Three NCCI state filings show the same pattern in the 2026 cycle. Connecticut approved an overall 3.8% voluntary loss cost decrease effective January 1, 2026, its 12th straight year of cuts. Texas accepted NCCI advisory loss costs at an overall 3.8% decrease effective July 1, 2026. South Carolina filed for an April 1, 2026 effective date. In each of these jurisdictions, and across the 36-plus states where NCCI provides ratemaking services, the same structural change rides underneath the headline: loss costs, rates, and Expected Loss Rates (ELRs) now extend from two decimal places to three, effective for filings on and after January 1, 2026.
NCCI frames this as a precision fix, not a rate action. In its decimal extension FAQ, the bureau states it is “not proposing to change any methodology underlying the calculation of loss costs, rates, or ELRs,” and the change is premium-neutral statewide. Connecticut’s public notice for NCCI’s 2026 filing shows the move from concept to live regulatory filing: loss costs, rates, and ELRs extend to three decimals for all classification and statistical codes. Experience rating modification factors stay at two decimals, but the ELRs that feed those mods move to three once a state approves the filing.
That reads like a rating-system cleanup. For self-insured employers, it quietly sharpens a benchmark used well outside the premium invoice.
Why the third decimal exists
The rationale is arithmetic, and it matters for understanding the reserve angle. NCCI explains that the workers compensation system has seen “an ongoing rise in the payroll exposure base, coupled with a decline in loss costs and rates,” a trend that accelerated after the onset of the COVID-19 pandemic. As loss costs fall toward small absolute values, two decimals stop being granular enough: the smallest possible change in a low-rated class is 0.01, which can swamp the true indicated movement. After decimal extension, changes can be as small as 0.001, a tenfold gain in resolution. To keep the methodology internally consistent, NCCI also extended intermediate values such as indemnity and medical pure premiums by one place, from three decimals to four.
The benefit is concentrated, not uniform. NCCI says decimal extension “will be particularly beneficial for classification codes with lower rates because it will minimize rounding limitations that are currently more likely to impact these class codes.” A high-hazard roofing or logging class carrying a loss cost of several dollars per $100 of payroll barely notices a 0.001 refinement. A clerical or professional class sitting near 0.10 or 0.20 can see its filed value shift in a way that two decimals would have rounded away entirely.
The benchmark hides inside the statewide average
The statewide headline is the wrong number to anchor a reserve study on. Connecticut’s 3.8% voluntary decrease is a weighted average that conceals real dispersion by industry group: Manufacturing is down 5.6%, Contracting is down 4.9%, Goods & Services is down 4.0%, Miscellaneous is down 3.6%, and Office & Clerical is actually up 1.3%. In the assigned risk market the spread is wider still; Office & Clerical rises 4.9% even though the assigned risk rate level overall is down roughly 0.4%. A self-insured employer whose payroll concentrates in clerical and professional codes is moving in the opposite direction from the statewide average it might casually cite.
Across NCCI states the dispersion is larger by an order of magnitude. Virginia approved a 7.7% loss cost decrease effective April 1, 2026, while Nevada took a 21.6% increase effective March 1. A multi-state program that blends a single expected-loss assumption across these jurisdictions, rather than applying approved values state by state and class by class, is averaging away the very signal the filings carry.
Quantifying the reserve lever
The mechanism is the expected loss ratio, not case severity on already-reported files. Decimal extension does not make an old back injury more expensive and does not touch a single case reserve; NCCI’s methodology is unchanged and the action is premium-neutral statewide. The effect is entirely prospective, and it runs through one place in the model.
A workers compensation IBNR analysis typically anchors the most recent accident years with an expected-loss benchmark built from payroll, class mix, and an ELR or loss-cost rate. In a Bornhuetter-Ferguson calculation, the ultimate for an immature year is the reported loss plus the expected loss times the unreported percentage. On a green accident year, the reported triangle has earned almost no credibility, so the expected-loss pick carries most of the weight. That is precisely where a two-decimal versus three-decimal ELR, a statewide average versus a payroll-weighted class blend, propagates straight into unpaid claims.
Sizing it requires holding the right thing constant. On any single low-rate class, a 0.001 refinement is immaterial; that is the correct intuition and the reason this is not a rate story. The exposure is multiplicative. A finance, retail, or healthcare employer can run hundreds of millions of dollars of payroll across dozens of clerical, professional, and light-duty classes, all clustered in the low-loss-cost band where rounding distortion was largest. Apply a benchmark that is blunt by 0.001 to 0.01 per $100 of payroll across a nine- or ten-figure payroll base concentrated in those classes, and the aggregate expected-loss anchor for the immature years moves by a margin a reserve study should be selecting deliberately, not inheriting from a stale two-decimal table.
Class mix is the second multiplier. The payroll-weighted average ELR is only as good as the weights. After an acquisition, a divestiture, an outsourcing shift, or a return-to-office change, the distribution of payroll across codes moves, and a model still carrying last year’s weights misstates the expected-loss pick even before the decimal question. Decimal extension simply makes the underlying class values fine enough that a refreshed, correctly weighted blend now resolves differences the old table flattened.
The same logic governs collateral and retained-risk budgets. A surety, fronting carrier, or excess insurer may describe a filing as premium-neutral statewide, but a collateral study cares about the employer’s retained layer and its payroll-weighted class mix, not the bureau’s headline. The reserve and collateral question is identical: does the model use the approved three-decimal values by state and class, or a statewide proxy.
The call
The programs most affected are large self-insured employers with payroll spread thin across many low-rate classes, and any program whose class mix has shifted materially since the last study, especially post-M&A. School districts, municipalities, health systems, multi-state retailers, and professional-services firms fit the first profile; recent acquirers fit the second. High-hazard monoline programs (a logging contractor, a single-trade construction firm) are the least affected, because their loss costs are large enough that the third decimal changes nothing.
A correct refresh should show two things. First, expected loss rates in the 2026 reserve model rebuilt from the approved three-decimal values, by state and by class, not a two-decimal statewide proxy. Second, a sensitivity that swaps payroll-weighted class ELRs in for statewide loss-cost movement and reports the change in the immature accident-year ultimate; for a concentrated low-rate book the gap should be visible, and for a high-hazard book it should be near zero. If the sensitivity is flat across a program with a low-rate, multi-class payroll base, the model is probably still running statewide averages and the refresh has not actually happened. And the expected-loss anchor should not be cut mechanically just because the headline fell; NCCI’s own Medical Inflation Insights warns that 1.8% medical price growth is temporary and likely to revert, while accelerating job growth is pushing the payroll exposure base higher. The decimal refresh is about precision in the pick, not a license to lower it.
The answer belongs in the expected-loss or Bornhuetter-Ferguson section of the study, by state and class, not in the premium exhibit.
Sources
- NCCI, Decimal Extension of Loss Costs, Rates, and Expected Loss Rates, https://www.ncci.com/Articles/Pages/II_Decimal-Extension-Loss-Costs-Rates-Expected-Loss-Rates.aspx
- Connecticut Insurance Department, Notice Of Public Comment For National Council On Compensation Insurance Rate and Loss Cost Filing Effective January 1, 2026, https://portal.ct.gov/cid/department-resources/notices/wc-ncci-rate-filing-for-2026
- PIA Northeast, Conn.: NCCI 2026 WC loss cost decrease of 3.8% approved (industry-group splits), https://blog.pia.org/2025/10/31/conn-ncci-2026-wc-loss-cost-decrease-of-3-8-approved/
- Hartford Business Journal, CT workers’ compensation insurance premiums to drop average of 3.8% in 2026, https://hartfordbusiness.com/article/ct-workers-compensation-insurance-premiums-to-drop-average-of-3-8-in-2026/
- Texas Department of Insurance, Commissioner’s Bulletin B-0001-26, https://www.tdi.texas.gov/bulletins/2026/b-0001-26.html
- NCCI, Summary of the Proposed South Carolina Workers Compensation Loss Cost Filing Effective April 1, 2026, https://www.ncci.com/Articles/Documents/II_StateAdvisoryForumState_SC_2025.pdf