Normalizing Catastrophe and Large Loss Experience

Why Normalize Historical Losses?

Historical losses are evidence for estimating future costs, but a single catastrophe or a few unusually large claims can make one period look unlike the risk the rates are intended to cover. Simply accepting every observed dollar can overstate the prospective cost when the event is unlikely to recur at the same level; deleting unusual losses altogether can understate it. Normalization makes the treatment explicit: remove or limit an atypical component, then replace it with a selected provision for the cost expected prospectively.

The adjustments below are made to losses on a consistent basis. For example, if losses have already been developed and trended to the prospective cost level, use a catastrophe provision and large-loss threshold on that same level. The hypothetical examples exclude loss adjustment expenses and other premium provisions, so the rate comparison is based only on losses and a stated target loss ratio.

Definition: Catastrophe normalization replaces actual losses assigned to catastrophe events with a selected expected catastrophe provision for the experience period or exposure base. Large-loss normalization limits individual claims at a stated threshold and adds a selected provision for expected losses above that threshold. The methods estimate different components and must be applied so the same loss is not removed or provided for twice.

Catastrophe Normalization

A common approach is to separate losses into catastrophe and non-catastrophe components. Let $L_{\mathrm{noncat}}$ be the observed non-catastrophe losses, $L_{\mathrm{cat}}$ the observed losses assigned to catastrophe events, and $C_{\mathrm{cat}}$ the selected expected catastrophe provision on the same basis. The normalized losses are

L_{\mathrm{norm}}=L_{\mathrm{noncat}}+C_{\mathrm{cat}}.

This is a replacement, not an addition to all observed losses: the actual catastrophe amount is removed before the selected provision is included. The provision might be selected per exposure, per year, or by another justified basis. Its amount is an assumption informed by the intended risk, available experience, and the chosen method; it is not automatically equal to the catastrophe losses in the period being adjusted.

The classification rule matters. Identify which events count as catastrophes and apply it consistently across the experience. If an event loss is included in $L_{\mathrm{cat}}$, it should not also remain in $L_{\mathrm{noncat}}$. Further, a large-claim cap applied to non-catastrophe claims should not inadvertently change the catastrophe amount that is being replaced.

Worked Example: Replacing an Unusually High Catastrophe Year

A hypothetical portfolio has $10,000 exposure-years and $1,800,000 of non-catastrophe losses. An identified catastrophe produced a further $600,000. The selected prospective catastrophe provision is $40 per exposure-year. On-level earned premium is $4,000,000, and the target loss ratio is 55%. Find the normalized loss ratio and indicated rate change using the loss-ratio method. Ignore expenses and other provisions.

Method. Remove the actual catastrophe losses and replace them with the selected provision. Then divide normalized losses by on-level premium and compare that loss ratio with the target.

The expected catastrophe provision is $10{,}000\times \$40=\$400{,}000$. Therefore, $L_{\mathrm{norm}}=\$1{,}800{,}000+\$400{,}000=\$2{,}200{,}000$.

The normalized loss ratio is $\$2{,}200{,}000/\$4{,}000{,}000=0.55$, or 55%. The indicated factor is $0.55/0.55=1.0000$; the indicated change is $1.0000-1=-0.0000$, or 0.00%.

As a check, without normalization the loss ratio would be $(\$1{,}800{,}000+\$600{,}000)/\$4{,}000{,}000=0.60$. That would indicate a factor of $0.60/0.55=1.0909$, or a 9.09% increase. The method avoids treating one period’s actual catastrophe amount as the prospective catastrophe cost. The correction is appropriate only if the $400,000 selected provision is suitable for the prospective risk.

Large-Loss Capping and an Excess Provision

A large-loss adjustment typically limits each claim at a selected threshold $u$, rather than limiting the total losses for a period. For a claim amount $X$, the limited amount is $\min(X,u)$, and the excess is $(X-u)_+=\max(X-u,0)$. Applying the threshold to each claim preserves the ordinary portion of a large claim while separately recognizing its excess portion.

X=\min(X,u)+(X-u)_+.

Thus, for a set of observed claims, capped losses are the sum of their limited amounts. If $e(u)$ is the selected expected excess loss above $u$ per exposure, the normalized losses for $E$ exposures are

L_{\mathrm{norm}}=\sum_{j=1}^{n}\min(X_j,u)+E\,e(u).

The expected excess provision must be selected for the same threshold, exposure basis, and cost level as the capped claims. It can be informed by a severity model or other appropriate analysis. The formula does not imply that the sample’s observed excess is the best prospective provision. If the provision is already included elsewhere, adding it again would double count.

The identity above explains the logic: each claim consists exactly of its limited part plus its excess part. Replacing observed excess with an expected excess provision preserves the capped portion of experience while substituting a selected prospective estimate for the volatile tail. This is useful when a small number of very large claims would otherwise dominate the period. It does not, by itself, determine whether the threshold or the expected excess provision is appropriate.

Worked Example: Capping Claims and Restoring Expected Excess

A hypothetical portfolio has three exposure-years and observed claims of $80,000, $140,000, and $300,000. Use a $100,000 per-claim threshold and a selected expected excess provision of $15,000 per exposure-year. On-level premium is $200,000 per exposure-year, and the target loss ratio is 65%. Find normalized losses and the indicated rate change. Ignore expenses and other provisions.

Method. Cap each observed claim separately, add the selected excess provision for all three exposure-years, and apply the loss-ratio indication.

The limited claims are $\min(\$80{,}000,\$100{,}000)=\$80{,}000$, $\min(\$140{,}000,\$100{,}000)=\$100{,}000$, and $\min(\$300{,}000,\$100{,}000)=\$100{,}000$. Their sum is $\$80{,}000+\$100{,}000+\$100{,}000=\$280{,}000$. The expected excess provision is $3\times \$15{,}000=\$45{,}000$, so normalized losses are $\$280{,}000+\$45{,}000=\$325{,}000$.

On-level premium is $3\times \$200{,}000=\$600{,}000$. The normalized loss ratio is $\$325{,}000/\$600{,}000=0.5417$ (to four decimal places). The indicated factor is $0.5416667/0.65=0.8333$ (to four decimal places), so the indicated rate change is $0.8333333-1=-0.1666667$, or $-16.67\%$ to two decimal places.

As a check, the unadjusted claims total $\$80{,}000+\$140{,}000+\$300{,}000=\$520{,}000$, giving a loss ratio of $520{,}000/600{,}000=0.8667$ and an indicated increase of $0.8666667/0.65-1=33.33\%$. The normalization reduces the indicated rate in this example. It does not establish that the capped result is always lower; the selected expected excess provision and the actual claim pattern determine the result.

Combining Adjustments and Indicating a Rate

When both methods are used, define their boundaries before calculating. One workable sequence is to identify and remove catastrophe losses, normalize the remaining claims for large losses, and then add the selected catastrophe provision. This sequence assumes the large-loss method is intended for non-catastrophe claims. If the selected methods instead treat catastrophe claims as part of the large-loss layer, specify that treatment and avoid including those claims in both adjustments.

For on-level earned premium $P_{\mathrm{on}}$ and target loss ratio $r_{\mathrm{target}}$, the simplified loss-ratio method is

\text{Indicated factor} = \frac{L_{\mathrm{norm}}/P_{\mathrm{on}}}{r_{\mathrm{target}}}, \qquad \text{indicated change} = \text{Indicated factor}-1.

This comparison assumes the premium and losses are on a matched basis and that the target loss ratio is appropriate for the indication being calculated. It isolates the effect of loss normalization; it is not a complete premium calculation with expenses or other provisions.

Worked Example: Applying Both Adjustments Without Double Counting

A hypothetical portfolio has 100 exposure-years. Its non-catastrophe claims consist of three claims of $90,000, $160,000, and $250,000, plus $400,000 of other claims. A separate catastrophe produced $300,000. Cap non-catastrophe claims at $100,000 each, and use an expected excess provision of $1,200 per exposure-year for those claims. Replace the catastrophe loss with a selected provision of $150,000. On-level premium is $1,500,000, and the target loss ratio is 60%. Find the normalized loss ratio and indicated change.

Method. First cap only the non-catastrophe claims and add the corresponding expected excess provision. Then replace, rather than retain, the catastrophe loss. This keeps the two adjustments separate.

The three specified non-catastrophe claims are limited to $\$90{,}000+\$100{,}000+\$100{,}000=\$290{,}000$. The $400,000 of other non-catastrophe claims is below the scope of this claim-level adjustment as stated, so capped non-catastrophe losses total $\$290{,}000+\$400{,}000=\$690{,}000$. The expected excess provision is $100\times \$1{,}200=\$120{,}000$. Normalized non-catastrophe losses are therefore $\$690{,}000+\$120{,}000=\$810{,}000$.

Replace the actual catastrophe amount with $150,000: total normalized losses are $\$810{,}000+\$150{,}000=\$960{,}000$. The normalized loss ratio is $\$960{,}000/\$1{,}500{,}000=0.64$, or 64%. The indicated factor is $0.64/0.60=1.0667$ (to four decimal places), giving an indicated increase of $1.0666667-1=6.67\%$ (to two decimal places).

As a check, the unadjusted losses total $\$90{,}000+\$160{,}000+\$250{,}000+\$400{,}000+\$300{,}000=\$1{,}200{,}000$. The unadjusted loss ratio is $1{,}200{,}000/1{,}500{,}000=0.80$, which would indicate $0.80/0.60-1=33.33\%$. The normalized total is not formed by adding the catastrophe provision to the $300,000 actual catastrophe loss; doing so would count both.

Consistency Checks and Exam Traps

Normalization is only interpretable if the inputs share a basis. Check the following before calculating:

Trend and development: Match the selected catastrophe provision and large-loss threshold to the loss level being indicated. If historical claims are trended to the prospective period, do not apply an untrended threshold as though it were already at that level.
Units: Multiply a per-exposure provision by the matching number of exposure-years, not by the number of claims. Keep dollars and thousands of dollars consistent.
Claim-level threshold: Apply $u$ to each claim when the method specifies a per-claim cap. A cap on total annual losses is a different method.
Replacement versus addition: Catastrophe normalization replaces the actual catastrophe component. Large-loss normalization replaces observed excess with the selected expected excess, rather than adding a provision to uncapped losses.
Scope: State whether catastrophe claims are included in or excluded from the large-loss adjustment. Do not remove them in one step and still count them in the other.

For instance, if a claim is $80,000 at the historical level and the applicable trend factor is 1.10, its prospective-level amount is $88,000. Against a prospective-level threshold of $100,000, the limited amount is $88,000; it is not $80,000. Applying the threshold before trending would mix levels and understate the limited loss in this example. If a different threshold basis is explicitly specified, follow that basis consistently rather than assuming it is prospective.

Key takeaway: State exactly which losses are being removed or limited, what replaces them, and the exposure and dollar basis of each amount. Then recalculate the loss ratio using normalized losses. A sound adjustment is transparent about its assumptions and does not count the same loss twice.

Check Your Understanding

For each hypothetical case, use only the method and assumptions stated. Round dollar amounts to the nearest dollar when needed and rate changes to two decimal places.

A portfolio has 2,000 exposure-years, $720,000 of non-catastrophe losses, and $180,000 of catastrophe losses. The selected catastrophe provision is $50 per exposure-year. What are normalized losses and the normalized pure premium per exposure-year? (Answer: $820,000 and $410 per exposure-year.)
Three claims are $55,000, $130,000, and $210,000. Cap each at $75,000 and add a selected expected excess provision of $20,000 per exposure-year for three exposure-years. Find normalized losses and normalized pure premium. (Answer: $265,000 and $88,333.33 per exposure-year.)
Non-catastrophe losses are $1,100,000 and actual catastrophe losses are $400,000. Replace catastrophe losses with a selected $250,000 provision. On-level premium is $5,000,000 and the target loss ratio is 50%. Find the indicated rate change. (Answer: $-46.00\%$.)
A historical claim is $80,000 and is trended once by 10% to the prospective level. The prospective per-claim cap is $100,000. For one exposure-year, the selected expected excess provision is $12,000. Find the normalized amount for this claim and its provision. (Answer: $100,000.)
For four exposure-years, non-catastrophe claims are $40,000, $120,000, and $250,000, plus $390,000 of other losses; a separate catastrophe caused $160,000. Cap the three specified claims at $100,000 each, add expected excess of $25,000 per exposure-year, and replace the catastrophe with an $80,000 provision. On-level premium is $1,350,000 and the target loss ratio is 60%. Find the indicated rate change. (Answer: 0.00%.)

What You'll Learn