Combining Mature And Immature Periods Without Equal Weighting

Why Periods Need Different Weights

A loss period with more exposure usually contains more information about the underlying loss rate than a period with less exposure. Maturity matters too: a period that has reported most of its expected losses generally provides more information than an equally exposed period with little loss development. Consequently, combining period estimates by taking a simple average can give too much influence to a small or immature period.

The central idea is to weight each period by its effective exposure: its exposure adjusted for the proportion of ultimate losses expected to have emerged. The result is a pooled estimate that gives greater influence to experience supported by more expected reported claims. It is aligned with the SOA Exam ASTAM (Advanced Short-Term Actuarial Mathematics) syllabus.

Definition: For period $j$, let $E_j$ be its exposure, $q_j$ the expected proportion of ultimate losses reported by the valuation date, and $C_j$ the cumulative reported loss amount. The period’s effective exposure is $E_jq_j$. Its development-adjusted loss-rate estimate is $r_j=C_j/(E_jq_j)$, provided $E_jq_j>0$.

The $q_j$ value is a maturity proportion, not a credibility factor selected independently of development. For example, if a period is expected to have reported 40% of its ultimate losses, then $q_j=0.40$. If a cumulative development factor is used instead and is defined as ultimate divided by reported, then $q_j$ is its reciprocal. Consistent definitions and valuation dates are essential.

The weighting rule below uses a common underlying loss rate across the periods being combined. It is not a rule for forcing periods with genuinely different expected rates to have the same rate. If there are known differences in coverage, mix, trend, or risk level, the periods may need to be adjusted or grouped before pooling.

Model and Weighted Estimate

Suppose cumulative losses arise from a common loss process across periods. Let $m$ be the expected ultimate loss amount per unit of exposure. The expected reported amount for period $j$ is then $E_jq_jm$. For the weighting argument, assume periods are independent, have the same loss process and maturity assumptions, and have a variance for reported losses proportional to $E_jq_j$. This proportionality holds, for example, under a compound Poisson model with common claim frequency and severity and independent emergence of claims.

The estimate based on period $j$ alone is $r_j=C_j/(E_jq_j)$. Under the stated model, it is unbiased for $m$, and its variance is inversely proportional to $E_jq_j$. Thus, a period with twice the effective exposure has half the variance of its rate estimate, all else equal. Inverse-variance weighting therefore gives weights proportional to effective exposure.

\widehat{m} = \frac{\sum_j (E_jq_j)r_j}{\sum_j E_jq_j} = \frac{\sum_j C_j}{\sum_j E_jq_j}.

The second expression follows because $(E_jq_j)r_j=C_j$. It is often the quickest way to calculate the pooled estimate: add reported losses and divide by total effective exposure. The corresponding estimated ultimate losses for the included periods are $\widehat{U}_j=E_j\widehat{m}$. If all those periods are being reserved together, the estimated reserve is the estimated total ultimate less the total already reported.

Condition: Effective-exposure weighting is appropriate when the periods can reasonably be treated as observations of a common loss rate and the reported amount has information proportional to $E_jq_j$. If rates differ systematically, maturity proportions are incompatible, or periods are dependent, the pooled estimate may not be appropriate as written.

Why the Weighting Formula Holds

A short derivation makes the role of maturity explicit. To keep the probability model concrete, suppose each claim costs exactly $1,000, and losses are recorded in thousands of dollars. Claim counts in period $j$ are Poisson with mean $E_jq_j\theta$, where $\theta$ is the common expected claim count per exposure unit at ultimate. This model makes $C_j$, measured in thousands, equal to the reported claim count. It gives $E[r_j]=\theta$ and $\operatorname{Var}(r_j)=\theta/(E_jq_j)$.

For independent periods, consider a weighted average $\sum_j w_jr_j$, where the weights sum to 1. Its variance is the sum of the weighted period variances. Minimizing that variance subject to the weights summing to 1 gives weights proportional to the reciprocal variances. Since the reciprocal variance for period $j$ is proportional to $E_jq_j$, normalizing the weights gives the formula below.

w_j = \frac{E_jq_j}{\sum_k E_kq_k}, \qquad \sum_j w_jr_j = \frac{\sum_j C_j}{\sum_j E_jq_j}.

The same calculation explains why an immature period is not discarded: it contributes information in proportion to its effective exposure, even when that contribution is modest. If $q_j=0$, the period has no reported experience under this model and cannot supply a rate estimate on its own. It may still contribute exposure when projecting ultimate losses after a rate has been estimated from other periods.

Worked Applications

Worked Example: Combining Two Periods With Different Maturity

A hypothetical portfolio has two independent accident periods. Exposure is 100 units in the first period and 200 units in the second. Their expected reported proportions are $q_1=0.80$ and $q_2=0.30$, and cumulative reported losses are $800,000 and $360,000. Estimate the common loss rate and the total reserve. Assume a common Poisson claim rate per exposure unit, claims of exactly $1,000 each, no discounting, and no mortality assumption (mortality is not relevant to this property-loss model). Losses and reserves are stated in thousands of dollars.

Method. Calculate each effective exposure, pool the experience using those weights, and subtract reported losses from estimated ultimate losses. The common mistake to avoid is giving the two period rates equal weight merely because there are two periods.

The effective exposures are $E_1q_1=100(0.80)=80$ and $E_2q_2=200(0.30)=60$. The two period loss-rate estimates are $800/80=10$ and $360/60=6$, in thousands of dollars per exposure unit.

The pooled rate is the total reported loss divided by total effective exposure:

\widehat{m} = \frac{800+360}{80+60} = \frac{1{,}160}{140} = 8.285714\ldots

Total exposure is $100+200=300$, so estimated ultimate losses are $300(8.285714\ldots)=2{,}485.714\ldots$ thousand dollars. Reported losses total $800+360=1{,}160$ thousand dollars. Therefore, the reserve estimate is $2{,}485.714\ldots-1{,}160=1{,}325.714\ldots$ thousand dollars. Rounding only the final monetary answer to the nearest cent gives a reserve of $1,325,714.29.

Check: the period ultimate estimates are $100(8.285714\ldots)=828.571\ldots$ and $200(8.285714\ldots)=1{,}657.143\ldots$ thousand dollars. Subtracting the respective reported losses gives reserves of $28.571\ldots$ and $1{,}297.143\ldots$ thousand dollars; their sum is $1{,}325.714\ldots$, as above.

Worked Example: Converting Development Factors to Weights

A hypothetical insurer reviews two independent lines that can reasonably be treated as having a common ultimate loss rate after adjustment for coverage mix. The first has exposure 120 and a cumulative development factor of 1.25; reported losses are $1,080,000. The second has exposure 100 and a cumulative development factor of 2.50; reported losses are $210,000. Estimate the common loss rate and combined reserve. Assume claims are Poisson with a common rate per exposure unit, each claim costs exactly $1,000, there is no discounting, and mortality is not relevant. Amounts are in thousands of dollars.

Method. Convert each ultimate-to-reported development factor to a reported proportion by taking its reciprocal. Then use effective exposure, rather than exposure alone, to pool the rates.

For the first line, $q_1=1/1.25=0.80$, so effective exposure is $120(0.80)=96$. Its rate estimate is $1{,}080/96=11.25$. For the second line, $q_2=1/2.50=0.40$, so effective exposure is $100(0.40)=40$. Its rate estimate is $210/40=5.25$.

The combined reported losses are $1{,}080+210=1{,}290$ thousand dollars, and total effective exposure is $96+40=136$. Therefore:

\widehat{m}=\frac{1{,}290}{136}=9.485294\ldots

Total exposure is $120+100=220$, so estimated ultimate losses are $220(9.485294\ldots)=2{,}086.765\ldots$ thousand dollars. The reserve is $2{,}086.765\ldots-1{,}290=796.765\ldots$ thousand dollars, or $796,764.71 to the nearest cent.

The calculation checks because the same result is obtained from the period ultimate estimates: $120(9.485294\ldots)+100(9.485294\ldots)=2{,}086.765\ldots$ thousand dollars. The common trap is to use $1.25$ and $2.50$ as reported proportions; they are development factors, so the corresponding proportions are their reciprocals.

Worked Example: Three Periods With Unequal Information

A hypothetical property program has three independent periods with exposures 50, 100, and 200. The respective reported proportions are 0.80, 0.40, and 0.10, and reported losses are $400,000, $240,000, and $200,000. Estimate the pooled rate and total reserve. Assume a common Poisson claim rate per exposure unit, a fixed claim amount of $1,000, no discounting, and no mortality assumption (not applicable here). All loss amounts are in thousands of dollars.

Method. Weight the period rates by effective exposure, then project the common rate across all 350 exposure units. This method avoids allowing the third period’s small emerged fraction to count as much as a more mature period.

Effective exposures are $50(0.80)=40$, $100(0.40)=40$, and $200(0.10)=20$. The period rates are $400/40=10$, $240/40=6$, and $200/20=10$. The total reported losses are $400+240+200=840$, and total effective exposure is $40+40+20=100$. Thus:

\widehat{m}=\frac{840}{100}=8.4

Estimated ultimate losses are $(50+100+200)(8.4)=350(8.4)=2{,}940$ thousand dollars. The reserve is $2{,}940-840=2{,}100$ thousand dollars, or $2,100,000.00 to the nearest cent.

Check by period: estimated ultimate losses are $420$, $840$, and $1{,}680$ thousand dollars. Their sum is $2{,}940$; subtracting reported losses of $400$, $240$, and $200$ gives reserves of $20$, $600$, and $1{,}480$, which sum to $2{,}100$ thousand dollars.

Common Trap and Practical Checks

A tempting but incorrect calculation in the first example is the unweighted average $(10+6)/2=8$. That gives the smaller effective exposure in the second period as much influence as the first period, despite the difference in information. The corrected estimate is $(80(10)+60(6))/(80+60)=8.285714\ldots$, which is exactly the pooled reported losses divided by total effective exposure.

Before using the estimate, check the following:

Use matching bases. Exposure, reported losses, and the maturity proportion must refer to the same period, coverage, and valuation date.
Convert development factors correctly. When a factor is ultimate divided by reported, the corresponding reported proportion is its reciprocal.
Keep the projection target clear. Estimate a common rate using effective exposure, but project ultimate losses using total exposure, not total effective exposure.
Assess whether pooling is defensible. Known shifts in expected loss level, exposure mix, or development pattern may call for adjustments or separate estimates.
Reconcile the reserve. The combined reserve should equal estimated ultimate losses less cumulative reported losses on the same basis.

Unequal weighting is not an optional refinement when periods have materially different exposure and maturity under a common-rate model: it is the way the model accounts for differing amounts of information. Its value depends on using appropriate maturity proportions and a defensible common-rate assumption.

Check Your Understanding

For each problem, use effective exposure to combine the period experience. Unless stated otherwise, assume independent periods, a common Poisson claim rate per exposure unit, fixed claim amounts of $1,000, no discounting, and no mortality assumption. Amounts are in thousands of dollars.

Two periods have $(E_1,q_1,C_1)=(80,0.75,480)$ and $(E_2,q_2,C_2)=(160,0.25,320)$. Find the pooled rate and the combined reserve. (Answer: rate $8$; reserve $1,120,000.)
Two periods have exposures 50 and 150, reported proportions 0.80 and 0.20, and reported losses 360 and 180. Find the pooled rate per exposure unit. (Answer: $540/70=7.7143$.)
Three periods have $(E_j,q_j,C_j)$ equal to $(40,0.50,120)$, $(100,0.30,240)$, and $(200,0.10,100)$. Find the pooled rate and combined reserve. (Answer: rate $460/70=6.5714$; reserve $1,774,285.71.)
Two development-adjusted rate estimates are 11 and 7. Their effective exposures are 20 and 80. Find their effective-exposure-weighted rate. (Answer: $7.8$.)
Two periods have exposures 120 and 80, reported proportions 0.50 and 0.25, and reported losses 420 and 140. Find the pooled rate and combined reserve. (Answer: rate $7$; reserve $840,000.)

What You'll Learn