Updating Experience Ratings As Exposure Changes

Why an Experience Rating Must Be Updated Carefully

An experience-rated premium is based on both the risk’s observed loss cost and the amount of experience supporting that observation. When new exposure and claims accrue, both can change: more exposure generally gives the risk’s experience greater credibility, while the added losses may raise or lower its observed loss cost. A correct update therefore recalculates the loss rate and credibility weight separately before combining them.

The calculations below use the Bühlmann-Straub model, which is designed for experience periods with unequal exposure. The examples treat losses as being on a common cost level and use the same collective complement and credibility parameters before and after the update. If those bases or parameters change, that is a separate recalibration, not merely an update for new experience.

Definition: Let $w_j$ be the exposure in period $j$, and let $S_j$ be the aggregate losses for that period. For total exposure $W=\sum_j w_j$ and total losses $S=\sum_j S_j$, the exposure-weighted observed loss rate is $\bar X=S/W$. Let $\mu$ be the collective expected loss rate, $v$ the expected process variance per unit exposure, and $a$ the variance of the underlying risk means. The Bühlmann-Straub credibility parameter is $K=v/a$, and the credibility weight is $Z=W/(W+K)$. The credibility premium per exposure unit is $P=Z\bar X+(1-Z)\mu$.

The exposure total is not the number of periods unless every period has one exposure unit. Likewise, the observed loss rate is not the simple average of period rates when the periods have different exposure. Aggregate exposure and losses first; divide total losses by total exposure only after the totals are combined.

What Changes When New Experience Arrives?

Suppose a risk has accumulated exposure $W$ and losses $S$, then receives additional exposure $\Delta W$ and losses $\Delta S$. The updated quantities are

W_{\text{new}}=W+\Delta W,\qquad S_{\text{new}}=S+\Delta S,\qquad \bar X_{\text{new}}=\frac{S+\Delta S}{W+\Delta W},\qquad Z_{\text{new}}=\frac{W+\Delta W}{W+\Delta W+K}.

With the collective rate and $K$ held fixed, the updated premium is $P_{\text{new}}=Z_{\text{new}}\bar X_{\text{new}}+(1-Z_{\text{new}})\mu$. A zero-loss addition still increases exposure and reduces the observed rate; it does not mean that the risk’s total losses are zero. Conversely, new losses affect the updated rate through $\Delta S$, not through the credibility formula itself.

Worked Example: Adding Exposure With No Losses

A hypothetical risk has two exposure units and aggregate losses of $300. The collective expected loss rate is $200 per exposure unit, and $K=2$. It then earns one additional exposure unit with no losses. Find the old and new credibility premiums per exposure unit.

Method. Apply Bühlmann-Straub to the combined exposure and losses, using the unchanged collective parameters $\mu=\$200$ and $K=2$.

Before the update. The observed rate is $300/2=\$150$. The credibility weight is $Z=2/(2+2)=1/2$. Thus

P_{\text{old}}=\frac12(150)+\frac12(200)=75+100=\$175.00.

After the update. Total exposure is $2+1=3$, and total losses remain $300. The updated observed rate is $300/3=\$100$. The updated credibility is $Z=3/(3+2)=3/5$, so

P_{\text{new}}=\frac35(100)+\frac25(200)=60+80=\$140.00.

The premium falls from $175.00 to $140.00. The additional no-loss exposure lowers the observed rate and increases the weight placed on that lower rate. Common mistake avoided: reporting the old premium as $160.00. That amount does not follow from the stated data: the old calculation is one-half of $150 plus one-half of $200, which equals $175.00.

Why the Credibility Weight Has This Form

The formula $Z=W/(W+K)$ expresses the balance between variation in the underlying risk means and variation in the observed experience around a risk’s mean. To see this, consider the Bühlmann-Straub assumptions: conditional on a risk mean $\Theta$, period loss rates have mean $\Theta$, conditional variance $v/w_j$, and are conditionally independent across periods. The process variance of the exposure-weighted rate is then $v/W$; the variance of $\Theta$ across risks is $a$.

For a linear estimate of $\Theta$ of the form $c\bar X+(1-c)\mu$, the mean squared error is minimized when the coefficient on the observed rate is the variance of the risk means divided by the total variance of the observed rate around the collective mean. Since the two variance components are $a$ and $a+v/W$, respectively, the minimizing coefficient is

c=\frac{a}{a+v/W} =\frac{aW}{aW+v} =\frac{W}{W+v/a} =\frac{W}{W+K}.

This gives the Bühlmann-Straub credibility weight and the credibility premium as a linear predictor. It is the best linear predictor under the stated variance assumptions; it need not equal the exact Bayesian posterior mean for every possible distribution. As $W$ increases, $v/W$ decreases, so the observed rate is given more weight. The collective complement remains important when exposure is small.

Combine the experience.
Add old and new exposures, and separately add old and new aggregate losses.

Recalculate the loss rate.
Divide combined losses by combined exposure; do not average period rates unless their exposure weights are equal.

Recalculate credibility.
Use the updated total exposure and the same $K$, provided the model parameters have not been changed.

Blend and interpret.
Combine the updated rate with the collective rate, then explain whether the premium changed because of credibility, observed cost, or both.

Separate the Effect of More Data From the Effect of New Losses

A useful diagnostic is to calculate a counterfactual premium that uses the new credibility weight but holds the old observed rate fixed. The difference between that counterfactual and the old premium isolates the effect of changing credibility while keeping the rate unchanged. The difference between the actual updated premium and the counterfactual reflects the change in observed rate at the new credibility weight.

Worked Example: More Exposure and a Higher New Loss Rate

A hypothetical risk has $W=2$ exposure units and $200 in losses. Its collective rate is $\mu=\$120$, and $K=3$. It then earns two more exposure units with $360 in losses. Calculate the premium before and after the update, and separate the exposure-only effect from the effect of the new experience rate.

Old premium. The observed rate is $200/2=\$100$, and $Z_{\text{old}}=2/(2+3)=0.4$. Therefore,

P_{\text{old}}=0.4(100)+0.6(120)=40+72=\$112.00.

Counterfactual with more exposure but the old rate. Total exposure would be $2+2=4$, so the new weight is $Z_{\text{new}}=4/(4+3)=4/7$. If the observed rate stayed at $100, the premium would be

P_{\text{cf}}=\frac47(100)+\frac37(120) =\frac{400+360}{7} =\frac{760}{7} =\$108.57.

Actual updated premium. Combined losses are $200+360=\$560$, and the updated rate is $560/4=\$140$. Thus

P_{\text{new}}=\frac47(140)+\frac37(120) =\frac{560+360}{7} =\frac{920}{7} =\$131.43.

With the rate held at $100, added exposure would reduce the premium from $112.00 to $108.57 because it gives more weight to experience below the collective rate. The actual premium is higher, at $131.43, because the new experience raises the observed rate to $140. The rate change adds $(4/7)(140-100)=\$22.86$ to the counterfactual premium. The check is $108.57+22.86=\$131.43$, subject to rounding.

Worked Example: Updating Unequal-Exposure Periods

A hypothetical risk has two past periods: exposure 1.5 with losses of $180, and exposure 0.5 with losses of $20. It then earns 1.5 more exposure with losses of $300. Let the collective rate be $\mu=\$150$ and $K=4$. Find the old and updated premium per exposure unit.

Method. Use exposure totals, not an unweighted average of the three period rates.

Before the new period. Exposure is $1.5+0.5=2$, and losses are $180+20=\$200$. The observed rate is $200/2=\$100$, and $Z_{\text{old}}=2/(2+4)=1/3$. Therefore,

P_{\text{old}}=\frac13(100)+\frac23(150) =\frac{100+300}{3} =\$133.33.

After the new period. Total exposure is $2+1.5=3.5$, and total losses are $200+300=\$500$. The observed rate is $500/3.5=\$142.8571$, while $Z_{\text{new}}=3.5/(3.5+4)=3.5/7.5=7/15$. Thus

P_{\text{new}}=\frac7{15}\left(\frac{500}{3.5}\right)+\frac8{15}(150) =\frac{1000}{15}+80 =\$146.67.

The new period’s own rate is $300/1.5=\$200$, above both the old rate of $100 and the collective rate of $150. The updated premium therefore rises even though the credibility weight also rises. As a check, the updated credibility premium is between the updated observed rate, $142.86, and collective rate, $150, as a convex weighted average must be.

Common Update Errors

Updating only the credibility weight. New losses can change the observed rate too. Recompute both before blending.
Using a period count in place of exposure. If periods have different exposure, $W$ is the sum of exposure amounts, not the number of periods.
Averaging rates without weights. The combined rate is total losses divided by total exposure. A low-exposure period should not have the same influence as a high-exposure period solely because each is one row in a table.
Treating no reported losses as zero exposure. A period with positive exposure and no losses increases $W$ and leaves $S$ unchanged.
Changing the collective complement unnoticed. In a pure experience update, keep $\mu$ and $K$ fixed. If the rating basis or parameter estimates are deliberately revised, label that change and recalculate consistently.

Key takeaway: An exposure update is not just a change in credibility. Add exposure and losses separately, recompute the weighted loss rate and credibility weight, and then blend with the unchanged collective complement. Explain the premium movement by distinguishing the effect of more data from the effect of a changed observed rate.

Check Your Understanding

For each hypothetical risk, use the stated Bühlmann-Straub parameters and give the updated premium per exposure unit.

A risk has $W=1$, losses of $120, collective rate $\mu=\$90$, and $K=1$. It adds one exposure unit with no losses. What is the updated premium? (Answer: $70.00)
A risk has $W=2$, losses of $400, $\mu=\$250$, and $K=3$. It adds one exposure unit with $200 in losses. Calculate its updated premium. (Answer: $225.00)
A risk has $W=4$, losses of $400, $\mu=\$80$, and $K=2$. It adds two exposure units with $40 in losses. Find the updated observed rate, credibility, and premium. (Answer: observed rate $73.33, credibility 0.7500, premium $75.00)
A risk has two past periods: exposure 1.5 with $180 in losses, and exposure 0.5 with $20 in losses. With $\mu=\$140$ and $K=2$, it adds exposure 2 with losses of $360. Find the updated premium and compare it with the old premium. (Answer: updated premium $140.00, up from $120.00)
A risk has $W=2$, losses of $320, $\mu=\$100$, and $K=2$. It then adds six exposure units with $600 in losses. Find the updated observed rate, credibility, and premium. (Answer: observed rate $115.00, credibility 0.8000, premium $112.00)

What You'll Learn