Handling Territories With Little or No Experience

Why Sparse Territory Experience Needs a Deliberate Method

A territory with few claims can show a very high or very low loss cost simply because a small number of outcomes has a large effect on the average. A territory with no observed claims presents the sharpest version of the problem: its observed loss cost is zero, but that does not establish that its expected loss cost is zero. The estimate must use information beyond the local claim total.

That information can come from a specified complement, such as a statewide or selected benchmark loss cost; from exposure information, including the mix of risks in the territory; or from a broader geographic grouping with more experience. The choice should reflect which information is relevant to the territory, not merely which calculation is easiest.

Definition: For territory $t$, let $E_t$ be its exposure count and $L_t$ its losses on the valuation and coverage basis being used. When $E_t>0$, the observed territory loss cost is $LC_t=L_t/E_t$. Let $C_t$ be a specified complement loss cost for that territory, and let $Z_t$, between 0 and 1 inclusive, be the selected credibility weight on the territory’s own experience. A blended estimate is $\widehat{LC}_t=Z_tLC_t+(1-Z_t)C_t$. If $LC_{\text{base}}$ is the loss-cost benchmark used for the rating plan, the corresponding territory relativity is $R_t=\widehat{LC}_t/LC_{\text{base}}$.

The complement is an estimate for the territory, not automatically the statewide average. It might be a statewide benchmark, the expected loss cost for a specified mix of classes, or a loss cost from a suitably broad and comparable group. State what the complement represents and use the same loss, exposure, and coverage definitions on both sides of the calculation.

The credibility weight is also an input, not a number that can be inferred from the loss total alone. It may be selected using the credibility analysis and standards appropriate to the data. The formula here shows how to combine the resulting weight and complement; it does not prescribe a universal credibility standard. If there is no usable local experience, set $Z_t=0$, so the estimate is the complement. If local experience is fully credible under the chosen approach, $Z_t=1$, so the estimate is the local loss cost.

Using Local Experience and a Complement

The blend is a weighted average of two estimates of the same quantity: the territory’s expected loss cost. The local estimate receives weight $Z_t$; the complement receives the remaining weight $1-Z_t$. Both estimates must be expressed per the same exposure unit and on a consistent basis.

\widehat{LC}_t=Z_tLC_t+(1-Z_t)C_t

Once the blended loss cost is found, compare it with the stated base to obtain a relativity. Do not compare a territory’s observed loss cost with a premium base or with a complement expressed on a different exposure unit.

Worked Example: A Territory With Zero Observed Losses

A hypothetical territory has 100 exposure units and no observed losses in the experience period. The selected complement is a loss cost of $200 per exposure unit, and the selected credibility weight for the territory experience is $Z_t=0.25$. The rating plan’s loss-cost base is $180 per exposure unit. Estimate the territory loss cost and relativity. Name the method and identify the mistake it avoids.

Method. Use the credibility-weighted complement blend. Since the observed losses are zero and exposures are positive, the observed loss cost is $LC_t=\$0/100=\$0$ per exposure unit.

Substitute the local estimate, complement, and credibility weight:

\widehat{LC}_t=0.25(\$0)+0.75(\$200)=\$0+\$150=\$150

The indicated relativity is:

R_t=\frac{\$150}{\$180}=0.833333\ldots

Rounded to three decimal places, the territory relativity is 0.833. As a check, the weights sum to 1, and the blended loss cost lies between $0 and $200, as it should. At 100 exposure units, the estimated losses corresponding to this blended cost are $100(\$150)=\$15{,}000$.

Common mistake avoided: assigning a zero loss cost and a zero relativity just because no claims were observed. That would ignore the complement and the nonzero expected risk indicated by it. The calculation also avoids giving the complement full weight: the selected $Z_t=0.25$ gives local experience some, but limited, influence.

Why the Blend Is a Reasonable Estimate

The formula’s basic justification is that it combines two estimates of the same target on the same scale. If the local estimate and the complement are each unbiased for the territory’s expected loss cost $\theta_t$, and $Z_t$ is fixed, linearity of expectation gives:

E[\widehat{LC}_t] =E[Z_tLC_t+(1-Z_t)C_t] =Z_t\theta_t+(1-Z_t)\theta_t =\theta_t

This explains why the weights sum to 1: under those assumptions, the blend preserves the target mean. Credibility gives more weight to the estimate judged to be more informative. This justification does not show that any particular $Z_t$ is correct, nor that the complement is unbiased. If the complement is poorly matched to the territory, the blend can still be systematically wrong. If the two estimates share data, they may also be dependent; do not use an independence-based variance calculation without checking that assumption.

When experience is pooled across territories, exposure weighting follows directly from the definition of loss cost. For a group of territories with losses $L_j$ and exposures $E_j$, the group loss cost is total losses divided by total exposures:

LC_{\text{group}}=\frac{\sum_j L_j}{\sum_j E_j} =\sum_j\left(\frac{E_j}{\sum_k E_k}\right)LC_j

Thus each territory’s loss cost is weighted by its share of the group’s exposure. This is not generally the same as taking a simple average of territory loss costs.

Exposure Information and Broader Geographic Groups

Exposure data can help even when the territory has few or no claims. For example, a territory’s exposure distribution across classes can be combined with class-specific complement loss costs to produce a complement appropriate to its own mix. The result is an expected loss cost for that territory’s exposure mix, rather than an unadjusted average across classes.

A broader geographic group is another possible source of information. The group should be defined on a defensible basis: its territories should be sufficiently comparable for the pooled experience to inform the target. Calculate the group loss cost from the group’s total losses and total exposures. If that pooled estimate is then blended with another complement, the weight and the data roles must be clear. Avoid counting the same losses once in the pooled estimate and again as if they were independent local experience.

Worked Example: Blending a Sparse Local Estimate

A hypothetical territory has 40 exposure units and $10,000 of observed losses. Its selected complement is $170 per exposure unit, and its selected credibility weight is $Z_t=0.30$. The loss-cost base is $200 per exposure unit. Estimate its blended loss cost and relativity. Name the method and identify the mistake it avoids.

Method. Calculate the local loss cost, then blend it with the complement. The local loss cost is:

LC_t=\frac{\$10{,}000}{40}=\$250

The territory’s own estimate exceeds the complement. Applying the specified weight:

\widehat{LC}_t=0.30(\$250)+0.70(\$170) =\$75+\$119 =\$194

The relativity is:

R_t=\frac{\$194}{\$200}=0.970

The indicated relativity is 0.970. The blended loss cost is below the local $250 because the complement is lower and receives 70% of the weight; it remains above the $170 complement because the local estimate receives 30%. The calculation is checked by verifying that the weighted contributions, $75 and $119, sum to $194.

Common mistake avoided: using the local loss cost of $250 as the final indication simply because it is directly observed. With sparse experience, a large local loss cost can be volatile. The specified credibility weight reduces its influence without discarding it.

Worked Example: Using a Broader Geographic Pool

A hypothetical target territory has no usable losses. Three comparable territories in a pre-defined broader region, excluding the target, have the following experience: Territory A has 300 exposures and $48,000 in losses; Territory B has 400 exposures and $84,000 in losses; Territory C has 300 exposures and $36,000 in losses. Use the pooled group as a complement, with a selected credibility weight of $Z_t=0.50$ on that group estimate and a statewide complement of $160 per exposure unit. The target’s base loss cost is $160. Estimate its loss cost and relativity. Name the method and identify the mistake it avoids.

Method. Pool losses and exposures first, then blend the group estimate with the statewide complement. The pooled amounts are:

L_{\text{group}}=\$48{,}000+\$84{,}000+\$36{,}000=\$168{,}000 $$ $$ E_{\text{group}}=300+400+300=1{,}000

Therefore, the group loss cost is:

LC_{\text{group}}=\frac{\$168{,}000}{1{,}000}=\$168

Blend that estimate with the statewide complement:

\widehat{LC}_t=0.50(\$168)+0.50(\$160)=\$84+\$80=\$164

The territory relativity is:

R_t=\frac{\$164}{\$160}=1.025

The indication is a loss cost of $164 per exposure unit and a relativity of 1.025. As a check, the group loss cost is $8 above the statewide complement, so a half-weight blend is $4 above it. The result, $164, has that property.

Common mistake avoided: taking a simple average of the three territory loss costs. Those costs are $160, $210, and $120; their simple average is $163.33, not the correctly pooled $168. The pooled estimate gives each exposure equal weight, so the territory with 400 exposures contributes more than the territory with 300.

Adjusting a Complement for Exposure Mix

A broad or statewide complement may not represent the target territory if the territory has a different distribution of risks. Suppose the complement provides expected loss costs by class. Multiply each class’s loss cost by the target territory’s exposure in that class, add the expected losses, and divide by the territory’s total exposures. This produces a mix-adjusted complement for that territory.

The calculation is an exposure-weighted average. It does not require local claims for the territory, but it does require appropriate class-level complements and reliable exposure counts. If the class definitions or coverage basis do not match, the result is not a valid mix adjustment.

Worked Example: A Complement Based on Territory Class Mix

A hypothetical territory has 40 exposure units in Class A and 60 in Class B. The specified class complements are $90 per exposure unit for Class A and $210 for Class B. The base mix is 60% Class A and 40% Class B. Estimate the territory’s mix-adjusted complement and its relativity against the base-mix loss cost. Name the method and identify the mistake it avoids.

Method. Calculate expected losses for the target territory’s actual class exposures. Class A contributes $40(\$90)=\$3{,}600$; Class B contributes $60(\$210)=\$12{,}600$. Total expected losses are:

\$3{,}600+\$12{,}600=\$16{,}200

There are $40+60=100$ exposure units, so the target mix-adjusted complement is:

C_t=\frac{\$16{,}200}{100}=\$162

For the base mix, the loss cost is:

LC_{\text{base}}=0.60(\$90)+0.40(\$210)=\$54+\$84=\$138

The relativity is:

R_t=\frac{\$162}{\$138}=1.173913\ldots

Rounded to three decimal places, the relativity is 1.174. The check is consistent with the mix: the territory has more exposure in the higher-cost Class B than the base mix does, so its mix-adjusted complement exceeds the base-mix loss cost.

Common mistake avoided: using the statewide or base-mix complement of $138 for the territory without checking its exposure distribution. Another error would be taking an unweighted average of the two class complements, which gives $150 and ignores that the territory has 60% of its exposure in Class B.

Practical Checks Before Selecting an Indication

A sparse-data estimate is only as sound as its inputs and definitions. Before using it in a territory rating indication, confirm the following:

Comparable basis. Local experience, complement, and base use consistent coverage, loss definitions, exposure units, and measurement periods.
Appropriate complement. The benchmark or broader group is relevant to the target territory, and any class mix adjustment uses matching class definitions.
Clear weight. The credibility weight is supported by the selected credibility approach; it is not chosen merely to produce a preferred relativity.
Correct pooling. Group loss cost is total losses divided by total exposures, not an unweighted average of territorial loss costs.
No double counting. The same experience is not included in multiple components of an estimate as if it were separate information.
Zero claims interpreted carefully. Zero observed losses can occur with positive exposure; it does not by itself establish a zero expected loss cost.

These methods help produce a defensible indication, not an automatic final factor. A complement may be inappropriate, a broad group may not be comparable, or the chosen exposure mix may be unreliable. State those limitations rather than presenting a calculated result as more precise than its information supports. This treatment of sparse territory data is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Check Your Understanding

For each problem, calculate the requested estimate before checking the answer.

A hypothetical territory has 60 exposure units and $6,600 in observed losses. Its complement is $150 per exposure unit, its selected credibility weight is $Z_t=0.25$, and its base loss cost is $160. Find the blended loss cost and relativity. (Answer: $140 per exposure unit; relativity $0.875$)
A territory has positive exposure but no observed losses. Its selected complement is $240 per exposure unit, its credibility weight is $Z_t=0.20$, and its base is $200. Find the blended loss cost and relativity. (Answer: $192 per exposure unit; relativity $0.960$)
A comparable geographic pool has three components: 300 exposures with $48,000 losses, 450 exposures with $72,000 losses, and 150 exposures with $30,000 losses. Blend the pooled loss cost at $Z_t=0.60$ with a $155 complement. (Answer: pooled loss cost $166.67; blended loss cost $162.00 per exposure unit)
A territory has 75 Class A exposures and 125 Class B exposures. Their class complements are $90 and $210. The base mix is 60% Class A and 40% Class B. Find the territory mix-adjusted complement and relativity to the base. (Answer: $165 per exposure unit; relativity $1.196$, rounded to three decimals)
A territory has 80 exposures and $12,000 in observed losses. Its complement is $180 per exposure unit, its selected credibility weight is $Z_t=0.40$, and its base is $175. Calculate the local loss cost, blended loss cost, and relativity. (Answer: local loss cost $150; blended loss cost $168 per exposure unit; relativity $0.960$)

What You'll Learn