Surety's 2025 Loss-Ratio Retreat Meets an 8.8-Month Backlog

The Surety & Fidelity Association of America's 2025 industry figures, cited in TSIB's 2026 Surety & Construction Forecast, put the top 100 writers' direct loss ratio at roughly 20.7%, down from 26.4% in 2024 and a run of 24.8% in 2023. ABC's Construction Backlog Indicator was still at 8.8 months in June 2026. A retreating loss ratio against a holding backlog looks like room to ease rate on the strength of the book. It is closer to the opposite: the losses that produced 2024's spike were bonded 18 to 30 months earlier, in a construction cycle that has only gotten larger since, not smaller.

What the Retreat Actually Measures

Surety's loss ratio is a lagging default signal on a contractor credit book, not an attritional frequency metric the way a property or auto loss ratio is. The 2024 spike to a five-year high, confirmed independently by AM Best's February 2025 market segment report at 25.0% through the first nine months of 2024, "the highest in five years" (AM Best, February 2025), was driven by a handful of large-account defaults, not a broad deterioration across the book. TSIB's forecast is blunt about the mechanism shift underneath both years: "subcontractor claims have increased while loss severity has ticked higher," and execution risk, not balance-sheet insolvency, is now the primary driver of project failure (TSIB, 2026). That reframes what the 2025 retreat to 20.7% is actually telling an underwriter. It says the specific large accounts that failed in 2024 have cycled out of the current-year triangle. It says nothing about whether the next cohort of jumbo accounts, bonded against a materially larger 2025 to 2026 contract base, carries the same tail.

That distinction matters because the premium base the 20.7% is applied against has grown far faster than the rate has been tested. TSIB puts surety premium volume up more than 40% since 2021, reflecting both construction market expansion and, in the forecast's own words, "competitive carrier appetite" (TSIB, 2026). A loss ratio computed on a denominator that has grown 40%-plus on construction-cost inflation and bonded-volume growth is not directly comparable across years unless the rate per dollar of exposure has kept pace. If the filed rate per thousand has been flat or only modestly adjusted while contract values inflated, the 2025 ratio is diluted by a bigger, not necessarily better-priced, book.

The Backlog Lag the Current Ratio Can't See

A construction backlog above eight months is a solvency support for the contractor book being bonded today, not a de-risking signal for the bonds already on the books. ABC's backlog reading has held in the 8 to 9 month range through 2026, from 8.1 months in February to 8.8 months in June, with the June reading still 0.1 months above a year earlier (ABC News Releases, 2026). Losses on a bonded project typically emerge 18 to 30 months after the bond is written, once a project has progressed far enough for cost overruns, payment disputes, or a contractor's working-capital strain to surface as a claim. That means the accounts driving 2024's five-year-high loss ratio were bonded in roughly 2022 to 2023, a period when backlog and bonded volume were already climbing. The bonds written against 2025 and 2026's elevated backlog have not yet reached the development age where their loss experience would show up in either year's reported ratio. A backlog that stays historically strong is evidence the exposure base keeps growing, not evidence that the exposure already on the books has de-risked.

The backlog data also shows where that exposure is concentrating. ABC's June 2026 release found contractors under contract for data center work carrying an 11.0-month backlog against 8.5 months for those without, and firms with more than $100 million in revenue running a 12.3-month backlog (ABC, July 2026). In February, the split was similar: 11.2 months for data-center-exposed contractors against 7.6 months for the rest of the market (ABC, March 2026). The book most exposed to jumbo, low-frequency accounts is precisely the segment carrying the longest, least-tested backlog, which is the opposite of where a filed rate built on a blended, aggregate loss ratio wants its confidence to sit.

DCCE as the Quiet Combined-Ratio Drag

Defense and cost containment expense rising from 1.9% of earned premium in 2022 to 2.9% in 2024 looks like a rounding error next to a 20-point swing in loss ratio, but on a line running sub-25% loss ratios, a one-point creep in DCCE is a proportionally larger bite out of the margin than the same point move would be on a 65% loss ratio casualty line. DCCE is also a leading indicator of claim complexity rather than claim count: it rises when more defaults are being litigated, when completion costs are contested by obligees, or when indemnitors are fighting collection under the general indemnity agreement rather than paying it. A DCCE ratio climbing alongside a headline loss ratio that is falling is a signal that the claims still open are harder to close, not that the book generating them has gotten simpler.

Selection and Structure, Not the Filed Rate

Surety's filed rate per thousand is thin, regulated, and slow to move by design; TSIB's forecast describes a market that is "not constrained, but selective," where bonding capacity follows underwriting discipline rather than broad rate increases. That leaves contractor selection, limit-setting, and reinsurance structure as the levers that actually reprice the tail. Underwriting is already tightening on those dimensions: cleaner financials, current work-in-progress reporting, and stronger subcontractor controls are the price of capacity in 2026, not a higher rate per thousand.

Jumbo-account structuring is where that discipline is most visible. Combined 2026 capital-expenditure guidance from Amazon, Microsoft, Google, and Meta approaches $700 billion, with roughly three-quarters aimed at AI infrastructure, and data center construction spending alone topped $52 billion in 2025 (Grit Insurance, 2026, citing hyperscaler capex disclosures). Willis Towers Watson has described surety credit facilities as "a leading mechanism for the data center industry," supporting bond syndications from $5 million to $500 million per project, with prime-contractor capacity needs running $100 million or higher on the largest scopes (WTW, via Grit Insurance, 2026). Single projects at that scale exceed what any one surety wants to hold net, which is why Janus Assurance Re's 2026 capacity outlook expects "localized tightening in contractor classes or regions most exposed to data center and power project clustering," with stronger contractors seeking enlarged single-job limits and expanded aggregate programs rather than broad market softening. Aon's megaproject brief goes further: on the largest jobs, "the market is moving away from traditional 100-100-50 structures" because an $8 billion project cannot feasibly be supported by an $8 billion performance bond, pushing capacity toward co-surety and syndicated programs precisely as reinsurance treaty terms tighten "following losses in renewable energy" (Aon, 2026).

That combination, a filed rate that barely moves and a reinsurance market simultaneously tightening terms on the treaties surety carriers lean on for jumbo capacity, means the single- and aggregate-limit decision on a mega-project account is doing the pricing work the manual rate cannot. A carrier holding its single-job limit flat while contract values on data center and power work climb is implicitly cutting its rate per dollar of tail exposure on every renewal, even if the filed rate per thousand has not changed at all.