RateCompare — Personal Auto Rate-Filing & Loss-Trend Intelligence
Approved rate actions, loss-trend exhibits, and carrier selections on free public data — CDI approved dispositions, NAIC experience, FRED driver projections, an SEC nowcast — every number with a source and an as-of date. Built for the CAS 2026 Ratemaking paper. The pipeline is state-agnostic: filings from any state, retrieved manually through SERFF's intended workflow, run through it unchanged; California is shown because its channels permit automated retrieval within terms of service. Loss trend here is the worked case study of justification-element extraction — the same recipe extracts any rating exhibit a filing discloses. Read-only dashboard — refreshed weekly by an offline pipeline that harvests new CA filings and extracts them with a local open-weights model (no hosted API).
Disclaimer — please read before relying on anything on this page. This is a research/educational dashboard built for the CAS 2026 Ratemaking paper.
- No warranty. All data and calculations are provided "as is," without warranty of accuracy, completeness, or timeliness. Every figure carries an as-of date and reflects public data that lags real experience.
- Not actuarial advice. Nothing here is an actuarial work product, opinion, or rate indication, and it is not a substitute for the professional judgment of a qualified actuary working under the applicable Actuarial Standards of Practice (ASOPs). Several Werner & Modlin steps are deliberately not implemented.
- Not legal advice.
- Antitrust / information-sharing caution. This tool aggregates only rate figures that carriers have already filed or published publicly; it is a transparency aid, not a mechanism for competitor coordination or price signaling, and must be used consistently with antitrust and unfair-trade-practices law.
- Data-source terms respected. Public sources (CA CDI public notices, NAIC reports, SEC EDGAR, FRED, CAS Schedule P) are used per their public terms; SERFF Filing Access is never scraped; no personal or confidential data is stored.
- Your responsibility. Although this tool was built with care to respect all applicable legalities, it is the user's sole responsibility to ensure that any use complies with all applicable state, federal, and other laws and regulations, including insurance rating and filing laws.
- Open source under MPL-2.0 for full auditability and reproducibility.
Historical / teaching view — not the current product. Where every carrier sat on realized Personal Passenger Auto loss ratios in the public CAS Schedule P window (1998–2007) and how the market moved — plus the extracted rate-filing corpus ranked by indicated-minus-filed 'rate need'. Current-vintage loss-trend analytics live on the Loss Trend tab; this view anchors the CAS paper's historical context and the SERFF extraction corpus.
Carriers ranked by realized ultimate loss ratio
Extracted rate-filing corpus (by rate-need gap)
Historical benchmark — CAS Schedule P (1998–2007) — data, method & assumptions
Data source & as-of
- CAS Loss Reserve Database / Schedule P (casact.org public CSVs); accident years 1998–2007, 10 development years; join key GRCODE = NAIC company code.
- SERFF rate-filing corpus (manually retrieved CA CDI public filings — this tool never scrapes SERFF).
How it's calculated
- Realized ultimate loss ratio = incurred losses ÷ earned premium (net), taken at development lag 10.
- Market index = median / p25 / p75 of carrier ultimate loss ratios by accident year.
- Rate-need gap = indicated − approved (filed) rate change, ranked descending.
Assumptions
- Ultimate = the development-lag-10 value (assumes full development by year 10; no LDF or tail factor).
- Loss ratio is net of reinsurance; the ceded ratio is shown only as a reinsurance proxy.
- No premium on-leveling and no premium/loss trend applied; window frozen at AY 1998–2007 (historical/teaching context, not a current indication).
- Loss ratios above 3.0 are clipped for display (tiny-premium carriers).
Werner & Modlin grounding
Ch. 6 (losses & their development — here a fixed 10-year proxy, not chain-ladder) feeding the Ch. 8 loss-ratio method.
Disclaimer — please read before relying on anything on this page. This is a research/educational dashboard built for the CAS 2026 Ratemaking paper.
- No warranty. All data and calculations are provided "as is," without warranty of accuracy, completeness, or timeliness. Every figure carries an as-of date and reflects public data that lags real experience.
- Not actuarial advice. Nothing here is an actuarial work product, opinion, or rate indication, and it is not a substitute for the professional judgment of a qualified actuary working under the applicable Actuarial Standards of Practice (ASOPs). Several Werner & Modlin steps are deliberately not implemented.
- Not legal advice.
- Antitrust / information-sharing caution. This tool aggregates only rate figures that carriers have already filed or published publicly; it is a transparency aid, not a mechanism for competitor coordination or price signaling, and must be used consistently with antitrust and unfair-trade-practices law.
- Data-source terms respected. Public sources (CA CDI public notices, NAIC reports, SEC EDGAR, FRED, CAS Schedule P) are used per their public terms; SERFF Filing Access is never scraped; no personal or confidential data is stored.
- Your responsibility. Although this tool was built with care to respect all applicable legalities, it is the user's sole responsibility to ensure that any use complies with all applicable state, federal, and other laws and regulations, including insurance rating and filing laws.
- Open source under MPL-2.0 for full auditability and reproducibility.
Approved rate actions married to realized loss ratios — approved dispositions only. For each of the ten largest CA personal-auto writers: the median approved rate change per approval year (% Rate Chng Appvd from the CDI Approval-and-Closed lists — never the notice-time ask) under the group's NAIC direct loss ratio vs the premium-weighted market. The small multiples show the cycle carrier by carrier: the 2021–22 approval freeze, the 2023–24 catch-up wave, and who was granted rate before vs after their loss ratio peaked.
Approved actions vs loss-ratio path
2019 | 0 | +14.0% | 66.3% | 64.3% |
Asked vs. granted (the regulatory haircut). Median requested vs approved change and the share of approved filings trimmed below the ask, per line and approval year — state-published, no extraction:
Requested vs approved, CA auto (CDI Approval-and-Closed lists)
Commercial auto | 2020 | 128 | +13.4% | +12.5% | 19% |
Rate actions vs. results (approved dispositions) — data, method & assumptions
Data source & as-of
- CA CDI Approval and Closed lists (YTD Excel editions, 2020–present): per filing, the requested AND the approved rate change, status, and closed date. Approved filings only — never the notice-time ask.
- NAIC Market Share Reports (frozen editions 2019–2024): per-group direct loss ratios, top-10 writers per state-line; join key = CDI 'Grp #' = NAIC group code.
How it's calculated
- Bars = median % Rate Chng Appvd per approval year for the group's STATUS=APPROVED personal-auto rate filings (rule/form-only filings excluded).
- Lines = the group's NAIC direct loss ratio and the premium(DPE)-weighted market loss ratio, by calendar year.
- Asked-vs-granted table = median requested vs approved and the share of filings trimmed below the ask, per line × approval year.
Assumptions
- Approval-year dating: a filing counts in the year the CDI closed it as approved, not the year it was requested or took effect.
- The 'market' loss ratio covers only the report's top-10 writers per state-line (the NAIC publication's truncation).
- Calendar-year direct loss ratios (not developed accident-year); year-over-year movements carry development noise.
Werner & Modlin grounding
Ch. 8 loss-ratio framing for the realized side; the approved-change series is a regulatory disposition, an input Werner & Modlin treat as given (Ch. 13 'other considerations' / regulatory constraints).
Disclaimer — please read before relying on anything on this page. This is a research/educational dashboard built for the CAS 2026 Ratemaking paper.
- No warranty. All data and calculations are provided "as is," without warranty of accuracy, completeness, or timeliness. Every figure carries an as-of date and reflects public data that lags real experience.
- Not actuarial advice. Nothing here is an actuarial work product, opinion, or rate indication, and it is not a substitute for the professional judgment of a qualified actuary working under the applicable Actuarial Standards of Practice (ASOPs). Several Werner & Modlin steps are deliberately not implemented.
- Not legal advice.
- Antitrust / information-sharing caution. This tool aggregates only rate figures that carriers have already filed or published publicly; it is a transparency aid, not a mechanism for competitor coordination or price signaling, and must be used consistently with antitrust and unfair-trade-practices law.
- Data-source terms respected. Public sources (CA CDI public notices, NAIC reports, SEC EDGAR, FRED, CAS Schedule P) are used per their public terms; SERFF Filing Access is never scraped; no personal or confidential data is stored.
- Your responsibility. Although this tool was built with care to respect all applicable legalities, it is the user's sole responsibility to ensure that any use complies with all applicable state, federal, and other laws and regulations, including insurance rating and filing laws.
- Open source under MPL-2.0 for full auditability and reproducibility.
The loss-trend exhibit. For a Personal Auto coverage and state: the experience rows (exposure, incurred claim count, incurred losses, frequency, severity, pure premium, loss ratio) and a multi-window exponential trend block (annual windows, clamped to available history; longer windows shown n/a). Source: NAIC Auto Insurance Database, stitched across editions into a current-vintage panel — all-carrier state aggregate; per-carrier detail comes from CA filings. Export to Excel gives a filing-workpaper sheet.
Experience by accident year
Exponential annual trend by trailing window
Layer 2 — exogenous driver projection (current vintage)
Selections — carrier SERFF trend vs market
Loss Trend Exhibit — data, method & assumptions
Data source & as-of
- NAIC Auto Insurance Database Report (all-carrier state aggregate; incurred losses and incurred claim counts), stitched across editions into a current-vintage panel (a year in two editions is taken from the most recent; restatement delta logged).
- FRED driver series (repair/parts/used-car/medical CPI, VMT, gas) — Layer 2.
- SEC EDGAR public-carrier quarterly loss ratios — Layer 3 nowcast.
- SERFF-filed carrier selections (CA CDI public filings) — optional overlay.
How it's calculated
- Frequency = claims ÷ exposure; severity = losses ÷ claims; pure premium = losses ÷ exposure = frequency × severity.
- Trend = exp(slope of ln(y) regressed on year) − 1, over each trailing window (windows longer than available history shown n/a).
- Layer 2: OLS of the metric's annual %Δ on a driver's annual %Δ, projected off the latest driver YoY.
- Layer 3: composite public-carrier loss-ratio move since the latest NAIC accident year.
Assumptions
- NAIC losses are incurred (not paid) and lag ~2–3 years; reserve development is not removed.
- Exponential trend assumes a constant % change per year and needs strictly positive values.
- The driver projection is a single-predictor fit over a short annual overlap — a directional read, not a credibility-weighted trend.
- The SEC nowcast is company-wide (all lines), not coverage-level frequency/severity — directional only.
Werner & Modlin grounding
Ch. 6 — loss trend. The engine reproduces the W&M Appendix A Personal Auto exhibit (p352 / Table 6.14) within ±0.15pp in the test suite. The driver and nowcast layers are extensions beyond W&M.
Disclaimer — please read before relying on anything on this page. This is a research/educational dashboard built for the CAS 2026 Ratemaking paper.
- No warranty. All data and calculations are provided "as is," without warranty of accuracy, completeness, or timeliness. Every figure carries an as-of date and reflects public data that lags real experience.
- Not actuarial advice. Nothing here is an actuarial work product, opinion, or rate indication, and it is not a substitute for the professional judgment of a qualified actuary working under the applicable Actuarial Standards of Practice (ASOPs). Several Werner & Modlin steps are deliberately not implemented.
- Not legal advice.
- Antitrust / information-sharing caution. This tool aggregates only rate figures that carriers have already filed or published publicly; it is a transparency aid, not a mechanism for competitor coordination or price signaling, and must be used consistently with antitrust and unfair-trade-practices law.
- Data-source terms respected. Public sources (CA CDI public notices, NAIC reports, SEC EDGAR, FRED, CAS Schedule P) are used per their public terms; SERFF Filing Access is never scraped; no personal or confidential data is stored.
- Your responsibility. Although this tool was built with care to respect all applicable legalities, it is the user's sole responsibility to ensure that any use complies with all applicable state, federal, and other laws and regulations, including insurance rating and filing laws.
- Open source under MPL-2.0 for full auditability and reproducibility.
How a carrier used frequency / severity / pure-premium trend across its own rate filings (CA CDI public filings, selected trends extracted from each filing's exhibits), filterable by year, with the carrier's latest available data shown.
Selected trends by filing (newest first) — per coverage (Werner & Modlin). Rate impact is TOTAL company (all coverages); approved-level preferred over the requested/public-notice ask.
Carrier trend history — data, method & assumptions
Data source & as-of
- SERFF-filed rate filings retrieved from CA CDI public filings (this tool never scrapes SERFF); per-coverage selected frequency/severity/pure-premium trends extracted from each filing's exhibits.
How it's calculated
- For each filing (newest first) the carrier's selected/projected trend per coverage is shown, company-specific separated from Fast-Track/industry.
- Rate impact is the total-company overall %; the approved level is preferred over the requested / public-notice ask.
- Per-coverage frequency × severity → pure-premium identity is checked (Ch. 6).
Assumptions
- Trend disclosure is missing-not-at-random: many carriers withhold trend exhibits, so this is a disclosed subset, not a census.
- Carrier/subsidiary rollup uses NAIC group code with a name fallback — fuzzy and can mis-group.
- Overall rate impact is total company, not per-coverage.
Werner & Modlin grounding
Ch. 6 — per-coverage loss-trend selection. Reflects what carriers publicly filed, not an independent re-derivation.
Disclaimer — please read before relying on anything on this page. This is a research/educational dashboard built for the CAS 2026 Ratemaking paper.
- No warranty. All data and calculations are provided "as is," without warranty of accuracy, completeness, or timeliness. Every figure carries an as-of date and reflects public data that lags real experience.
- Not actuarial advice. Nothing here is an actuarial work product, opinion, or rate indication, and it is not a substitute for the professional judgment of a qualified actuary working under the applicable Actuarial Standards of Practice (ASOPs). Several Werner & Modlin steps are deliberately not implemented.
- Not legal advice.
- Antitrust / information-sharing caution. This tool aggregates only rate figures that carriers have already filed or published publicly; it is a transparency aid, not a mechanism for competitor coordination or price signaling, and must be used consistently with antitrust and unfair-trade-practices law.
- Data-source terms respected. Public sources (CA CDI public notices, NAIC reports, SEC EDGAR, FRED, CAS Schedule P) are used per their public terms; SERFF Filing Access is never scraped; no personal or confidential data is stored.
- Your responsibility. Although this tool was built with care to respect all applicable legalities, it is the user's sole responsibility to ensure that any use complies with all applicable state, federal, and other laws and regulations, including insurance rating and filing laws.
- Open source under MPL-2.0 for full auditability and reproducibility.
Quarterly per-carrier loss ratio (SEC EDGAR XBRL) for public auto insurers vs a tracked-carrier composite market line, plus the carrier's latest qualitative frequency/severity commentary from its 10-Q. Clean quarterly numeric freq/sev per carrier isn't publicly disclosed — this is the closest available.
Quarterly loss ratio
Quarterly carrier trends — data, method & assumptions
Data source & as-of
- SEC EDGAR XBRL (10-K/10-Q) for public auto insurers; quarterly loss ratios plus qualitative frequency/severity commentary from the latest 10-Q MD&A.
How it's calculated
- Single-quarter values reconstructed from cumulative YTD by differencing (Q4 from the 10-K full year).
- Loss ratio = incurred losses ÷ premium earned per quarter.
- Market line = unweighted mean loss ratio across the tracked ticker set.
Assumptions
- Figures are largely company-wide, not coverage-level freq/sev — a directional nowcast only.
- The composite is an unweighted mean over a small, hand-picked set of public carriers (not premium-weighted, not the whole market).
- Clean per-carrier numeric quarterly freq/sev is not publicly disclosed — this is the closest public proxy; ISO Fast Track / Verisk is the paid one.
Werner & Modlin grounding
Extension beyond Werner & Modlin — a public-carrier nowcast, not a textbook ratemaking procedure; grounded in the Ch. 8 loss-ratio framing.
Disclaimer — please read before relying on anything on this page. This is a research/educational dashboard built for the CAS 2026 Ratemaking paper.
- No warranty. All data and calculations are provided "as is," without warranty of accuracy, completeness, or timeliness. Every figure carries an as-of date and reflects public data that lags real experience.
- Not actuarial advice. Nothing here is an actuarial work product, opinion, or rate indication, and it is not a substitute for the professional judgment of a qualified actuary working under the applicable Actuarial Standards of Practice (ASOPs). Several Werner & Modlin steps are deliberately not implemented.
- Not legal advice.
- Antitrust / information-sharing caution. This tool aggregates only rate figures that carriers have already filed or published publicly; it is a transparency aid, not a mechanism for competitor coordination or price signaling, and must be used consistently with antitrust and unfair-trade-practices law.
- Data-source terms respected. Public sources (CA CDI public notices, NAIC reports, SEC EDGAR, FRED, CAS Schedule P) are used per their public terms; SERFF Filing Access is never scraped; no personal or confidential data is stored.
- Your responsibility. Although this tool was built with care to respect all applicable legalities, it is the user's sole responsibility to ensure that any use complies with all applicable state, federal, and other laws and regulations, including insurance rating and filing laws.
- Open source under MPL-2.0 for full auditability and reproducibility.
Methodology & Sources
Provenance is the moat: every number is traceable to a free public source with an as-of date. This page is generated from the data actually loaded.
Public data inventory
| Source | Grain | Fields | Vintage | Status |
|---|---|---|---|---|
| NAIC Auto Insurance Database Report | State × coverage (all-carrier aggregate) | earned premium, earned exposure, incurred losses, incurred claim count | Annual, ~2–3y lag | in use |
| FRED (Federal Reserve Bank of St. Louis) | National monthly/weekly driver series | repair/parts/used-car/medical CPI; vehicle-miles-traveled; gas price | Monthly/weekly, current | in use |
| SEC EDGAR (10-K/10-Q/XBRL) | Per public carrier | personal-auto segment premium/losses/loss ratio; MD&A freq/sev commentary | Quarterly (Progressive monthly) | in use |
| [SERFF rate filings (user-uploaded)](manual download — this tool never scrapes SERFF) | Per filing (carrier × state × line) | indicated/selected rate change, selected trends, experience series | As filed | in use |
| [California / Texas / Massachusetts DOI experience reports](insurance.ca.gov ; tdi.texas.gov ; cardata.org) | Sub-state (county/territory/program) | earned premium, exposure, claims, incurred losses | Annual | available — not yet ingested |
Formulas
- Frequency = incurred claim count ÷ earned exposure
- Severity = incurred losses ÷ incurred claim count
- Pure premium = incurred losses ÷ earned exposure (= frequency × severity)
- Loss ratio = incurred losses ÷ earned premium
- Exponential trend = exp(slope of ln(y) regressed on year) − 1, over each trailing window (windows longer than history shown n/a)
- Driver projection = OLS of experience-metric annual %Δ on driver annual %Δ over overlapping years; projected off the latest driver YoY
NAIC editions stitched into the current panel
- ADB 2016-2018 — accident years 2016–2017; accessed 2026-06-14
- ADB 2018-2020 — accident years 2018–2019; accessed 2026-06-14
- ADB 2020-2022 — accident years 2020–2022; accessed 2026-06-14
Driver series (Layer 2)
- CPI: Used Cars & Trucks (
CUSR0000SETA02, severity) — FRED; fixture as of 2026-06-14, live whenFRED_API_KEYis set. - CPI: Motor Vehicle Maintenance & Repair (
CUSR0000SETD, severity) — FRED; fixture as of 2026-06-14, live whenFRED_API_KEYis set. - CPI: Motor Vehicle Parts & Equipment (
CUSR0000SETC, severity) — FRED; fixture as of 2026-06-14, live whenFRED_API_KEYis set. - CPI: Medical Care Services (
CUSR0000SAM2, severity) — FRED; fixture as of 2026-06-14, live whenFRED_API_KEYis set. - Vehicle Miles Traveled (12-mo total) (
TRFVOLUSM227NFWA, frequency) — FRED; fixture as of 2026-06-14, live whenFRED_API_KEYis set. - US Regular Gas Price (
GASREGW, frequency) — FRED; fixture as of 2026-06-14, live whenFRED_API_KEYis set.
Werner & Modlin grounding (Basic Ratemaking, CAS 5th ed.)
| Calculation | Werner & Modlin grounding |
|---|---|
| Loss trend — frequency, severity, pure premium; exponential fit exp(slope)−1; freq×sev→pp identity | Ch. 6 (Losses & LAE — trend). Engine reproduces the W&M Appendix A Personal Auto exhibit (p352 / Table 6.14) within ±0.15pp in the test suite. |
| Multi-window trailing trend selection | Ch. 6 — trend selected across experience windows. |
| Realized ultimate loss ratio = incurred losses ÷ earned premium at development lag 10 (Schedule P) | Ch. 6 (loss development) — implemented as a fixed 10-year proxy for ultimate, NOT a chain-ladder projection. |
| Market loss-ratio index (median / p25 / p75 of realized ultimate LR) | Ch. 8 loss-ratio method framing; realized losses developed per Ch. 6. |
| Rate-need gap = indicated − approved (filed) rate change | Ch. 8 (overall rate-level indication) — indicated vs. adopted rate change. |
| Driver overlay (FRED) and SEC quarterly nowcast | Extensions beyond W&M — external economic drivers and a public-carrier nowcast; W&M selects trend judgmentally. |
Werner & Modlin methods deliberately NOT implemented
- On-leveling / current rate level / parallelogram method (Ch. 5) — historical premium is not brought to current rate level; loss-ratio reads are off-level.
- Chain-ladder loss-development triangle (Ch. 6) — ultimate is proxied by the development-lag-10 diagonal; no link ratios or tail factor.
- Credibility weighting (Ch. 12) — trends and indications are shown on their own data; no complement of credibility is blended.
- Expense-derived permissible loss ratio (Ch. 7–8) — the diagnostic's PERMISSIBLE_LR (0.70) and ASSUMED_CURRENT_LR (0.78) are illustrative placeholders, not a filed expense analysis.
Limitations (read these)
- NAIC ADB is annual and lags ~2–3 years; the nowcast layer flags movement since the latest edition but cannot replace current experience.
- ADB losses are incurred (not paid) and the claim count is incurred (not closed) — labelled as such; reserve development is not removed.
- NAIC reuses one rolling URL per edition; history is stitched from archived editions, and a year present in two editions is taken from the most recent edition (the restatement delta is logged).
- The driver projection is a single-predictor fit over a short annual overlap — an illustrative directional read with r and n shown, NOT a credibility-weighted trend.
- SEC figures are largely company-wide, not coverage-level freq/sev; treated as a directional nowcast only.
- Clean per-carrier quarterly frequency/severity is not publicly disclosed — this tool does not claim to provide it. ISO Fast Track / Verisk would be the paid source.
Disclaimer — please read before relying on anything on this page. This is a research/educational dashboard built for the CAS 2026 Ratemaking paper.
- No warranty. All data and calculations are provided "as is," without warranty of accuracy, completeness, or timeliness. Every figure carries an as-of date and reflects public data that lags real experience.
- Not actuarial advice. Nothing here is an actuarial work product, opinion, or rate indication, and it is not a substitute for the professional judgment of a qualified actuary working under the applicable Actuarial Standards of Practice (ASOPs). Several Werner & Modlin steps are deliberately not implemented.
- Not legal advice.
- Antitrust / information-sharing caution. This tool aggregates only rate figures that carriers have already filed or published publicly; it is a transparency aid, not a mechanism for competitor coordination or price signaling, and must be used consistently with antitrust and unfair-trade-practices law.
- Data-source terms respected. Public sources (CA CDI public notices, NAIC reports, SEC EDGAR, FRED, CAS Schedule P) are used per their public terms; SERFF Filing Access is never scraped; no personal or confidential data is stored.
- Your responsibility. Although this tool was built with care to respect all applicable legalities, it is the user's sole responsibility to ensure that any use complies with all applicable state, federal, and other laws and regulations, including insurance rating and filing laws.
- Open source under MPL-2.0 for full auditability and reproducibility.