Manaz Afzal
Insurance · Pricing

A pricing review that shows which motor customers the rating table overcharges, and which it undercharges

A yearly refit of claim frequency on 678,013 real motor policies, compared with a conventional rating table and tested on policies the models never saw.

2.1×what the table charges the tenth it overprices most, against claims seen
50%of expected claims charged to the tenth it underprices most
0.33Gini for the boosted model; rating table 0.28
38%of claim cost in the largest 1% of claims

Executive summary

A boosted frequency model finds where a motor rating table charges good drivers double and risky drivers half, on 678,013 real policies, without changing the total premium.

2.1×what the table charges the tenth it overprices most, against claims that arrive
50%of expected claims charged to the tenth it underprices most
0.33Gini for the boosted model against 0.28 for the rating table (gain 0.044 to 0.062, 95% CI)

Situation

A small insurer or underwriting agency reviews motor rates once a year with a factor-by-factor rating table. Combinations the table does not see are mispriced: good customers leave at renewal, risky ones stay.

Approach

Claim frequency refitted on 678,013 French motor policies with a Poisson GLM (the rating table) and LightGBM, on a group split so near-identical policies never sit on both sides. 135,920 test policies.

Result

All methods predict the 7,149 claims in total; the gain is in who pays. Deviance falls 8.1% from a flat rate with the boosted model against 4.9% for the table. 73% of policies would move more than 10%, so changes must be capped and phased.

The business problem

A motor insurer prices from a rating table built on factors such as driver age, vehicle and area, set years ago and adjusted in small steps. Pricing teams review it with GLMs of claim frequency and severity, and those reviews often find factors that are weak or out of line, which means some customers subsidise others.

I measured where the rating table and a model of expected claims disagree most, and how much the table overcharges good drivers and undercharges risky ones in those cells.

Team
One pricing analyst, an underwriting manager and a claims lead
Book
20,000 to 700,000 policies; claims in about 1 in 10 policy-years
Systems
Policy admin system export (Guidewire, Duck Creek or in-house), Excel rating tables, a claims system
Constraint
General Insurance Code of Practice, the Privacy Act and pricing that can be explained to a customer; the model proposes, the underwriter signs off

What was built

A yearly pricing review that refits claim frequency on every policy-year, compares the rating table with a boosted model, and lists where they disagree. The underwriter gets the factor tables, a list of segments the table misprices, and the price change each segment would see. Nothing changes in the rating engine until it is signed off.

Policy and claims exportPolicy-years, factors,claim counts and costsAuditDuplicates, caps,missing claim costsRefitRating table andboosted modelCompareWhere they disagree,by segmentUnderwriter reviewAccept, cap orreject each change
Yearly pricing workflow. The model never prices a live quote; it proposes changes to the table.

Results

0.000.100.200.3012345678910Boosted modelClaims seenRating tableTable charges for 0.15;0.07 claimedTable charges for 0.13;0.26 claimedTenth of policies by boosted ÷ table price (1 = table dearest)
Where the two disagree most, the boosted model is right. In the tenth the table overprices most, it charges for 0.15 claims per policy-year and 0.07 arrive. In the tenth it underprices most, it charges for 0.13 and 0.26 arrive. Data: freMTPL2, 678,013 French motor third-party liability policies; test set 135,920 policies

Both ends cost money. The first group are good customers paying double; they are the ones a cheaper competitor wins at renewal. The second group pay about half their expected claims.

0%25%50%75%100%0%20%40%60%80%100%Boosted modelEven splitFlat rateRating table (GLM)Cheapest half of the book:30% of claimsShare of policy-years, lowest predicted risk first
Sorting the book by the boosted model separates risk better (Gini 0.332 against 0.279 for the rating table; difference 0.044 to 0.062, 95% CI).
MethodMean deviance (lower is better)GiniClaims predictedClaims seen
Flat rate0.3278 (0.3219–0.3334)0.0667,2267,149
Rating table (Poisson GLM)0.3118 (0.3064–0.3169)0.2797,2357,149
Boosted model (LightGBM, Poisson)0.3011 (0.2953–0.3066)0.3327,2257,149

All three methods predict close to the 7,149 claims seen in the test year, so the gain is in who pays, not in the total. Mean Poisson deviance falls 4.9% from flat to the table and 8.1% to the boosted model. The boosted model's gain over the table has a 95% interval of 0.0095 to 0.0119, clear of zero.

What the price change looks like

Under 60%13,886 policies10.2%60–70%11,932 policies8.8%70–80%16,797 policies12.4%80–90%19,797 policies14.6%90–100%20,123 policies14.8%100–110%16,456 policies12.1%110–120%11,403 policies8.4%120–130%7,217 policies5.3%130–140%4,356 policies3.2%140% or more13,953 policies10.3%
73% of policies would move more than 10% from the table price. A real rollout caps the change per renewal and phases it over two or three years.

As a pure premium (frequency times capped average claim cost, no expenses or profit), the flat price is 171 euros. The table spreads the middle 80% of customers from 101 to 368; the boosted model from 80 to 330.

Factor by factor

Rating tableClaims seen18–200.2621–220.1923–250.1426–300.1031–350.0936–400.0841–450.1046–500.1051–550.1056–600.0961–650.1066–700.1171–750.1076–800.1081+0.08
The rating table and the claims agree for most ages. The gaps are at 21 to 25 and over 65, where few drivers sit and each extra claim moves the rate.
Bonus-malus 503,615 claims0.08Bonus-malus 51–55409 claims0.08Bonus-malus 56–60479 claims0.11Bonus-malus 61–70730 claims0.14Bonus-malus 71–80683 claims0.13Bonus-malus 81–90421 claims0.14Bonus-malus 91–100568 claims0.21Bonus-malus 101–125213 claims0.33Bonus-malus 126+31 claims0.42Book 0.10
Claims record is the strongest factor: drivers above 100 on the bonus-malus scale claim three to four times the book rate. They are 729 policy-years of 71,663 in the test year, so a table built on averages leaves them underpriced.
Rating factorShare of model gain
Bonus-malus (claims record)25.2%
Vehicle age22.1%
Driver age11.7%
Vehicle make9.8%
Region9.2%
Vehicle power band8.5%
Population density8.3%
Fuel type3.8%
Area code1.4%

Claim costs

Frequency is half the price. Claim costs in this book are dominated by a few large claims: the top 1% of 26,444 paid claims carry 38% of the cost, and the largest is 4,075,401 euros. The average claim is 2,266 euros; the median is 1,172.

1,204 exactly4,792 claims18.1%1,128.12 exactly3,056 claims11.6%1,172 exactly2,071 claims7.8%1,128 exactly831 claims3.1%602 exactly433 claims1.6%1,320 exactly175 claims0.7%
18% of claims are exactly 1,204 euros and 12% exactly 1,128.12. These repeat too often to be measured damage; they look like fixed settlement amounts. A claim-cost model fitted without knowing that learns noise.

The review caps claims at the 99.5th percentile (34,377 euros) and spreads the excess, 25% of all cost, across the book as a loading. An Australian book needs the same check for fixed amounts, such as excess-only claims or recovery settlements.

Data and method

Source
freMTPL2freq and freMTPL2sev, French motor third-party liability (CASdatasets), OpenML 41214 and 41215
Size
678,013 policies, 358,499 policy-years, 36,102 claims; 26,444 claims with a cost
Split
Group split 80/20 on identical risk profiles: 542,093 train, 135,920 test, so near-identical policies never sit on both sides
Audit
38% of rows share a profile with another row; 9,116 policies with claims have no cost record; 9 policies with 5+ claims capped at 4; 1,224 exposures over a year capped at 1

Three methods with exposure as the offset: a flat rate, a Poisson GLM on one-hot factors (the rating table), and LightGBM with a Poisson objective (943 rounds, early stopping on a held-out slice of train). Confidence intervals from 200 bootstrap redraws of the test policies.

A random row split, which lets near-duplicate policies sit on both sides, flatters the boosted model: it explains 8.7% of deviance that way against 8.1% on the group split. The figures above use the group split.

Assumptions and limitations

What an insurer would do next

Technical depth

Deviance is the mean Poisson deviance per policy, weighted by exposure through the offset. The Gini is exposure-weighted, from the ordered Lorenz curve. The double-lift chart sorts test policies by the ratio of boosted to table prediction and compares each with claims seen in tenths. Driver-age partial dependence and factor A/E tables are in the analysis output.

Analysis code (Python, LightGBM, scikit-learn)

Financial impact

About €1.5m of claim cost a year sits uncovered in the most underpriced tenth of a 50,000-policy book

This sizes the mispricing, not a guaranteed gain. How much is recovered depends on caps per renewal and how many customers stay, which needs the insurer’s renewal data.

InputValueSource
Policies in force50,000 mid-sized motor book
Claims charged vs seen, most underpriced tenth0.13 vs 0.26 per policy-yearMeasured
Claims charged vs seen, most overpriced tenth0.15 vs 0.07 per policy-yearMeasured
Average claim cost€2,266Measured freMTPL2, uncapped mean
One policy-year per policyyes study exposure averages under a year

Calculation

  1. Policies in each tenth5,000
  2. Uncharged claims, underpriced tenth650 a year
  3. Claim cost not covered€1,472,900
  4. Overcharge in the overpriced tenth€906,400
  5. Total mispricing at the two ends€2,379,300

Claim cost uncovered in the underpriced tenth€1,472,900

The overcharged tenth is a retention risk rather than a loss; its size is shown for balance. Figures are in euros because the data is French; an Australian book must be refitted on its own history.

Implementation and rollout

One renewal cycle to a capped, signed-off change on one product

  1. Weeks 1 to 2

    Export

    Three years of policy-years, claims and costs from the policy admin and claims systems; audit duplicate profiles and fixed settlement amounts.

    Pricing analyst, Manaz

  2. Weeks 3 to 4

    Refit and review

    Refit, compare with the current table, walk the largest disagreements and factor tables with the underwriter.

    Underwriting manager, Manaz

  3. Next renewal

    Capped pilot

    Apply capped changes to one product or channel, hold the rest as a comparison; fairness and explainability review before release.

    Underwriting manager, compliance

  4. 12 months

    Track

    Loss ratio and retention by segment against the comparison group; refit yearly.

    Pricing analyst

Tools

Guidewire, Duck Creek or in-house policy admin exports, Python, LightGBM and scikit-learn, Excel rating tables for sign-off, A segment report for underwriters

Team

  • Underwriting manager signs off every rate change
  • Pricing analyst owns the table and the yearly refit
  • Claims lead checks claim-cost anomalies
  • Manaz model, review pack and refit pipeline

The timeline is a plan for a typical engagement; it depends on data access and how many systems are involved.

Risks and mitigations

What could go wrong in production, and the control for each

RiskMitigation
Large rate moves drive customers awayCap the change per renewal and phase over two or three years; test on one channel first.
Factors such as age or area raise fairness concernsFairness and explainability review under the General Insurance Code of Practice; the model proposes, the underwriter decides.
Fixed settlement amounts distort claim costsDetect repeated amounts (18% of claims here were exactly €1,204) and treat them separately before any severity model.
Overfitting to near-duplicate policiesGroup split on identical risk profiles; a random split would have overstated the gain (8.7% against 8.1% of deviance).

Key takeaways

Key takeaways

  • The total is right in every method; the money is in who pays, so judge a pricing model on Gini and double-lift, not total claims.
  • Where table and model disagree most, the table charges 2.1× the claims at one end and half at the other.
  • Claims record and vehicle age carry almost half the model’s gain; combinations of factors are where the table goes wrong.
  • Large claims dominate cost (top 1% carry 38%), so frequency and severity need separate treatment.