Tariff-Driven Severity Has Entered P&C Triangles: How to Find It

Auto parts prices rose 6.0% year over year in 2025, with June at 5.0% and July at 6.6% as pre-tariff inventory ran out. That increase entered loss triangles not as a trend ramp averaging smoothly across accident years but as a hard break in the 2025 diagonal, produced by effective dates that are known and datable. The diagnostic sits in the 12-to-24 month column, and the treatment is not to weight it away.

Key Takeaways

• A 25% tariff on assembled vehicles took effect April 3, 2025, with 150 auto parts categories following on May 3, after Q1 cost signals were suppressed by pre-tariff inventory.
• The average US import tariff rate went from 2.6% to roughly 13% to 14% by mid-year, a step concentrated in a calendar window rather than spread across accident years.
• Average total cost of repair rose only 1.7% to $4,818, the lowest annual increase since 2017, which is a mix effect rather than a severity result.
• Total-loss frequency hit a record 23.1% as tariff-elevated estimates pushed repairs past the threshold, removing the cheapest repairable claims from the pool.
• A three-year weighted average understates the 12-to-24 factor by about 1.6 points if the 2025 link ratio runs 5% above its predecessors.

The Hard Event Dates

A 25% tariff on assembled vehicles took effect April 3, 2025 under Section 232 of the Trade Expansion Act, with a second tranche covering 150 auto parts categories on May 3. Before either date suppliers and distributors ran down pre-tariff inventory, which suppressed cost signals in Q1. By the third quarter the buffer was gone, and average part prices rose 5.0% in June and 6.6% in July on post-depletion purchasing at tariff-inclusive prices.

The magnitude rules out quiet absorption into a multi-year selection. The average tariff rate on US imports went from 2.6% before 2025 to roughly 13% to 14% by mid-year. A change of that size in input cost, concentrated in a specific window, is the structural break that Berquist-Sherman adjustments exist to handle: as a shift in case reserve adequacy distorts link ratios in prior years, an effective date puts a step in severity that averaging historical data dilutes rather than reflects.

The claims cost consequence is sized. APCIA puts the increase to personal auto claims costs from tariffs on parts from China, Mexico, and Canada at $26 to $52 billion annually, with approximately 60% of US replacement parts sourced from those three countries. The width of that range is policy uncertainty about which rates hold through the development window, which is a scenario problem rather than a data problem.

Why the Headline Severity Understates the Step

Auto physical damage is where the signal is cleanest. APD claims close fast, most within 12 to 18 months, and repair estimates reference parts prices when the invoice is written rather than at policy inception, so tariff costs land in paid amounts without a case reserve lag.

The headline reads the other way. Average total cost of repair reached $4,818 in 2025, up only 1.7% year over year, the lowest annual increase since 2017. Two mix effects produce that. Parts per repair fell from 13.6 in 2024 to 13.0 in 2025, a 4.4% reduction that lowers total repair cost arithmetically while price per part rose 6.0%. And total-loss frequency hit a record 23.1%.

The second explains the first. When a tariff-elevated estimate crosses the economic total-loss threshold, typically 70% to 80% of actual cash value, the carrier pays ACV and the claim leaves the repairable pool. The claims that cross are the ones nearest the threshold, so the surviving repairable pool shifts toward complex or low-ACV vehicles where repair still pays, and average cost of repair falls while the per-part cost that caused the shift keeps running at 6.0%.

The consequence for triangle work is that average paid per claim is the wrong input. Blending repairable and total-loss payments across a distribution that moved this far in one year underweights the cost environment for the repairable segment that drives development at 24 and 36 months. Severity per paid repairable estimate, isolated from the total-loss component, is the series to run.

The factor arithmetic follows directly. A three-year weighted average of the 12-to-24 factor assigns roughly a third of the weight each to 2023, 2024, and 2025, so if the 2025 link ratio runs 5% above its predecessors the average understates expected development by about 1.6 points. On $200 million of APD estimated ultimate at 12 months that is roughly $3.2 million of understated IBNR for one accident year in one line.

Bornhuetter-Ferguson helps only conditionally. Its a priori anchor makes it more resilient to an outlier diagonal, and that resilience disappears if the a priori loss ratio was itself built from 2023 or 2024 experience. Progressive's Susan Griffith named the shape of the exposure before the April effective dates, calling tariffs "one-directional risk to loss costs." A selection weighting pre-tariff and post-tariff experience equally does not carry that asymmetry.

Step or Trend, and What Cannot Yet Separate Them

The prescription above depends on a distinction that one accident year of post-tariff data cannot settle.

Plotting age-to-age link ratios by accident year rather than averaging them is the test. A genuine trend, social inflation or ADAS calibration adoption or medical cost growth, rises roughly linearly across accident years with no discontinuity. A discrete effective date produces stable ratios for 2022, 2023, and 2024 and then a jump at 2025 disproportionate to the year-over-year change around it.

The two readings invert the pricing conclusion. If tariff severity is a new trend with continuing velocity, carriers that filed increases in 2024 are only partly caught up. If it is a level step that holds at the 2025 parts cost without further acceleration, those same carriers may be closer to adequate than emerging data suggests, because the step is already in the most recent accident year at full size. Running both as scenarios is more defensible than committing while post-tariff observations cover one year.

Property lines cannot even isolate the column. Steel mill products rose 20.7% year over year and aluminum mill shapes 33% by 2026, with Canadian softwood lumber carrying a separate 35.2% duty, and property claims develop over 24 to 48 months, so the signal spreads across the 12-to-24 and 24-to-36 periods where catastrophe losses, labor shortages, and local supply disruption compound it.

The cleaner property read is not in the triangle at all. Policies written in 2024 on pre-tariff reconstruction cost indices are closing on tariff-elevated material costs, and when reconstruction runs 20% above the values the policy was written on, severity is higher before any development factor is applied. Development-only methods book that as adverse development when it is underinsurance at original valuation, which is a different problem with a different fix.