Why Development Bases Must Be Chosen Together
A frequency-severity estimate splits projected losses into two components: the number of claims per exposure and the average cost per claim. Developing counts and losses separately can be appropriate because reported claims and claim dollars mature differently. But the calculation is only meaningful when both components refer to the same experience and are brought to a consistent ultimate basis.
A count factor does not generally serve as a loss factor, and a loss factor does not generally serve as a severity factor. Loss development reflects both the emergence of additional loss and changes in the amounts recorded for claims already known. Count development concerns the emergence of additional claims under a specified claim-count definition. Choosing bases without checking what each factor develops can produce components that do not reproduce the projected aggregate loss.
Here \(C_a\) is the count at age \(a\), \(L_a\) is the corresponding loss amount, and exposure \(E\) is the exposure measure for that experience period. The resulting estimates are
The formulas presume that the count and loss data describe the same experience cohort and compatible claim populations. The count definition must be clear—for example, reported claims rather than closed claims—and the loss basis must also be specified. Differences in coverage, gross-versus-net treatment, expense inclusion, or claim limits can make the numerator and denominator describe different things even when their dates match.
Selecting and Aligning the Bases
Each development factor should be selected for the data it is applied to. A reported-count factor should reflect the maturity pattern of the chosen reported-claim definition. A loss factor should match the loss measure used in the data, such as paid or incurred losses, and its valuation age. A factor selected for one basis should not be transferred to another just because both are labeled “development.”
The count and loss observations do not have to be recorded on the same calendar date if each is developed from its own documented valuation age to ultimate. They do, however, need to refer to the same experience period, and both projections need to represent a consistent ultimate view. If the observations come from different dates, use factors appropriate to each age; do not compare one immature observation with another more mature observation as if they were equally developed.
Specify the exposure period, exposure count, claim-count definition, loss basis, and whether the amounts are gross or net and include any expenses.
For both the count and loss observations, identify the valuation date or development age. A shared accident period does not imply a shared maturity.
Use the selected count factor for the count at its age and the selected loss factor for the loss amount at its age. Each factor must project its own measure to ultimate.
Check that estimated frequency times estimated severity equals ultimate losses divided by exposures. Investigate any failure of this identity before using the estimate.
Worked Example: Developing Count and Loss on Separate Bases
A hypothetical book has 5,000 exposures, 400 reported claims, and $1,440,000 of incurred losses at the selected valuation age. The selected reported-count factor is 1.25 and the selected incurred-loss factor is 1.50. Estimate ultimate frequency, severity, and pure premium. State the method and check the result.
Method. Develop the count and incurred losses separately to ultimate, then divide ultimate values to obtain frequency and severity. This method assumes both observations describe the same experience period and claim population, and that each factor matches its observation’s maturity and basis.
Ultimate claim count: \(400(1.25)=500\) claims.
Ultimate losses: \($1,440,000(1.50)=$2,160,000\).
Ultimate frequency: \(500/5,000=0.1000\) claims per exposure.
Ultimate severity: \($2,160,000/500=$4,320.00\) per claim.
Pure premium: \($2,160,000/5,000=$432.00\) per exposure. The component check gives \(0.1000($4,320.00)=$432.00\), agreeing with the aggregate calculation. The common mistake is to divide the undeveloped loss amount by developed claims, or to report a severity that fails this check.
Why the Implied Severity Factor Is Not Usually the Loss Factor
A useful way to understand the choice of bases is to compare developed severity with observed severity. Suppose the count and losses are measured on matching bases at the same experience period, with observed count \(C_a\) and observed losses \(L_a\). Observed severity is \(L_a/C_a\). After developing both components, estimated ultimate severity is \(L_aD_L(a)/(C_aD_C(a))\). Dividing this by observed severity gives the severity development factor implied by the two selected factors:
Thus, when both factors apply to the same observation age and population, the implied severity factor is the loss factor divided by the count factor—not the loss factor by itself. This follows because loss development includes the effect of additional claims entering the loss total, while count development accounts for those claims in the severity denominator.
This ratio is a reconciliation identity, not a guarantee that every factor selection is appropriate. If count and loss observations have different ages, use each factor with its own observation to calculate ultimate count and ultimate losses. Do not treat their ratio as a severity factor applied to a single observed severity unless the underlying observations and ages support that comparison.
Worked Example: Converting Loss Development to an Implied Severity Factor
A hypothetical portfolio has 2,000 exposures, 100 reported claims, and $500,000 of losses on a consistent valuation basis. The selected count factor is 1.20 and the selected loss factor is 1.50. Calculate ultimate frequency, severity, and pure premium, and show why applying the loss factor directly to observed severity is wrong.
Method. Develop each component separately. The ratio of the loss and count factors provides the implied severity factor because both factors refer to this same valuation basis.
Ultimate claims are \(100(1.20)=120\). Ultimate losses are \($500,000(1.50)=$750,000\). Therefore, frequency is \(120/2,000=0.0600\) claims per exposure, and severity is \($750,000/120=$6,250.00\) per claim.
Observed severity is \($500,000/100=$5,000.00\). The implied severity factor is \(1.50/1.20=1.25\), and \(5,000(1.25)=$6,250.00\), which confirms the developed severity. Pure premium is \($750,000/2,000=$375.00\) per exposure; the component check is \(0.0600($6,250.00)=$375.00\).
Common mistake. Applying the loss factor directly to observed severity gives \($5,000(1.50)=$7,500.00\). Multiplying that by the correctly developed frequency gives \($450.00\) per exposure, which does not equal the developed-loss estimate of \($375.00\). The error is counting the claim-emergence effect in the severity adjustment even though frequency has already been developed.
Different Valuation Ages: A Reconciliation Example
Different valuation ages are not automatically an inconsistency. A count observed at one age and losses observed at another can still yield a coherent estimate if they refer to the same experience cohort and each is developed using a factor selected for its own maturity and measurement basis. The key is to compare the resulting ultimate estimates, not raw values from unequal maturity dates.
Worked Example: Reconciling Two Valuation Snapshots
For a hypothetical cohort of 4,000 exposures, the reported count is 300 at 12 months and 360 at 24 months. Selected count factors to ultimate are 1.20 at 12 months and 1.00 at 24 months. Incurred losses are $1,080,000 at 12 months and $1,440,000 at 24 months; selected loss factors are 1.50 and 1.125, respectively. Estimate ultimate frequency, severity, and pure premium using each snapshot as a cross-check.
Method. Develop each snapshot using the factor for that snapshot’s age, then compare the implied ultimate count and losses. This assumes the two snapshots are for the same cohort, on unchanged definitions and loss basis.
At 12 months, ultimate claims are \(300(1.20)=360\), and ultimate losses are \($1,080,000(1.50)=$1,620,000\). At 24 months, ultimate claims are \(360(1.00)=360\), and ultimate losses are \($1,440,000(1.125)=$1,620,000\). Both snapshots therefore imply 360 ultimate claims and $1,620,000 ultimate losses.
Frequency is \(360/4,000=0.0900\) claims per exposure. Severity is \($1,620,000/360=$4,500.00\) per claim, and pure premium is \($1,620,000/4,000=$405.00\) per exposure. The component check is \(0.0900($4,500.00)=$405.00\), so the snapshots reconcile.
Common mistake. Dividing the 24-month loss amount by the 12-month count gives \($1,440,000/300=$4,800.00\), but that ratio mixes maturity dates and is not the ultimate severity. Developing both observations to ultimate, or consistently using a single valuation snapshot with its appropriate factors, avoids this mismatch.
Common Inconsistencies and Their Consequences
A mismatch may be caused by more than different ages. Before accepting a frequency-severity estimate, confirm that the count and loss data use compatible claim definitions and cover the same exposure period. A count of reported claims cannot be paired casually with losses that exclude some of those claims or include a different coverage population. Similarly, switching between paid and incurred losses requires factors selected for the corresponding basis.
Also distinguish development from trend. Development projects the observed experience period to its ultimate value. A trend adjustment changes the cost level to a different period. If the selected loss development already reflects changes in claim amounts as claims emerge, applying an additional adjustment for the same changes can duplicate an effect. Any separate trend step should have a clearly identified purpose and timing, and should not be confused with bringing a claim count or loss amount to ultimate.
A practical diagnostic is to compute pure premium both ways:
These are algebraically equal when the same developed claim count and losses are used. If reported values do not agree, there is a calculation error, inconsistent rounding, or a mismatch in the data or adjustments. The equality cannot prove that a factor selection is actuarially suitable; it checks internal arithmetic and alignment.
Check Your Understanding
For each problem, assume the stated development factors are selected for the stated bases and experience periods. Check that the component calculation agrees with developed losses per exposure.
- A hypothetical book has 2,500 exposures, 200 reported claims, and $900,000 incurred losses. The count factor is 1.30 and the loss factor is 1.40. Find ultimate frequency, severity, and pure premium. (Answer: 0.1040 claims per exposure; $4,846.15 per claim; $504.00 per exposure.)
- For 6,000 exposures, 450 reported claims at 12 months have a count factor of 1.20. Losses of $2,160,000 at 24 months have a loss factor of 1.10. Assuming the data describe the same cohort and compatible claim populations, calculate the ultimate components. (Answer: 540 claims; $2,376,000 ultimate losses; 0.0900 claims per exposure; $4,400.00 per claim; $396.00 per exposure.)
- Observed severity is $3,000.00 on a consistent basis. The count factor is 1.25 and the loss factor is 1.50. Find the implied severity factor and ultimate severity. Explain why using 1.50 as the severity factor is incorrect. (Answer: implied severity factor 1.20; ultimate severity $3,600.00 per claim; applying 1.50 directly incorrectly gives $4,500.00.)
- A cohort has 4,000 exposures. At 12 months it has 300 claims and $1,080,000 losses, with count and loss factors 1.20 and 1.50. At 24 months it has 360 claims and $1,440,000 losses, with factors 1.00 and 1.125. Do the snapshots reconcile, and what are ultimate frequency, severity, and pure premium? (Answer: Yes; 360 claims and $1,620,000 losses from either snapshot; 0.0900 claims per exposure; $4,500.00 per claim; $405.00 per exposure.)
- A hypothetical portfolio has 10,000 exposures, 800 reported claims, and $3,200,000 in losses. The selected count factor is 1.25 and the loss factor is 1.40. Calculate ultimate severity and pure premium, then identify the implied severity factor relative to observed severity. (Answer: 1,000 claims; $4,480,000 ultimate losses; $4,480.00 ultimate severity; $448.00 per exposure; implied severity factor 1.12.)