Blog

Trend selection in rate filings: how actuaries project future costs

How frequency and severity trends, loss development, exposures and credibility feed a filed auto rate, and how a one-point change in trend moves the request.

Trend selection is the step in a rate filing where an actuary decides how much claim frequency, claim severity and premium per car will change between the period the data covers and the period the new rate will be in force. A trend is an annual percentage rate of change fitted to past data and then applied forward; a one-point difference in the annual trend typically moves the requested rate by about two points when the data is two years from the future policy period, as the worked example below shows. Trend is separate from loss development, which adjusts immature claims for what is still to be paid.

This article is about reading the trend line, not about the full rate indication; for that, see loss ratios and rate filings: how insurers justify an increase. It relies on the actuarial standards that govern the work, on California's published trend regulation as a concrete rulebook, and on federal price data for examples. Your own state's rules will differ in detail, so use state car insurance requirements lookup or the state department of insurance to learn how filings are reviewed where you live.

Trend and development are different things

Actuarial Standard of Practice No. 13 (ASOP 13), Trending Procedures in Property/Casualty Insurance, draws the line that filings depend on. A trending procedure, in its words, evaluates how changes over time affect items such as claim costs, claim frequencies, expenses and exposures, and it works by "analyzing changes between exposure periods (for example, accident years or underwriting years)." It does not cover development, which "estimates changes over time in losses (or other items) within a given exposure period" (Actuarial Standards Board, ASOP 13, last revised May 2011, checked November 2026).

In plain terms:

  • Development asks: the claims reported so far for accident year 2025 will grow as late claims arrive and open claims settle; by how much?
  • Trend asks: compared with accident year 2025, what will an average claim, and the number of claims per car, look like for policies written in the future?

Both are needed because a rate must cover claims that have not yet happened and have not yet been fully paid.

The two pieces of a loss trend: frequency and severity

California's regulation on trend, 10 CCR 2644.7 (checked November 2026), gives a concrete definition of how the two pieces are measured for a filing in that state. Frequency trend is calculated as reported or closed claims divided by exposures. Severity trend is calculated on paid losses divided by closed claims, or on total paid losses including partial payments in previous calendar years on closed claims divided by closed claims. The insurer must submit the calculations on all of those bases and show that its selection is the most actuarially sound.

Frequency is how often a car has a claim. It moves with miles driven, traffic, distracted driving, weather and a driver's willingness to file. Severity is what a claim costs. It moves with repair prices, vehicle values, medical costs, legal costs and the mix of claim types. The pure-premium trend, the cost per car, is the product of the two. A hypothetical example: a frequency trend of -2% and a severity trend of +8% combine to 0.98 times 1.08 = 1.0584, a pure-premium trend of +5.8%, not +6%.

Where the underlying cause differs, the regulation requires separate trends. If the trend within a line varies significantly by subline, policy limits, region of the state or coverage, separate trend factors must be calculated according to that evidence (10 CCR 2644.7(c)). That is why filings show trends by coverage, such as bodily injury, property damage, collision and comprehensive, rather than one number for the whole policy.

Which data window: the first big choice

The regulation requires a California insurer to file the most recent 8, 12, 16, 20 and 24 quarters of rolling calendar-year data excluding catastrophes, then pick the single period it considers most actuarially sound; the Commissioner may require an alternative period (10 CCR 2644.7(b)). Trend factors are based on an exponential curve of best fit, meaning a constant percentage change per year is fitted through the points.

The choice of window is where a trend becomes a judgment. A short window responds to recent changes but is noisy; a long window is stable but slow. Here is an illustration using public price data rather than any insurer's claims. The U.S. Bureau of Labor Statistics' index for motor vehicle maintenance and repair (checked November 2026) shows these December-to-December changes:

Year Change in repair index
2021 +4.8% (309.888 to 324.688)
2022 +13.0% (to 366.960)
2023 +7.1% (to 392.897)
2024 +6.2% (to 417.312)

Annualized, December 2022 to December 2024 is 6.6% a year; December 2020 to December 2024 is 7.7%; December 2018 to December 2024 is 6.3%. The numbers are close here, but the 2022 spike shows how a four-year window carries it and a two-year window does not. An actuary would test several windows and ask which describes the period ahead. This index is a stand-in used for illustration; a company's trend uses its own paid claims, and the repair index includes routine maintenance.

ASOP 13 tells the actuary to consider, when selecting data, "known biases or distortions" such as catastrophic influences, seasonality, coverage changes, nonrecurring events, claim practices, and changes in the distribution of deductibles, risks and policy limits. It also tells the actuary to consider economic and social influences and the timing of those influences (ASOP 13, sections 3.2 and 3.3). The first item is why the California rule excludes catastrophes (hail and flood claims distort an auto trend). The last item is why a company that raised its deductibles or changed its mix of coverages must adjust the data before reading a trend into it.

Exposure: the denominator

A frequency trend is claims divided by exposures, so the choice of exposure base matters. ASOP 53, Estimating Future Costs for Prospective Property/Casualty Risk Transfer and Risk Retention, defines an exposure base as "a basic unit that is used to measure the future cost of risk transfer and risk retention," and notes the unit can vary by element of cost (ASB, ASOP 53, last revised December 2017, checked November 2026). In personal auto the usual unit is the earned car-year: one car insured for one year. A mix change, such as a book that tilts to newer cars or more physical damage coverage, changes the cost per exposure without any change in the world.

ASOP 53 also directs the actuary, in considering trends, to consider past and prospective changes in claim costs, claim frequencies, exposures and premiums and to refer to ASOP 13 for selecting trends (section 3.8.3).

Premium trend

The rate indication compares projected losses with the premium the current rates would produce, so premium also needs a trend. The California rule develops premium trend from company-specific premium per exposure. If the average premium per car has been drifting up because customers are buying higher limits or newer cars, the indication should not treat that drift as a rate increase it has not yet earned; the premium trend handles it.

Credibility and the small-sample problem

Trend lines drawn on little data wobble. The standard that governs this is ASOP 25, Credibility Procedures, which defines credibility as a measure of the predictive value in a given application that the actuary attaches to a particular set of data (ASB, ASOP 25, last revised December 2013, checked November 2026). In a filing, credibility is usually applied as a weight between the company's own trend and a broader benchmark.

California's rule fixes the weights for private passenger auto liability and physical damage: full credibility for loss trend requires 6,000 total claims over the same number of quarters as the data window, for each coverage. Partial credibility is the square root of the ratio of actual claims to 6,000. The complement, the benchmark that takes the remaining weight, is built from the latest California Fast Track data, using the exponential curve of best fit for the same number of quarters (10 CCR 2644.7(d)).

A hypothetical example, with the 6,000 standard applied:

Claims in the window Credibility Z Company trend Benchmark trend Selected trend
6,000 or more 1.00 12% 6% 12.0%
1,500 0.50 12% 6% 9.0%
600 0.32 12% 6% 7.9%

The selected trend is Z times the company trend plus (1 minus Z) times the benchmark. A smaller insurer with 600 claims sees about two-thirds of its own signal replaced by the benchmark. Other states may set different standards; ASOP 25 itself stresses professional judgment in choosing and using a credibility procedure.

Selected versus indicated trend

Analysts in a filing distinguish the trend the data indicates from the one the company selects. The indicated trend is what the fitted curve yields. The selected trend is what the actuary adopts after considering reasonableness. ASOP 13 requires the actuary to evaluate each trending procedure's results for reasonableness and revise where appropriate (section 3.6), and permits presenting the estimate as a point estimate, a range, a point estimate with a margin for adverse deviation, or a probability distribution (section 3.1).

Reasons a selected trend departs from the indicated one include a known change the data predates (a new law, a repair-cost shock), an unusual quarter, or a conflict between measures: paid severity says one thing and incurred severity another. The California rule requires insurers to present the calculations on all bases and demonstrate the selection, which is the regulatory way of forcing the explanation onto paper. ASOP 13 also requires documentation sufficient for another qualified actuary to assess the reasonableness of the work (section 3.8), and ASOP 23, Data Quality, separately covers selecting, reviewing and using data (ASB, ASOP 23, last revised December 2016, checked November 2026).

How reviewers challenge a trend

A regulator's actuary typically tests a filing along these lines. The list is a summary of what the cited standards and rule make checkable, not a description of any one department's checklist.

  1. Is the window supported? Does the filing show the alternatives (in California, all five windows) and explain the selection?
  2. Do the bases agree? Do reported counts and closed counts, and paid and incurred severity, tell the same story?
  3. Is there separate evidence by coverage? Does each coverage have its own trend where limits, regions or coverages differ?
  4. Are distortions removed? Are catastrophes excluded and mix changes adjusted?
  5. Is the credibility defensible? Is a small book leaning on its own noise, or on a relevant benchmark?
  6. Is the selection reasonable against outside data? Does it fit published price indexes, such as the BLS repair and used-car series?

In California the Commissioner may also require an alternative data period under 10 CCR 2644.7(b). Development gets similar scrutiny: 10 CCR 2644.6 requires loss development presented as a triangle based on the dollar-weighted average of ratios for the three most recent accident years, policy years or report years available, with paid losses and case reserves stated separately, and with catastrophes excluded (Cornell LII, 10 CCR 2644.6, checked November 2026).

A one-point change in trend, worked through

The numbers here are hypothetical, defined for illustration.

Assume the experience period is centered two years before the average date the new rate will be in force, so the trending period is 2.0 years. Assume the permissible loss ratio (the share of premium available for claims) is 65%, and the projected loss ratio at a 6% annual trend is 68.0%.

  • At 6% a year, the trend factor is 1.06 squared, 1.1236. Indicated change: 68.0 divided by 65 = 1.0462, or +4.6%.
  • At 7% a year, the factor is 1.07 squared, 1.1449, which is 1.9% higher. The projected loss ratio becomes 68.0 times 1.0190 = 69.3%. Indicated change: 69.3 divided by 65 = 1.0660, or +6.6%.

A one-point trend change moved the indicated change by about two points. If the trending period were 2.5 years, the same one-point change would raise projected losses by about 2.4%. Longer distance between data and policy period means a larger effect, which is why filings that rely on older data are more sensitive to trend selection, and why a filing's trending period is worth checking.

What to open first in a filing

Filing formats and names vary by state and company, and this article could not open a public filing database during research, so the guidance below follows the structure the California regulation describes rather than a particular state's exhibit names.

  1. The summary of the rate indication or overall rate change. It shows the total requested and how much comes from each coverage. Find the coverages with the largest request.
  2. The trend exhibit for those coverages. Look for the window selected, the alternatives, the frequency and severity separately, and the credibility weight. A selected trend far from the indicated one needs an explanation.
  3. The loss development exhibit. Check that the factors apply to immature years only and that paid and incurred agree.
  4. The data exhibits. Look for claim counts: a small count in the window explains noise.

If you are comparing your own renewal increase with a filing, the renewal rate increase checker and how to read car insurance rate studies and average-cost claims are good starting points. For cost drivers behind the numbers, see why car repairs cost so much more than they used to.

Common questions

Is a trend a forecast of what will happen to my premium?

No. It is an estimate of the change in average claim cost and frequency for a group of policies, applied to the rate for all customers. Your premium also depends on your own rating factors and on what regulators approve.

Why does one insurer's trend differ from another's?

Their data, mix of business, claim handling, selected data windows and credibility weights differ. ASOP 13 expects the actuary to choose procedures appropriate to the purpose and data (section 3.4).

Can regulators reject a trend?

In California the Commissioner may require an alternative data period, and under 10 CCR 2644.1 no rate may be approved or remain in effect that is excessive or inadequate; other states have their own standards and processes. A licensed attorney or the state insurance department is the place for specifics about a particular filing.

In the California rule, trend and development data exclude catastrophes. Other jurisdictions may handle them differently.

Key takeaways

  • Trend projects frequency, severity and premium per car from the data period to the future policy period; development adjusts immature claims within a period.
  • The data window, the credibility weight and the selected-versus-indicated judgment are where trends are most open to challenge.
  • In a hypothetical 2-year trending period, one point of annual trend changed the indicated rate by about two points.
  • When reading a filing, start with the overall change by coverage, then the trend exhibit, then development and claim counts.

Sources

Terms used in this guide

Related states

Related guides

Related tools

Latest on this topic