb part of any loss. The deductible works as an incentive device: a household that pays the first slice of damage has a reason to prevent it. Because this device works, the direct cost of moral hazard that survives in the contract is modest.
That modest figure is what remains after the risk has been shifted. The economic weight of moral hazard shows up in the exposure households keep. By the authors' estimate, households retain 29 percent of expected losses through deductibles and coverage limits. That is a large uninsured position held by people who, according to the same model, place a high value on getting rid of risk.
So "moral hazard is cheap" and "households are badly underinsured" are the same finding seen from opposite sides of the contract. Read only the first half and the market looks efficient. Read both and it looks like a cost allocation that nobody priced in public.
The Households That Value Insurance Most Keep the Most Risk
The distribution of that retained exposure is where the result turns uncomfortable. Sorted by FICO decile, four estimated quantities fall steadily as credit scores rise: risk aversion, risk premia paid, the cost associated with moral hazard, and residual uninsured exposure. The estimates already control for property value, so this is not just cheaper homes carrying lower-limit policies.
Lower-FICO policyholders appear more risk averse. They pay more to transfer risk. Their contracts still leave them holding more of it.
The authors call this a paradox, and it is a sharp one. Standard insurance theory would expect the most risk-averse buyers to buy the most complete protection. Here the households with the strongest revealed preference for coverage, and the least financial cushion to absorb a large unexpected loss, end up with the thinnest effective coverage. The higher estimated moral hazard cost in this group points to one mechanism: if insurers see more incentive risk among these policyholders, the contract response is a larger retention, which sends more of the loss back to the buyer who most wanted it gone.
The authors are explicit that these are correlations rather than causal estimates. FICO is a noisy proxy that tracks income, liquidity, local housing stock, and carrier mix as well as financial constraint. The direction of the gradient is consistent across all four measures, though, and the analytical point does not depend on FICO being the cause. It identifies where residual risk concentrates, and the answer is the balance sheets with the least room for it.
Tail Risk Compounds the Exposure
The second concentration is geographic. County-level estimates show that areas with greater exposure to severe disaster losses are also areas where homeowners retain more risk through their contracts. This is the worst pairing available. Retained exposure matters most when losses are large and correlated, and tail-disaster regions are exactly where losses arrive together and large.
A deductible scaled to the value of a home functions very differently in a county facing routine hail claims than in one facing a hurricane or wildfire that hits every policyholder on the same street at once. The incentive argument for retention is also weakest in these places. A homeowner can clear brush or reinforce a roof, but no maintenance effort stops a severe regional catastrophe. In the highest-hazard regions, more of the retained exposure is risk that the policyholder cannot prevent and still has to bear.
The maps show wide variation in all four quantities, reflecting differences in underlying property risk, in the households living there, and in the insurers writing policies. These factors overlap. Lower-credit households in high-hazard counties sit at the intersection of both gradients, and the post does not separate how much of the combined exposure comes from each.
How Carrier Capital Sorts the Risk
Insurer finances add a third layer. Carriers with weaker risk-based capital ratios tend to insure riskier properties and collect higher dollar risk premia. The risk premium as a share of total premium is roughly flat across insurer financial constraint. Constrained carriers are not obviously earning fatter margins; they are matched with riskier homes and different policyholders.
This matters for how the market should be read. Thinly capitalized insurers end up holding the riskier book, in the same places and plausibly with the same households who already retain the most exposure. The authors frame this as endogenous matching: households, properties, and insurers sort into one another rather than being assigned at random. Any analysis that studies pricing alone, without asking which carrier holds which risk, will misread the market.
Listed carriers offer a separate, partial lens on the same pressure. The study does not analyze them, and the link here is an inference rather than a finding. Carrier earnings across the recent catastrophe cycle show how quickly underwriting results swing when tail losses arrive, and how carriers repair them through rate and terms.
When margins are rebuilt partly through tighter terms rather than price alone, some of the repair is the same mechanism the paper describes: risk handed back to the policyholder through the deductible. The paper's 29 percent figure is a cross-section, not a time series, so it cannot show whether retention has risen through this cycle. That is the obvious next question.
The Case for Keeping Households on the Hook
The strongest counterargument is the authors' own, and it deserves full weight. Retention is not a design flaw. It is the device that keeps moral hazard cheap. Remove the deductible, and the incentive to maintain and protect the property weakens. The cost of moral hazard that now looks small could grow, and premiums would rise for everyone, including the lower-credit households the reform was meant to help.
The validation tests support the claim that the incentive channel is real rather than a modeling artifact. Homeowners with lower loan-to-value ratios, who have more equity at stake, show lower estimated moral hazard costs after controls. States that require inspections or verification, which narrow the information gap between insurer and homeowner, also show lower costs. Neither test is causal, and the authors say so. Both move in the direction the mechanism predicts.
The inspection result points to an alternative policy lever. If verification reduces moral hazard, it could stand in for some of the retention. Carriers that can observe maintenance need fewer financial incentives to enforce it, and could in principle offer lower deductibles without the incentive loss. That trade is cleaner than a coverage mandate, because it reduces the information problem instead of overriding the contract response to it.
The results hold up under the robustness checks reported. Dropping flood exposure, which standard homeowner's policies usually exclude, has little effect. Adjusting for non-disaster claims by comparing modeled damages with realized claims does not change the conclusion. Extensions for insurer competition and for lenders' requirement that borrowers carry coverage preserve the household, location, and insurer patterns.
Where the Retained Loss Goes Next
The mandatory-coverage detail is the thread worth following beyond the insurance market. Most of these households are mortgage borrowers who must buy insurance as a loan condition. A homeowner who retains a large share of a disaster loss and lacks liquid savings has a short list of options after the event: drain savings, take on high-cost credit, or fall behind on the mortgage. Concentrated retained exposure among lower-FICO borrowers in tail-risk counties is, plausibly, a mortgage credit exposure that standard credit models do not tag as insurance risk. The authors do not draw this link. It follows from the population they studied, and it is the channel through which a household insurance gap could reach lenders and mortgage guarantors.
The authors' companion post will use the model to estimate what happens if insurers must offer full coverage. That result decides the policy reading. If full insurance pushes the cost of moral hazard up enough to price lower-credit households out or sharply raise their premiums, the current allocation is defensible as the lesser cost, and the policy focus should shift to verification regimes that reduce the information gap directly. If the cost of moral hazard stays modest under full coverage, the 29 percent retention looks less like a necessary incentive and more like a burden placed on the households least equipped to carry it.
